UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Instrumentation and ultrasound for epidural anesthesia Tran, Denis 2010

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2010_fall_tran_denis.pdf [ 7.02MB ]
Metadata
JSON: 24-1.0064993.json
JSON-LD: 24-1.0064993-ld.json
RDF/XML (Pretty): 24-1.0064993-rdf.xml
RDF/JSON: 24-1.0064993-rdf.json
Turtle: 24-1.0064993-turtle.txt
N-Triples: 24-1.0064993-rdf-ntriples.txt
Original Record: 24-1.0064993-source.json
Full Text
24-1.0064993-fulltext.txt
Citation
24-1.0064993.ris

Full Text

INSTRUMENTATION AND ULTRASOUND FOR EPIDURAL ANESTHESIA   by  Denis Tran  B. Eng., McGill University, 2003 M. Eng., McGill University, 2005  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  Doctor of Philosophy  in  The Faculty of Graduate Studies  (Electrical and Computer Engineering)             THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  August 2010  © Denis Tran, 2010  ii Abstract  Lumbar epidural anesthesia is used for alleviating the pain of labor and for surgery. Here, a catheter is threaded through a Tuohy needle that is traditionally inserted using the loss- of-resistance technique to confirm entry into the epidural space.  This research begins with a study of the loss-of-resistance through instrumentation. Sensors measure 1)the force applied at the plunger by the anesthesiologist, 2)the pressure at the needle tip and 3)the position of the plunger relative to the syringe. The “feel” in different tissues is quantified for porcine subjects ex vivo and human subjects in vivo.  A vertebra counting protocol is developed to identify the desired vertebral interspaces. Ultrasound is then used to measure anatomical distances such as the distance between the skin and ligamentum flavum and surrogate measures compared to the actual needle insertion depth. Good correlation is only found between skin-to-ligamentum flavum and the actual needle insertion depth.  Next, a real-time in-plane ultrasound technique is developed with a needle guide fixing the needle trajectory to the ultrasound transducer. This allows the anesthesiologist to guide the insertion of the epidural needle as an “aim-and-insert” method. In 18 of 19 subjects, the procedure was successfully performed.  The key limitation of ultrasound in this application is the image quality that inhibits interpretation of the images. A median-based spatial compounding with warping is performed to align the anatomical features of different beam-steered images and combine them to obtain a single enhanced image. This method is tested on image sets of phantoms and lumbar anatomy of 23 human subjects and shows a significant improvement in noise reduction and clarity.  Another limitation is the interpretation of ultrasounds of the spinal anatomy requires understanding of ultrasound. An automatic detection algorithm is developed based on the experienced sonographer’s method of detecting the ligamentum flavum in ultrasounds. This novel method is tested on ultrasounds of the lumbar anatomy in 20 human subjects and shows the method successfully detects the ligamentum flavum in 34 of 39 cases.  The main conclusion is that specialized ultrasound tools and protocols are needed to accomodate the range of patients and levels of experience of practitioners.  iii Preface  This manuscript conforms to the traditional format. My supervisor Robert N. Rohling provided the general research topic of providing ultrasound guidance for epidurals, including the concept of instrumenting the loss-of-resistance technique and spatial compounding to improve ultrasound quality. He provided ongoing feedback and assisted in editorial revising of manuscripts including journals, conferences and this manuscript. Otherwise, all of the content of this manuscript is original material by the author, including the literature review, methods, data analysis, and conclusions. Part of data acquisition was done with the aid of medical personnel, including Dr. Allaudin Kamani (MD FRCPC FACA, Dept. Anesthesiology, Pharmacology and Therapeutics, University of British Columbia), Dr. Carly E. Peterson (MD, Dept. Anesthesiology, Pharmacology and Therapeutics, University of British Columbia), Dr. Simon Massey (MD, Dept. Anesthesiology, British Columbia Women’s Hospital and Health Centre), Dr. Fahed Marzouqi (MD, Dept. Anesthesiology, Pharmacology and Therapeutics, University of British Columbia), Dr. Elias al-Attas (MD, Dept. Anesthesiology, Pharmacology and Therapeutics, University of British Columbia) and Victoria A. Lessoway (RDMS, Department of Ultrasound, British Columbia Women’s Hospital and Health Centre).  Five journal papers and twelve conference papers have resulted from this research, and the author list includes those listed above. For some publications, the work of student King-Wei Hor (M. Sc.) was incorporated and he also appears on the author list. For this thesis manuscript, the work by King-Wei Hor is omitted and replaced with a brief description and a reference to the appropriate publication. Since each chapter roughly  iv corresponds to one journal paper, a more detailed description of co-authorship is provided.  A version of Chapter 2 appeared in the Institute of Electric and Electronic Engineers (IEEE) Transactions on Biomedical Engineering [Tran-2009b]. This article was co- authored with King-Wei Hor, Dr Allaudin A Kamani, Victoria Lessoway and Dr Robert N. Rohling. King-Wei Hor calibrated the sensors and modelled the pressure using the force and displacement data. He also compared the intermittent and continuous pressure techniques. His work has been presented in his Master’s thesis. The author calibrated a second position sensor, developed the experimental setup to measure the loss-of- resistance in porcine subjects to compare the anesthesiologist feel in the paramedian approach against the midline approach. The author also modified the apparatus so that a sterile set of sensors could be used in the operating room to conduct experiments on human subjects and acquired the sensor data for all subjects. The author wrote the journal paper. Victoria Lessoway is the sonographer and participated in discussions. Dr Allaudin Kamani is the anesthesiologist who performed the epidural needle insertions in human and porcine subjects. Dr Robert N. Rohling provided feedback and edited the manuscript.  A version of Chapter 3 appeared in Anesthesia and Analgesia [Tran-2009a]. This article was co-authored with Dr Allaudin A Kamani, Victoria Lessoway, Dr Carly E. Peterson, King-Wei Hor and Dr Robert N. Rohling. Dr Allaudin A. Kamani is the anesthesiologist who performed the epidural needle insertions. Victoria Lessoway is the sonographer and helped implement the counting and scanning protocol. King-Wei Hor’s contribution has  v already been mentioned in Chapter 2 (the journal paper included a section on sensors). The author designed the scanning and counting protocol from the 12th rib down and acquired and analyzed data. The author also wrote the journal paper. Dr Robert N. Rohling provided feedback and edited the manuscript.   A version of Chapter 4 appeared in the Canadian Journal of Anesthesia [Tran-2010a]. This article was co-authored with Dr Allaudin A Kamani, Dr Simon Massey, Dr Elias Al- Attas, Victoria Lessoway and Dr Robert N. Rohling. Drs Kamani, Massey and Al-Attas are the anesthesiologist who performed the epidural needle insertions using the apparatus. Victoria Lessoway performed all the ultrasound scans and measurements. The author designed and assembled the sterile apparatus for real-time needle insertion, calibrated the needle trajectory, designed a new paramedian approach for epidural needle insertion. The author derived all the geometric relations between the measurements. The author acquired and analyzed data and wrote the journal paper. Dr Robert N. Rohling provided feedback and edited the manuscript.   A version of Chapter 5 appeared in Computerized Medical Imaging and Graphics [Tran- 2009c]. This article was co-authored with King-Wei Hor, Dr Allaudin A Kamani, Victoria Lessoway and Dr Robert N. Rohling. Dr Rohling developed the idea of the adaptive median compounding and King-Wei Hor tested it on the cube phantom. Dr Allaudin A. Kamani is the anesthesiologist who performed the epidural needle insertions.  vi Victoria Lessoway is the sonographer who performed all the ultrasound scans. The original idea of using linear prediction to reduce computational cost was developed by the author. The author developed all the framework and C++ code for testing and comparison of the algorithms. The author also did all the derivations and analysis for the refraction and speed of sound errors for a multilayered structure with a linear transducer and a curvilinear transducer when spatial compounding is used. The author built a multilayered spine phantom and tested the algorithm on the phantom and on human subjects. The author also wrote the journal paper. Dr Robert N Rohling provided feedback and edited the manuscript.  A version of Chapter 6 is in press in IEEE Transactions on Biomedical Engineering [Tran-2010b]. This article was co-authored with Dr Robert N. Rohling. The original idea of template matching and design of the templates were developed by the author. The author designed, implemented and tested the method on porcine and human subjects. The author also wrote the journal paper. Dr Robert N Rohling provided feedback and edited the manuscript. Preliminary and ongoing results of this research were also presented at 12 conferences: [Tran-2007] [Tran-2008a] [Peterson-2007] [Kamani-2007] [Peterson-2008] [Hor-2007b] [Tran-2008b] [Kamani-2008] [Al-Attas-2009] [Tran-2010c] [Tran-2010d] [Lo-2010].  Ethics approval was obtained from the Clinical Review Ethics Board of the British Columbia Women’s Hospital and Health Center (C05-0409).  vii Table of contents Abstract .......................................................................................................................................................... ii Preface........................................................................................................................................................... iii Table of contents .......................................................................................................................................... vii List of tables .................................................................................................................................................. ix List of figures ................................................................................................................................................. x List of abbreviations ..................................................................................................................................... xii Glossary....................................................................................................................................................... xiii Acknowledgments ........................................................................................................................................ xv 1 Introduction........................................................................................................................................... 1 1.1 Motivation ................................................................................................................................... 1 1.2 Background ................................................................................................................................. 8 1.3 Thesis objectives ....................................................................................................................... 42 1.4 Chapter summary ...................................................................................................................... 44 2 Instrumentation of epidural needle insertion....................................................................................... 47 2.1 Introduction ............................................................................................................................... 47 2.2 Methods..................................................................................................................................... 51 2.2.1 Experiment 1: comparison of the midline approach and the paramedian approach on porcine tissue ...................................................................................................................................... 54 2.2.2 Experiment 2: instrumentation of human subjects........................................................... 54 2.3 Results ....................................................................................................................................... 58 2.3.1 Experiment 1: comparison of the midline approach and the paramedian approach on porcine tissue ...................................................................................................................................... 58 2.3.2 Experiment 2: instrumentation of human subjects........................................................... 60 2.4 Discussion ................................................................................................................................. 62 2.5 Conclusion and future work ...................................................................................................... 64 3 Pre-insertion ultrasound for epidural needle insertion ........................................................................ 66 3.1 Introduction ............................................................................................................................... 66 3.2 Methods..................................................................................................................................... 67 3.3 Results ....................................................................................................................................... 77 3.4 Discussion ................................................................................................................................. 82 4 Real-time ultrasound guidance for epidural needle insertion.............................................................. 86 4.1 Introduction ............................................................................................................................... 86 4.2 Methods..................................................................................................................................... 89 4.2.1 Subject selection .............................................................................................................. 89 4.2.2 Intervertebral level identification..................................................................................... 90 4.2.3 Geometric measurement for aim-and-insert technique .................................................... 91 4.2.4 Needle insertion and block procedure.............................................................................. 97 4.2.5 Statistical analysis............................................................................................................ 98 4.3 Results ....................................................................................................................................... 99 4.4 Discussion ............................................................................................................................... 109 5 Adaptive spatial compounding for ultrasound images of lumbar anatomy....................................... 114 5.1 Introduction ............................................................................................................................. 114 5.2 Methods................................................................................................................................... 115 5.2.1 The need for registration................................................................................................ 116 5.2.2 Refraction and speed of sound error in a curvilinear transducer.................................... 121 5.2.3 The registration and compounding algorithm ................................................................ 129 5.3 Results ..................................................................................................................................... 138 5.4 Discussion ............................................................................................................................... 147 5.5 Conclusions ............................................................................................................................. 151 6 Automatic detection of ligamentum flavum in ultrasound images of the lumbar anatomy .............. 153 6.1 Introduction ............................................................................................................................. 153 6.2 Methods................................................................................................................................... 154  viii 6.2.1 Overview of algorithm................................................................................................... 154 6.2.2 Ridge map generation .................................................................................................... 154 6.2.3 Lamina extraction .......................................................................................................... 160 6.2.4 LF extraction.................................................................................................................. 164 6.2.5 Performance measures ................................................................................................... 167 6.2.6 Statistical significance ................................................................................................... 167 6.2.7 Clinical trial ................................................................................................................... 168 6.3 Results ..................................................................................................................................... 169 6.4 Discussion ............................................................................................................................... 175 6.5 Conclusions ............................................................................................................................. 184 7 Conclusion and future work.............................................................................................................. 185 7.1 Summary ................................................................................................................................. 185 7.2 Contributions........................................................................................................................... 190 7.3 Future work ............................................................................................................................. 194 7.4 Anecdotal comments ............................................................................................................... 199 Appendix A: Consent form ........................................................................................................................ 214 Appendix B: Ethics approval...................................................................................................................... 220 Appendix C: Pressure modeling ................................................................................................................. 221 Experiment 1: Sensor reliability ............................................................................................................ 221 Experiment 2: Pressure estimate for both continuous and intermittent techniques on porcine tissue.... 222 Appendix D: Warping and linear prediction .............................................................................................. 229   ix List of tables  Table 1-1 Visibility in different ultrasound planes....................................................................................... 20 Table 1-2 Success of ultrasound controlled epidural anesthesia................................................................... 24 Table 1-3 Results for ultrasound-guided epidurals for the children 6months and older group..................... 27 Table 2-1 Experiment 1: flow rate, force and pressure for the interspinous ligament, muscle, LF, and epidural space  for porcine subjects using the midline and paramedian  approaches.......................... 59 Table 2-2 Experiment 2: flow rate, force and estimated pressure  for human subjects ................................ 61 Table 3-1 Subject biometrics and data on ES depth and surrogate measures............................................... 77 Table 4-1 Subject biometrics and comments.............................................................................................. 101 Table 4-2 Difference between anesthesiologist mark and sonographer mark (mm) to compare palpation versus ultrasound method.................................................................................................................. 102 Table 4-3 Skin-to-epidural space depth with and without transducer force ............................................... 103 Table 4-4 Distance A.................................................................................................................................. 106 Table 5-1 Values of crosscorrelation coefficients for the spine phantom................................................... 135 Table 5-2 Maximum Laplacian at the LF for the spine phantom ............................................................... 136 Table 5-3 NCC coefficients for a human patient for four selected beam-steered images........................... 141 Table 5-4 Maximum Laplacian of line profile of LF for a human subject ................................................. 141 Table 5-5 Maximum Laplacian at the LF and gradient of the lamina for the spine phantom..................... 145 Table 5-6 Maximum Laplacian at the LF and the gradient of the lamina for human subjects ................... 145 Table 5-7 SNR for light and dark regions and CNR for the spine phantom............................................... 146 Table 5-8 SNR for light and dark regions and CNR for human subjects ................................................... 146 Table 5-9 Computational costs ................................................................................................................... 147 Table 6-1 Patient detailed information ....................................................................................................... 170 Table 6-2 The accuracy of the automatic detection of LF .......................................................................... 173 Table 6-3 Computational cost of automatic detection of LF ...................................................................... 175 Table C-1 Root-mean square (RMS) error for the continuous and intermittent technique......................... 227 Table C-2 Flow rate, force (Fa) and pressure (P) for  midline approach..................................................... 228  x  List of figures  Figure 1-1 Lumbar vertebral anatomy............................................................................................................ 3 Figure 1-2 Tuohy needle and epidural catheter .............................................................................................. 4 Figure 1-3 Tuffier’s line ................................................................................................................................. 6 Figure 1-4 Learning curves in obstetric epidural anesthesia .......................................................................... 7 Figure 1-5 Nerve stimulation needle insertion apparatus ............................................................................. 10 Figure 1-6 Pressure transduction signal........................................................................................................ 11 Figure 1-7 Episure syringe ........................................................................................................................... 12 Figure 1-8 Fluoroscopy for epidurography................................................................................................... 13 Figure 1-9 Computed tomography fluoroscopy............................................................................................ 14 Figure 1-10 Ultrasound image of the lumbar anatomy in 1984.................................................................... 15 Figure 1-11 Ultrasound images of the lumbar anatomy ............................................................................... 17 Figure 1-12 Ultrasound image with colour Doppler overlay........................................................................ 21 Figure 1-13 Blood patch observed with real-time ultrasound ...................................................................... 23 Figure 1-14 Diagram of real-time observation of epidural needle being inserted for a CSE........................ 25 Figure 1-15 Real-time ultrasound-guided epidural needle insertion performed on a child with a two person method ................................................................................................................................................ 26 Figure 1-16 Ultrasound image of a single-operator real-time in-plane ultrasound guided epidural needle insertion .............................................................................................................................................. 28 Figure 1-17 The “bat” representation of the lumbar vertebral structures ..................................................... 30 Figure 1-18 Dr Malcolm Watson’s split array transducer ............................................................................ 31 Figure 1-19 Some post-processing techniques on ultrasound images .......................................................... 34 Figure 1-20 Example of beam-steering ........................................................................................................ 36 Figure 1-21 Bone segmentation using shadows ........................................................................................... 39 Figure 1-22 Bone segmentation using phase symmetry ............................................................................... 40 Figure 2-1 A comparison of a transversal cross-sections of human subject and porcine subject. ................ 48 Figure 2-2 A 3-way stopcock is used to connect the syringe, pressure sensor and epidural needle ............. 50 Figure 2-3 The force sensor, pressure sensor, and displacement sensor mounted to the syringe ................. 52 Figure 2-4 The set of metal tools to be sterilized after every procedure....................................................... 56 Figure 2-5 Redesigned casing for easy assembly stage ................................................................................ 58 Figure 2-6 A sample graph of force, pressure and displacement measurements .......................................... 60 Figure 2-7 A sample graph of a midline approach on a human subject showing ......................................... 62 Figure 3-1 Experimental setup ..................................................................................................................... 68 Figure 3-2 Counting intervertebral spaces.................................................................................................... 71 Figure 3-3 Ultrasound image of the sacrum ................................................................................................. 71 Figure 3-4 Ultrasound image of the transverse processes ............................................................................ 72 Figure 3-5 Ultrasound image of the facet joints ........................................................................................... 72 Figure 3-6 Paramedian lumbar anatomy....................................................................................................... 73 Figure 3-7 Subject with a significant layer of subcutaneous fat ................................................................... 75 Figure 3-8 Skin-to-LF depth measured by ultrasound correlated with needle insertion depth..................... 79 Figure 3-9 Bland-Altman plot of the difference between ultrasound-measured skin-to-LF depth and the needle insertion depth ......................................................................................................................... 79 Figure 3-10 The correlation of needle insertion depth with biometrics........................................................ 80 Figure 3-11 Needle insertion depth at L3-4 correlated to the distance from skin to the tip of the transverse process L4 and the thickness of overlaying subcutaneous fat. ............................................................ 81 Figure 4-1 a) Transverse processes b) facet joints c) ultrasound image with dashed needle guide line representing the predicted needle path................................................................................................ 91 Figure 4-2 Assembled sterile transducer with needle guide and epidural needle attached........................... 92 Figure 4-3 Schematic of probe and needle guide geometry ......................................................................... 93 Figure 4-4 Geometry relating the measurement of skin-to-LF depth ........................................................... 94 Figure 4-5 Schematic of distance A measurement ....................................................................................... 95  xi Figure 4-6 Distance B versus 1.27*distance E – 8.7 mm ........................................................................... 104 Figure 4-7 Bland-Altman plot of the measured distance B versus calculated from eq. 4.3........................ 105 Figure 4-8 Epidural space depth along the needle guide line overlay versus actual needle depth.............. 107 Figure 4-9 Bland-Altman plot of the epidural space depth along the needle guide line overlay (distance B+C+D) versus measured actual needle depth ................................................................................. 108 Figure 5-1 Linear ultrasound transducer image of a two-layer structure.................................................... 118 Figure 5-2 Difference between the reference frame apparent position and the beam-steered frame apparent position for a point at the midline for a linear transducer ................................................................. 120 Figure 5-3 Curvilinear ultrasound transducer image of a two-layer structure ............................................ 122 Figure 5-4 Angles φ 1 and φ 2 for a point at x = 20mm from the midline ................................................... 124 Figure 5-5 Angles φ 1 and φ 2 for a point at the midline (x = 0), φ 1=0. ...................................................... 125 Figure 5-6 Difference between the reference frame apparent position and the beam-steered frame apparent position for a point at x = 20 mm from the midline for a curvilinear transducer .............................. 127 Figure 5-7 Difference between the reference frame apparent position and the beam-steered frame apparent position for a point at the midline (x = 0) for a curvilinear transducer ............................................. 128 Figure 5-8 Comparison of human and phantom subjects ........................................................................... 133 Figure 5-9 Adaptive spatial compounding on spine phantom .................................................................... 139 Figure 5-10 Adaptive spatial compounding with linear prediction on spine phantom ............................... 140 Figure 5-11 Adaptive spatial compounding on human subject................................................................... 144 Figure 5-12 Beam-steered images of a difficult to image patient in the first clinical study ....................... 149 Figure 6-1 Overview of the LF detection algorithm................................................................................... 154 Figure 6-2 Log-Gabor filter........................................................................................................................ 157 Figure 6-3 Orientations of the Log-Gabor filter used for phase symmetry ................................................ 158 Figure 6-4 Flow chart of the ridge map generation .................................................................................... 158 Figure 6-5 Phase symmetry response using the 3 angles of interest........................................................... 159 Figure 6-6 Ridge map................................................................................................................................. 159 Figure 6-7 Templates for the detection of the lamina and LF .................................................................... 163 Figure 6-8  Search for the laminae ............................................................................................................. 164 Figure 6-9  Search of the LF....................................................................................................................... 166 Figure 6-10  The steps of the LF detection on ultrasound image of a human subject ................................ 171 Figure 6-11  The five cases of failed LF detection ..................................................................................... 172 Figure 6-12  Bland-Altman plots of automatic detection ........................................................................... 174 Figure 6-13   The distribution of the LF detections in the crosscorrelation with lamina and LF templates space ................................................................................................................................................. 181 Figure 6-14 An ultrasound image of the lumbar anatomy with the LF detected ........................................ 182 Figure 7-1 Anesthesiologist using a prototype of the dual-probe............................................................... 195 Figure 7-2 Graphical user interface of the dual-probe................................................................................ 196 Figure 7-3 Porcine thoracic paramedian spinal anatomy............................................................................ 199 Figure C-1 Example of epidural needle insertion on porcine tissue using intermittent technique ............. 225 Figure C-2 Example of epidural needle insertion on porcine tissue using continuous technique............... 226 Figure C-3 RMS error between the measured and calculated pressure ...................................................... 227 Figure D-1 Block diagram of the adaptive spatial compounding algorithm............................................... 229 Figure D-2 The warped image and the reference frame in interleaved checkered pattern ......................... 230 Figure D-1 The order of block-matching displacement vector estimation ................................................. 231     xii List of abbreviations  BMI: body mass index CNR: contrast-to-noise ratio CPU: central processing unit CSE: combined spinal-epidural CSF: cerebrospinal fluid ES: epidural space FPGA: field-programmable gate array GB: gigabyte GPS: global positioning system IEEE: Institute of Electrical and Electronic Engineers kg: kilogram kPa: kilopascal L2: lumbar vertebra 2 (similar nomenclature for other lumbar vertebrae) L2-3: interspace between lumbar vertebrae 2 and 3 LF: ligamentum flavum LOR: loss-of-resistance LP: linear prediction – a search strategy based on prior estimates LP2: linear prediction with search area reduction of 2 LP2+: same as LP2 but with a starting point where the no. of features is greatest mA: milliampere mg: milligram MHz: megahertz mL: milliliter mm: millimeter N: Newton NCC: normalized crosscorrelation RAM: random-access memory RMS: root mean square s: second SNR: signal-to-noise ratio T4: fourth thoracic vertebra (similar nomenclature for other lumbar vertebrae) T4-5: interspace between thoracic vertebrae 4 and 5    xiii Glossary  Accidental cannulation: anesthesia catheter is inserted in a blood vessel Anterior: front of the body Beam-steering: ultrasound imaging at different insonation angle Biometric: measurement of biological data Biopsy: removal and examination of sample tissue from living body Blood patch: procedure by which blood is injected in the subdural space to form a clot for repairing a dural puncture Caudad: toward the bottom Cephalad: toward the top Clinical trial: a controlled test on human subjects Contraindication: medical reason that makes it inadvisable to prescribe a procedure Ex-vivo: outside the living body Fluoroscopy: x-ray imaging permitting visualization of inner organs Foramina: space between the bones Freehand: performed without mechanical aids Gauge: thickness of needle, higher gauge is smaller diameter Gradient: first derivative Haptic: relating to the sense of touch Hypoechoic: low ultrasound response Infusion pump: device delivering measured amounts of fluid intravenously Insonation: exposure to ultrasound waves Interpolation: calculation of value between known values In-vivo: inside the living body Ionizing radiation: high-energy radiation that can remove electrons from atoms Laplacian: second derivative Lateral: toward the sides Medial: toward the midline Median: middle value in a sorted list Midline: line along the center of the spine, going from caudad to cephalad Neuraxial: surrounding a major nerve Obstetrics: branch of medicine related to the care of women during pregnancy, childbirth and recuperative period after delivery Oedema: having excessive liquid in intercellular space of tissue Paramedian (parasagittal):plane parallel to the spine, slightly lateral Parturient: in labour Perfusion: passage of fluid into an organ Peripheral block: anesthesia procedure to block the main nerve of a limb  xiv Pitch: frequency of sound Posterior: back of the body Refraction: change of direction of propagating wave Scan-convert: mapping the ultrasound pulse-echo times and intensity into a Cartesian ultrasound image Scoliosis: abnormal curvature of the spine Sterility/aseptic: using methods free of pathological micro-organisms Subcutaneous: beneath the skin Surrogate: substitute Systematic error: an error biased in measurement Therapeutic: ability to restore health Transduction: transfer of energy from mechanical to electrical or vice versa Transverse: crossing from left to right, plane perpendicular to the axis of the spine Visual analogue score: a scale of pain determined by the facial expression  xv Acknowledgments  Professor Robert N. Rohling, Ph. D., for being a very patient and generous supervisor. Dr Allaudin Kamani, Vickie Lessoway and all others from the clinical research team, for sharing interesting conversation between patients and teaching me the clinical aspect of research. Vicky Earle for drawing some of the figures. Special thanks to Dr Jose Carvalho at Mount Sinai Hospital in Toronto for hosting and teaching me the different nuances and state of the art techniques for ultrasound scanning of the lumbar spine.  My family, Tran Van Thach (father), Tran Van Thi An (Mother), Michel and Nathalie for always supporting me and reminding me that they will always be there for whatever my career orientation. The distance has kept us close for all five years. And in particular, my father for proofreading this thesis.  À mes vieux amis du secondaire, le dénommé O5. Je vous aime pour tout, ces jours à ne rien faire, dans des pays bizarres, souvenirs tels que MR, classique. Andrew Berthe, Nobar Khanikian, Tung Tang et Luc Gervais. Avec vous, j’ai assez ri pour toute une vie.  Albert Sutrisno, Budi Kartono, Thomas Diego Prananta, Andre Lei, Ali Baghani, David Yao, Joanna Lam, Cory Lang and Shelly Jang, the friends without whom living in a rainy Vancouver would have been tasteless. John Ko, Brad Howie, Klement and Kevin Mui, Kevin Mark, and Dragon Hearts Magnum, building this team together, the closest to raising a kid together.  And Mandy Tang, the woman who showed me to never give up.    1 1 Introduction 1.1 Motivation Epidural anesthesia is a regional anesthesia technique created in the early 1920’s by Fidel Pages to inject continuous anesthesia into the epidural space (ES) to alleviate pain [Pages-1921]. It is most commonly used in obstetrics i.e. for labour and cesarean delivery, where it is often called an “epidural”. To deliver the anesthesia, a catheter is inserted into the ES through an epidural needle that has been previously inserted into the spine. The challenge is to place the needle tip accurately in the ES on the first attempt.  To understand this challenging procedure, a brief overview of spinal anatomy is described. The spine consists of 33 vertebrae: 7 cervical, 12 thoracic, 5 lumbar, 5 sacral and 4 coccygeal. Epidurals are most commonly performed on the lumbar spine. The 5 lumbar vertebrae are labelled L1 (first lumbar) to L5. Inferior to the lumbar vertebrae is the sacrum, which is formed of 5 fused vertebrae. As shown in Figure 1-1a, the lumbar vertebrae contain each a spinous process, transverse process, articular process or facet joints, and lamina. Connecting the vertebrae are several ligaments, including the supraspinous ligament, interspinous ligament and ligamentum flavum (LF) (Figure 1-1c). The dura mater (Figure 1-1b) is thin membrane protecting the spinal cord. Descending inferiorly, at the lumbar level, the spinal cord separates into branches and progressively becomes the cauda equina, which is a structure containing relatively fewer nerves. Spinae erector muscles are located on both sides of the spine. The ES, situated between the LF and the dura mater, is where the anesthesia should be delivered. In some cases, it is a potential space that opens when saline or air is injected. This means the target is at most a  2 few millimiters (mm) thick and lies immediately behind a stiff LF and thin dura mater. For this reason, a Tuohy needle is used that has a hole at the tip that faces upward so that the catheter can be threaded up the spine in the gap between the LF and the dura mater. The epidural catheter is a tube made of polyamide, polyurethane and silicone. It is inserted through the needle into the ES (Figure 1-2).  It should be clear that the needle insertion encounters multiple tissue types and the trajectory and penetration depth must be carefully chosen to avoid contact with bone and to deliver the anesthesia into the right place (i.e. ES).  3    Figure 1-1 Lumbar vertebral anatomy a) bony structures in the midline b) tissues  c) ligaments (images reproduced with permission from [Tran-2009a])   Articular process Articular process  a) b) Needle c)  4  Figure 1-2 Tuohy needle and epidural catheter shown with a ruler for size comparison.   Epidurals are considered one of the most difficult needle insertion procedures in anesthesia [Filho-2002][Konrad-1998]. Although epidural anesthesia has been practiced for decades, the failure rate remains relatively high (6-25% [LeCoq-1998] [Watts-1992]). The learning of regional anesthesia is also among the most difficult of all anesthesia techniques. In [Kopacz-1996], it is shown that about 20-25 spinals and epidurals should be practiced by a resident anesthesiologist before achieving a statistically significant (p<0.05) performance improvement. It was shown that 90% success is only obtained after an average of 45 spinal and 60 epidural blocks. Needle insertion can fail if the needle trajectory encounters bone before reaching the ES, requiring a subsequent withdrawal of  5 the needle and reinsertion, adding time and discomfort to the patient. Needle insertion can also fail if the needle overshoots the ES and punctures the dura mater, the thin membrane that covers the spinal nerves. This overshoot is typically detected by leakage of cerebrospinal fluid (CSF) from the base of the needle. Complications may include block failure, backache, infection (localized and central nervous system), and headache. Rarer are accidental intravascular injection (0.67%), inadvertent dural puncture (0.61%), paresthesia (0.16%) (ranging from temporary peripheral nerve injury to paralysis) and death or brain injury [Tanaka-1993].  For lumbar epidural needle insertion, the needle’s puncture site has traditionally been selected by palpation of external anatomical landmarks, such as the tips of the spinous processes and Tuffier’s intercrest line [Chestnut-2004], which is a transverse line between the iliac crests that aligns approximately with the fourth lumbar vertebra (L4) (Figure 1-3). The needle is normally inserted midline into the intervertebral space, through the skin, fat, supraspinous ligament, ISL and finally the LF, which is a stiff ligament lying directly above the ES (Figure 1-1c). Just after the start of the procedure, when the needle is partly inserted, the anesthesiologist attaches a syringe to the needle, filled with either air or saline. The position of the needle tip as it passes through the various tissues is then felt through the tactile feedback the anesthesiologist receives at the syringe’s plunger upon injection of saline or air into the tissue. The sensation of a “loss- of-resistance” (LOR) is felt as the needle tip goes from the stiff LF into the ES where fluid is easily injected. This procedure is the current standard of practice and the anesthesiologist relies on experience and qualitative “feel” to estimate the location of the  6 epidural needle tip. Quantitative measurements may be beneficial for improving the training of anesthesiologists in this procedure.  Figure 1-3 Tuffier’s line is an imaginary line between the left and right iliac crests that aligns with the spinous process of L4.    Moreover, the choice of the needle puncture site and insertion trajectory is based on prior knowledge of normal human anatomy with little information about each specific patient. This means that the optimal needle trajectory and needle insertion depth are unknown prior to initial insertion. Medical imaging, such as ultrasound, could be used for guidance.   7 Very experienced anesthesiologists often claim they do not need such ultrasound guidance as their failure rate is low. This is not the case, however, for less experienced anesthesiologists. In 2003, Grau studied the learning of resident anesthesiologists and found that ultrasound can increase the success rate during the learning phase. Two groups of residents were to perform 60 epidurals under supervision, one group using ultrasound for prepuncture information, and one group using the traditional “blind” method. Failure was defined by failing 3 needle insertion attempts, a visual analogue score larger than 1, relocation of the insertion site, or any intervention by the supervisor. It was shown that the control group achieved a success rate of 60% after 10 insertions and a rate of 84% on the next 50 insertions. The ultrasound group achieved a rate of 86% in the first 10 insertions and 94% on the next 50 insertions. Ultrasound was therefore shown to increase the rate of learning, see Figure 1-4.   Figure 1-4 Learning curves in obstetric epidural anesthesia. CG is control group and UG is the ultrasound group. The low is the resident with the lowest rate of success, high is the resident with the highest rate of success and mean is the average of all residents (reproduced with permission from [Grau-2003])    8 Ultrasound imaging, which uses ultrasound waves to help see the anatomy below the skin, is therefore useful to gain more knowledge on the patient’s anatomy prior to the needle insertion. It may also provide real-time guidance during needle insertion.  Ultrasound has not yet been adopted as the standard procedure for epidural needle insertion in anesthesiology, whereas another needle insertion procedure - central line placement - has already adopted ultrasound guidance [Scott-2004]. One possible reason is that ultrasound image interpretation requires some expertise to correctly place the ultrasound transducer on the body and to recognize the relevant structures in the image. Interpreting the ultrasound images is complicated by the presence of ultrasound artifacts and the variability of human anatomy. Since epidural needle insertion is performed usually by non-ultrasound specialised users (anesthesiologists), a computer-assisted interpretation of the ultrasound image may be helpful for the user to gain confidence and efficiency in the detection of the LF and ES. The next section provides a review of these challenges dealt with in previous works by others. More detailed literature reviews on the specific techniques are provided in each of the chapters.  1.2 Background The traditional use of the Tuffier’s line (Figure 1-3) to determine the L3-4 intervertebral level has been shown to be only accurate in 29% of cases versus 71% when using ultrasound [Furness-2002]. The ultrasound technique described in [Furness-2002] uses the sacrum as the reference, however, sacralisation and lumbarisation alter the vertebral  9 count leading to errors in localization of the intervertebral level. Sacralisation (estimated incidence ranging from 1.3% in subjects without lower back pain [Steinberg-2003] to 4.5% [Hahn-1992]) is the fusion of the 5th lumbar vertebra to the sacrum. Lumbarisation (estimates include 3.4% [Moore-1925], 9.5% in subjects without lower back pain [Steinberg-2003], 7.5% [Hahn-1992]) is the separation of the first vertebra of the sacrum from the sacrum. Nerves rooted at each vertebral level affect different levels and limbs of the body so proper identification of the level is important.  Different studies have been done to find a methodology to pick a puncture site and then guide the needle toward and stop at the ES to deliver anesthesia.  Nerve stimulation  Instrumentation has been used previously in different types of needle insertions for regional anesthesia. For example, nerve stimulators (Figure 1-5) have been used on peripheral blocks (infraclavicular brachial plexus, axillary brachial plexus, femoral nerve [Fanelli-1999]) and in lumbar epidurals [Tsui-1999]. With nerve stimulation, an insulated needle is inserted while a current is applied and a motor response is obtained from nearby muscles. As the needle gets closer to a nerve, the current can be reduced to obtain the same motor response. The needle is considered close enough to a nerve when a current of 0.5 milliampere (mA) still provokes a muscle reaction. The anesthetic solution is then administered at the location of the nerve. If the anesthesia is successful, further nerve stimulation will not provoke a muscle reaction [De Andres-2001]. This procedure can be  10 however uncomfortable for the patient as the patient is repeatedly given impulses of electric current, and there may be a significant quantity of anesthesia injected in surrounding tissues before the desired nerve location is reached. This technique has not been adopted widely for lumbar epidurals. Also, the needle insertion remains a blind procedure.  Figure 1-5 Nerve stimulation needle insertion apparatus (images reproduced with permission from [Tsui-1999])   Pressure transduction  Pressure transduction has also been used to confirm the entry of the needle in the ES as well as for catheter placement [Lennox-2007]. Pressure transduction is based on the principle of detecting a pulsatile pressure when the needle tip has reached the ES, see Figure 1-6. The heart pulsations compress the spinal column which is detected as changes in pressure in the ES. It has been shown to be more reliable than the loss-of-resistance method at confirming successful entry into the ES in thoracic epidural needle insertion (79 true positives, 0 false positive, 2 false negative, 2 true negative [Lennox-2007]). The  11 principle is to measure the pressure changes following the cardiac cycle in the spinal column affecting the pressure in the ES. If the insertion is successful, a pulsatile waveform can be detected. This however remains a blind needle insertion method and is only used as a confirmation of successful needle or catheter placement.  It does not provide feedback of needle location relative to the different overlying tissues.   Figure 1-6 Pressure transduction signal (epidural) compared with the ECG signal. When the catheter or needle is in the ES, a signal is observed matching the heart rate. ECG is electrocardiogram, SpO2 is oxygen saturation, ABP is arterial blood pressure, CVP is central venous pressure and EPIDURAL is the pressure measured through the epidural needle or catheter. The pressure in the epidural needle is linked to the ECG only when the needle is in the ES. (Images reproduced with permission from [Lennox-2007])   12 The hanging drop method consists of placing a drop of saline at the hub of the epidural needle. The needle is advanced without syringe. As the needle bevel is entered into the ES, the negative pressure difference (equivalent to the intrapleural pressure) causes the drop of saline to be drawn into the ES. This method is now rarely used, and has been replaced by LOR.  Although the LOR is the current standard of care for epidural needle insertions, it requires the anesthesiologist to keep one hand on the syringe plunger to apply pressure. The Episure™ AutoDetect Syringe (Indigo Orb, Inc., Irvine, CA, USA)[Riley-2007] shown in Figure 1-7 is a spring-loaded LOR syringe providing constant pressure. This relieves the anesthesiologist from the requirement of applying pressure on the plunger.  Figure 1-7 Episure syringe (Image reproduced with permission from [Riley-2007])  X-ray image-guided epidural needle insertion  Fluoroscopy [Johnson-1999] [Johnson-2000] [Diez-1998] has been used previously for real-time guidance for epidural needle insertion. In this procedure, the needle  13 advancement is subsequently observed in real-time and when the needle reaches the ES, a contrast agent is injected (epidurography) followed by a therapeutic injection (Figure 1-8).  Figure 1-8 Fluoroscopy for epidurography (images reproduced with permission from [Johnson- 1999])  Computed-tomography fluoroscopy [Wagner-2004] uses a computed-tomography machine to generate a view of the anatomy not visible in real-time through traditional fluoroscopy (Figure 1-9). This technique allows for precise and real-time visualization of the anatomy and the needle tip. The insertion can be planned to avoid bone structures and be performed rapidly. Unfortunately, both fluoroscopy and computed-tomography fluoroscopy are based on ionizing X-ray radiation and so are unsuitable for parturient patients because of concerns over radiation exposure to the mother and fetus.  14  Figure 1-9 Computed tomography fluoroscopy.  Epidural needle insertion is shown in the transverse view (images reproduced with permission from [Wagner-2004])  Ultrasound has been used in both peripheral anesthesia procedures [Sandhu-2003] [Nowakowski-2007] [Casati-2007] [Grau-2005] [Bonazzi-1995] and in epidural anesthesia [Grau-2001b] [Arzola-2007] to visualize the anatomy and obtain pre-puncture anatomical information as well as for real-time guidance [Grau-2004] [Willschke-2006] [Karmakar-2009]. Interest in epidural ultrasound is growing quickly as image quality improves with the latest generation of ultrasound machines.  The history of ultrasound in epidural anesthesia  Cork was the first to use ultrasonography to measure the skin-to-ES depth in the lumbar spine in 1980 [Cork-1980]. The coefficient of determination (referred to as correlation coefficient) between the ultrasound measured depth and the needle depth was R2=0.84  15 (where R2=1 is perfect correlation and R2=0 is no correlation). Near the same time, Currie also used ultrasound to measure the skin-to-ES depth [Currie-1984]. The ultrasound image quality was relatively low compared to modern standards, as seen in Figure 1-10. Currie needed to use a bag of saline and adjust the image for maximum contrast in order to observe the lamina. Even under these circumstances, they achieved a correlation between the ultrasound skin-to-ES depth and the needle depth of R2=0.96. Some errors were thought to be due to misinterpretation of the echoes. Although image quality was low, the usefulness of ultrasound scanning for measuring the ES depth was recognized as a useful tool to facilitate performance of epidurals, decrease complication rate and help as a teaching tool. Nevertheless, ultrasound was not adopted by practitioners at that time for guidance of epidural needle insertions, and research in this area slowed.  Figure 1-10 Ultrasound image of the lumbar anatomy in 1984 . The top line is the lamina, bars B and C are the dura mater (images reproduced with permission from [Currie-1984])   16 In 2001, Grau [Grau-2001b] re-introduced the use of ultrasound in epidural anesthesia. Modern ultrasound machines have greatly improved image quality, which permits a better differentiation of the echo structures. Grau used ultrasound mainly to obtain prepuncture information for the epidural technique and described the key structures seen in the transverse and longitudinal views of the lumbar anatomy. The key structures (Figure 1-11) include the spinous process, articular process, LF, anterior and posterior dura mater and vertebral body. The depth to the ES is referred to as skin-to-LF depth as the LF is the key visible structure in the ultrasound image and therefore is where measurements are normally taken in the ultrasound image.  17  Figure 1-11 Ultrasound images of the lumbar anatomy. The top image is a transverse ultrasound view and the bottom image is a longitudinal view (images reproduced with permission from [Grau- 2001b])  The correlation coefficient between measured and actual needle insertion depth in [Grau- 2001b] was R2=0.79. A study of the changes in lumbar anatomy during pregnancy [Grau-2001d] was also conducted. In that paper, Grau studied the tissue alterations of pregnancy on the epidural technique by comparing the quality of ultrasound images and skin-to-LF depth and LF thickness measurements during pregnancy and after giving birth. The study shows depths to be deeper when the subject is pregnant, and the ultrasound depicted lower overall image quality of the key structures: LF, ES, dura mater and spinal canal.  18 Grau also proceeded to study the use of ultrasound for difficult epidural punctures such as on subjects with obesity, scoliosis and oedema [Grau-2001a]. He found the use of ultrasound resulted in a reduction in the number of puncture attempts, changes of puncture site and complications with a lower (better) visual analogue score (a scale of pain determined by the facial expression) and higher satisfaction of the patient. This study also showed a correlation coefficient of R2= 0.87 between the ultrasound skin-to- LF depth and the needle depth and R2=0.30 between the ultrasound measured angle and the actual needle insertion angle. This means the ultrasound measured angle and the actual needle insertion angle have little correlation and that ultimately, the actual needle insertion angle cannot be predicted precisely from the prepuncture measured angle. Grau stated the three key measurements that ultrasound can provide: puncture site, puncture angle and skin-to-LF depth. In that paper, Grau showed that the ultrasound prepuncture examination added little time to the procedure: from 4.4 min in the control group, to 5.2 min in the ultrasound group. Additionally, Grau specified three sources of error of ultrasound measured skin-to-LF depth: 1- The pressure from the epidural needle during insertion can compress anatomical structures and alter the measured distances. 2- The ultrasound measurement is measured to the top of the echo (LF) which is different from where the epidural needle tip is inserted into (ES). An added systematic error will be of the thickness of the LF. 3- The insertion angle measured during the prepuncture examination can be very different from the actual needle insertion. A deviation of the angle can cause a significant error in the distance measurements.  19 In [Grau-2001c], the three main ultrasound scanning planes were compared: 1- The median longitudinal plane: the ultrasound transducer is placed on the subject’s back with the transducer elements along the spine giving therefore a view of the spinous processes. 2- The transverse plane: the ultrasound transducer is placed on the subject’s back with the transducer elements along a line perpendicular to the spine. This is also the view of choice by some researchers [Carvalho-2008] [Arzola-2007]. 3- The longitudinal paramedian plane: the ultrasound transducer is placed on the subject’s back with the transducer elements parallel to the spine but slightly off midline (≈5mm). The view is off midline to avoid the spinous processes seen in the midline longitudinal plane and allow ultrasound transmission through the erector spinae muscles. These three planes were compared based on the visibility of the LF, dura mater, blood vessels and nerves. The shadow/window ratio was also computed by taking the width of the L3 lumbar vertebra shadow with the adjacent window in the longitudinal approaches. The paramedian plane provides the best visibility of all three planes and has a larger shadow/window ratio as shown in Table 1-1. The paramedian plane achieved the best overall visibility of all examined structures.   20 Table 1-1 Visibility in different ultrasound planes, visibility (1 = very good, 6 = insufficient) (table reproduced with permission from [Grau-2001c])   Grau also explored the use of colour Doppler imaging (Figure 1-12) to improve detection of blood vessels in the lumbar anatomy [Grau-2001e]. In that study, 4 megahertz (MHz) and 7MHz ultrasound transducers were used in B-mode imaging and in Doppler colour imaging. The images were compared on the basis of vessel visibility, pulsation perceptibility, LF, ES and dura mater visibility. The vessel visibility was better on the Doppler colour imaging mode, the 4 MHz Doppler colour imaging could depict vessels larger than 1.0 mm in diameter while the 7 MHz Doppler could depict smaller vessels as small as 0.5 mm in diameter. Most vessels large enough for accidental cannulation (catheters have diameter larger than 0.8 mm) were marginally visible on the 4 MHz Doppler colour images but clearly visible on the 7 MHz Doppler. It was concluded that Doppler imaging might help reduce accidental blood vessel catheterization in epidural anesthesia. Plane Ratio Shadow/window LF Dura Vessels Nerves Transversal  2.5±0.7 4.4±1.1 5.7±0.4 5.9±0.3 Longitudinal Median 0.34±0.14 2.3±0.7 3.6±1.1 5.5±0.7 5.7±0.5 Longitudinal paramedian 0.55±0.11 1.4±0.5 1.7±0.6 1.9±0.8 1.9±0.7  21  Figure 1-12 Ultrasound image with colour Doppler overlay to help detect veins and arteries in the lumbar anatomy (images reproduced with permission from [Grau-2001e])  Grau then used ultrasound to obtain prepuncture information on the lumbar anatomy prior to a combined spinal-epidural (CSE) [Grau-2001f]. A combined spinal-epidural is a procedure where an epidural needle is inserted in the ES, through which a spinal needle is inserted into the subarachnoid space by purposefully puncturing the dura mater, and spinal anesthesia administered. As the spinal needle is removed, an epidural catheter is threaded through the epidural needle into the ES [Cook-1999]. This procedure is typically used because spinal anesthesia has a faster onset time, and the epidural catheter is used as a safety measure in case the surgery takes longer than expected, and for post-operative pain relief. It was shown that ultrasound can help reduce the number of needle insertion attempts (in the ultrasound group: 75%, 22% and 3% were successful in the first, second and third attempts respectively; in the control group: 20%, 60% and 20% were successful respectively). In other words, with ultrasound technique, the rate of success is already  22 75% at the first attempt and after the second attempt, the cumulative rate of success is 97%, so that on only 3% of cases, a third attempt is necessary. With the traditional technique, the rate of success is only 20% at the first attempt and attains 80% after a second attempt. 20% of time, a third attempt is needed with the traditional method. The ultrasound technique reduces the uncertainty of the needle insertion depth to ±2.55 mm according to the ultrasound-measured skin-to-LF depth. The insertion depth range is normally 20-90 mm, so using ultrasound can reduce the uncertainty by one order of magnitude.  In 2002, Grau used ultrasound to observe a blood patch being formed [Grau-2002a]. The blood patch is a procedure commonly performed when a dural puncture occurs. The patient’s blood is injected in the ES at the location of the puncture with the expectation that blood clotting will seal the dura mater.   The needle was directed to the site of the puncture with real-time ultrasound guide (Figure 1-13), and LOR was used to confirm entry into the ES. Then, 20 milliliter (mL) of blood from a brachial vein was slowly injected into the ES. Dural fiber reorganization were observed to happen after about 10±14 seconds (s). Headache symptoms were alleviated within a few minutes after the injections (visual analogue score after 15 minutes at 0±1) in all patients.  23  Figure 1-13 Blood patch observed with real-time ultrasound (left) punctured dura, (right) patched dura images (reproduced with permission from [Grau-2002a])  In 2002, Grau conducted a larger study on the use of ultrasound for obstetric anesthesia [Grau-2002b]. That study had 300 subjects. The ultrasound depth measurement had a correlation coefficient of R2=0.83 with the actual needle depth. The ultrasound scans added approximately 75 s to the needle insertion procedure. The precision was 6.96 mm, which means the search region is effectively 10 times smaller than when using the traditional “blind” LOR technique (20 mm to 90 mm). Similar to previous studies, fewer complications and side effects, lower visual analogue score, and fewer puncture attempts were recorded when using ultrasound.   In 2004, Grau used real-time ultrasound to guide CSE [Grau-2004]. The subjects were separated into 3 groups: control group using the traditional “blind” method, prepuncture ultrasound group and a real-time ultrasound guidance group. Each group performed 10 procedures. Real-time guidance requires sterile ultrasound to be used, so sterile sleeves and ultrasound gel were used. An assistant was also used to hold the ultrasound transducer. The needle was inserted using a free-hand ultrasound method. The epidural needles were seen in the ultrasound images but the catheter could not be observed (Figure  24 1-14). The ES was identified in all cases. The number of attempts is shown in Table 1-1. Real-time ultrasound was shown to be an improvement over the prepuncture information because the anticipated ideal path is rarely achieved, due to tissue deformation during the needle insertion. Both ultrasound techniques reduce the number of attempts when compared to the control group without ultrasound. However, the feasibility of this real- time ultrasound-guided epidural needle insertion is limited because of the need of an assistant proficient in the lumbar sonoanatomy. Table 1-2 Success of ultrasound controlled epidural anesthesia (table reproduced with permission from [Grau-2004])    25  Figure 1-14 Diagram of real-time observation of epidural needle being inserted for a CSE (reproduced with permission from [Grau-2004])  Another study for real-time ultrasound for epidural needle insertion was performed in 2006 [Willschke-2006] on pediatric subjects. The study was conducted on 64 children, 0- 6 years of age requiring epidurals in the lumbar or thoracic levels. Results were presented in two groups, children 0-6months and 6months to 6 years old and categorized by the ultrasound and LOR techniques. For the ultrasound group, the LOR was not used and only ultrasound was used to guide the needle into the ES. It was shown that in children older than 6 months, the time to perform an epidural decreased from 286 s to 142 s when  26 using ultrasound and the number of bone contacts was reduced from 89 to 9 (Table 1-3). Moreover, the catheter was seen in ultrasound images of children. Since the children in this study had a needle insertion depth of about 13.5 mm, which is far below the 20-90 mm range in adults, it is not clear how the results would translate to adult epidurals. Finally, this was still a two-person technique with an ultrasound assistant/operator (Figure 1-15). A single-operator technique would be more practical and cost efficient.  Figure 1-15 Real-time ultrasound-guided epidural needle insertion performed on a child with a two person method (reproduced with permission from [Willschke-2006]).  27 Table 1-3 Results for ultrasound-guided epidurals for the children 6months and older group. Using ultrasound reduced the time to perform the epidural and the number of bone contacts. P-value denotes the statistical strength of a comparison, a low P-value is more statistically significant. (reproduced with permission from [Willschke-2006])   In 2009, Karmakar devised a method for performing real-time ultrasound-guided paramedian epidurals using only one operator [Karmakar-2009]. The LOR method was performed using an Episure™ Autodetect LOR syringe instead of the traditional air or saline loaded syringe. The Episure needle contains an internal compression spring that applies constant pressure. This eliminates the need for a hand to apply pressure on the syringe and perform the LOR, making the procedure easier to perform with one operator (Figure 1-16). The method was used on 15 patients of body mass index (BMI) < 35.This  28 technique is a freehand technique and requires some expertise to align the ultrasound imaging plane to the needle trajectory. Karmakar stated that success depends on the quality of ultrasound images.  Figure 1-16 Ultrasound image of a single-operator real-time in-plane ultrasound guided epidural needle insertion (reproduced with permission from [Karmakar-2009]).  So far, ultrasound was mostly performed in the paramedian plane, stated by Grau to be the optimal plane for imaging, because the image quality and depiction of the important features is best viewed in the paramedian plane, avoiding the adverse effects from the spinous processes. Carvalho re-evaluated the transverse approach in the perspective of midline epidural needle insertion [Arzola-2007] [Carvalho-2008]. The premise is that the preferred needle insertion path for obstetric lumbar epidurals is the midline approach and that the paramedian ultrasound plane does not image the anticipated midline needle trajectory. It is argued that this makes the transverse approach more suitable than the  29 paramedian approach for imaging because the anticipated needle trajectory is seen in the transverse plane. Figure 1-17 shows the ideal transverse ultrasound image of the lumbar anatomy for epidurals. Carvalho describes the image as a “bat”, in which the articular processes are the ears, the transverse processes are the wings of the bat, and the anterior and posterior dura form the head of the bat. Another advantage of the transverse view is that one can obtain the puncture site, the puncture angle (equivalent to the angle the transducer forms the image) and the skin-to-ES depth in one single image. In this work, the skin-to-ES depth is measured from the skin to the bottom of the echo, which is closer to where the ES actually is, so it is expected to be a better approximation of the needle insertion depth. Moreover, the depth is measured on an image acquired along the needle trajectory so the approximation should be even closer. The R2=0.881 and Bland-Altman 95% limits of agreement of -6.66 mm to 6.87 mm partly confirm these hypotheses. Visibility of the key anatomical structures continues to be a critical aspect of ultrasound guided epidurals.    30  Figure 1-17 The “bat” representation of the lumbar vertebral structures in an ultrasound image in the transverse plane (reproduced with permission from [Carvalho-2008]).  Most manufacturers offer ultrasound transducers designed to be coupled with a needle, but few are designed for epidural guidance. One exception is the split array transducer  31 designed by Dr Malcolm Watson [Watson-2007] shown in Figure 1-18. The novel prototype transducer is shaped like a computer mouse and produces images at 90° from the patient’s skin surface with a hole in the transducer array for the needle. This ensures the needle path is in the ultrasound plane. The transducer can be held stable on the patient’s back due to the large surface of contact. Although it provides convenience for aligning the needle with the imaging plane, the combination of a fixed needle guide and large flat transducer face means the needle angle can only be perpendicular to the skin surface. This reduces the range of trajectories available to the anesthesiologist. In particular, it cannot provide a midline needle insertion with real-time paramedian imaging (which produces the best image of the epidural space) of the target and needle insertion. needle Ultrasound beam Ultrasound array cable  Figure 1-18 Dr Malcolm Watson’s split array transducer. The transducer provides an image at 90° to the patient’s skin, and a hole designed for needle insertion acts as a needle guide to ensure the needle is in plane with the ultrasound image.    32 Ultrasound image enhancement  There are several methods which improve aspects of ultrasound images. In [Cobbold- 2007], a review of some methods is presented, ranging from harmonic imaging, to varying the aperture of the ultrasound beam, to speckle reduction methods. A small selection of ultrasound enhancement methods suitable for epidural imaging is presented next.  As seen in previous studies, ultrasound of the lumbar region produces an image filled with speckle and artifacts which can impede detection of important features such as the LF. Ultrasound uses the pulse-echo technique to generate images. The recorded echo is based on reflections from large-scale (relative to wavelength) structures, such as bone (i.e. specular reflection) and reflections from small-scale structures, such as cells (i.e. random scattering). If the specular reflection is strong enough, it casts shadows in the beam direction. Random scattering creates speckle from constructive and destructive interference [Christensen-1988]. Although the texture of the speckle can be related to tissue type, in general it is considered as a noise present throughout the image. Other artifacts such as reverberations and refraction also affect image quality. Variable image quality is one of the factors why ultrasound-guided epidurals have not progressed past the research stage to become the standard of care for epidurals. As mentioned, there are many potential methods for improving aspects of image quality [Cobbold-2007], some of which may be suitable for epidural imaging which requires both high spatial resolution and contrast in order to depict the LF. Moreover, the LF may only appear clearly at certain  33 angles of insonation. Generally, an ultrasound beam that is perpendicular to an interface will yield the most reflected energy received by the ultrasound transducer, producing the brightest echo compared to other beams. Both speckle reduction methods and various multi-beam imaging/compounding methods are briefly reviewed.  Many post-processing methods employ filters to reduce speckle (e.g. homomorphic Wiener filtering [Achim-2001], modified homomorphic Wiener filtering [Michailovich- 2006], adaptive weighted median filter [Loupas-1989] and anisotropic diffusion filter [Perona-1990]) as shown in Figure 1-19, but all suffer to some degree from loss of fine details through the removal of content with high spatial frequency.  34  Figure 1-19 Some post-processing techniques on ultrasound images. Homomorphic wavelet despeckling (HWDS) is based on a multiplicative model of noise and uses a logarithmic transformation to convert the multiplicative noise into additive noise, followed by a wavelet denoising. The total variation despeckling (TVDS)[Rudin-1992] is known for filtering out the noise without blurring out  edges. The TVDS can be implemented as a signal-dependent filter. The anisotropic diffusion despeckling (ADDS) [Perona-1990] uses an anisotropic diffusion filter at the denoising stage. It uses the locality and anisotropy of differential equations (reproduced with permission from [Michailovich-2006]) ( © 2006 IEEE)  Another category of speckle reduction includes compounding techniques. Compounding involves acquiring several images with varying levels of decorrelated speckle pattern, and  35 then averaging these images to create a compounded image with an increase in signal-to- noise ratio (SNR). There are three main compounding methods. Temporal compounding averages several images of the same anatomical region over a period of time and relies on small tissue movement or transducer motion to obtain images with different speckle patterns. Frequency compounding varies the transmitted frequencies while acquiring several images of the same region, so the received images show different speckle patterns [Amir-1986] [Silverstein-1988]. However, frequency compounding tends to reduce the image spatial resolution. The speckle reduction performance of temporal and frequency compounding is modest especially when the change in speckle pattern is small, in other words, when the images are highly correlated. The third method, spatial compounding, uses beam-steering [Berson-1981] [Carpenter-1980] [Jespersen-1998] [Jespersen-2000] [Huber-2002] [Anderson-1997] to acquire several frames of the same anatomy at different beam angles, see Figure 1-20.  In spatial compounding, the speckle pattern is expected to be different for each frame as the diagonal distances between the small-scale reflectors are different. Averaging those images reduces speckle noise, improves the SNR and improves boundary continuity of features, given decorrelated noise patterns. The level of correlation depends on the beam angle, with larger angles giving lower speckle correlation. Many modern ultrasound machines have all three compounding methods as optional features. Of these three methods, spatial compounding appears to be the most suitable for epidural imaging. This is because spatial compounding acquires several images of the same anatomy but taken from different beam angles, thus improving the probability of clearly depicting the LF and dura mater interfaces.   36  Figure 1-20 Example of beam-steering to create 3 images of the same region of interest but at different beam angles. The three images are combined into a single compounded image.  Registration  Compounding methods still need registration as capturing several frames requires time and there are small geometric image distortions (such as variations in speed of sound through different tissues) which can be as large as 14% [Christensen-1988]. Distortion amoung the images results in small misalignment of the features which, in turn, blurs these features in the compounded image. Registration of the images to align the features would be beneficial [Groves-2004].  Image registration is the process of transforming an image from one coordinate system to another. For instance, registration can be performed on an ultrasound image of a patient in order to compare it to a magnetic resonance image of the same patient, or it can be used to compare two images taken at different moments in time in order to see the  37 progression of a tumor. For spatial compounding, registration can be performed using the known beam angles, but would be improved with an estimate of the distortion.  Image registration is a widely researched topic. Good reviews are provided by [Maintz- 1998] and [Zitova-2003]. Registration methods can be roughly classified into two categories: rigid/affine transformations and non-rigid transformations.  Affine transformation registration methods can be described by a linear transformation matrix. The image can be translated, rotated, sheared and stretched. Rigid transformations are a sub-category where only rotations and translations are allowed. Non-rigid transformations allow a potentially large number of degrees of freedom to “warp” an image.  Re-alignment of ultrasound features using non-rigid (elastic) registration was studied by several groups [Groves-2004] [Ruekert-1999] [Thirion-1998] [Bro-Nielsen-1996] [Christensen-1996]. Groves provides a typical example where registration was used to align beam-steered images [Groves-2004]. The method, called “warping”, uses a block matching method followed by a smooth interpolation of the warping vectors using radial- basis function interpolation. A vector field with a unique vector for each pixel was then produced and used for realigning the beam-steered images. The result was an ultrasound image with increased edge sharpness when compared with spatial compounding without registration, but with significant additional computational cost. Given the variety of ways  38 to perform registration and compounding, the question is what aspect of the image should be optimized for epidural ultrasound?  Bone detection and LF detection  Detection of the LF in the ultrasound image is the key aspect of image interpretation for epidurals, but remains a challenge. No previous work has been done on optimizing the appearance of the LF in ultrasound. The detection of bones is closely related to the detection of the LF because both create largely specular reflections due to a large mismatch of acoustic impedance with surrounding tissues. There has been some work on automatic bone detection in ultrasound images. The shadow cast by a bone has previously been used to help find the location of the bone using image fusion [Daanen-2004], adaptive thresholding [Filho-2006](Figure 1-21) and by using dynamic programming [Foroughi-2007].   39  Figure 1-21 Bone segmentation using shadows (reproduced with permission from [Filho-2006])  Phase techniques have also been used to detect bones [Hacihaliloglu-2006] [Hacihaliloglu-2009] (Figure 1-22). It is shown that phase information contains more significant information in an image than amplitude information [Oppenheim-1981]. Moreover, phase techniques have been shown to perform better than gradient-based techniques because local phase is invariant to image brightness, and therefore should be suitable for ultrasound images [Mellor-2004]. The principle of phase congruency is that many of their frequency components have the same phase at the location of an edgelike feature. Phase symmetry is another phase method which decomposes the signal into even and odd parts; the location of a ridge is where the even part is strongest compared to the odd part. This can be used to detect bone features as they are often seen as strong ridges followed by a shadow. It is possible that phase-based techniques can also be used to highlight the LF.  40    a)  b) Figure 1-22 Bone segmentation using phase symmetry a) ultrasound image b) phase symmetry image. In this work, phase symmetry is compared to the gradient image and Canny edge detector. Phase symmetry is shown to produce clear images with detected features matching bone surfaces. (reproduced with permission from [Hacihaliloglu-2009])  Summary of limitations of previous studies on ultrasound-guided epidurals  In most studies, the skin-to-ES depth is compared to the needle insertion depth using LOR to define the endpoint for insertion. Although the success of anesthesia can be used to support the belief that the needle tip is exactly in the ES, some uncertainty remains because LOR is still a subjective measure; it can occur at locations outside of the ES  41 [Sharrock-1979] and the success of anesthesia is an indirect measure. Instrumentation of the LOR may help to quantify this endpoint, so a study of both ultrasound and instrumentation is warranted. Also, most previous research has been on measurement of the skin-to-ES depth, but little work has been done to correctly identify the intervertebral level and to choose the puncture site. Furthermore, little work has been done on real-time guidance of needle insertions as opposed to pre-puncture imaging. Performing real-time guidance reliably with a single inexperienced operator is the ultimate goal. Achieving this likely involves the creation of high quality ultrasound images specifically for epidurals and providing automated tools to assist image interpretation and measurements. Finally, little work has been done on finding surrogate measures for skin-to-ES depth when the LF is not visible. We propose to do such studies.  In summary, ultrasound imaging is a suitable choice for epidural guidance because it is noninvasive, harmless at low power, portable, accurate and inexpensive compared to other imaging modalities. Nevertheless, ultrasound has not been widely accepted for epidurals, despite the number of recent papers showing its potential benefits, for the reasons listed above  The ideal epidural needle insertion guidance system should have the following: 1- real-time feedback 2- single-person operation 3- low cost 4- no radiation  42 5- clearly visible epidural needle during the course of insertion 6- clearly visible lamina and ligamentum flavum 7- localization of the desired intervertebral level and puncture site 8- guidance for choosing proper needle insertion angle 9- incorporation of the traditional LOR technique for redundancy and ease-of- acceptance.  1.3 Thesis objectives The purpose of this thesis is to establish the overall hypothesis that that new instrumentation techniques can quantify the subjective LOR technique, and new ultrasound imaging techniques can detect and depict the anatomy of the ES and surrounding tissue more accurately and reliably than conventional ultrasound for the purpose of guiding needle insertion . The following objectives are set for the preparation of this thesis: - quantify the LOR technique in porcine subjects and in human subjects - use ultrasound to measure the skin-to-ES depth and identify surrogate measures for that depth - use real-time ultrasound to guide an epidural needle to the target with a single operator - improve the quality of ultrasound images of the lumbar anatomy - automatically locate the LF in the ultrasound images of the lumbar anatomy    43  These contributions were made toward reaching the objectives: Chapter 2: - measurements and analysis of force, displacement and pressure of epidural needle insertion in porcine subjects ex-vivo to compare paramedian and midline needle insertion approaches - development of a set of sensors compatible with sterile conditions and subsequent measurements and analysis of force and displacement on epidural needle insertion in human subjects Chapter 3: - development of a protocol to count the intervertebral spaces with ultrasound using the 12th rib and the sacrum as references - comparison of the actual needle insertion depth to ultrasound measured skin-to- LF depth using instrumented LOR as the endpoint - comparison of the actual needle insertion depth to surrogate measures (skin-to- transverse process tip, fat thickness and subject biometrics) to be used when the LF is not clearly seen Chapter 4: - using a needle guide and calibration to perform a single-operator aim-and-insert in-plane real-time 2D ultrasound-guided epidural needle insertion - calculation of some geometrical limitations of the 2D ultrasound-guided approach and introduction of an opportunity for 3D ultrasound   44  Chapter 5: - calculation of expected ultrasound image distortion due to speed of sound errors as a function of the fat layer and beam-steering angle with both linear and curvilinear ultrasound transducers - building of a novel agar gelatin phantom with fat and muscle mimicking layers to find a set of suitable parameters for non-rigid registration for spatial compounding of beam-steered images - design of an algorithm to speed up the non-rigid registration of beam-steered images - development of tests to compare various spatial compounding techniques Chapter 6: - design of a template matching strategy for automatic lamina and LF detection - development of a confidence measure for the LF detection step based on the normalized crosscorrelation ratio - comparing automated LF detection versus manual LF detection, sonographer measurements, and actual epidural needle depth on human subjects  1.4 Chapter summary The overall goal of this thesis is to introduce several new tools and techniques to help anesthesiologists perform the epidural needle insertion procedure.   45 In Chapter 2, a set of force, pressure and position sensors are used to measure the force applied on the plunger of the syringe, pressure at the tip, and flow rate as the anesthesiologist performs the LOR. The sensors are used to quantify the LOR in porcine subjects in the midline approach and the paramedian approach, and in humans in the midline approach, and show the differences in what the anesthesiologist can “feel” when the needle tip traverses each tissue.  In Chapter 3, ultrasound is used to count the intervertebral levels starting at the 12th rib to locate the interspace between the second lumbar vertebra and third lumbar vertebra (L2- 3) and L3-4. The positions of the L2-3 and L3-4 intervertebral spaces measured by ultrasound are compared with the positions located by traditional palpations and bony landmarks. Ultrasound is then used to measure the paramedian skin-to-LF depths, and surrogate measures are compared with the actual midline epidural needle insertion depths.  In Chapter 4, ultrasound is further used to guide lumbar epidural needle insertions in real- time with a needle guide defining the angle and position between the ultrasound transducer and the needle. In this way, a single operator can perform both ultrasound and needle insertion. The geometrical limitations are studied and the need for a transducer dedicated for epidural needle insertion is defined. Tests are conducted on human subjects in a clinical trial. Pre-puncture measurements are compared to the actual needle insertion depths.   46  In Chapter 5, adaptive median-based spatial compounding is used to improve the image quality of speckle noise while keeping the features sharp. The expected speed of sound errors are derived from beam-steering angle, fat thickness, speed of sound, and pixel position in the image. Beam-steered images are registered through warping using speed- of-sound calculations to apply a bound to the amount of warping, and combined using a median-based compounding algorithm. Moreover, linear prediction is used to speed up the warping stage of the algorithm. This algorithm is tested on images of an agar gelatine spine phantom to determine a suitable set of parameters. A clinical study on 20 human subjects is performed to test the performance.  In Chapter 6, an algorithm for automatically localizing the LF and lamina from the ultrasound image is presented. The algorithm starts by applying adaptive spatial compounding to an image, extracts a ridge map using phase symmetry, then a template matching method based on actual lamina and LF parameters is used to extract the position of the LF from the ultrasound image with a confidence factor. This is tested on 20 human subjects, with the automatic LF detection compared with the manual segmentation, the sonographer measurements, and the actual needle insertion depth.  In Chapter 7, the thesis is concluded by a summary. The contributions of the work in this thesis are put in perspective of the field of instrumentation and ultrasound of epidural needle insertion. Finally, some future work directions are presented.   47 2 Instrumentation of epidural needle insertion1 2.1 Introduction   As mentioned in Chapter 1, the needle insertion is considered a blind procedure. The standard method of confirming entry into the ES is through the “feel” of LOR [Miller- 2005]. For a midline insertion, the epidural needle is partly inserted in the interspinous ligament and a saline-filled syringe is attached to the needle. Force is applied at the plunger to get a “feel” of the resistance to saline injection by the tissue at the needle tip. This haptic feedback (“feel”) helps the physician determine the location of the needle tip within the different tissues. The needle is further inserted until the tip reaches the LF, and a high resistance to injection is felt. As the needle tip traverses the LF, it enters the ES, a fluid-filled cavity, where saline is injected with no resistance, hence the LOR. The LOR is therefore an inferred phenomenon for detecting the ES [Wilson-2007]. There are several variations in the way the LOR technique can be performed. The technique normally consists of the epidural needle being inserted in the midline, and traversing the interspinous ligament, LF and into the ES. Another variation uses a paramedian approach [Miller-2005] in which the needle is inserted laterally, about 2 cm to the midline, and at an angle of 10°-25° [Miller-2005] into muscle and then into the LF and the ES. Both techniques are shown in Figure 2-1. It is impractical to provide substantial training on animals or cadavers, and current simulators only capture a portion of the full realism of the actual tissue resistance [Magill-2004] [Dang-2001]. Practice on human subjects remains the standard and  1 The material in this chapter has been published in [Tran-2009b] ( © 2009 IEEE)  48 proficiency is gained slowly (e.g. about 50 procedures to reach a success rate of 90% [Kopacz-1996]). Instrumentation of the technique of LOR would help gain a better understanding of the variations in feel, including subtle changes in resistance that experienced physicians report. Instrumentation would also provide quantitative differences in technique. Finally, instrumentation would also provide more information for the next generation of haptic epidural simulators which can help in learning as the “feel” is only learned through experience.   Figure 2-1 A comparison of a transversal cross-sections of a) human subject (image reconstructed from the Visible Human Project, The National Library of Medicine) (reproduced with permission from [Ackerman-1998]) ( © 1998 IEEE) , and b) porcine subject, this photographic image was created by slicing a frozen porcine cadaver in the transverse plane.   49 The pressure in the ES has been measured in the past [Vas-2001]. Vas et al. studied the pressures for infants and measured a pressure of about 9.2 kilopascal (kPa) when the needle tip is in the LF and 0.13 kPa when the needle tip reaches the ES and the LOR is felt. The setup to measure pressure in [Vas-2001] is shown in Figure 2-2. In [Rodiera- 1995], a pressure of 67.6±12.2 kPa when the needle tip is in the LF for adults has been reported. The pressure throughout the epidural needle insertion as well as LOR has been documented. In [Lechner-2003], an infusion pump is used to provide continuous pressure of saline injection. The pressure is measured throughout the epidural needle insertion and converted to an acoustic signal. With the pressure associated to the pitch of the acoustic signal, and the infusion pump set to a desired flow of 100 mL/h, tissues of high resistance would cause a high pitch signal to be audible, while low resistance regions such as the ES would cause a low pitch signal to be audible. A drop in pressure at the LOR can therefore be clearly detected audibly. The infusion pump adds cart and complexity and has not been widely adopted. Moreover, the detection of LOR is still a subjective task as the sound pitch would decrease but there is no quantitative analysis.  50  Figure 2-2 A 3-way stopcock is used to connect the syringe, pressure sensor and epidural needle (reproduced with permission from [Vas-2001]).   In this chapter, the LOR to saline is instrumented with sensors on a standard syringe and plunger. In particular, the paramedian approach was compared to the midline approach to quantify the differences between the “feel” of each approach. This is in anticipation of the need for a paramedian needle insertion approach in real-time 2D ultrasound guided epidural needle insertion described in Chapter 4. Finally, the force and displacement sensors were used in a clinical setting to quantify the tactile feedback or “feel” of LOR in human subjects in vivo. The goal is to provide a better understanding of flow measurements in different tissue and better understanding of differences between porcine subjects ex vivo and human subjects in vivo since work in following chapters will be done on both sets of subjects. These results may also be helpful for possible design of haptic simulators.  51 2.2 Methods  The first experiments were performed in a laboratory environment on excised porcine tissue (sus scrofa domestica). The porcine lumbar spinal anatomy is similar to the human anatomy with the main difference being that the porcine lumbar region generally contains 6 vertebrae rather than 5 in humans. Additionally, the porcine subjects were slaughtered one or two days prior to the tests, causing some dehydration and increased stiffness in the tissue. The porcine tissue was prepared and fixed in a vertical orientation to mimic a seated position of human subjects as shown in Figure 2-3b. The excised portion of tissue contained the entire spine, muscle and ligaments. The advantage of using porcine tissue is that multiple needle insertions can be performed for repeated measurements on the same subject. The porcine tissue was obtained through a certified butcher following guidelines and notification of the UBC Animal Care and Biosafety Committee. Three sensors were used to quantify the LOR: a force sensor, a pressure sensor and a displacement sensor, as shown in Figure 2-3a. The SLB-25 force sensor (Transducer Techniques, Temecula, CA) was mounted on a custom-built stainless steel harness and fitted to the physician’s thumb to measure the force applied to the plunger of the syringe. The CSPR IP65 magnetostrictive displacement sensor (MTS System, Cary, NC) was attached to the body of the syringe and a ring magnet was used to track the position of the plunger relative to the syringe. The change in displacement measurements were converted into the change of volume of saline remaining in the syringe by multiplying the displacement by the cross-sectional area of the syringe (121.9 mm2). The ring magnet did not touch the rod and therefore the additional friction was negligible. The PX302 pressure  52 sensor (Omega Engineering, Stamford, CT) was connected to a three-way stop-cock between the syringe and the epidural needle so that the pressure measurements were approximately equal to the pressure at the tip of the needle. The three sensors were connected to a computer workstation and data were acquired using a Q8 data acquisition board (Quansar, Markham, ON) at a sampling period of 0.01s. From the manufacturers’ specifications, the position sensor accuracy is 0.2 mm (when mounted to the syringe, the saline volume accuracy is then 24.5 mm3), the force sensor accuracy is 0.18 Newton (N) (0.6% of the maximum force of around 30 N) and the pressure sensor accuracy is 0.25 kPa (0.25% of the maximum pressure of around 100 kPa). Glass syringes (JH-0550 Epidural Catheterization Kit, Arrow International, Reading, PA) were used as they provide minimal friction when properly wetted and are commonly used.   a) b) Figure 2-3 The force sensor, pressure sensor, and displacement sensor mounted to the syringe.   53  It was shown in [Hor-2007a] that these sensors could be used to accurately detect a LOR. In those experiments, the sensors were used to measure the force, position and pressure of the “feel” of the LOR. The anesthesiologist verbally communicated the success of the needle insertion and the time stamp was recorded. This time stamp was then compared to the time LOR was observed on the sensor signals. It was found that the average time difference between the time indicated by the sensors and the anesthesiologist is 0.8±0.3 s (the anesthesiologist confirmed it later than the sensors showed). Moreover, the sensors were used in a midline epidural needle insertion in porcine subjects. The experiment showed that the “feel” was different depending on whether the epidural needle was in the interspinous ligament or in the LF.   For all epidural needle insertions, the LOR is used. The epidural needle is inserted in the spinal tissues while a syringe filled with saline and an air bubble at the base is attached to the needle. Continuous pressure is applied to the plunger and tactile feedback is felt at the thumb. The measured pressure is expected to be lower before reaching the LF, higher at the LF and dropping abruptly at the ES. These two experiments are shown in more detail in Appendix C.   54 2.2.1 Experiment 1: comparison of the midline approach and the paramedian approach on porcine tissue As mentioned, two approaches are used by physicians for performing a needle insertion: midline and paramedian [Muranaka-2001]. The two approaches were compared by performing needle insertions on excised porcine tissue using a similar protocol to [Hor- 2007a] on 10 midline insertions and 20 paramedian insertions (n=10 for midline, n=20 for paramedian). The “feel” of LOR is inherently different in the two techniques because, in the midline case, the epidural needle first traverses interspinous ligament and in the paramedian case, the needle first traverses muscle. This knowledge of the anatomy allowed the measurements to be divided into portions and, each portion associated with the interspinous ligament/muscle (midline/paramedian), LF and ES at the LOR. Data from the same interspace but using paramedian and midline approaches are compared using a paired t-test. A p-value smaller than 0.05 means there is a 5% likelihood that the hypothesis of the samples being differnt is wrong and so is considered as significantly different.  2.2.2 Experiment 2: instrumentation of human subjects The 11 human subjects were recruited (n = 11) using signed consent. This study was approved by the clinical review ethics boards of both the University of British Columbia and the British Columbia Women’s Hospital (C05-0409). Subjects were women in labour prior to vaginal delivery, or women scheduled for cesarean section. A standard catheter, epidural needle and syringe were supplied in a sterile package (FlexTip Plus Catheter, model JH-05500, Arrow International Inc., Reading, PA). The  55 force sensor was worn on the physician’s thumb and a sterile glove was worn on top of it to ensure sterility. Since the position sensor contains sensitive electronics not suitable for high temperature sterilization, the position sensor was covered by a general purpose ultrasound cover (Cone Instruments, Solon, Ohio). The sterile position sensor apparatus was connected to the syringe according to the sterility procedure described below. Since only one needle insertion can be performed per patient, the midline approach is chosen because it is the standard of practice in the hospital where the experiments were conducted. The patients were placed in the seated position. A 17-gauge epidural needle was inserted and the continuous pressure technique was used while measuring force and displacement.  For all tests, the physician gave a verbal indication of which tissue the needle tip was believed to be traversing. The times of each verbal indication was used to help segment the graphs into the following regions: interspinous ligament, LF, and the ES where LOR occurred.  The paired t-test is used to obtain p-values for assessing statistical significance of the difference between the feel in the different tissues encountered by the needle in the midline approach.      56 Sterilization Because the sensors were to be used in clinical trials on human subjects during the epidural needle insertion, sterility is critical for percutaneous procedures so a protocol for assembly of the sensorized needle was developed in coordination with the hospital’s sterilization unit. Each of the three sensors needs to be carefully considered. None of the sensors can be sterilized because of the high temperature and pressure possible effect on the electronics. Because of the 3 hour sterilization processing time, duplicate sets of components (shown in Figure 2-4) were used to ensure availability.  Figure 2-4 The set of metal tools to be sterilized after every procedure: stainless steel hex key, screws, sterilizable magnet, small aluminum clamp (bottom) to hold the magnet and the syringe plunger, large aluminum clamp (top) to hold the position sensor body to the syringe, and a stainless steel casing for the magnetic position rod.   The thumb force sensor is to be worn on the anesthesiologist’s thumb to measure the force applied on the syringe plunger. To maintain sterility, the anesthesiologist routinely  57 wears sterile surgical gloves. For this experiment, the force sensor is to be worn under the glove at all times. The pressure sensor is required to measure the pressure at the tip of the epidural needle. Since saline actually needs to be in contact with the pressure sensor for the measurements to be made, the sensor needs to be sterilized or omitted. Therefore, the pressure has been modeled from force and displacement using a decay model [Hor-2007a] allowing the omission of the pressure sensor in the clinical trials on human subjects (to be described in the next section). The decay model is given by        0)( 0)( )( dt dDe A tFk dt dD A tFk tP itt aa aa         eq 2.1 where ti is the time at which the plunger stops moving and τ is a decay time constant determined empirically on bench-top tests with a closed needle to be 23±8 s. The sensors were finally used to compare the intermittent and continuous pressure on the syringe plunger techniques in order to determine which one was more compatible with the pressure model expressed by the position and force sensors. The continuous pressure technique matched the model better, so it is selected for our work.  The position sensor is directly attached to the syringe, and the anesthesiologist will be using this syringe directly for the LOR. Although the sensor itself cannot be sterilized, the clamps, casing and screws connecting the position sensor and the syringe can be sterilized. The body of the sensor was covered by a general purpose sterile non-latex ultrasound transrectal probe cover (Cone Instruments, Solon, Ohio). Because of the short time available in the operation room, the covered sensor and clamps were assembled  58 approximately 30 minutes prior to the procedure. A redesigned sterilizable casing (Figure 2-5) was custom fabricated for easy assembly using a locking mechanism rather than a set of screws. The covered position sensor was then connected to the syringe by a set of aluminum clamps. The clamps, screws and tools were sterilized after every procedure using steam at high temperature (132 °C) and high pressure (186 kPa) for 4 minutes and wrapped and stored until assembled.  Figure 2-5 Redesigned casing for easy assembly stage. The casing for the rod locks to the position sensor.  2.3 Results 2.3.1 Experiment 1: comparison of the midline approach and the paramedian approach on porcine tissue Table 2-1 shows sensor data for the 3 regions of interest: before LF, at LF and in the ES (i.e., when LOR is felt). Before the LF, in the midline approach, the needle tip is in the interspinous ligament and the flow rate (131±86 mm3/s) is smaller (paired t-test p<0.175) than the flow rate in the paramedian approach where the needle tip is in the muscle (172±133 mm3/s). The force and pressures are not significantly different for the two approaches at this point of insertion so the differences are again due mainly to the tissue  59 properties, not the operator. When the needle tip is on the LF, the flow rate in the paramedian plane is 147±141 mm3/s, which is again significantly larger (paired t-test p<0.02) than the flow rate in the midline of 92±48 mm3/s so the differences arise from the tissue. This is an unexpected result since the tissue type is the same for both approaches. Although the force is slightly higher paramedian than midline, the pressure is not significantly different. Finally, when the needle tip is in the ES and the LOR to saline is felt, the flow rates are not significantly different (1077±530 mm3/s in the midline compared to 1064±723 mm3/s in the paramedian approach). Again, the force is not significantly different for the two approaches although the pressure is slightly higher paramedian than midline. Figure 2-6 shows typical force, pressure and displacement measurements obtained when performing midline and paramedian epidural needle insertion in the same subject. P-value denotes statistical significance of a comparison, where a low P-value means more statistical significance. Table 2-1 Experiment 1: flow rate, force (Fa) and pressure (P) for  the interspinous ligament, muscle, LF, and epidural space (LOR endpoint) for porcine subjects using the midline and paramedian approaches. Region Flow rate (mm3/s) Fa (N) Max Fa (N) P (kPa) Max P (kPa) Midline: interspinous ligament 131±86 8.9±5.3 14.7±6.6 31.3±12.8 59.4±24.1 Paramedian: muscle 172±133 9.4±5.4 13.7±7.0 34.0±17.4 56.2±31.3 p-value <0.175 >0.25 >0.25 >0.25 >0.25 Midline: LF 92 ± 48 8.9±5.3 14.7±6.6 31.3±12.8 59.4±24.1 Paramedian: LF 147 ± 141 12.9±5.3 16.1±5.3 50.4±25.8 65.6±33.5 p-value <0.02 <0.075 <0.175 >0.25 >0.25 Midline: LOR 1077 ± 530 11.4±8.3 18.2±9.6 40.1±32.7 61.2±40.0 Paramedian: LOR 1064 ± 723 12.9±5.1 17.4±5.5 57.9±28.1 82.0±34.7 p-value >0.25 >0.25 >0.25 <0.125 <0.125  60    Figure 2-6 A sample graph of force (blue), pressure (red) and displacement (green) measurements using continuous pressure in the porcine subject for a) midline approach, and b) paramedian approach. Note the several bone contacts in the paramedian approach that occur occasionally when performing the needle insertion in the paramedian plane. 2.3.2 Experiment 2: instrumentation of human subjects In the clinical trial on human subjects, the measurements in the interspinous ligament can be compared to the LF. The subject biometrics were as follows: average age of 33.8 ± 4.6 years, weight of 73.6 ±16.5 kg, height of 162.6 ± 8.2 cm and skin-to-epidural depth of 51.9 ± 11.8 mm. Sample measurements are shown in Figure 2-7. Table 2-2 shows the average flow rate and force applied. The flow rate at the interspinous ligament is 60±30 mm3/s which is significantly larger than the flow rate in the LF at 12±13 mm3/s. The  61 average force is also significantly larger in the LF (5.0±3.0 N) than in the interspinous ligament (2.0±1.4 N) despite the lower flow. The maximum force applied is 6.0±3.0 N in the LF and is also significantly larger than in the interspinous ligament at 4.6±1.3 N. Since the continuous pressure technique rather than intermittent pressure technique [Hor- 2007a] is used, the pressure can be approximated as a function of the force and displacement. Using the decay model, the estimated pressures are calculated to be 15.0±5.3 kPa for the interspinous ligament and 37.5±20.0 kPa for the LF.  Table 2-2 Experiment 2: flow rate, force (Fa) and estimated pressure (Pest)  for human subjects using a midline approach.  Flow rate (mm3/s) Fa (N) Max Fa (N) Pest (kPa) Max Pest (kPa) Interspinous ligament 60±30 2.0±1.4 4.6±1.3 15.5±12.0 34.9±17.4 LF 12±13 5.0±3.0 6.0±3.0 31.5±28.0 39.5±30.3 p-value <0.05 <0.05 <0.05 <0.05 >0.25   62  Figure 2-7 A sample graph of a midline approach using the continuous pressure technique on a human subject showing a) displacement, b) force, and c) estimated pressure from the decay model.  2.4 Discussion  In experiment 1, the epidural needle traverses the interspinous ligament in the midline approach and muscle in the paramedian approach. For a comparable force and pressure, the flow rate is higher in muscle than in interspinous ligament; therefore the interspinous ligament is more resistant to injection and a physician performing a needle insertion should be careful not to mistake the relative ease of injection in muscle to be an indication of LOR in the ES. Also, when the needle is in the LF, a similar level of force and pressure produces a greater flow rate in the paramedian approach than midline. This phenomenon was unexpected as the needle tip is in the same LF. However, it is noticed  63 that the needle does not smoothly penetrate the LF but instead “pops” through. This means the needle tip is still partially in the interspinous ligament in the midline approach and in the muscle in the paramedian approach. When the needle tip is in the ES, similar forces, pressures and flow rates are measured for both approaches. Because the flow rate is higher in the paramedian approach than in the midline approach prior to the LOR region, it is said that the LOR is more subtle in the paramedian approach. Experiment 2 showed that the LF in live human subjects is more resistant to injection (12±13 mm3/s) than excised porcine tissue (92±48 mm3/s) for the same midline approach. The applied force was less on human subjects (5.0±3.0 N) than porcine tissue (8.9±5.3 N) but the very large difference in flow rate cannot be completely attributed to differences in applied force. The differences between excised porcine tissue (ex vivo) and human tissue in vivo include changes in temperature, blood pressure, perfusion, and dehydration, amoung others. This means that excised porcine tissue cannot adequately replicate human tissue in vivo for the purposes of training, and that practice on human subjects will likely remain the gold standard. The overall relationship between the interspinous ligament and LF in human subjects is similar to the relationship between interspinous ligament and LF in porcine tissue. The estimated pressure generated from the decay model is in the range of the pressures previously published. Previous research has reported a pressure of 67.6±12.2 kPa when the needle tip is in the LF for adults [Rodiera-1995] and 9.22±4.93 kPa for infants [Vas-2001]. The Tuohy needle used for these tests facilitates the introduction of the catheter into the ES with the exit orifice perpendicular to the central axis of the needle. This allows some lateral leakage of saline into the surrounding tissues in the interspace between muscle  64 tissue fibers which are perpendicular to the needle. The leakage of saline into the tissues, especially when performing a paramedian needle insertion, was not taken into account in the modeling.  Also, as more saline is injected, the surface of contact between the plunger and the barrel increases and this in turn increases the friction. The friction force is therefore non-linear. However, this effect is considered small enough to be negligible. More research should be performed to address more realistic friction models in a glass syringe. Additionally, modeling of the pressure by incorporating the instantaneous flow rate may be more accurate than the current method. For instance, in the region of LOR, the pressure is low but one could apply a high force. Only when looking at the high flow rate one would realize that the pressure is low and not only dependent on the force. The commonly used air bubble in the syringe for LOR to saline also affects the pressure readings as air is compressible. Errors from the presence of the glove between the sensor and thumb are very small. For the relatively slow movements of the plunger, the inertial effects of the glove material are negligible.  If the glove is modeled as an elastic element in series with the plunger and the thumb, then the force is unchanged by the transmission through the glove. This also applies to the compressible portion of the thumb. As long as the thumb, glove, sensor, and plunger are modeled as a series connection, and inertial effects are ignored, then the force is the same in all elements at all times. 2.5 Conclusion and future work The data collected from this work can be used to implement a more accurate haptic simulator for anesthesiology training which would contain force, pressure and  65 displacement components to characterize the feel of LOR. To further refine the models, a larger clinical trial is needed to detect differences among the subjects. The data captured also improves the overall understanding of the LOR. Using this data, the difference often reported by anesthesiologists between the paramedian and midline approaches can be quantified. The claim that anesthesiologists often make about the abruptness of the LOR in the paramedian versus midline approaches is quantified by the flow-rate data. The sensors used in this work can be used for education by senior physicians as an additional cue since they now have a way to quantitatively gauge how much force a student is applying in the procedure.  This data can be used to build simulators which can train the resident in using the LOR as part of the anaesthesiology simulator and high-fidelity simulation team training described in [Small-2008].  With the knowledge built in this chapter, the “feel” of LOR can be quantified, so real- time feedback can be given about the needle insertion, however, information about the needle insertion site and angle, as well as needle insertion depth remains unknown. In the next chapter, ultrasound will be used for prepuncture planning of the epidural needle insertion.   66 3 Pre-insertion ultrasound for epidural needle insertion1 3.1 Introduction Needle insertion for catheter placement in epidural anesthesia is a challenging procedure, particularly for non-experienced anesthesiologists [Grau-2003] [Wilson-2007]. In the previous chapter, a system to quantify the tactile feel of the LOR was presented but epidural needle insertion remains a blind procedure. Understanding the geometry of the lumbar anatomy for a given patient is needed to determine a suitable puncture site, needle trajectory and depth of needle insertion to reach the ES. As mentioned in the Introduction, there is growing interest in ultrasound imaging for epidural anesthesia. The main goal has been on estimating the depth to the ES, so that the anesthesiologist can anticipate the approximate region where the LOR will be used. In other words, ultrasound is expected to supplement, not to replace the LOR technique. Moreover, as described in Chapter 1, the current practice uses palpation to locate the intervertebral spaces which is unreliable.The primary aim of the work presented in this chapter was to create, and describe in detail, an ultrasound imaging technique that can be used for identifying the level of the vertebrae and estimating the depth of the ES. The goal is also to see if surrogate measures, such as weight, height, age, BMI or depth of the transverse process tip can be correlated to the needle insertion depth. Such surrogate measures may be useful for patients where ultrasound cannot locate the ES.   1 The material in this chapter has been published in [Tran-2009a].  67 3.2 Methods Ethics approval was obtained from the Clinical Review Ethics Board of the British Columbia Women’s Hospital and Health Center (C05-0409) to scan pregnant subjects. The subject’s age, weight, and height were recorded. Informed written consent was obtained for all subjects (n=20).  The exclusion criteria were the usual contraindications to neuraxial anesthesia [Miller- 2005] (infection at the injection site, bleeding diathesis, known left ventricular outflow obstruction, hypovolemia, and increased intracranial pressure) as well as the inability to speak English. Four subjects went through labor and 16 subjects went through cesarean section, for which a combined spinal-epidural anesthesia was administered [Cook-1999]. The LOR technique used saline with continuous pressure on a glass syringe plunger. The subjects were scanned in a seated position with spinal flexion in a similar posture as when performing the epidural needle insertion. The subjects were given leg, arm and head supports for comfort and stability. The bed was raised to allow the sonographer to sit behind the patient and support the scanning arm by placing the elbow on the bed surface.  This provided transducer and scanning plane stabilization.  This experimental setup is shown in Figure 3-1. The anesthesiologist used the Tuffier’s line to locate the L3- 4 intervertebral level by palpations. Then, the midline was also identified by palpation of the spinous process by the anesthesiologist. Then the sonographer used the ultrasound technique described below to locate the L2-3 and L3-4 interspaces. In case of disagreement, the count from the ultrasound examination was chosen because ultrasound was previously shown to be more accurate for counting the intervertebral levels [Furness-  68 2002]. The needle insertion was performed in the midline at either the L2-3 or L3-4 interspace depending on the subject’s anatomy. The widest interspace was chosen by palpation. If palpation did not provide a clear indication of interspace width, ultrasound was used to visually assess the width of the interspace by visualizing the interlaminar space.  Figure 3-1 Experimental setup: patient in sitting position with arms resting on a table. Sonographer scans with elbow rested on the elevated bed.  The same sonographer (Victoria Lessoway) performed all pre-puncture ultrasound examinations and the same anesthesiologist (Allaudin A Kamani) performed all epidural needle insertions. Scanning and data capture were performed by an experienced Registered Diagnostic Medical Sonographer (RDMS) using an Ultrasonix RP500 and a 1-5  MHz broadband curvilinear transducer (Ultrasonix Medical Corp., Burnaby, BC, Canada). The ultrasound images were taken in the paramedian plane, described below, as it was found to be the best window for seeing the LF.  Two ultrasound scanning methods  69 were used to identify the intervertebral level: counting-down from the 12th rib and counting-up from the sacrum.  The following description is for identifying the L3-4 interspace, but a similar technique applies to L2-3 which is simply one level above. The counting-up method starts with the ultrasound transducer over the sacrum, as shown in Figure 3-2a. The echoes from the sacrum are seen arising from the flat bony surface that is readily palpated just under the skin surface. Then, the ultrasound transducer is moved upward through L5, L4, and L3, counting the interlaminary or interfacet joint spaces as shown in Figure 3-3. The transducer is then placed over the spinous processes and moved slightly laterally 2-3 mm, or until the spinous processes no longer appear in the images and the laminas are observed. Then the transducer is angled toward the midline 0-10° to visualize the LF and thus locate the LF. The angle depends on the geometry between the skin-to-LF depth and the distance the transducer is lateral to the midline. This is the paramedian plane.  The counting up technique relies on the subject having 5 lumbar vertebrae and a normal sacrum, in other words, with no lumbarization or sacralization. The subject was placed in the sitting position. The transducer was initially placed (with the top of the transducer oriented cephalad), on the back, 5-10 cm lateral to the midline where the 12th rib could be seen ultrasonically superficial to the kidney. The 12th rib was then followed medially where the transverse process of L1 could be seen to arise 3-5 cm from the midline. The transducer was then moved caudad parallel to the midline. The transverse processes (Figure 3-4) of L2 to L5 could be identified and the acoustic shadows appear like fingers.  70 On moving the transducer medially, while maintaining the same orientation, the transverse process became less defined and was replaced by a wider shadow. This is the intervertebral facet joint (Figure 3-5) and looks like a thumb.  More medially, the facet joint shadow diminished and a deeper flatter “wavy” pattern was seen (Figure 3-6a). This is the lamina, where the intervertebral spaces were seen as gaps in the wavy line. The transducer then laid immediately paramedian to the midline spinous processes. By gently angling the transducer toward the midline by 5-10°, the ultrasound beam could be directed at the base of the spinous process where it intersected with the laminae. This enabled better definition of the lamina, the foramina and the LF/dura mater complex, that was seen as a small white line deep to the foramina. This paramedian plane provided the optimum window for ultrasound images of the lumbar anatomy [Grau-2001c]. The L3-4 vertebral interspace generally lies directly medial to the L4 transverse process.  The success of the counting down method depends on the rib count abnomalities (estimated occurrence is 8% of the population have greater or fewer than 24 ribs [Loder-2007]).  Small adhesive markers were fixed to the skin surface lateral to the L2-3 and L3-4 interspaces observed in the ultrasound image to ensure a match between the ultrasound images of a particular level and the actual level chosen by the anesthesiologist for needle insertion. Figures 3-3 to 3-6 were acquired using a C3-7 3D transducer.   71 a) b) Figure 3-2 Counting intervertebral spaces a) Counting-up from the sacrum b) and counting-down from the 12th rib. The dark rectangle shows the ultrasound transducer position.   Figure 3-3 Ultrasound image of the sacrum. The dark rectangle shows the ultrasound transducer position.   72  Figure 3-4 Ultrasound image of the transverse processes (finger-like structures).  The dark rectangle shows the ultrasound transducer position.   Figure 3-5 Ultrasound image of the facet joints (thumbprint-like structures). The dark rectangle shows the ultrasound transducer position.   73 muscle lamina ligamentum flavum dura mater epidural space (a) (b) (c) (d) vertebral body  Figure 3-6 Paramedian lumbar anatomy a) ultrasound image resulting from a paramedian ultrasound plane b) idealized echoes from the paramedian ultrasound of the lumbar anatomy c) Visible Human project slice of the paramedian lumbar anatomy [Ackerman-1998] ( © 1998 IEEE) d) schematic of paramedian lumbar anatomy.  Ultrasound images of the location of the ES and transverse processes were also captured. In this adult population, the location of the LF (which is immediately above the ES) is identified by a small pair of parallel linear echoes (“the doublet”) seen at right angles to the ultrasound beam at a depth of 20-90 mm [Grau-2001a].  These doublet echoes are difficult to observe and it requires some practice and anatomical familiarization to identify them. The doublet echoes are seen superior to the upper margin of the base of the spinous process at its junction with the lamina.  In [Grau-2001b], the first echo of the doublet is assumed to arise from the interface between the muscle and LF, and the second  74 echo is from the interface between the ES and the dura mater. All ultrasound measurements were made to the leading edge of the first doublet echo and are therefore the skin-to-LF depth. The paramedian plane was used for all ultrasound scans and all the epidural needle insertions were performed in the midline.  The distances on the captured ultrasound images were measured using the electronic calipers provided by the ultrasound machine software. The skin-to-LF depth for L2-3 and L3-4, and the distance from skin to the transverse processes of L2, L3 and L4 were recorded. For needle insertion at L2-3, the right transverse process of L3 was used for comparisons. For needle insertion at L3-4, the right transverse process of L4 was used for comparisons. The skin-to-LF depth was measured on the ultrasound image as the distance from the skin surface to the leading edge of the first echo in the doublet (Figure 3-6a). The skin to the transverse process was measured on the ultrasound image as the distance from the skin surface to the leading edge of the echo from the tip of the transverse process (Figure 3-4).  The acquired ultrasound images were also transferred to the laboratory for offline measurements of the subcutaneous fat layer over the region of interest. Fat was measured as the thickness of the hypoechoic region directly beneath the skin surface and above the muscle (Figure 3-7).  75  Figure 3-7 Subject with a significant layer of subcutaneous fat shown by the white caliper  After completion of the ultrasound scans, the subjects were prepared for epidural catheter placement using a FlexTip Plus Catheter model JH-05500 kit (Arrow International Inc., Reading, PA, USA) in the operating room. The L2-3 and L3-4 interspaces were identified for needle puncture by the anesthesiologist using palpation and confirmed by the skin markers placed by the sonographer.  The site of insertion is anesthesized with local anesthetics. Three mL of sterile saline and 1 mL of air was used for the LOR technique. Then, the epidural needle was inserted in the midline approach into the interspinous ligament. The saline-filled syringe was then attached to the needle. The epidural needle was inserted with relatively constant pressure on the plunger. When the needle reached the ES, the LOR was felt and measured on the sensors described in Chapter 2. The needle shaft was marked at the point of skin puncture, withdrawn, and the distance from the mark to the needle tip was measured by a millimetric ruler (±0.5 mm). This measurement is the distance the needle traveled and is considered as the actual needle insertion depth and consequently the depth to the ES. For combined spinal-epidural procedures, a smaller  76 spinal needle was inserted into the epidural needle and advanced slightly further to puncture the dura mater. The flow of cerebrospinal fluid confirmed the position of the tip of the spinal needle that had moved beyond the interface of the ES and dura mater. The catheter was then threaded into the ES and anesthesia delivered.  The Pearson correlation ratio was used to assess how a measure, such as the age of the subject, was related to the needle insertion depth. The square of the correlation coefficient, R2 (coefficient of determination) is used for convenient comparison to previous work. A higher R2 value means the two measurements are very closely dependent whereas a value close to zero means very little or no dependence. 2 2 |2 1 x yxR             eq 3.1 where σx2 is the variance of the needle insertion depth x and σx|y2 is the square error of the ultrasound measured skin-to-LF depth to the best fit linear regression line.  The Bland-Altman analysis is used to assess the accuracy of the measurement of the skin- to-LF depth using ultrasound compared with the needle insertion depth. A bias value corresponding to the average of the differences of the two measurements is provided as well as 95% limits of agreement corresponding to μ ± 2.086 σ, where μ is the bias of the differences, σ is the standard deviation and 2.086 corresponds to the T-distribution coefficient when the sample size is n = 20. In our case, the bias is a value by which the ultrasound measured depth to the ES is expected to be different from the actual needle insertion depth and the 95% limits of agreement are the limits between which 95% of ultrasound measured depth to the ES should fall around the needle insertion depth.  77 3.3 Results Subject biometrics were measured and shown in Table 3-1. The average subject age was 35 ± 4 years (mean ± standard deviation), weight was 76 ± 15 kg, height was 161 ± 7 cm, BMI was 29 ± 7, and the depth to the ES was 51 ± 11 mm. The time needed for navigating to the L2-3 and L3-4 interfaces, and obtaining a clear depiction of the LF, ranged from approximately ten minutes for the first subjects to approximately three minutes for the last subjects.  Table 3-1 Subject biometrics and data on ES depth and surrogate measures. The transverse process depth on subject 8 is missing due to lack of time in the surgery preparation room. patient no age weight (kg) height (cm) BMI Actual needle depth Ultrasound needle Depth Associated transverse process depth 1 35 62 168 22.0 41.0 43.1 43.1 2 32 62 160 24.2 52.0 44.2 46.0 3 33 80 157 32.5 49.2 50.5 49.6 4 40 66 157 26.8 49.2 40.6 39.1 5 37 70 164 26.0 38.0 40.7 37.3 6 36 73 179 22.8 49.0 40.7 37.6 7 40 56.5 155 23.5 39.8 39.8 39.4 8 35 100 162 38.1 68.0 57.7 n/a 9 36 55 160 21.5 62.0 52.1 52.3 10 36 84 166 30.5 48.0 50.1 59.0 11 35 82.5 169 28.9 47.7 45.0 50.1 12 33 86 155 35.8 60.0 55.8 61.9 13 30 74 169 25.9 56.0 48.2 42.7 14 33 114 149 51.3 80.0 77.4 66.3 15 38 91 170 31.5 52.0 43.3 48.6 16 32 62 155 25.8 39.5 37.6 33.4 17 44 64 166 23.2 46.0 41.0 33.5 18 34 61 157 24.7 45.0 41.4 36.6 19 23 71 163 26.7 51.0 37.7 31.7 20 38 86 150 38.2 64.0 55.1 45.5   The ultrasound measured skin-to-LF depth is correlated to the actual needle insertion depth using a least-squares fit (Figure 3-8; R2 = 0.7962). The Bland-Altman analysis for  78 comparing the skin-to-LF depth on the ultrasound to the needle insertion depth has a bias of 4.8 mm and 95% limits of agreement of -5.2 mm to 14.7 mm(Figure 3-9). The R2 values correlations of age, height, weight and BMI (using only the 16 subjects who underwent needle placement at L3-4) are 0.023, 0.043, 0.159 and 0.249 respectively (Figure 3-10), demonstrating that subject biometrics are a poor predictor of the needle insertion depth. The comparison of the needle insertion depth and the depth to the tip of the nearest transverse process for subjects on which the needle insertion was performed in the L2-3 space was not analyzed because of the small sample size (n=4). Figure 3-11 shows that the least square fit for the correlation between the depth to transverse process of L4 and the needle insertion depth L3-4 produces an R2 = 0.35. The Bland-Altman test for this pair yields a bias of 2.7 mm and 95% limits of agreement of -13.8 mm to 19.1 mm. These results suggest that the transverse process is not suitable as a surrogate measure for the ES for depth measurement. Figure 3-11b shows the correlation between fat thickness and needle insertion depth. The least-squares fit produces R2 = 0.66. The weight and BMI are also weakly correlated with fat thickness: linear correlations of the fat thickness to weight and BMI produce R2 values of 0.19 and 0.31 respectively.   79 y = 1.0034x + 4.6164 R2 = 0.7962 30 40 50 60 70 80 90 30 35 40 45 50 55 60 65 70 75 80 Ultrasound skin-to-LF depth (mm) N ee dl e in se rti on  d ep th  (m m ) line of equality  Figure 3-8 Skin-to-LF depth measured by ultrasound (x axis) correlated with actual needle insertion depth (y axis). There is a systematic error of 4.6 mm between the ultrasound skin-to-LF depth and the needle insertion depth.  ‐6 ‐1 4 9 14 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 er ro r b et w ee n t he  ne ed le  ins er tio n d ep th   an d u ltr as ou nd  sk in ‐to ‐LF  de pt h Subject number  Figure 3-9 Bland-Altman plot of the difference between ultrasound-measured skin-to-LF depth and the needle insertion depth. It is observed that the ultrasound-measured distance is generally shorter than the needle insertion depth. The bias is 4.8mm and is shown by the dash line. The 95% limits of agreement are shown by the dotted lines at -5.2 mm and 14.7 mm.   80 R2 = 0.2493 35 40 45 50 55 60 65 70 20 25 30 35 40 BMI N ee dl e in se rti on  d ep th  (m m )  a) R2 = 0.1587 30 35 40 45 50 55 60 65 70 40 50 60 70 80 90 100 weight (kg) N ee dl e in se rti on  d ep th  (m m )  b) R2 = 0.043 30 35 40 45 50 55 60 65 70 140 145 150 155 160 165 170 175 180 height (cm) N ee dl e in se rti on  d ep th  (m m )  c) R2 = 0.023 30 35 40 45 50 55 60 65 70 20 25 30 35 40 45 50 age N ee dl e in se rti on  d ep th  (m m )  d)  Figure 3-10 The correlation of needle insertion depth with biometrics such as a) BMI b) weight c) height d) age. There is little to no correlation between these biometrics and the needle insertion depth.  81  R2 = 0.3475 30 35 40 45 50 55 60 65 70 30 35 40 45 50 55 60 65 70 skin-to-L4 transverse process tip depth (mm) N ee dl e in se rti on  d ep th  o f L 3- 4 (m m )  a) R2 = 0.6605 30 35 40 45 50 55 60 65 70 0 2 4 6 8 10 12 14 fat thickness (mm) N ee dl e in se rti on  d ep th  L 3- L4  (m m )  b) Figure 3-11 a)Needle insertion depth at L3-4 correlated to the distance from skin to the tip of the transverse process L4. b) Needle insertion depth correlated with the thickness of overlaying subcutaneous fat.  82  3.4 Discussion  The paramedian ultrasound skin-to-LF depth is well correlated with the actual needle insertion depth but the limits of agreement still suggest the use of the LOR as the needle gets closer to the predicted ES depth. This conclusion agrees with [Grau-2001a] as ultrasound is only intended to be used to reduce the region of needle insertion that uses LOR. The other measurements (fat thickness, skin-to-transverse process tip depth and biometrics) provide less accurate estimates and should not be used to predict the actual needle insertion depth.  The R2=0.7962 of ultrasound skin-to-LF depth compared to needle insertion depth is similar to previous work (R2=0.79-0.92 [Arzola-2007] [Grau-2001b] [Grau-2002b] [Grau-2001f] [Grau-2001a]). The bias of -4.8 mm in this work can be attributed to the measurement of the skin-to-LF depth measured to the top of the echo from the LF (as opposed to the bottom of the echo), as the LF is typically about 5 mm thick. The limits of agreement obtained were 5.2 mm to -14.8 mm, however, if the thickness of the LF is taken into account and the bias of 4.8 mm is corrected for, the limits of agreement become -10 mm to 10 mm. These limits are slightly larger than those obtained in previous work with the limits of agreement of 5.1 mm-7.7 mm [Arzola-2007] [Grau- 2001b] [Grau-2002b] [Grau-2001f] [Grau-2001a]. The subjects in this study had an average depth to the ES of 51.9 mm which is comparable to the average needle depths of 46.5 mm-57.5 mm [Arzola-2007] [Grau-2001b] [Grau-2002b] [Grau-2001f] [Grau- 2001a] in previous research.  83  The ultrasound scan is performed in the paramedian plane and the needle insertion is performed in the midline, so small geometric errors may occur. The midline needle path is angled to match the angle between the spinous processes whereas the paramedian ultrasound is measured straight down. However, the paramedian ultrasound is slightly angled toward the midline (5° to 10°, giving a distance roughly 0.4% to 1.5% larger) which adds a small error to the measurements. The paramedian ultrasound is also taken from the top of the erector spinae muscle which can add some distance compared to the midline distance since the subjects’ backs are not perfectly flat. There are no obvious outliers in the correlations of ultrasound measurements to actual depth of ES in Figure 3-8, which also suggests that the needle entered the ES in all subjects. This means that the errors between ultrasound and actual needle measurements are due to other sources than mistaken LOR. And finally, another source of error is the different compression levels caused by the ultrasound transducer on the skin surface compared with the compression caused by the epidural needle in the operating room. This effect will be studied in Chapter 4. Because of the large limits of agreement, the paramedian skin-to-LF depth measured by ultrasound should only be used as an approximate estimate of the needle insertion depth.  As anticipated, patient biometrics were poorly correlated with the needle insertion depth. This was also the case in previous work by Stamatakis [Stamatakis-2005]. Fat thickness was slightly better correlated but still weak. This result may still be useful in the  84 definition of a region of interest for locating the LF, especially in the context of an automatic LF detection system as described in Chapter 6.  It is appealing to replace the measurement to the small echo produced at the ES with a stronger, hyperechoic anatomical feature, but the transverse process is not a suitable choice. Other vertebral features, such as the superior or inferior articular processes or the lamina, should be investigated since bone is a strong reflector of ultrasound and can be accurately identified [Hacihaliloglu-2006]. In Chapter 6, the strong reflection of the lamina is used as a first step to automatically locate the LF.  The fat thickness was shown to have a weak correlation with the depth to the ES (R2=0.66). If a stronger relationship can be found between fat thickness and skin-to-LF depth, then a distance could be provided to the anesthesiologist about the depth of the trajectory through the muscle, the beginning of which can often be felt during insertion. An estimate of fat thickness cannot be made directly from weight (R2=0.19) or BMI (R2=0.31) so other measures are needed. This result also suggests that if an anesthesiologist can estimate the fat thickness through pressure, it would provide a smaller region of interest in which one would search for the LF using the LOR.  Previous work has concluded that the paramedian plane gives the optimum window for ultrasound of the ES [Grau-2001c]. This means the paramedian plane scanning technique deserves careful study, which is especially helpful for anesthesiologists inexperienced with ultrasound. Although some researchers have found success with the transverse plane  85 [Arzola-2007], others have compared the transverse, midline longitudinal, and paramedian planes and found the paramedian plane offers the best image quality [Grau- 2001c].  In our work, we agree the paramedian plane provides best quality images and so efforts will be made to develop techniques built around a paramedian ultrasound image.  In Chapter 4, a method for real-time ultrasound-guidance for epidural needle insertion will be described. As the paramedian plane gives the best images, it will be the plane of choice in that study. The real-time system will attempt to solve the problem of the discordance of measurements made prior to the needle insertion as opposed to directly observing the needle reach the target. Some of the issues pertaining to the compression caused by the ultrasound transducer will also be addressed more in depth.    86 4 Real-time ultrasound guidance for epidural needle insertion1 4.1 Introduction As described in the previous chapters, epidural needle insertion is most commonly performed using an anatomical landmark-based technique using palpation to identify the chosen intervertebral level and site of insertion, and the position of the needle tip (insertion depth) is estimated using the LOR to saline/air injection. This method relies on “feel” to confirm entry of the epidural needle in the ES.  Ultrasound pre-puncture scanning can identify the intervertebral level [Wallace-1992] [Furness-2002] and the midline for needle insertion. It can also give information about best angle, direction of approach and depth to the ES as means to improve successful placement [Grau-2003] [Grau-2001a] [Grau-2002b] [Arzola-2007]. It is observed however in [Grau-2001a], that the measured needle angle at the time of prepuncture examination and the actual needle insertion angle only have a weak correlation of R2=0.30. A difference of a few degrees on the angle can cause redirections and a large error on the expected skin-to-LF depth which can affect the success of the needle insertion. The ultimate goal is to plan the trajectory of the needle to achieve successful needle placement in one single attempt without redirections (needle is withdrawn and reinserted). This can be done with real-time ultrasound-guided needle insertions. This is true for both epidural and combined spinal-epidural anesthesia [Cook-1999]. We use the  1 The material in this chapter has been published in [Tran-2009c]  87 term epidural needle insertion for both cases for simplicity in this chapter. Although small differences exist, the first step of the combined spinal-epidural is an epidural needle insertion.  In efforts to achieve real-time ultrasonic observation of the epidural needle insertion, investigators have previously described a two-operator real-time ultrasound-guided technique of epidural needle placement [Grau-2004] [Willschke-2006]. One operator performs a midline epidural needle insertion whilst observing its entry into the ES on a paramedian ultrasound image from a second operator. As seen in Chapter 1, on Figure 1-15, having 2 operators can be cumbersome, especially when sterile conditions should be respected. A single-operator in-plane ultrasound-guided epidural needle insertion method has been recently described [Karmakar-2009] but it requires a specialized Episure spring-loaded needle (Indigo-Orb, Irvine, CA). It is also a free-hand ultrasound technique, meaning that the needle trajectory cannot be superimposed on the ultrasound image for planning before insertion. A free-hand technique allows flexibility in scanning, but requires considerable operator expertise [Matalon-1990]. The needle insertion angle still cannot be set accurately to guarantee a successful epidural needle insertion in a single attempt.  In this chapter, a fixed in-plane needle angle relative to the ultrasound probe is used to facilitate the pre-puncture determination of the needle insertion depth from a simple measurement along this path, as shown in the ultrasound image as a superimposed line.  88 This superimposed needle trajectory line also helps the anesthesiologist choose a needle insertion site and angle suitable for a direct path toward the ES.  This chapter first describes usage of the systematic method for ultrasound pre-puncture scanning to identify the lumbar level for the puncture site defined in Chapter 3 and compares it to the palpation-based method. Then, a method for real-time guidance of needle insertion is described. The primary objective is to determine whether it is feasible and practical to perform this aim-and-insert technique with a single operator. The secondary objectives are to compare ultrasound and palpation-based identification of the intervertebral space, study the effect of transducer pressure on measured distances in the lumbar paramedian plane, compare the straight-down depth to the ES to the diagonal needle trajectory and the relationship to the needle guide geometry and, finally, compare the site of insertion selected in the pre-puncture examination with the actual insertion site. The secondary objectives are expected to provide insight into the limitations of the proposed method.    89 4.2 Methods 4.2.1 Subject selection The study was approved by the Children and Women’s Health Centre of British Columbia and the University of British Columbia’s clinical research ethics boards (CW05-0262 / H05-70409). Subjects were recruited using informed written consent. The subject conditions for exclusion were the inability to speak English and the standard contraindications for spinal anesthesia. Subject age, weight, height were collected and the BMI was calculated. As this is a feasibility study, test subjects were those scheduled for cesarean delivery so that labour pains did not disrupt the time or subject stability needed to learn the new technique. For such subjects, a combined spinal-epidural procedure was used. The sample size was 20. This number was chosen because a minimum of 10 subjects was found to give an 85% success rate [Grau-2003] for other ultrasound guidance studies, a success rate of 90% was achieved after 20 patients which is the goal in [Kopacz-1996]. Studying 20 subjects allows more robust statistical analysis while keeping the study within a reasonable amount of time.  For consistency, one sonographer (Victoria Lessoway) performed all pre-puncture measurements. Using ultrasound guidance, one experienced anesthesiologist (Dr Allaudin Kamani) performed all epidural needle insertions. All complications (insufficient spinal block, failure to obtain LOR, paresthesia, failure to obtain CSF aspirations, patient discomfort) during the course of the procedure were noted.   90 4.2.2 Intervertebral level identification In the pre-surgical assessment area the anesthesiologist identified and marked the lumbar interspaces, L2-3 and L3-4, using traditional palpation-based techniques and Tuffier’s line [Chestnut-2004]. The sonographer, using our designed systematic approach described in Chapter 3, also identified and marked the L2-3 and L3-4 interspaces, using a curvilinear 1-5MHz ultrasound transducer (Model C5-1/60, Ultrasonix Medical Corporation, Richmond, Canada).  The difference between the anesthesiologist’s palpation-based intervertebral space identification and the ultrasound-based identification indicated with skin surface marks was measured using a flexible millimetric ruler. The distance between the two facet joints was also measured as shown in Figure 4-1b. and called the interfacet distance. If the difference was greater than one full interfacet distance then it was considered a misidentification of the intervertebral level by the landmark technique, as interspace identification by ultrasound has been shown to be more reliable [Furness-2002]. Although ultrasound is not always correct in identifying the intervertebral space count (off by one space in 14.7% of cases), it remains a more reliable count than palpation. The measurement of the intervertebral spacing (Figure 4-1b) was calculated as the distance between the facets of L2-3 and L3-4 as measured with ultrasound using software callipers by the experienced sonographer.    91   a)  b)  c) Figure 4-1 a) Transverse processes (finger-like appearance), b) facet joints (thumb-like appearance) with the interfacet distance measurement taken from the centre of one facet  to the centre of the neighbouring facet, c) ultrasound image with dashed needle guide line representing the predicted needle path. The wave-like structures are the lamina and, at the base of the lamina, the LF can be seen as a bright reflector, with the ES underneath. The needle guide line is aimed at the visible portion of the LF. Each large white dot represents 10 mm marking.  4.2.3 Geometric measurement for aim-and-insert technique A needle guide bracket (Ultrasonix Corp., Richmond, Canada) for the curvilinear 1-5 MHz transducer was mounted (Figure 4-2). The needle guide sets the needle angle β (as epidural space lamina anterior side of spinal canal  92 defined in Figure 4-3) to 23° with respect to the transducer centreline. This angle would permit the needle to intersect with a target that is approximately 50mm deep at the center of the image, where the resolution is highest.  Figure 4-2 Assembled sterile transducer with needle guide and epidural needle attached, showing top view. The needle bracket was attached to the ultrasound transducer and the needle guide was attached to the bracket. The epidural needle was then mounted on the needle guide at an angle of 23° with respect to the ultrasound transducer.   Transducer cover Epidural needle Needle guide Needle guide bracket  93 AB α C D β  Figure 4-3 Schematic of probe and needle guide geometry, viewed from the side. The distance A is the distance from the felt tip pen mark, indicating the center of the ultrasound transducer when it is properly positioned for needle insertion in the surgery preparation room, with the needle guide line aligned with the ES. The distance B is the distance seen on the ultrasound image along the needle guide line on the ultrasound image, going from the LF to the edge of the image. C is the distance the needle is inserted in the skin that is not seen in the ultrasound image. D is the length of the needle guide. α is the angle the transducer is placed against the normal to the skin surface. β is the angle of the needle with respect to the ultrasound transducer.    94 The use of a needle guide and the compression of tissue from contact pressure of the transducer require careful analysis of the geometrical measurements. With the transducer perpendicular to the skin surface, the skin-to-LF depth (distance E in Figure 4-4) was measured when no pressure was applied to the ultrasound transducer and measured again with a typical contact pressure required by an experienced sonographer to achieve an adequate image. The distance A is the distance between (1) the mark of the center of the transducer made in the pre-puncture examination where the needle guide line is aligned with the ES and (2) the actual needle insertion site and is shown in Figure 4-5.   a)  E B α+β C α  b) Figure 4-4 Geometry relating the measurement of skin-to-LF depth a) straight down (distance E) and b) the measurement along the diagonal needle guide line (distance B) where there is a need to angle the transducer by an angle α accounting for tissue deformation at the transducer location, and a needle angle β with respect to the transducer axis.  95 A actual puncture site actual position of transducer, angled medially by 5-10° pre-puncture transducer footprint marks made from center of pre-puncture transducer position shown relative to actual puncture site  Figure 4-5 Schematic of distance A measurement, viewed from the back.   The depiction of the expected needle trajectory was superimposed digitally on the ultrasound image as a dashed line. The transducer was repositioned so the expected needle path targeted the LF echoes whilst ensuring that the bottom of the needle guide apparatus rested on the skin surface.  This involved aiming the transducer slightly cephalad by an angle α while maintaining the same scanning plane as shown in Figure 4-3. Distance B, the diagonal ultrasound-measured needle depth along the guide line, was measured by the software callipers and recorded. The total of the ultrasound distance B plus the fixed length of the guide (distance D) of 31.9 mm, and the blind region (distance C) of 8.7mm, was the length required for needle insertion.   96 As shown in Figure 4-3 and Figure 4-4, the distances B, C, E and angles α, β are related geometrically by CB E  )cos(           eq 4.1 or CEB  )cos(           eq 4.2 where C is 8.7mm, α is the 15° transducer tilting angle (the approximate transducer angle with respect to the normal to the subject’s skin as it is angled so the biopsy guide touches the skin), and β is the 23° angle of the needle relative to the transducer. Angle α is an approximate angle because the transducer can be tilted more or less to get a better image quality, although the guideline is to have the guide touching the skin. Angle β is fixed by the needle guide. Eq. 4.2 can then be written as  mmEB 7.827.1  .         eq 4.3  The anesthesiologist then marked the vertical and central plane of the ideal transducer position on the skin as shown in Figure 4-5. These marks were used later for faster placement of the ultrasound transducer at the correct level in the operating room where less time is available for scanning.   97 4.2.4 Needle insertion and block procedure The second part of the experiment was in the operating room. Standard aseptic methods were used including a sterile transducer cover (Cone Instruments, Solon, Ohio, USA) and a sterile disposable needle guide (Protek Medical Products, Coralville, Iowa, USA). The skin and subcutaneous tissues were anesthesized at the predicted needle insertion site, about 30-40mm (distance A) below the felt-tipped pen mark corresponding to the centre of the transducer as shown in Figure 4-5. The ultrasound transducer was placed on the previously marked locations and was positioned and angled so that the superimposed needle trajectory guide line was aimed at the ES on the ultrasound image. A 125mm New Gertie Marx® CSE-Set epidural needle (IMD Incorporated, Huntsville, Utah, USA) was inserted in-plane with the paramedian plane of the ultrasound. The epidural needle was used because it is more rigid than the spinal needle and guaranteed the actual needle path to follow more closely the predetermined needle path on the ultrasound image. Using this in-plane needle arrangement, the needle could be observed during the entire course of the insertion. Care was taken to ensure that the predicted needle path, superimposed on the scanned image, targeted the LF throughout the procedure. The epidural needle, with the stylet still inside the shaft, was inserted until the tip was seen to be approximately 10 mm away from the expected position of the ES (the 95% Bland-Altman limits of agreements from measurements in Chapter 3 are approximately 10 mm). Then, a saline-filled syringe was mounted on the epidural needle and LOR to saline injection was used to advance the needle until it went through the LF and into the ES.  Care was taken not to inject saline until being about 10 mm from the LF to avoid degrading the ultrasound image. This is the aim-and-insert technique. Although this procedure is a combined spinal-epidural, it is  98 expected that the success or failure of single-operator ultrasound guidance would also apply to epidural needle insertion.  The endpoint for a successful procedure under ultrasound guidance is defined as a LOR permitting either successful anesthesia or a successful threading of a catheter. For an elective cesarean delivery, a spinal neuraxial block is our standard. The spinal anesthesia is then injected in the subarachnoid space. If the spinal needle insertion is successful and the dura is punctured, a few drops of CSF should be seen as confirmation. In case of absence of CSF, an epidural catheter was threaded into the ES to confirm successful epidural placement of the needle. Then, a standard midline spinal needle insertion would be performed to administer the spinal anesthesia. Spinal anesthesia was used to deliver intrathecal hyperbaric bupivacaine (12 milligram (mg)), fentanyl (10 μg), morphine (100 μg) for regional anesthesia for cesarean delivery. Successful subarachnoid block was verified through the standard tests for cesarean delivery (ice at the fourth thoracic vertebra (T4) and pin prick at T6).  4.2.5 Statistical analysis The Bland-Altman analysis is used to assess the agreement of the measurement of the actual distance B compared with the distance calculated from eq. 4.3 using the measurement of distance E. The 95% limits of agreement are the limits between which 95% of distance B should fall around the calculated distance from eq. 4.3.  99  4.3 Results  Twenty subjects gave signed consent, but one subject had unrecognizable bony landmarks and so did not participate in the study. The participating subjects (n=19) were aged 35±5.3 years old (mean±standard deviation), weighed 80.3±13.2 kg, had a height of 160.0± 5.6 cm and a BMI of 31.5±5.9 (with 11 obese BMI > 30, 5 overweight 25 < BMI < 30 and 3 normal subjects BMI < 25) as shown in Table 4-1.  All subjects underwent attempted needle placement using the new real-time ultrasound- guided single-operator aim-and-insert technique. The epidural needle was successfully guided into the ES, as defined by good LOR, in 18 of the 19 subjects. Real-time ultrasound guidance failed to locate the ES in one subject (subject 9) in which case, despite good ultrasound and needle views, LOR to saline could not be elicited. In 14 out of the 18 subjects, there was CSF aspiration confirming that the spinal needle had passed through the ES giving further evidence, though not a guarantee, that the needle tip was in the ES. In the 4 subjects with no CSF aspirated, a catheter was easily threaded into the ES giving partial confirmation of a successful epidural needle insertion. Spinal anesthesia provided a satisfactory block in all 18 subjects.  Measurements of the interspace identification in 19 subjects, listed in Table 4-2, show the differences between the palpation-based mark and the ultrasound-based mark. In 2 out of 19 subjects (subjects 5 and 7), a difference of 25mm was found. The magnitude of this  100 difference was compared to the respective interfacet distances. The differences for subject 5, with 24.1 mm interfacet distance, and subject 7, with 23.7 mm interfacet distance, were considered to be a miscount of intervertebral level.  101 Table 4-1 Subject biometrics and comments   Subject Age (years) Weight (kg) Height (cm) ES: LOR obtained with ultrasound guidance Comments about success of overall procedure 1 33 59 157 Yes 2 31 80 155 Yes 3 43 98 163 Yes 4 36 93 157 Yes 5 26 61 163 Yes 6 41 95 158 Yes 7 27 99 156 Yes 8 41 65 161 Yes 9 25 74 161 No LOR to saline could not be elicited 10 36 82 160 Yes 11 33 94 169 Yes No CSF aspirated, epidural catheter passed, midline spinal performed for cesarean 12 40 80 170 Yes 13 41 78 156 Yes 14 33 80 165 Yes No CSF aspirated, epidural catheter passed, midline spinal performed for cesarean 15 37 83 155 Yes No CSF aspirated, epidural catheter passed, midline spinal performed for cesarean 16 37 95 150 Yes 17 37 79 159 Yes No CSF aspirated, epidural catheter passed, midline spinal performed for cesarean 18 30 57 155 Yes 19 31 73 170 Yes  102  Table 4-2 Difference between anesthesiologist mark and sonographer mark (mm) to compare palpation versus ultrasound method. Data that is missing due to time constraints are labeled “n/a”. Subject Difference(mm) Associated interfacet distance 1 0 31.5 2 15 27.4 3 0 31.6 4 0 28.7 5 25 (different level) 24.2 6 17.5 27.5 7 25 (different level) 23.7 8 15 27.5 9 10 27.7 10 0 36.6 11 2.5 31 12 2.5 32.6 13 0 32.2 14 7.5 28.0 15 2.5 25.8 16 0 36.5 17 0 29.8 18 10 n/a 19 9 24.1   The ultrasound-measured ES depths E (Figure 4-4) with, and without, contact pressure (n = 17) are compared. Due to time constraints in the preparation room, measurements of the depth of ES were not obtained from two subjects, so these were excluded. The average difference is 2.8±1.1 mm and the 95% limits of agreement were 0.4 mm to 5.2 mm according to Bland-Altman analysis. Differences in measurements are shown in Table 4-3.  103  Table 4-3 Skin-to-epidural space depth with and without transducer force. The average error is 2.9mm. There is missing data for 12 and 16 (insufficient time in the surgery preparation room). Subject Distance E with transducer pressure (mm) Distance E without transducer pressure (mm) Difference (mm) 1 44.3 46.7 2.4 2 49.5 50 0.5 3 58.5 61.2 2.7 4 53.7 57.8 4.1 5 35.3 37.6 2.3 6 49.2 52 2.8 7 57.1 62.2 5.1 8 36.7 40.2 3.5 9 38.9 42.1 3.2 10 46.5 50.8 4.3 11 48.8 51.3 2.5 12 51.2 n/a n/a 13 44.1 46.5 2.4 14 42.8 45.7 2.9 15 47.7 50.4 2.7 16 49.5 n/a n/a 17 42.6 43.5 0.9 18 35.5 37.2 1.7 19 35.9 38.9 3    The measured diagonal distance B (n=15) is compared to the distance B calculated from measurement of E with eq. 4.3 shown in Figure 4-6, and the Bland-Altman plot is shown in Figure 4-7. The bias is 0.8 mm and the 95% limits of agreement are -4.7 mm to 6.4 mm.  104 R2 = 0.9718 30 35 40 45 50 55 60 65 70 75 30 40 50 60 70 1.27 x distance E - 8.7mm di st an ce  B  in  m m  Figure 4-6 Distance B versus 1.27*distance E – 8.7 mm, i.e. the predicted distance B using eq. 4.3. There is a high correlation between distance B and the predicted distance B using eq. 4.3 because of the static geometry between transducer and needle guide.  105 -8 -6 -4 -2 0 2 4 6 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 subject di ffe re nc e B  a nd  p re di ct ed  B  Figure 4-7 Bland-Altman plot of the measured distance B versus the distance calculated from eq. 4.3, i.e. the predicted distance B.   The distance A (n = 13), which is the distance between the actual insertion point and the mark of the middle of the transducer at the pre-puncture examination, shown in Figure 4-5, is measured and shown in Table 4-4. Distance A is on average 34±6 mm, with the values ranging from 25 mm to 45 mm. The actual measurement of this distance on the transducer (Figure 4-2) from the middle of the transducer to the tip of the needle guide is 35 mm which agrees closely with the average of 34 mm.  106  Table 4-4 Distance A Subject Distance A (mm) 1 n/a 2 n/a 3 n/a 4 33 5 n/a 6 n/a 7 45 8 25 9 40 10 n/a 11 35 12 35 13 30 14 45 15 30 16 30 17 35 18 30 19 27.5 Average 34 Standand deviation 6   The actual needle insertion depth is also compared to the pre-punctured distance B+C+D and shown in Figure 4-8 and Bland-Altman shown in Figure 4-9.   107 R2 = 0.433 70 75 80 85 90 95 100 105 110 70 75 80 85 90 95 100 105 110 Actual needle insertion depth (mm) Ep id ur al  s pa ce  d ep th  B +C +D  (m m )  Figure 4-8 Epidural space depth along the needle guide line overlay (distance B+C+D) versus measured actual needle depth in the operating room.   108 -20 -15 -10 -5 0 5 10 15 0 5 10 15 20 subject number er ro r ( m m )  Figure 4-9 Bland-Altman plot of the epidural space depth along the needle guide line overlay (distance B+C+D) versus measured actual needle depth in the operating room. Most error is small except for 2 cases.  109 4.4 Discussion It was demonstrated here that the single-operator aim-and-insert real-time ultrasound- guided, lumbar paramedian epidural technique is feasible, with success on 18 of the 19 subjects. The only failure of 19 subjects was in the first 10 subjects, and the following cases were felt to be considerably smoother and faster after the operator became more familiar with the technique. We chose LOR as the endpoint, similar to previous studies on ultrasound guidance. We further use the presence of CSF or insertion of the catheter as additional evidence that the target had been reached with the needle, but admittedly these are still not perfect measures of successful placement. Our rates of absence of CSF in subjects where LOR was felt (4/18) are comparable to previous studies, with the added difficulty of performing the needle-through-needle technique in the paramedian plane [Cook-1999].  It is noted in previous studies that a CSE using a Tuohy epidural needle has a similar rate of failure: absence of CSF aspirations has been reported to happen 16% [Lyons-1992], 25% [Tanaka-2004] and 18% [Bluvol-2009].  The described technique has the advantage of being performed by a single operator, and by using a fixed angle guided needle instead of a free-hand technique. The fixed needle guide allows the needle to be held in the plane of the ultrasound image, and allows the operator to release their grip on the needle momentarily without losing its alignment in the plane. Conversely, the free-hand technique requires more operator expertise to see the needle tip in the ultrasound image [Bluvol-2009]. This arrangement also allows a standard needle and syringe to be used, compared to the need for the Episure needle used in the freehand technique described previously [Karmakar-2009].  110  Identification of the interspace level by palpation disagreed with the ultrasound in only 2 out of 19 subjects (11%), which is much lower than indicated in [Furness-2002]. It is still recommended that ultrasound guidance be used because previous studies [Furness-2002] have suggested that palpation misidentifies the actual space in 58% of the cases [Furness- 2002].  The curvilinear transducer head shape and size of footprint presented three practical problems.  Firstly, with the subject sitting flexed forward, a better image was obtained by pressing the transducer onto the skin surface as compared to having the transducer barely touching the skin. This caused a difference in depth to the ES of approximately 2.8 mm. A less curved transducer shape, with better skin contact enabling light pressure should reduce this practical error. This known source of error from transducer pressure, both considered by Grau and our group in Chapter 3, is shown to give distances approximately 2.8 mm shorter than the skin-to-LF distance without pressure.  Secondly, our projected needle tract was substantially longer than the more direct midline trajectory used in the blind technique as suggested by the 27% longer distance for distance B than E in eq. 4.3. The needle guide length also adds to the required needle minimum length. Our measured skin-to-ES distance was 88-111 mm requiring the use of a long 125 mm New Gertie Marx® CSE-Set epidural needle. Future developments will require a transducer with a smaller footprint to reduce the track distance through the subject’s tissues.  111  Thirdly, distance A, which is the vertical distance between the mark made from the center of the transducer face during the pre-puncture scan and the actual puncture site, (Figure 4-5) is on average 34 mm, ranging from 25 to 45 mm. The magnitude of this distance is important because the puncture site is 35mm below the desired intervertebral space, it necessitates the choice of L2-3 instead of L3-4 to avoid contact with the pelvic bones. This choice is entirely due to the limitation of the size of the transducer held parallel to the midline for real-time scanning.  Also, the variability of distance A shows that the pre-puncture ultrasound is useful but that ultimately, only real-time guidance can ensure the best site and angles as the transducer position found in the surgical preparation room is slightly different from the position found in the operating room by up to 10 mm. We believe the pre-puncture ultrasound gives the anesthesiologist time to familiarize and mark the approximate location of the transducer, thus saving time in the operating room but that ultimately, real-time ultrasound needs to be used to ensure successful needle insertion in one single attempt because the angle, site of insertion and depth at the prepuncture stage are different than in the operating room.  As shown in Figure 4-8 and Figure 4-9, the pre-puncture ultrasound measured distance B+C+D does not correlate well with the actual needle insertion depth, suggesting that the pre-puncture measurement is not sufficiently reliable to recommend it be used for guidance alone; loss-of-resistance should still be used as the needle approaches the  112 epidural space. Moreover, this shows that a real-time guidance system is needed as the prepuncture distance can be short or long by 15 mm.  The current protocol requires extra equipment (125 mm CSE set, probe cover, biopsy needle guide, biopsy bracket, ultrasound machine and curvilinear transducer), additional work for the nurse and the anesthesiologist (the ultrasound transducer must be covered by the sterile cover and the sterile biopsy guide must be attached prior to the procedure) and extra time for the pre-puncture orientation and skin marking. This extra time may be mitigated by a reduction in the number of needle insertion attempts. Also, as the injection of saline into the subcutaneous tissues obscures the image, the anesthesiologist should minimize the amount of saline injected prior to the needle approaching the LF. Needle visibility is not at the state where it can replace LOR because the epidural needle was observed in very few cases. Better needle visibility would be key to the success of this technique. We believe there are 3 reasons why the needle was not visible in most human subjects. First, there was a significant amount of local anesthetic injected (2-10mL) in the tissues prior to real-time ultrasound-guided epidural needle insertion. Local anesthetic usually contain microbubbles of air which reflect ultrasound, so injecting it in the tissues prior to ultrasound visualization will decrease visibility of underlying structures. The second reason is the specular reflectivity of the needle. Ultrasound beams reflect best when the angle of incidence is perpendicular to a surface. The needle is inserted at an angle of 23º and so the echo component reflected  back to the transducer may not be strong enough to be detected. This issue can be resolved by using more echogenic needles, such as those with commercially available coatings. The third reason is that the  113 needle may not always be perfectly in the plane of the ultrasound because the ultrasound transducer could still be moved after the needle is inserted. This in turn bends the needle causing it to be out of plane. Care must be taken to insert the needle without moving the ultrasound transducer throughout the procedure.  In conclusion, in 18 of the 19 subjects, a real-time aim-and-insert ultrasound-guided epidural needle insertion was successfully performed by a single operator. This includes overweight and obese subjects. This technique includes a systematic approach for both a pre-puncture scan and ultrasound guidance during needle insertion. Several limitations were also described, mostly related to the needle guide geometry. Current efforts are now underway on the design of a specialized transducer and needle guide to improve practicality, which would enable shorter needles, midline insertion and real-time guidance.  For successful guidance, the anesthesiologist relies on a good ultrasound image quality. It is often the case that the ultrasound image contains speckle noise, making it harder to depict anatomical structures and most importantly the LF. Moreover, as the trajectory is to be aligned with the LF for a proper aim-and-insert procedure, it is paramount for the anesthesiologist to be able to see the LF and the lamina, in case the LF is unclear. In the next chapter, spatial compounding with warping will be used to improve image quality of the LF and lamina.  114 5 Adaptive spatial compounding for ultrasound images of lumbar anatomy1 5.1 Introduction In Chapters 3 and 4, we used pre-puncture and real-time ultrasound to view the patient's lumbar anatomy and for the localization of the ES. However, ultrasound of the lumbar spine often shows an image filled with speckle and artifacts that can impede visualization of important and hard to detect features such as the LF and ES. Many post-processing methods employ filters to reduce speckle but all suffer to some degree from loss of fine details as they tend to reject content with high spatial frequency [Michailovich-2006].  There are many other techniques for improving ultrasound appearance [York-1999] [Cobbold-2007], but the spatial compounding technique is chosen in the present study since it can be implemented on standard commercial machines and is well-suited for the epidural detection problem. Spatial compounding consists of imaging at various angles. The interface between the LF and ES is usually only visible at certain angles depicted as a short horizontal line pair or “doublet” and is the target for needle insertion. The goal is to improve visibility of the epidural anatomy with a new adaptive spatial compounding technique. This chapter starts with a description of the key concepts, followed by extensions to the technique and performance evaluation and validation.   1 The material in this chapter has been published in [Tran-2009c]  115 5.2 Methods  Spatial compounding uses beam-steering [Berson-1981] [Carpenter-1980] [Jespersen- 1998], to capture several images of the same region by sending the ultrasound pulses at different angles of incidence. Since speckle noise is dependent on the distribution of reflectors along the ultrasound path [Trahey-1986b], changing the beam angle will also change the speckle noise pattern. Given different noise patterns but similar anatomical features, averaging these images will reduce the noise and enhance features.  Spatial compounding has been investigated previously as a way to improve other diagnostic tasks [Jespersen-2000] [Huber-2002] [Anderson-1997] by reducing speckle noise and improving the boundary continuity. Spatial compounding still suffers from blurring due to misalignment of images. The speed of sound varies by as much as 14% in soft tissue [Christensen-1988] and this, in turn, causes the apparent positions of structures to be slightly different under different angles of incidence. Moreover, the beam-steered images are captured as a sequence of images, so there is inevitable patient movement or sonographer movement within a set of images. Re-alignment of the features using an additional non-rigid registration (warping) [Krücker-2002] was first used for beam- steered images by our group [Groves-2004]. The result was a sharper ultrasound image but with significant additional computational time (40 ms on a dual core P4 2.8 GHz). That method was tested on an artificial gelatin cube phantom and muscle tissue. Building on those results, the warping/compounding method is extended here to improve visibility of the LF - ES interface. The previous work does not make use of the dependency of the distortion between blocks of close promixity. A reduction in computational cost is also  116 attempted. The previous algorithm [Groves-2004] used the normalized cross-correlation (NCC) for registration, radial-basis function interpolation and inverse mapping so these components form the starting point of this new work.  5.2.1 The need for registration  Warping is a registration technique introduced for spatial compounding by [Groves-2004] aimed at reducing blur of the spatial compounded image due to patient motion, probe motion and speed of sound and refraction errors. It is worthwhile to have an idea of how large of an error expected from each type of error. In our case, it takes about 0.5 s to steer the beam with the Ultrasonix research interface, so the 9 angles would take 4.5 s. During this time, there is breathing, patient movement and probe motion which cause some error.  The refraction errors are inherent to the beam-steering and we will derive a formula for the refraction error for a two-layer model where the top layer mimics fat with a lower speed of sound c1 and the bottom layer with a higher speed of sound c2 mimicking muscle. The beam-steering θi also affects the error as a larger angle causes greater refraction. This simple model should capture the majority of errors because fat is the tissue type with the most different speed of sound. The error from beam-steering in a two-layer model is derived both for a linear transducer and for a curvilinear transducer.  Here, the reference image is the image to which the other images will be registered and is chosen to be the image with beam angle 0o. There is no refraction error in the reference image because the angle of incidence is normal to the interface of the two layers but there  117 is still speed of sound errors. The beam-steered image is the image acquired with the ultrasound machine beam-steering set to a selected angle θ. There is refraction error in this case because the angle of refraction is related to the angle of incidence by Snell’s law. The apparent position is the position of an object with refraction and speed of sound errors. The true position is the actual position of an object without the refraction and speed of sound errors. The position of an object is denoted X = (x,y).  As shown in Figure 5-1, the beam comes from the top at an angle of incidence of θi. The apparent position is indicated as position X’ for the reference image, X” for the beam- steered image and the true position is at X. The speed of sound used by the machine to convert echo time into distance is ci=1540 m/s. The first layer has speed of sound of c1 and the second layer has a speed of sound of c2. For simplifying, the calculations and the schematic, we assume c1<ci<c2, which is expected because c1 is 1479 m/s for fat and c2 is 1566 m/s for muscle. ci is the speed of sound of the top-most rubber membrane of the ultrasound transducer, usually designed to be close to 1540 m/s. 1540m/s is the speed of sound assumed by the machine for image formation because it is the average speed of sound in tissues. There is variation of the tissues’ sound speeds, as reported by several labs [Hofer-2005] [Hill-2005], and will vary within and amoung patients, so the numbers given above are only to illustrate the nature of the problem.  118  Figure 5-1 Linear ultrasound transducer image of a two-layer structure  The angles of refraction can be calculated using the Snell-Descartes law.     i ic c  sinarcsin 11          eq 5.1        i ic c c c  sinarcsinsinarcsin 21 1 2 2       eq 5.2 and the distances l1 and l2 by trigonometric relationship 1 1 1 cos dl            eq 5.3 2 2 2 cos dl  .          eq 5.4  119 While the beam of sound traverses medium 1 for a distance l1, a distance of l1’ is the apparent distance, and while the beam of sound traverses medium 2 for a distance l2, a distance of l2’ is the apparent distance. If the speed of sound is lower than 1540 m/s (ci), then the apparent distance is longer than the actual distance [Salter-2008]. 1 1 ' 1 c cll i           eq 5.5 2 2 ' 2 c c ll i           eq 5.6  For a point at X, a beam steered at θi will yield the point at the apparent position X” of: ii llx  sinsin" '2'1           eq 5.7 ii lly  coscos" '2'1          eq 5.8 This same point at the position x will be observed in the reference image as being at position X’. There is no error in the x direction for X’because the angle of incidence is 0, so only speed of sound errors occur. 2211 sinsin'  llx           eq 5.9 2 2 1 1' c cd c cdy ii           eq 5.10  Then, the difference between the apparent positions will be seen as the amount of misalignment in the compounded image. The error is due to both the speed of sound and refraction error: ii llllxxx  sin'sin'sinsin"' 212211       eq 5.11  120 ii ii ll c cd c cdyyy  cos'cos'"' 21 2 2 1 1  .     eq 5.12 If we choose θi=10°, ci=1540 m/s, c1=1479 m/s, c2=1566 m/s, d1=10 mm, d2=40 mm (realistic values close to the case of lumbar anatomy), we get that the error between the beam-steered frame and the reference frame to be Δx=0.153 mm and Δy=0.0135 mm, which is approximately 0.5 pixel in the ultrasound image acquired.  Figure 5-2 shows the effect of increasing the fat thickness layer to the errors in the x and y directions.  Figure 5-2 Difference between the reference frame apparent position and the beam-steered frame apparent position for a point at the midline for a linear transducer  121 5.2.2 Refraction and speed of sound error in a curvilinear transducer The effect of refraction in a curvilinear transducer also exists. As seen on Figure 5-3, the point X is seen by an angle φ1 in the reference frame at position X”. The beam fired at angle φ1 following the green line through medium 1 with no refraction because the angle of incidence is normal to the surface of the transducer, and the beam is refracted at the interface between medium 1 and medium 2 toward the point X. This is the real position. The apparent position seen by the reference frame follows the dotted line to position X”, ignoring refractions at the interfaces. The steered beam is fired at angle φ2 following the red line in the transducer and the angle θi is added at the interface of the transducer and medium 1. This angle is refracted following Snell’s law. It is refracted again at the interface between medium 1 and 2. The apparent position seen by the beam-steered frame follows the dotted line to position X’, ignoring refractions at interfaces. The apparent positions X’ and X” will be different from the true position X due to refraction and speed of sound errors. As in the case of the linear transducer, calculations will be made to predict the magnitude of the error caused by refraction and speed of sound errors which will need to be compensated by the warping.  122 Φ θ θ 1 2 r d 2 d1 c c2 1 X’ X” X Φ1 2 θ i x 0  Figure 5-3 Curvilinear ultrasound transducer image of a two-layer structure  These two angles need to be calculated for each point in the image. The angles are obtained by solving for φ in these equations:    2211 tantancos1sin  drdrx       eq 5.13 21 ddry           eq 5.14 where the refraction is defined by Snell’s law         i ic c sinsin 111         eq 5.15      1 1 21 2 sinsin  c c          eq 5.16 x and y are the real (x,y) coordinates of the point X, d1 is the depth of the first layer, and d2 is the depth of the second layer. θ1 and θ2 are the angles of the beam in mediums 1 and 2 with speeds of sound c1 and c2. x is calculated as the sum of the horizontal distance the  123 beam travels in each medium, with x = 0 at the centre of the transducer. y is as the sum of the horizontal distance the beam travels in each medium, with y = 0 at the centre of the transducer.  Using the steepest descent optimization method, φ 1 (reference frame) can be calculated iteratively for θi=0, and φ 2 (reference frame) can be found for θi as the beam-steered angle.  Figure 5-4 shows how the two angles change with the fat layer thickness, for a beam- steering angle θi=10o, ci=1540 m/s, c1=1479 m/s (fat), c2=1566 m/s (muscle), d2=40 mm (muscle layer), the position of the point x = 20 mm from the middle and a transducer radius of r = 30 mm and varying d1. It is interesting to observe that φ 1+ θi ≠ φ2.   124  Figure 5-4 Angles φ 1 and φ 2 for a point at x = 20mm from the midline  When changing the position of the point to x = 0, the angles become as shown in Figure 5-5.  125  Figure 5-5 Angles φ 1 and φ 2 for a point at the midline (x = 0), φ 1=0.  The apparent position for the reference frame X’ would be a beam at angle φ 1 with no refraction and a speed of sound correction. The apparent position for the beam-steered frame X” would be a beam at angle φ 2 and θi with no refraction and a speed of sound correction (ci/c1 for medium 1 and ci/c2 for medium 2) as shown in the speed of sound correction for linear transducer.          2 22 1 1111 'tan'tancos1sin' c cd c crdrx ii      eq 5.17         2 2 1 111 cos1cos' c cd c crdry ii      eq 5.18 where, in the reference frame,  126 121 ''             eq 5.19 And for the beam-steered frame         2 22 1 1212 "tan"tancos1sin" c c d c c rdrx ii      eq 5.20         2 2 1 212 cos1cos" c c d c c rdry ii      eq 5.21 where i  221 ""          eq 5.22 The error between the two apparent positions is the amount of misalignment in the compounded image. '" xxx            eq 5.23 '" yyy            eq 5.24 For the case of fat and muscle, from these formulas, a relationship between the fat thickness and the error can also be obtained. Since the fat has speed of sound of 1479 m/s and muscle is 1566 m/s, the two mediums would have errors in opposite directions. There is a ratio of fat and muscles that would minimize the errors.  For typical parameters θi=10o, ci=1540 m/s, c1=1479 m/s (fat), c2=1566 m/s (muscle), d1=10 mm (fat layer), d2=40 mm (muscle layer), the position of the point x = 20 mm from the middle and a transducer radius of r = 30 mm, Δx = 0.155 mm (0.5 pixel) and Δy = 0.6269 mm (2 pixels).  Figure 5-6 shows how the errors depend on the fat thickness.  127  Figure 5-6 Difference between the reference frame apparent position and the beam-steered frame apparent position for a point at x = 20 mm from the midline for a curvilinear transducer.  And if we are looking at a pixel in the middle of the image, x = 0, the errors become as shown in Figure 5-7.   128 0 5 10 15 20 25 30 35 40 45 50 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 fat layer thickness er ro r i n (m m )   error in x error in y (mm) er ro r i n (m m )  Figure 5-7 Difference between the reference frame apparent position and the beam-steered frame apparent position for a point at the midline (x = 0) for a curvilinear transducer.  These calculations show that refraction and speed of sound errors represent a very small error but that each pixel has a unique alignment error. In addition to the refraction and speed of sound errors, since the images are acquired at a slow frame rate, there are probe motion and patient movement errors which can be expected to also contribute to feature misalignment. The warping algorithm can be used to calculate the vector field so each feature in a beam-steered image is mapped correctly to the same feature in the reference frame.  129  5.2.3 The registration and compounding algorithm  The beam-steered images are divided into blocks and each block is registered to the reference image. The reference image is the image to which the other images will be registered and is chosen to be the image with beam angle 0°. Once the individual warping vectors have been found for each block, each pixel is assigned a warping vector by interpolation of the block warping vectors such that a smooth warping transition occurs from one block to its neighbours. When comparing the different interpolation methods (bilinear, cubic, B-spline), performance was found to be very similar but for very different computational costs so cubic interpolation was chosen and no visible discontinuities were produced between blocks after interpolation. After interpolating the warping vectors, the individual pixels must be mapped following the warping vector field to the reference frame. Inverse mapping is used here. More detail on warping is shown in Appendix D.  In order to reduce computational cost, several approaches can be taken. A popular method is a coarse-to-fine or multi-resolution approach where lower resolution blocks are registered and then higher resolution blocks are registered using a smaller search region [Groves-2004]. This method, however requires larger scale features (visible at the coarse resolution), additional memory overhead, and does not make use of the trends in the warping vectors of neighbouring blocks. Linear prediction [Gersho-1992] is a standard technique in audio coding in which one would give an initial guess of the current value  130 based on a function of the previous values and calculate an error signal from this initial guess. The error signal is typically much smaller than the original signal. The same idea can be used in making an initial guess of the value of the warping vector of the current block based on surrounding blocks, under the assumption that the warping vector field is smoothly varying. The search region for the current block can then be reduced while still ensuring no discontinuities are created between blocks after interpolation. The key step is the choice of the first block to register. Setting the starting block at the center of the image and using linear prediction to reduce the search space by a factor 2 is tested here and abbreviated LP2. A second method, called LP2+, uses the Canny edge detector [Canny-1986] to detect areas containing edges; the block with the highest edge count is assumed to be a good candidate as starting point for the linear prediction algorithm. This block may be a better choice than the center block because a block with many strong features is less susceptible to misregistration.  After the set of beam-steered images are aligned in the reference coordinate system, the overlapping pixels are compounded to form a new image. With conventional spatial compounding, the images are averaged with equal weights but other techniques are possible [Wilhjelm-2000] [Behar-2006] [Shankar-1986]. A new gradient-based compounding method was proposed [Hor-2007a] whereby an image is considered to have high information content at a given pixel when the image gradient is large. Accordingly, any edges are weighted more than homogeneous regions. Many edge detection methods can be used. To avoid a response from relatively small-scale speckle noise, the edge information of a smoothed image is calculated using the Laplacian of a Gaussian. Here,  131 the key parameter is the size of the Laplacian of a Gaussian filter, which should be set according to the ultrasound machine and transducer characteristics since speckle scale depends on properties such as ultrasound frequency and imaging depth. By using this filter, a gradient-based compounded image is formed by calculating a weighted average of the pixels in different images, according to their edge strength    I i i I i ii gradient yxG yxpyxG yxp ),( ),(),( ),(        eq 5.25 where pi(x,y) and Gi(x,y) are the respective pixel value and edge strength of the image i at position (x,y). pgradient(x,y) is the weighted average of I beam-steered images at location (x,y).  This concept was further extended by introducing an adaptive median-based method [Hor-2007a] (labeled “median-based compounding”). I}i,),(|),({),(  thresholdii GyxGyxpyxP      eq 5.26     otherwiseyxp I yxPyxPmedianyxp gradient median ),( 2 ),()),((),(      eq 5.27 where P(x,y) is the set of all pixels pi(x,y) that have a Gi(x,y) that is larger than a set threshold Gthreshold. Gthreshold is set according to the speckle size that, in turn, depends on image scale and probe frequency. pmedian(x,y) is the adaptive median of location (x,y): the median of all points in P(x,y) if P(x,y) contains more than half of the number of images, otherwise it is pgradient(x,y). |P(x,y)| is the cardinality of the set P(x,y). With this approach, the pixel retains a sharper response near edges as it is no longer a weighted average there,  132 but retains the weighted averaging needed for more homogeneous regions. The median- based method builds on the gradient-based method with an added decision-making component: if a pixel combines mainly data with high feature content, the median of those points is used, otherwise, the gradient-based technique is used.  In summary, for the purpose of comparison, the “reference image” is defined as the image taken at beam angle 0o. In practice, the reference frame should be the frame with the most features so that registration of the beam-steered frames is correct. It is chosen as 0º in this work because the test images are captured by scanning with a beam angle of 0º to get the best view and capturing the 9 frames at angles around this best view frame. The “spatial compounding” method produces an image through simple averaging of the overlapping regions of the angle-corrected beam-steered images without any registration. The “average-based compounding with warping” is defined as the image resulting from the average of the registered images. The “median-based compounding with warping” is defined as the image resulting from adaptive median-based compounding of the registered images. Performance evaluation and validation are performed on both tissue- mimicking phantoms and human subjects.  The purpose of the tests on a spine phantom is to investigate the sensitivity of the parameter settings for warping/alignment in a controlled environment. For this purpose, a spine phantom was built to emulate the structure of the region of interest in the spine, in particular vertebrae L2, L3, L4.   133 A large plastic container was used as the base for the phantom to hold the simulated tissues and avoid reflections from the walls and bottom. To emulate the various tissues in the lumbar region, different materials were used. For the vertebra, plastic models were used since the hardness of plastic is sufficient to fully reflect ultrasound. To emulate fat and muscle, a mixture of agar gelatin, cellulose, glycerol and water was used [Burlew- 1980]. First, the muscle layer (3% agar, 2.5% cellulose, 20.7% n-propanol) was made, then the thin fat layer (3% agar, 1% cellulose, 2% n-propanol) was poured on top of the solidified muscle layer. Finally, a water-saturated coagulated mixture of rice flour was used to emulate the LF. These materials permit, at some angles, a clear double echo (doublet) of the interface between the LF and ES to be seen. Figure 5-8 is used to compare the phantom at L3-L4 to the image of a human subject at L3-L4. Although differences exist, the key bone and ligament features are sufficiently visible for evaluation.   a)  b) Figure 5-8 Comparison of human and phantom subjects. a) Ultrasound image of the L3-L4 of a human subject. b) Ultrasound image of L3-L4 of the spine phantom.   epidural space epidural space bone bone  134 The number of images was set to 9, spanning angles from -8o to 8o with a step of 2o. The block size and search region were two parameters that need to be optimized on the spine phantom and confirmed on real images. A set of images from each data set was warped using several sets of parameters. The normalized cross-correlation was used in conjunction with the Laplacian at the location of the doublet to find the best set of parameters.  Good alignment should produce sharp depictions of the LF and bone boundaries. In particular, the Laplacian of the ES and the gradient of the bone surfaces at the lamina of the vertebrae were calculated along vertical line profiles in the vicinity of the features. The Laplacian is used because the LF is a ridge so it is seen as a peak in the vertical profile. These are suitable measures since the epidural application requires localization of bones for choosing an appropriate needle puncture site and trajectory, and localization of the ES for choosing the depth of needle insertion. The bone is seen as a transition from a light to dark region, so the gradient was used for assessing the strength of the edge. The LF usually appears as two bright bars or doublet, so it is advantageous to detect a peak of the convolution with the Laplacian operator in the region of the LF. Moreover, the speckle reduction of each method was compared by calculating the SNR at bright and dark regions and also of the contrast-to-noise ratio (CNR) as defined in [Groves-2004]. The crosscorrelation coefficients for different parameter sets are shown for the spine phantom on Table 5-1. The highest crosscorrelation coefficient would indicate better alignment. The Laplacian was calculated on the LF of the spine phantom for different  135 parameter sets and shown in Table 5-2. Higher Laplacian means the ridge will be stronger. The optimal parameters are shown to be similar for both measures. Table 5-1 Values of crosscorrelation coefficients for the spine phantom. The NCC is computed over a small region of interest for several parameter sets. A higher NCC indicates a better alignment. The last row, labeled ‘Compound’, indicates the NCC when no warping is performed. Highest values are in bold. Search region Block size Horizontal Vertical 192 96 48 32 24 1 1 0.9223 0.9241 0.9244 0.9255 0.9245  2 0.9184 0.9258 0.9258 0.9252 0.9234  4 0.9203 0.9357 0.9357 0.9267 0.9277  8 0.9203 0.9357 0.9347 0.9185 0.9299 2 1 0.9291 0.931 0.9309 0.9309 0.9298  2 0.9255 0.9329 0.9329 0.9322 0.9275  4 0.9295 0.9445 0.9445 0.9445 0.9299  8 0.9295 0.9445 0.9429 0.9251 0.9388 4 1 0.9382 0.94 0.94 0.9413 0.9394  2 0.9355 0.9423 0.9423 0.9404 0.9436  4 0.9425 0.9566 0.9567 0.9552 0.9413  8 0.9425 0.9566 0.9555 0.9551 0.9429 8 1 0.9364 0.9377 0.9423 0.9456 0.9425  2 0.9368 0.9374 0.9439 0.9485 0.9482  4 0.9441 0.9436 0.9496 0.9514 0.9486  8 0.9343 0.9415 0.9495 0.9466 0.9476 16 1 0.9267 0.9265 0.9422 0.9409 0.9497  2 0.9281 0.9284 0.9437 0.9322 0.9467  4 0.9343 0.9324 0.9486 0.9466 0.9467  8 0.9343 0.931 0.9482 0.9421 0.9505 Compound 0.924   136  Table 5-2 Maximum Laplacian at the LF for the spine phantom. The Laplacian is computed over a vertical line profile over the LF for several parameter sets. A higher Laplacian indicates a clearer feature. The last row, labeled ‘Compound’, indicates the Laplacian when no warping is performed. Highest values are in bold. Search region Block size Horizontal Vertical 192 96 48 32 24 1 1 80 80 69 67 66  2 110 110 110 107 130  4 208 187 215 239 201  8 208 187 216 205 202 2 1 102 105 79 62 88  2 119 119 104 110 110  4 218 216 227 222 178  8 212 216 232 199 199 4 1 98 119 120 110 75  2 141 152 153 139 107  4 227 219 230 225 171  8 227 219 234 237 187 8 1 60 81 116 102 80  2 141 130 130 151 108  4 203 190 195 181 157  8 203 177 193 187 162 16 1 72 69 112 85 97  2 89 117 127 149 134  4 181 137 198 179 179  8 199 140 199 189 191 Compound 63   The reference image, spatial compounded image, average-based compounding with warping, and median-based compounded with warping (with and without LP2 and LP2+) were compared. Computational costs of each of the key processing steps were recorded.  Ultrasound images were acquired using an Ultrasonix 500RP with a 38 mm linear 4-9 MHz probe with a depth setting of 60 mm. The scan-converted captured B-mode images  137 (192 × 384 raw data becomes a 192 × 304 scan converted image) had a resolution of 5 pixels/mm.  Phantom tests were followed by a study on human subjects. The purpose of these tests is to evaluate performance and validate the algorithm in vivo. The study was approved by the Clinical Review Ethics Board of the University of British Columbia and the British Columbia Women’s Hospital and Health Center (C05-0409).  Written informed consent was obtained from all subjects. Pregnant women at term in active labour or scheduled for elective cesarean section were included (n=23). Subjects who had contraindications to neuraxial anesthesia or who could not communicate in English were excluded.   The same tests and quantitative measures used on the spine phantom were used. The block size and search region were used based on the results from the spine phantom and further fine tuned for human subjects. The quantitative measures were performed on all 23 sets of data at the region of interest (doublet and lamina) and averaged. Ultrasound images (ranging from 190 × 192 to 190 × 384 pixels deep depending on the imaged depth become a 726 × 423 pixel scan converted image) were acquired using an Ultrasonix 500RP with a 40 mm curvilinear 1-5 MHz probe with a depth setting of 60-120 mm, depending on the subject. The captured B-mode images had a resolution of 3.2 pixels/mm. The spatial compounding methods were implemented in the C language on a 3.0 GHz Intel Pentium 4 processor with 1 gigabyte (GB) of random access memory (RAM).  138  5.3 Results  In the spine phantom, it is observed that a block size of 48 × 48 and a search range of ±4 × ±4 gives the highest NCC and Laplacian of the LF value between blocks of the reference image and blocks of the warped beam-steered images. This translates to a block size of 9.6 mm and a maximum warping of 0.8 mm vertically and horizontally. These parameters are used for the rest of the spine phantom tests.  Figure 5-9 shows different compounding methods using this search region and block size. The reference image shown in Figure 5-9a has sharp edges, however, it has substantial speckle noise. The spatial compounded image shown in Figure 5-9b has much less speckle but at the cost of blurred features – the LF doublet is almost indiscernible. The average-based compounding with warping is shown in Figure 5-9c. One can see that the speckle is reduced and that the features remain sharp. Finally, the median-based compounding with warping is shown in Figure 5-9d. The image is slightly sharper and still retains good speckle reduction. It should be noted that a relatively clear depiction of the doublet is achievable in the reference image through careful probe positioning on the phantom, but this is more difficult on human subjects as described later.     139  a)  b)  c)  d) Figure 5-9 Adaptive spatial compounding on spine phantom. a) Reference image. b) Spatial compounded image. c) Average-based compounding with warping. d) Median-based compounding with warping.  Starting with median-based compounding with warping, LP2 is used to reduce computational cost in Figure 5-10. The LF doublet is blurred (comparing Figure 5-10a to Figure 5-9d). When using the feature detector to choose a more appropriate starting point for linear prediction (LP2+), the resulting image has a sharper doublet than with LP2 (comparing Figure 5-10a and Figure 5-10b). Here, the benefit of using the feature detector to find the starting block for the linear prediction algorithm is clear.   140  a)  b) Figure 5-10 Adaptive spatial compounding with linear prediction on spine phantom. a) Median- based compounding with warping and LP2. b) Median-based compounding with warping and LP2+.  As seen in the methods, finding the right parameters can affect the performance of warping. The parameters are adjusted slightly again on human subjects, with a procedure similar to the phantom, on a sample data set. The NCC is calculated on a human subject, from 4 images to the reference frame, and shown in Table 5-3. The Laplacian of the LF is shown in Table 5-4.  141  Table 5-3 NCC coefficients for a human patient for four selected beam-steered images  Block size Horizontal×Vertical 193 96 51 31 26 4×2 0.983 0.983 0.983 0.986 0.984 8×2 0.983 0.983 0.983 0.986 0.985 8×4 0.984 0.983 0.983 0.986 0.985 4×2 0.99 0.99 0.99 0.989 0.988 8×2 0.989 0.989 0.99 0.99 0.988 8×4 0.989 0.987 0.99 0.99 0.989 4×2 0.983 0.983 0.982 0.982 0.982 8×2 0.983 0.984 0.984 0.984 0.986 8×4 0.983 0.983 0.984 0.984 0.985 4×2 0.982 0.982 0.983 0.984 0.983 8×2 0.98 0.982 0.983 0.983 0.983 8×4 0.98 0.98 0.982 0.981 0.982  Table 5-4 Maximum Laplacian of line profile of LF for a human subject. ‘Compound’, indicates the maximum laplacian when no warping is performed.  Block size Horizontal×Vertical 193 96 51 31 26 4×2 123 114 101 87 83 4×4 110 102 106 98 85 8×2 118 112 139 126 99 8×4 126 112 136 120 93 Compound 87   Table 5-5 and Table 5-6 list the quantitative measures on the LF and bone boundaries for the spine phantom and human subjects respectively and Table 5-7 and Table 5-8 the values for SNR and CNR for the spine phantom and human subjects respectively. It is observed that warping significantly improves the Laplacian value at the LF from 76 to 180 (p<0.05). When using LP2, the value of the Laplacian falls significantly from 180 to 107 (p<0.05), but when using LP2+, the Laplacian increases to 154. Similarly, the gradient of the bone boundaries changes significantly (p<0.05) from 40 to 62 to 45 to 54 for standard spatial compounding, adding warping, adding LP2, and adding LP2+  142 respectively. For median-based compounding with warping the maximum Laplacian is 192, decreasing to 158 when using LP2 and rising to 162 when using LP2+. The median- based compounding is seen to significantly outperform the average-based compounding with a Laplacian value of 192 compared to 180 (p<0.05) respectively, and a bone gradient of 66 compared to 62 (p<0.05).  Table 5-7 and Table 5-8 show a large increase in SNR and CNR with compounding. There is a slight reduction when adding warping, but SNR and CNR are still significantly above the reference image (p<0.05).  The human subject sample biometrics comprised ages of 35±4 years, weights of 76±15 kg, heights of 161±7 cm and epidural depths of 51±11 mm. The set of subjects included both easy and difficult patients as assessed by the anesthesiologist and sonographer.   Since the image sizes for the human subjects are slightly different than the spine phantom, the selection of block size and search range is repeated based on the spine phantom results and a block size of 51 × 51 (15 mm × 15 mm) with a search region of ±8 × ±2 (±2.5 mm × ±0.6 mm) is seen to give best NCC and Laplacian at the LF.  Compounding by itself gives a clearly visible improvement to the appearance of the LF – ES interface or doublet. The appearance of the doublet is barely detectable in Figure 5-11a (the reference image) being embedded in speckle, and clearer in Figure 5-11b (the  143 spatial compounded image). When adding warping in Figure 5-11c, the features become slightly sharper. And finally, the adaptive median-based compounding with warping shown in Figure 5-11d further sharpens the features. If linear prediction is used to reduce the computational cost, the image sharpness remains approximately the same for both LP2 and LP2+.  144   a)  b)  c)  d)  e) Figure 5-11 Adaptive spatial compounding on human subject. a) reference image b) spatial compounded image c) average-based compounding with warping d) median-based compounding with warping e) the full ultrasound image showing the region of interest. The doublet appears slightly below the middle of the region of interest.  The Laplacian at the location of the LF and the gradient at the bone boundaries are calculated on the spine phantom (Table 5-5) and on 23 sets of human subjects (Table  145 5-6). The result is a significantly higher Laplacian and gradient (p<0.05) for average- based compounding with warping over simple compounding, and significantly higher for adaptive median-based compounding with warping over average-based compounding with warping. Again LP2 and LP2+ have no significant detrimental effect.  Table 5-5 Maximum Laplacian at the LF and gradient of the lamina for the spine phantom (mean±standard deviation)  Table 5-6 Maximum Laplacian at the LF and the gradient of the lamina for human subjects (mean±standard deviation)   The SNR for light and dark regions and CNR of every compounding method is significantly higher (p<0.05) than the reference image in the spine phantom (Table 5-7) and on 23 human subjects (Table 5-8). The SNR and CNR are ordered from the highest Method Laplacian of LF Gradient of bone boundaries Reference image 275±67 78±14 Spatial compounding 76±14 40±6 Average-based compounding w/ warping 180±52 62±8 Median-based compounding w/ warping 192±53 66±10 Average-based compounding w/ warping w/ LP2 107±20 45±5 Median-based compounding w/ warping w/ LP2 158±36 51±10 Average-based compounding w/ warping w/ LP2+ 154±42 54±6 Median-based compounding w/ warping w/ LP2+ 162±31 62±15 Method Laplacian of LF Gradient of bone boundaries Reference image 211±59 59±14 Spatial compounding 132±39 45±16 Average-based compounding w/ warping 148±40 47±16 Median-based compounding w/ warping 163±38 51±17 Average-based compounding w/ warping w/ LP2 147±38 47±17 Median-based compounding w/ warping w/ LP2 159±36 53±16 Average-based compounding w/ warping w/ LP2+ 163±44 50±14 Median-based compounding w/ warping w/ LP2+ 155±35 51±16  146 to lowest as follows: compounding without warping, average-based compounding with warping, median-based compounding with warping and LP2+, average-based compounding with warping and LP2+, median-based compounding with warping and, finally, the reference image. The SNR is slightly higher when using LP2+ than LP2.  Table 5-7 SNR for light and dark regions and CNR for the spine phantom  Table 5-8 SNR for light and dark regions and CNR for human subjects (mean±standard deviation)    Table 5-9 summarizes the computational time. Standard spatial compounding by averaging adds very little extra cost. Adding warping is among the most costly operation (94.6 ms) because of the calculation of the cross-correlation coefficients and exhaustive search through the search region. This reduces the maximum frame rate to ten images per Method SNR light SNR dark CNR Reference image 6.08 1.34 4.08 Spatial compounding 12.23 1.91 6.93 Average-based compounding w/ warping 9.98 1.75 6.09 Median-based compounding w/ warping 9.35 1.68 5.71 Average-based compounding w/ warping w/ LP2 11.49 1.87 6.71 Median-based compounding w/ warping w/ LP2 10.38 1.79 6.13 Average-based compounding w/ warping w/ LP2+ 11.49 1.80 6.62 Median-based compounding w/ warping w/ LP2+ 10.39 1.74 6.10 Method SNR light SNR dark CNR Reference image 6.0±1.0 4.5±1.5 1.9±0.5 Spatial compounding 7.4±1.6 7.5±2.8 2.4±0.7 Average-based compounding w/ warping 7.2±1.5 7.4±2.8 2.4±0.7 Median-based compounding w/ warping 7.0±1.4 6.8±2.5 2.3±0.6 Average-based compounding w/ warping w/ LP2 7.1±1.7 7.2±2.7 2.3±0.8 Median-based compounding w/ warping w/ LP2 7.1±1.5 6.6±2.4 2.3±0.6 Average-based compounding w/ warping w/ LP2+ 7.3±1.8 7.3±2.7 2.3±0.8 Median-based compounding w/ warping w/ LP2+ 7.1±1.7 6.6±2.4 2.3±0.8  147 second which borders on real-time implementation. When using linear prediction, reducing the size of the search region by a factor of two in each direction (LP2), the cost of finding the warping vectors is reduced (27.4 ms). The cost of using the Canny edge detector is negligible so LP2+ is almost as fast as LP2. The highest cost is with the adaptive median compounding (164 ms). This is mainly due to the sorting step. It is possible to implement a sorting algorithm on the field programmable gate array (FPGA) portion of the Ultrasonix system or using more clever ordered data structures to achieve real-time performance.  Table 5-9 Computational costs. Central processing unit (CPU) time is calculated using an Intel P4 3.0GHz with 1GB RAM personal computer, an image of 192 × 304, a block size of 96 × 96 and a warping search region of ±4 × ±4 pixels. LP2 indicates warping with linear prediction and a reduction of the search space by a factor of two. Method Time (ms) Image resizing and angle correction 1.8 Averaging images 4.2 Interpolating warping vectors 1.6 Remapping the image 1.3 Canny edge detection 1.8 Finding warping vectors 94.6 Finding warping vectors with LP2 27.4 Adaptive median compounding 164.6  5.4 Discussion The reference image Figure 5-9a shows a high quality LF and bone surface. This is because the reference image in this case was deliberately chosen to have maximum clarity of these structures for the purpose of comparison. However, the beneficial effect of spatial compounding with warping is even more apparent when comparing with frames that are not the reference frame and thus represent more typical images captured without  148 such careful alignment of the reference beam angle to the doublet. Figure 5-12 shows the sensitivity of the echo from the LF to the angle of the ultrasound images taken at angles differing by 4° but on Figure 5-12a the doublet is clear whereas on Figure 5-12b, the doublet is not visible. This means a good quality standard ultrasound image of the target epidural anatomy is more difficult to quickly achieve than the adaptive spatial compounded image. This issue of ease-of-scanning is especially important in epidural anesthesia since the operators are typically non-specialist in ultrasound (anesthesiologists).  149   a)   b) Figure 5-12 Beam-steered images of a difficult to image patient in the first clinical study. a) Image showing a distinct echo of the LF. b) Image without distinct echo. The difference in beam angles between (a) and (b) is 4°.   150  Inspecting the Laplacian at the LF, the LP2+ is seen to outperform the LP2 in the spine phantom, but it is not the case for human subjects, where the LP2 and LP2+ performance are very close. The sonographer generally centred the structures of interest in human subjects, which causes the LP2 to be closely related to the LP2+. Also, the LF doublet is the structure of interest but the structure which is easiest to observe and most likely to get best registration are the bony lamina, where LP2+ would initiate the registration.  In summary, the best choice including all considerations of the described methods for scanning human subjects is median-based compounding with warping and linear prediction using the feature detector (LP2+) to choose a starting point for block matching. It should be noted that these differences, although significant, are small as evidenced by inspecting the images directly. The small improvement is still extremely important for the epidural anesthesia application since, in our experience, detection of the ES in a difficult patient is the most challenging aspect and a small improvement in quality is valuable. It is not uncommon to get subjects where the doublet is barely detectable. In general terms, the use of adaptive compounding methods not only gives a small improvement to overall image quality, but allows the tradeoff between different quality measures to be selected.   151  5.5 Conclusions Spatial compounding is shown to improve the key aspects of image quality in this application of spinal imaging. This is especially apparent in human subjects. The reason is that the ES is only visible from some angles, and perception of the characteristic doublet is limited by noise. The reference image in comparison, although having a high gradient and Laplacian for the features, has a low SNR which impedes detection of the features, so the doublet may look like speckle. Adding the intermediate step of warping to the compounding algorithm further improves image quality. The spine phantom and human subjects had a substantial fat layer, patient motion and/or transducer movement thus requiring warping. When adding LP2+, the computational cost reduces to levels permitting real-time operations while maintaining good sharpness and speckle reduction. As can be seen from the results on the spine phantom and human subjects, the median- based compounded warped images gives the sharpest overall images while retaining good speckle reduction. This is to be expected as pixels from images without features may be omitted through the median calculation and thus do not obscure the feature. The gradient and Laplacian measurements of image quality show significant improvements over the other methods.  An automatic algorithm for choosing a reference frame should be developed as when using spatial compounding in real-time, there is no guarantee that the frame with the most visible features is the frame at 0º.   152 The methods described here can be extended to other clinical applications by following the same paradigm: - implement spatial compounding with beam-steering - pick an appropriate measure of image quality for a specific clinical application - choose warping parameters and method of compounding based on optimization of the image quality measure. This approach can be easily implemented in real-time on the latest generation of open architecture ultrasound machines. The end result can be a significant improvement in image quality for the specific anatomical feature.  Even with an image containing less speckle noise and enhanced features, detection of the LF can be non-trivial. In the next chapter, an algorithm for automatically detecting the lamina and LF is presented.  153 6 Automatic detection of ligamentum flavum in ultrasound images of the lumbar anatomy1 6.1 Introduction As seen in Chapters 3 and 4, being able to recognize the LF is crucial both for prepuncture planning and for real-time ultrasound guidance. In Chapter 5, an image enhancement method based on spatial compounding was used to emphasize the gradient of the lamina and Laplacian of the LF, but even with these improvements, the structures can be hard to depict for the inexperienced anesthesiologist. To ensure a proper paramedian plane, the lamina and LF need to be seen and recognized in the image. Moreover, needle insertions are performed in a sterile environment, and the anesthesiologist needs to measure relevant distances (skin-to-LF depth) while preserving sterile conditions, so interaction with the ultrasound controls must be minimal.   A fully automatic method for detecting the laminae and LF would be useful for helping the anesthesiologist choose the proper paramedian plane and measure the skin-to-LF depth that represents the desired needle insertion depth from the skin surface. Such an automatic method is developed based on the way the anesthesiologist identifies the features in the lumbar anatomy in the ultrasound image and recognizes the lamina and the LF. From the detected features, the skin-to-LF depth can be calculated.   1 The material in this chapter has been published in [Tran-2010b] ( © 2010 IEEE)  154 6.2 Methods 6.2.1 Overview of algorithm Figure 6-1 shows an overview of the algorithm used to estimate the skin-to-LF depth. The first step is to acquire beam-steered images and use spatial compounding with warping and combine them using the adaptive median method described in Chapter 5. The purpose of compounding here is to improve the uniformity of the lamina and LF. Phase symmetry [Hacihaliloglu-2006] is then used to extract the lamina and LF from this enhanced image in a ridge map. Then, a template matching algorithm is used to segment the laminae from the ridge map. The LF is subsequently segmented from the image using another template around the lamina. The skin-to-LF depth is finally measured. Each step is now explained in detail.   Figure 6-1 Overview of the LF detection algorithm  6.2.2 Ridge map generation For bone detection in ultrasound, Canny [Canny-1986] and other gradient-based methods are sensitive to the choice of parameters and thresholds [Kovesi-1999] [Hacihaliloglu- 2009]. In particular, there is a need for a technique independent of image intensity because the ultrasound intensity changes with the time gain compensation curves and image depth parameters selected by the operator.  155  Phase is suggested to often contain more information than magnitude [Oppenheim-1981]. Recently, phase techniques have gained popularity as they perform well at detecting bones in ultrasound images [Hacihaliloglu-2006]. As the lamina and LF have the appearance of a ridge, phase symmetry is used because it is a phase-based measure designed to detect ridges [Kovesi-2006].  Phase symmetry is defined as              n n n nn xA Txoxe xPS )(         eq 6.1 where en(x) is the even part of the signal and on(x) is the odd part of the signal produced by the Log-Gabor filter (eq. 6.5), ε is a small constant to prevent division by 0, T is a threshold used to offset the phase symmetry by the expected noise value RRT             eq 6.2 where η is a user defined variable, μR and σR are the mean and standard deviation of the Rayleigh distribution of the noise energy response. We are interested in the distribution of magnitude An(x) of the energy vector [Kovesi-1997]:      22 xoxexA nnn          eq 6.3          onennn MxIMxIxoxe  ,, .      eq 6.4  The Log-Gabor filter G(ω) is used to extract the even and odd parts to be used in the phase measures and is defined as:  156            2 2 /log2 /log exp)( o o k G   ,       eq 6.5 and          GFM GFM o n e n 1 1 Im Re              eq 6.6 where k/ωo is a ratio to be held constant for constant-shape ratio filters which defines the bandwidth of the Log-Gabor filter, and ωo is the filter’s center frequency. G(ω) is the Log-Gabor filter and F-1(G(ω)) is the inverse Fourier transform of G(ω).  The Log-Gabor filter is shown in Figure 6-2a. To define the direction and spread of the filter to be used to extract the even and odd parts of the image, an additional filter is applied to the Log-Gabor filter. The directional spread is defined by an angle of the orientation φ and an angular bandwidth θ/σ. The resulting filter is shown in Figure 6-2b.  157   a)  b) Figure 6-2 Log-Gabor filter in frequency domain a) original Log-Gabor filter built using ω0=1/15 k/ω0=0.65 b) Log-Gabor filter after a directional spread of angle φ=2π/3 and an angular bandwidth of θ/σ=1.5.   The 192 line pre-scan converted images of parturient patients were acquired with a resolution of 0.3125 mm/pixel and a depth varying between 60 mm and 120 mm depending on the subject. By performing the autocorrelation in the axial direction of a small block containing only speckle, we obtained a speckle size of approximately 7-9 pixels. The minimum wavelength of 15 pixels was used to avoid response from speckle at different angles. We used 3 scales with a multiplier of 2.1 between the wavelengths of each scale. The Log-Gabor filter was designed with a k/ω0 (frequency spread) of 0.65 and θ/σ (angular spread) of 1.5. These parameters are determined experimentally to result in minimal overlap necessary to achieve even spectral coverage using a Log-Gabor filterbank. The value of η used for noise threshold is also determined experimentally and is set to 5 [Kovesi-1999].  For bone detection in ultrasound, these parameters are similar to those used previously in other applications [Hacihaliloglu-2006] so the algorithm does not require parameter tuning to a specific subject.  158  The Log-Gabor filter was run on a test image of the lumbar anatomy and it was observed that some filter orientations did not contain information on the ridges of interest (lamina and LF) as the features mostly appeared at specific orientation angles (near oblique to the ultrasound beam). Only three orientations (90°, 120° and 150°) (shown in Figure 6-3) are needed.  Figure 6-3 Orientations of the Log-Gabor filter used for phase symmetry (90°, 120° and 150° corresponding to the blue arrows)   In the ultrasound image, the lamina and LF can also be seen by their bright edges relative to the muscle, creating high intensity ridges. The phase symmetry is multiplied with the amplitude of the image and is called the ridge map (Figure 6-4).  Figure 6-4 Flow chart of the ridge map generation  159   When applying phase symmetry at these three angles to the ultrasound image in Figure 6-3, the phase symmetry result is shown in Figure 6-5. As can be seen, there are symmetric parts of the image which do not correspond to features. Because phase symmetry is a ridge detector that is independent of image intensity, a pixel can have high phase symmetry and low intensity. We define a bone or LF as having high phase symmetry and intensity. As shown in Figure 6-4, the ridge map is defined by the phase symmetry combined with the image intensity. The ridge map is shown in Figure 6-6.  Figure 6-5 Phase symmetry response using the 3 angles of interest using eq. 6.1.   Figure 6-6 Ridge map obtained by combining the phase symmetry response with the image intensity.  160 6.2.3 Lamina extraction The lamina and LF need to be extracted from the ridge map. The lamina is defined geometrically by a slightly curved line. Because of the variable curved nature of the lamina that depends on each patient and transducer placement, a template matching with a wide angle capture range approach is used in this work.  A diagonal line of angle matching the average of the lamina angles is blurred by a Gaussian filter where the angle of the diagonal line has a Gaussian shape with standard deviation depending on the expected range of lamina angles. This is determined by manually measuring the angles of the ridge map from Chapter 3 (20 subjects with two data sets per subject, except for one subject, giving n = 39). The angles for the lamina are measured to be 32° (μlam) with standard deviation 11.2° (σlam).        2 2 2 exp 2 1),( lam lamlam lam lamrF      ,     eq 6.7 Additionally, the diagonal angular Gaussian blurred filter is blurred in the direction perpendicular to its length in order to widen the capture range. F(r,θlam) is the function defining the angular Gaussian blurred filter, μlam and σlam  are the average and standard deviation of the angles of the laminae. θlam is the angle in the lamina template defined in polar coordinates equivalent to θlam = tan-1(y/x). r is the radius from the center but is not used in the formula.  S(x,y) is the function defining the Gaussian blurring filter [Fan-2001], μthick is the average thickness of the laminae in the ridge map.  161       2 2 2 exp 2 1, thickthick yyxS  .            eq 6.8 S(x,y) is rotated by a rotation matrix                 y x y x o lam o lam o lam o lam 90cos90sin 90sin90cos ' '        eq 6.9 so it follows the direction perpendicular to the angular Gaussian blurring filter F(r,θlam) and becomes S(x’,y’). μthick is measured to be 2.1 pixels from extracting the ridge map of a wire phantom and measuring the thickness of the response from a point, assuming the response to be Gaussian shape, 2 σ  (or 2μthick in this application) would be the point where the response falls to 5%. This is the lamina template    ',', yxSrFT lamlam   .        eq 6.10 The lamina template size is 41 × 41, which is sufficiently large to capture a single lamina in adults. The template is shown in Figure 6-7a.   This template is matched to the image using the Pearson correlation [Filev-2005] such that a lamina similarity map is generated. An image can contain more than one lamina. From knowledge of adult anatomy and previous experience with ultrasound of the spine acquired in the clinical trials in Chapters 3 and 4, the adult subjects in this study have at most three laminae in the ultrasound image with a skin-to-LF depth ranging from 30 mm to 80 mm. To provide a marginal error, up to five laminae candidates are allowed by the algorithm. The areas of highest correlation in the lamina similarity map are chosen as the potential locations of the laminae: the highest similarity point in the image is chosen as the location of the first lamina, then, as the distance between two laminae is  162 approximately 30 mm, the search region for the second lamina is set to at least 20 mm away from the first lamina location. This is repeated for the maximum number of possible laminae allowed by the algorithm (five in this study). The search for the lamina locations with an example is depicted in Figure 6-8.  163   a)  b) Figure 6-7 Templates for the detection of the lamina and LF  a) template for the detection of the lamina (41 × 41 pixels) composed of a line with an angular Gaussian blur and a diagonal thickness blur b) template for the detection of the LF (21 × 21 pixels) composed of a line from the corner to halfway through the template so to detect an end of the line, the line is also blurred with an angular Gaussian blur and a diagonal thickness blur.   164      a) b) Figure 6-8  Search for lamina a) Flow chart of the search for the laminae, b) steps of the search for laminae from top row to bottom: choose the point of highest lamina correlation, then set the area around that point to zero, choose the next point of highest correlation. The circles denote the positions of the three first detected laminae (red is first and strongest lamina detected, green is the second and blue is the third lamina detected).  6.2.4 LF extraction For manual detection, after finding the lamina, the sonographer mentally locates the LF by following the lamina down and searching for the LF around the tip of the lamina because the LF attaches two laminae together. For each lamina, there is one potential LF at the base. The automated algorithm follows a similar approach. The LF template is  165 similar to the template for the detection of the lamina in that it is a line with Gaussian cross section to have a wide capture range of angles. However, the LF is often a line of different thickness and angle from the lamina. The angles for the LF of adults are measured to be 13.2° with a standard deviation 12.7° on the ultrasound images (Chapter 3, n = 39). σthick is set to 2.1 pixels as well. Moreover, the template for the LF needs to locate the end of the ridge because the sonographer typically follows the lamina until the end of the line. A different template is used and is shown in Figure 6-7b. The crosscorrelation RLF of the LF template with the ridge map is only computed for pixels around the lamina. To further emphasize the end of the line rather than any section of the lamina, the LF detector looks for the point with the largest difference between RLF and Rlam defined as       jiRjiRyxLF lamLFjik ,,maxarg, ,        eq 6.11 where LFk(x,y) is the location of the kth LF. An example showing the search for the LF is shown in Figure 6-9. The template size is 21 × 21 for the LF template, which is sufficient for the LF. The position of the lamina is detected and circled in red. The RLF-Rlam is shown superimposed on Figure 6-9b. As expected, the position of strongest RLF-Rlam is at the base of the lamina and is chosen to be the position of the LF.  166   a)  b) Figure 6-9 Search for LF a) Flow chart for the search of the LF k,  b) the region below the position of the lamina is correlated with the lamina template and the LF template and the highest point of the difference between the two correlations is chosen as the position of the LF.     167 6.2.5 Performance measures The correctness of the LF detection was first evaluated manually by visual inspection of the results. The LF was considered detected successfully if the detection was on the body of the LF.  In order to assess the accuracy of the algorithm, the calculated distance from skin to LF was compared to three other measures: manual segmentation of the same ultrasound image, independent sonographer prepuncture examination measurement on a different image and the actual needle insertion depth measured by the anesthesiologist. The average error, root-mean-squared (RMS) error and Bland-Altman 95% limits of agreement were calculated on correctly identified LF. Although the images can contain three LF, only the measurements for the most central LF were used because it is the one measured by the sonographer and to which the anesthesiologist inserted the needle.  Finally, the computational cost was measured. The algorithm was written in Matlab (The Mathworks, Natick, MA, USA) except for the spatial compounding which was written in C++ as described in Chapter 5. The algorithm was tested on images of size 262 × 500 pixels, searching for 5 potential LFs. The code was run on an Intel Core 2 Quad CPU at 2.83 GHz with 3.25 GB of RAM.  6.2.6 Statistical significance The paired t-test was used to assess the statistical significance of the difference between the skin-to-LF depth measured by the algorithm, manually, by the sonographer and the  168 needle insertion depth. On those measurements, the omnibus K2 test was first performed [D’Agostino-1971] to verify normality of the data.  6.2.7 Clinical trial Ethics approval was obtained from the Clinical Review Ethics Board of the British Columbia Women’s Hospital and Health Center (C05-0409) to perform ultrasound scans on twenty parturients in labour or scheduled for cesarean delivery. Informed written consent was obtained for all subjects. Two sets of images were recorded on L2-3 and L3- 4 on each subject except for one case in which only L2-3 was recorded (n = 20 × 2 – 1 = 39) due to time constraints. The sonographer was unable to measure the skin-to-LF depth on 1 of 39 cases, so 38 measurements are available.  The exclusion criteria were the usual contraindications to neuraxial anesthesia and the inability to speak English. Four subjects (1-4) went through labor and 16 subjects (5-20) went through Cesarean delivery, for which a combined spinal-epidural anesthesia was administered. Needle insertion was guided by the LOR technique, using saline and continuous pressure on a glass syringe plunger.  The same anesthesiologist (Dr Allaudin Kamani) performed all epidural needle insertions. Scanning and data capture were performed by an experienced Registered Diagnostic Medical Sonographer (Victoria Lessoway) using an Ultrasonix RP500 and a 1–5 MHz broadband curvilinear transducer (Ultrasonix Medical Corp., Richmond, BC,  169 Canada). The sonographer performed the measurements directly on the ultrasound image and performed the image capture for the offline processing at a later time.  The images were first enhanced by spatial compounding as described in Chapter 5 before being processed with the algorithm described in this chapter.  6.3 Results The subject biometrics are shown in Table 6-1. Tests were run on 39 sets of images of varying image quality. The proposed method successfully detected the LF in 34 cases out of 39 (87%). Figure 6-10 shows the intermediate images throughout the automatic detection of the LF. Among the 5 failed detections, 1 case mis-detected the vertebral body as the LF, because the lamina was not properly detected above the LF (Figure 6-11a), 3 had a disconnected lamina-LF complex and very poor image quality (Figure 6-11bcd), and 1 image had no discernable LF (Figure 6-11e). The low quality images (Figure 6-11de) were of an obese subject with a significant layer of fat (16 mm).  170  Table 6-1 Patient detailed information. The success column shows success of the LF detection on each of the two data sets of one patient  patient age weight height BMI success 1 35 62 168 22 Y N 2 32 62 160 24.2 Y N 3 33 80 157 32.5 Y Y 4 40 66 157 26.8 Y Y 5 37 70 164 26 Y N 6 36 73 179 22.8 Y Y 7 40 56.5 155 23.5 Y Y 8 35 100 162 38.1 Y n/a 9 36 55 160 21.5 Y Y 10 36 84 166 30.5 Y Y 11 35 82.5 169 28.9 Y Y 12 33 86 155 35.8 Y Y 13 30 74 169 25.9 Y Y 14 33 114 149 51.3 N N 15 38 91 170 31.5 Y Y 16 32 62 155 25.8 Y Y 17 44 64 166 23.2 Y Y 18 34 61 157 24.7 Y Y 19 23 71 163 26.7 Y Y 20 38 86 150 38.2 Y Y   171  a) b) c) d) Figure 6-10  The steps of the LF detection on ultrasound image of a human subject a) Spatial compounded image of a typical subject b) ridge map c) lamina similarity map and five strongest points of correlation d) five most likely position of the LF, note that there are cases where one lamina contains two of the five positions, in those cases, the same position for the LF is chosen twice. Here the subsequent detections of the LF are colour-coded with blue, green, turquoise, red and purple corresponding to the 1st, 2nd, … 5th LF detections.  172   a)  b) c)  d)  e)  Figure 6-11  The five cases of failed LF detection: a) in this image, the region directly underneath the LF was detected as a lamina and the search for the LF found the vertebral body as the LF (image set 2); b) the structures were hard to discern, no lamina was detected on top of what was manually chosen as the lamina (image set 3); c) the structures were hard to discern but a few laminae were still found (image set 9); d) The lamina is not connected to the LF so the search found the position of the lamina (image set 27); e) a case of low image contrast, no discernible LF. The lamina was found but could not lead to the LF (image set 28). The white arrow denotes the actual location of the LF as determined by manual segmentation.  173  The accuracy on the 34 successful detections of the LF is shown in Table 6-2. The RMS error of manual versus automatic detection of the LF is 0.64 mm, equivalent to 2 pixels, the average error is 0.04 mm and 95% limits of agreement of -1.3 mm to 1.4 mm. The RMS error of the automatic detection of the LF versus sonographer measurement of the distance on a different image during the pre-puncture examination is 3.7 mm, equivalent to 10 pixels, the average error is 2.5 mm (an overestimate from the automatic detection algorithm) and 95% limits of agreement of -3.1 mm to 8.1 mm. The RMS error between the automatic detection of the LF versus the needle insertion depth as measured by the anesthesiologist is 5.1 mm, equivalent to 16 pixels, the average error is -2.8 mm (an underestimate from the automatic detection algorithm) and 95% limits of agreement of - 12.3 mm to 6.7 mm. The Bland-Altman plots are shown in Figure 6-12.  Table 6-2 The accuracy of the automatic detection of LF compared to manual segmentation, sonographer measurements and anesthesiologist needle insertion depth Automatic detection versus RMS error (mm)  Average error (mm) Bland-Altman 95% limits of agreement (mm) Manual detection 0.64 0.04 -1.3 1.4 Sonographer prepuncture examination 3.7 2.5 -3.1 8.1 Needle insertion depth 5.1 -2.8 -12.3 6.7   174   a)  b)  c) Figure 6-12  Bland-Altman plots of automatic detection versus a) manual segmentation (n=39, 2 interspaces per subject except one case), b) sonographer measurements (n=38, the sonographer was unable to measure the LF depth in one case), c) needle insertion depth (n = 20, only defined for one interspace per subject as only one insertion was performed).   175   Computational costs for the main functions are measured and shown in Table 6-3. The cost breakdown for compounding can be found in Chapter 5. The three main operations of the detection algorithm are the phase symmetry (0.454 s), the lamina template crosscorrelation (3.019 s) and the LF template crosscorrelation (0.392 s).  Table 6-3 Computational cost of automatic detection of LF using Matlab on a Intel Core 2 Quad CPU at 2.83GHz with 3.25GB of RAM on a 262 × 500 pixels image, searching for 5 LF. Function Computational cost (s) Template matching of the lamina 3.019 Template matching of the LF 0.392 Phase symmetry 0.454 Median-based spatial compounding 0.202  6.4 Discussion In 31 cases, the algorithm detects other LFs in addition to the central LF as there can be more than one foramina observed on each image. The algorithm was set to detect the 5 most likely positions of the LFs (Figure 6-10). If phase symmetry provides a ridge map corresponding to the lamina and LF, the lamina will be detected and the ridge will be followed until it reaches the LF or will remain at the edge of the lamina if the LF is missing or disconnected (Figure 6-11de).   176  Looking at the cases where the automatic detection algorithm failed (Table 6-1), only one of the four subjects with failures was obese (defined as BMI > 30 [WHO-1997]). Six other obese subjects were successful in automatic detection, so there does not appear to be a relationship between obesity and success of the detection algorithm. On the other hand, the subject that produced failures on both automatic detection attempts had a BMI of 51.3, which is much larger than all other subjects and can be classified as morbidly obese. For that subject, the image quality was relatively poor.  It is possible that patients with very high BMI still present a challenge to the algorithm because very large imaging depths produce images with poorer image quality. To investigate this issue further, the thickness of the fat layer above the ES in the ultrasound image was manually measured for each subject because obesity may not correlate with fat thickness on the back. Indeed, the subcutaneous fat thickness only had a correlation coefficient of R2=0.71 with obesity. It is observed from the data that out of 7 subjects with a fat thickness greater than 5 mm, only one case had a failed LF detection, whereas for fat thickness less than 5 mm, 3 out of 13 cases had a failed LF detection. In summary, it is more appropriate to state that the success of the LF detection depends mainly on the ultrasound image quality, but not directly related to obesity or fat thickness. This means that the automatic detection algorithm may provide valuable assistance on those patients where epidural needle insertion targeting is difficult. Since there is still a trade-off between the benefit of ultrasound guidance, versus the additional equipment, training and time needed to perform the ultrasound scan, it is expected initially that ultrasound will be used mainly on patients after a failed insertion with traditional LOR technique. As more anesthesiologists gain experience with ultrasound, this technology may be used more frequently as a pre-  177 puncture imaging technique to identify and assist “difficult” patients. Given the increasing popularity of ultrasound in anesthesiology, coupled with miniaturization and cost-reduction of ultrasound scanners, ultrasound may eventually be used on a wide range of patients, but likely still coupled with the LOR technique.  All measures are normally distributed as shown by the omnibus K2 test. The bias was only 0.04 mm between the automatic detection and the manual detection of the LF and is not statistically significant (p>0.05). There is a significant error of 2.5 mm overestimate (p < 0.001) of the skin-to-LF depth when comparing the sonographer and the automatic detection measurements. Most of this error likely arises because the sonographer always measured the depth from the skin to the leading edge of the LF, as opposed to the algorithm which was designed to locate the central part of the LF. The average thickness of the LF echo measured on the ultrasound images is 2.6 mm and the actual thickness of the LF is 5.0-6.0 mm [Chestnut-2004]. In comparison, there is also a significant error (p < 0.03) of 2.8 mm when comparing the anesthesiologist needle insertion depth and the automatic detection is an underestimate of the skin-to-LF depth and this also likely reflects the difference in measurement method. This is expected because the anesthesiologist inserts the needle until the tip passes the LF and goes into the ES, experiencing the LOR thereafter to confirm entry into the ES [Chestnut-2004]. This means the distance measured by the anesthesiologist is the skin-to-ES depth whereas the automatic detection is measuring the distance from the skin-to-the-middle-of -the-LF and the sonographer is measuring the skin-to-top-of-the-LF.   178 There is also an RMS error between the automatic detection and the sonographer (3.7 mm). It has been shown that the repeatability of a measurement by a sonographer is about 4.75% to 7% [Balint-2001]. The average skin-to-LF depth in this study was 46.2 mm, so an intraobserver repeatability error around 2.2 mm to 3.2 mm is likely. The RMS error between the automatic detection and anesthesiologist measured needle insertion depth is 5.1 mm. In addition to the bias discussed earlier, the ultrasound is in the paramedian plane and the needle insertion is performed in the midline. These sources of error are explored in more detail in Chapter 3.  The compression from the ultrasound transducer can cause a small but significant difference in the measurement of the skin-to-LF depth. In Chapter 4, it was measured that the skin-to-LF depth changed by 2.8 mm on average (6.4%), on an average skin-to-LF depth of 43.8 mm.  The majority of computation time is spent on the Pearson cross-correlation to match the lamina template to the ridge map (3.0 s on Matlab), thus obtaining the lamina similarity map. The correlation can also be performed on a coarse resolution ridge map as the position of the lamina is not required to be precise. Overall computation cost can also be reduced by defining a region of interest. Computational speed should be judged sufficient if the algorithm can run in less than one second on a general purpose CPU with additional hardware or acceleration techniques. The automated algorithm can achieve this goal on the CPU of a machine comparable to the ultrasound machine used in this study. Other  179 ultrasound machines have similar performance so the algorithm can be implemented on a wide range of ultrasound machines.  The typical weakness of many automated image interpretation algorithms is the need to tune the parameters for optimal performance on each image used. The parameters of the algorithm are set to values determined by the range of variation in geometry of adult lumbar anatomy. All parameters remained fixed for all subjects in this study. Failures are attributed to image quality and not to suboptimal parameter values.  Although the images were acquired by an experienced sonographer, the scanning parameters remained unchanged throughout the 20 subjects, with the exception of the depth and focus. Since few parameters are changed, it is expected that anesthesiologists can achieve comparable image quality to sonographers. In Chapter 4, with a similar scanning protocol, the sonographer was replaced with an anesthesiologist for real-time ultrasound guidance of epidural needle insertion. The key is to learn the correct location where to place the probe to see the anatomy clearly, which is learned quickly by searching for the familiar wave-like pattern of the lamina, as explained in Chapter 4. Moreover, it is suggested that anesthesiology residents, with little or no ultrasound experience, can rapidly learn and improve their speed and accuracy in performing a simulated interventional ultrasound procedure [Sites-2004].  The user is expected to have an understanding of how to interpret the structures seen in the ultrasound image so that the results of the automatic detection algorithm can be  180 assessed for possible failures. Since the automatic skin-to-LF depth measurement is shown as a graphic overlap on the ultrasound image, an incorrect identification of the LF should be obvious if the overlay does not lie near the expected location of the LF relative to the lamina. As an additional measure, it may be possible to look at the level of cross- correlation with the lamina template and the LF template. Figure 6-13 shows a distribution of the LF detections in the cross-correlation with the lamina and LF templates space. Note that sometimes the first LF detected is a wrong detection but that the second LF is correct. For these patients, a cross-correlation coefficient of the lamina template greater than 0.5 was always associated with success of the algorithm. The display of the cross-correlation coefficient should therefore be investigated in future research.   181  Figure 6-13   The distribution of the LF detections in the crosscorrelation with lamina and LF templates space. These are the detection of the first lamina and LF. The circles are the successful detections and the crosses are the failed detections.   A depiction of the ultrasound image with the LF detection and the measurement of the skin-to-LF depth are overlayed on the image, as shown in Figure 6-14. This automatic LF detection can be used in the following proposed clinical protocol: - place the ultrasound transducer in the paramedian plane over the target intervertebral level using the method described in Chapter 3 to identify the levels. - show the ultrasound with overlay locating the LF - adjust transducer placement until the clearest view of the LF is obtained during prepuncture scanning; then the anesthesiologist confirms that the LF is correctly  182 identified by the algorithm and memorizes the provided skin-to-LF depth measurement for subsequent manual needle insertion without further use of ultrasound - if ultrasound is used for real-time needle guidance, then the anesthesiologist confirms that the LF is correctly identified by the algorithm and aligns the needle trajectory as shown in Chapter 4 with the LF.  42.5mm  Figure 6-14 An ultrasound image of the lumbar anatomy with the LF detected, showing the skin-to- LF depth and the automatic measurement as an overlay.    The needle is inserted with precaution such that the actual needle insertion depth (using the markings on the needle) only comes close to the skin-to-LF depth measured on the  183 image, and traditional LOR is used for the final approach. The needle tip is then placed between the LF and dura mater as confirmed by the LOR technique.   184  6.5 Conclusions An automatic algorithm can detect the lamina and LF and measure the skin-to-LF depth with a success rate of approximately 87% in adult subjects. The algorithm is meant to assist the anesthesiologist in quickly finding the ES and measure the skin-to-LF depth because of time limitations in the operating room and the need to remain sterile. Therefore, limited interaction with the ultrasound machine is required. It should be emphasized that this algorithm is meant to be an aid and not a replacement for image interpretation by the anesthesiologist. The presence of small errors and occasional failures to detect the LF mean the anesthesiologist should view the actual ultrasound image with the automated measurement shown as an additional guide. The algorithm uses image processing techniques that can be implemented to run sufficiently fast on the general computing hardware of typical ultrasound machines. Failures occurred mainly in images with poor image quality so continued improvement in image quality should also improve performance of the algorithm.  Current interest in ultrasound for epidural needle insertion is toward real-time needle insertion in which in-plane needle insertion is performed with a needle guide and an associated predefined trajectory line. Future algorithms will look for the LF along a region of interest defined around this trajectory line used for real-time needle insertion, in which case there will not be any ambiguity about which is the LF of interest, as well as decreased computational cost.   185 7 Conclusion and future work In this final chapter, a summary of the thesis is presented. Our contributions to the ultrasound technique and instrumentation applied to epidural anesthesia are also summarized and future directions for work in this research field are also suggested.  7.1 Summary Instrumentation of the LOR  In Chapter 2, we presented a method to measure some aspects the anesthesiologist monitors that are part of the “feel” of the LOR technique. In this technique, as the anesthesiologist inserts the epidural needle into the patient’s spinal anatomy, force was applied at the syringe plunger to inject saline into the tissues. Depending on the tissue the needle tip is located, the feedback will be different, as the muscles were easier to inject saline than the interspinous ligaments, and the latter were easier to inject saline than the LF, followed by a sudden drop in resistance when the needle tip reached the ES where saline was easily injected. The force applied by the thumb of the anesthesiologist to the plunger was measured using a force sensor mounted onto a metal clip worn by the anesthesiologist during needle insertion. The position of the plunger with respect to the syringe was also monitored so to be able to measure the flow rate of saline being injected at any given time. The pressure at the tip of the syringe was also measured by a pressure sensor attached between the epidural needle and the syringe.  186 These measurements were performed on porcine subjects to compare the “feel” when the needle insertion is performed in the midline approach, which has the needle tip traverse the interspinous ligament and LF before reaching the ES, to the paramedian approach which has the needle tip traverse the muscle and LF before reaching the ES. The flow rates in tissues were shown to be significantly higher in the paramedian approach compared to the midline approach while the force applied and pressure were not significantly different. At the epidural space, the flow rates were not significantly different. Finally, the force and flow rate sensors were used to measure the LOR in the midline approach in human subjects to obtain quantitative measures for the human anatomy. Such measures may be useful for future work on haptic simulators that can be used for training.   Preinsertion ultrasound guidance for lumbar epidurals  In Chapter 3, we studied how ultrasound can be used to measure the skin-to-LF depth directly or indirectly with surrogate measures.  We defined a paramedian ultrasound scanning technique for correctly identifying the vertebral level which starts counting from the 12th rib. We also identified possible surrogate measures of the skin-to-LF depth to improve the ease of scanning when the LF is not clearly visible. Twenty parturient subjects were examined with preinsertion ultrasound in the paramedian plane and the predicted depth was compared to the actual midline depth. The actual  187 needle insertion depth was also compared to subject biometrics, skin-to transverse process tip and thickness of subcutaneous fat. The scanning technique allowed the skin-to-LF depth to be measured in all subjects. The skin-to-LF depth measured in ultrasound was strongly correlated to the actual needle insertion depth, while lower correlation was observed with patient biometrics, the skin- to- transverse process tip, or the thickness of subcutaneous fat. The duration of the ultrasound scan decreased throughout the trial suggesting there is a learning curve for using ultrasound to image the lumbar anatomy.  Paramedian ultrasound can be used to estimate the midline needle insertion depth. The surrogate measures are not sufficiently correlated with the needle insertion depth to recommend surrogate measures as a replacement for a direct depth measurement to the target (LF).  Real-time ultrasounds for lumbar epidurals  In Chapter 4, ultrasound was used as real-time guidance for the epidural needle insertion. A new aim-and-insert single-operator ultrasound-guided epidural needle placement was developed, described in detail, and demonstrated.  19 subjects undergoing elective cesarean delivery were consented to undergo both a pre- puncture ultrasound scan and real-time paramedian ultrasound-guidance for needle insertion.  The objectives were to measure the success of a combined spinal-epidural  188 needle insertion under real-time guidance, compare the locations of the chosen interspinous level determined by ultrasound and palpation, measure the change in skin-to- LF depth from the skin surface as pressure is applied to the ultrasound transducer, and investigate the geometrical limitations of using a fixed needle guide.  The measurements of insertion lengths agreed with the geometrical model of the needle guide, but the needle requires a larger insertion angle than without the guide.  This small study demonstrates the feasibility of a real-time ultrasound-guidance technique. Areas for further development are identified for both ultrasound software and physical design.   Adaptive spatial compounding  As the image quality is an important factor in the acceptance of ultrasound in epidural needle guidance, in Chapter 5, an adaptive spatial compounding method with warping was developed and tested on images of phantoms and human subjects. Spatial compounding can emphasize structures; however, features in the beam-steered images are not always aligned. A non-rigid registration method, called warping, shifted pixels of the beam-steered images to best match the reference image. A simple method to improve computation speed, called linear prediction, was used to find the warping vectors in the registration step. An adaptive median-based combination technique for compounding was also investigated. The results showed a significant improvement in quality when using warping with adaptive median-based compounding.  189   Automatic detection of LF  In Chapter 6, an algorithm to automatically detect and measure the skin-to-LF depth was developed. The anesthesiologist needs to measure relevant distances while preserving sterile conditions, so interaction with the ultrasound controls must be minimal. Automated measurement is therefore needed, and such a method was developed.  Beam-steered ultrasound images were captured and spatial compounding was used to improve image quality. Phase symmetry was used to enhance bone (lamina) and LF ridges. A lamina template was matched to this ridge map using Pearson’s crosscorrelation and the most likely lamina positions were found. Then, the lamina was traversed using a LF template with the Pearson’s crosscorrelation and the location of the LF was obtained.  The accuracy, reliability and speed suggest that this method may be valuable for helping guide epidurals in conjunction with the traditional LOR method.  In summary, we believe we have helped the problem of blind insertion. The proposed methods may also be used for a variety of other blind needle insertion procedure such as thoracic epidurals, spinals, as well as non-anesthesia procedures such as facet joint steroid injections.   190 7.2 Contributions Quantifying LOR  The LOR felt through the epidural needle insertion in the midline and paramedian in porcine subjects was successfully quantified by using the force, position and pressure sensors. It was shown that the anesthesiologist applied statistically insignificantly different forces and pressures yet obtained a significantly different flow rate in the two approaches. The force and displacement data were also captured on humans during midline epidural needle insertions and differences in the flow rate and force were measured in interspinous ligament, LF and ES.  Ultrasound protocol for counting vertebra  Ultrasonography is becoming a standard of care in several anesthesia procedures. It follows that lumbar epidural anesthesia, being one of the most difficult anesthesia procedures, should also benefit from ultrasound. There are many possible uses of ultrasound. First, it can be used to count the vertebra and locate the proper site of insertion, as it has been shown [Furness-2002] that counting via palpations often gives incorrect identification. We have developed a counting protocol for counting up from the sacrum and counting down from the 12th rib. Using the two methods together should reduce the errors from missing 12th rib, 13th rib, sacralization and lumbarization which each have a small probability of occurrence.   191 Paramedian ultrasound can predict depth of ES  The second use of ultrasound is to measure the skin-to-LF depth. There have been several proposed methods of predicting the needle insertion depth by ultrasound: a transverse midline approach [Arzola-2007], a midline longitudinal approach [Grau-2001b], and a paramedian longitudinal approach [Grau-2001c]. Our work focuses on the paramedian longitudinal approach because it provides the best images and we showed that the measured distance in this approach is well correlated with the actual needle insertion depth. Quantifying LOR was used as the endpoint for the first time. We also showed that other biometrics such as skin to transverse process tip, age, sex, weight, height are not well correlated to the needle insertion depth. Fat thickness has a better correlation compared to the other indirect biometrics, but is still insufficient for guidance.  Ultrasound for real-time guidance of epidural needle insertion  The third use of ultrasound is the real-time guidance of epidural needle insertion. Real- time guidance has been used on adults as a two-operator method in [Grau-2004], and as a free-hand method with a single-operator and specialized syringe in [Karmakar-2009]. We designed a single operator real-time guidance needle insertion in which a standard needle and syringe are attached to the ultrasound probe. This permits the needle trajectory to be predictable and shown as an overlay on the ultrasound image. This method permits an aim-and-insert method in which the needle trajectory is aimed at the target LF and needle insertion can be achieved in one attempt. This method necessitates a paramedian  192 ultrasound plane with an in-plane paramedian needle insertion as well as a longer needle as it will have a steeper insertion angle and extra length required for the needle guide. Needle visibility was still very poor, however, because the needle angle to the ultrasound beam was not steep enough to produce strong echoes. This work shows the need for a dedicated ultrasound system for epidurals that overcomes the geometric limitations of in-plane ultrasound, and needle visibility.  Multi-layered ultrasound phantom  A novel two-layered phantom was built of agar gelatine to mimic the lumbar anatomy containing a layer of fat and a layer of muscle. The layer mimicking fat has a speed of sound of about 1479 m/s and the layer mimicking muscle of about 1566 m/s using the method in [Burlew-1980]. The speeds of sound are controlled by measuring the depth of a column of gelatine with both a millimetric ruler and ultrasound and correlating the two measurements. A plastic set of lumbar vertebra is submerged in the muscle layer and a water-saturated mixture of rice flour is used to mimic the LF as all that is required is a reflector.  Spatial compounding using linear prediction for warping  Beam-steering was used to capture features at different angles and the images were compounded together. In this work, a warping method was used to align the features just prior to compounding, a median-based compounding method was used to combine the  193 images by weighting strong features more heavily than the background. Finally, a new linear prediction stage was added to reduce the computational cost. Metrics for comparing image quality in this clinical application were developed. The median-based adaptive spatial compounding was tested on a spine phantom and in a clinical trial on 23 patients and showed significant improvement of image quality.  Calculate the refraction and speed of sound errors in beam-steering  The warping step was used to realign the features in the different beam-steered images. Misalignment can arise from refraction of the beams, speed of sound variation and tissue motion during image acquisition. Without warping, the features on the compounded images appeared blurred. A formula was derived for a linear ultrasound probe and a curvilinear ultrasound probe and showed that refraction produces small errors. It was also shown that refraction and speed of sound errors can partly cancel each other, but errors depend on fat thickness, beam angle and other properties of the tissue and acquisition system. Registration results will therefore vary amoung subjects and ultrasound settings.  LF detection algorithm  Finally, because recognizing the LF in the ultrasound image is difficult for non- experienced users, and measuring the skin-to-LF depth in the images can be difficult in sterile conditions, a novel automatic method for detection and measurement was developed. The algorithm used phase symmetry for ridge detection in the ultrasound  194 image of the lumbar anatomy. The algorithm then used template-matching to locate most likely lamina and then another template for detecting the LF at the base of the lamina. The algorithm can detect several LF as there are typically 3 visible LF in each ultrasound image. This algorithm has been tested on 20 patients and achieved a success rate of 87% and an RMS error of 5.1 mm when compared with the actual needle insertion depth. 7.3 Future work Dual-probe needle insertion system  The technique in Chapter 4 is a paramedian needle insertion technique. There is much debate as to whether midline or paramedian approaches are superior with proponents on each side claiming their approach holds different advantages [Rabinowitz-2007] [Leeda- 2005]. For proponents of the midline approach, interest remains in having an ultrasound- guided midline in-plane epidural needle insertion. A 3D ultrasound probe could be used for this purpose. The volume generated with a 3D probe can be sliced in the sagittal plane containing the epidural needle and shown in the display. In this way, a midline real-time epidural needle can be observed in real-time without obscuring the puncture site. The frame rate for the 3D ultrasound is however usually slower compared to 2D imaging with only one volume produced approximately every second, so the slice of interest is updated relatively slowly. A slow frame-rate real-time guidance system is not ideal for an anesthesiologist, therefore, a second 2D ultrasound probe could be added to the apparatus, on the opposite side of the spine, in a transverse view to provide real-time ultrasound observation of the needle insertion [Lo-2010] [Rasoulian-2010] (Figure 7-1). Preliminary tests on a porcine subject retain adequate LOR sensation and prepuncture  195 depth measurement accuracy comparable to current methods. At present, image formation alternates from one probe to the other. Real-time probe switching capability is currently being developed to obtain true real-time ultrasound guidance. The graphical interface of a prototype system is shown in Figure 7-2.  Figure 7-1 Anesthesiologist using a prototype of the dual-probe to simultaneously acquire transverse and a midline 3D resliced frame, while performing a midline needle insertion on a porcine subject.  196  Figure 7-2 Graphical user interface of the dual-probe with a transverse view on the left and a midline longitudinal 3D resliced view on the right. The red line is the calibrated needle trajectory.  The dual probe concept also has the potential for better visualization of the needle. The needle may appear at certain angles and not at others as observed when performing the real-time needle insertions with spatial compounding. It has also been observed that spatial compounding may help in visualizing the needle during clinical trials. Compounding can be performed between the 3D volume and the 2D image as well as standard spatial compounding in the 2D transverse image plane to improve visibility of the needle and structures.    197   Thoracic epidural guidance  This research has focused on lumbar epidural anesthesia but thoracic epidural anesthesia is also of interest. Although lumbar and thoracic epidurals both require epidural needle insertion, and both techniques use the LOR as a confirmation of the needle tip being in the ES, thoracic epidural needle insertions are typically performed in the paramedian plane as the space between the spinous processes is barely enough for the needle. The needle is inserted to purposely hit the lamina, then the angle is changed so as to “walk up” the lamina until the needle reaches the LF. The needle is typically inserted at an angle of 15° inward and 55° upward at a distance of 1 cm lateral to the midpoint between two spinous processes [Cousins-1998]. Since the needle insertion plane is paramedian, the method described in Chapter 4 may be suitable. Moreover, thoracic epidural anesthesia is used mostly for post-surgical pain relief; this means the procedure could be performed while the patient is under general anesthesia, provided sufficient clinical evidence of success and confidence in imaging the anatomy and the needle tip. The main issue is that the anesthesiologist needs confirmation that there are no nerve damage and a good catheter insertion. If ultrasound can be made reliable enough to guarantee visualization of the procedure, i.e. good visualization of the needle and LF, then thoracic epidurals can be performed under general anesthesia. Beam-steering shows promise in needle visualization and electromagnetic trackers may also be employed to predict the needle insertion trajectory.  198  Certain ultrasound systems have already integrated magnetic sensors in the ultrasound transducers and needles (e.g. the Ultrasonix “global positioning system (GPS)”) to provide a predicted needle path relative to the probe position which can be overlayed onto the ultrasound image. This can be potentially very helpful as it will permit free-hand needle insertion to be performed while using ultrasound to plan for the insertion and observe the insertion of the needle in real-time. However, the accuracy of such a system, needle visibility, and sterility remain issues to be solved.   199 The automated measurement technique described in Chapter 6 may also be applied to the thoracic spine to detect the LF. Using similar parameters as with the human lumbar anatomy, the lamina and LF were successfully detected in porcine subjects [Tran- 2010c](Figure 7-3). A lower success rate was found (60%) compared to the lumbar spine. This is mainly a reflection of the need for a different template for the lamina and the LF. Nevertheless, as there was some success on a porcine study, research on automated measurement should continue.  Figure 7-3 Porcine thoracic paramedian spinal anatomy. In red is the ridge map. The white circles are the detected LF positions.  7.4 Final comments  There are many ways to perform a successful epidural needle insertion, and experienced anesthesiologists do so rountinely. Nevertheless, it remains a very challenging procedure,  200 especially for those with less experience, so additional guidance is needed with specialized ultrasound tools. In my opinion, epidural needle insertions should begin with identification of the desired intravertebral level by using the method of counting from the 12th rib down, as described in Chapter 3. Next, the needle insertion should be performed under real-time ultrasound-guidance using the procedure described in Chapter 4, in conjunction with the automatic LF detection described in Chapter 6. The anesthesiologist would then just need to align the needle trajectory with the detected LF location and insert the needle.  This is a paramedian approach for needle insertion, as used in several European and African countries. In North America, the standard of care is a midline needle insertion, and there are arguments from both sides, each claiming less pain than the other approach. In our experience, the pain perception was very subjective, with some patients feeling pain when the ultrasound transducer was placed on their backs, even without a needle. In some countries, epidurals are performed with the same 17G needle in the paramedian plane without the use of local anesthesia to ease pain, and it is accepted by the patient population. Given this anectodal evidence, it is difficult to discount a paramedian approach given that it can be facilitated by 2D ultrasound guidance.  It should also be made clear that the usefulness of this technology is greatest for physicians with the least experience with epidurals. An experienced anesthesiologist can achieve an extremely high rate of success and would be able to perform most epidurals without difficulty. Residents and inexperienced anesthesiologists, however, would benefit from learning and using our methods. The LOR should still be taught because it is still the standard of care. I would recommend that it be used but under ultrasound guidance in  201 order to give the inexperienced physicians a better understanding and confidence in performing the epidural.  202 References  [Achim-2001]  Achim A, Bezerianos A, Tsakalides P. Novel bayesian multiscale method for speckle removal in medical ultrasound images. IEEE Trans Med Imaging. 2001; 20:772-783. [Ackerman-1998] Ackerman MJ. The visible human project. Proc. of the IEEE 1998; 86:504-511. [Al-Attas-2009] Al-Attas E, Tran D, Kamani AA, Massey S, Lessoway VA, Rohling RN. Ultrasound scanning technique for epidural needle insertion by a single operator. Society for Obstetric Anesthesia and Perinatology (SOAP) Annual Meeting, 2009. [Amir-1986]  Amir I, Bilgutay NM, Newhouse V. Analysis and comparison of some frequency compounding algorithms for the reduction of ultrasonic clutter. IEEE trans ultrasonics ferroelectrics and frequency control 1986; 33(4):402-411. [Anderson-1997]  Anderson ME, Soo MS, Bentley RC, Trahey GE. The detection of breast microcalcifications with medical ultrasound. J Acoust Soc Am 1997; 101:29-39. [Arendt-2008] Arendt K, Segal S. An effectiveness study of air versus saline for identification of the epidural space by loss of resistance. 40th annual meeting of society for obstetric anesthesia and perinatalogy; 2008 April 30-May 4; Chicago, USA; Anesthesia and Analgesia. 2008;106:A104. [Arzola-2007]  Arzola C, Davies S, Rofaeel A, Carvalho JCA. Ultrasound using the transverse approach to the lumbar spine provides reliable landmarks for labor epidurals. Anesthesia and Analgesia 2007; 104:1188-1191. [Balint-2001]  Balint PV, Sturrock RD. Intraobserver repeatability and interobserver reproducibility in muscoskeletal ultrasound imaging measurements. Clin. Exp. Rheumatol. 2001; 19:89-92. [Behar-2006]  Behar V, Nikolov M. Statistical analysis of image quality in multi- angle compound imaging. Proceedings of the IEEE John Vincent Atanasoff 2006 International Symposium on Modern Computing; 2006 Oct 3-6; JVA;Washington, DC;IEEE Computer Society;197- 201. [Beilin-2000]  Beilin Y, Arnold I, Telfeyan C, Bernstein HH, Hossain S. Quality of analgesia when air versus saline is used for identification of the epidural space in the parturient. Regional Anesthesia and Pain Medicine 2000;25(6):596–599. [Berson-1981]  Berson M, Roncin A, Pourcelot L. Compound scanning with an electrically steered beam. Ultrason Imaging 1981;3:303–8. [Bluvol-2009] Bluvol N, Komecki A, Shaikh A, Fernandez DDR, Taves DH, Fenster A. Freehand versus guided Breast biopsy: comparison of accuracy, needle motion, and biopsy time in a tissue model. American Journal of Roentgenology 2009; 192:1720-5.  203 [Bonazzi-1995] Bonazzi M, de Gracia LB. Individuazione ecoguidata dello spazio epidurale lombare. Minerva Anestesiol 1995;61:201-205. [Bonica-1956]  Bonica JJ. Continuous peridural block. Anesthesiology 1956; 17:626-630. [Bro-Nielsen-1996]  Bro-Nielsen M, Gramkow C. Fast fluid registration of medical images. LNCS 1996; 1131:265-276. [Burlew-1980]  Burlew M, Madsen E, Zagzabski J, Banjavic R, Sum S. A new ultrasound tissue-equivalent material. Radiology. 1980; 134:517- 520. [Canny-1986]  Canny J. A computational approach to edge detection. IEEE Trans Pattern Anal Mach Intell. 1986; 8:679-714. [Carden-2006]  Carden E, Ori A. The bip test: a modified loss of resistance technique for confirming epidural needle placement. Pain Physician 2006;9(4):323–325. [Carpenter-1980]  Carpenter DA, Dadd MJ, Kossoff G. A multimode real time scanner. Ultrasound Med Biol 1980;6:279–84. [Carvalho-2008]  Carvalho JCA. Ultrasound-facilitated epidurals and spinals in obstetrics. Anesthesiol Clin. 2008 Mar;26(1):145-158. [Casati-2007]  Casati A, Baciarello M, Di Cianni S, Danelli G, De Marco G, Leone S, Rossi M, Fanelli G. Effects of ultrasound guidance on the minimum effective anaesthetic volume required to block the femoral nerve. British Journal of Anaesthesia 2007; 98(6):823-7. [Chaudhry-2008]  Chaudhry SY, Phelps AL, Mantha V, Dalby PL, Romeo RC, Vallejo MC. Longitudinal vs. Transverse View to Determine Actual Clinical Depth for Labor Epidural Placement using the Sonosite® Micromaxx® Ultrasound System. 40th annual meeting of society for obstetric anesthesia and perinatalogy; 2008 April 30- May 4; Chicago, USA; Anesthesia and Analgesia. 2008;106:A- 136. [Chen-2003]  Chen Y, Yin R, Flynn P, Broschat S. Aggressive region growing for speckle reduction in ultrasound images. Pattern Recognition letters 2003; 24:677-691. [Chestnut-2004]  Chestnut DH, Obstetric anesthesia: principles and practice. 2nd ed, USA: Elsevier Mosby, 2004. [Christensen-1988]  Christensen D. Ultrasonic bioinstrumentation. New York: John Wiley and Sons; 1988 [Christensen-1996]  Christensen GE, Rabbitt RD, Miller MI. Deformable templates using large deformation kinematics. IEEE transactions on image processing 1996; 5:1435-1447. [Cobbold-2007] Cobold RSC. Foundations of biomedical ultrasound. New York: Oxford University Press; 2007. [Cohen-2001]  Cohen B, Dinstein I. New maximum likelihood motion estimation schemes for noisy ultrasound images. Pattern Recognition. 2001; 35:455-463. [Cook-1999] Cook TM. Combined spinal epidural anaesthesia: a new technique. Int J Obstet Anesth 1999; 8:3-6.  204 [Cork-1980]  Cork RC, Kryc JJ, Vaughan RW.  Ultrasonic localization of the lumbar epidural space.  Anesthesiology 1980; 52(6):513-6. [Cousins-1998] Cousins MJ, Veering BT. Epidural neural blockade. In: Cousins MJ, Bridenbaugh PO, editors. Neural blockade in clinical anesthesia and management of pain. Philadelphia7 Lippincott- Raven Publishers; 1998. p. 243- 321. [Currie-1984] Currie JM. Measurement of the depth to the extradural space using ultrasound. Br J Anaesth 1984; 56:345-7. [D’Agostino-1971]  D’Agostino RB. An omnibus test of normality for moderate and large size samples. Biometrika 1971; 58:341-348. [Daanen-2004] Daanen V, Tonetti J, Troccaz J. A Fully Automated Method for the Delineation of Osseous Interface in Ultrasound Images. LNCS 2004, 3216:549-557. [Dahl-2003]  Dahl JJ, Guenther D, Trahey GE. Performance evaluation of combined spatial compounding/adaptive imaging: simulation, phantom and clinical trials. Proc of the IEEE Ultrasonics Symposium; 2003 Oct 5-8; Honolulu, Hawaii; 2:1532-1536. [Dang-2001]  Dang T, Annaswamy TM, Srinivasan MA. Development and evaluation of an epidural injection simulator with force feedback for medical training. Stud Health Technol Inform 2001; 81:97– 102. [De Andres-2001]  De Andres J, Sala-Blanch X. Peripheral nerve stimulation in the practice of brachial plexus anesthesia : a review. Regional anesthesia and pain medicine 2001; 26(5):478-483. [Diez-1998]  Diez RLA, Reinoso-Barbero F, Sanabria P, Rodriguez E, Suso B. Usefulness of radiologic monitoring of epidural catheters using epidurography. Rev Esp Anestesiol Reanim 1998;45:416-420. [Fan-2001]  Fan L, Santago P, Riley W, Herrington DM. An adaptive template matching method and its application to boundary detection of brachial artery ultrasound scans. Ultrasound Med. Biol. 2001; 27:399-408. [Fanelli-1999]  Fanelli G, Casati A, Garancini P, Torri G. Nerve stimulator and multiple injection technique for upper and lower limb blockade : failure rate, patient acceptance, and neurologic complications. Anesth analg 1999; 88:847-52. [Filev-2005]  Filev P, Hadjiiski L, Sahiner B, Chan HP, Helvie MA. Comparison of similarity measures for the task of template matching of masses on serial mammograms. Med. Phys. 2005; 32:515-529. [Filho-2006]  Filho EDS, Saijo Y, Tanaka A, Yambe T, Li S, Yoshizawa M. Automated calcification detection and quantification in intravascular ultrasound images by adaptive thresholding. IFMBE Proceedings, World Congress on Medical Physics and Biomedical Engineering 2006, August 27 – September 1, 2006 COEX Seoul, Korea “Imaging the Future Medicine”  205 [Filho-2002]  Filho GRO. The construction of learning curves for basic skills in anesthetic procedures: an application for the cumulative sum method. Anesth Analg 2002; 95:411-6. [Foroughi-2007]  Foroughi P, Boctor E, Swartz MJ, Taylor RH, Fichtinger G. Ultrasound bone segmentation using dynamic programming. IEEE ultrasonics symposium 2007. pp 2523-2526. [Furness-2002]  Furness G, Reilly MP, Kuchi S. An evaluation of ultrasound imaging for identification of lumbar intervertebral level. Anaesthesia 2002; 57:277–280. [Gersho-1992]  Gersho A, Gray RA. Vector quantization and signal compression. 1st edition; USA:Springer-Verlag; 1992. [Gonzales-2000]  Gonzales RC, Woods RE. Digital image processing. 2nd edition, USA:Prentice-Hall; 2000. [Grau-2003]  Grau T, Bartusseck E, Conradi R, Martin E, Motsch J.  Ultrasound imaging improves learning curves in obstetrical epidural anesthesia: a preliminary study.   Can J Anesth 2003; 50(10):1047- 1050. [Grau-2001a]  Grau T, Leipold R, Conradi R, Martin E. Ultrasound control for presumed difficult epidural puncture.  Acta Anaesthesiol Scand 2001; 45:766-71. [Grau-2002a]  Grau T, Leipold RW, Conradi R, Martin E, Motsch J. The visualisation of dura perforation and blood patches with ultrasound. Anasthesiol Intensivmed Notfallmed Schmerzther 2002; 37 :149-53. [Grau-2002b]  Grau T, Leipold RW, Conradi R et al. Efficacy of ultrasound imaging in obstetric epidural anesthesia. J Clin Anesth 2002; 14:169–175. [Grau-2001b]  Grau T, Leipold RW, Conradi R, Martin E, Motsch J. Ultrasonography and epidural anaesthesia : technical possibilities and boundaries of ultrasonic examination of the epidural space. Anaesthesist 2001; 50:94-101. [Grau-2001f]  Grau T, Leipold RW, Conradi R, Martin E, Motsch J. Ultrasound imaging facilitates localization of the epidural space during combined spinal and epidural anesthesia. Reg Anesth Pain Med. 2004; 26:64-67. [Grau-2004]  Grau T, Leipold RW, Fatehi S, Martin E, Motsch J. Real-time ultrasonic observation of combined spinal-epidural anaesthesia. Eur J Anaesthesiol. 2004 Jan; 21(1):25-31. [Grau-2001c]  Grau T, Leipold RW, Horter J, Conradi R, Martin EO, Motsch J. Paramedian access to the epidural space: the optimum window for ultrasound imaging. Journal of Clinical Anesthesia. 2001; 13:213- 217. [Grau-2001d]  Grau T, Leipold RW, Horter J, Conradi R, Martin E, Motsch J. The lumbar epidural space in pregnancy: visualization by ultrasonography. British Journal of Anaesthesia 2001; 86:798-804.  206 [Grau-2001e]  Grau T, Leipold RW, Horter J, Martin E, Motsch J. Colour Doppler imaging of the interspinous and epidural space. Eur J Anaesth 2001; 18:706-712. [Grau-2005]  Grau T. Ultrasonography in the current practice of regional anaesthesia. Best Practice & Research Clinical Anaesthesiology 2005; 19(2):175-200. [Groves-2004]  Groves A, Rohling R. Two dimensional spatial compounding with warping. Ultrasound Med Biol. 2004; 30:929-942. [Hacihaliloglu-2006] Hacihaliloglu I, Abugharbieh R, Hodgson AJ, and Rohling RN. Enhancement of bone surface visualization from 3D ultrasound based on local phase information. Proc. IEEE International Ultrasonics Symposium. Vancouver, BC. October 2006; 21-24. [Hacihaliloglu-2009]  Hacihaliloglu I, Abugharbieh R, Hodgson AJ, Rohling RN. Bone surface localization in ultrasound using image phase-based features. Ultrasound in Medicine and Biology 2009; 35:1475-1487. [Hahn-1992]  Hahn PY, Strobel JJ, Hanh FJ. Verification of lumbosacral segments of MR images: identification of transitional vertebrae. Radiology 1992; 182: 580-581. [Hao-1999]  Hao X, Gao S, Gao X. A novel multiscale nonlinear thresholding method for ultrasonic speckle suppressing. IEEE Trans Med Imaging. 1999; 18:787-794. [Herrera-1996]  Herrera A, de las Mulas M. In memoriam Fidel Pages Mirave (1886–1923) on the 75th anniversary of the publication of ‘Anesthesia metamerica’. Rev Esp Anestesiol Reanim 1996; 43: 59–66 [Hill-2004] Hill CR, Bamber JC, ter Haar GR. Physical principles of medical ultrasonics. West Sussex: John Wiley and Sons; 2004. [Hirsch-1963]  Hirsch C, Ingelmark BE, Miller M. The anatomical basis for low back pain: studies on the presence of sensory nerve endings in ligamentous, capsular and intervertebral disc structures in the human lumbar spine. Acta Orthopaedica 1963; 33:1-17. [Hofer-2005]  Hofer M. The basics of performing and interpreting ultrasound scans. New York: Georg Thieme Verlag; 2005 [Hor-2007a]  Hor KW, Instrumentation and ultrasound imaging for epidural anesthesia. Master thesis, University of British Columbia [Hor-2007b]  Hor KW, Tran D, Kamani AA, Lessoway VA, Rohling RN. Instrumentation for epidural anesthesia. Medical Image Processing and Computer Aided Interventions (MICCAI), Brisbane, Australia. LNCS 2007; 4792:918-925. [Huber-2002]  Huber S, Wagner M, Medl M, Czerbirek H. Real-time spatial compound imaging in breast ultrasound. Ultrasound Med Biol. 2002; 28:155-163. [Jespersen-1998]  Jespersen SK, Wilhjelm JE, Sillesen H. Multi-angle compound imaging. Ultrason Imaging 1998;20:81–102.  207 [Jespersen-2000]  Jespersen SK, Wilhjelm SE, Sillesen H. In vitro spatial compound scanning for improved visualization of atherosclerosis. Ultrasound Med Biol 2000;26:1357–62. [Johnson-1999]  Johnson BA, Schellhas KP, Pollei SR. Epidurography and therapeutic epidural injections: technical considerations and experience with 5334 cases. Am J Neuroradiol 1999; 20:697-705. [Johnson-2000]  Johnson BA. Image-guided epidural injections. Neuroimaging Clin N Am 2000 Aug; 10 (3):479-91. [Kamani-2007]  Kamani AA, Hor KW, Tran D, Lessoway VA, Rohling RN. Pressure displacement comparing continuous vs intermittent technique. Society for Obstetric Anesthesia and Perinatology Annual Meeting. Banff Hot Springs, AB. May 2007. Published in Anesthesiology, Lippincot, Williams & Wilkins, Philadelphia, PA 2007. 106: 65. [Kamani-2008]  Kamani AA, Tran D, Lessoway VA, Rohling RN. . Identification of the epidural space using real-time ultrasound for needle guidance in porcine subjects. Society for Obstetric Anesthesia and Perinatology (SOAP)Annual Meeting. Published in Anesthesia & Analgesia, Lippincot, Williams & Wilkins, Philadelphia, PA 2008. 106: A-18. [Karaman-1995]  Karaman M, Kutay MA, Bozdagi G. An adaptive speckle suppression filter for medical ultrasonic imaging. IEEE trans med img 1995; 14(2):283-292. [Karmakar-2009] Karmakar MK, Li X, Ho AMH, Kwok WH, Chui PT. Real-time ultrasound-guided paramedian epidural access: evaluation of a novel in-plane technique. British Journal of Anaesthesia 2009; 102 (6):845-54. [Konrad-1998]  Konrad C, Schupfer G, Wietlisbach M, Gerber H. Learning manual skills in anesthesiology: Is there a recommended number of cases for anesthetic procedures? Anesth Analg 1998; 86: 635–639. [Kopacz-1996]  Kopacz DJ, Neal JM, Pollock JE. The regional anesthesia “learning curve”. What is the minimum number of epidural and spinal blocks to reach consistency? Reg Anesth 1996; 21:182-90. [Kovesi-1997]  Kovesi P. Symmetry and Asymmetry From Local Phase. AI'97, Tenth Australian Joint Conference on Artificial Intelligence. 2 - 4 December 1997. Proceedings - Poster Papers. pp 185-190. [Kovesi-1999]  Kovesi PD. Image features from phase congruency. Videre: Journal of Computer Vision Research 1999; 1:1-26. [Krücker-2002] Krücker JF, LeCarpentier GL, Fowlkes JB, Carson PL. Rapid elastic image registration for 3D ultrasound. IEEE Trans Med Imag 2002; 11: 1384-1394.–1394. [Krucker-2004]  Krucker JF, Fowlkes JB, Carson PL. Sound speed estimation using automatic ultrasound image registration. IEEE Trans Ultrasonics, Ferroelectrics, and Frequency Control. 2004; 51:1095-1106.  208 [Lavallee-1996] Taylor RH, Lavallee S, Burdea GC, Mosges R. Computer- integrated surgery, Technology and clinical applications. MIT Press, Cambridge, MA. pp. 77–97. [LeCoq-1998]  Le Coq G, Ducot B, Benhamou D. Risk factors of inadequate pain relief during epidural analgesia for labour and delivery. Can J Anesth 1998;45: 719–723. [Lechner-2003] Lechner TJ, van Wijk MG, Maas AJ, van Dorsten FR, Drost RA, Langenberg CJ, Teunissen LJ, Cornelissen PH, van Niekerk J. Clinical results with the acoustic puncture assist device, a new acoustic device to identify the epidural space. Anesth Analg 2003;96:1183–7. [Leeda-2005] Leeda M, Stienstra R, Arbous MS, Dahan A, Th Veering B, Burm AG, Van Kleed JW. Lumbar epidural catheter insertion: the midline vs. the paramedian approach. Eur J Anaesthesiol 2005; 22:839-42. [Lefebvre-2000]  Lefebvre F, Meunier M, Thibault F, Laugier P, Berger G. Computerized ultrasound B-scan characterization of breast nodules. Ultrasound Med Biol. 2000; 26:1421-1428. [Lehmann-1999] Lehmann TM, Gonner C, Spitzer K. Survey: interpolation methods in medical image processing. IEEE Trans Med Imaging. 1999; 18:1049-1075. [Lennox-2007]  Lennox PH, Umedaly HS, Grant RP, White SA, Fitzmaurice BG, Evans KG. A pulsatile pressure waveform is a sensitive marker for confirming the location of the thoracic epidural space. J Cardiothorac Vasc Anesth 2007; 21(6):920-1. [Leroy-2004]  Leroy A, Mozer P, Payan Y, Troccaz J. Rigid registration of freehand 3D ultrasound and CT-scan kidney images. LNCS 2004; 3216:837-844. [Lewis-2000] Lewis MC, Lafferty JP, Sacks MS, Pallares VS, TerRiet M. How much work is required to puncture dura with Tuohy needles? Br. J. Anaesth. 2000 85: 238-241 [Lin-2005]  Lin CH, Weng CM, Sun YN. Ultrasound image compounding based on motion compensation. Engineering in Medicine and Biology 27th annual conference; 2005 Sept 1-4; Shanghai, China.; 6445-6448. [Lo-2010] Lo C, Rasoulian A, Rohling RN, Abolmaesumi P, Tran D, Kamani AA. Dual-source ultrasound identification of the lumbar epidural space allows real-time guidance of midline lumbar epidural needle insertion in a porcine model, Society for obstetric anesthesia and perinatology (SOAP) Annual Meeting. San Antonio, Texas, 2010. [Loder-2007]  Loder RT, Huffman G, Toney E, Wurtz LD, Fallon R. Abnormal rib nuber in childhood milgnancy: implications for the scoliosis surgeon. Spine 2007; 32:904-910. [Loupas-1989]  Loupas T, McDicken WN, Allan PL. An adaptive weighted median filter for speckle suppression in medical ultrasound images,” IEEE Trans. Circuits Syst. 1989; 36 :129–135.  209 [Lubbers-1998]  Lubbers J and Graaff R. A simple and accurate formula for the sound velocity in water. Ultrasound Med Biol. 1998. 24:1065- 1068. [Lyons-1992] Lyons G, MacDonald R, Mikl B. Combined epidural/spinal anaesthesia for caesarean section: through the needle or in separate spaces? Anaesthesia 1992;47:199 –201. [Magill-2004]  Magill J, Anderson B, Anderson G, Hess P, Pratt S. Multi-axis mechanical simulator for epidural needle insertion. In: Medical Simulation. LNCS, International Symposium on Medical Simulation 2004; 3078:267–76. [Maintz-1998]  Maintz JBA, Viergever MA. A survey of medical image registration. Medical Image Analysis 1998; 2:1-36. [Matalon-1990] Matalon TA, Silver B. US guidance of interventional procedures. Radiology 1990; 174: 43-47. [Marhofer-2005]  Marhofer P, Willschke H, Greher M, Kapral S. New perspectives in regional anesthesia: the use of ultrasound past, present, and future. Can J Anesth 2005; 52(6):R1–R5. [Mellor-2004]  Mellor, M. (2004). Phasemethods for non-rigidmedical image registration. Ph.D. thesis, Oxford University. [Michailovich-2006]  Michailovich OV, Tannenbaum A. Despeckling of medical ultrasound images. IEEE Trans Ultrason Ferroelectr Freq Control 2006;53:64–78. [Miller-2005]  Miller RD. Miller’s Anesthesia. 6th edition. Philadelphia: Elsevier Churchill Livingstone, 2005. [Moore-1925]  Moore BH. Sacralization of the fifth lumbar vertebra. J Bone Joint Surg Am 1925; 7:271-8. [Moore-1976]  Moore DC. Regional block. Springfield Illinois: Thomas, 1976 [Muranaka-2001]  Muranaka K, Mizutani H, Seo K, Yoshida M, Gohara T, Miyawaki H. A Comparison between midline and paramedian approaches for combined spinal-epidural anesthesia. Japanese Journal of Anesthesiology 2001. 50(10):1085-1088. [Nowakowski-2007]  Nowakowski P, Wielgus M, Kosson D, Mayzner-Zawadzka E. Percutaneous electrode guidance reduces local anesthetic dose in axillary brachial plexus block. Regional Anesthesia and Pain Medicine 2007; 32(5):136. [Oppenheim-1981]  Oppenheim AV, Lim JS. The importance of phase in signals. Proc. of IEEE 1981; 69:529-541. [Pages-1921] Pages F. Anestesia metamerica. Rev de la sanidad militar 1921. 11: 351-85. [Penney-1998]  Penney GP, Weese J, Little JA, Desmedt P, Hill DLG, Hawkes DJ. A comparison of similarity measures for use in 2-d-3-d medical image registration. IEEE Trans Med Imaging. 1998; 17:586-595. [Perona-1990] Perona P, Malik J. Scale space and edge detection using anisotropic diffusion. IEEE Trans Pattern Anal Mach Intell. 1990; 12:629-639.  210 [Peterson-2007]  Peterson CE,  Tran D, Kamani AA, Lessoway VA, Rohling RN. Pre-estimate distance from skin-to-epidural space using ultrasound can be helpful. Society for Obstetric Anesthesia and Perinatology Annual Meeting. Published in Anesthesiology, Lippincot, Williams & Wilkins, Philadelphia, PA 2007; 106:72. [Peterson-2008]  Peterson CE, Tran D, Rohling RN, Kamani AA, Lessoway VA. Ultrasonography of the lumbar spine and instrumentation of the loss-of-resistance technique for epidural anesthesia. World Congress of Anesthesia. Cape Town, South Africa. March 2008. [Pluim-2003]  Pluim JPW, Maintz JBA, Viergever MA. Mutual-information- based registration of medical images: a survey. IEEE Trans Med Imaging. 2003; 22:986-1004. [Rabinowitz-2007]  Rabinowitz A, Bourdet B, Minville V, Chassery C, Pianezza A, Colombani A, Eychenne B, Samii K, Fourcade O. The paramedian technique: a superior initial approach to continuous spinal anesthesia in the elderly. Anesthe Analg 2007; 105: 1855-1857. [Rasoulian-2010] Rasoulian A, Rohling RN, Abolmaesumi P, Kamani AA, Lo C, Lessoway VA. Porcine thoracic epidural depth measurement using 3D ultrasound resliced images. Canadian anesthesia society 2010. [Riley-2007] Riley ET, Carvalho B. The episure syringe: a novel loss of resistance syringe for locating the epidural space. Anesth Analg 2007. 105:1164-6.  [Roche-2001]  Roche A, Pennec X, Malandain G,  Ayache N. Rigid registration of 3D ultrasound with MR images : a new approach combining intensity and gradient information. IEEE Trans on Medical Imaging 2001; 20:1038-1049. [Rodiera-1995]  Rodiera J, Calabuig R, Aliaga L, et al. Mathematical analysis of epidural space location. International Journal of Clinical Monitoring and Computing 1995. 12:213-7 [Rudin-1992] Rudin L, Osher S, Fatemi E. Nonlinear total variation based noise removal algorithms. Phys D 1992; 60:259-268. [Ruekert-1999]  Ruekert D, Sonoda LI, Hayes C, Hill DLG, Leach MO, Hawkes DJ. Nonrigid registration using free-form deformations: application to breast MR images. IEEE transactions on medical imaging 1999; 18:712-721. [Salter-2008] Salter BJ, Wang B, Szegedi MW, Rassiah-Szegedi P, Shrieve DC, Cheng R, Fuss M. Evaluation of alignment error due to a speed artifact in stereotactic ultrasound image guidance. Phys Med Biol 2008; 53: N437-N445. [Sandhu-2003]  Sandhu NS, Capan LM. Ultrasound-guided infraclavicular brachial plexus block. Br J Anaesth 2003; 90(1):107-8. [Scott-2004]  Scott DHT. Editorial II: the king of the blind extends his frontiers. British Journal of Anaesthesia 2004 93(2):175-177. [Shankar-1986]  Shankar PM. Speckle reduction in medical ultrasound via spatial compounding. IEEE Trans Ultrasonics Ferroelectrics Frequency Control. 1986; 33(6):754-758.  211 [Sharrock-1979] Sharrock NE. Recordings of, and anatomical explanation for false positive loss of resistance during lumbar extradural analgesia. Br J Anaesth 1979; 51:253-8. [Silverstein-1988]  Silverstein SD, O’Donnell M. Theory of frequency and temporal compounding in coherent imaging: speckle suppression and image resolution. J Opt Soc Am A 1988; 5(1):104-113. [Sites-2004]  Sites BD, Gallagher JD, Cravero J, Lundberg J, Blike G. The learning curve associated with a simulated ultrasound-guided interventional task by inexperienced anesthesia residents. Regional Anesthesia and Pain Medicine 2004; 29:544-548. [Small-2008]  Small SD, Wuerz RC, Simon R, Shapiro N, Conn A, Setnik GS. Demonstration of high-fidelity simulation team training for emergency medicine. Academic emergency medicine 2008; 6:312- 323. [Stamatakis-2005] Stamatakis E, Moka E, Siafaka I, Argyra E, Vadalouca A. Prediction of the distance from the skin to the lumbar epidural space in the Greek population, using mathematical models. Pain Practice 2005; 5:125-134. [Steinberg-2003]  Steinberg EL, Luger E, ARbel R, Menachem A, Dekel S. A comparative Roentgenographic analysis of the lumbar spine in mal army recruits with and without lower back pain. Clinical Radiology 2003; 58:985-989. [Tanaka-1993]  Tanaka K, Watanabe R, Harada T, Dan K. Extensive application of epidural anesthesia and analgesia in a university hospital: incidence of complications related to technique. Regional anesthesia and pain medicine 1993; 18:34-8. [Tanaka-2004]  Tanaka N, Ohkubo S, Takasaki M. Evaluation of a lockable combined spinal-epidural device for use with needle-through needle technique. Masui 2004;53:173–7. [Thévenaz-2000]  Thévenaz P, Blu T, Unser M. Interpolation revisited. IEEE Trans Med Imaging. 2000; 19:739-758. [Thirion-1998]  Thirion JP. Image matching as a diffusion process: an analogy with Maxwell's demons. Medical image analysis 1998; 2:243-260. [Trahey-1986a]  Trahey GE, Allison JW, Smith SW, Von Ramm OT. A quantitative approach to speckle reduction via frequency compounding. Ultrason Imaging. 1986; 8:151-164. [Trahey-1986b]  Trahey GE, Smith SW, Von Ramm OT. Speckle patterns correlation with lateral aperture translation: experimental results and implications for spatial compounding. IEEE Trans Ultrason Ferroelectr Freq Control 1986;33:257–64. [Tran-2007] Tran D, Kamani AA, Lessoway VA, Rohling RN. Adaptive spatial compounding for improving ultrasound images of the epidural space. Medical Imaging 2007: Ultrasonic Imaging and Signal Processing, edited by Stanislav Y. Emelianov, Stephen A. McAleavey, Proc. Of SPIE 2007, SPIE Press, Bellingham, WA; 6513-65130W:1-12. [Tran-2008a] Tran D, Hor KW, Kamani AA, Lessoway VA, Rohling RN. Adaptive spatial compounding for improving ultrasound images of  212 the epidural space on human subjects. Medical Imaging 2008: Ultrasonic Imaging and Signal Processing, edited by Stephen A. McAleavey, Jan D’hooge, Proc. Of SPIE 2008, SPIE Press, Bellingham, WA. 6920-6900I:1-12. [Tran-2008b]  Tran D, Kamani AA, Lessoway VA, Rohling RN. Comparison of loss-of-resistance along midline versus paramedian approach for epidural needle insertion in porcine subjects. Society for Obstetric Anesthesia and Perinatology (SOAP) Annual Meeting. Published in Anesthesia & Analgesia, Lippincot, Williams & Wilkins, Philadelphia, PA. Vol.106: pp. A-164, 2008. [Tran-2009a]  Tran D, Kamani AA, Lessoway VA, Peterson CE, Hor KW, Rohling RN. Preinsertion paramedian ultrasound guidance for epidural anesthesia. Anesth Analg 2009; 109:661-667. [Tran-2009b]  Tran D, Hor KW, Kamani AA, Lessoway VA, Rohling RN. Instrumentation of the loss-of-resistance technique for epidural needle insertion. IEEE Trans Biomed Eng 2009; 56:820-7. [Tran-2009c]  Tran D, Hor KW, Lessoway VA, Kamani AA, Rohling RN. Adaptive ultrasound imaging of the lumbar spine for guidance of epidural anesthesia. Computerized Medical Imaging and Graphics 2009; 33:593-601. [Tran-2010a]  Tran D, Kamani AA, Al-Attas E, Lessoway VA, Massey S, Rohling RN. Single-operator real-time ultrasound-guidance to aim and insert a lumbar epidural needle. Can J Anaesth 2010; 57:313- 21. [Tran-2010b]  Tran D, Rohling RN. Automatic detection of lumbar anatomy in ultrasound images of human subjects. IEEE Trans Biomed Eng 2010. (in press)  [Tran-2010c]  Tran D, Rohling RN, Kamani AA, Lessoway VA. Automatic detection of the ligamentum flavum on ultrasound images of the lumbar spine. Society for Obstetric Anesthesia and Perinatology (SOAP) 42nd Annual Meeting, San Antonio, Texas, 2010. [Tran-2010d]  Tran D, Rohling RN, Kamani A, Lessoway VA. Lo C, Lohser J. Ligamentum flavum detection in ultrasound of porcine thoracic spine. Canadian Anesthesiologist’s Society Annual Meeting, Montréal, Québec, 2010. [Tsui-1999]  Tsui BC, Gupta S, Finucane B. Determination of epidural catheter placement using nerve stimulation in obstetric patients. Reg Anesth Pain Med 1999;24:17-23. [Vas-2001]  Vas L, Raghavendran S, Hosalkar H, Patil B, A study of epidural pressures in infants. Paediatric Anesthesia 2001; 11(5):575-583. [Viola-1997]  Viola PA, Wells WM. Alignment by maximization of mutual information. International Journal of Computer Vision. 1997; 24(2):137-154. [Wagner-2004]  Wagner AL. CT Fluoroscopy-guided epidural injections: technique and results. Am J Neuroreadiol 2004; 25:1821-3.  213 [Wallace-1992] Wallace DH, Currie JM, GIlstrap LC, Santos R. Indirect sonographic guidance for epidural anaesthesia in obese pregnant patients. Reg Anesth 1992; 17:233-6. [Wantman-2006]  Wantman A, Hancox N, Howell NR. Techniques for identifying the epidural space: a survey of practice amongst anesthetists in the UK. Anesthesia 2006; 61: 370–375. [Watson-2007] Cochran S, Corner GA, Kirk KJ, Lines DIA, Watson MJ. Design and validation of an ultrasound array optimized for epidural needle guidance. IEEE Ultrasonics Symposium 2007; 2007 Oct 28 - Oct 31; New York, NY; 2255-2258. [Watts-1992]  Watts RW. A five-year prospective analysis of the efficacy, safety, and morbidity of epidural anesthesia performed by a general practitioner anesthetist in an isolated rural hospital. Anesth Intensive Care 1992; 20(3):348–353. [White-1980]  White AH, Derby R, Wynne G. Epidural injections for the diagnosis and treatment of low-back pain. Spine 1980 Jan-Feb; 5(1):78-86. [WHO-1997] World Health Organization. Obesity: Preventing and managing the global epidemic: Report of a WHO Consultation on Obesity. Geneva: 1997. [Wilhjelm-2000]  Wilhjelm JE, Jensen MS, Brandt T, Sahl B, Martinsen K, Jespersen SK, Falk E. Some imaging strategies in multi-angle spatial compounding.  Proc of the IEEE Ultrasonics Symposium 2000;2000 Oct 22 – Oct 25; San Juan, Puerto Rico; 2:1615-1618. [Willschke-2006]  Willschke H, Marhofer P, Bösenberg A, Johnston S, Wanzel O, Sitzwohl C, Kettner S, Kapral S. Epidural catheter placement in children: comparing a novel approach using ultrasound guidance and a standard loss-of-resistance technique. Br J Anaesth. 2006; 97:200-207. [Wilson-2007]  Wilson MJA, Epidural endeavour and the pressure principle. Anaesthesia 2007; 62(4):319-322. [Windisch-2009]  Windisch G, Ulz H, Feigl G. Reliability of Tuffier’s line evaluated on cadaver specimens. Surg Radiol Anat 2009; 31:627:30. [York-1999] York G, Kim Y. Ultrasound processing and computing: review and future directions. Annual Review in Biomedical Engineering 1999; 1: 559-588. [Zitova-2003] Zitova B, Flusser J. Image registration methods: a survey. Image and vision computing 2003; 21:977-1000.    214 Appendix A: Consent form  T H E  U N I V E R S I T Y  O F  B R I T I S H  C O L U M B I A       Subject Information and Consent Form Research into Ultrasound Guided Epidural Needle Insertions Principal Investigator:   Dr. Ali Kamani, M.D.     Department of Medicine     University of British Columbia     604-875-2158  Co-Investigators:  Dr. Robert Rohling, Ph.D., P.Eng. Department of Electrical and Computer Engineering     University of British Columbia     604-822-2045  Mr. King-Wei Hor. (M.A.Sc. student) Department of Electrical and Computer Engineering     University of British Columbia     604-822-2045  Mr. Brandon Lin (M.A.Sc. student) Department of Electrical and Computer Engineering     University of British Columbia     604-822-2045  Mr. Denis Tran (M.A.Sc. student) Department of Electrical and Computer Engineering     University of British Columbia     604-822-2045  Sponsors:  Canada Institute of Health Research & Natural Sciences and Engineering Research Council Emergency Telephone Number: 604-875-2158  215 1.  INTRODUCTION  You are being invited to take part in this research study because you are scheduled to receive anesthesia via an epidural needle insertion, or anesthesia via a spinal needle insertion for Caesarean section.  2.  YOUR PARTICIPATION IS VOLUNTARY  Your participation is entirely voluntary, so it is up to you to decide whether or not to take part in this study.  Before you decide, it is important for you to understand what the research involves.  This consent form will tell you about the study, why the research is being done, what will happen to you during the study and the possible benefits, risks and discomforts.  If you wish to participate, you will be asked to sign this form.  If you do decide to take part in this study, you are still free to withdraw at any time and without giving any reasons for your decision.  If you do not wish to participate, you do not have to provide any reason for your decision not to participate nor will you lose the benefit of any medical care to which you are entitled or are presently receiving.  Please take time to read the following information carefully and to discuss it with your family, friends, and doctor before you decide.  3.  WHO IS CONDUCTING THE STUDY?  The study is being conducted by Dr. Ali Kamani at the BC Women’s Hospital, and Dr. Robert Rohling at the University of British Columbia. It is sponsored by the Canada Institute for Health Research, in collaboration with the Natural Sciences and Engineering Research Council.  4.  BACKGROUND  Epidural needle insertions are common medical procedures. Often it is used as a way to deliver anesthesia for pain control, especially during childbirth. The doctors normally perform the needle insertions using their sense of touch to feel the bony parts of the spine and the resistance of the needle as it is inserted. Despite the relatively safe nature of the procedure, some drawbacks remain. One drawback is that doctors need to go through a learning stage where they develop their hand-eye coordination and how to interpret their sense of touch. The other drawback is the small risk of complications; the most common of which are headaches, backaches, lack of anesthesia control, and occasionally nerve damage.     216  5.  WHAT IS THE PURPOSE OF THE STUDY?  The purpose of this study is to investigate how ultrasound can help the doctor to choose a suitable puncture point and needle angle so that it enters the epidural space with high accuracy.  This study looks at various ultrasound techniques as a way of providing additional information to the doctor so that the procedure can be done with greater accuracy. If this study is successful, it will provide the basis for a new generation of ultrasound machines designed specifically to help in epidural needle insertions.  6.  WHO CAN PARTICIPATE IN THE STUDY?  Any patient who is expecting to receive an epidural needle insertion or Caesarean section can participate.  7.  WHO SHOULD NOT PARTICIPATE IN THE STUDY?  No restrictions.  8.  WHAT DOES THE STUDY INVOLVE?  If you agree to take part in this study, the procedures you can expect will include the following:  An initial ultrasound scan will be performed on your lower back immediately prior to the epidural needle insertion or Caesarian section. This will take approximately an extra 5 minutes, which may be done while waiting in the surgical daycare for Caesarean section subjects. The ultrasound data is saved for later study – it will not be used here to guide the anesthesiologist (Dr. Kamani).  One technical researcher may observe this ultrasound scan to save and store the ultrasound images.  The doctor then performs the epidural or combined spinal-epidural needle insertion in the standard way.  The doctor performing the needle insertion will also have a small sensor on their thumb under the glove to measure the force they press on the needle and the distance the syringe is pushed.  The technical researcher records and stores the sensor readings.  In some cases, the ultrasound may be used during the needle insertion to observe the needle and the anatomy. This may also take an extra 5 minutes.  Except for the extra time required for the ultrasound scans, you will not notice any difference in the regular needle insertion procedure.  The ultrasound images and the sensor recording will be saved anonymously on a computer, and then analyzed at a later date. Your name will not be recorded, but your age, sex, weight, height, ethnic group and complicating health issues will be recorded.  217  To assist the research, one or two persons (maximum two) will be present in addition to the anesthesiologist.  This person(s) is one of the following: a graduate student, a sonographer, or Dr. Rohling (the technical lead investigator).  The graduate students are from the Faculty of Applied Science and have received at least six months training in the area of ultrasound imaging. They will be using these studies for their research, but, like the others, will keep patient confidentiality. The sonographer (Ms. Vickie Lessoway) has more than ten years of experience in ultrasound scanning in a hospital. Dr. Rohling may be present simply to observe. No other people will be present during the epidurals.   9.  WHAT ARE MY RESPONSIBILITIES?  None.   10.  WHAT ARE THE POSSIBLE HARMS AND SIDE EFFECTS OF PARTICIPATING?  Since ultrasound does not pose any known health hazards, there are no known harms or side effects from participating in this study. The additional time required for the ultrasound scans should not have any effect on the delivery of anesthesia.  11.  WHAT ARE THE BENEFITS OF PARTICIPATING IN THIS STUDY?  Since much of the work is done later in the laboratory, there is no benefit to you. Ultrasound provides additional information for the doctor, but the doctor performing the procedure will not likely benefit from this information, since it needs further processing. We hope that the information learned from this study can be used in the future to benefit other people needing epidural needle insertions.  One of the possible benefits for patients in the future is the commercialization of the epidural guidance techniques. Ultrasonix Medical Corporation of Burnaby BC is supplying the ultrasound equipment for these tests. If the research is successful, Ultrasonix has expressed an interest in eventually licensing the ideas from UBC and making a commercial product to help perform epidurals elsewhere. No financial benefits will be provided to the subjects in this study.  12. WHAT HAPPENS IF I DECIDE TO WITHDRAW MY CONSENT TO PARTICIPATE?  Your participation in this research is entirely voluntary.  You may withdraw from this study at any time.  If you decide to enter the study and to withdraw at any time in the future, there will be no penalty or loss of benefits to which you are otherwise entitled, and your future medical care will not be affected.   218 The study doctor(s)/investigators may decide to discontinue the study at any time, or withdraw you from the study at any time, if they feel that it is in your best interests.  If you choose to enter the study and then decide to withdraw at a later time, all data collected about you during your enrolment in the study will be retained for analysis.  By law, this data cannot be destroyed.  Signing this consent form in no way limits your legal rights against the sponsor, investigators, or anyone else.  13.  WHAT HAPPENS IF SOMETHING GOES WRONG?  In the event you become injured or unexpectedly ill while participating in this study, necessary medical treatment will be available at no additional cost to you.  14. WHAT WILL THE STUDY COST ME?  Nothing. You will also not be paid for participating in this study.  15. WILL MY TAKING PART IN THIS STUDY BE KEPT CONFIDENTIAL?  Your confidentiality will be respected.  No information that discloses your identity will be released or published without your specific consent to the disclosure.  However, research records and medical records identifying you may be inspected in the presence of the Investigator or his or her designate by representatives of the UBC Research Ethics Board for the purpose of monitoring the research.  However, no records which identify you by name or initials will be allowed to leave the Investigators' offices.  20.  WHO DO I CONTACT IF I HAVE QUESTIONS ABOUT THE STUDY DURING MY PARTICIPATION?  If you have any questions or desire further information about this study before or during participation, you can Dr. Kamani at 604-875-2158.  21.WHO DO I CONTACT IF I HAVE ANY QUESTIONS OR CONCERNS ABOUT MY RIGHTS AS A SUBJECT DURING THE STUDY?  If you have any concerns about your rights as a research subject and/or your experiences while participating in this study, contact the Research Subject Information Line in the University of British Columbia Office of Research Services at 604-822-8598.  219 22. SUBJECT CONSENT TO PARTICIPATE   I have read and understood the subject information and consent form.  I have had sufficient time to consider the information provided and to ask for advice if necessary.  I have had the opportunity to ask questions and have had satisfactory responses to my questions.  I understand that all of the information collected will be kept confidential and that the result will only be used for scientific objectives.  I understand that my participation in this study is voluntary and that I am completely free to refuse to participate or to withdraw from this study at any time without changing in any way the quality of care that I receive.  I understand that I am not waiving any of my legal rights as a result of signing this consent form.  I understand that there is no guarantee that this study will provide any benefits to me.  I have read this form and I freely consent to participate in this study.  I have been told that I will receive a dated and signed copy of this form.    SIGNATURES      Printed name of subject    Signature  Date      Printed name of witness    Signature   Date      Printed name of principal investigator/ designated representative    Signature   Date  220 Appendix B: Ethics approval   221 Appendix C: Pressure modeling  The sensors had previously been used in [Hor-2007a]. The sensors were calibrated to obtain the conversion from voltage to N (for the force sensor), mm (for the position sensor) and kPa (for the pressure sensor). The calibration shows that for these sensors  NVF forceouta 166.2707.5 ,         eq C-1  kPaVP pressureout 409.101.24 ,         eq C-2  mmVD ntdisplacemeout 86.3108.26 ,         eq C-3 where Fa is the force applied by the thumb to the syringe plunger, Vout,force is the voltage received from the force sensor to the data acquisition board, P is the pressure between the needle and the syringe and Vout,pressure is the voltage received from the pressure sensor to the data acquisition board, and D is the position of the plunger with respect to the syringe, Vout, displacement is the voltage received from the position sensor to the data acquisition board.  Experiment 1: Sensor reliability  The sensors were then tested for confirming the LOR in porcine tissues. The sensors were first evaluated to ensure the reliability of detection of the LOR endpoint. For these tests, the anesthesiologist performed a needle insertion using LOR with the sensors measuring the 3 parameters of interest at L2-3 and L3-4. Upon entry into the epidural space, the  222 anesthesiologist verbally communicated the success and a time stamp was recorded on the computer recording the sensor measurements. This time stamp was then compared to the time LOR was observed on the sensor signals. Both the continuous and the intermittent techniques were evaluated by performing 5 needle insertions with each technique at each of the two lumbar intervertebral spaces (L2-3 and L3-4). This gives n=10 for each technique. The midline approach was used for all insertions. Since the LOR endpoint could be clearly observed as a rapid drop in all three sensors, the times of the drops were averaged and compared to the time the anesthesiologist indicated LOR.  The time believed to be associated with the detection of the LOR using each of the three sensors is compared to the anesthesiologist’s confirmation of entry into the epidural space. Both the intermittent and continuous techniques show a clear LOR as shown by a rapid fall in values. The errors are similar for both techniques. Combining all measurements, the overall average of the difference between the time indicated by the sensors and the anesthesiologist is 0.8±0.3s. This is interpreted as being the time taken by the anesthesiologist to confirm the feel is indeed the loss of resistance.  Experiment 2: Pressure estimate for both continuous and intermittent techniques on porcine tissue The pressure is believed to be more closely related to the properties of the tissues where the needle tip is located because it is what the anesthesiologist attempts to feel when using the LOR technique. The saline comes in contact with the pressure sensor so  223 sterilization is difficult and the pressure sensor cannot be used for clinical trials on human subjects. By deriving a model, it may be possible to estimate the pressure from the measured force and displacement. The simplest model is a static model: A tFk tP aa )( )(           eq C-4 where P(t) is the estimated pressure, Fa(t) is the force applied from the thumb to the plunger, A is the area of the cross-section of the syringe and ka is a unit-less constant which accounts for losses such as friction, viscosity and off-axis forces. ka is determined empirically from bench-top tests using a range of test forces to be 0.900 [Hor-2005a]. The value of ka is determined by correlating 3 applied forces (6.05N, 14.86N and 20.90N) and the associated measured pressure taking into account the dimensions of the syringe. A linear fit is applied to the points and the slope is found to be 0.900. It is observed that there is some leakage at the plunger-barrel interface, which causes a drop in the pressure over time for a constant force. The pressure decays in regions where there is force on the plunger but no displacement. A decay term is added to the model in order to more closely estimate the pressure:        0)( 0)( )( dt dDe A tFk dt dD A tFk tP itt aa aa         eq C-5 where ti is the time at which the plunger stops moving and τ is a decay time constant determined empirically on bench-top tests with a closed needle to be 23±8s [Hor-2007a].  224 The plunger is considered stationary for plunger displacements smaller than 0.18mm/s (or 21mm3/s), just beyond the noise level of the sensors. As mentioned, there are two standard methods of applying pressure with the saline-filled syringe: the continuous pressure technique and the intermittent pressure technique. Given two models and two techniques, tests were performed on excised porcine tissue to determine the most accurate of the four combinations. Using a similar protocol to Experiment 1 for tissue preparation, the estimated pressure values were compared to the actual pressure measurements. Needle insertions (n=5) were performed using each of the continuous and intermittent technique at L2-3 and L3-4. Knowledge of the porcine anatomy and verbal indication from the anesthesiologist allowed different portions of sensor measurements to be associated with the interspinous ligament, ligamentum flavum and the epidural space where the LOR endpoint occurred. Volume flow rate was calculated for each point in time by multiplying the rate of change of displacement by the syringe cross-section (a constant). Average flow rate was then calculated for each tissue type.  Examples of the sensor signals when using the intermittent technique and the continuous technique are shown on Fig. C-1 and Fig. C-2 respectively. For comparison, Fig. C-1d shows the estimated pressure using the static model and Fig. C-1e shows the estimated pressure using the decay model. Figures 6d and C-2e can be compared similarly. The errors for each of the 10 trials are shown in Fig. C-3 and are summarized in Table C-1. It is observed that when using the continuous pressure technique for LOR to saline, the error of the estimated pressure from the static model compared to the actual pressure  225 sensor signal (6.89kPa RMS error) is significantly larger (p<0.05 using the paired t-test) than the error of the estimated pressure from the decay model compared to the actual pressure sensor signal (4.8kPa RMS error). These represent 20% and 14% errors of the peak pressure for the static and decay models respectively. When using the intermittent technique, both models estimate the pressure sensor with a higher level of error (8.3kPa RMS error), which is 24% of the peak pressure.  0 2 4 6 8 10 12 14 16 18 0 1 2 3 4 5 6 7 8 9 10 Time (s) Fo rc e (N ) a) 0 2 4 6 8 10 12 14 16 18 5 10 15 20 25 30 35 Time (s) D is pl ac em en t D  (m m ) b) 0 2 4 6 8 10 12 14 16 18 0 10 20 30 40 50 60 70 Time (s) Pr es su re  (k Pa ) c) 0 2 4 6 8 10 12 14 16 18 0 10 20 30 40 50 60 70 Time (s) Pr es su re  (k Pa )  d) 0 2 4 6 8 10 12 14 16 18 0 10 20 30 40 50 60 70 Time (s) Pr es su re  (k Pa )  e) Figure C-1 Example of epidural needle insertion on porcine tissue using the intermittent technique of a) force, b) displacement, c) measured pressure, d) estimated pressure using the static model, and e) estimated pressure using the decay model.  226   a) b) c)  d)  e) Figure C-2 Example of epidural needle insertion on porcine tissue using the continuous technique of a) force, b) displacement, c) measured pressure, d) estimated pressure using the static model, and e) estimated pressure using the decay model.   227 0.0 2.0 4.0 6.0 8.0 10.0 12.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Trial R M S Er ro r ( kP a) Static Decay Continuous Intermittent  Figure C-3 Root-mean-square (RMS) error between the measured pressure and the pressure calculated from the decay model, and between the measured pressure and the pressure calculated from the static model.Both the continuous and intermittent techniques are compared on porcine tissue.   Table C-1 Experiment 2: root-mean square (RMS) error for the continuous and intermittent techniques using either a static pressure model or decay pressure model.  Mean Error Static model RMS Error Static model Mean Error Decay model RMS Error Decay model Continuous 0.69±5.5 kPa 6.89 kPa -0.69±4.1 kPa 4.8 kPa Intermittent 1.4±6.89 kPa 8.3 kPa 1.2±6.89 kPa 8.3 kPa  Table C-2 shows the different average flow rates, forces and pressures as well as the estimated pressures using the decay model for the porcine tissue. The flow rate is shown to be significantly higher (p<0.05 using the paired t-test) for the interspinous ligament (29±9mm3/s) compared to the flow rate in the ligamentum flavum (9±7mm3/s). The force  228 and pressure are not significantly different so the differences come from the tissue properties, not the operator. The estimated pressure is also not significantly different from the measure pressure from the sensors giving confidence that the decay model can accurately estimate the tip pressure using the force and displacement measurements. Table C-2 Experiment 2: flow rate, force (Fa) and pressure (P) for the interspinous ligament and ligamentum flavum for porcine subjects using the midline approach.  Region Flow rate (mm3/s) Fa(N) Max Fa(N) P (kPa) Max P (kPa) Pest (kPa) Max Pest (kPa) interspinous ligament 29±9 2.7±1.6 4.5±1.6 20±10 31±13 20±11 34±13 ligamentum flavum 9±7 3.3±1.4 4.1±1.5 27±6 30±7 25±7 32±10  229 Appendix D: Warping and linear prediction  The previous work on warping and median-based compounding [Groves-2004][Hor- 2007a] is further explained here.  Here is the overall block diagram followed by justification of design decisions.   Figure D-1 Block diagram of the adaptive spatial compounding algorithm  There are many registration methods available. Spatial compounding is expected to be used in real-time systems and so low computational cost is an important aspect. Also, because the deformation caused by patient movement, transducer motion, speed of sound and refraction is not expected to be large, the registration should be limited to small deformations. This means a method that can register small deformations with a small computational cost is desired. Finally, because the features are bones and ligaments, and the rest of the image contains mostly speckle, a registration method which focuses on local anatomical details (but larger than speckle) is needed. A block-matching method is used because it satisfies these criteria.  Each block from the reference image is matched to the beam-steered image by using the crosscorrelation coefficient. The crosscorrelation coefficient is sensitive to small bright features in the image and is therefore suitable for our needs as the LF is a small bright echo. The block-matching displacement vectors are interpolated so that each pixel has a unique warping vector and so the warping vector field is constructed.  230  The beam-steered image is warped to have the features best match the reference image. Inverse mapping is performed to avoid the typical issues associated with forward mapping such as having multiple pixels mapped to certain pixels and other pixels not having any pixels mapped to that location. To perform inverse mapping, a pixel at the location of the reference image needs to get a coordinate for where it would obtain its value in the beam-steered image from. Therefore, rather than obtaining the warping vector field for matching the beam-steered image to the reference image, the warping field for matching the reference image to the beam-steered image is calculated and applied to inverse map the beam-steered image.   Figure D-2 The warped beam-steered image and the reference frame interleaved in a checkered pattern with block size corresponding to the block size used for block matching. The blocks can be better observed around the edges of the ultrasound image.   The linear prediction algorithm LP2+ starts the block-matching at the location where there are most features. The Canny edge detector is used to find features here as it is known to give one single line for each edge. Then, an initial guess is assigned for each  231 neighboring block and the block-matching with initial guess is performed in a spiral around the location of the first block.   Figure D-3 The order of block-matching displacement vector estimation. Starting from the location of most features and following a spiral for initial guess and estimation.

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0064993/manifest

Comment

Related Items