UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Ultrasound elastography for intra-operative use and renal tissue imaging Schneider, Caitlin Marie 2017

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_may_2017_schneider_caitlin.pdf [ 9.38MB ]
JSON: 24-1.0343641.json
JSON-LD: 24-1.0343641-ld.json
RDF/XML (Pretty): 24-1.0343641-rdf.xml
RDF/JSON: 24-1.0343641-rdf.json
Turtle: 24-1.0343641-turtle.txt
N-Triples: 24-1.0343641-rdf-ntriples.txt
Original Record: 24-1.0343641-source.json
Full Text

Full Text

Ultrasound Elastography for Intra-Operative Use andRenal Tissue ImagingbyCaitlin Marie SchneiderB.Sc., Johns Hopkins University, 2009M.A.Sc., The University of British Columbia, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORALSTUDIES(Electrical and Computer Engineering)The University of British Columbia(Vancouver)April 2017c© Caitlin Marie Schneider, 2017AbstractThe kidney is a vital organ within the human body and improvements in the abil-ity to characterize the kidney tissue can create benefits for patients with kidneytumours and for kidney transplant recipients. Often, changes in tissue health ordevelopment of cancer are manifested in changes in tissue structure that affect tis-sue elastic properties. For example, the cancerous tissue of renal cell carcinoma isstiffer than healthy kidney tissue, and the development of fibrosis, which impairskidney function, also causes the tissue to become stiffer over time. These changescan be imaged with ultrasound elastography, a technique for quantitatively assess-ing tissue elasticity. If proven effective, elastography tissue characterization canreplace biopsy.The ultrasound elastography method used in this thesis, called Shear WaveAbsolute Vibro-Elastography, or SWAVE, measures the wavelength of inducedsteady-state multi-frequency mechanical shear waves to calculate tissue elasticity.SWAVE can employ standard ultrasound transducers that image the kidney thoughthe skin above the organ, or custom miniaturized transducers that are placed di-rectly on the surface of the organ during surgery. The accuracy of SWAVE is vastlyimproved by the use of 3D ultrasound data.We propose and evaluate 3D SWAVE imaging based on the use of a trackedintra-operative ultrasound transducer designed for use with the da Vinci Robot.Different tracking methods are evaluated for future intra-operative use. Elasticityimages of tissue phantoms are obtained using interpolated 3D tissue displacementdata acquired with the da Vinci robot and the intra-operative transducer. The useof tracked ultrasound transducer opens the way for introducing registered preoper-ative imaging, including elastography, to improve surgical guidance.iiDifferent methods of characterizing kidney tissue using SWAVE imaging areexamined. The elastic and viscous properties are estimated kidney tissue ex-vivo.The effect of arterial pressure on the measured kidney elasticity is characterized. Itwas found that increasing input pressure increases the measured elasticity. Finally,ultrasound and ultrasound elastography are applied to kidney transplant recipientsin-vivo to assess the level of fibrosis development. A preliminary study indicatesthat it is possible to transmit shear waves into the transplanted kidney and measurethe elastic properties of the kidney tissue.iiiPrefaceA version of the material presented in Chapter 2 was published as:Caitlin Schneider, Christopher Nguan, Robert Rohling, Septimiu Sal-cudean. “Tracked ‘Pick-up’ Ultrasound for Robot-assisted MinimallyInvasive Surgery”. IEEE Transactions on Biomedical Engineering, 63(2), 2016.I was responsible for all the data collection and analysis. Drs Nguan, Rohlingand Salcudean advised on data analysis and manuscript editing.The material presented in Section 3.3 is based on material orally presented atand in the proceedings of the Medical Image Computation and Computer AidedInterventions (MICCAI) conference as:Caitlin Schneider, Ali Baghani, Robert Rohling, Septimiu Salcudean.“Remote ultrasound palpation for robotic interventions using absoluteelastography”. MICCAI LNCS: Vol. 7510, 2012.For this publication, I was responsible for the integration of the robot APIcode into the elastography scanning system, data collection and analysis. Dr. AliBaghani worked on the development of the phasor fitting and Local Frequency Es-timation (LFE) C++ and CUDA code, and all authors contributed to the manuscriptediting.The second experiment described in Chapter 3, Section 3.4 has been submittedand is in process:ivCaitlin Schneider, Christopher Nguan, Jeff Abeysekera, Julio Lobo,Robert Rohling, Septimiu Salcudean. “Intra-operative shear wave elas-tic imaging for robotic interventions”.I was responsible for all the integration between the dVRK (da Vinci robotsoftware) and the current elastography software (eScan). Jeff Abeysekera and JulioLobo wrote the elastography software. I completed all the data collection andanalysis, as well as writing the manuscript. Drs Salcudean and Rohling supervisedand edited the manuscript.Chapter 5 contains material that was presented at two conferences in 2016.Section 5.4 contains material presented as:Caitlin Schneider, Julio Lobo, Mohammad Honarvar, Samir Bidur,Robert Rohling, Septimiu Salcudean. “Blood Pressure DependentElasticity Measurements of Porcine Kidney ex-vivo”. IEEE Ultrason-ics, Ferroelectrics and Frequency Control, International UltrasonicsSymposium. 2016.I was responsible for the coordination of the kidney harvesting and assisted bySamir Bidur. I completed all the data collection, while Julio Lobo and MohammadHonarvar wrote and developed the elastography software. Drs Nguan, Rohling andSalcudean advised on data analysis and manuscript editing.The second half of Chapter 5 contains material that as:Caitlin Schneider, Mohammad Honarvar, Robert Rohling, SeptimiuSalcudean, Christopher Nguan. “Apparatus for Imaging and ModelFitting of ex-vivo Porcine Kidney”. Fifteenth International TissueElasticity Conference.Mohammad Honarvar constructed the apparatus for holding the ex-vivo kidneyand processed the Finite Element Model (FEM) results. I was responsible for theexperimental design, kidney harvest, data collection, processing the results for theLFE method and data analysis. Drs Nguan, Rohling and Salcudean advised on dataanalysis and manuscript editing.vTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiGlossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Overview and Motivation . . . . . . . . . . . . . . . . . . . . . . 11.2 The Kidney . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2.1 Kidney Anatomy . . . . . . . . . . . . . . . . . . . . . . 21.2.2 Kidney Cancer and Partial Nephrectomy . . . . . . . . . . 31.2.3 Kidney Transplant and Transplant Fibrosis . . . . . . . . 71.3 Ultrasound Imaging . . . . . . . . . . . . . . . . . . . . . . . . . 101.4 Ultrasound Elastography . . . . . . . . . . . . . . . . . . . . . . 131.5 Shear Wave Absolute Vibro-Elastography (SWAVE) . . . . . . . . 151.5.1 Overview of Elastography Methods . . . . . . . . . . . . 171.6 Minimally Invasive Surgery, Intra-operative Ultrasound and the daVinci Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18vi1.6.1 Minimally Invasive Surgery . . . . . . . . . . . . . . . . 181.6.2 Intra-operative Ultrasound . . . . . . . . . . . . . . . . . 201.6.3 Robot-assisted Surgery with the da Vinci Surgical System 201.6.4 Intra-operative Ultrasound for the da Vinci . . . . . . . . 231.6.5 Intra-operative Ultrasound Elastography . . . . . . . . . . 281.7 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 291.8 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 302 Ultrasound Transducer Tracking and 3D Reconstruction . . . . . . 322.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332.2.1 Transducer Tracking . . . . . . . . . . . . . . . . . . . . 332.2.2 Repeatability Tests . . . . . . . . . . . . . . . . . . . . . 342.2.3 Tracking Tests . . . . . . . . . . . . . . . . . . . . . . . 362.2.4 Image Processing, 3D Reconstructions and Registration . 372.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392.3.1 Repeatability . . . . . . . . . . . . . . . . . . . . . . . . 392.3.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . 402.3.3 3D Reconstructions and Registration . . . . . . . . . . . . 412.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513 Elastography using the da Vinci Surgical Robot . . . . . . . . . . . 523.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.2 Elastography . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.3 Experiment 1: Freehand Technique in Clinical Mode . . . . . . . 563.3.1 Experimental Setup and Results . . . . . . . . . . . . . . 573.4 Experiment 2: Automated 3D Elastography with da Vinci ResearchKit (DVRK) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.4.1 Robotic Integration . . . . . . . . . . . . . . . . . . . . . 583.4.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . 613.4.3 Experiment 2: Results . . . . . . . . . . . . . . . . . . . 643.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68vii3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 714 Review of Elastography Measurements of Kidney . . . . . . . . . . 734.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.2 Type of Elastography Imaging . . . . . . . . . . . . . . . . . . . 754.2.1 Strain Ratios . . . . . . . . . . . . . . . . . . . . . . . . 754.2.2 Acoustic Radiation Force Impulse (ARFI) and SuperSonicImagine (SSI) . . . . . . . . . . . . . . . . . . . . . . . . 784.2.3 FibroScan . . . . . . . . . . . . . . . . . . . . . . . . . . 784.2.4 Magnetic Resonance Elastography (MRE) . . . . . . . . . 824.3 Challenges of Kidney Imaging . . . . . . . . . . . . . . . . . . . 854.3.1 Patient Heterogeneity . . . . . . . . . . . . . . . . . . . . 854.3.2 Kidney Structure . . . . . . . . . . . . . . . . . . . . . . 854.3.3 Perfusion . . . . . . . . . . . . . . . . . . . . . . . . . . 864.3.4 Transducer Pressure . . . . . . . . . . . . . . . . . . . . 884.3.5 Viscous Properties of Tissue . . . . . . . . . . . . . . . . 884.3.6 Incongruent Results . . . . . . . . . . . . . . . . . . . . 884.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 895 Tissue Characteristics of Porcine Kidney ex-vivo . . . . . . . . . . . 925.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 925.2 Imaging Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . 935.3 Kidney Harvest . . . . . . . . . . . . . . . . . . . . . . . . . . . 945.4 Kidney Pressure Tests . . . . . . . . . . . . . . . . . . . . . . . . 955.4.1 Infusion Pump . . . . . . . . . . . . . . . . . . . . . . . 955.4.2 Kidney Pressure Results . . . . . . . . . . . . . . . . . . 985.5 Viscous Properties of Kidney Tissue . . . . . . . . . . . . . . . . 985.5.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 1015.5.2 Tissue Models . . . . . . . . . . . . . . . . . . . . . . . 1015.5.3 Viscosity Results . . . . . . . . . . . . . . . . . . . . . . 1035.5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 1105.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1135.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116viii6 Measurements of Transplant Kidneys in vivo . . . . . . . . . . . . . 1176.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1176.2 Patient Populations . . . . . . . . . . . . . . . . . . . . . . . . . 1186.3 Patient Histology . . . . . . . . . . . . . . . . . . . . . . . . . . 1186.4 Imaging Procedure . . . . . . . . . . . . . . . . . . . . . . . . . 1206.5 Patient Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1236.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1286.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1307 Conclusions and Future Work . . . . . . . . . . . . . . . . . . . . . 1317.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1317.2 Summary of Findings . . . . . . . . . . . . . . . . . . . . . . . . 1317.3 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1337.4 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1357.5 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1367.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140A Supporting Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 161A.1 dVRK and its Accuracy . . . . . . . . . . . . . . . . . . . . . . . 161A.2 Individual Model Fitting Results . . . . . . . . . . . . . . . . . . 169ixList of TablesTable 2.1 Standard deviations of the fixed checkerboards, used to deter-mine the repeatability of the wide baseline optical camera track-ing system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Table 2.2 Position Errors for Hand-eye Calibration between the da Vinci(dV), electromagnetic sensor (EM) and Optotrak . . . . . . . . 42Table 2.3 Position errors for the hand-eye calibration of the Optotrak andda Vinci stereo camera . . . . . . . . . . . . . . . . . . . . . . 42Table 2.4 Reconstruction and registration errors for the volumes of thevessel phantom and targets collected with the da Vinci as a tracker. 44Table 2.5 Reconstruction and registration errors for the volumes of thevessel phantom and targets collected with the EM sensor as atracker. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45Table 2.6 Reconstruction and registration errors for the volumes of thevessel phantom and targets collected with the stereo camera asthe tracker. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46Table 3.1 Results from the CIRS QA Elastography Phantom . . . . . . . 58Table 3.2 The results of the elastogram volumes taken from CIRS phan-tom and compared to results of the same phantom capturedwith a 3D motorized ultrasound transducer, the Siemens VirtualTouch Image Quantification (VTIQ) system and the manufac-turer specifications . . . . . . . . . . . . . . . . . . . . . . . 64xTable 4.1 Existing studies using strain and strain ratios which are relevantto kidney fibrosis stiffness measurements. . . . . . . . . . . . 77Table 4.2 Existing studies using FibroScan relevant to kidney fibrosis stiff-ness measurements. . . . . . . . . . . . . . . . . . . . . . . . 81Table 4.3 Existing MRE studies relevant to kidney fibrosis stiffness mea-surements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84Table 5.1 Results from LFE analysis for the Voigt and Maxwell modelspresented. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106Table 5.2 Results from LFE analysis for the Zener model presented. . . . 107Table 5.3 Results from FEM analysis for the Voigt and Maxwell modelspresented. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108Table 5.4 Results from FEM analysis for the Zener model presented. . . 109Table 5.5 The standard deviations in the parameters for each of the threemodels after 10 trials of up to ±10% perturbations in the origi-nal data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111Table 6.1 Characteristics of the patients in this study. . . . . . . . . . . . 119Table 6.2 The Voigt model fitting results for patients. Each cross sectionalview for each patient is represented separately. The top halfof the table represents the long axis view and the lower halfrepresents the short axis view of the kidney. . . . . . . . . . . 128Table A.1 TLE for Lego locations 1-10 . . . . . . . . . . . . . . . . . . 166Table A.2 TLE for Lego locations 11-36 . . . . . . . . . . . . . . . . . . 167Table A.3 The mean and standard deviation of the training and testing datafor these 50 registrations . . . . . . . . . . . . . . . . . . . . . 167xiList of FiguresFigure 1.1 A diagram of the kidney anatomy (right) and their placementwithin the abdomen (left). Image credit: philschatz.com Anatomyand Physiology [122]. . . . . . . . . . . . . . . . . . . . . . 3Figure 1.2 The detailed diagram of the nephron structure including theGlomerulus, Bowman’s capsule and filtrate flow. Image Credit:Medical Science Navigator [135]. . . . . . . . . . . . . . . . 4Figure 1.3 The transplant kidney is placed in the pelvis rather than inplace of the failing kidney. It is attached to the iliac arteryand vein. Image Credit: University of Utah Healthcare [119]. . 7Figure 1.4 The pitch is the spacing between the piezoelectric crystals. Thethree directions of the image are defined as axial (away fromthe probe), lateral (along the direction of the crystal array) andelevational (out of plane). . . . . . . . . . . . . . . . . . . . . 12Figure 1.5 Left: Acquisition timeline for two frames of B-mode data fora twelve scan-line field of view. Right: Acquisition timelinefor two frames of high-frame-rate acquisition data for a fourscan-line field of view. T is the time required for acquisitionof a single RF line. . . . . . . . . . . . . . . . . . . . . . . . 16Figure 1.6 The da Vinci Surgical System. Image courtesy of Intuitive Sur-gical Inc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Figure 1.7 Example ultrasound images created with the intraoperative “pick-up” transducer of the carotid artery and vein. Left: B-modeimage. Right: Doppler image of the arteries after bifurcation. . 25xiiFigure 1.8 Rendered images of the da Vinci interface. . . . . . . . . . . 25Figure 1.9 Top: Diagram showing the cross section of the “pick-up” trans-ducer, and the location of the electromagnetic (EM) sensor andcrystal stack. Bottom: a photograph of the ultrasound trans-ducer, with checker-boards used for camera tracking. . . . . . 27Figure 1.10 Rendered images of the “pick-up” transducer. Left and middle:cross sectional view (blue) of the lap-handle. The angled facesand locking pins can be seen. Right: the tool fits tightly againstthe angled faces. The practicality of adding visual trackingmarkers is demonstrated. . . . . . . . . . . . . . . . . . . . . 28Figure 1.11 The pick-up intra-operative transducer is grasped and manipu-lated by the Pro-GraspTM tool of the da Vinci Surgical System. 28Figure 2.1 Grasping Repeatability testing set-up. The transducer was pickedup 30 independent times and the transformation between thecheckerboard on the tool jaws and on the transducer was cal-culated. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Figure 2.2 Vessel Phantom. Top left: B-mode, top right: CT, bottom left:Power Doppler and bottom right: Phantom photo . . . . . . . 38Figure 2.3 Grasping Repeatability. The mean-centered components (cen-tered at zero) of the transformation between the tool jaws andthe transducer computed using stereo tracking. See Figure 2.1. 41Figure 2.4 Top: Examples of the vessel phantom reconstructed from eachmodality, from left to right, the Computed Tomography (CT)scan, the da Vinci kinematics, the Electromagnetic (EM) sen-sor and the stereo camera tracking. The blue stars representthe locations of the ball bearing targets for each of the sensorbased reconstructions. Bottom: An example of the CT volumeregistered with a volume reconstructed using the da Vinci kine-matics. The red mesh and stars represent the surface of the CTvolume and the ball bearing targets, while the blue mesh andstars represent the da Vinci reconstructed volume. . . . . . . . 48xiiiFigure 3.1 A) Elastography set-up for the 3D transducer method. B) Set-up for the extension to freehand scanning using a 2D trans-ducer and the da Vinci robot. In both cases, sector subdivisionhigh-frame rate imaging is applied. . . . . . . . . . . . . . . 55Figure 3.2 Left: Image of 6 kPa CIRS phantom lesion. Right: Image ofthe 54 kPa CIRS phantom lesion. The colour bar shows theelasticity in kilopascals (kPa). . . . . . . . . . . . . . . . . . 57Figure 3.3 A series of phasor volumes for the same trajectory in the CIRSphantom, of the stiff lesion. The different wave patterns can beseen at each different frequency. . . . . . . . . . . . . . . . . 58Figure 3.4 The transducer used in this study was custom designed for usewith the da Vinci Robot [143]. It has 128 elements, an imagingface that is 28 mm wide and a diameter of 15mm. . . . . . . . 59Figure 3.5 Example trajectory from the ex vivo kidney scans. The originaltrajectory of the da Vinci tool (top), the resampled trajectory(middle) and the achieved trajectory as reported by the da Vinci(bottom). All locations are reported in millimetres and the x,y, and z directions are indicated by red, blue and green linesrespectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . 62Figure 3.6 Top: The first six frames from the resampled trajectory andthe achieved trajectory overlaid from Figure 3.5. The desiredtransforms are shown in solid lines, while the achieved trans-forms are shown with dotted lines. The magenta lines betweenthe points indicate corresponding points within the trajectory.Bottom: A graph showing the values of the position and rota-tional errors for the six frames. . . . . . . . . . . . . . . . . . 65Figure 3.7 Results from the stiff lesion of the CIRS phantom. The re-constructed B-mode image of the phantom (top). The recon-structed real part of the phasor image (middle) and the result-ing 3D elastogram (bottom). The blue sphere outline desig-nated the area of the inclusion used to calculate the elasticity.Phasors are shown on a scale from -5 microns to 5 microns andthe elastogram is shown on a scale for 0 to 60 kPa. . . . . . . 66xivFigure 3.8 The cross sectional graphs of the elastogram taken throughthe center of the stiff inclusion in the phantom with the 2Dtransducer and da Vinci system. The red line is the calcu-lated Young’s Modulus and the vertical blue lines indicatedthe edges of the stiff inclusion. . . . . . . . . . . . . . . . . 67Figure 3.9 The B-mode (left), real part of the phasor image (center) andelastogram (right) of the ex vivo kidney. Phasors are shownon a scale from -3 microns to 3 microns and the elastogram isshown on a scale from 0 to 60 kPa. . . . . . . . . . . . . . . . 67Figure 5.1 A) A diagram of the shaker set-up. B) The Solidworks ren-dering of the set-up, showing the frame, ultrasound transducerand the kidney. . . . . . . . . . . . . . . . . . . . . . . . . . 93Figure 5.2 The pump and flushing set-up. The preservation solution ispumped into the kidney through a cannula which is insertedinto the renal artery. A pressure gauge is in the system to mea-sure the input pressure into the kidney. . . . . . . . . . . . . . 96Figure 5.3 A diagram that shows the cross section of the kidney that isimaged during these experiments. . . . . . . . . . . . . . . . 96Figure 5.4 The relationship between measured elasticity and excitationfrequency for three different input renal pressures. Connect-ing lines are used for visualization purposes only. . . . . . . . 97Figure 5.5 Top: a histogram of the average elasticity for each specimen.The average elasticity is measured at three different input pres-sures, zero, simulated diastolic and simulated systolic. Bot-tom: The elasticity, estimated at 125Hz, for all the kidneyspecimens at each of their measured input pressures. . . . . . 99Figure 5.6 Cross-sectional images of the B-mode image of the kidney andthe resulting elastograms at each input pressure. . . . . . . . . 100Figure 5.7 There are three rheological models used in these studies. a)Voigt, b) Maxwell and c) Zener. The springs represent the elas-tic components and the dash-pots represent the viscous com-ponents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102xvFigure 5.8 The graphic results of the LFE model fitting for all the experi-ments. The stars represent the experimental data, the red, blueand green lines represent the results of the Voigt, Maxwell andZener Models respectively. . . . . . . . . . . . . . . . . . . 104Figure 5.9 The graphic results of the FEM model fitting for all the experi-ments. The stars represent the experimental data, the red, blueand green lines represent the results of the Voigt, Maxwell andZener Models respectively. . . . . . . . . . . . . . . . . . . . 105Figure 5.10 Cross-sectional images of the B-mode image of the kidney atdifferent input pressures. The bars on the right side each crosssection denote the two surfaces of the kidney. The width of thekidney increases with pressure. The width was measured as 2cm at 0 mmHg, 2.25 cm at 80 mmHg, 2.75 cm at 130 mmHgand 3.25 cm at 155 mmHg. . . . . . . . . . . . . . . . . . . . 115Figure 6.1 Two example ultrasound images of the transplant kidneys, longaxis views. Top: Patient 3. Bottom: Patient 4. . . . . . . . . . 121Figure 6.2 An example sketch of the placement of the ultrasound trans-ducer (blue oval) and shaker (red circle) on the lower abdomenof the patient for both the long axis and short axis views. . . . 122Figure 6.3 A flow chart describing the steps of image processing. . . . . 123Figure 6.4 The patients’ estimated glomerular filtration rate vs the BanffScore from the most recent biopsy. . . . . . . . . . . . . . . . 124Figure 6.5 The average measured Young’s modulus vs the estimated Glomeru-lar Filtration Rate (GFR) for both cross sectional views for allpatients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124Figure 6.6 B-mode and wave images of Patient 4 at each frequency usedin this study. The outline of the cortex (solid line) and collect-ing system (dotted line) have been overlaid on the images. . . 126Figure 6.7 The measured Young’s modulus of the cortex and collectingsystem vs the excitation frequency. . . . . . . . . . . . . . . . 127xviFigure 6.8 The patient results measured Young’s modulus as compared tothe patients’ blood pressure taken on the day of the ultrasoundscan. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127Figure A.1 An image of the experimental setup. The OptoTrak stylus wasplaced in a divot in the center of the custom Lego, while theMicro Forceps were placed in a divot to the left. . . . . . . . . 163Figure A.2 A sketch of the tip of a da Vinci Instrument and the transformsthat describe the transforms that describe the orientation of thetool. The control point is located at O6. This image is takenfrom the DVRK user guide. . . . . . . . . . . . . . . . . . . . 164Figure A.3 Point distributions of the locations defined by the DVRK. . . . 165Figure A.4 Resulting registration, in the OptoTrak coordinate frame. Thered stars represent the Lego locations defined by the OptoTrakand the black diamonds represent the DVRK locations after be-ing multiplied by this average transform. . . . . . . . . . . . 168Figure A.5 Individual model fitting results for each of the ex-vivo kidneysused in the study. . . . . . . . . . . . . . . . . . . . . . . . . 169Figure A.6 Individual model fitting results for each of the ex-vivo kidneysused in the study when viscous components are set to zero. . . 170Figure A.7 Individual model fitting results for each 10 perturbation trials. 171Figure A.8 Individual model fitting results for each patient in the study(First 8). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172Figure A.9 Individual model fitting results for each patient in the study(Second 8). . . . . . . . . . . . . . . . . . . . . . . . . . . . 173xviiGlossaryAPI Application Programming InterfaceARFI Acoustic Radiation Force ImpulseBMI Body Mass IndexCAD Chronic Allograft DysfunctionCAN Chronic Allograft NephropathyCT Computed TomographyDOF Degree of FreedomDVRK da Vinci Research KitECG ElectrocardiogramEGFR Estimated Glomerular Filtration RateEM ElectromagneticESRD End Stage Renal DiseaseFEM Finite Element ModelGFR Glomerular Filtration RateICP Iterative Closest PointIF/TA Interstitial Fibrosis and Tubular AtrophyxviiiIR LED Infrared Light Emitting DiodeLED Light Emitting DiodeLFE Local Frequency EstimationMAP Mean Arterial PressureMIS Minimally Invasive SurgeryMRE Magnetic Resonance ElastographyMRI Magnetic Resonance ImagingPCA Principal Component AnalysisPLUS Public Library of UltrasoundPRA Point Reconstruction AccuracyRF Radio FrequencySWAVE Shear Wave Absolute Vibro-ElastographySDK Software Development KitSDUV Shearwave Dispersion Ultrasound VibrometrySSI SuperSonic ImagineTLE Target Localization ErrorTRE Target Registration ErrorVGH Vancouver General HospitalVISR Viscoelastic ResponseVTIQ Virtual Touch Image QuantificationxixAcknowledgmentsThe mountains are calling and I must go, and I will work on while I can, studyingincessantly.–John MuirI would like to acknowledge all the help that I have received throughout theyears of my PhD. My supervisors, Tim, Rob and Chris have provided me withvaluable guidance during the course of my journey. My friends in the RCL labwere always there if I needed help with experiments, a good idea, or just someoneto talk to. My friends in the outdoors taught me to take a break and get a new lookat the world. I would like to thank my family, whose continuous motivation keptme going. I could not have made it without each of you.xxChapter 1Introduction1.1 Overview and MotivationThis thesis takes a comprehensive look at how to improve the characterization ofkidney tissue through the application of ultrasound imaging and elastography ofthe kidney. There are multiple clinical applications for this work. The first ap-plication in which tissue characterization can improve clinical outcomes and sur-gical workflow is during kidney cancer identification and removal. In order toaccomplish this goal, a miniaturized ultrasound transducer used for intra-operativeimaging was developed and characterized. With this transducer, 3D ultrasound andelastography volumes can be produced for surgical guidance for robotic-assistedMinimally Invasive Surgery (MIS) procedures. The other major clinical applica-tion revolves around measurement of progressive kidney Interstitial Fibrosis andTubular Atrophy (IF/TA) in kidney transplants, also known as Chronic AllograftNephropathy (CAN), which is the leading cause of long term graft failure. Themonitoring and measurement of CAN, through the use of ultrasound elastography,could provide insight into graft health and potentially lead to changes in treatmentthat would prolong graft life without the need for additional biopsies of the kidney.11.2 The Kidney1.2.1 Kidney AnatomyThe kidneys are a paired organ that are responsible for filtering the blood of ionsand proteins, regulating the water content in the body and creating urine to removewaste from the body. The kidney also plays a role in the release of hormones, suchas renin. The kidney is approximately 5 cm long, 3 cm wide and 2 cm thick. Thekidneys sit under the liver and diaphragm respectively and are attached to the aortaand vena cava through the renal artery and vein. Although they seem to be locatedinside the abdomen, they are outside, anterior, of the abdominal wall. At any giventime, the blood flow into the kidney is equal to approximately 25% of the cardiacoutput [82].The kidney is composed of a three different sections: the kidney cortex (theoutside of the kidney), the medulla (the middle of the kidney) and the collectingsystem (the internal section of the kidney). Figure 1.1 illustrates each of the struc-tures mentioned. The cortex is the outside of the kidney and is the upper end of thenephrons, the functioning units of the kidney. The lower ends of the nephrons formpyramidal structures known as the medulla. The vertex of each pyramid terminatesin the minor calyx, and the numerous minor calyces expand into two or three majorcalyces. The collecting system, made up of the minor and major calyces, collectsthe urine from the pyramids and pools it in the renal pelvis before sending it downthe ureter to the bladder. The collecting system is lined by thicker connective tissuethat appears as hyperechoic on ultrasound images.The nephron is the functioning unit of the kidney, and is made up of loops thatoriginate in the cortex and descend into the medulla. Its structure and function areillustrated in Figure 1.2. The blood is first filtered by the nephron at the glomeru-lus, a bundle of small vessels enclosed in a capsule, called Bowman’s capsule.Further filtering and chemical exchanges take place in the ascending and descend-ing Henle’s loops. The measured amount of creatinine (a substance created bythe muscles) in the blood and/or the urine, along with the patients physical char-acteristics is a commonly used way to estimate the filtration rate of the kidneys.The Estimated Glomerular Filtration Rate (EGFR) is used as a way to track how2Figure 1.1: A diagram of the kidney anatomy (right) and their placementwithin the abdomen (left). Image credit: philschatz.com Anatomy andPhysiology [122].well the kidney is functioning, and thus a measure of the overall kidney health.For example, the normal estimate is around 100 mL/min/1.73m2 and a sustainedmeasurement of 60-90 mL/min/1.73m2 is indicative of kidney damage.1.2.2 Kidney Cancer and Partial NephrectomyRenal cell carcinoma is a one type of aggressive kidney cancer that is often treatedthrough the removal of all or part of the kidney. It is the sixth most diagnosed ma-lignancy in Canada in men and the tenth most common in women, with estimated4,800 new cases and about 1,650 deaths from the disease in 2010. In the UnitedStates, 61,560 new cases of kidney cancer were diagnosed in 2015 [1]. Since themid-1980s, death rates have decreased by 0.3 per cent per year for males, and forfemales by 0.7% per year. Despite the death rates decreasing, this type of canceris also increasing its rate of incidence by approximately 1.3% per year. (CanadianCancer Statistics: www.cancer.ca).When the tumour in the kidney is small, less than 4 cm, or mostly on the outsideof the kidney, the kidney tissue can be spared by only removing the canceroustissue, rather than the entire kidney. The remaining kidney tissue continues tofunction as each section of the kidney is independent, as long as the blood flow is3Figure 1.2: The detailed diagram of the nephron structure including theGlomerulus, Bowman’s capsule and filtrate flow. Image Credit: Medi-cal Science Navigator [135].uninterrupted.Partial nephrectomy, or removal of only a section of the kidney, is a relativelynew procedure. In a study of matched patients receiving partial or radical nephrec-tomy, the patients receiving partial nephrectomy had equivalent long-term can-cer outcomes, but a lower risk of long-term decreases in kidney function [96]. Itis increasingly being performed but the complexity of the surgery has limited itswidespread acceptance [130].A detailed description of the surgical procedure is given in [157], but the fol-lowing list outlines the steps in the robot-assisted procedure as experienced at ourinstitution, the Vancouver General Hospital:1. Prepare the patient and robot for surgery2. Dock robot to the patient3. Locate and dissect the gonadal vein and the ureter, follow this back to therenal hilum, identifying their entire path4. Expose the renal hilum, where the vessels enter and exit the kidney5. Determine location and expose the tumour, dissect additional fat if needed46. Use laparoscopic ultrasound transducer to find the exact location of the tu-mour7. Use ultrasound to determine tumour margins and mark them on the kidneysurface using electrocautery8. Place the clips on the renal hilum (25 minute warm ischemia time limit startsnow)9. Begin resection of tumour10. Remove tumour, checking visually for clean margins11. Suture the defect closed12. Release the clamps on the hilum (25 minute warm ischemia time limit ends)13. Check to make sure that there is no bleeding14. Undock robot and finish closing using laparoscopic toolsThe two major steps are 5) the localization of the kidney and tumor and 9-11) the removal of the tumor and defect reconstruction (repairing the hole left byremoving the tumor). The localization and exposure of the renal hilum and thelocalization of the ureter is the most time consuming portion of the surgery. Thetotal average time over 12 patients for the procedure was reported as 289.5 minutes(range 145-369 minutes) [36]. Since the exact locations of the vessels, includingthe renal artery and vein, are not known and the consequences of damaging thesevessels would cause significant blood loss for the patient, surgeons proceed slowlyand carefully. During the surgery, there is a time limit due to the necessity ofclamping the major vessels to the kidney while the tumor removal is taking place[37]. After 25 minutes of warm ischemia time, it has been found that permanentdamage to the kidney occurs [169]. The majority of the difficult suturing and de-fect reconstruction must be completed during the warm ischemia time, the time inwhich the blood flow to the kidney is cut off. The time for resection of the tumorand suturing was only 35.3 minutes (range 15-49 minutes) [36]. This means onlyabout 12% of the total surgical time is spent during tumor resection and suturing,5while the rest of the time is spent locating the tumor and vessels, an area whereimproved ultrasound guidance could decrease the total time required for the proce-dure. There have been concerns that the use of partial nephrectomy could cause anincrease in the rate of positive margins and cancer reoccurrence [134].In general, surgeons try to leave 0.5 to 1 cm margins around the edge of thetumour to ensure that all cancerous tissue has been removed [161]. It has beenfound that any positive margin leads to the same (low) risk of cancer reoccurrence,but any positive margin, indicating some cancerous tissue had been left behind,increases the reoccurrence rate [134]. Intra-operative laparoscopic ultrasound isused to determine the extent and location of the tumor during surgery and helpsdetermine the margins of the resection, providing the surgeon with another tool tominimize the occurrence of positive margins.As the localization of the vessels is an important and potentially time consum-ing step within the procedure, one of the methods to provide additional guidancewould be to register the ultrasound images of the vessels to pre-operative Com-puted Tomography (CT) scans of the patient [143]. Vessels have been previouslyused to register intra-operative data to pre-operative CT and Magnetic ResonanceImaging (MRI) scans [39, 127, 136]. For example, the cortical vessels of the brainwere used to orient the surgeon and account for brain deformation during imageguided surgery through video tracking of the vessels [39] or tracked ultrasound[136]. Image guidance during liver surgery is another application in which ves-sels were used for registration [93, 98]. The vessels of the liver are prominentfeatures in surgical navigation for tumour resection. In addition, in complicatedstructures such as the liver, hierarchical methods of vessel registration were alsoimplemented [124]. Previous work on vessel extraction and registration have in-volved voxel-based registration [98], image-to-model registration [74], model-to-model registration [136] and the integration of landmark and intensity information[93]. Model-to-model registration methods typically use a modified version of theIterative Closest Point (ICP) algorithm [22] applied to the vessel centerlines. Othermethods use Doppler images to create a model from the ultrasound images withregion growing segmentation [92] or color-based segmentation [136].Registering the pre-operative CT would allow the surgeon to see the relation-ships between the vessel and kidney structure. In section 2, the possibility of regis-6Figure 1.3: The transplant kidney is placed in the pelvis rather than in placeof the failing kidney. It is attached to the iliac artery and vein. ImageCredit: University of Utah Healthcare [119].tering 3D ultrasound reconstructions of vessel structures with their correspondingCT structures is explored further.1.2.3 Kidney Transplant and Transplant FibrosisKidneys are one of the most commonly transplanted organs with 1265 transplantsfrom live and deceased donors taking place across Canada in 2014 (Canadian In-stitute for Health Information). In this case, the donor organ is removed from thedonor and placed into the pelvic region of the recipient. It is not placed where thenative kidney resides since in many cases the native kidneys are not removed. Dueto the placement of the transplant kidneys, they are very close to the surface of thepatient’s skin. This makes them very accessible for ultrasound and elastographyinterrogation.The kidneys are a redundant organ, meaning that a single kidney is sufficient tofilter the entire blood flow. This means that a person can function well with only asingle kidney, making living kidney donors an option and that a transplant recipientonly needs a single kidney to live a full and healthy life.Renal transplantation represents the current gold standard in the treatment ofpatients with End Stage Renal Disease (ESRD). The major causes of ESRD are dia-7betes and hypertension, but other causes include glomerulonephritis, infection andpolycystic kidney disease [73]. ESRD is characterized by low EGFR and eventualkidney failure. When diagnosed with ESRD, the patient must be put on dialysis ormust receive an organ transplant. Great strides have been made in the managementof renal transplant recipients in terms of supportive care, immunology, and drugtherapeutics. While treatment of acute problems such as rejection has significantlyimproved over the past two decades, overall long term kidney graft survival is stilla major problem. Indeed, the half-life for the graft was 36 and 19.5 years respec-tively (when removing those patients who died with a functioning graft) [57]. In2013, the 5 year survival rates for kidney transplant patients who received theirkidney from deceased donors is about 85% respectively [107]. It has been shownthat clinical acute rejection within the first year decreases the long term survivaland that chronic rejection is an important factor in graft failure [57].CAN or as it is sometimes known, Chronic Allograft Dysfunction (CAD), is theterm that describes the build up of interstitial fibrosis and tubular atrophy, whichis characterized by thickening of the basement membranes and flattening of theepithelial cells [150]. In a study of risk factors for CAN, Schwarz et al. found thatthe patients at the highest risk for CAN were those with early histological changes(found on serial biopsy), a deceased donor and long cold ischemia times (the timeduring which the kidney was outside of a body during the transplant process) [150].CAN is one of the common causes for eventual kidney transplant failure [57].The fibrosis may be caused by the accumulation of rejection episodes that might nothave obvious outward symptoms, known as subclinical rejection. Over time, theepisodes cause fibrosis and scar tissue to build up and prevent the nephrons fromfunctioning properly and the transplant organ fails. If the level of fibrosis couldbe measured and identified in the early stages of rejection episodes, these could bemitigated by making changes in the patient’s immunosuppressant regimen.Currently, the only method for measuring the level of fibrosis is through theuse of biopsy. Biopsy of the kidney removes a small section of the kidney usingan ultrasound guided biopsy needle. This tissue section is then sent to pathologywhere the level of fibrosis is identified. This process is painful for the patient andmay lead to complications and morbidity [149]. Schwarz et al. reported on thesafety of graft biopsy [149]. In a study of 1171 biopsies of 508 patients, they found8that arterio-venous fistulas occurred in 7.3% of the biopsies and gross hematuriaoccurred in 3.5% of the biopsies. 1% of the biopsies resulted in major complica-tions. Although the complication rate is low, there are still risks, and patients whoare on blood thinning medications must stop these medications before the biopsyis completed, which introduces other risks.Some institutions have completed studies using serial biopsies in order to bet-ter understand the formation of fibrosis, and how subclinical rejection episodesinfluence the long-term graft survival. It is possible that these serial biopsies coulddetect the subclinical rejection episodes, and changes in medication and treatmentcould have resulted in better long term outcomes [110]. In one study, subclinicalrejection was found in 14% of the patients [150]. CAN was found in 5% of the firstround of these biopsies at 6 weeks and 37% of patients at 26 weeks. Another studyfound that CAN was present in one third of the patients at 1 year [116]. This showsthat the early development of CAN, which could affect long term graft survival, isoften progressing undetected.To standardize the assessment of the tissue from biopsy, a scoring system wasdeveloped [155]. This scoring system measures different aspects of the tissue, in-cluding tubular atrophy, interstitial fibrosis, arteriolar hyaline thickening, glomeru-lopathy and vascular fibrous intimal thickening.In particular, for this thesis, the interstitial fibrosis measurement is of mostinterest as it is most likely to cause distinguishable effects on the measured kidneystiffness, as well as functional effects on the kidney. According to the Banff scoringsystem, the amount of interstitial fibrosis is broken down into four levels regardingthe amount of cortical area affected by the fibrosis.• Level 0: less than 5% of the tissue shows evidence of fibrosis• Level 1: between 6% and 25% of the tissues shows fibrosis• Level 2: between 26% and 50% of the tissue shows evidence of fibrosis• Level 3: More than 50% of the tissue shows evidence of fibrosis.Using a non-invasive method for measuring fibrosis, such as ultrasound, wouldallow for constant and ongoing monitoring of the kidney health. A non-invasive9method would mean that the early detection of CAN is potentially more likely andcould result in changes in patient treatment. This will hopefully lead to kidneygrafts that will last longer and improve the long-term outcomes for transplant pa-tients.1.3 Ultrasound ImagingUltrasound imaging is a non-invasive, non-ionizing and real-time method of med-ical imaging. Ultrasound offers real-time imaging with typical frame rates of 40Hz, it is non-ionizing which makes it safer to use for both the patient and the op-erator and it is inexpensive compared to other imaging modalities. All of thesecharacteristics make ultrasound an advantageous imaging modality for intraoper-ative navigation and guidance. A full explanation of ultrasound imaging can befound in textbooks [23], but a short overview will be presented here.Ultrasound sound uses high frequency pulses to create images of tissue under-neath the surface of a patient’s skin or organ surface.The system consists of a transducer, which is placed on the patient, and a pro-cessing machine or ultrasound scanner, which sends the energy to the transducerand receives the signals from the transducer. The ultrasound scanner then createsthe images for the doctor or technician. The transducer is made of an array ofpiezoelectric crystals. These crystals vibrate in response to an electric stimulusand also produce an electric signal is response to a vibration stimulus.With an electrical stimulus, the piezoelectric crystals of the transducer send apulsed vibration at approximately 3-10 MHz into the body. These pulses travelthrough the tissue at approximately 1540 m/s. 1540m/s is the average speed forthese waves in soft tissue. The true speed ranges from 1450 m/s in fat to 1580m/s in muscle. The reflections from these waves are sensed by the piezolelectriccrystals over time. The depth of a particular reflection can be calculated from thetime of flight between when the pulse was created and when the echo was detectedby the crystal and the associated electronics. The speed of sound generally changesvery little depending on the tissue type [62], so this average is used for all tissuetypes unless specified otherwise.The depth, d, is calculated as:10d = (t ∗ c)/2 , (1.1)where t is the time from the pulse generation to the reception of the reflectionor echo, and c is the speed of sound within the tissue.The reflections are created when there is a change in the acoustic impedance Zof the tissue. The acoustic impedance is calculated as Z = ρc where ρ is the tissuedensity and c is the speed of sound in tissue.The ratio of the energy passing through a boundary and the reflected energy isknown as the reflection coefficient. This coefficient of any perpendicular reflectionfrom a boundary between materials of two different impedances Z1 and Z2 is asfollowsR = (Z2−Z1)/(Z2 +Z1). (1.2)In general the reflection coefficient between soft tissues is small, such that mostof the energy of the wave is transmitted through the tissue.After the reflections are recorded, envelope detection is performed via a HilbertTransform and the resulting signal is compressed to an appropriate range for dis-play. After compression, the data is mapped to the transducer coordinate frame todisplay a geometrically accurate image of echo reflection strength as a function ofecho location. This image is called a B-mode image (B for ”Brightness”).In the cases of bone or air, there can be large reflection coefficients, where mostof the wave is reflected at the boundary. This creates a shadowing effect, where theimage past this boundary is mostly dark.A general characteristic of ultrasound images is determined by sub-wavelengthinhomogeneities in the tissue that create small echoes known as speckle. Thisspeckle pattern depends on the transducer used and the type of image processingemployed to generate the B-mode image.Ultrasound transducers come in a variety of shapes, sizes and frequencies de-pending on their intended use. Transducers designed for examining small regionsclose to the surface use higher frequency pulses, often close or in excess of 10MHz. The higher frequency waves allow for higher resolution imaging becausethey generate narrower pulses that can distinguish between reflectors that are close11Figure 1.4: The pitch is the spacing between the piezoelectric crystals. Thethree directions of the image are defined as axial (away from the probe),lateral (along the direction of the crystal array) and elevational (out ofplane).to each other. However, the attenuation of ultrasound pulses in tissue increaseswith frequency, so the total depth of imaging is lower. Transducers that are meantfor imaging the abdomen transmit longer pulses at a lower frequency, generally be-tween 3-5 MHz. These transducers are able to image at a much greater depth thanthose at higher frequency. The depth for typical ultrasound images ranges from2-16 cm depending in the transmitting frequency.The piezoelectric crystals of many transducers are arranged in a linear fashionat the face of the transducer. Other possible arrangements include a curvilineararrangement along a convex surface (convex arrays) to provide a wider field ofview, and a matrix arrangement, which would allow for fast 3D imaging. Thespacing between the crystal elements is known as the pitch. The direction of therow of piezoelectric crystals is known as the lateral direction, the direction of wavetransmission (away from the transducer face in the body) is known as the axialdirection and the final direction, out of plane of the produced image, is known asthe elevational direction (Figure 1.4).Ultrasound images have the greatest resolution in the axial direction, whichis determined by the pulse length. The resolution in the lateral direction is deter-12mined by the beam width. This can be improved using beam focusing [164]. Theresolution in the elevation direction is also determined by beam width via a fixedlens on the transducer face, and cannot be changed using software.For all work in this thesis, an Ultrasonix Touch ultrasound machine (Analogic,Richmond, BC) was used. This machine is equipped with an Software Develop-ment Kit (SDK), which allows access to programming the underlying signal trans-missions and processing. An image is typically formed by the sequential transmitand receive of groups of crystals along the lateral direction of the image. The timerequired to complete the sending and receiving of the signals will determine theframe rate of the imaging system. By adjusting the sequence of sending and re-ceiving, possible with the Ultrasonix SDK, it is possible to create a small section ofan image at a very high frame rate. This concept will be described in more detailin Section Ultrasound ElastographyIn a broad sense, elastography is the measurement of tissue stiffness. A compre-hensive overview of elastography can be found in the thesis by Jeffrey Abeysekera[3] and a short review is presented here.Although there are several methods for measurement, all methods use the sameunderlying principle. The tissue is physically pushed and then ultrasound imagingis used to measure the tissue response to this push. For a dynamic push, the shearwave that is created moves through the tissue at a speed and wave length that isdetermined in large part by the shear modulus or elasticity of the tissue.The original method for measurement of tissue stiffness is using quasi-staticstrain measurements. This requires a user to slightly compress and decompress thetissue. The tissue is compared in the two states and the relative motion is calcu-lated. This method allows the relative displacement or strain of the tissues withinthe image to be compared, but does not allow for comparison between imaging ses-sions or between patients, because strain depends on the push and on the boundarycondition surrounding the region of interest, so it is not an intrinsic property oftissue. However, strain measurement opened the gateway for subsequent develop-ment of absolute or quantitative measurement methods by providing proof that the13tissue stiffness could be correlated with tissue health.Absolute elasticity methods allow for the tissue stiffness to be measured ina quantitative manner by observing the dynamic wave propagation, producing aquantitative measurement of shear modulus or Young’s modulus in kilopascals(kPa). This allows for intra- and inter-patient comparisons. The excitation canbe generated by an acoustic force [21, 40, 118] or by external vibration [12, 34].We use an external vibration source and a standard ultrasound machine. This sys-tem was named Shear Wave Absolute Vibro-Elastography (SWAVE) and will bereferred to as such for the remainder of this thesis.For a harmonic excitation, shear wave propagation in an isotropic homoge-neous elastic solid, and for incompressible tissues undergoing small strain and lin-ear approximations, is described in the frequency domain by the following waveequation:ρ( jω)2uˆ(x, jω) = µO2uˆ(x, jω) , (1.3)where ρ is the tissue density, ω is the frequency, uˆ(x, jω) are the displacementphasors and µ is the shear modulus.We assume that the density of tissue ρ is homogeneous and equal to that ofwater. This is a common assumption in the elastography community even thoughtissue density does change with tissue type and affects the wave speed. We alsoassume that the displacement phasors uˆ(x, jω) generated in the tissue by the exciterhave only a shear wave component, no pressure wave component. Then, if onecomponent of the displacement phasors uˆ(x, jω) is known over a region of interest,in this case the axial displacement with respect to the transducer, the Helmholtzequation (1.3) can be used to find the shear modulus µ .There are various ways to define the elastic modulus, or elasticity. In much ofthe literature, the Young’s Modulus (E) is used and describes a material’s resistanceto deformation under unilateral compression. The shear modulus (µ) specificallydescribes a material’s resistance to shear. Poisson’s ratio (ν) defines the ratio ofdeformation that occurs orthogonal to the applied force due to tissue compress-ibility. It is generally assumed that the Poisson’s ratio for tissue is ≈ 0.5, or thattissue is incompressible. The shear modulus, Young’s modulus and Poisson’s ratio14are related in the equation below, for a homogeneous, isotropic, linear elastic andincompressible material:µ =E2(1+ν)(1.4)When ν is = 0.5, this reduces to E ≈ 3µ .With the above assumptions, the speed of the shear wave cs, the shear wavewavelength, λ , at a known frequency, f , can be related to Young’s Modulus, E, asfollows:E = 3µ (1.5)µ = ρc2s (1.6)cs = fλ (1.7)E = 3ρ( fλ )2 . (1.8)1.5 SWAVEIn the SWAVE imaging method used in this thesis, the displacement phasors fromeach 2D image are interpolated over a regular 3D grid and the distribution of shearmoduli over the tissue is estimated by using a local spatial frequency estimator,such as Local Frequency Estimation (LFE) [105]. As the vibration source is notin any particular position with relation to the ultrasound probe, it is important tomeasure the wave in three dimensions to measure the full length of wave, whichmay not be within the imaging plane. The local frequency of the 3D volume isestimated after applying a bank of directional filters. These filters are directionallyoriented and cover a wide band of spatial frequencies. The weighted average of theratios between the outputs from these filters determines the local frequency at thatlocation within the volume. This 3D volume can be created in a variety of waysincluding a motorized 3D probe, a tracked freehand 2D probe or a motorized 2D15Figure 1.5: Left: Acquisition timeline for two frames of B-mode data fora twelve scan-line field of view. Right: Acquisition timeline for twoframes of high-frame-rate acquisition data for a four scan-line field ofview. T is the time required for acquisition of a single RF line.probe.Strain is inversely proportional to elasticity under certain conditions, such asuniform stress. Changing boundaries violates this assumption, which results inimage artifacts that do not represent the elasticity of tissue, but are the result ofa non-uniform stress field within the tissue. SWAVE however, measures the localwavelength of the waves traveling in the tissue, which is dependent on the localintrinsic elasticity of the tissue.With a steady state excitation, the axial displacement phasors can be estimatedover a volume at a high effective frame rate using sector based imaging. In otherwords, using the software development kit provided by Ultrasonix, it is possibleto program the sequence of imaging on the ultrasound machine. By repeatedlyimaging a small section of the image, the frame rate can be much higher than whenimaging the entire width of the image (Figure 1.5) [12].Our system uses an external mechanical exciter, a voice coil shaker, drivenby waveforms created by a signal generator (Agilent 33220A) controlled throughsoftware loaded on the Ultrasonix machine. The excitation is applied directly tothe phantom or tissue. No particular positioning of the exciter with respect to the16transducer in needed. When the tissue motion is analyzed in the frequency domain,the motion caused by each of the frequencies of excitation can be separated.1.5.1 Overview of Elastography MethodsThere are several methods to measure tissue elasticity. In this Section, an overviewof the other systems are presented with their advantages and limitations.Two of the other methods of quantitative elastography are acoustic radiationforce impulse imaging [118] and supersonic shear wave imaging [21]. Both ofthese methods use the acoustic power from the ultrasound transducer to create dis-placements in the tissue. The impulses are focused along the axial direction of theultrasound image to create a wave in the tissue that propagates in the lateral direc-tion of the image. Using high speed imaging techniques the wave is tracked as itmoves away from the focus line of the push, or uses the ultrasound to measure thetissue response at the site of the push.One limitation of this technique is that it requires a higher power machine inorder to create the acoustic push. It is available on the Siemens Acuson Machine(Erlangen, Germany), the Philips EPIQ (Philips Healthcare, Boston, Mass), and theSupersonic Imagine Aixplorer (Aix-en-Provence, France). One advantage of thistechnique is that no external hardware is required and the push is created withinthe ultrasound transducer. It is also very localized, and has a high potential res-olution. This method has been used extensively for breast cancer measurements[40, 44, 117]. The depth of measurement using this technique is limited as thepower needed to create the push cannot penetrate further than about 6 cm. Thislimits its use in deeper organs, such as liver.Another commercially available method of stiffness measurement is the Fi-broScan system (Echosens, Paris, France). FibroScan is an elasticity measurementsystem that uses a mechanical impulse and a single ultrasound element. The sys-tem measures the response of the tissue to the mechanical impulse along the lineof the ultrasound element. One limitation of this system is that there is no B-modeultrasound guidance for the measurement to guide the user. FibroScan is suitedfor large homogeneous organs such as the liver. The system is easy to use and ithas been proven to accurately measure the liver stiffness and level of liver cirrhosis17[72].External vibration is used in Magnetic Resonance Elastography (MRE) and inthe steady state shear wave ultrasound elastography used in this thesis. In thismethod, a constant vibration source is used to create steady state waves at a knownfrequency of excitation [106]. The displacements of the tissue are measured asthe tissue moves and the local wavelength within the tissue can be measured withimage processing techniques [187].One advantage of using MRE is that the imaging technique is already in 3D andthe waves can be measured naturally in three dimensions. The disadvantage is thatimaging can be very slow, one acquisition taking several minutes. The approachused in this thesis is similar in principle to that used in MRE. The advantages ofour approach are that we eliminate the challenges of penetration depth and tissueheating that are associated with using acoustic impulse forces for excitation. Inaddition, we can excite the tissue at specific frequencies, allowing us to examinethe frequency dependency of tissue stiffness [154]. Using multiple frequenciesalso decreases the problems caused by nodes in the wave patterns [167]. Nodescan cause artifacts in the elasticity image because there is no displacement at thesepoints, and therefore the wavelength measurement is unreliable. One disadvantageof this method is the desirability of having multiple wavelengths within the tissueof interest in order to get an accurate measure the wavelength; this is becauseultrasound can only accurately measure axial displacement of the tissue.1.6 Minimally Invasive Surgery, Intra-operativeUltrasound and the da Vinci Robot1.6.1 Minimally Invasive SurgeryMinimally invasive surgery is slowly becoming the standard of care for many pro-cedures, and continues to grow in popularity. Minimally invasive operations oftenrequire several small incisions, of approximately 2 cm each, to be made throughthe patient’s skin, instead of the 10-15 cm incision required for a traditional opensurgery. The surgeon works using long specialized instruments and views the surgi-cal scene through a laparoscopic camera. This type of procedure decreases the pa-18tient’s morbidity and shortens hospital stays [47, 137]. Many abdominal, thoracicand pelvic procedures are now completed as minimally invasive (or laparoscopic)operations.Although there are benefits to the patient, additional physical and cognitiveloads are placed on the surgeon [99, 173]. First of all, the surgeon must workwith instruments that lack the degrees of freedom available in the human hand.The majority of laparoscopic instruments do not have any wrist-like joints at theend-effector. Additionally, during laparoscopic surgery, the surgeon must over-come working through the ‘fulcrum effect’ caused by the instrument having topass through a narrow opening in the mostly rigid abdominal wall. This constraintmakes the tip of the instrument move in the opposite direction of the surgeon’shand. Also, the view of the surgical field becomes considerably more restrictedwhen moving from open surgery to laparoscopic surgery. The surgeon is no longerable to see the entire surgical field at once, but must rely on the narrow view throughthe laparoscopic camera. Last, they are now looking at a single camera image, theylose binocular vision, and therefore, depth perception.The limitations of laparoscopic surgery are especially noticeable in more dif-ficult surgeries, such as laparoscopic partial nephrectomy [130]. Partial nephrec-tomy is one treatment for kidney cancer. Kidney cancer refers to any tumours thatoccur in the parenchyma of the kidney, of which approximately 80% are renal cellcarcinomas.Because partial nephrectomy allows the cancerous tissue to be removed whilepreserving as much of the healthy kidney tissues as possible, it is the preferredmethod over radical nephrectomy for tumours under 4 centimetres in size [77].Laparoscopic partial nephrectomy has also gained popularity as it is minimally in-vasive and has shown to have comparable outcomes to open procedures [56, 130,141]. Partial nephrectomy, with its innate challenges, as described above in Section1.2.2, combined with the limitations of laparoscopic surgery, make partial nephrec-tomy an operation which could be improved through the application of new tech-nologies, such as intra-operative ultrasound.The use of intra-operative ultrasound could also be pivotal to the improve-ment of surgery. As a real-time imaging technique, ultrasound has the potentialto be used for registration to pre-operative data and also as a method of direct vi-19sualization of critical structures such as with elastography. Through the use ofintra-operative ultrasound combined with robotic technology, the challenges of la-paroscopic surgery and partial nephrectomy have the potential to be minimized[143].1.6.2 Intra-operative UltrasoundIntra-operative ultrasound was first introduced in the 1950s [104]. In 1958, the firstlaparoscopic ultrasound transducer was used during cholecystectomy, but intra-operative ultrasound did not gain widespread usage until B-mode images becameavailable in the 1970s and interpretation of images became more straightforwardsince 2D B-mode images could now be visualized as a ‘picture’ rather than a singleecho line. Laparoscopic ultrasound also gained popularity with the spread of min-imally invasive surgery, where the ultrasound images enhance guidance when us-ing laparoscopic tools [104]. Traditional laparoscopic intra-operative ultrasound iscurrently used for a variety of procedures, including resection of liver cancer [16],gall bladder removal [31] and resection of kidney cancer [129]. Intra-operativeultrasound provides high-quality, real-time intra-operative imaging for assessmentof tumour margins, guidance around vessels and locating tumour resection planes.Va˚penstad et al. presents a survey on the current use of laparoscopic ultrasound[174].1.6.3 Robot-assisted Surgery with the da Vinci Surgical SystemThe introduction of robotic systems such as the da Vinci Surgical System attemptsto mitigate the shortcoming of laparoscopic surgery by replicating an environmentmore similar to that of traditional open surgery. The system consists of three sec-tions: the surgeon’s console, the patient side cart and the vision cart [Figure 1.6].The surgeon is able to sit comfortably and ergonomically at the console (a possibleimprovement over traditional open surgery), and the patient is positioned on thetable by the patient side cart. The surgeon sits at the surgeon’s console and his/hermotions are mapped to the motions of the instruments on the patient-side cart, thisis known as a teleoperated robot, or a master (the surgeon) and slave (the instru-ments of the patient side cart) configuration. Although the da Vinci system does not20Figure 1.6: The da Vinci Surgical System. Image courtesy of Intuitive Surgi-cal Inc.currently included automation, the surgeon’s motions are scaled and filtered to im-prove dexterity and to remove tremor. The instruments used during robotic surgeryare wristed at the end-effector, allowing the surgeon to regain natural control of thedegrees of freedom of the tools.The orientation of the surgeon’s hands on the ‘master manipulators’ is con-strained to match the orientation of the end-effector as seen by the surgeon throughthe cameras. The surgeon uses the ‘master manipulators’ in the surgeon’s consoleto control all of the Degree of Freedom (DOF) available to the instrument, generallysix or seven. In six DOF instruments, there are three translational degrees and threerotational degrees of freedom while in seven DOF instruments, the surgeon alsocontrols the grasping motion of the instrument. The camera arrangement is alsodifferent from laparoscopic procedures. The da Vinci laparoscope has two camerachannels that run along its length. The video feeds are kept separate until viewedby the surgeon through the console. This camera system is stereoscopic, simulatingbinocular vision and allows for depth information to be provided to the surgeon.This, along with the degrees of freedom of the instruments, makes manipulationsof tissue, knot tying, needle passing, and discrimination of layers more effective[30, 68, 184].Although the da Vinci robot offers advantages over traditional laparoscopic21surgery, some studies have found that experienced laparoscopic surgeons have notseen a significant benefit [101]. On the other hand, the da Vinci robot is a goodplatform for integration of additional aids and is not being currently used to its fullpotential. With additional navigational aids, such as intra-operative imaging, webelieve the surgical guidance can be improved. The stereoscopic vision allows thesurgeons to visualize intra-operative and pre-operative imaging [175], and can beused for tissue tracking in 3D [159]. Using the joint angles of the robot encoders,the tool position and orientation can be tracked within the accuracy of other com-monly used surgical localizers, approximately 1 mm [89].One of the main limitations cited by researchers and users is the lack of hapticfeedback to the surgeon in the console during the surgery. This means that the sur-geon can no longer feel the tissue in the traditional way using the sensors in theirfingers. The surgeons often feel for tumours under the tissue surface. Tumoursare often much stiffer than the surrounding tissue, which means the surgeon canpalpate or touch the tissue to determine its approximate depth and diameter evenwhen it is not visible on the surface of the tissue [86, 88, 109]. This would thenguide the surgeon’s further dissection or resection. This limitation could be miti-gated through the use of ultrasound elastography as a tool for identifying stiff areasbelow the surface.There are two interfaces used to get information to and from the da Vinci Robot.The clinical robot Application Programming Interface (API) is used to read the jointangles from the robot in real time. These joint angles can be used to locate the toolsin space through the use of forward kinematics. This interface can be used whilethe robot is in clinical use.The other interface used is the da Vinci Research Kit (DVRK). The DVRK dif-fers from the previously mentioned API, in that it does not use the clinical da Vincisoftware, but uses a separate controller to command the joint positions, velocitiesor torques of each of the robot arms [33, 79]. These controllers are linked to acomputer and each arm is controlled individually. This allows for a computer toautomate the motion of the robot, in addition to the usual teleoperation mode of theda Vinci.221.6.4 Intra-operative Ultrasound for the da VinciAlthough traditional laparoscopic ultrasound instrumentation is a valuable imag-ing tool in many types of surgery, the manoeuvrability, and thus imaging ability,with these types of instruments is limited. Furthermore, during robotic procedures,laparoscopic ultrasound probes are typically controlled by the patient bedside sur-gical assistant instead of the operating surgeon. Additionally, the ultrasound instru-ment requires a dedicated laparoscopic port, meaning that another instrument mustbe removed from the surgical field and the operation must be halted while scan-ning is taking place. It is well established that ultrasound is highly user dependentmodality in two main ways. Firstly, the quality of the ultrasound image views de-pends on the experience of the user. Secondly, only some users have the ability tomake three-dimensional connections between the images and the actual anatomy.Manipulating the transducer allows the surgeon’s natural hand-eye coordinationto aid in the interpretation of the 3D anatomy from a series of 2D cross-sectionalimages.In order to provide the operating surgeon with control of the transducer, and tointegrate intra-operative ultrasound into robotic surgery, an ultrasound transducerthat is controlled directly from the da Vinci surgeon’s console was proposed [100].Later, another transducer was designed and modeled after a typical da Vinci 5mmtool and had greater mobility inside the patient [142]. However, both of thesecombined ultrasound/da Vinci tools still require a dedicated port or da Vinci toolchanges, and this version of the 5mm da Vinci tool is more limited in its rangeorientation than a typical 8mm da Vinci tool.The first miniature “pick up” style probe was developed for cardiac procedures.A 13 MHz Aloka (Hitachi Aloka Medical, Wallingford, CT) mini-transducer (15×9× 6mm) has been used for characterization of the coronary arteries during tho-racic laparoscopic surgery [28, 29]. This transducer was outfitted with a small finthat allowed the da Vinci needle drivers to grasp and maneuver the transducer.Two commercial products have recently been introduced, one by Aloka (Walling-ford, CT), and one by Analogic Ultrasound (named the ProArt R©) (Peabody, MA).Both these ultrasound transducers are designed with a fin on the ventral proximalsurface that fits the da Vinci Pro-GraspTM tool (model 420093). The fin is placed in23such a way that the long axis of the transducer and the jaws of the tool are parallel.They fit through a standard sized trocar and use the TilePro display in the da VinciS and Si. Both transducers are high frequency and thus have high resolution withrelatively low penetration depth. The Aloka model is a linear array transducer andthe ProArt R© uses a curved array for wider visibility.The grasping fin designs of the three transducers mentioned above do not en-able a repeatable, stable grasp by a da Vinci tool, such as the Pro-GraspTM forceps.The Pro-GraspTM forceps are a very commonly used dissection tool and their slot-ted shape is conducive to holding onto an ultrasound transducer. While the findesign of [28, 29] makes it easy to grasp the device, the location of the grasp isnot well defined. A stable, repeatable grasp is needed for accurate 3D tissue recon-struction or localization with respect to the da Vinci coordinate system.In order to take advantage of the degrees of freedom available with the da Vinciwristed tools, a new intra-abdominal pick-up transducer was designed in previousresearch to fulfill the design parameters (Figure 1.10) [143]. As outlined in thispaper, the design requirements for this new transducer were determined throughconsultations with the surgeons at our center and a transducer manufacturer. Thecompleted transducer is small enough to fit through a standard laparoscopic inci-sion and be maneuverable within the abdominal cavity. The da Vinci tools cangrasp the transducer in a repeatable manner and create a fixed transformation be-tween the tool location and the ultrasound image. The transducer’s motion shouldalso be fully constrained once grasped. Finally, the tool self-aligns with the ‘lock-ing’ mechanism on the transducer, such that the tool and transducer need not bewell aligned initially to guarantee capture.The transducer:• is small enough to be maneuvered inside the patient• has a small enough diameter to fit through a surgical incision• has a consistent and self-aligning interface with the da Vinci grasper• has no sharp or breakable components• can be tracked via multiple modalities24Figure 1.7: Example ultrasound images created with the intraoperative “pick-up” transducer of the carotid artery and vein. Left: B-mode image.Right: Doppler image of the arteries after bifurcation.Figure 1.8: Rendered images of the da Vinci interface.• can be sterilized using standard methods.The transducer is made of 128 elements spaced over an imaging face of 28 mmand has a center frequency of 10 MHz. These are similar characteristics to othercommercially available intraoperative transducers which provide high-resolutionimaging to a depth of 4 to 6 cm. Pin connections are currently compatible with Ul-trasonix ultrasound machines (Analogic Medical Corp., Richmond, Canada). Ex-amples of the B-mode and Doppler images are displayed in Figure 1.7.Unique to this transducer, and different from the commercially available trans-ducers, is the specially designed interface between the transducer and the da VinciPro-GraspTM forceps. This metal interface was designed such that it forms a lock-25ing type grasp with the Pro-GraspTM tool [Figure 1.8]. A groove [Figure 1.8],shown in blue (A), was built to match the width of the Pro-GraspTM. A small pin,shown in red (B), was designed to catch in the end of the tool slot and prevent thetool from sliding off. The walls of the groove are angled, shown in green (C, D),to increase the capture range. One wall is slanted to the floor of the groove, whilethe other ends just short, shown in yellow (E). The solid wall helps to constrain themotion between the transducer and the tool while the fully slanted wall prevents thetool from ever becoming jammed into the groove. The angle between the two sidesof the groove matches that of the Pro-GraspTM when the tool is completely closedon to the metal walls. This allows for a fixed transformation between the da Vincitool and the transducer, which in turn transfers all the degrees of freedom and rangeof motion available to the da Vinci tool to the transducer. The jointed wrist of theda Vinci and direct connection to the transducer makes manipulation easier thantraditional hand-held laparoscopic transducers [148] and because the two are nowfixed, it is possible to track the ultrasound transducer by measuring the location ofthe tool (Section 2. The transducer was manufactured by Vermon (Vermon, Tours,France) and is approved for STERRAD (ASP, Irvine, CA) sterilization methodsthat are commonly used in hospitals to sterilize ultrasound transducers. STER-RAD is a low temperature method of sterilization for sensitive surgical instrumentsusing hydrogen peroxide gas plasma technology and manufactured by AdvancedSterilization Products.Another unique aspect of this transducer is the inclusion of an electromagneticsensor inside the body of the transducer [Figure 1.9]. A Model 180 (Ascension,Vermont, USA) 6 degree of freedom sensor is used and compatible with the soft-ware of the Ultrasonix Ultrasound GPS system.As an alternative, a flat checkerboard for visual tracking can be placed on eitherflat section of the transducer, which is approximately 24mm× 10mm. An exampleof this checkerboard is shown in Figure 1.9, and can be used, as described later,for 3D tracking with a stereo camera. The checkerboard is surgical identificationtape (Key Surgical Inc., MN, USA) that is approved for human use and designedto withstand repeated autoclave sterilization cycles [42].The transducer can be placed into the abdominal cavity before the laparoscopiccannulas. The transducer tether will run alongside the camera cannula, but through26Figure 1.9: Top: Diagram showing the cross section of the “pick-up” trans-ducer, and the location of the electromagnetic (EM) sensor and crystalstack. Bottom: a photograph of the ultrasound transducer, with checker-boards used for camera tracking.27Figure 1.10: Rendered images of the “pick-up” transducer. Left and middle:cross sectional view (blue) of the lap-handle. The angled faces andlocking pins can be seen. Right: the tool fits tightly against the angledfaces. The practicality of adding visual tracking markers is demon-strated.Figure 1.11: The pick-up intra-operative transducer is grasped and manipu-lated by the Pro-GraspTM tool of the da Vinci Surgical System.the same incision. Thus, ultrasound is available throughout the operation withoutthe need for tool changes and provides further surgeon autonomy. This minimizesthe effect of using the transducer on surgical workflow.1.6.5 Intra-operative Ultrasound ElastographyUsing the transducer described above, 2D and 3D B-mode ultrasound images andvolumes can be used during the procedure to provide the surgeon with additionalguidance. In some cases, B-mode imaging alone of the renal tumor may not be28enough to provide the surgeon with adequate guidance.23-46% of small renal cell carcinomas, those treated with partial nephrectomy,appear iso- or hypo-echoic in B-mode imaging [60]. This means that the bound-aries of the tumor would be difficult to define, resulting in either positive margins(effecting patient outcomes) or more healthy tissue removed than necessary [91].Renal tumors can also have complicated structures, including necrotic cores andfat tissues [59].Ultrasound elastography has the potential to add contrast to imaging in order tobetter define the tumor boundary, as tumors are generally stiffer than the surround-ing tissues [86]. This is particularly important when the tumor has a complicatedstructure.In addition, elastography also has the potential to differentiate between benignangiomyolipomas and malignant renal cell carcinoma. The differences in stiffnessof these two types of tumors could potentially differentiate them intra-operatively.Elastography imaging could also measure other tissue characteristics, just as vis-cous properties [154] or poroelasticity [84]. The fat content of the tumors mightbe measurable from elastography measurements. Elastography could also helpdistinguish hypoechoic cysts, where shear waves do not propagate, from other hy-poechoic lesions.The tissue characteristics determined from elastography, along with B-modeimaging, could improve intra-operative imaging guidance for surgeons.1.7 Thesis Objectives1. To use ultrasound and ultrasound elastography to better characterize kid-ney tissue for use in robotic-assisted surgical guidance and kidney transplantmonitoring2. To determine the tracking accuracy of the new intra-operative “pick-up” ul-trasound probe previously built and tested. Looking at and testing three dif-ferent tracking methods, including the robotic kinematics, electromagnetictracking and stereo camera tracking3. To use the “pick-up” ultrasound probe for the construction of 3D elastograms29that could be used in surgery4. To determine tissue characteristics of the kidney. Previous research has pro-duced very different quantitative measures of kidney elasticity. To examinethe relationship between pressure and the elasticity properties and to esti-mate the viscous properties of the ex-vivo kidney by analyzing the frequencydependency of elasticity measurements5. To measure the correlation of kidney health on the elasticity measurementsof in-vivo transplant kidneys1.8 Thesis OverviewChapter 2 describes the experiments which test the tracking accuracy and recon-structions of the “pick-up” ultrasound probe. Three different tracking modes aretested and the accuracy of the reconstructions are tested against a known geometryof a vessel phantom.Chapter 3 details the use of the “pick-up” ultrasound transducer for use in elas-tography. In this chapter, two sets of experiments are described. The first set ofexperiments uses the Clinical Mode of the da Vinci robot, where the surgeon stillhas direct control of the robot and ultrasound transducer and the second set of ex-periments leverages the ability to automate the robot using the da Vinci ResearchKit.Chapter 4 provides an in depth look at the current state of the art research inkidney elastography measurements. In particular, the different and varying resultsof previous studies are examined and also the complications in kidney imaging thathave arisen from the different studies.Motivated by the complications of kidney imaging, Chapter 5 examines twomain features of kidney tissue. First, the relationship between blood pressure andmeasured elastography is investigated. The second experiment measures the vis-cous properties of the kidney. There is a distinctive relation between the excitationfrequency (assuming a completely elastic tissue) and the measured elasticity, andby measuring at multiple frequencies, a model of the tissue which includes theviscous elements can be fit to the resulting data.30The next chapter of this thesis moves to the realm of in-vivo patient imaging.Elastography data using the SWAVE method is collected from kidney transplantrecipients without biopsy. The initial study and results of four patients is presentedin this chapter.The final chapter of this thesis provides an overview of the work presented.The long term implications and limitations of the work are also discussed. Finally,the future of the work and possible improvements are considered.31Chapter 2Ultrasound Transducer Trackingand 3D Reconstruction2.1 IntroductionLaparoscopic MIS has become the standard of care for many abdominal operations.Benefits for the patient include lower morbidity and faster recovery time [47, 137].However, MIS is more technically challenging for the surgeon, and this has his-torically limited the widespread adoption of laparoscopic approaches for complexsurgical procedures such as partial nephrectomy. This operation is compounded bypressures including time constraints, restricted visualization, and elevated risks ofbleeding [157].As described in Section 1, knowledge of the vessels’ location during the pro-cedure would assist the surgeon in conducting a safer and more efficient dissec-tion. Our center, Vancouver General Hospital (VGH), performs about 150 open orlaparoscopic partial nephrectomy procedures per year, and 5 to 10 robotic proce-dures. We expect that a surgical navigational adjunct to robotic assisted surgeryusing the da Vinci, as proposed here will increase surgeon confidence, improvepatient safety and decrease dissection time required for this operation.With the increasing use of the da Vinci, better visualization and intra-operativeA version of this chapter was previously published [147]32imaging could be incorporated into the robotic procedures. As ultrasound is al-ready regularly used during laparoscopic procedures, we focused on how ultra-sound and ultrasound reconstruction could be integrated. An important applicationof ultrasound during laparoscopic procedures is 3D ultrasound reconstruction andmapping of the vasculature. Localized vessels can be used to define surgical planesand to register intra-operative images to pre-operative scans.In this chapter, we describe the use of a novel, custom-designed intra-operative“pick-up” ultrasound transducer, described in detail in Section 1.6.4. We demon-strate that it can create 3D ultrasound volumes via transducer tracking methods andthat the volumes can be used for pre-operative imaging registration. A preliminarystudy of the “pick-up” transducer was presented using one tracking method [143].In this chapter, we study the accuracy of each of the three tracking methods usedwith this novel transducer and the potential for vessel reconstruction with eachtracking method. These tracking methods include using the robotic kinematics, theElectromagnetic (EM) sensor and the stereo camera.In particular, for registration, the 2D contours are combined to create a 3Dmodel of the vessels and branches. The surface representations which are extractedfrom both the ultrasound images and pre-operative models can be registered. Dueto the potential advantages that the transducer could bring to robot-assisted partialnephrectomy, this is the primary application discussed in this chapter.2.2 Methods2.2.1 Transducer TrackingLocalization and tracking of the ultrasound transducer can therefore be achievedusing several different methods: robot kinematics [142], electromagnetic tracking[46, 83] and vision based tracking [133]. In addition, it is possible to create 3Dvolumes without any external tracking equipment, using such techniques as speckledecorrelation [5, 52].For the “pick-up” transducer described in this thesis, three methods are used totrack the transducer while inside the body: robotic kinematics, embedded electro-magnetic tracker and stereoscopic vision. For this transducer, all of these methods33minimize the lever arm effects which can multiply error in calibration and track-ing because the tracked sensor/element is very close to the ultrasound linear arrayand associated imaging plane. This effect is minimized by keeping the distancebetween the tacking point (the actual tool tip, the checkerboard or EM sensor) andthe corner of the ultrasound image as short as possible.Previous literature states that the expected accuracy of the da Vinci kinematicsis approximately 1 mm, when tracking the activated joints of a single da Vinci armusing the clinically approved robot software [89]. In this situation, that accuracycompares favourably to other tracking techniques. The transducer can be graspedin a repeatable manner so that the transformation from robot to tool is fixed andknown a priori. This is important because it would be impractical to do a calibra-tion to calculate the transformation during surgery.2.2.2 Repeatability TestsTo test the repeatability of grasping and to perform the optical tracking ultrasoundcalibration, a wide baseline stereo-camera was used. A checkerboard with 3.175mm squares was placed on the transducer and instrument jaws and was tracked bythe camera. The checkerboard was tracked using the wide baseline camera, whichhas a 75 mm baseline and consists of two Flea2 cameras (Point Grey Research,Richmond, Canada), each with a resolution of 1280 x 960 [43]. The intrinsic andextrinsic parameters of the wide-baseline camera and the location of the checker-board on the tools within the camera frame were found using the Caltech CameraCalibration Toolbox [27].Some studies claim that electromagnetic sensors have an accuracy of about 1%of the workspace [69], which, in this case, is approximately 0.5 mm. These testswere performed under optimal conditions; we would expect larger errors, morelikely around 1 to 1.5 mm, due to the large metal objects in the operating room.In particular, the operating table and the da Vinci interfere with the EM sensorreadings. The six degrees of freedom EM sensor is located within the body ofthe transducer as seen in Figure 1.9. It is close to the center of the transducer tominimize lever arm effects and maximize tracking accuracy with respect to theimage plane. This sensor also does not interfere with the sterilization process as34the sensor is approved for use with the STERRAD system as well.Another approach is to optically track the ultrasound transducer using the la-paroscopic camera image(s). This has been done previously with a monocular la-paroscope tracking a checkerboard pattern on an ultrasound transducer to create anaugmented reality laparoscopic ultrasound system [133]. We used a stereo laparo-scope and larger checkerboard pattern to do the same type of tracking. Specifically,we used a wide angle NTSC da Vinci stereo laparoscope from the da Vinci SurgicalSystem (Standard). It has a baseline of 5 mm and a resolution of 720×486 pixels.The 6×2 checkerboard pattern on the pick-up ultrasound transducer had 3.175 mmsquares. Camera calibration was performed using the Caltech Camera CalibrationToolbox [27]. The checkerboard tracking and stereo triangulation functions fromthe calibration toolbox were also used to track the checkerboard on the ultrasoundtransducer. This includes locating the checkerboard corners with a Harris cornerdetector and using the known geometry of the checkerboard to accurately deter-mine the checkerboard pose. For each camera frame, the user selected the fourextreme corners of the checkerboard to initialize the software.In order to track the ultrasound image, a calibration must be completed to de-termine where the ultrasound plane is located relative to the respective sensor (daVinci kinematics, EM sensor or stereo camera) [112].N-wire calibration was used to calibrate the EM sensor and checkerboard tothe ultrasound image. We used the PLUS framework for the EM sensor calibration[94] and a similar implementation in Matlab for the checkerboard calibration. EMsensor to ultrasound image calibration was completed away from large metal ob-jects and other interference. The calibration of the stereo camera tracking to theultrasound image was completed using the wide baseline stereo camera set-up pre-viously described. Because locating an N-wire phantom within the reference frameof the da Vinci kinematics was particularly difficult, the transformation from theultrasound image to the da Vinci tool was found through calibration using a singlewall phantom [132].352.2.3 Tracking TestsSeveral tests will be undertaken to determine the accuracy of the different trackingmethods.First, before the da Vinci kinematics can be used for tracking, the graspingrepeatability must be tested. The same wide baseline camera system that was de-scribed above was used to test the repeatability, so the error of the wide baselinecamera system and checkerboard tracking also needed to be determined. To testthis camera system error, two checkerboards were placed in a fixed relative posi-tion. These checkerboards were the same 3.175mm square pattern used for trackingthe transducer. The Harris corner detection methods of extraction from the CaltechCamera toolbox used on the transducer and da Vinci tools were applied to thesefixed checkerboards. Thirty images of the two fixed checkerboards in differentposes were used to determine the camera repeatability. The error was calculated asthe difference in the thirty resulting transforms.For testing the grasping repeatability, the transducer was picked up 30 inde-pendent times (released and picked up again) using the robotic instrument. Imageswere saved from the stereo camera and the transformation between the tool andtransducer was determined for each of the 30 trials [Figure 2.1].To understand the errors present in the different tracking methods, we per-formed hand-eye calibration using the Kronecker product implementation [152],solving the system AX = ZB. Using the EM sensor and da Vinci robot as an ex-ample, A is the set of 4× 4 matrix transforms corresponding to the location andorientations of the EM sensor with respect to its transmitter base and B is the col-lection of homogenous transformations of the tool tip with respect to the robotbase. We then solve for X , the transform between the tool tip and the EM sensorand Z, the transformation between the robot base and the EM transmitter base. Z issubject to change with changes in the experimental set-up, but X should be consis-tent as the transformation between the da Vinci tool tip and the EM sensor insidethe ultrasound transducer is repeatable, as described above. The position error isdefined as the Cartesian distance between points in B and in the points defined bythe transform: Z−1AX .We use the Optotrak Certus Motion Capture System (Northern Digital Inc.,36Figure 2.1: Grasping Repeatability testing set-up. The transducer was pickedup 30 independent times and the transformation between the checker-board on the tool jaws and on the transducer was calculated.Waterloo, Canada), as the ground truth during these experiments because it is atracking system that uses active Infrared Light Emitting Diode (IR LED)s, has anaccuracy of 0.1 mm and resolution of 0.01 mm over its working volume [138].A rigid body with four active infrared Light Emitting Diode (LED)s was attachedto the transducer and the transducer was moved through 75 to 100 poses whilegrasped by the da Vinci. Three sets of data were collected with slightly differentexperimental set-ups and for each set, X and Z were calculated. Three hand eyecalibrations were completed with this data, the da Vinci and EM sensor, both withrespect to the Optotrak and the da Vinci to the EM sensor. Finally, because ofdifferences in the experimental setup, we compared the tracking accuracy of theda Vinci stereo camera tracking to the Optotrak. In order to complete this, a rigidbody of Optotrak markers was created surrounding a checkerboard of the same sizeand dimensions as the one attached to the transducer. Two sets of data of 100 poseswere collected.2.2.4 Image Processing, 3D Reconstructions and RegistrationThe contours of the vessels within the ultrasound images of the reconstructed vol-ume were segmented manually for each of the 18 volumes (6 da Vinci kinematics,37Figure 2.2: Vessel Phantom. Top left: B-mode, top right: CT, bottom left:Power Doppler and bottom right: Phantom photo6 EM, and 6 stereo camera) after reconstruction. Once the contours from each im-age are found, a surface was created from the points on the surface of vessels usingregularized marching tetrahedra [171]. Segmentations were all completed by thesame person for consistency and to minimize bias. This surface structure was thenused for registration with a previously created surface representation of the vesselstructure created from pre-operative imaging. In this case, pre-operative CT scansof the phantom were used and contours were created through manual segmentation.A custom designed flow phantom (Blue Phantom, Redmond, WA) was used dur-ing the experiments [Figure 2.2]. This phantom consists of a single vessel whichbranches once midway through the phantom. The vessel diameter varies from 4 to6 mm.The surface-to-surface registration was completed using Principal ComponentAnalysis (PCA) [11] followed by ICP [22]. Ball bearings were placed under thevessel phantom during the CT scans and subsequent ultrasound scans be used forregistration validation. To validate the reconstructions, the distances between theball bearings in the CT scan (#1 to #2, #1 to #3 and so on) were compared to thedistances between the bearings in the ultrasound volumes. To verify the registra-tions, the distance between corresponding ball bearings was calculated, and theDice coefficient [32] and volume error (the difference in the two overall volumes)were also calculated. The distance between corresponding ball bearings is defined38as the Target Registration Error (TRE). In addition, the average distance was cal-culated between every point on the CT surface and its closest counterpart on thereconstructed surface.Table 2.1: Standard deviations of the fixed checkerboards, used to determinethe repeatability of the wide baseline optical camera tracking system.Roll Pitch YawAngles (deg) 0.13 0.15 0.07X Y ZTranslation (mm) 0.012 0.010 0.022.3 Results2.3.1 RepeatabilityFirst, the tracking error of the wide baseline camera system needed to be deter-mined. The standard deviations for the three Euler angles and Cartesian compo-nents of the transforms resulting from testing two fixed checkerboards are shownin Table 2.1. The axis are defined on Figure 2.1, roll is the rotation around the Zaxis, pitch, the rotation around the X axis and yaw, the rotation around the Y axis.From the 30 release and capture trials, a transformation between the tool jawsand the ultrasound transducer was calculated using the position of the checker-board markers in the frame of the stereo camera [Figure 2.1]. The transformationswere converted to rotations and positions and centered on the mean values. Thedistributions of these values are shown in Figure 2.3.The grasping repeatability results demonstrate repeatability better than the ac-curacy of the da Vinci kinematics and the EM tracking and approaches the accuracyof the checkerboard tracking itself. This shows that the ability to grasp the probeis repeatable and the transformation between the tool and the transducer can beconsidered fixed.392.3.2 CalibrationThe results for the ultrasound calibration using the da Vinci kinematics, EM sensorcalibration and stereo camera laparoscope tracking are described in turn. For eachsensor, a slightly different measure of calibration accuracy was used. For the daVinci kinematics calibration, the residual error using a single wall phantom was2.2 mm, averaged over three calibrations using 60-65 images. For the EM sen-sor calibration, the re-projection error for EM sensor calibration using the N-wirephantom was 0.94 mm, using 200 calibration images. The re-projection error is thedifference in position of the N-wires, projected using the calibration matrix com-pared to the segmented N-wires in the ultrasound images. For the stereo camerato ultrasound image calibration, the pinhead Point Reconstruction Accuracy (PRA)was 1.3 mm. The point reconstruction accuracy is the Euclidean distance fromthe average of the estimated pinhead location to the actual pinhead location. Theestimated pinhead location was determined by segmenting the pinhead from theultrasound image and transforming that point to the stereo laparoscopic coordinatesystem. The actual location is the location of the pin, as seen by the stereo laparo-scope, after the fluid in the ultrasound imaging bath is drained. During calibration,Edgcumbe et al. used a wide baseline camera to track the ultrasound transducer,an N-wire phantom for ultrasound calibration and the da Vinci stereo laparoscopewas used to determine the accuracy of the calibration [43]. 30 and 22 transducerposes were used for ultrasound calibration and calculating the PRA respectively.Since there is an unknown amount of error between the EM sensor measure-ments and the da Vinci kinematic measurements, both were compared to the NDICertus Optotrak. Table 2.2 shows the mean position errors and standard deviationsfor three calibration set-ups using 50, 75 and 75 poses for the calibration data re-spectively and 15, 25 and 25 separate, unique poses for the test data. The resultspresented are mean and standard deviations of 50 randomized iterations of the cal-ibrations and tests, such that the poses in each data set were randomly assigned tocalibration or test sets.One hundred poses of a rigid body were also collected with both the da Vincistereo camera and the Optotrak. The data was separated randomly into 75 posesto be used for the hand-eye calibration and 25 poses to be used for testing. Fifty40Figure 2.3: Grasping Repeatability. The mean-centered components (cen-tered at zero) of the transformation between the tool jaws and the trans-ducer computed using stereo tracking. See Figure 2.1.iterations of the hand-eye calibration were completed, in which the poses wererandomly assigned to either the calibration or test sets. The average results of the50 iterations are shown in Table 3D Reconstructions and RegistrationWe created 3D ultrasound reconstructions of a vessel phantom using the EM sen-sor, da Vinci kinematics and stereo camera tracking systems. With each system, sixvolumes were reconstructed and registered to the CT scan of the same vessel phan-tom. The contours in each image were segmented manually using B-mode images.The ball bearing fiducials were located and the distance between each fiducial wascalculated and compared to the distances from the CT volume. The mean errorand standard deviations between the bearing locations for each tracking modality41Table 2.2: Position Errors for Hand-eye Calibration between the da Vinci(dV), electromagnetic sensor (EM) and OptotrakData Set Mean and Standard Deviation (mm)CalibrationData Error(mm)Test Data Er-ror (mm)1 dV to Optotrak 2.6 ± 1.8 2.5 ± 1.9EM to Optotrak 5.6 ± 2.9 5.5 ± 3.3dV to EM 7.9 ± 3.1 7.7 ± 3.52 dV to Optotrak 3.5 ± 2.5 3.7 ± 2.6EM to Optotrak 6.8 ± 3.2 6.9 ± 3.5dV to EM 10.7 ± 4.6 10.8 ± 5.13 dV to Optotrak 3.2 ± 2.3 3.3 ± 2.4EM to Optotrak 5.9 ± 2.9 5.9 ± 3.2dV to EM 10.1 ± 4.2 10.6 ± 4.4Table 2.3: Position errors for the hand-eye calibration of the Optotrak and daVinci stereo cameraData Set Calibration Data Error (mm) Test Data Error (mm)1 1.58 ± 0.92 1.62 ± 0.942 0.78 ± 0.54 0.82 ± 0.54are shown in Table 2.4, Table 2.5 and Table 2.6. This error is a representation ofreconstruction accuracy, while the TRE, the mean distance between correspondingfiducial bearings, represents the registration error.The errors between the CT volumes and volumes created from the three sen-sors were also calculated. The volume error, the difference in the volumes of theCT and the sensor-based reconstructions, was calculated, as it was seen that thevolume segmented from the ultrasound images was typically smaller than that ofthe CT scan. The relative volume error reported in Table 2.4, Table 2.5 and Table2.6 is the ratio of the volume of the CT reconstruction and the reconstruction fromthe respective sensor. To find a measure of registration error, we used the Dicecoefficient, which is twice the volume overlap divided by the sum of the two vol-42umes. Note that the Dice coefficients drop very quickly with errors in registration.For example, a mis-registration of 1mm along the x-axis of the CT volume withitself results in a coefficient of 0.8. Table 2.4, Table 2.5 and Table 2.6 outline theresults from these volumes. Note that the volumes collected and reconstructed witheach method correspond, ie volume 1 from the da Vinci and volume 1 from the EMsensor are the same volume, collected at the same time while the camera volumewere collected separately and correspond to volumes 7-12.43Table 2.4: Reconstruction and registration errors for the volumes of the vessel phantom and targets collected with theda Vinci as a tracker.Volume Mean Target SpacingError (mm)Mean TRE(mm)DiceCoefficientRelative VolumeError (ratio)Mean Closest SurfacePoint (mm)1 2.8 ± 1.12 5.35 ± 1.88 0.41 0.60 1.33 ± 0.942 2.7 ± 0.98 5.45 ± 1.16 0.42 0.55 1.31 ± 0.873 2.8 ± 1.28 4.91 ± 1.25 0.49 0.57 1.18 ± 0.794 0.8 ± 0.64 5.97 ± 1.49 0.41 0.53 1.33 ± 0.935 2.68 ± 0.74 5.01 ± 2.74 0.54 0.51 1.01 ± 0.636 3.60 ± 1.37 5.73 ± 1.87 0.33 0.54 1.66 ± 1.29Averages 2.56 ± 1.02 5.40 ± 1.73 0.43 0.55 1.30± 0.9144Table 2.5: Reconstruction and registration errors for the volumes of the vessel phantom and targets collected with theEM sensor as a tracker.Volume Mean Target SpacingError (mm)Mean TRE(mm)DiceCoefficientRelative VolumeError (ratio)Mean Closest SurfacePoint (mm)1 2.00 ± 1.20 8.05 ± 2.94 0.25 0.67 2.08 ± 1.532 1.49 ± 1.12 8.03 ± 1.97 0.23 0.86 2.49 ± 2.153 1.03 ± 1.05 6.9 ± 2.11 0.23 0.64 2.30 ± 1.964 1.38 ± 0.40 6.16 ± 1.95 0.21 0.61 2.48 ± 1.915 1.44 ± 0.33 6.09 ± 1.90 0.23 0.64 2.19 ± 1.686 1.41 ± 1.34 4.06 ± 1.33 0.35 0.55 1.71 ± 1.25Averages 1.46 ± 0.91 6.55 ± 2.03 0.25 0.66 2.21 ± 1.7545Table 2.6: Reconstruction and registration errors for the volumes of the vessel phantom and targets collected with thestereo camera as the tracker.Volume Mean Target SpacingError (mm)Mean TRE(mm)DiceCoefficientRelative VolumeError (ratio)Mean Closest SurfacePoint (mm)7 2.30 ± 0.68 5.15 ± 1.49 0.29 0.49 1.92 ± 1.448 5.09 ± 1.16 8.52 ± 2.15 0.37 0.37 1.50 ± 1.229 2.47 ± 0.80 7.38 ± 2.54 0.43 0.57 1.65 ± 1.5910 5.18 ± 2.76 6.39 ± 2.97 0.41 0.35 1.37 ± 1.0311 5.53 ± 2.34 7.81 ± 1.60 0.45 0.27 1.36 ± 1.0112 2.48 ± 0.80 4.59 ± 1.91 0.39 0.27 1.45 ± 1.06Averages 3.84 ± 1.42 6.64 ± 2.11 0.39 0.39 1.54 ± 1.2346The vessel surfaces were registered using PCA and visual inspection was com-pleted to verify that the surfaces were correctly registered and reconstructed. Fig-ure 2.4 shows examples of the reconstructions using each tracking method andexamples of the registration with a vessel reconstructed using the da Vinci track-ing.2.4 DiscussionWe have presented tracking methods that make an ultrasound transducer usefulduring minimally invasive surgery. The three methods for tracking the transducerhave been validated for a vessel phantom. The camera tracking has been indepen-dently validated by Edgcumbe et al. [43]. The transducer follows all the guidelinesfor intra-operative use and discussions with several surgeons have confirmed thatthis is a transducer they would like to use during their surgeries.We were able to test the repeatability of grasping the transducer with the daVinci tool and found that errors introduced by variations in the grasping were muchsmaller than the tracking errors introduced by the robot kinematics. In fact, theerrors that were found between the tool jaws and the ultrasound transducer areclose to those that were found between the fixed checkerboards, indicating that thegrasping of the transducer is repeatable.The results of the hand-eye calibrations are particularly interesting as they givea good indication of the errors introduced by the different tracking methods. Theerrors for the calibration between the EM and the da Vinci averaged 9.7 mm. It isnot clear though, which tracking method is closer to the ground truth. When thetwo tracking methods were then compared to the Optotrak, which has much lowerand known tracking error, the errors in the EM sensor calibration had an average of6.1 mm while the errors between the da Vinci and the Optotrak had an average of3.2 mm. The larger errors in the EM sensors could be related to the metal within theenvironment. In such situations, the operating room table and the da Vinci itselfcould contribute to warping of the EM fields.The da Vinci camera tracking was also compared to the Optotrak and the aver-age error between the Optotrak and the stereo camera positions was 1.22 mm. Thedata for this comparison was collected within the optimal distance range for the da47Figure 2.4: Top: Examples of the vessel phantom reconstructed from eachmodality, from left to right, the CT scan, the da Vinci kinematics, theEM sensor and the stereo camera tracking. The blue stars represent thelocations of the ball bearing targets for each of the sensor based recon-structions. Bottom: An example of the CT volume registered with avolume reconstructed using the da Vinci kinematics. The red mesh andstars represent the surface of the CT volume and the ball bearing tar-gets, while the blue mesh and stars represent the da Vinci reconstructedvolume.Vinci cameras, so larger errors are expected as the checkerboard is moved furtherfrom the camera.The vessel reconstructions with the da Vinci had an inter-fiducial spacing errorthat was within the tracking errors of the robot and ultrasound calibration. Vi-sually, the reconstructions looked correct, without major distortions in the ves-sel structure. The surfaces were well matched, as seen by the low closest pointdistances. The registration errors may decrease with a more rigorous registrationmethod, or a more complicated vessel surface. The blood vessel phantom used forthe experiments is Y-shaped and does not well constrain all the degrees of freedom,48presenting a challenging but anatomically realistic model.The EM sensor had similar results for the reconstruction of the vessel phan-tom. The mean target spacing error was slightly lower than with the da Vinci, butthe correspondence errors were larger. However, visual inspections of the volumesreveal larger distortions in the vessel structure than were seen with the da Vincivolumes. This is partly attributed to the larger closest point means and the largevariations. The larger distortions are due to the EM environment around the daVinci, as reconstructions performed in a metal-free environment did not displaythese distortions and were more consistent. Additional testing needs to be com-pleted to determine the primary cause of the EM field warping. Warping of the EMfield by the operating room table is static and warping from the moving da Vinciarms is dynamic. In the case of static warping, it would be possible to correct thedistortions using one of the other tracking methods, either camera or kinematicswith methods described by Kindratenko et al. [80]. In the case of warping fromthe operating room table, the use of a flat panel EM transmitter would negate someof the effects. The flat panel would be placed under that patient, blocking some ofthe effects from the table.The reconstruction using the stereo camera had the largest mean spacing er-ror and TRE. The reconstructed trajectory of the ultrasound and associated bloodvessel path has several unexpected jumps that were caused by inaccurate opticaltracking of the ultrasound transducer. These jumps are visible in the stereo camerareconstruction in Figure 2.4. Specifically, at the extreme ends of the vessel thereare visible distortions. This is likely because the distance between the camera andtransducer was outside the optimal distance for the stereo camera tracking. At oneend of the vessel, it was inside the focal distance of the cameras and in the extremelower corner of the images. At the far end, the checkerboard was very small inthe images, limiting the accuracy of the corner detection. The distance betweenthe camera and transducer, (z) was between 50 mm to 200 mm and the camera hada narrow baseline of 5 mm. Thus the depth to baseline ratio ranged from 10 to40 during the experiment. The range of depth is significant because the 3D stereolocalization error scales in a nonlinear manner (z2) according to the following for-mula:49δ z =−z2f Bm , (2.1)where f is the focal length (pixels), B is the stereo baseline (mm), and m is theuncertainty in disparity (pixels). In Section 2.2.3, for a checkerboard at a depthof 10 cm from the camera, we calculated the accuracy of stereo point localizationto be 0.8 mm. If we assume that all the error was caused by the uncertainty indisparity, then the point localization error is 0.2 mm and 3.2 mm at depths of 50mm and 200 mm respectively. This corresponds with the range of error that wasobserved for the blood vessel surface reconstruction.The three tracking methods each have advantages and disadvantages. Thestereo camera tracking gives high tracking accuracy while the transducer is close tothe camera, and has the benefit of being tracked directly in the camera frame, mak-ing integration of augmented and virtual reality relatively simple. However, it hasthe disadvantages that the markers must be visible in the camera and clean. Thiswill be most accurate early in the surgery, but may become obscured later. Anotherdisadvantage is the limited angle for which they can be seen by the camera; themarkers on the transducer can only be accurately tracked when they are facing thecamera. There will be times during the procedure when the markers will not bevisible.On the one hand, the EM sensor has the disadvantage that it may be warpedby the electromagnetic fields and metal objects in the surgical environment. Onthe other hand, it has the advantage that the transducer can be tracked anywherein the surgical field, even when out of view of the camera. We hypothesize thatit can be used to direct the surgeon to the location of the transducer, through theuse of on-screen cues, if the transducer were to be placed somewhere outside thecamera’s view.The robot kinematics has high accuracy and can be used throughout the surgerywithout any degradation in tracking accuracy. It should be noted that this accuracyis only valid for the activated joints on a single arm; the errors in the set-up joints ofthe da Vinci can be very large as different types of potentiometers are used in thesejoints. In order to use the kinematic tracking (or the EM sensor) for augmentedreality, the tracking coordinate frames need to by registered at the beginning of the50surgery, likely using a hand-eye calibration method. We may be able to use thestereo markers to calibrate the other trackers to the camera frame and un-warp theinconsistencies in the EM field.By appropriate decision making throughout the surgery to determine whichtracking methods to use, on their own or in combination, we will provide the sur-geon with the best possibly localized ultrasound image.For the remainder of this thesis, the da Vinci kinematics alone will be usedto track the ultrasound transducer. The da Vinci kinematics offer an acceptableaccuracy over a large working volume, can be read in real-time during imaging,and are not affected by the metal objects in the room or the angle of transducerwith respect to the camera. These practical considerations along with the accuracyof the reconstructions, mean that the da Vinci kinematics were the best choice fortracking the transducer at this point.2.5 ConclusionThe transducer tracking techniques presented in this chapter are applicable to anultrasound transducer that can be easily grasped, aligned and released. In addition,the grip on the transducer is stable and repeatable, less than 0.1 mm and 0.2 deg,enabling the use of the robot kinematics to track the transducer and create accu-rate 3D ultrasound volumes as it is maneuvered using the da Vinci Robot. Thethree tracking methods allow for volume reconstruction when tested using a vesselphantom with internal targets.51Chapter 3Elastography using the da VinciSurgical Robot3.1 IntroductionAn increasing number of surgeries are being carried out as minimally invasivesurgeries. During this type of surgery, surgeons have limited haptic feedback, sincethey can only touch the organs with the distal ends of long surgical instrumentsthat must pass through the patient’s abdominal wall. This configuration createsinterference from friction at the trocar (the device that allows the instrument to beinserted in the body during laparoscopic surgery) as well as a fulcrum effect inwhich the tool tip moves in the opposite direction to that of the surgeon’s hand.In order to mitigate some of these challenges, surgical robots have been intro-duced and the most widely used laparoscopic robotic system is the da Vinci Sur-gical System (Intuitive Surgical, Sunnyvale, CA). The da Vinci Surgical Systemhas overcome some of the challenges of MIS by offering the surgeon a stereoscopicvision system for 3D viewing of the surgical scene and increased tool dexterity.Both of these innovations improve the surgeon’s performance [68], but the issueof haptic feedback remains unsolved. In fact, while surgeons using traditional MIStools had limited haptic feedback, the surgeons using the da Vinci have no hapticA version of this chapter was previously published [144]52feedback. The da Vinci tools have no force sensors and the surgeon receives noforce feedback. Thus, during da Vinci robotic surgery, the surgeons rely solely onvisual cues to estimate the forces they exert on the tissue.Robotic systems have been used to measure and recreate manual palpation sen-sations in an attempt to solve the ongoing issue of haptic feed back during mini-mally invasive surgery [165, 166]. Graphical displays can also be used to replacedirect haptic feedback [180]. Adding force sensing and haptic feedback to the daVinci robot have been shown to allow for lower and more controlled forces to beapplied to the tissue and suture [114, 120, 121]. The addition of force sensing canbe used for vessel and tumor identification [108, 165]. Adding force sensors to theda Vinci tools can be difficult from a practical standpoint due to the sterilizationprocess and the fact that the tools are disposable. Using motor torques as a forcesurrogate is also more difficult in this environment due to the fact that the da Vincirobot uses cables to control the end effector, which can stretch over time.We believe that ultrasound elastography is a promising alternative to directforce sensing. Ultrasound elastography provides a full image of tissue stiffness andviscous properties – the very properties that surgeons try to measure during manualpalpation. Conventional ultrasound has been integrated previously into the da VinciSurgical System using multiple types of ultrasound transducers [28, 143, 179].We have identified the two major uses of haptics needs for minimally invasivesurgery to be tumour and vessel palpation (Section 1.6.3). In open surgery, thesurgeon uses his/her fingers to feel the underlying vessels, through their complianceand pulsatile motion. In contrast, surgeons using the da Vinci robot cannot feeltumors and vessels because the da Vinci robot does not collect or transmit thisinformation.The absolute elastography technique is a promising adjunct for a broad range ofsurgeries; quantitatively measures the Young’s Modulus rather than a relative valuein comparison to the surrounding tissue. For radical prostatectomy, it will helpdelineate the prostate gland and the tumours within it. For partial nephrectomy, itwill help navigate towards the kidney and delineate the tumour boundary. For liversurgery it will help determine appropriate ablation boundaries. Once an elasticitymap of the tissue is constructed, haptic feedback based on deformation simulationcan also be provided without affecting the da Vinci teleoperation system stability.53Previous ultrasound elastography with the da Vinci robot has been primarilybased on strain imaging. Ultrasound strain imaging, which provides images ofrelative tissue deformation in response to various compression levels applied bythe ultrasound transducer [123], has also been integrated with the da Vinci SurgicalSystem [24]. That system uses the ‘Read-Write’ API [38] to overlay a palpationmotion onto the movements of the surgeon. The API is an interface with the clinicalda Vinci software.This removes some of the user-related difficulties of creating quality strain im-ages by moving the transducer with a known amplitude and frequency. Strainimaging can be used to determine the tumour extent and for image registration[158]. Because of the oversimplified assumption of a constant stress field, strainimages often contain artifacts caused by tissue inhomogeneity and stress concen-tration. Strain images are also more affected by boundary conditions.This chapter demonstrates the feasibility of obtaining absolute elastography byusing the da Vinci robot in order to create an environment in which the surgeon isable to estimate the tissue stiffness during robotic surgery. An external exciter isused to induce low frequency vibrations at multiple frequencies into the patient’sbody, while a 2D intra-operative transducer [143] acquires 1D axial displacementamplitude and phase over a given volume. These displacements are then used tocreate a 3D displacement volume used to calculate the elastic properties. Differentthan calculating the elastography from each planar 2D ultrasound image, and thencreating a volume, measuring the displacements of volume allows for a more accu-rate measurement of the wavelength through the tissue, and thus a more accuratemeasurement of tissue properties.The first section describes a method of 3D elastography that uses the clinicalda Vinci software, but is a bit more cumbersome for the user, as the user must movethe transducer slowly over the given region of interest. The second section of thischapter proposes a novel, 3D absolute elastography method with external excitationthat is suitable for the da Vinci robot using the da Vinci Research Kit (dVRK). TheDVRK and a da Vinci ”Standard” (1st generation) robot system [33, 79] are usedto determine the position and orientation of the ultrasound transducer so that thedisplacement phasors are acquired at known 3D locations. More specifically, as thesurgeon traces out a trajectory, equally spaced ultrasound images are reconstructed54Figure 3.1: A) Elastography set-up for the 3D transducer method. B) Set-upfor the extension to freehand scanning using a 2D transducer and the daVinci robot. In both cases, sector subdivision high-frame rate imagingis applied.into a regular 3D volume when the DVRK traces out the surgeons trajectory inreverse, under autonomous control.These experiments represent the first time that 3D ultrasound elastography vol-umes have been created using a freehand scanning or automated freehand tech-niques. This process requires that the imaging be synchronized with the trans-ducer motion in order to accuracy capture the tissue displacements. It requiresthat transducer position be tracked accurately in order to create a 3D volume ofdisplacements. The methods and results presented here represent a step forwardfor elastography imaging, expanding the applications to intra-operative proceduresand other freehand imaging.3.2 ElastographyIn these experiments, the SWAVE elastography system was used, along with a soft-ware interface, eScan, developed in our laboratory and described in detail in JeffreyAbeysekera’s doctoral dissertation [3]. This elastography system was described inmore detail in Section 1.4. This system is generally based on the idea of induc-ing mechanical vibrations into phantoms and tissues using an external excitation55source. Here, sector based imaging was used to image the wave displacements athigh frame rate.3.3 Experiment 1: Freehand Technique in Clinical ModeA mechanical 3D transducer has been used previously to capture a 3D volume oftissue displacements [Figure 3.1A] [14]. Such a transducer mechanically sweepsa 1D crystal array to create regularly spaced ultrasound planes, but can be largeand bulky. In this chapter, we describe the use of a tracked 2D transducer (thetransducer described and characterized in Section 2 to create the same type of 3Dvolume [Figure 3.1B].The custom-designed transducer creates a static and repeatable transform be-tween the da Vinci tool (Prograsp) and the ultrasound image. The transducer has128 elements, is 28 mm long and is operated at 7 to 10 MHz (Section 1.6.4). Forthis study in this Chapter, a sector size of 8 crystals and a sector frame rate of 625frames per second was used (Figure 1.5). The tool-to-image transform was foundusing the single-wall phantom method implemented in Stradwin [132].To synchronize the external exciter with the ultrasound image acquisition, thesurgeon using the transducer triggers the image capture by using the clutch pedalin the robot console, an event captured by the da Vinci API. This approach hasthe benefit of keeping the ultrasound transducer still while imaging takes place buthas the drawback that it interferes with the natural, smooth motion of ultrasoundscanning. See Section 3.5 for methods of improving upon the natural movementlimitation.Image acquisition begins with the collection of Radio Frequency (RF) dataalong the axial direction of the ultrasound image. The RF data is used to com-pute the displacement phasors uˆ(x, jω) from equation 1.3, at all depths in the imagewith respect to the transducer face. When a full volume of displacements phasors iscaptured, typically 15-30 frames, the real and imaginary parts of the axial displace-ment phasors (scalar values) are reconstructed into a volume on a regular grid. Thevolume reconstruction is performed using the Public Library of Ultrasound (PLUS)software architecture [95] and uses linear interpolation between points for the realpart and the imaginary part of the estimated displacement phasors. The field of56Figure 3.2: Left: Image of 6 kPa CIRS phantom lesion. Right: Image of the54 kPa CIRS phantom lesion. The colour bar shows the elasticity inkilopascals (kPa).view with an intra-operative transducer is small, about 30 mm by 30 mm. Theabsolute elastic properties of the samples are found from the grid displacementsusing local frequency estimation (Section 3.2). This final volume of elasticity canbe displayed to the operating surgeon or used to create a local model for hapticfeedback.3.3.1 Experimental Setup and ResultsA CIRS Elastic Quality Assurance Phantom, model 049 (Computerized ImagingReference Systems, Norfolk, VA), was used to evaluate the accuracy of the elasticproperties found using this method. The soft (6 kPa) and stiff (54 kPa) shallowlesions were imaged against a neutral background (29 kPa). Image acquisition wasperformed simultaneously at four different frequencies: 180, 210, 230 and 270 Hz.The soft and stiff lesions were imaged at a depth of 4 cm and a frequency of7 MHz and the resulting elastograms are shown in Figure 3.2. The diameter andaverage elasticity were measured in two images (corresponding to the middle ofthe lesion) for 5 trials for each lesion. The diameters of each lesion were alsomeasured using the caliper function of the ultrasound software [Table 3.1]. Theresults for the stiff lesion were within 6% error of the manufacturer’s specificationsand within 4% of the values found using MRE on the same phantom model [13].The softer lesion was stiffer than the manufacturer specified value but within 2%of the values reported with MRE. The results achieved through freehand ultrasound57Figure 3.3: A series of phasor volumes for the same trajectory in the CIRSphantom, of the stiff lesion. The different wave patterns can be seen ateach different frequency.elastography are repeatable, with narrow standard deviations in both the diametermeasurements and the elastic properties.Table 3.1: Results from the CIRS QA Elastography PhantomDiameter of Soft Lesion Diameter of Stiff LesionB-Mode 8.9 ± 0.6 mm 10.8 ± 0.2 mmElastogram 10.4 ± 1.6 mm 10.7 ± 1.6 mmElasticity of Soft Lesion Elasticity of Stiff LesionManufacturer Specifications 6 kPa 54 kPaMR Elastography [13] 11.1 ± 2.1 kPa 49.4 ± 16.9 kPaFreehand Elastography 10.9 ± 0.6 kPa 51.1 ± 5.2 kPa3.4 Experiment 2: Automated 3D Elastography withDVRK3.4.1 Robotic IntegrationIn previous work on SWAVE, a 3D motorized ultrasound transducer has been usedto create 3D volumes of tissue elasticity [4, 12]. These types of transducers tendto be large and there are none designed for intraoperative use. Instead, we havedesigned a custom ultrasound transducer that is controlled by a da Vinci surgicalrobot also described in Section 1.4 and shown in Figure 3.4 [143, 147].58Figure 3.4: The transducer used in this study was custom designed for usewith the da Vinci Robot [143]. It has 128 elements, an imaging facethat is 28 mm wide and a diameter of 15mm.In order to create accurate 3D volumes of the area of interest using the da Vincirobot, the DVRK is used to control the position and orientation of the tool and thusthe location of the ultrasound transducer and image [33, 79]. In order to createaccurate 3D elastic reconstructions, the transducer must be held still during eachindividual image acquisition within a volume, as the displacements in an imageplane must be captured over several phases of the excitation signal. A furtherdescription and accuracy assessment of the DVRK is given in Appendix A.1.The DVRK is used to record the desired trajectories traced out by the surgeon.The surgeon moves the transducer over the surface of a given area of interest instandard teleoperation mode, while the robot records the motion. Using this learnedtrajectory, the robot is commanded to recreate the trajectory, stopping at givenlocations to capture images with more regular spacing. A spacing of approximately0.5 mm is used.Within an ultrasound image, the displacement can only be accurately measuredalong the axial direction, which has the highest resolution, since it is the direction59of the transmitted and received acoustic waves. When measuring the tissue dis-placements, it is best if the axial directions of each image are approximately in thesame orientation. In using axial ultrasound displacements, only a single compo-nent, or projection, of a three dimensional wave is measured. If this component ischanging throughout the acquisition, it is very difficult to reconstruct the wave am-plitude, since we do not collect overlapping data. Thus, since the reconstruction isbased on Helmholtz’s equation (1.3), the most accurate reconstruction is achievedif the planes are parallel.Because the trajectory recorded from the surgeon may have changes in the axialdirection, the trajectory that is given back to the robot aligns the axial direction ofeach plane, while covering the same volume (Figure 3.1). In order to compensatefor the changing axial direction of the phasors, the average axial direction of theimage plane is found over either the volume or a sub volume depending on thetotal size and directions of the images. The average orientation is applied to eachtransformation of the desired transducer trajectory. The desired trajectory of thetransducer passed to the da Vinci robot will have all the axial directions aligned.The average rigid transformation of all the recorded orientations is found using analgorithm based on Kavan’s method as described in [78].The DVRK and the elastography software on the ultrasound machine commu-nicate through TCP/IP sockets. The resampled trajectory is loaded into the ul-trasound machine and before a new displacement plane is captured, the softwarecommands the robot to move to the next desired position. The robot is allowedsome time to complete the motion (usually 1 second as the motions are small) andreturns its ‘achieved” position. Through this communication, the software recordsthe actual achieved position and orientation of the robot for every image.At each 3D location, 25 images are collected at an effective high frame rate,using the sector based method described above. The tissue displacements betweenthese 25 images are found and the phasors are calculated at each of the desired 3Dlocations [187]. Using the measured trajectory locations from the DVRK, the pha-sors are reconstructed on a regular grid using the PLUS software architecture [95].Reconstruction is completed for the real and imaginary part of the displacementphasors to recreate a volume where each value is located in its correct geometricallocation in 3D. LFE was then used on the 3D volume to estimate the absolute value60of the tissue Young’s modulus [81].3.4.2 Experimental SetupTwo sets of experiments were used to validate this method of elastography integra-tion on the da Vinci robot. The first set of experiments use a CIRS Elastic QualityAssurance Phantom, model 049 (Computerized Imaging Reference Systems, Nor-folk, VA). The phantom has embedded lesions of known stiffness and size that canbe used to validate the accuracy of the elasticity. This phantom was used to validatethe accuracy of the elasticity reconstructions.The second set of experiments tested the system in ex vivo tissue. Ex vivo liveror kidney was purchased from a local butcher. The surface was arranged to mimicthe surface curvature of the human liver or kidney.CIRS Phantom Elastography ValidationFrom the CIRS phantom, 5 volumes were collected of the stiffest small sphericallesion and 5 volumes of the softest lesion. Each volume consisted of 30-50 imageslices, enough to image the 10 mm lesion and some surrounding tissue. Eachvolume collected used a unique trajectory.B-mode imaging was used to determine the size of the lesion and then a virtualsphere of that diameter was manually overlaid onto the elasticity volume. Theelasticity values inside the sphere are then averaged to determine the value of theelasticity.To determine the elasticity of the background, a similar method was used. Inthe B-mode image, an area of background was located, and the average elasticitywithin a sphere of 1.5 cm in diameter was calculated. This process was repeatedat two separate locations for each volume and combined to determine the averagebackground elasticity.Because the elasticity calculation has been found to be dependent on the ex-citation frequency, we measured the values at several different frequencies usingsteady state excitation. For this phantom and transducer, the frequencies were: 175,200 and 225Hz. The optimal frequencies of excitation depend on the tissue beingimaged and the frame rate of the machine. It is also recommended to use multiple61Figure 3.5: Example trajectory from the ex vivo kidney scans. The originaltrajectory of the da Vinci tool (top), the resampled trajectory (middle)and the achieved trajectory as reported by the da Vinci (bottom). Alllocations are reported in millimetres and the x, y, and z directions areindicated by red, blue and green lines respectively.62frequencies as some create nodal patterns in the tissue depending on the reflections.In addition, in places of very low displacement, such as the nodes, it is hard to de-termine the tissue motion. At different frequencies, the nodes appear at differentphysical locations and averaging over the frequencies allows the mitigation of anyartifacts caused by the nodes.As part of internal lab testing for accuracy and repeatability, the phantom wasimaged with a 3D motorized transducer, using the same methods and software pre-sented in this paper. The manufacturer values are also presented in the Table 3.2.In addition, we also compared the measured results with measurements taken withthe Siemens Virtual Touch Image Quantification (VTIQ) system on a Siemens Acu-son S2000 (Malvern, PA). This commercial system uses ARFI to measure tissuewave speed as described in Section 1.4. These wave speeds were then convertedto elasticity measurements in order to compare with the other results. For thesemeasurements, 14 individual locations were sampled for the stiff and soft lesions,as well as the background. The mean and standard deviation of these results areshown in Table 3.2.Ex vivo TissueVolumes of two different tissue types were collected. Calf liver tissue was obtainedfrom the butcher shop and fresh porcine kidney was procured from the hospital.The kidney was scanned at 175, 200, 225, 250 Hz, while the liver was scannedat 75, 100 and 125 Hz. The difference in excitation frequencies is due to thedifference in expected elastic properties. The overall elasticity of the tissue willdetermine the shear wavelength within the volume, and it is desirable for accurateelastic measurements to have at minimum one wavelength within the area scanned.One wavelength is the minimum, but increasing the number of wavelengths withinthe volume will increase the resolution of the elastic measurements.The set-up for both types of ex vivo tissue required that the tissue be restrained.The tissue was placed in a metal container lined with rubber designed to absorbultrasound. This is needed to minimize the reflections that would have been causedby the container. A plastic film of the same type that is used in ultrasound trans-ducer covers is placed over the tissue and held in place with magnets.633.4.3 Experiment 2: ResultsUnique tool path trajectories were created for each volume imaged. A subset ofone trajectory is shown in Figure 3.5. The figure shows the original trajectory col-lected from the da Vinci and the resampled trajectory used as input during volumecollection.When comparing the resampled and the achieved positions and orientationsacross all the trajectories used in this study, the average angular error was 2.87 ±0.55 degrees and the average translational error was 0.73 ± 0.18mm. The posi-tion error is defined as the 3D distance between the location of the tool tip in theresampled trajectory and the achieved trajectory. The angular error is defined asthe average inverse cosine of the three primary axes. Figure 3.6 illustrates the po-sition error between the resampled trajectory and achieved trajectory from Figure3.5. The errors are small, but some bias can be seen. With a consistent bias, thereconstructed volume should not be affected.The mean elasticity of the lesions in the CIRS phantom is found in Table 3.2and compared to the previous results on the same phantom found using the sameelastography processing techniques, but collected with a motorized 3D ultrasoundtransducer, specifically a 4DL14-5/38 transducer from Ultrasonix Medical. Theseprevious results of internal lab testing for accuracy and repeatability. All resultsare given as mean ± standard deviation in kPa.Table 3.2: The results of the elastogram volumes taken from CIRS phantomand compared to results of the same phantom captured with a 3D motor-ized ultrasound transducer, the Siemens VTIQ system and the manufac-turer specificationsSofter Lesion Stiff Lesion Background2D and DVRK 16.1 ± 2.3 63.9 ± 6.4 38 ± 23D Motorized Transducer 9.5 ± 0.8 66.7 ± 0.8 29.3 ± 5Manufacturer Specifications 6 54 29Siemens VTIQ 12.78 ± 1.36 62.44 ± 10.7 28.4 ± 3.1MRE [13] 11.1 49.4 22.3The phasor images from each volume, which indicate the tissue displacement64Figure 3.6: Top: The first six frames from the resampled trajectory and theachieved trajectory overlaid from Figure 3.5. The desired transformsare shown in solid lines, while the achieved transforms are shown withdotted lines. The magenta lines between the points indicate correspond-ing points within the trajectory. Bottom: A graph showing the values ofthe position and rotational errors for the six frames.and thus wave tracking, were examined along with the resulting elastogram to val-idate a smooth reconstruction. The phasor images should show smooth waves thatpropagate throughout the volume. Any jumps or discontinuities will cause artifactsin the resulting elastogram. These jumps or discontinuities could be caused by gapsin the reconstruction or errors in the transducer tracking; they would be viewed bythe algorithm as areas of high frequency change causing artifacts. One example ofthe phasors taken from the CIRS phantom at a 200 Hz excitation frequency and theresulting elastogram is shown in Figure 3.7. All elastograms of the CIRS phantomare shown on a scale from 0 to 60 kPa. The inclusion in this volume is clearlyvisible relative to the background within the phantom. All of the volumes imagedhad similar contrast between the inclusion and the background.65Figure 3.7: Results from the stiff lesion of the CIRS phantom. The recon-structed B-mode image of the phantom (top). The reconstructed realpart of the phasor image (middle) and the resulting 3D elastogram (bot-tom). The blue sphere outline designated the area of the inclusion usedto calculate the elasticity. Phasors are shown on a scale from -5 micronsto 5 microns and the elastogram is shown on a scale for 0 to 60 kPa.66Figure 3.8: The cross sectional graphs of the elastogram taken through thecenter of the stiff inclusion in the phantom with the 2D transducer andda Vinci system. The red line is the calculated Young’s Modulus andthe vertical blue lines indicated the edges of the stiff inclusion.Figure 3.9: The B-mode (left), real part of the phasor image (center) and elas-togram (right) of the ex vivo kidney. Phasors are shown on a scale from-3 microns to 3 microns and the elastogram is shown on a scale from 0to 60 kPa.67To demonstrate that the method works on real tissue, ex vivo kidney and liverwere scanned. The B-mode, phasors and elastograms are shown in Figure 3.9.The structure of the kidney is much more complicated than a phantom, causingdifferent wave patterns. The central calyx of the kidney is still filled with urineand the vessels in the same area are filled with blood. As the fluid filled regions donot propagate shear waves, there is a lack of displacement data in that area. Thislack of displacement causes the soft central region on the elastogram. In the phasorimages, the upper surface of the kidney can be seen as the waves follow the cortexof the kidney.The resulting measured Young’s modulus of the upper half of the kidney was37.9 kPa and the lower half of the kidney was 48.1 kPa.3.5 DiscussionThis Chapter presents two methods and associated results for 3D absolute elastog-raphy with a freehand scanning technique. Specifically, a small 2D intra-operativetransducer allows robotic laparoscopic surgeons access to the elastic values of tis-sue. The results with freehand imaging and the clinical API are comparable to boththose achieved with a mechanical 3D transducer [14] and within 5% of the resultsachieved with magnetic resonance imaging techniques [13].The methods are currently implemented using the da Vinci robotic system butcan easily be extended to any 2D transducer and any tracking system, such aselectromagnetic or optical tracking systems. This would allow absolute elasticityvalues to be found in nearly every clinical setting. We have used the da Vinci robotas an initial platform for integration for several reasons; the robotic environmenthas the most room for improvement with regards to haptic feedback, and at thesame time provides a stable and accurate platform for transducer tracking. Thequality of the elastogram is dependent on the tracking error, which in turn dependson the tracker involved. Generally speaking though, if the planes are dense, a betterelastogram is obtained, because of the higher signal to noise ratio due to averaging.One current limitation of the method using the clinical API is synchronizationof the imaging and the external exciter. This is currently addressed by using theclutch pedal in the surgeon’s console to trigger imaging at a specified phase of the68exciter. But is addressed in a more comprehensive manner in the second methodpresented here.The second method developed using the DVRK automates the scanning process.The robot is only automated on a path that is provided by the surgeon. The automa-tion of the robot path allows for evenly spaced image acquisition and alignment ofthe axial direction, which is needed for accurate quantitative results. Automationof the imaging makes for reproducible results and makes the acquisition of dataless dependent on the experience of the user. One remaining challenge is the in-troduction of automation into the operating room. The system currently relies onthe use of the DVRK in order to automate the image collection process. This typeof automation has not yet received regulatory approvals. In its current form, thesystem described attempts to minimize the potential risk to the patient, by keep-ing the amount of autonomous trajectory planning to a minimum, and keeping theautonomous trajectory in line with what was performed by the surgeon.The resulting position and orientation of the achieved da Vinci trajectoriesmatch well the desired trajectories. Across all the trajectories used in this study, theaverage angular error was 2.87 ± 0.55 degrees and the average translational errorwas 0.73 ± 0.18 mm. Internal lab testing similar to that done by Kwartowitz et al.[89] as described in Appendix A.1, found that the DVRK has a target localizationerror accuracy of 1.16 ± 0.46 mm, over a total of 36 individual points. This meansthat the DVRK can localize the tool tip within 1.16 mm, so each achieved positionhas the possibility of deviating around 1 mm. The commanded trajectory of theDVRK had about 0.5 mm spacing between the Cartesian positions. Since there issome uncertainty in the location of the DVRK, as well as in the image calibrationprocedure, the close spacing of the imaging locations allows for some averagingin the case of errors. The larger angular error compared to that of the translationalerror could be attributed to the differences in controller gains on the DVRK. TheDVRK uses a proportional, integral and differential (PID) controller to commandthe location of the position and orientation of the tool tip. For the studies in thispaper, the gains on translational joints of the robot were set much higher than thegains for the orientation joints. This configuration of gains was found to allownormal teleoperation of the robot without causing instability within the system.Higher gains on the orientation joints were tried, but the system became unsta-69ble under normal teleoperation. In the future, we would like to tune the gains toachieve tighter bounds on the rotations.The results of the elastography images compiled from multi-frequency sweepsof different phantoms and ex vivo tissue gives us high confidence in the usability ofthis method. The use of multiple frequencies diminishes the effects of nodes in thedisplacement measurements and allows for a more consistent picture of tissue elas-ticity. The reconstructed phasor images showed smooth wave patterns consistentto what is seen using the 3D motorized transducers.The results of the elastograms created with the DVRK were slightly biased com-pared to those that have previously been acquired from the same phantom using amore conventional, constrained 3D motor sweep. Recent results taken with the3D motorized transducer for the same phantom and using the same algorithms, areindicated in Table 3.2. In general the results found with the 2D da Vinci trans-ducer were about 6 kPa higher for the softer lesions than those found with the 3Dmotorized transducer, but about 3 kPa lower for the stiffer lesion. The bias of theresults could be due to a range of possible errors, some of these include errors dueto calibration, DVRK tracking, and irregular spacing. These errors could also leadto larger variance in the elastic measures, but the variance remains small relativeto the mean. Most important, these results show that it is possible to create 3Dquantitative elastograms with robotic automation, but some differences still existand further testing is needed to investigate and these differences further.When comparing the results of the da Vinci method with those of the manu-facturer and the VTIQ system, there is an issue with averaging over the volume ofthe inclusion. The VTIQ measurements are taken as point values, near the centerof the inclusion, while the presented values using the da Vinci robot system areaveraged. Looking at the cross sectional values in Figure 3.8, it can be seen thatthe inclusion is depicted as a peak in the elasticity value. The peak value is closerto what was measured with the VTIQ system and the manufacturer specifications.Since the expected wavelength in the phantom is close to the size of measured in-clusion, the LFE algorithm is unable to correctly identify the boundaries of smallinclusions. Both the 2D and 3D transducers had results that were higher than themanufacturer’s specifications for both the soft and stiff lesions. Higher frequencyexcitation could be used, but as one increases the frequency, the wave penetration70and wave displacement amplitudes decrease to the point of being undetectable orunreliably measurable. It may also be possible that the phantom has changed overtime and the manufacturer specifications are not longer truly valid.The study of ex vivo tissue shows that the method is not constrained to onlywork within phantoms. The phasor images show that the waves are propagatingwell within the tissue and elastography images are consistent with the structure ofthe kidney. In some other studies of in vivo porcine kidneys using supersonic shearwave imaging, the measured elasticity ranged from 6.9 to 8.7 kPa [53].This comparison indicates that the elastic properties of the kidney tissue couldchange significantly when taken out of the body and vary depending on the studyand method in which they were collected. Future laboratory studies, described inChapter 5, investigate how blood flow and blood pressure affect the wave propa-gation in in-vivo tissue. It is also possible that elasticity measured in the tissue ismore frequency dependent, as described by Sinkus et al. [154] and thus the steadystate shear wave measurement would result in a different elasticity than a transientshear wave as was used by Gennisson et al. [53].The liver tissue scanned does show consistent elastic results with other liverstudies. MRE liver studies report elastic properties ranging from 2 kPa to about8 kPa, depending on the level of fibrosis [71, 72, 181]. Since the technique ofmeasuring the elastic properties in MRE is similar to the method used in this paper,the comparison is fair. As with the ex vivo kidneys, the properties of the livershould be further studied in an in vivo environment.One limitation of this approach, in comparison to the 3D motorized transducer,is the changes in surface pressure caused by the changes in transducer positionduring the acquisition of the 3D volume. The pressure of the transducer on thetissue can, in itself, cause changes in the measured stiffness due to pre-compressioneffects [19], and the changes in transducer position can cause changes in the wavepattern that is measured, which is assumed to be in steady-state.3.6 ConclusionIn this chapter, we present an absolute elastography method for a tracked 2D ul-trasound transducer using two different methods, both a clinical freehand method71and an automated method using the research da Vinci Robot interface. Absoluteelastography methods provide quantitative information about the underlying tissuecharacteristics to the user. The system was implemented on the da Vinci in orderto provide surgical guidance to the surgeon.72Chapter 4Review of ElastographyMeasurements of Kidney4.1 IntroductionKidney transplant is the gold standard treatment for end stage renal disease. Asthe kidney function decreases, and the EGFR falls below 15 mL/min, kidney failureis diagnosed. The only treatments at this stage of kidney failure are dialysis orkidney transplantation. The leading causes of death for dialysis patients includecancer, cardiovascular disease, valvular disease and infections [170]. Transplant isthe preferred method as the long term survival of the patient is greatly increasedover those on dialysis [131].Unfortunately, kidney transplants fail over time. As discussed in Section 1.2.3,chronic allograft dysfunction or CAD is the main reason for graft failure [57]. CADis caused by the accumulation of IF/TA. Over time, the fibrosis and atrophy impairthe kidney’s ability to filter the blood plasma and regulate water levels and ionconcentrations [150].The progression of CAD often goes undetected as the EGFR may not fall belowclinically relevant levels until after significant fibrosis build up [116]. Biopsies al-low detection of histological changes that can be used to diagnose IF/TA. On theone hand, biopsy involves risk and pain for the patient and thus frequent biopsiesare not recommended [149]. On the other hand, it has been reported that the av-73erage transplanted kidney loss from biopsy was approximately 0.03% whereas theloss from rejection was 5% [110]. It is possible that the losses of these transplantscould have been prevented using the histological knowledge gained from a biopsy.Thus, it is important to have a non-invasive method to provide a quantifiable mea-surement of fibrosis. A non-invasive method could be used more often than biopsyand could provide a better monitoring protocol, providing more quantitative dataover time.Ultrasound elastography presents a non-invasive method for fibrosis detectionthat has been used successfully to diagnose the stages of fibrosis in the liver. Un-fortunately, the extension from the liver to the kidney has proven to be difficult[25]. If the elastic properties of the tissue, as measured with ultrasound, could beconsistently correlated with the histological changes within the kidney, it would bepossible to better monitor the progression of IF/TA and make changes in a patient’streatment that would slow the progression and prolong the life of the allograft. Thismethod of interrogation could be used for either transplant or native kidneys. Therehas been previous research on the topic of relating the measured elastic propertiesof the kidney with the clinical prognosis.This chapter present an overview of current literature and will focus on thecomplications and difficulties of kidney elastography. It builds upon the reviewfrom Grenier et al., which presents a comprehensive overview of the kidney elas-tography literature prior to 2013 [55]. The literature reviewed in this chapter usesa variety of elastography techniques that were described previously in Section 1.4,such as strain ratio, Acoustic Radiation Force Impulse (ARFI) (Siemens Medical)and SuperSonic Imagine, FibroScan, and MRE. Results for both native and trans-plant kidneys will be presented including literature addressing the difficulties ofimaging the complicated anatomy of the kidney.The papers presented were selected to provide a broad overview and illustratethe current state of the literature. This chapter will concentrate on the most currentliterature and the most important contributions, according to the author’s judgment.744.2 Type of Elastography Imaging4.2.1 Strain RatiosStrain imaging measures the compression of the kidney or section of the kidney inresponse to pressure applied to the tissue by the transducer. Most papers report astrain ratio, in which the strain of two areas is compared. In the case of the kidney,these two areas are often anatomical sections of the kidney, the cortex or medulla.In other papers, the compression of the cortex of the kidney is normalized by thecompression of the entire kidney or across the entire image [48].Using this normalization method, a study of 20 patients with either mild ormoderate fibrosis found that the levels of fibrosis as defined in the Banff Score,could be distinguished [48]. Comparing the compression of the cortex of the kid-ney with that of the medulla, cortico-medullary strain was found to be useful indetermining the level of IF/TA in a study of 45 patients [50]. Another method ofstrain ratio calculation is to use the strain ratio measured between the parenchymaand the central collecting system of the kidneys [76]. Kahn et al. looked at 112patients, but only 19 had biopsy results. They found that there was some correla-tion with high grade of fibrosis (three patients). Ozkan et al. also measured thestrain ratio using the parenchyma and central collecting system [125]. They foundthe method promising but the variability of this type of elasticity measurement washigh enough to warrant further study.Not directly looking at fibrosis, Menzilcioglu et al. compared the strain ratiowith the measured resistive index (a measure of blood flow) of the kidney [111].For measuring the strain ratio they measured the decompression of the tissue withthe fat surrounding the kidney as a reference. They looked at 121 CAN patients and40 healthy volunteers and found that the measured strain index was more sensitivein detecting CAN than the resistivity index.All the papers mentioned above use semi-quantitative strain ratios, meaningthat the numerical results cannot be compared across patients, making the resultsof these studies less applicable across large patient populations. Many of them usedifferent tissues types to regularize their measurements. Doing so minimizes theuser and pressure dependent effects that negatively impact strain imaging. Strain75and strain ratios do have an advantage in their simplicity. Acquisition and cal-culations are straightforward and simple and generally do not require specializedequipment. All the literature presented here found positive correlations betweenthe strain ratios and kidney health indicators such as level of fibrosis or resistiveindex. However, strain and strain ratios are operator dependent and often show ahigh variability depending on the user [125].76Author Year Method Number & type ofpatientsResults & NotesWeitzel[178]2004 Normal-izedstrain2 patients Compared one normal patient with one patient withconfirmed fibrosisOzkan [125] 2012 Strain ratio 42 transplant patients Positive correlation with resistive index, but lowrepeatabilityGao [48] 2013 Strain 20 patients with biopsy Compared developed and normalized strain tobiopsy resultsKahn [76] 2013 Strain ratio 112 patients Correlation only with high fibrosis patientsGao [50] 2015 Strain ratio 45 patients Used the strain ratio between cortex and medullaGao [49] 2015 Strain ratio 2 canine models Measured the strain as a function of time after renalligationMenzil-cioglu[111]2016 Strain ratio 121 CAN patients and 40healthy volunteersFound strain ratio more sensitive than resistiveindex in predicting CAN patientsTable 4.1: Existing studies using strain and strain ratios which are relevant to kidney fibrosis stiffness measurements.774.2.2 ARFI and SuperSonic Imagine (SSI)ARFI and SSI are quantitative stiffness measurements that use the acoustic powerof the transducer to create a “push pulse” in the tissue and then use fast imagingtechniques to track the speed of the resulting shear wave. Doherty et al. present acomprehensive review of elasticity imaging methods that use an acoustic radiationpush pulse [40]. These methods have a high resolution, measuring the stiffness ofa small section of tissue, about 1 cm by 1 cm. The kidney is a complicated organand the likely area of interest is the outer cortex and medulla of the kidney, wherethe functioning nephrons are located and filtering takes place. It is advantageousin this setting to be able to measure specific area within the kidney.ARFI and SSI are the most widely used methods to measure kidney stiffness,and as such, there are review papers that outline the results from these studies.Grenier et al. provides an overview of all renal elastography [55], while Zaffanelloet al. specifically targets the use of ARFI in renal assessment [185, 186]. We willdiscuss the most pertinent specifics of these articles in the following sections.These review papers noted the wide variation in reported results from renalallograft scans. Despite these variations, Zaffanello et al. point out that there isa consensus among the papers, that higher levels of fibrosis, or lower levels ofrenal function, lead to higher measured stiffness [186]. They noted the technicaldifficulties in creating repeatable and reliable measurements, citing many of thesame factors discussed below, including transducer pressure, perfusion, anatomi-cal structures and patient heterogeneity. They conclude that ARFI can be a usefultechnique in long term renal allograft assessment, but note that these are prelimi-nary studies and further improvements and standardized guidelines are necessary.As mentioned in Chapter 1, one of the limitations of this type of imaging is thedepth penetration, which limits its use for kidneys farther from the patient’s skin.4.2.3 FibroScanFibroScan is a commercial product often used in liver imaging and fibrosis staging[25]. It has been proven to provide good correlation between its stiffness measure-ment and the level of fibrosis in the liver; a few studies report its application tomeasure the stiffness of kidneys. This type of elasticity imaging uses a mechan-78ical impulse to create a wave in the tissue and then measures the resulting shearwavefront speed with a single element transducer. Unfortunately, there is only asingle image line to guide where the measurement is taking place. There are twodepth settings available, but the resulting measurement is an average along the lineof the single element transducer. The sample volume is fixed at 4cm long between25mm and 65mm below the skin surface. Without a clear indication of where themeasurement is taking place it is possible that the measurement is an average offat, cortex and collecting system of the kidney, rather than just one specific area[55].Despite some technical challenges of using the FibroScan system on the com-plicated anatomy of the kidney, researchers have looked at the correlation betweenthe FibroScan results and kidney health and function.The FibroScan system was used to measure the stiffness in renal transplants[8]. The stiffness of 55 patients was measured. The study found that there was asignificant positive correlation between the percentage of fibrosis and the measuredstiffness, reporting the average of 10 measurements. When comparing the stiffnessof stable (EGFR > 50ml/min) and impaired function (EGFR < 50 mL/min) grafts,they report a value of 22.2 ± 11 and 37.1 ± 14.2 kPa respectively. Of the 55patients measured with FibroScan, 20 had biopsies, and the rest were assumed to bestable grafts with < 5% fibrosis. In these patients, the results were also comparedto the Banff Score [155] and the different fibrosis levels could be differentiated (seeSection 1.2.3 for more details of the Banff Scoring system) based on stiffness.A study of 164 transplant patients also found positive correlations betweenkidney stiffness and advanced fibrosis [156]. The authors also looked at measure-ments from different parts of the kidney and found an average stiffness of 35.0 ±19.9 and 33.2 ± 18.6 kPa in the pole and center of the kidney respectively. Inpatients with advanced fibrosis, the average values were 42.0 ± 17.0 and 42.8 ±15.7 kPa in these two areas of the kidneys. They also looked and found inter- andintra-observer variation to be less than 4 kPa. It should be noted that the stan-dard deviations of the reported values is large in comparison to the values reported,where the standard deviation is often half of the reported value. To overcome someof the technical challenges, Sommerer et al. use software available from Echosansto reassess the elastograms from the device in order to adjust the angle, depth and79length of the shear wave to limit the area of interest to the kidney cortex. Withoutthis reassessment, they experienced high rates of measurement failure.Nakao et al. and Lukenda et al. also presented similar results [103, 115]. Theyboth found that the FibroScan measurements showed a positive significant corre-lations between the measured stiffness and the level of fibrosis and were relatedto the EGFR. Of interest, Nakao et al. reported that patients with high grade ofinterstitial fibrosis (Banff grade 2) had a mean value of 53.87 ± 17.8 kPa which isthe highest reported stiffness for advanced fibrosis. Lukenka et al. report a meanstiffness of 32.2± 5.6 kPa which is close to average, 37 kPa across the studies pre-sented here. Similar to Ardnt et al., they found no correlation with the resistivityindex of the patients [8], where resistive index is a quantification of blood flow intothe kidney.Across all the studies presented here, the average Young’s modulus for healthykidneys was approximately 28 kPa, and 37.5 kPa for those with impaired function.But the values that are given in each of the studies overlap, with normal rangingfrom a possible 11 to 50 kPa and impaired ranging from 28 to 63 kPa. Althougheasy to use, the complications due to averaging along the line of measurement andthe lack of detailed imaging feedback mean that it is very difficult to provide con-sistent and reliable results for the kidney. The results from the FibroScan systemare also higher compared to those of the ARFI and SSI systems, which are generallyin the 10 to 20 kPa range.80Author Year Young’s Modulus (GFR> 50 mL/min vs GFR <50 mL/min)Correlationwith KidneyHealthNumber& type ofpatientsNotesArdnt [8] 2010 22.2±11 vs 37.1±14.2 yes 55patients,20 withbiopsyGood correlation with eGFR cutoffsSom-merer[156]2012 32.7±18.7 vs43.1±20.8not definitive 117patientsCorrelation with Banff score, variationsof 3 kPa over 3 months, correlation withincreased creatine levelsLukenda[103]2014 28±2.7 vs 33.9±5.5 yes 52 patients Correlation with eGFR and with BanffScoringNakao[115]2014 25±5.5 vs 37.6±4.2 yes 35patients,27 withbiopsyPositive correlation with increasingfibrosisTable 4.2: Existing studies using FibroScan relevant to kidney fibrosis stiffness measurements.814.2.4 MREAlthough this chapter has focused on the use of ultrasound for imaging kidneys,other papers using magnetic resonance imaging have also looked at kidney tissuecharacteristics.When using MRE, a separate excitation source is placed on the patient’s skinand steady state waves are transmitted through the patient’s abdomen. MRI can thenimage the wave propagation in three dimensions and inversion algorithms are usedto recover the elastic properties of the tissue [106]. This method of elastographyis similar to SWAVE, described in Section 1.4 and used throughout this thesis. Itis important to note that SWAVE and MRE measure the wave speed of steady statewaves, which may have different properties than the transient waves measured bythe imaging methods above. All results should also be reported with the excitationfrequency when possible since there are some frequency dependent variables thataffect the measured shear modulus. In addition, all the results presented here areof native kidneys (either human or swine), rather than transplanted kidneys.Kruse et al. reported early results on the feasibility of using MRI to collect elas-tography data of healthy native kidneys and found that the average shear stiffnesswas 16± 5 kPa [87]. Other papers focused on the reliability and reproducibility ofthe method. Low et al. imaged 16 healthy volunteers at excitation frequencies of90 and 60 Hz [102]. They scanned each patient twice in a 30 minute interval andfound that it was a reliable technique with an average renal stiffness of approxi-mately 3.5 kPa at 60 Hz and 5.7 kPa at 90 Hz.Rouviere et al. looked at 10 healthy patients and attempted to measure thestiffness in the different sections of the kidney at 45 and 76 Hz [139]. They wereable to measure the stiffness in the parenchyma (4.9 ± 0.5 kPa and 9.4 ± 0.8 kPa).It should be noted that at these excitation frequencies, the wavelength in the kidneyis 50 mm at 45 Hz and 40 mm at 76 Hz which might be a limiting factor in this typeof imaging, because the size of the parenchyma is about the same as a wavelength,so observing a full wavelength within the parenchyma may be impossible. Similarto this study, Bensamoun et al. also looked at healthy kidneys in order to assessthe normal kidney value [20]. When measuring at 60 Hz, they reported an averageshear modulus of 4.1 ± 0.2 kPa.82Other MRE papers looked at the influence of renal stenosis on the measuredstiffness of swine kidneys. Korsmo et al. found that the Young’s modulus (E) ofthe medulla changed (10.7± 0.2 kPa to 12.7± 0.4 kPa) while the cortical stiffnesswas unaffected after 10 weeks of induced stenosis [85]. In histological studies, thetwo kidney areas showed similar fibrotic levels. Warner et al. looked at the resultsfrom acute graded stenosis [177]. They found a decrease in the stiffness duringacute stenosis, which they attributed to decrease in renal blood flow.The MRE papers were more focused in establishing the baseline results for thismethod of kidney imaging and determining an appropriate protocol. It is a bit moredifficult to compare these studies, as the excitation frequency should be taken intoaccount, but, of the studies looking at normal healthy patients, the reported shearmodulus fell around 3 to 8 kPa, with one paper reporting much lower results nearto 1-2 kPa.All the papers that looked at the results from artificial stenosis of a swine kid-ney reported that the contralateral kidney (the one without the artificial stenosis)measured stiffer than the one with compromised blood flow. In the longer termstudies, the stenosed kidney also became stiff as it developed fibrosis.It is interesting to note the excitation frequencies used in these papers, rangingfrom 45 Hz to 120 Hz (with one instance of 150 Hz). These are similar frequenciesto those used during liver imaging, which generally is a more homogenous organwith a lower stiffness (4 kPa for healthy subjects) [10]. At 80 Hz, and a shearmodulus of 5 kPa, the expected wavelength within the kidney is about 2.7 cm,where the entire length of the kidney is about 7 cm. This means that the resolutionof the MRE is low compared to the size of the kidney and may not be able to resolvethe stiffness within the anatomical structures of the kidney.83Author Year Frequency(Hz)Shear Modulus (µ (kPa)) Results Number & Type ofpatientsYin [182] 2009 90 Control: 5.1±0.6Contralateral kidney:4±0.2Shear stiffness of the contralateral kidney increased comparedto one with artificial stenosisSwine models: 5 withstenosis and 4 normalYin [183] 2010 90 3-5 Increased stiffness with increased pressure, decreasing stiffnesswith decreasing flow4 swine modelsBensamoun[20]2011 60 4.2±0.2 Established protocol for kidney imaging, determine normalvalues11 healthy subjectsRouvier[139]2011 45 and 76 4.9 @ 45 Hz, 9.5 @ 76 Hz Average variability of 6% within the same subject 10 healthy patientsWarner[177]2011 120 With stenosis: 4.8±0.6Normal: 7.6±0.3Decreases in renal blood flow leads to decreased stiffness.Chronic decrease could offset increase from fibrosis6 swine models withgraded ischemiaLee [97] 2012 90, 120 and150Normal: 7.9 With Fibrosis:9.2 kPa @ 120 HzIncreased stiffness with increasing frequency and increasingfibrosis11 patients with biopsyKorsmo[85]2013 120 Normal: 10.7±0.2 WithFibrosis: 12.7± .4Correlation with fibrosis only within the medulla of the kidney 17 swine modelsStreitberger[160]2013 30-60 Medulla: 2.67 Cortex:1.64 Hilum: 1.17Bladder pressure seemingly irrelevant 9 healthy volunteersLow [102] 2014 90 and 60 3.5 @ 60 Hz, 6 @ 90 Hz 6 kPa and 3.5 kPa 16 healthy subjectsTable 4.3: Existing MRE studies relevant to kidney fibrosis stiffness measurements.844.3 Challenges of Kidney Imaging4.3.1 Patient HeterogeneityThe liver has shown promising results when comparing the level of fibrosis to theresults of elastic imaging [25]. Unlike the liver, the kidney is less homogeneous,with a complicated structure that may influence wave propagation. In addition tothe structure, the kidney also experiences a wide variety in pressures from the renalartery and possible pressure for the ureter that may affect the elastic measurements.Other factors that may influence the measurements could include age, gender, ar-teriosclerosis, transducer pressure, frequency of excitation and when looking atex-vivo specimens, storage conditions.Bota et al. looked at general factors that influence the shear wave speed in91 healthy volunteers [26]. Looking at Body Mass Index (BMI), kidney length,renal parenchyma thickness, age and gender, they found that age and gender werecorrelated to the elasticity measurements and also found some influence from themeasurement depth. These types of factors should be factored in when measuringelasticity under other conditions.4.3.2 Kidney StructureThe structure of the kidney has been shown to influence the speed of the waves.The nephrons in the cortex and medulla are aligned in such a way that tubes all runradially from the center collecting system towards the surface of the kidney in a fanshape. This means that when an ARFI transducer, for example, is placed facing thecenter of the kidney, the push pulse is travelling down the tubes of the nephron andthe shear waves will be moving across the alignment of the loops of the nephrons.Alternatively, when the transducer is placed near the pole and imaging towards theopposite pole through the cortex, the push pulse will be aligned across the tubes ofthe nephron.In particular, Gennisson et al. measured the shear wave speed in different ori-entations [53]. They found that elasticity values were higher when the ultrasoundimage and the push pulse was aligned perpendicular to the main pyramid axis.This means that the measured wave were traveling fast (higher elasticity) along85the nephron tubes than against them. The orientation dependency is not somethingthat is not well standardized in kidney elastography. In general, most of the pa-pers using ARFI and SSI align the ultrasound image parallel to the renal pyramids.However, it is not easy to be consistent in imaging the transplanted kidney sincethe kidney is not always in the same position with respect to the patient’s skin.4.3.3 PerfusionApproximately 25% of the cardiac output flows through the kidney with each heart-beat, which creates large changes in the input blood pressure. The kidney has twoautoregulation mechanisms to create constant filtering pressure inside the kidney,such that the changes in blood pressure do not affect the functionality of the kid-ney [82]. The first mechanism is the myogenic mechanism, which is the intrinsicproperty of the smooth muscle. When the pressure rises suddenly the muscle of therenal afferent arteriole is stretched and contracts to compensate. This mechanismsmoothens out the changes due to the heart cycle. The second mechanism compen-sates for larger changes in the general systemic pressure. The tubule-glomerularfeedback uses the NaCl concentration to release vasoconstriction that affects the af-ferent arteriole [82]. The vasoconstriction also causes the vessels within the kidneyto constrict, causing the apparent input pressure to increase.It has been hypothesized that the blood pressure could affect the elastographymeasurements. In particular, researchers have looked at two areas, stenosis [49, 53,85, 177, 182], where the renal artery or vein are blocked and perfusion, where theincrease in blood pressure could affect the measurements [9, 61, 183].Of the papers looking at stenosis, Warner et al. noted that the stiffness de-creased with increasing stenosis (and decreasing blood flow) [177]. Gennisson etal. also found that decreasing the blood flow through the renal artery decreased themeasured stiffness and that ligating the renal vein (increasing the internal pressure)increased the measured stiffness [53].Korsmo et al. looked at the fibrosis caused 10 weeks after the induced steno-sis [85]. They found that the stenosis increased the fibrosis in the medulla of thekidney, with no change in the cortex. Yin et al. also looked at the kidney stiffness10 weeks after induced stenosis, but instead of looking at the kidney with stenosis,86they measured the stiffness of the contralateral kidney [182]. It was suggested thatthe contralateral kidney increases blood flow to make up for the stenosis of theoriginal kidney. The authors found that the measured stiffness in the contralateralkidney was increased. In a different study, Yin et al. decreased the blood flow tothe kidney and this time measured the shear stiffness in both the stenotic and con-tralateral kidney [183]. This study found that with increasing stenosis, the stenotickidney decreased in stiffness and the contralateral kidney increased slightly.Only a few papers look at the effects of perfusion directly as it is difficult tomeasure and control. Asano et al. measured the ankle pulse velocity as a surrogatefor blood flow in the kidney [9]. The ankle pulse velocity, the velocity of theblood flow through the artery near the patient’s ankle, would indicate the levelof sclerosis. They found that with decreasing shear wave velocity, there was adecrease in EGFR, but surmised that the changes in shear wave velocity were causedmore by the change in blood flow than by changes in fibrosis.The connection between chronic kidney disease (and the development of fibro-sis) and arterial stiffness (and cardiovascular disease) is described by Garnier etal. [51]. In that paper, the authors point out that patients with compromised renalfunction (with and without hypertension) have much higher incidences of arterialstiffness. Cardiovascular death is 10-30 times higher in patients with end-stagerenal disease than in the general population. This association persists even afteradjustments for traditional cardiovascular risk factors. In light of this connection,it is possible that the differences in ankle wave velocity that Asano et al. saw couldhave been part of this association [9].Using ex-vivo kidneys, it is possible to control the arterial input pressures from60 to 120 mmHg, and 60 mmHg in the ureter [61]. In that study, porcine kid-neys (n = 11) were harvested and examined four hours after death. They foundthat with increasing the pressure, the measured shear wave velocity also increased.This the only study found that looks specifically at ex-vivo kidneys in a controlledenvironment.874.3.4 Transducer PressureStrain ratio is used to measure the changes in kidney structure and function, andmeasuring strain involves putting pressure on the kidney and measuring its re-sponse. Thus, it would not be surprising that transducer pressure on the kidneywould influence the measurements of other elastography measurement methods.Syversveen et al. looked at the relationship between ARFI measurements and trans-ducer pressure [163]. In patients with varying degrees of fibrosis, they found cor-relation only between pressure and shear wave velocity, but no correlation betweenthe shear wave velocity and the degree of fibrosis. The transducer pressure wouldbe especially important for transplant kidneys, which do not have the protection ofthe ribs from outside pressures. No subsequent papers have attempted to standard-ize the transducer pressure, so this is an unknown variable in all the papers.4.3.5 Viscous Properties of TissueMost elasticity measurement systems assume that the tissue is entirely elastic,even though tissue can have significant viscous properties as well. Two meth-ods, Shearwave Dispersion Ultrasound Vibrometry (SDUV) and Viscoelastic Re-sponse (VISR) have begun to measure the viscous properties of different tissuetypes [6, 7, 66, 151, 172]. The studies involving kidney tissue, though, are verylimited. The viscous properties of the kidney could also be contributing to thedifferences in results, as the viscous components are dependent on the excitationfrequency, which is not controlled in all measurement methods.Other models of tissue including poroelasticity, could be used to characterizethe tissue properties not explained with a linear elastic model. A poroelastic modeldescribes the tissue as a material with an elastic skeleton saturated with fluid. Thisfluid is allowed to flow freely through the pores of the matrix. This type of modelhas not been extensively applied to kidney tissue but has been used to characterizeother tissue types [45, 128].4.3.6 Incongruent ResultsThe majority of papers have reported that increasing fibrosis results in increasingmeasurements of stiffness. There have been a few notable exceptions. One of the88first papers to use ARFI to look at the correlation between fibrosis and shear wavevelocity found there was no significant correlation [162], disagreeing with the firstpositive correlation found with FibroScan [8].In a paper looking at 43 patients, Grenier et al. found that the stiffness mea-surements correspond only to the sum of all the Banff Scores for a given biopsy,but not with any particular value [54]. They also reported very high interobserverviability which might account for the lack in correlation. The measured stiffnesswas also notably higher than other reported values using supersonic shear waveimaging, with a mean stiffness of 22 kPa in the cortex and 16 kPa in the medulla.Using ARFI to measure only the cortex, Wang et al. determined from their datathat there was no correlation between the shear wave velocity and the level of renalfibrosis [176]. They did find, though, that the resistivity index was higher in thelater stages of chronic kidney disease.Contradicting most other published papers, Hu et al. found that ARFI mea-surements of impaired kidneys showed the shear wave velocity in the kidneys de-creased with increasing kidney impairment [67]. This means that as the kidneysfailed functionally, the kidneys became softer instead of harder. They also noticedthat the parenchymal thickness decreases as the function decreases.As both of these studies took place in China, it is possible that differences inthe causes of kidney failure in this region account for the differences in results. InNorth America, the main causes are hypertension and diabetes, whereas the maincause of kidney failure in China is glomerulonephritis [51].4.4 ConclusionsElastography has the potential to provide vital information about the health ofthe transplant and native kidneys. This information could potentially change themethod of treatment and prolong the life of a transplant.The results of the previous research of kidney elasticity have been mixed andvariable. Different elastography methods have produced different results, espe-cially considering the absolute measurement results. With ranges from 2 kPa to50 kPa, this shows that results across methods cannot be compared reliably. Theintra- and inter-observer coefficient of variation was found to be 20% and 12%89respectively [54] and another paper found the coefficient of variation on repeatedmeasurements to be between 8% and 15% depending on the region within the kid-ney [58]. The results from the native kidney studies and the transplant kidneystudies may also not be directly comparable. The transplantation process and im-munosuppressant drugs may affect the measurements of the kidney in a way thatis of yet ill defined, complicating the comparison of MRE results (mostly nativekidneys) to the ultrasound results in transplant patients.Many of the authors point out the confounding factors in making measurementsof the kidney. Sommerer et al. identified confounding factors with FibroScanincluding scanning depth, BMI, fluid accumulation and kidney structure in general.They also provided an outline for acceptable criteria for valid measurements [156].Due to the inconsistency of the measurements, it is important to better under-stand how confounding factors directly affect the measurement results and howthey might be addressed in the future. The confounding factors to be addressed areperfusion and how the kidney anatomy affects wave travel.It is important to look at perfusion and anatomy in an environment where mostof the possible variables can be controlled, such as an ex-vivo setting. Even in anex-vivo setting, factors such as storage temperature could potentially have an effecton the ARFI measurements [168]. Fortunately, it was found that storing the kidneysat either +4 degrees Celsius or at 20 degrees Celsius did not significantly changethe elasticity measurements from when the kidneys were measured fresh. So usingex-vivo kidneys stored overnight should be representative of fresh kidneys. Someinitial results for changes in pressure were reported, but only looked at static pres-sure rather than pressure with flow, which is more representative of kidneys in-vivo[61].In the following chapters, we examine the relationships between the input renalpressure and elasticity measurements using steady state shear wave elastography.We also examine the wave pattern within the kidney and how the anatomy of thekidney affects the waves. It is often assumed in elastography measurements thatthe tissue is entirely elastic, without viscous components. In the next chapter wewill also determine the viscous properties of the kidney tissue, as these can be aconfounding factor when measuring at different frequencies or with different typesof acoustic pulses.90Although most of the measurements are taken in the context of native andtransplant kidneys imaged trans-abdominally, all the results of the experimentsdescribed in the following chapters can be and will be applied to intra-operativeultrasound elastography. All the renal tissue characteristics should be similar, andchanges in these characteristics could increase differentiation between healthy andcancerous tissue.91Chapter 5Tissue Characteristics of PorcineKidney ex-vivo5.1 IntroductionAs described in Chapter 4 it is imperative to better understand the parameters ofkidney elasticity imaging. In many elastography algorithms, assumptions are madeincluding that the tissue is completely elastic, and that the tissue is homogeneous(Chapter 1.4). In this chapter, we look specifically at two poorly understood aspectsof the kidney imaging. The first aspect examined is the effect of input pressure onthe kidney elasticity measurements. In particular, how the elasticity of the changeswith pressure throughout the heart cycle and possible long term changes with hy-pertension. The effect of heart beat is poorly controlled during imaging in-vivoand has been shown to influence the results as discussed in Chapter 4. The secondaspect that was examined is the presence of a viscous component of the kidneytissue. The viscous effects are generally not taken into account during traditionalelasticity imaging, but could affect the results and cause the differences betweendifferent methods of elasticity imaging.92Figure 5.1: A) A diagram of the shaker set-up. B) The Solidworks renderingof the set-up, showing the frame, ultrasound transducer and the kidney.5.2 Imaging Set-upIn order to image the kidney in a controlled environment ex-vivo, a stainless steelapparatus was designed to hold the ultrasound transducer, the exciter coil and thekidney.In order to best allow vibrations to propagate through the kidney, the kidneyshould be allowed to move freely at the boundaries. To allow this, the kidney isheld away from the ultrasound transducer with a flexible silicone sheet. The sheetis stretched on the frame, supporting the kidney and holding it in place duringscanning.On the other side of the kidney, the exciter coil and the kidney are separatedby a stand off pad of silicone. This pad is approximately 2cm thick and servesto transfer the vibrations from the exciter coil into the kidney with minimal loses.This also means that the waves arrive at the kidney surface across a wider area,rather than from a point source, or small disk, as some of our previous work hasA much abridged version of this chapter was previously published [145, 146]93used [3, 14, 113].The frame is adjustable to deal with the different sizes of kidneys, and the factthat the kidney expands as the pressure increases during some parts of the testing.It is important to maintain constant pressure on the kidney, as pressure can affectthe elasticity measurement [163].The kidneys were scanned using a multi frequency steady-state wave imagingtechnique and using external excitation and a 3D motorized ultrasound transducer(Ultrasonix 4DL14, Analogic, Richmond, BC, Canada) [2, 14]. Each volume wascreated from 25 image planes, each placed approximately 0.5mm apart. Elasticityof the volume was calculated using LFE [105] and/or Finite Element Model (FEM)[64].5.3 Kidney HarvestAll the kidneys used in this thesis were harvested from pigs used in research andtraining studies that left the kidneys intact. These pigs were all female pigs andweighed approximately 50 kilograms. Within moments of euthanasia, the kidneyswere harvested in a similar manner to kidneys harvested for donor nephrectomy.The kidney is identified within the abdomen and the tissue surrounding the kidneyis removed. The vessels entering and exiting the kidney are then identified andisolated from the surrounding tissue and fat. The length of the vessel, from thekidney to the aorta and vena cava was isolated and the vessels are clamped andcut close to these major vessels. Particular care was taken to harvest as much ofthe renal vessels as possible, as these will be used later in the study. Unlike thosekidneys used for transplant, for this study, the ureter was not needed.After harvest, the kidneys were flushed with graft preservation solution, untilthe solution coming from the renal vein flowed clear. A 16 or 18 gauge cannula wasplaced into the renal artery, and tied in place with suture ties. The fluid was pushedinto the kidney with a syringe, and then a slow pump was used to continuouslypush fluid for several minutes.Both water and saline have been previously used to perfuse the kidney, but inour experience, these fluids would not flush out of the vein.The kidneys were then stored in solution at +4 ◦C until imaged, no more than9412 hours after harvest. Storage in these conditions should not have caused changesto the elastic properties of the kidney [168].5.4 Kidney Pressure Tests5.4.1 Infusion PumpTo change the pressure into the kidney, and to flush the kidney slowly, a peristalticpump was used (Fisher Scientific, part number 13-876-3). The pump has severalsizes of tubing and variable speeds that allow changes in the flow rate of the pump.A constant tubing size was used for all the kidneys for consistency. To measurethe pressure on the renal side of the kidney, a pressure gauge (Kodiak ControlsKC25-3# Low Pressure Gauge, 5 PSI) was attached to the tubing between thepump and the kidney. The renal vein was left open and pressure on that side wasassumed to be zero. Since within a live subject, the renal vein would be connecteddirectly to the vena cava, which has a pressure of about 1 mmHg, we considered thedifference between our set-up and the correct anatomic pressure to be negligiblein comparison to the pressure on the input side, which was between 60 and 130mmHg.Figure 5.2 shows the general set-up of the kidney, pressure gauge and the pump.The three components are connected through silicone tubing.For this series of tests, 5 kidneys were harvested and measured.We measured the elasticity of each kidney at three gauge pressures: zero (0mmHg), simulated diastolic (60-80 mmHg) and simulated systolic (125 mmHg) tocorrespond with average diastolic and systolic pressures [15, 35].Each kidney was measured at multiple excitation frequencies ranging from50Hz to 175Hz. As there is an intrinsic viscous component of the kidney biome-chanics, causing the apparent elasticity to increase with increasing excitation fre-quency, a linear estimation of the relationship between measured elasticity andfrequency was made and the estimated elasticity at 125Hz is presented here.95Figure 5.2: The pump and flushing set-up. The preservation solution ispumped into the kidney through a cannula which is inserted into therenal artery. A pressure gauge is in the system to measure the inputpressure into the kidney.Figure 5.3: A diagram that shows the cross section of the kidney that is im-aged during these experiments.96Figure 5.4: The relationship between measured elasticity and excitation fre-quency for three different input renal pressures. Connecting lines areused for visualization purposes only.975.4.2 Kidney Pressure ResultsIn the examination of the 5 porcine kidneys, it was found that the measured elastic-ity of the kidney was dependent on the input pressure of the pump. Increasing theinput pressure resulted in an increase in the measured elasticity, increasing froman average of 21 ± 3 kPa at zero mmHg to 34.1 ± 7 kPa at simulated diastolicto approximately 34.4 ± 9 kPa at simulated systolic pressure (130 mmHg). Whencomparing the results in a paired t-test, there was a statistically significant differ-ence between the results of the 0 mmHg and simulated systolic (p ≤ 0.05) andbetween the 0 mmHg and simulated diastolic (p≤ 0.05) but the p−value betweenthe simulated diastolic and the simulated systolic was 0.8. Although only usingfive kidneys in this experiment, we believe that the effect would be notable in allkidneys.Figure 5.4 shows the results at different input pressures for each of the differentexcitation frequencies (Figure 5.5). It can be seen that the measured elasticityincreases as the pressure increases at all excitation frequencies. This frequencydependence is further investigated in later sections of this chapter.When looking at the resulting elastograms, the differences in the different areasof the kidney become visible (Figure 5.6). The upper section of the B-mode imageis the parenchyma of the kidney, while in the center of the image is the collectingsystem. In the elastograms, it is possible to distinguish these two distinct anatom-ical structures. The collecting system shows up as soft in the elastogram, as it ismostly made of fluid. Similar to the results seen in vessels, fluid filled areas do notrepresent the true stiffness of the area [144]. The shear waves propagate in the softtissue matrix and fluid of the collecting system, and appear much shorter than innormal renal tissue.5.5 Viscous Properties of Kidney TissueIn most elasticity measurements, there is an assumption that the tissue behaviourcan be modeled solely by using a linear elastic model in which the shear modulusis assumed to only have a storage component. In reality, tissue has both elastic andviscous properties due to the physical construction of the cells and intercellularmatrix. These viscous properties become more apparent when measuring at differ-98Figure 5.5: Top: a histogram of the average elasticity for each specimen. Theaverage elasticity is measured at three different input pressures, zero,simulated diastolic and simulated systolic. Bottom: The elasticity, esti-mated at 125Hz, for all the kidney specimens at each of their measuredinput pressures.99Figure 5.6: Cross-sectional images of the B-mode image of the kidney andthe resulting elastograms at each input pressure.ent excitation frequencies, since the tissue will respond differently when vibratedat higher vs lower frequency.As the literature and different methods of measuring elasticity use various ex-citation frequencies, it is very important to understand how the tissue responds tothese changes when trying to compare results across methods and studies.Two methods to measure tissue viscous properties have recently emerged. Theyare SDUV and VISR. SDUV uses a single element “push transducer” and a “detecttransducer” [7, 172]. The detect transducer measures the phase of the wave, pro-duced by the push transducer, at two different locations along its path of travel. Attwo known locations the wave speed can be estimated for the phase velocity. Thepush transducer can push a different frequencies to measure the dependency of theshearwave speed on frequency. The measured wave speed at each frequency, rang-ing from 50Hz to 500Hz, is then fit to a mechanical model to solve for the elasticand viscous properties of the tissue. This method has been applied to an ex-vivoswine kidney [6, 7].VISR imaging also leverages the Voigt model for tissue (described in detailbelow) [66, 151]. This method employs two acoustic radiation force push pulses ata single location within the tissue. The relaxation times for the tissue are measuredand fit to the Voigt Model [151] or to a mass, spring, damper tissue model [66].This method then reports τ , the estimate of the ratio of the viscosity coefficient tothe relaxed elastic modulus. VISR is able to estimate the tissue properties in a verylocalized manner without reliance on shearwave propagation, so it is less affectedby local changes in the tissue.In a study of transplanted kidneys in-vivo using VISR, τ , the relative elasticityand relative viscosity were measured [66]. The relaxation time and these rela-100tive measurements were taken from multiple anatomical areas within the kidney.Using different combinations of these ratios, for different anatomical locations, theauthors found that this method could differentiate patients with different diagnoses.Unfortunately, the results of this method are difficult to compare to other methodssince the resulting measurements (ie. τ) are different than those reported by othersystems (ie. Young’s Modulus or shear wave velocity).These two methods both use a model fitting approach to estimate the tissueproperties, similar to what is described in this chapter.5.5.1 MethodsUsing a similar imaging set-up to that used during the pressure tests, all kidneysused for this portion of the study were imaged without flow, in the same imagingapparatus.The kidneys were all flushed with preservation solution after harvest to removeany blood within the organ, as the clotted blood could change the measurements.The kidneys are imaged in the stainless steel apparatus at six different frequen-cies: from 50 Hz to 225 Hz at intervals of 25 Hz. For each volume, 25 slicesare imaged at a spacing of 0.5 mm. The 4DL14 ultrasound transducer was used,imaging at a depth of 4 cm.5.5.2 Tissue ModelsThe mechanical models for tissue include the Voigt, b) Maxwell and c) Zener rhe-ological models. These include both the elastic (µ (Pa)) and viscous (η (Pa s))elements.These models are simple representation of the tissue models. Real tissue ismade up of cells, connective fibers, and intra- and extracellular matrix. Each ofthese components has its own elastic and viscous properties. The models presentedhere act as an aggregation of this very complicated system. Using these models wehope to gain insight into how much influence the viscous components have on themeasured elasticity.The Voigt model represents the two components placed in parallel, whereasthe Maxwell model has these components in series. The Zener model combines101Figure 5.7: There are three rheological models used in these studies. a) Voigt,b) Maxwell and c) Zener. The springs represent the elastic componentsand the dash-pots represent the viscous components.these two ideas into a single model. The derivation of these models can be foundin detail In Jeffrey Abeysekera’s thesis [3].For each model the wave speed (cs) as a function of frequency (ω) is calculatedas follows:Voigt:cs =√2(µ2 +ω2η2)ρ(µ+√µ2 +ω2η2)(5.1)Maxwell:cs =√√√√ 2µρ(1+√1+ µ2ω2η2 )(5.2)102And Zener:cs =√√√√ 2(µ21µ22 +ω2η2(µ1 +µ2)2)ρ(µ1µ22 +ω2η2(µ1 +µ2)+√(µ21µ22 +ω2η2(µ1 +µ2)2)(µ22 +ω2η2))(5.3)For each of these models, the model parameters (X) representing the elastic andviscous properties were optimized using the following, where ωi is the excitationfrequency in radians.minimizex6∑i=1(cs(ωi,X)− cˆs(ωi))2 (5.4)And the error, ξ is defined as follows:ξ =√166∑i=1(cs(ωi,X)− cˆs(ωi))2 (5.5)Eleven experiments were performed. Each experiment represents a differentkidney. All the elasticity results here were analyzed using thin volume 3D LFEtechniques [14, 113] and with FEM methods [63].The results of the these tests were fit to each different model using the MatlabOptimization Toolkit.5.5.3 Viscosity ResultsThe quality of the model fits to the observed data can be seen in the followingfigure. And the results of the variable fitting are shown in Table 5.1 and Table 5.3below. Individual graphs of each kidney used in this study can be found in SectionA.2.103Figure 5.8: The graphic results of the LFE model fitting for all the experi-ments. The stars represent the experimental data, the red, blue and greenlines represent the results of the Voigt, Maxwell and Zener Models re-spectively.104Figure 5.9: The graphic results of the FEM model fitting for all the exper-iments. The stars represent the experimental data, the red, blue andgreen lines represent the results of the Voigt, Maxwell and Zener Mod-els respectively.105Table 5.1: Results from LFE analysis for the Voigt and Maxwell models presented.Voigt Maxwellµ (kPa) η (Pa s) error (m/s) µ (MPa) η (Pa s) error (m/s)1 1.28 8.83 0.42 5.70 7.63 0.492 0.69 4.46 0.29 1.05 3.83 0.333 0.98 6.27 0.36 2.25 5.37 0.424 0.79 4.99 0.40 4.41 4.26 0.455 0.71 4.73 0.28 1.22 4.02 0.336 0.64 4.91 0.23 0.54 4.25 0.287 0.74 4.88 0.26 2.14 4.16 0.298 1.05 6.33 0.26 0.40 5.31 0.319 0.71 4.73 0.28 1.22 4.02 0.3310 0.64 4.91 0.26 1.69 4.24 0.3111 0.74 4.88 0.26 2.14 4.16 0.29Mean 0.82±0.2 5.45±1.3 0.30 2.07±1.6 4.66±1.1 0.35106Table 5.2: Results from LFE analysis for the Zener model presented.Zenerµ1 (kPa) µ2 (GPa) η (Pa s) error (m/s)1 1.28 240.93 8.83 0.422 0.69 37.85 4.46 0.293 0.98 118.25 6.27 0.364 0.79 132.86 4.99 0.405 0.71 49.86 4.73 0.286 0.64 57.80 4.91 0.237 0.74 80.03 4.88 0.268 1.05 84.03 6.33 0.269 0.71 49.86 4.73 0.2810 0.64 164.90 4.91 0.2611 0.74 80.03 4.88 0.26Mean 0.82±0.2 99.67±61 5.45±1.3 0.30107Table 5.3: Results from FEM analysis for the Voigt and Maxwell models presented.Voigt Maxwellµ (kPa) η (Pa s) error (m/s) µ (MPa) η (Pa s) error (m/s)1 0.42 2.55 0.13 136.99 2.20 0.172 0.52 1.97 0.16 10.97 1.60 0.173 0.38 2.12 0.13 20.56 1.79 0.164 0.54 1.78 0.09 20.54 1.42 0.115 1.59 2.72 0.13 0.00 3.17 0.116 0.69 2.73 0.04 15.05 2.20 0.077 0.61 2.79 0.08 20.23 2.30 0.108 1.80 2.54 0.06 0.00 2.76 0.049 1.66 3.08 0.04 0.00 3.06 0.0510 0.69 2.73 0.05 15.05 2.20 0.06Mean 0.89±0.6 2.50±0.4 0.09 23.94±41 2.27±0.6 0.10108Table 5.4: Results from FEM analysis for the Zener model presented.Zenerµ1 (kPa) µ2 (GPa) η (Pa s) error (m/s)1 416.33 79.96 2.55 0.132 516.57 2.91 1.97 0.163 381.68 30.84 2.12 0.134 539.06 11.36 1.78 0.095 376.00 0.00 3.58 0.106 692.67 16.30 2.73 0.047 606.66 7.48 2.79 0.088 449.17 0.00 3.18 0.039 1659.27 17.10 3.08 0.0210 692.67 16.30 2.73 0.04Mean 633.01±380 18.22±24 2.65±0.6 0.08109The kidneys do show a distinctive increase of measured elasticity with increas-ing excitation frequency. The models show that the viscous component is presentand influential.The FEM results do show a better correlation with the models at lower frequen-cies. The wavelength within the kidney at these low frequencies is close to the sizeof the image. When this happens it is difficult to determine the exact wavelength.The overall error in the FEM results is less than that of the LFE results as well.For both the LFE and FEM processing methods, the Voigt model had lowest overallerror for all the kidneys imaged.Parenchyma ScanningIn order to look only at the tissue of interest, the parenchyma of the kidney, wedissected the kidney and isolated the parenchyma. In addition to scanning thekidney tissue with ultrasound elasticity, we were able to use a manual indenterto test the static elasticity of the parenchyma. This indenter is made of a preciselinear stage and a force sensor [63]. The force sensor is pressed into the tissueat 5 micron increments and force measurements taken after the tissue has settled.Using these results to calculate the elasticity, it was found that the parenchyma wasapproximately 7 kPa. It is important to note these measurements are taken after thetissue has settled, and in a static manner. This means that the measurements fromthe indenter are not dependent on frequency, and should only represent the elasticproperties of the tissue.5.5.4 DiscussionWe have looked in depth at the viscous tissue characteristics of kidney ex-vivo.Constructing an apparatus to provide means of excellent wave propagation through-out the kidney, we have looked at the frequency response of the kidney in orderidentify the elastic and viscous properties. Using two different elasticity algo-rithms we have fit different models to estimate these properties. We found thatthe FEM results generally measured lower elasticity values than LFE on the samephasor data.When fitting the models to the observed data, the observed error in the Voigt110and Maxwell models was a 0.30 and 0.35 m/s respectively for the LFE analysis and0.09 and 0.10 m/s respectively for the FEM analysis.In addition, the sensitivity of the parameters to small changes in the data wasmeasured. Using the original data from experiment 6, which had very low fittingerror, small perturbations to the data were made. A random perturbation was madeto the original data, up to ±10%, for 10 trials. From these 10 trials, the standarddeviation of the parameters is shown in Table 5.5. For the Voigt, the resulting stan-dard deviations in the parameters were small, 7.5% and 2% of the mean parametervalues. For the Maxwell model, the deviation in the µ and η parameter were ap-proximately 82% and 2% of the mean value respectively. For the Zener model,the µ2 parameter varied hugely from trial to trial, while the error in the fits do notchange much, ranging from 0.15 to 0.10 (m/s). From these results, it shows thatthe Voigt model is the most sensitive to small changes in the data results.Table 5.5: The standard deviations in the parameters for each of the threemodels after 10 trials of up to ±10% perturbations in the original data.Voigtµ (kPa) η (Pa s)0.047 0.092Maxwellµ (MPa) η (Pa s)0.985 0.096Zenerµ1 (kPa) µ2 (GPa) η (Pa s)30.98 126.53 0.15Given the above results, we explored how changing the final parameters wouldchange the fitting errors. After optimization, each of the parameters was multipliedby a factor of 10, and the fitting error was calculated. For the Maxwell model, theµ parameter could be changed by a factor 1000 with less than a 0.2 (m/s) changein the fitting error. The fit for this model is dominated by the viscous component,and the value of the elastic component has little or no effect on the model.For the Zener model, the µ2 parameter could also be multiplied by up to 1000111with little change in the fitting error, less than 0.2 (m/s). This parameter is basicallya solid bar, in which case the model simplifies to the Voigt model.The results of previous ex-vivo kidney studies using the SDUV method, foundthat the elasticity and viscosity of the cortex ranged from 1.7 to 2.3 kPa and 1.8to 2.2 PaS, respectively [7]. The resulting elasticity component is larger than thevalues reported for the Voigt model here. Amador et al. were able to measure athigher frequencies, up to 500Hz, more than was possible here. It is possible thatthe higher frequencies changed the model fitting parameters. They also reportedonly values for a single kidney, some of the kidneys results reported in this chapterfall within this range.When examining the waves within the kidney, we can see that the waves are de-flected and changed by the internal kidney anatomy. When using the steady-statewave technique, it is possible to see the wave propagation throughout the entirekidney and spot these anomalies. They are sometimes correlated with visible dif-ferences in the B-mode, but sometimes there is no easily discernible cause. Otherissues also arise when there are reflections within the kidney. These are sometimescaused in ex vivo situations due to the kidney-water interface. Interestingly, theyhave detrimental effects on the apparent wavelength but do not show as poorlycorrelated areasAs mentioned above there have been numerous studies looking at the correla-tion between kidney fibrosis and elastic measurement, but little consistency in theresulting measurements, we hypothesize that these disruptions in the wave may bethe cause for some of the variety in other kidney elasticity results.One limit of this study is that the measurements were taken on ex-vivo kidneyswithout blood flow. As we have seen from the results of Section 5.4, the inputpressure can have an effect on the elasticity results. It is possible that the measure-ments could be very different quantitative results, but it seems that the trends arefairly consistent so we would expect similar results. These studies go to prove thatthere are significant viscous properties of kidney tissue that will have an effect onelasticity measurements. For example, when the optimization was run on the sameLFE data, assuming no frequency dependency viscous components, the root meansquare fitting error for the Voigt model increased from 0.33 to 0.92 (m/s). Individ-ual graphs of the results when viscous components are set to zero can be found in112Section A.2 Figure A.6.5.6 DiscussionFrom previous results published in this area it can be seen that there are confound-ing factors that are complicating the elasticity measurements in kidneys. In theseexperiments, we have attempted to create a more controlled environment to mea-sure the effects of two of the possible factors, blood pressure with flow and vis-cosity. From our results, we see that there are pressure and viscous effects on theelastography measurements, and that these are factors that should be taken intoconsideration when measuring kidney elasticity in-vivo.Unfortunately, even in this controlled environment, it is difficult to control allthe possible variables. In a study looking at transducer pressure, Syversveen et al.found that transducer pressure was a factor in elasticity measurements [163]. Itwas difficult to standardize the exact pressure that was placed on the kidney, butcare was taken to keep the pressure required to hold the kidney in place constantthroughout the measurement process. We used the constant elasticity of the thinsilicone to maintain the distance between the transducer face and the kidney, andthus maintain a constant pressure on the kidney. In the future, we would like to adda pressure sensor to our apparatus in order to measure the pressure applied to thekidney during scanning.In a previous paper describing the effects on elasticity with changes in staticpressure, Helfenstein et al. also found that increasing input pressure increased themeasured elasticity in ex-vivo kidneys [61]. Using ARFI imaging, they measuredthe elasticity as it changed from 0 mmHg to 120 mmHg. Interestingly, the mea-sured elasticity at 120 mmHg was much more variable than that measured at 0mmHg, with measurements ranging from 8 kPa to 35 kPa. This is similar to ourfindings, in that the standard deviation in the measurements at 0 mmHg was 3 kPabut at 130 mmHg our measured standard deviation was 9 kPa. They also measuredthe changes with depth, but found they had trouble measuring the entire depth ofthe kidney into the collecting system because of the limitation of using ARFI athigher depths. Fortunately, our imaging method, SWAVE, does not suffer from thislimitation.113Helfenstein et al. noted that the kidney increases in size as perfusion pressureincreases; something we noted as well (Figure 5.10). This happens as the kidneyexpands to hold and filter the blood. As the same organ must now hold more fluid,changes in the tissue result in changes in the measured elasticity. The tissue muststretch to accommodate the extra fluid.The kidney reacts to the dynamic changes in blood pressure as well. The kidneyhas an autoregulation mechanism to create constant filtering pressure inside thekidney, such that the changes in blood pressure throughout the heart cycle do notaffect the functionality of the kidney [82]. Depending on when during the heartcycle each elasticity measurement is made, the result may vary. These changesmay account for the variability in the results of previous papers.As changes in blood pressure and the dynamic changes in the kidney couldaffect the elasticity, it becomes reasonable that the measurement timing should besynchronized with one temporal location in the heart cycle. One paper mentionedthat synchronizing the measurements of healthy volunteers with the Electrocardio-gram (ECG) measurements resulted in better correlation between the EGFR of thevolunteers and the measured shear wave velocities [9]. EGFR is used as a measureof the function of the kidney and an estimate of the level of CAN. This only goesto reinforce the idea that ECG synchronizing should become standard when takingelasticity measurements.Presented here is a preliminary investigation of the effects of changes in bloodpressure on the elastography measurements of kidneys within a controlled environ-ment. We have seen that the measured elasticity increases with increasing pressureand in order to consistently use elastography as a method to measure fibrosis andthe progression of CAN, the perfusion pressure and flow need to be measured orsynchronized.Looking at the results of the viscous measurements, we have found that thetissue has a viscous component that should be accounted for during elasticity mea-surements. This also applies when comparing different types of elasticity imaging.These comparisons can be difficult as frequency dependency is not accounted forduring the comparisons of shear wave speed or estimated Young’s modulus. Themeasurement of viscous components could also be used in conjunction with otherinformation to make more definitive diagnosis. In Section A.2 Figure A.5, the indi-114Figure 5.10: Cross-sectional images of the B-mode image of the kidney atdifferent input pressures. The bars on the right side each cross sec-tion denote the two surfaces of the kidney. The width of the kidneyincreases with pressure. The width was measured as 2 cm at 0 mmHg,2.25 cm at 80 mmHg, 2.75 cm at 130 mmHg and 3.25 cm at 155mmHg.vidual graphs for each kidney are shown separately. When looking at these graphs,it is important to note how little difference there is between the fit of the threedifferent models. When looking at the sensitivity to both changes in the data andchanges in the final parameters, the Voigt model was least affected by these. Whenusing these models in the future, only the Voigt model will be presented as it is themost representative of the tissue of the three.In some cases, the slope of the fit is not close to the data as desired. This couldbe an indication that none of these models are entirely applicable to renal tissueand other models should be considered, such as a poroelastic model [84].When looking at the results with the Voigt model, the fitted elastic componentis about 0.5 kPa, with is considerably lower than all the results reported in literaturewith any machine. It is also lower than the static elastic component measured withthe indenter, as it was about 7 kPa. It is possible that these methods are too simpleand that other tissue properties need to be taken into account, such as the porousnature of tissue [84] and compressibility. But it is important that there is a largeviscous component in kidney tissue that cannot be taken for granted.In particular, this viscous component does not just affect the elastographymethods where excitation frequency is inherent, such as SWAVE or MRE. Theviscous component also affects ARFI and SSI wave speed calculations. The push115pulse used in these methods has a spectrum of frequency components. The result-ing wave front speed is measured either by measuring the peak displacement of thetissue some distance away from the push pulse [41], or by measuring the velocityof the tissue motion. When measuring the velocity, the derivative of the displace-ments is taken. This derivative acts as a high pass filter on the spectrum of thewave, weighting more heavily on higher frequencies and resulting in differences inmeasured elasticity [126].During the viscous property testing, we looked at the same data with two differ-ent analysis techniques, LFE and FEM and presented the results. We were also ableto examine the wave images of the kidney and determine that it is a complicatedorgan, and careful measurements are required in order to avoid artifacts caused byultrasound and anatomy.5.7 ConclusionAn apparatus was created for imaging ex-vivo kidney tissue. In doing so, the kidneywas imaged at multiple different input pressures and excitation frequencies. Astudy of how the tissue characteristics change with increasing input kidney pressurewas completed. It was found that input pressure does affect the measured elasticproperties.When quantifying the viscous component of the tissue by fitting the measuredwave speed to three different mechanical models, it was found that the measuredwave speed increases with increasing excitation frequency, showing that the kidneyhas viscous characteristics. Of the three models used in these experiments, theVoigt model was least sensitive to changes in the data and most sensitive to changesin the final parameters. This model will be used to model tissue in the remainderof this thesis.Very limited data regarding the kidney is currently available, so it is hoped thatthe results presented in this chapter will expand the understanding of kidney tissuecharacteristics.116Chapter 6Measurements of TransplantKidneys in vivo6.1 IntroductionThe ultimate goal of calculating the elastic properties of the transplanted kidney isto be able to monitor the health of the kidney in a non-invasive manner. It wouldbe of great benefit to the patient to be able to obtain the same information whichwould be collected from histological interrogation but without the need for biopsy[149].We have chosen to examine transplant kidney patients specifically due to thefact that transplant recipients are an at risk population, with only a single functionalkidney and taking immunosuppressants. This means that these patients could bene-fit from incremental improvements in monitoring. It has been described in Section1.2.3, a transplant is preferred to long term dialysis, but transplanted organs have alimited lifetime [131]. It is vital that the transplanted kidney be preserved to avoidthe need for repeated transplants, which creates additional demands on the numberof available transplants, not to mention an additional procedure for the patient, withall the associated risks. The location of the transplanted kidney, near the surfaceof the skin, makes them easy to image and locate, in comparison to native kidneyswhich are more protected by the ribs.For this preliminary study, our goal was to determine the feasibility of using117SWAVE to measure the wavelength and elasticity of transplanted kidneys in-vivo ina range of patients, and compare the results to the patients’ current EGFR and latestbiopsy findings. Since all patients in this study have recent graft biopsy results, thepossible correlation between the level of fibrosis as reported in the biopsy and theinitial elastic measurements can be examined in this small patient population. Wehave started this study with results in 23 patients.As described in the previous chapter, SWAVE has been used to image the elasticproperties of phantoms and ex-vivo kidneys. The difficulties and complications ofimaging kidneys have been explored and some limitations exposed. In this chapter,this knowledge is now applied to patients who in the future could derive somebenefit from this work.6.2 Patient PopulationsPatients were recruited from the Transplant Clinic at VGH. Patients were requestedto come for an additional ultrasound elastography scan after their normally sched-uled appointment at the Transplant Clinic. Patients were called at least two daysahead of their appointment at VGH. All patients had received at least one kidneytransplant, though the time since transplant varied widely. All imaging was per-formed under the ethics protocol H09-02455 Elastography in Transplantation.Patient characteristics are outlined in Table 6.1. The patient’s blood pressureand EGFR were measured on the same day as their Transplant Clinic appointmentand elastography scan. Previous work has compared the kidney stiffness to EGFR[8, 103, 115] as well as biopsy, so this variable will be used along with the biopsyresults.6.3 Patient HistologyThe gold standard for fibrosis assessment is histology from a biopsy of the kidney.An ultrasound guided biopsy is taken from one or both poles of the kidney. Thetissue is then sent to pathology to be stained and examined under a microscope.A biopsy report from the pathologist is submitted and read by the physician todetermine the level of fibrosis. Many pathologists use the Banff Scoring systemdescribed in Section 1.2.3 [155]. In the event that the pathologist does not explicitly118Table 6.1: Characteristics of the patients in this study.Patient Age(Years)BMI BloodPressureeGFR Time SinceTransplant (Days)BanffScore1 57 30.1 148/87 95 406 02 55 31.5 148/85 56 7987 23 56 22.5 138/75 26 470 04 52 30.4 126/60 45 2378 05 34 18.3 156/94 11 2935 26 73 26.5 132/62 81 5258 17 77 20.0 110/79 41 4392 08 43 28.1 119/89 5 2340 29 64 23.4 140/64 26 29 110 56 33.3 120/72 39 26 111 37 21.8 120/70 45 1956 012 48 21.5 118/74 31 422 013 50 25.0 110/78 5 3083 114 47 23.6 120/83 38 120 115 66 26.3 155/76 46 135 116 54 21.6 132/78 70 358 017 78 29.2 117/62 41 1470 018 32 20.0 100/69 106 494 019 45 17.7 105/70 57 46 020 64 21.4 157/89 17 2537 221 40 32.2 138/82 72 64 022 71 26.1 140/72 54 40 023 60 21.6 118/70 45 38 2119use this system, the level of fibrosis according to the Banff system is determinedfrom the pathologist’s notes and comments.The recruitment for this study is ongoing, and additional patients will be in-cluded in the future. The results of the most recent biopsy will be used for com-parison in this and future studies. In the case in which the patients were comingin for routine appointments, the most recent biopsy was used. For these patientswith stable grafts, it was assumed no major changes have occurred between themost recent biopsy and the elastography scan. For patients who had very recentlyreceived their transplant, results from the implant biopsy will be used during thecomparison.6.4 Imaging ProcedureEach patient was asked to lie in a supine position on the exam bed. The kidneywas first located and imaged with clinical ultrasound imaging software using a C4-7 transducer and an Ultrasonix Touch ultrasound machine. This transducer is acurvilinear transducer often used for abdominal scans. Once an optimal locationfor the ultrasound transducer had been identified, the ultrasound transducer wasfixed in place using a Civco Positioning Arm mounted on the side of the bed. Atthis time the optimal depth of imaging was also determined. It was desired thatthe entire kidney be visible within the image. It is important that the wave patternthroughout the graft be seen, to be certain there was adequate wave propagationduring image acquisition. All patients were imaged at depths between nine andtwelve centimeters, depending on the size of the patient and location of the trans-planted kidney. In the cases of these patients, the kidney was not seen to moveexcessively with breathing, so patients were not instructed to hold their breath.Each scan of 25 frames required about 20 seconds to acquire.Each kidney was imaged along its two major axes. The first position of thetransducer creates an image showing the long axis of the kidney, which allowsfor a visible distinction between the cortex and the collecting system. The secondposition of the transducer images the kidney in cross-section. Figure 6.1 showstwo of the optimal kidney ultrasound images from these patients.The shaker was placed on the patient’s skin as close as possible to the trans-120Figure 6.1: Two example ultrasound images of the transplant kidneys, longaxis views. Top: Patient 3. Bottom: Patient 4.ducer to allow for the most possible wave displacement within the kidney. Severalplacements of the shaker were tested around the transducer to determine wherethe most consistent waves were produced. The waves were monitored in real timeusing the eScan software. A lateral placement of the shaker with respect to thetransducer was typically best as this maximizes components of the shear wave mo-tion in the plane of the ultrasound image.A series of images were collected from each patient. Each set of images in-cluded 25 frames. Three sets of 25 frames were collected at each frequency andeach kidney was imaged at 50, 75, 100 and 125 Hz. In total, 600 2D images werecollected of each kidney.Due to issues with motion during acquisition, the frames were filtered and those121Figure 6.2: An example sketch of the placement of the ultrasound transducer(blue oval) and shaker (red circle) on the lower abdomen of the patientfor both the long axis and short axis views.showing motion artifacts were discarded. Motion artifacts were created when thepatient shifted position, coughed, or took an unusually deep breath. Patients wereinstructed to try to hold still during image acquisition, but this was not alwayspossible.The cortex and collecting system of the kidney were manually segmented fromthe B-mode images. Although automated and semi-automated segmentation algo-rithms exist for ultrasound images [18], it was more efficient in this case to segmentthe B-mode images manually, as most would have required manually seeding. Be-122Figure 6.3: A flow chart describing the steps of image processing.cause the measured stiffness in the cortex and the collecting system of the kidneycould be quite different, the cortex and the collecting system of the kidney weremanually segmented separately.The elasticity within the kidney area was calculated using LFE methods [105]and averaged across all the 2D images of that frequency that did not show motionartifacts [Figure 6.3]. These elasticity values were then compared to each patient’shealth characteristics and to the results of other studies.As discussed in Chapter 5, the kidney is not an entirely elastic tissue, but alsohas viscous properties. To compare the results of the ex-vivo studies with thesein-vivo studies, the measurements were fit to the same Voigt model described inFigure Patient ResultsThe results of the elasticity measurements using LFE were compared to a varietyof different patient characteristics including the patient histology as defined by theBanff Score, EGFR and patient blood pressure. Figure 6.4 shows that the generaltrend from pervious literature holds true, that patients with higher Banff Scores,also have lower filtration rates.Figure 6.5 shows the relationship between the measured Young’s modulus (es-timated at 85 Hz) and the patient’s EGFR.123Figure 6.4: The patients’ estimated glomerular filtration rate vs the BanffScore from the most recent biopsy.Figure 6.5: The average measured Young’s modulus vs the estimated GFR forboth cross sectional views for all patients.124Because some of the waves patterns within the kidney changed depending ontheir location with respect to the anatomical structures of the kidney, the phasor(wave) images of Patient 4 are presented in Figure 6.6. It is possible to see thedifferences between the waves within the cortex of the kidney and those of thecollecting system. The outline of the cortex and the collecting system are definedin the B-mode image and then overlaid onto the phasor images at each frequency.It can be seen the wavelength in the collecting system is much shorter than thewavelength in the cortex. Also, as expected, the wave pattern within the kidneychanges with increasing frequency.The shorter wavelength in the collecting system causes a change in the mea-sured Young’s modulus, creating a softer region within the center of the kidney.This was most apparent in the long axis view of the kidney, where more of thecollecting system was visible in the ultrasound image. In Figure 6.7, the averageYoung’s modulus for the cortex and collecting system at each excitation frequencyare shown. Using only the results from the long axis view of the kidney, a T-testwas performed (p = 0.07).It was found that the measured Young’s modulus for the transplant kidneysgenerally increased with excitation frequency (Figure 6.7). This indicates that thekidney exhibit some viscous properties, similar to what was found in the ex-vivostudies.The resulting elasticity measurements were fit to the Voigt model in order toestimate the viscous properties of the kidney tissue (Chapter 5) and outlined inTable 6.2. And the fitting error is defined as follows:ξ =√164∑i=1(cs(ωi,X)− cˆs(ωi))2 (6.1)The individual model fitting graphs for each patient are found in Appendix A.2.As the blood pressure was one aspect measured in the ex-vivo studies, the pa-tient’s blood pressure was compared to the measured Young’s modulus. Figure 6.8shows how the patient’s blood pressure corresponds with the measured Young’smodulus. The patients’ systolic, diastolic, Mean Arterial Pressure (MAP) and thedifference between systolic and diastolic (Delta P) are shown.125Figure 6.6: B-mode and wave images of Patient 4 at each frequency used inthis study. The outline of the cortex (solid line) and collecting system(dotted line) have been overlaid on the images.126Figure 6.7: The measured Young’s modulus of the cortex and collecting sys-tem vs the excitation frequency.Figure 6.8: The patient results measured Young’s modulus as compared tothe patients’ blood pressure taken on the day of the ultrasound scan.127Table 6.2: The Voigt model fitting results for patients. Each cross sectionalview for each patient is represented separately. The top half of the tablerepresents the long axis view and the lower half represents the short axisview of the kidney.Patient µ (kPa) η (Pa s) Fitting Error (m/s)1 1.3 15.8 0.293 2.2 16.2 0.274 6.4 13.7 0.605 7.5 16.1 0.507 13.2 12.9 0.528 1.9 16.5 0.3010 13.1 15.4 0.5912 1.4 11.3 0.4414 13.8 22.2 0.5015 4.5 11.0 0.3016 7.9 12.6 0.4418 7.7 17.5 0.7419 2.6 19.6 0.1521 10.3 8.8 0.4922 4.0 14.1 0.3923 9.4 13.2 0.06Mean 6.7 ± 4.2 14.8 ± 3.2 0.46.6 DiscussionThe results here are presented using either the Banff Score or the EGFR as an indi-cator of kidney heath. Unfortunately, both of these measurements, which are usedas a gold standard in practice, have limitations. The Banff Score is given based onbiopsy results from the kidney, but only has 4 levels of differentiation. Also, sincethis score is based on a small sample of the tissue, it is possible that the score doesnot represent the entirely of the kidney. It may be that the fibroses forms in patchesof the kidney and a biopsy could either over represent or under represent the levelof fibrosis.128The EGFR also has some limitations. The EGFR is based one of several formulaswhich include the level creatinine in the blood serum (plasma) and the patient’scharacteristics, such as age, gender, body size and race. The serum creatininethough, can be affected by other factors, not directly related to kidney function,including the muscle mass of the patient and even whether the patient has eaten alarge amount of red meat that day. A more accurate measure of glomerular filtrationrate would require taking a complete urine collection over a 24-hour period, whichis also difficult and prone to error.When looking at the differences between the Young’s modulus in the cortexwith compared to the collecting system, the collecting system was on average 10kPa softer than the cortex. Although the difference in the measurements were notstatistically significant (p= 0.07), this difference in measurements shows that thesetwo tissue types are distinct.Unfortunately not all the patients could be scanned at 125 Hz excitation fre-quency due to problems with wave propagation. In the patients with high BMI, thedistance between the excitation source and the kidney was greater than the wavepenetration. In this case, insufficient displacements were induced in the cortex ofthe kidney. Only those patients with all four excitation frequencies were includedin the model fitting.It is interesting to note the differences in Table 6.2 with the model fitting resultsof the ex-vivo kidneys. The average elastic parameter, µ , was 0.82 kPa and 5.45 Pas for the viscous parameter η for the ex-vivo kidneys. In comparison, the results offitting the in-vivo kidney resulted in 6.7 kPa and 14.8 Pa s. The differences in theseparameters may be attributed to the fact that the kidneys measured in-vivo wereinfluenced by uncontrolled variables such as abdominal pressure, bladder pressureand potential changes in perfusion. These differences could also be attributed todifferences between porcine and human kidneys.These in-vivo results were also calculated from 2D images, where as the ex-vivoresults used 3D volumes. It was assumed that most of the shear wave motion wasconfined to the plane of the ultrasound image, but this may not have been strictlythe case. With only a 2D image of the wave motion, it is difficult to compare theresults of the measured Young’s modulus as an absolute measure.1296.7 ConclusionsWe have shown that at least in our limited patient study, we were able to generateand image shear waves in transplanted kidneys, in a safe and comfortable manner,through the use of a small mechanical shaker placed on the skin of the patient.After optimization of the quality factor, the most reliable frequency range wasdetermined to be between 50 Hz and 125 Hz. This range is higher than the typ-ical range used for liver imaging, 45-60Hz, but closer to the range used for kid-ney imaging in previous MRE papers. This range provided a compromise in thetradeoff between the resolution of the imaging (higher frequency is better) and thewave propagation and penetration (lower frequency is better). At 50 Hz, the wave-length approximate the length of the kidney, resulting in possible inaccurate wave-length estimation. While at 125 Hz, there were challenges with wave propagationthroughout the kidney, especially in larger patients, as wave attenuation could bequite high, and the distance between the patients skin and the kidney, quite large.This study showed that the in vivo kidney, not surprisingly, also exhibits vis-cous properties like those found in ex-vivo porcine kidneys. These propertiesshould be taken into account in future examinations and when comparing resultsof different methods and excitation frequencies.130Chapter 7Conclusions and Future Work7.1 OverviewThis chapter provides an overview of work presented throughout this thesis andputs it in context of current research. The contributions to the field and the limita-tions of this approach are also described. The chapter concludes with a descriptionof the future directions for this work.7.2 Summary of FindingsChapter 2 presented the use of the intra-operative ultrasound transducer designedspecifically for use with the da Vinci surgical robot. This chapter looked at thedifferent tracking methods for the transducer, including the use of robot kinemat-ics, electromagnetic sensors and stereo camera tracking. A vessel phantom withembedded targets was used to validate the tracking and 3D reconstructions of eachof the three methods. The camera tracking turned out to have the lowest recon-struction accuracy, but may be improved through the use of different stereo cameratracking implementations. Due to overall accuracy and practical considerations, itwas decided to use the da Vinci kinematics as the main tracking method movingforward.The next chapter highlighted the use of the intra-operative ultrasound trans-ducer described in Chapter 2 for use in elastography measurements. Focusing on131measuring the tissue properties of kidney tumours during partial nephrectomy, thetransducer allows a surgeon increased autonomy during surgery. Two differentmethods for collecting elastography images were developed. The first uses the clin-ical mode on the da Vinci robot, which only allows the kinematics to be read fromthe robot. This method requires the surgeon to move the transducer very slowly anduse the ‘clutch’ pedal to activate the image collection. The second method lever-ages the research interface to the robot to automate the transducer motion duringimage collection. The results of the volumes collected on elastography phantomsusing these two different methods were presented here.In order to move elastography measurements into kidney tissue in-vivo it is im-perative to understand the results and the limitations of the past work in this area.Chapter 4 takes an in depth look at the different results from different elastographyimaging methods including strain imaging, FibroScan, ARFI and MRE. There is alarge amount of variation in the presented results both within the methods and be-tween the different imaging types, so other factors are also identified. The possibleconfounding factors are explored, such as arterial pressure, transducer pressure,and tissue heterogeneity. Two factors were identified to study further. The first isthe perfusion pressure of the kidney which has only been studied directly in onestudy [61]. The second factor to explore further is the tissue viscosity. Tissue vis-cosity is a known but little studied tissue property that could have large effects onthe measured elasticity of the kidney tissue.Chapter 5 looks at these two factors in more detail using a controlled ex-vivoset-up. Fresh porcine kidneys were harvested and flushed according to transplantkidney procedures. An apparatus was designed to hold the ex-vivo kidney in a waythat was well constrained and repeatable. Using a peristaltic pump and pressuregauge, the input pressure to the kidneys was carefully controlled and it was seenthat at increasing arterial pressure, the measured elasticity also increased. Thiscorroborates the earlier papers outlined in Chapter 4 that show some general in-vivo trends between the arterial pressure and measured elasticity.The apparatus also allowed for in depth exploration of the viscous propertiesof the kidney tissue. The ex-vivo kidneys were excited at frequencies from 50to 250 Hz. From these measurements, an increasing trend of measured elasticitywith excitation frequency can be seen. These results were fit to three different132mechanical models in order to identify the viscous and elastic properties of thetissue. It was determined that there are non-negligible viscous properties that likelycontributed to the variations seen across previous results. It was also found that theVoigt model was less sensitive small changes in the data and most sensitive tochanges in the final parameters than the other two tissue models.In the last chapter of this thesis, the measurement of kidney elastography wasapplied to transplanted kidneys in-vivo. Although the results of only 23 patientsare presented here, it was proved that this type of scanning is possible and canbe reliably performed. The SWAVE technique for measuring transplanted kidneyelasticity is feasible in this setting and good wave propagation within the kidneywas noted. Similar to the findings in ex-vivo kidneys, the in-vivo kidneys alsoexhibit strong viscous properties. Within these preliminary patients, there were nostrong correlations between measured elasticity and biopsy results.7.3 ImplicationsThe implications of the work presented in this thesis reach across the spectrumof kidney tissue characterization. The results from Chapter 2 and Chapter 3 mayhave an impact the surgical work flow of difficult kidney surgeries, while the workpresented in Chapters 5 and 6 attempts to explain the widespread variability ofelastography results and provide the first steps towards understanding how to stan-dardize kidney imaging.The ultrasound transducer described in Chapters 2 and 3 provides the surgeonwith a tool to extend his/her vision during surgery, specifically during the com-plicated and time sensitive partial nephrectomy procedure. The tracking and re-construction accuracy has been determined, such that the transducer can be usedduring surgery to provide the surgeon with a better understanding of anatomicalstructures under the surface, which are not visible through the camera. Creatinga transducer specifically designed for the da Vinci robot means that ultrasound in-tegration is more streamlined, the surgeon is more autonomous and can be moreefficient. Possible registration of preoperative images also provides the surgeonwith better context of the patient’s anatomy, minimizing errors and increasing thesurgeon’s confidence.133Adding quantitative ultrasound elastography to the da Vinci robot gives thesurgeon additional surgical guidance during robotic surgery. Surgeons can use theinformation provided with the ultrasound elastography to find the boundaries ofthe tumours within the kidney, as the tumours are generally stiffer than the healthytissue. Even when the extent of the tumour is not obvious within the traditionalultrasound image, surgeons would be able to assess the proper cutting planes toincrease the probability of positive surgical margins and complete removal of allcancerous tissue. This would lead to better long-term patient outcomes.The automated implementation of quantitative ultrasound elastography for theda Vinci robot is one step towards the integration within the robotic platform ina way that is acceptable and easy to use for the surgeon. In the lab, it has beenseen that creating a freehand 3D elastogram can be tricky and user dependent.The manual forward/automated backward scanning approach allows the surgeonto create an elastogram of a specific user-defined region in a simple and safe way,without needing to understand the particulars of the method. This novel use ofthe robot capability could be used not only for elastography, but also for generalimaging scans in which a slightly corrected trajectory is used for data acquisition.The implications of our work on patient health are expanded by the applicationof ultrasound elastography to the characterization of kidney transplants. If elas-tography is proven to correlate well with kidney health, the health of a patient’stransplanted kidney could be monitored more often without the need for additionalbiopsy, which presents additional risks for the patient.Looking at previous work in the area, it became clear that a more careful andtargeted approach was needed as confounding factors of the imaging were leadingto inconclusive results. Of the different factors, the impact of input pressure onelasticity measurements and defining the viscous properties of the kidney tissuewas examined in detail. From the experiments presented in this thesis, it has beenshown that these have an impact on the measured elasticity and should be consid-ered in kidney tissue models, regardless of the elastography imaging method used.In order for ultrasound elasticity of the kidney to come into more widespread useand regular practice, it is imperative to understand how different factors can affectthe resulting measurements. Without a better understanding of what can affect ameasurement and how the measurements can be confounded, it becomes impossi-134ble to compare results across patients, elastography imaging methods or platformsto gain an overall understanding of transplant kidney health.All the factors which were identified to influence the elastic measurements canbe applied to both trans-abdominal scanning of kidneys and intra-operative imag-ing described in Chapters 2 and 3. Understanding renal tissue characteristics willalso allow better understand of kidney health during a procedure. These charac-teristics could also enhance the contrast between the healthy tissue and canceroustissue, or allow better distinction between benign and malignant tumors.7.4 LimitationsAlthough extensive testing in ex-vivo setting has taken place on both phantoms andex-vivo tissue, the new ultrasound transducer has not yet been tested in an in-vivosetting. It is not expected to change the outcomes of any of the work presented here,but the specific design and expectations for tracking may change as a result of theconstraints of an in-vivo environment. In this type of setting, the space constraintsand lighting may change how the different methods of tracking the transducer areused.The integration of automated ultrasound elastography into the robotic surgicalsystem is limited by the use of the DVRK, which is not approved for clinical use.This is a proof of concept to show that this type of imaging may be a useful surro-gate for the lack of haptic feedback during robotic surgery, providing the surgeonwith additional guidance.During the ex-vivo studies, it was difficult to control for the variations of theporcine kidneys used. The kidneys were harvested from pigs used during otherstudies, and some may have had other unknown diseases which may have changedthe kidney. As an example, some of the kidneys had large cysts and were rejectedfor use in the work presented here. In addition, the original studies were often timesinvasive training labs for surgical residents. The kidneys may have undergone sometype of injury during the study, such as unusually high blood pressure or periods ofwarm ischemia. All kidneys were harvested very soon after euthanasia, but slightvariations in timing were not accounted for. Small changes in the kidneys couldhave an affect on the results here, but it is felt that all the presented conclusions are135valid.Another limitation within the ex-vivo work was that porcine kidneys were used.Porcine kidneys are often used in the literature as a substitute for human kidneys,but it is possible that the differences in the anatomy could contribute to differencesin results compared to human kidneys.The obvious limitation to the patient study is the number of patients, whichwill be remedied in the near future. These initial patients were also only imagedin two 2D cross-sectional planes. This changes the measurement of the correctwavelength if there is significant out of plane motion. The vibration source wasplaced laterally to the ultrasound transducer to limit this effect, but it would bemore accurate to take 3D volumes of the kidney. 3D volumes would also allow formore of the kidney to be measured effectively.7.5 Future WorkThe da Vinci transducer, described in Section 1.6.4, has received approval for hu-man trials, and clinical trials are in the planning stages. The accuracy of the track-ing within the surgical environment will be tested, also completing 3D reconstruc-tions on the vessel structures of the kidney. From a preoperative CT, the vesselscan be segmented and used for registration. The anatomical landmarks such as thevessel bifurcation locations and vessel direction vectors could be used as targetsfor the registration.Also, the three tracking methods have only been tested individually. Becausesome of these methods are more accurate at different stages of a procedure, devel-opment of a smart method for integrating all three is needed. Decisions need tobe made as to when one method should be trusted more than another, or weightingschemes could be developed. Testing within an in-vivo and surgical environmentwould also provide a better understanding of how each tracking method performsunder ‘real-world’ conditions. For example, blood and smoke can affect the stereocamera tracking, metal instruments cause errors in the EM sensors and large forcesapplied to the robotic instruments can create apparent kinematic errors. Each ofthese potential issues should be quantified.The next step in this project involves how to display the resulting ultrasound136volumes to the surgeon in a way that is understandable and does not distract fromtheir task. The method in which the information is displayed is critical for bothsurgeon acceptance and the proper interpretation of the information [153]. The useof augmented reality is a popular method for this type of display but challengesstill exist as to how best to show the images and which images are best used [70].These are research areas in which this ultrasound transducer can accelerate the cre-ation of different augmented reality systems since it can provide real-time trackedultrasound images integrated into a system which already uses a stereo display.The SWAVE elastography system could be enhanced through the addition ofalternate inversion methods, rather than the current LFE method [105]. For exam-ple, FEM inversion may be less prone to artifacts within the elastograms. Sinceframe rate is a limiting factor on the excitation frequency, the use of plane waveimaging to measure the displacements would allow for faster acquisition over theentire plane rather than relying on the sector method currently implemented. Thisincrease in frame rate would also limit the effects of patient motion on the images.More ex-vivo experiments should take place to increase the sample size ofthe measurements. Further adaptation of the apparatus to hold the ex-vivo kid-ney would allow better control over the direction of the shear waves created withinthe tissue. Due to the directional nature of the fibers within the kidney cortex, it isimportant to better understand the magnitude of the effect on the shear wave speed.It has been shown here that the pressure can affect the measured elasticity ofthe kidney. In the work presented, the change was large enough to warrant thatit be taken into consideration during future studies. One way to do this would bethrough ECG gating of the image acquisition. An initial study could collect imagesat different times within the heart cycle to measure the changes in measured elasticproperties of the tissue. Once an optimal time within the heart cycle has beendetermined, image collection could be timed to occur at the same time within theheart cycle and thus be consistent.Interesting connections between kidney disease and arterial stiffness have beenpointed out and also warrant further investigation [51]. The connection betweenmeasured kidney stiffness and fibrosis may not be as straightforward as it seems.Many other factors may be contributing to changes in the kidney and need to bebetter understood before general statements can be made.137In the future work for the patient studies, it is imperative to collect further datafrom more patients with a greater variety of kidneys. The full spectrum of kidneysshould be represented, from newly implanted kidneys to transplant kidneys whichhave failed to the point where the patient must return to dialysis. Also, additionalpatients who have had a ‘for cause’ biopsy will be invited to participate in thestudy. These are patients who have been referred to the Ultrasound department forbiopsy due to some concern or symptoms, such as a rejection episode. Patientswill be invited for elasticity scanning about 6 weeks post biopsy, in order to allowany inflammation of the kidney to subside. The short time between the biopsyand scan will provide the most accurate data for this study. If possible, it wouldalso be of interest to scan these patients on the day of biopsy and 6 weeks postbiopsy to compare any effects of the possible inflammation. In particular, thiswould allow the changes due to inflammation to be separated from changes dueto fibrosis build up. Inflammation or infection may have a greater effect on theviscosity measurements of the kidney as it increases the fluid within the tissue. Ifinflammation of the kidney can be reliably detected, changes in patient treatmentcan be applied soon, to prevent further rejection episodes and the longer term buildup of fibrotic tissue.Overall, any changes in the patients’ biomarkers such as biopsy results or EGFRcan be compared to elasticity measurements over time. This would give researchersand doctors a better idea of the relation between the patient’s health and the mea-sured elasticity.As was discussed in Chapter 5, the input pressure on the kidney has an effecton the measured elasticity properties. This has not yet been incorporated into thepatient scanning procedure, through the use of ECG gating, but will be incorporatedwithin the next months. The time of scanning with respect to the heart cycle canbe standardized for each patient.The current patient data sets were also limited to 2D frames of the transplantedkidney, in two different cross-sections. This limitation was in effect due to theissues with motion artifacts. Through the use of faster imaging techniques and mo-tion compensation, the image capture can be expanded to 3D volumes. With 3Dvolumes, more accurate elasticity measurements can be taken since the full wave-length in any direction can be measured. In the measurements presented in this sec-138tion, the position of the shaker was specifically positioned in the lateral directionof the ultrasound transducer, to maximize the amount of in-plane displacement.The differences between elastography imaging methods requires standardiza-tion so that it is possible to compare the results from patients across imaging tech-niques, manufacturers and care centers. In the realm of liver elastography, theRadiological Society of North America is currently attempting to standardize thescanning technique and provide quantification phantoms [17]. These phantomscould be used by all of the ultrasound manufacturers to calibrate their machinesand methods, such that all machines output the same quantifiable results. In thefuture, as the area of kidney elastography matures, similar efforts could be appliedfor the overall benefit of kidney elastography.7.6 ConclusionThis thesis presents the work to further the efforts in characterization and quantifi-cation of kidney tissue. Through the use of a small intra-operative ultrasound trans-ducer for use with the da Vinci surgical robot, kidney tissue can be better imagedduring robot-assisted partial nephrectomy. Accurate 3D ultrasound reconstructionscreated with the small transducer can be used in future surgical guidance. Withthe addition of ultrasound elastography to this system, the delineations betweenhealthy and cancerous tissue can be imaged in real-time, allowing for assessmentof kidney tumours and more accurate surgical margins.Ultrasound and ultrasound elastography were then expanded to characterizethe tissue of transplant kidneys, where the undetected development of fibrosis canimpair function of the graft and shorten its overall life. The kidney is a complicatedorgan and the work here has provided proof that certain aspects should be taken intoconsideration during future development of elastography systems for renal tissue.139Bibliography[1] American cancer society: Cancer facts and figures, 2015. URLhttps://www.cancer.org/research/cancer-facts-statistics/. [Online; accessed2016]. → pages 3[2] J. Abeysekera, R. Rohling, and S. Salcudean. Vibro-elastography:Absolute elasticity from motorized 3d ultrasound measurements ofharmonic motion vectors. In Ultrasonics Symposium (IUS), 2015 IEEEInternational, pages 1–4. IEEE, 2015. → pages 94[3] J. M. Abeysekera. Three dimensional ultrasound elasticity imaging. PhDthesis, University of British Columbia, 2016. → pages 13, 55, 94, 102[4] T. Adebar and et al. A robotic system for intra-operative trans-rectalultrasound and ultrasound elastography in radical prostatectomy. InInformation Processing in Computer-Assisted Interventions, pages 79–89.Springer, 2011. → pages 58[5] N. Afsham and et al. A generalized correlation-based model forout-of-plane motion estimation in freehand ultrasound. Medical Imaging,IEEE Transactions on, 33(1):186–199, 2014. → pages 33[6] C. Amador, M. W. Urban, J. F. Greenleaf, and L. V. Warner. Measurementsof swine renal cortex shear elasticity and viscosity with shearwavedispersion ultrasound vibrometry (sduv). In 2009 IEEE InternationalUltrasonics Symposium, pages 491–494. IEEE, 2009. → pages 88, 100[7] C. Amador, M. W. Urban, L. V. Warner, and J. F. Greenleaf. In vitro renalcortex elasticity and viscosity measurements with shearwave dispersionultrasound vibrometry (sduv) on swine kidney. In 2009 AnnualInternational Conference of the IEEE Engineering in Medicine andBiology Society, pages 4428–4431. IEEE, 2009. → pages 88, 100, 112140[8] R. Arndt, S. Schmidt, C. Loddenkemper, M. Gru¨nbaum, W. Zidek, M. VanDer Giet, and T. H. Westhoff. Noninvasive evaluation of renal allograftfibrosis by transient elastography–a pilot study. Transplant International,23(9):871–877, 2010. → pages 79, 80, 81, 89, 118[9] K. Asano, A. Ogata, K. Tanaka, Y. Ide, A. Sankoda, C. Kawakita,M. Nishikawa, K. Ohmori, M. Kinomura, N. Shimada, et al. Acousticradiation force impulse elastography of the kidneys is shear wave velocityaffected by tissue fibrosis or renal blood flow? Journal of Ultrasound inMedicine, 33(5):793–801, 2014. → pages 86, 87, 114[10] P. Asbach, D. Klatt, U. Hamhaber, J. Braun, R. Somasundaram, B. Hamm,and I. Sack. Assessment of liver viscoelasticity using multifrequency mrelastography. Magnetic Resonance in Medicine, 60(2):373–379, 2008. →pages 83[11] M. Audette, F. Ferrie, and T. Peters. An algorithmic overview of surfaceregistration techniques for medical imaging. Medical Image Analysis, 4(3):201–217, 2000. → pages 38[12] A. Baghani, A. Brant, S. Salcudean, and R. Rohling. A high-frame-rateultrasound system for the study of tissue motions. Ultrasonics,Ferroelectrics and Frequency Control, IEEE Transactions on, 57(7):1535–1547, 2010. → pages 14, 16, 58[13] A. Baghani, S. Salcudean, M. Honarvar, R. Sahebjavaher, R. Rohling, andR. Sinkus. Traveling wave expansion: A model fitting approach to theinverse problem of elasticity reconstruction. IEEE Medical Imaging, 30(99):1–1, 2011. → pages 57, 58, 64, 68[14] A. Baghani, H. Eskandari, W. Wang, D. Da Costa, M. N. Lathiff,R. Sahebjavaher, S. Salcudean, and R. Rohling. Real-time quantitativeelasticity imaging of deep tissue using free-hand conventional ultrasound.In Medical Image Computing and Computer-AssistedIntervention–MICCAI 2012, pages 617–624. Springer, 2012. → pages 56,68, 94, 103[15] J. R. Banegas, F. Rodrı´guez-Artalejo, J. J. de la Cruz Troca,P. Guallar-Castillo´n, and J. del Rey Calero. Blood pressure in spaindistribution, awareness, control, and benefits of a reduction in averagepressure. Hypertension, 32(6):998–1002, 1998. → pages 95141[16] D. Barbot. Improved staging of liver tumors using laparscopicintraoperative ultrasound. Journal of Surgical Oncology, 64:63–67, 1997.→ pages 20[17] R. G. Barr, G. Ferraioli, M. L. Palmeri, Z. D. Goodman, G. Garcia-Tsao,J. Rubin, B. Garra, R. P. Myers, S. R. Wilson, D. Rubens, and D. Levine.Elastography assessment of liver fibrosis: Society of radiologists inultrasound consensus conference statement. Radiology, 276(3):845–861,2015. → pages 139[18] J. G. Bartlett. Feature-based registration of preoperative CT tointraoperative 3-D ultrasound in laparoscopic partial nephrectomy using apriori CT segmentation. PhD thesis, University of British Columbia, 2011.→ pages 122[19] M. A. L. Bell, H. T. Sen, I. Iordachita, and P. Kazanzides. Force-controlledultrasound robot for consistent tissue pre-loading: Implications for acousticradiation force elasticity imaging. In Biomedical Robotics andBiomechatronics (2014 5th IEEE RAS & EMBS International Conferenceon, pages 259–264. IEEE, 2014. → pages 71[20] S. F. Bensamoun, L. Robert, G. E. Leclerc, L. Debernard, and F. Charleux.Stiffness imaging of the kidney and adjacent abdominal tissues measuredsimultaneously using magnetic resonance elastography. Clinical imaging,35(4):284–287, 2011. → pages 82, 84[21] J. Bercoff, M. Tanter, and M. Fink. Supersonic shear imaging: a newtechnique for soft tissue elasticity mapping. Ultrasonics, Ferroelectricsand Frequency Control, IEEE Transactions on, 51(4):396–409, 2004. →pages 14, 17[22] P. Besl and N. McKay. A method for registration of 3-D shapes. IEEETransactions on Pattern Analysis and Machine Intelligence, pages239–256, 1992. → pages 6, 38[23] J. Beutel, H. L. Kundel, and R. L. Van Metter. Handbook of medicalimaging, volume 1: Physics and psychophysics. 2000. → pages 10[24] S. Billings, N. Deshmukh, H. Kang, R. Taylor, and E. Boctor. System forrobot-assisted real-time laparoscopic ultrasound elastography. In SPIEMedical Imaging, 2012. → pages 54142[25] S. Bota, H. Herkner, I. Sporea, P. Salzl, R. Sirli, A. M. Neghina, andM. Peck-Radosavljevic. Meta-analysis: Arfi elastography versus transientelastography for the evaluation of liver fibrosis. Liver International, 33(8):1138–1147, 2013. → pages 74, 78, 85[26] S. Bota, F. Bob, I. Sporea, R. S¸irli, and A. Popescu. Factors that influencekidney shear wave speed assessed by acoustic radiation force impulseelastography in patients without kidney pathology. Ultrasound in medicine& biology, 41(1):1–6, 2015. → pages 85[27] J.-Y. Bouguet. Camera calibration toolbox for matlab. 2004. URLhttp://www.vision.caltech.edu/bouguetj/calib doc. → pages 34, 35[28] R. Budde and et al. Endoscopic localization and assessment of coronaryarteries by 13MHz epicardial ultrasound. The Annals of Thoracic Surgery,77(5):1586–1592, 2004. → pages 23, 24, 53[29] R. Budde and et al. Robot-assisted 13MHz epicardial ultrasound forendoscopic quality assessment of coronary anastomoses. InteractiveCardiovascular and Thoracic Surgery, 3(4):616, 2004. → pages 23, 24[30] J. Byrn, S. Schluender, C. Divino, J. Conrad, B. Gurland, E. Shlasko, andA. Szold. Three-dimensional imaging improves surgical performance forboth novice and experienced operators using the da Vinci Robot System.The American Journal of Surgery, 193(4):519–522, 2007. → pages 21[31] J. Catheline. A comparison of laparoscopic ultrasound versuscholangiography in the evaluation of the biliary tree during laparoscopiccholecystectomy. European Journal of Ultrasound, 10(1):1–9, 1999. →pages 20[32] S.-H. Cha. Comprehensive survey on distance/similarity measures betweenprobability density functions. Int. J. Math. Models Methods Appl. Sci., 4:300–307, 2007. → pages 38[33] Z. Chen, A. Deguet, R. Taylor, S. DiMaio, G. Fischer, and P. Kazanzides.An open-source hardware and software platform for telesurgical roboticsresearch. In Proceedings of the MICCAI Workshop on Systems andArchitecture for Computer Assisted Interventions, Nagoya, Japan, pages22–26, 2013. → pages 22, 54, 59, 161[34] P. T. Christopher and K. J. Parker. New approaches to the linearpropagation of acoustic fields. The Journal of the Acoustical Society ofAmerica, 90(1):507–521, 1991. → pages 14143[35] G. Danaei, M. M. Finucane, J. K. Lin, G. M. Singh, C. J. Paciorek, M. J.Cowan, F. Farzadfar, G. A. Stevens, S. S. Lim, L. M. Riley, et al. National,regional, and global trends in systolic blood pressure since 1980:systematic analysis of health examination surveys and epidemiologicalstudies with 786 country-years and 5· 4 million participants. The Lancet,377(9765):568–577, 2011. → pages 95[36] L. Deane, H. Lee, G. Box, O. Melamud, D. Yee, J. Abraham, D. Finley,J. Borin, E. McDougall, R. Clayman, et al. Robotic versus standardlaparoscopic partial/wedge nephrectomy: a comparison of intraoperativeand perioperative results from a single institution. Journal of Endourology,22(5):947–952, 2008. → pages 5[37] M. Desai, I. Gill, A. Ramani, M. Spaliviero, L. Rybicki, and J. Kaouk. Theimpact of warm ischaemia on renal function after laparoscopic partialnephrectomy. British Journal of Urology International, 95(3):377–383,2005. → pages 5[38] S. DiMaio and C. Hasser. The da vinci research interface. In MICCAIWorkshop on Systems and Arch. for Computer Assisted Interventions,Midas Journal, Sep, 2008. → pages 54[39] S. Ding and et al. Tracking of vessels in intra-operative microscope videosequences for cortical displacement estimation. Biomedical Engineering,IEEE Transactions on, 58(7):1985–1993, 2011. → pages 6[40] J. R. Doherty and et al. Acoustic radiation force elasticity imaging indiagnostic ultrasound. Ultrasonics, Ferroelectrics, and Frequency Control,IEEE Transactions on, 60(4):685–701, 2013. → pages 14, 17, 78[41] J. R. Doherty, J. J. Dahl, and G. E. Trahey. Harmonic tracking of acousticradiation force-induced displacements. IEEE transactions on ultrasonics,ferroelectrics, and frequency control, 60(11):2347–2358, 2013. → pages116[42] P. Edgcumbe and et al. Pico lantern: A pick-up projector for augmentedreality in laparoscopic surgery. In Medical Image Computing andComputer-Assisted Intervention–MICCAI 2014, pages 432–439. Springer,2014. → pages 26[43] P. Edgcumbe, C. Nguan, and R. Rohling. Calibration and stereo tracking ofa laparoscopic ultrasound transducer for augmented reality in surgery. InAugmented Reality Environments for Medical Imaging and144Computer-Assisted Interventions, pages 258–267. Springer, 2013. → pages34, 40, 47[44] A. Evans, P. Whelehan, K. Thomson, D. McLean, K. Brauer, C. Purdie,L. Jordan, L. Baker, and A. Thompson. Quantitative shear wave ultrasoundelastography: initial experience in solid breast masses. Breast Cancer Res,12(6):R104, 2010. → pages 17[45] S. Ferguson, J. Bryant, R. Ganz, and K. Ito. The acetabular labrum seal: aporoelastic finite element model. Clinical Biomechanics, 15(6):463–468,2000. → pages 88[46] M. Feuerstein and et al. Magneto-optic tracking of a flexible laparoscopicultrasound transducer for laparoscope augmentation. In Medical ImageComputing and Computer-Assisted Intervention, pages 458–466, 2007. →pages 33[47] K. Fuchs. Minimally invasive surgery. Endoscopy, 34(2):154–159, 2002.→ pages 19, 32[48] J. Gao, W. Weitzel, J. M. Rubin, J. Hamilton, J. Lee, D. Dadhania, andR. Min. Renal transplant elasticity ultrasound imaging: correlationbetween normalized strain and renal cortical fibrosis. Ultrasound inmedicine & biology, 39(9):1536–1542, 2013. → pages 75, 77[49] J. Gao, W. He, L.-G. Cheng, X.-Y. Li, X.-R. Zhang, K. Juluru, N. Al Khori,A. Coya, and R. Min. Ultrasound strain elastography in assessment ofcortical mechanical behavior in acute renal vein occlusion: in vivo animalmodel. Clinical imaging, 39(4):613–618, 2015. → pages 77, 86[50] J. Gao, J. M. Rubin, W. Weitzel, J. Lee, D. Dadhania, S. Kapur, andR. Min. Comparison of ultrasound corticomedullary strain with dopplerparameters in assessment of renal allograft interstitial fibrosis/tubularatrophy. Ultrasound in medicine & biology, 41(10):2631–2639, 2015. →pages 75, 77[51] A.-S. Garnier and M. Briet. Arterial stiffness and chronic kidney disease.Pulse, 3(3-4):229–241, 2016. → pages 87, 89, 137[52] A. H. Gee and et al. Sensorless freehand 3d ultrasound in real tissue:speckle decorrelation without fully developed speckle. Medical imageanalysis, 10(2):137–149, 2006. → pages 33145[53] J.-L. Gennisson and et al. Supersonic shear wave elastography of in vivopig kidney: influence of blood pressure, urinary pressure and tissueanisotropy. Ultrasound in medicine & biology, 38(9):1559–1567, 2012. →pages 71, 85, 86[54] N. Grenier, S. Poulain, S. Lepreux, J.-L. Gennisson, B. Dallaudie`re,Y. Lebras, E. Bavu, A. Servais, V. Meas-Yedid, M. Piccoli, et al.Quantitative elastography of renal transplants using supersonic shearimaging: a pilot study. European radiology, 22(10):2138–2146, 2012. →pages 89, 90[55] N. Grenier, J.-L. Gennisson, F. Cornelis, Y. Le Bras, and L. Couzi. Renalultrasound elastography. Diagnostic and interventional imaging, 94(5):545–550, 2013. → pages 74, 78, 79[56] G. Haber and I. Gill. Laparoscopic partial nephrectomy: Contemporarytechnique and outcomes. European urology, 49(4):660–665, 2006. →pages 19[57] S. Hariharan, C. P. Johnson, B. A. Bresnahan, S. E. Taranto, M. J.McIntosh, and D. Stablein. Improved graft survival after renaltransplantation in the united states, 1988 to 1996. New England Journal ofMedicine, 342(9):605–612, 2000. → pages 8, 73[58] W.-Y. He, Y.-J. Jin, W.-P. Wang, C.-L. Li, Z.-B. Ji, and C. Yang. Tissueelasticity quantification by acoustic radiation force impulse for theassessment of renal allograft function. Ultrasound in medicine & biology,40(2):322–329, 2014. → pages 90[59] O. Helenon, S. Merran, F. Paraf, P. Melki, J.-M. Correas, Y. Chretien, andJ.-F. Moreau. Unusual fat-containing tumors of the kidney: a diagnosticdilemma. Radiographics, 17(1):129–144, 1997. → pages 29[60] O. Helenon, J. Correas, C. Balleyguier, M. Ghouadni, and F. Cornud.Ultrasound of renal tumors. European radiology, 11(10):1890–1901, 2001.→ pages 29[61] C. Helfenstein, J.-L. Gennisson, M. Tanter, and P. Beillas. Effects ofpressure on the shear modulus, mass and thickness of the perfused porcinekidney. Journal of biomechanics, 48(1):30–37, 2015. → pages 86, 87, 90,113, 132[62] C. R. Hill, J. C. Bamber, and G. ter Haar. Physical principles of medicalultrasonics, volume 2. Wiley Online Library, 2004. → pages 10146[63] M. Honarvar. Dynamic elastography with finite element-based inversion.PhD thesis, University of British Columbia, 2015. → pages 103, 110[64] M. Honarvar and et al. Direct vibro-elastography fem inversion in cartesianand cylindrical coordinate systems without the local homogeneityassumption. Physics in medicine and biology, 60(9):3847, 2015. → pages94[65] B. K. Horn, H. M. Hilden, and S. Negahdaripour. Closed-form solution ofabsolute orientation using orthonormal matrices. JOSA A, 5(7):1127–1135,1988. → pages 164[66] M. M. Hossain, M. Selzo, R. Hinson, L. Baggessen, R. Detwiler,W. Chong, L. Brubaker, M. Caughey, M. Fisher, S. Whitehead, et al.Evaluation of renal transplant status using viscoelastic response (visr)ultrasound: A pilot clinical study. In Ultrasonics Symposium (IUS), 2016IEEE International, pages 1–4. IEEE, 2016. → pages 88, 100[67] Q. Hu, X.-Y. Wang, H.-G. He, H.-M. Wei, L.-K. Kang, and G.-C. Qin.Acoustic radiation force impulse imaging for non-invasive assessment ofrenal histopathology in chronic kidney disease. PloS one, 9(12):e115051,2014. → pages 89[68] G. Hubens, H. Coveliers, L. Balliu, M. Ruppert, and W. Vaneerdeweg. Aperformance study comparing manual and robotically assisted laparoscopicsurgery using the da Vinci system. Surgical Endoscopy, 17(10):1595–1599,2003. → pages 21, 52[69] S. Hughes and et al. Volume estimation from multiplanar 2D ultrasoundimages using a remote electromagnetic position and orientation sensor.Ultrasound in Medicine & Biology, 22(5):561–572, 1996. ISSN0301-5629. → pages 34[70] A. Hughes-Hallett, E. K. Mayer, H. J. Marcus, T. P. Cundy, P. J. Pratt,A. W. Darzi, and J. A. Vale. Augmented reality partial nephrectomy:examining the current status and future perspectives. Urology, 83(2):266–273, 2014. → pages 137, 161[71] L. Huwart and et al. Liver fibrosis: non-invasive assessment with mrelastography. NMR in Biomedicine, 19(2):173–179, 2006. → pages 71[72] L. Huwart and et al. Comparison of mr elastography and fibroscan for thenon-invasive assessment of liver fibrosis. In Joint Annual MeetingISMRM-ESMRMB, number EPFL-CONF-188693, 2007. → pages 18, 71147[73] V. Jha, G. Garcia-Garcia, K. Iseki, Z. Li, S. Naicker, B. Plattner, R. Saran,A. Y.-M. Wang, and C.-W. Yang. Chronic kidney disease: global dimensionand perspectives. The Lancet, 382(9888):260–272, 2013. → pages 8[74] J. Jomier and S. Aylward. Rigid and deformable vasculature-to-imageregistration: A hierarchical approach. Medical Image Computing andComputer-Assisted Intervention, pages 829–836, 2004. → pages 6[75] M. Jung, P. Morel, L. Buehler, N. Buchs, and M. Hagen. Robotic generalsurgery: current practice, evidence, and perspective. Langenbeck’sArchives of Surgery, pages 1–10, 2015. → pages 161[76] J. Kahn, T. Slowinski, A. Thomas, S. Filimonow, and T. Fischer. Tsiultrasound elastography for the diagnosis of chronic allograft nephropathyin kidney transplanted patients. Journal of Ultrasonography, 13(54):253,2013. → pages 75, 77[77] S. Kaul, R. Laungani, R. Sarle, H. Stricker, J. Peabody, R. Littleton, andM. Menon. da Vinci-assisted robotic partial nephrectomy: Technique andresults at a mean of 15 months of follow-up. European Urology, 51(1):186–192, 2007. → pages 19[78] L. Kavan and et al. Dual quaternions for rigid transformation blending.Trinity College Dublin, Tech. Rep. TCD-CS-2006-46, 2006. → pages 60,166[79] P. Kazanzides, Z. Chen, A. Deguet, G. Fischer, R. Taylor, and S. DiMaio.An open-source research kit for the da vinci r surgical robot. In Proc. IEEEInternational Conf. on Robotics and Automation (ICRA14), 2014. → pages22, 54, 59, 161[80] V. Kindratenko. A survey of electromagnetic position tracker calibrationtechniques. Virtual Reality, 5(3):169–182, 2000. → pages 49[81] H. Knutsson, C.-F. Westin, and G. Granlund. Local multiscale frequencyand bandwidth estimation. In Image Processing, 1994. Proceedings.ICIP-94., IEEE International Conference, volume 1, pages 36–40. IEEE,1994. → pages 61[82] B. Koeppen and B. Stanton. Renal Physiology. Elsevier Mosby,Philadelphia, 5th edition, 2013. → pages 2, 86, 114148[83] K. Konishi and et al. A real-time navigation system for laparoscopicsurgery based on three-dimensional ultrasound using magneto-optic hybridtracking configuration. International Journal of Computer AssistedRadiology and Surgery, 2(1):1–10, 2007. → pages 33[84] E. E. Konofagou, T. P. Harrigan, J. Ophir, and T. A. Krouskop.Poroelastography: imaging the poroelastic properties of tissues. Ultrasoundin medicine & biology, 27(10):1387–1397, 2001. → pages 29, 115[85] M. J. Korsmo, B. Ebrahimi, A. Eirin, J. R. Woollard, J. D. Krier, J. A.Crane, L. Warner, K. Glaser, R. Grimm, R. L. Ehman, et al. Magneticresonance elastography noninvasively detects in-vivo renal medullaryfibrosis secondary to swine renal artery stenosis. Investigative radiology,48(2):61, 2013. → pages 83, 84, 86[86] T. A. Krouskop, T. M. Wheeler, F. Kallel, B. S. Garra, and T. Hall. Elasticmoduli of breast and prostate tissues under compression. Ultrasonicimaging, 20(4):260–274, 1998. → pages 22, 29[87] S. Kruse, M. Dresner, and R. Ehman. Mr elastography of human kidney invivo: a feasibility study. In Proceedings of the 12th Annual Meeting of theInternational Society for Magnetic Resonance in Medicine, Kyoto, Japan,page 2600, 2004. → pages 82[88] C. Kut, C. Schneider, N. Carter-Monroe, L.-M. Su, E. Boctor, andR. Taylor. Accuracy of localization of prostate lesions using manualpalpation and ultrasound elastography. In SPIE Medical Imaging, pages726128–726128. International Society for Optics and Photonics, 2009. →pages 22[89] D. Kwartowitz, S. Herrell, and R. Galloway. Toward image-guided roboticsurgery: Determining intrinsic accuracy of the da Vinci robot.International Journal of Computer Assisted Radiology and Surgery, 1(3):157–165, 2006. → pages 22, 34, 69, 161, 166[90] D. M. Kwartowitz, S. D. Herrell, and R. L. Galloway. Update: Towardimage-guided robotic surgery: determining the intrinsic accuracy of thedavinci-s robot. International Journal of Computer Assisted Radiology andSurgery, 1(5):301–304, 2007. → pages 166[91] J. S. Lam, J. Bergman, A. Breda, and P. G. Schulam. Importance ofsurgical margins in the management of renal cell carcinoma. Natureclinical practice Urology, 5(6):308–317, 2008. → pages 29149[92] T. Lange and et al. Augmenting intraoperative 3D ultrasound withpreoperative models for navigation in liver surgery. Medical ImageComputing and Computer-Assisted Intervention, pages 534–541, 2004. →pages 6[93] T. Lange and et al. 3d ultrasound-ct registration of the liver usingcombined landmark-intensity information. International journal ofcomputer assisted radiology and surgery, 4(1):79–88, 2009. → pages 6[94] A. Lasso and et al. Plus: open-source toolkit for ultrasound-guidedintervention systems. IEEE Trans Biomed Eng, May 2014.doi:10.1109/TBME.2014.2322864. URLhttp://dx.doi.org/10.1109/TBME.2014.2322864. → pages 35[95] A. Lasso, T. Heffter, C. Pinter, T. Ungi, T. K. Chen, A. Boucharin, andG. Fichtinger. Plus: An open-source toolkit for developingultrasound-guided intervention systems. In 4th Image Guided TherapyWorkshop, volume 4, page 103, 2011. → pages 56, 60[96] W. K. Lau, M. L. Blute, A. L. Weaver, V. E. Torres, and H. Zincke.Matched comparison of radical nephrectomy vs nephron-sparing surgery inpatients with unilateral renal cell carcinoma and a normal contralateralkidney. In Mayo Clinic Proceedings, volume 75, pages 1236–1242.Elsevier, 2000. → pages 4[97] C. U. Lee, J. F. Glockner, K. J. Glaser, M. Yin, J. Chen, A. Kawashima,B. Kim, W. K. Kremers, R. L. Ehman, and J. M. Gloor. Mr elastography inrenal transplant patients and correlation with renal allograft biopsy: afeasibility study. Academic radiology, 19(7):834–841, 2012. → pages 84[98] D. Lee and et al. Non-rigid registration between 3d ultrasound and ctimages of the liver based on intensity and gradient information. Physics inmedicine and biology, 56(1):117, 2011. → pages 6[99] G. Lee, M. Lee, I. Green, M. Allaf, and M. Marohn. Surgeons physicaldiscomfort and symptoms during robotic surgery: a comprehensiveergonomic survey study. Surgical Endoscopy, pages 1–10, 2016. → pages19[100] J. Leven and et al. da Vinci canvas: A telerobotic surgical system withintegrated, robot-assisted, laparoscopic ultrasound capability. MedicalImage Computing and Computer-Assisted Intervention, pages 811–818,2005. → pages 23150[101] R. Link, S. Bhayani, and L. Kavoussi. A prospective comparison of roboticand laparoscopic pyeloplasty. Annals of Surgery, 243(4):486, 2006. →pages 22[102] G. Low, N. E. Owen, I. Joubert, A. J. Patterson, M. J. Graves, K. J. Glaser,G. J. Alexander, and D. J. Lomas. Reliability of magnetic resonanceelastography using multislice two-dimensional spin-echo echo-planarimaging (se-epi) and three-dimensional inversion reconstruction forassessing renal stiffness. Journal of Magnetic Resonance Imaging, 42(3):844–850, 2015. → pages 82, 84[103] V. Lukenda, I. Mikolasevic, S. Racki, I. Jelic, D. Stimac, and L. Orlic.Transient elastography: a new noninvasive diagnostic tool for assessmentof chronic allograft nephropathy. International urology and nephrology, 46(7):1435–1440, 2014. → pages 80, 81, 118[104] M. Makuuchi, G. Torzilli, and J. Machi. History of intraoperativeultrasound. Ultrasound in Medicine & Biology, 24(9):1229–1242, 1998. →pages 20[105] A. Manduca, R. Muthupillai, P. Rossman, J. Greenleaf, and R. Ehman.Local wavelength estimation for magnetic resonance elastography. InInternational Conference on Image Processing, volume 3, pages 527–530.IEEE, 1996. → pages 15, 94, 123, 137[106] Y. K. Mariappan, K. J. Glaser, and R. L. Ehman. Magnetic resonanceelastography: a review. Clinical anatomy, 23(5):497–511, 2010. → pages18, 82[107] A. J. Matas, J. M. Smith, M. A. Skeans, B. Thompson, S. K. Gustafson,D. E. Stewart, W. S. Cherikh, J. L. Wainright, G. Boyle, J. J. Snyder, A. K.Israni, and B. L. Kasiske. Optn/srtr 2013 annual data report: Kidney.American Journal of Transplantation, 15(S2):1–34, 2015. ISSN1600-6143. doi:10.1111/ajt.13195. URLhttp://dx.doi.org/10.1111/ajt.13195. → pages 8[108] S. McKinley, A. Garg, S. Sen, R. Kapadia, A. Murali, K. Nichols, S. Lim,S. Patil, P. Abbeel, A. M. Okamura, et al. A single-use haptic palpationprobe for locating subcutaneous blood vessels in robot-assisted minimallyinvasive surgery. In Automation Science and Engineering (CASE), 2015IEEE International Conference on, pages 1151–1158. IEEE, 2015. →pages 53151[109] A. L. McKnight, J. L. Kugel, P. J. Rossman, A. Manduca, L. C. Hartmann,and R. L. Ehman. Mr elastography of breast cancer: preliminary results.American journal of roentgenology, 178(6):1411–1417, 2002. → pages 22[110] M. Mengel, J. Chapman, F. G. Cosio, M. Cavaille´-Coll, H. Haller,P. Halloran, A. Kirk, M. Mihatsch, B. Nankivell, L. Racusen, et al.Protocol biopsies in renal transplantation: insights into patient managementand pathogenesis. American Journal of Transplantation, 7(3):512–517,2007. → pages 9, 74[111] M. S. Menzilcioglu, M. Duymus, S. Citil, G. Gungor, M. Saglam,O. Gungor, S. N. Boysan, A. Sarıca, and S. Avcu. The comparison ofresistivity index and strain index values in the ultrasonographic evaluationof chronic kidney disease. La radiologia medica, pages 1–7, 2016. →pages 75, 77[112] L. Mercier and et al. A review of calibration techniques for freehand 3-Dultrasound systems. Ultrasound in Medicine & Biology, 31(2):143–165,2005. ISSN 0301-5629. → pages 35[113] O. Mohareri, A. Ruszkowski, J. Lobo, J. Ischia, A. Baghani, G. Nir,H. Eskandari, E. Jones, L. Fazli, L. Goldenberg, et al. Multi-parametric 3dquantitative ultrasound vibro-elastography imaging for detecting palpableprostate tumors. In International Conference on Medical Image Computingand Computer-Assisted Intervention, pages 561–568. Springer, 2014. →pages 94, 103[114] O. Mohareri, C. Schneider, and S. Salcudean. Bimanual telerobotic surgerywith asymmetric force feedback: A davinci R© surgical systemimplementation. In Intelligent Robots and Systems (IROS 2014), 2014IEEE/RSJ International Conference on, pages 4272–4277. IEEE, 2014. →pages 53, 161[115] T. Nakao, H. Ushigome, T. Nakamura, S. Harada, K. Koshino, T. Suzuki,T. Ito, S. Nobori, and N. Yoshimura. Evaluation of renal allograft fibrosisby transient elastography (fibro scan). In Transplantation proceedings,volume 47, pages 640–643. Elsevier, 2015. → pages 80, 81, 118[116] B. J. Nankivell, R. J. Borrows, C. L.-S. Fung, P. J. OConnell, R. D. Allen,and J. R. Chapman. Natural history, risk factors, and impact of subclinicalrejection in kidney transplantation. Transplantation, 78(2):242–249, 2004.→ pages 9, 73152[117] K. Nightingale, M. Soo, R. Nightingale, and G. Trahey. Acoustic radiationforce impulse imaging: in vivo demonstration of clinical feasibility.Ultrasound in medicine & biology, 28(2):227–235, 2002. → pages 17[118] K. R. Nightingale, M. L. Palmeri, R. W. Nightingale, and G. E. Trahey. Onthe feasibility of remote palpation using acoustic radiation force. TheJournal of the Acoustical Society of America, 110(1):625–634, 2001. →pages 14, 17[119] U. of Utah Kidney Transplant Program. Kidney transplant, 2016. URLhttp://healthcare.utah.edu/transplant/kidney/. [Online; accessed Dec 12th2016]. → pages xii, 7[120] A. M. Okamura. Methods for haptic feedback in teleoperatedrobot-assisted surgery. Industrial Robot: An International Journal, 31(6):499–508, 2004. → pages 53[121] A. M. Okamura, L. N. Verner, C. Reiley, and M. Mahvash. Haptics forrobot-assisted minimally invasive surgery. In Robotics research, pages361–372. Springer, 2011. → pages 53[122] A. OpenStax College and P. O. CNX. Anatomy and physiology, 2016.URLhttp://cnx.org/contents/14fb4ad7-39a1-4eee-ab6e-3ef2482e3e22@8.26.[Online; accessed Dec 12th 2016]. → pages xii, 3[123] J. Ophir, I. Cespedes, H. Ponnekanti, Y. Yazdi, and X. Li. Elastography: aquantitative method for imaging the elasticity of biological tissues.Ultrasonic imaging, 13(2):111–134, 1991. → pages 54[124] E. M. V. Osorio and et al. Accurate ct/mr vessel-guided nonrigidregistration of largely deformed livers. Medical physics, 39(5):2463–2477,2012. → pages 6[125] F. Ozkan, Y. C. Yavuz, M. F. Inci, B. Altunoluk, N. Ozcan, M. Yuksel,H. Sayarlioglu, and E. Dogan. Interobserver variability of ultrasoundelastography in transplant kidneys: correlations with clinical-dopplerparameters. Ultrasound in medicine & biology, 39(1):4–9, 2013. → pages75, 76, 77[126] M. L. Palmeri, Y. Deng, N. C. Rouze, and K. R. Nightingale. Dependenceof shear wave spectral content on acoustic radiation force excitationduration and spatial beamwidth. In 2014 IEEE International UltrasonicsSymposium, pages 1105–1108. IEEE, 2014. → pages 116153[127] G. Penney and et al. Registration of freehand 3D ultrasound and magneticresonance liver images. Medical Image Analysis, 8(1):81–91, 2004. ISSN1361-8415. → pages 6[128] P. R. Perrin˜ez, F. E. Kennedy, E. E. Van Houten, J. B. Weaver, and K. D.Paulsen. Modeling of soft poroelastic tissue in time-harmonic mrelastography. IEEE transactions on biomedical engineering, 56(3):598–608, 2009. → pages 88[129] T. Polascik and et al. Intraoperative sonography for the evaluation andmanagement of renal tumors: Experience with 100 patients. The Journal ofUrology, 154(5):1676–1680, 1995. → pages 20[130] F. Porpiglia, A. Volpe, M. Billia, and R. Scarpa. Laparoscopic versus openpartial nephrectomy: analysis of the current literature. European Urology,53(4):732–743, 2008. → pages 4, 19[131] F. K. Port, R. A. Wolfe, E. A. Mauger, D. P. Berling, and K. Jiang.Comparison of survival probabilities for dialysis patients vs cadaveric renaltransplant recipients. Jama, 270(11):1339–1343, 1993. → pages 73, 117[132] R. Prager and et al. Rapid calibration for 3-D freehand ultrasound.Ultrasound in Medicine & Biology, 24(6):855–869, 1998. → pages 35, 56[133] P. Pratt and et al. Intraoperative ultrasound guidance for transanalendoscopic microsurgery. In Medical Image Computing andComputer-Assisted Intervention–MICCAI 2012, pages 463–470. Springer,2012. → pages 33, 35[134] D. N. Reddan, G. V. Raj, and T. J. Polascik. Management of small renaltumors: an overview. The American journal of medicine, 110(7):558–562,2001. → pages 6[135] D. M. T. Reece. Kidney physiology basics, 2016. URLhttp://www.medicalsciencenavigator.com/page/4/. [Online; accessed Dec12th 2016]. → pages xii, 4[136] I. Reinertsen and et al. Validation of vessel-based registration for correctionof brain shift. Medical Image Analysis, 11(4):374–388, 2007. → pages 6[137] T. Robinson and G. Stiegmann. Minimally invasive surgery. Endoscopy, 36(1):48–51, 2004. → pages 19, 32154[138] R. Rohling and et al. Comparison of relative accuracy between amechanical and an optical position tracker for image-guided neurosurgery.Computer Aided Surgery, 1(1):30–34, 1995. → pages 37, 162[139] O. Rouvie`re, R. Souchon, G. Pagnoux, J.-M. Me´nager, and J.-Y. Chapelon.Magnetic resonance elastography of the kidneys: feasibility andreproducibility in young healthy adults. Journal of Magnetic ResonanceImaging, 34(4):880–886, 2011. → pages 82, 84[140] A. Ruszkowski, O. Mohareri, S. Lichtenstein, R. Cook, and S. Salcudean.On the feasibility of heart motion compensation on the da vinci surgicalrobot for coronary artery bypass surgery: Implementation and user studies.In Proc. IEEE International Conf. on Robotics and Automation, pages4432–4439, 2015. → pages 161[141] J. Schiff, M. Palese, E. Vaughan Jr, R. Sosa, D. Coll, and J. Del Pizzo.Laparoscopic vs open partial nephrectomy in consecutive patients: Thecornell experience. British Journal of Urology International, 96(6):811–814, 2005. → pages 19[142] C. Schneider and et al. Robot-assisted laparoscopic ultrasound.Information Processing in Computer-Assisted Interventions, pages 67–80,2010. → pages 23, 33[143] C. Schneider and et al. Intra-operative ”pick-up ultrasound for robotassisted surgery with vessel extraction and registration: A feasibility study.Information Processing in Computer-Assisted Interventions, pages122–132, 2011. → pages xiv, 6, 20, 24, 33, 53, 54, 58, 59[144] C. Schneider, A. Baghani, R. Rohling, and S. Salcudean. Remoteultrasound palpation for robotic interventions using absolute elastography.In Medical Image Computing and Computer-AssistedIntervention–MICCAI 2012, pages 42–49. Springer, 2012. → pages 52, 98[145] C. Schneider, M. Honarvar, R. Rohling, S. Salcudean, and C. Nguan.Apparatus for imaging and model fitting of ex-vivo porcine kidney. InFifteenth International Tissue Elasticity Conference. ITEC, 2016. → pages93[146] C. Schneider, J. Lobo, M. Honarvar, S. Bidur, R. Rohling, andS. Salcudean. Blood pressure dependent elasticity measurements ofporcine kidney ex-vivo. In 2016 IEEE International UltrasonicsSymposium. IEEE, 2016. → pages 93155[147] C. Schneider, C. Nguan, R. Rohling, and S. Salcudean. Tracked pick-upultrasound for robot-assisted minimally invasive surgery. IEEETransactions on Biomedical Engineering, 63(2):260–268, 2016. → pages32, 58[148] C. M. Schneider, S. E. Salcudean, R. N. Rohling, and C. Y. C. Nguan.Ultrasound probe for laparoscopy, 2012. US Patent App. 13/525,183. →pages 26[149] A. Schwarz, W. Gwinner, M. Hiss, J. Radermacher, M. Mengel, andH. Haller. Safety and adequacy of renal transplant protocol biopsies.American journal of transplantation, 5(8):1992–1996, 2005. → pages 8,73, 117[150] A. Schwarz, M. Mengel, W. Gwinner, J. Radermacher, M. Hiss, H. Kreipe,and H. Haller. Risk factors for chronic allograft nephropathy after renaltransplantation: a protocol biopsy study. Kidney international, 67(1):341–348, 2005. → pages 8, 9, 73[151] M. R. Selzo and C. M. Gallippi. Viscoelastic response (visr) imaging forassessment of viscoelasticity in voigt materials. IEEE transactions onultrasonics, ferroelectrics, and frequency control, 60(12):2488–2500, 2013.→ pages 88, 100[152] M. Shah. Solving the robot-world/hand-eye calibration problem using thekronecker product. Journal of Mechanisms and Robotics, 5(3):031007(1–7), 2013. → pages 36[153] T. Sielhorst, M. Feuerstein, and N. Navab. Advanced medical displays: Aliterature review of augmented reality. Journal of Display Technology, 4(4):451–467, 2008. → pages 137[154] R. Sinkus and et al. Mr elastography of breast lesions: understanding thesolid/liquid duality can improve the specificity of contrast-enhanced mrmammography. Magnetic Resonance in Medicine, 58(6):1135–1144, 2007.→ pages 18, 29, 71[155] K. Solez et al. International standardization of criteria for the histologicdiagnosis of renal allograft rejection: the banff working classification ofkidney transplant pathology. Kidney international, 44(2):411–422, 1993.→ pages 9, 79, 118156[156] C. Sommerer, M. Scharf, C. Seitz, G. Millonig, H. K. Seitz, M. Zeier, andS. Mueller. Assessment of renal allograft fibrosis by transient elastography.Transplant International, 26(5):545–551, 2013. → pages 79, 81, 90[157] M. Spaliviero and I. Gill. Laparoscopic partial nephrectomy. BritishJournal of Urology International, 99(5b):1313–1328, 2007. → pages 4, 32[158] P. Stolka, M. Keil, G. Sakas, E. McVeigh, M. Allaf, R. Taylor, andE. Boctor. A 3D-elastography-guided system for laparoscopic partialnephrectomies. In SPIE Medical Imaging, 2010. → pages 54[159] D. Stoyanov, G. Mylonas, F. Deligianni, A. Darzi, and G. Yang. Soft-tissuemotion tracking and structure estimation for robotic assisted MISprocedures. Medical Image Computing and Computer-AssistedIntervention, pages 139–146, 2005. → pages 22[160] K.-J. Streitberger, J. Guo, H. Tzscha¨tzsch, S. Hirsch, T. Fischer, J. Braun,and I. Sack. High-resolution mechanical imaging of the kidney. Journal ofbiomechanics, 47(3):639–644, 2014. → pages 84[161] S. E. Sutherland, M. I. Resnick, G. T. Maclennan, and H. B. Goldman.Does the size of the surgical margin in partial nephrectomy for renal cellcancer really matter? The Journal of urology, 167(1):61–64, 2002. →pages 6[162] T. Syversveen, K. Brabrand, K. Midtvedt, E. H. Strøm, A. Hartmann, J. A.Jakobsen, and A. E. Berstad. Assessment of renal allograft fibrosis byacoustic radiation force impulse quantification–a pilot study. TransplantInternational, 24(1):100–105, 2011. → pages 89[163] T. Syversveen, K. Midtvedt, A. E. Berstad, K. Brabrand, E. H. Strøm, andA. Abildgaard. Tissue elasticity estimated by acoustic radiation forceimpulse quantification depends on the applied transducer force: anexperimental study in kidney transplant patients. European radiology, 22(10):2130–2137, 2012. → pages 88, 94, 113[164] T. L. Szabo. Diagnostic ultrasound imaging: inside out. Academic Press,2004. → pages 13[165] A. Talasaz and R. V. Patel. Integration of force reflection with tactilesensing for minimally invasive robotics-assisted tumor localization. IEEETransactions on Haptics, 6(2):217–228, 2013. → pages 53157[166] A. Talasaz, A. L. Trejos, S. Perreault, H. Bassan, and R. V. Patel. Adual-arm 7-degrees-of-freedom haptics-enabled teleoperation test bed forminimally invasive surgery. Journal of Medical Devices, 8(4):041004,2014. → pages 53[167] L. Taylor, B. Porter, D. Rubens, and K. Parker. Three-dimensionalsonoelastography: principles and practices. Physics in medicine andbiology, 45(6):1477, 2000. → pages 18[168] R. Ternifi, J.-L. Gennisson, M. Tanter, and P. Beillas. Effects of storagetemperature on the mechanical properties of porcine kidney estimatedusing shear wave elastography. Journal of the mechanical behavior ofbiomedical materials, 28:86–93, 2013. → pages 90, 95[169] R. H. Thompson, B. R. Lane, C. M. Lohse, B. C. Leibovich, A. Fergany,I. Frank, I. S. Gill, M. L. Blute, and S. C. Campbell. Every minute countswhen the renal hilum is clamped during partial nephrectomy. Europeanurology, 58(3):340–345, 2010. → pages 5[170] S. Thompson, M. James, N. Wiebe, B. Hemmelgarn, B. Manns,S. Klarenbach, M. Tonelli, A. K. D. Network, et al. Cause of death inpatients with reduced kidney function. Journal of the American Society ofNephrology, pages ASN–2014070714, 2015. → pages 73[171] G. Treece, R. Prager, and A. Gee. Regularised marching tetrahedra:improved iso-surface extraction. Computers & Graphics, 23(4):583–598,1999. → pages 38[172] M. Urban, S. Chen, and M. Fatemi. A review of shearwave dispersionultrasound vibrometry (sduv) and its applications. Current medicalimaging reviews, 8(1):27–36, 2012. → pages 88, 100[173] M. Van Veelen, E. Nederlof, R. Goossens, C. Schot, and J. Jakimowicz.Ergonomic problems encountered by the medical team related to productsused for minimally invasive surgery. Surgical Endoscopy, 17(7):1077–1081, 2003. → pages 19[174] C. Va˚penstad and et al. Laparoscopic ultrasound: A survey of its currentand future use, requirements, and integration with navigation technology.Surgical Endoscopy, pages 1–10, 2010. → pages 20[175] D. Wang, F. Bello, and A. Darzi. Augmented reality provision inrobotically assisted minimally invasive surgery. In International CongressSeries, volume 1268, pages 527–532, 2004. → pages 22158[176] L. Wang, P. Xia, K. Lv, J. Han, Q. Dai, X.-m. Li, L.-m. Chen, and Y.-x.Jiang. Assessment of renal tissue elasticity by acoustic radiation forceimpulse quantification with histopathological correlation: preliminaryexperience in chronic kidney disease. European radiology, 24(7):1694–1699, 2014. → pages 89[177] L. Warner, M. Yin, K. J. Glaser, J. A. Woollard, C. A. Carrascal, M. J.Korsmo, J. A. Crane, R. L. Ehman, and L. O. Lerman. Noninvasive in vivoassessment of renal tissue elasticity during graded renal ischemia using mrelastography. Investigative radiology, 46(8):509, 2011. → pages 83, 84, 86[178] W. F. Weitzel, K. Kim, J. M. Rubin, R. C. Wiggins, H. Xie, X. Chen, S. Y.Emelianov, and M. O’DONNELL. Feasibility of applying ultrasound strainimaging to detect renal transplant chronic allograft nephropathy. Kidneyinternational, 65(2):733–736, 2004. → pages 77[179] R. Yakoubi, R. Autorino, H. Laydner, J. Guillotreau, M. A. White,S. Hillyer, G. Spana, R. Khanna, W. Isaac, G.-P. Haber, et al. Initiallaboratory experience with a novel ultrasound probe for standard andsingle-port robotic kidney surgery: increasing console surgeon autonomyand minimizing instrument clashing. The International Journal of MedicalRobotics and Computer Assisted Surgery, 8(2):201–205, 2012. → pages 53[180] T. Yamamoto, B. Vagvolgyi, K. Balaji, L. L. Whitcomb, and A. M.Okamura. Tissue property estimation and graphical display for teleoperatedrobot-assisted surgery. In Robotics and Automation, 2009. ICRA’09. IEEEInternational Conference on, pages 4239–4245. IEEE, 2009. → pages 53[181] M. Yin and et al. Assessment of hepatic fibrosis with magnetic resonanceelastography. Clinical Gastroenterology and Hepatology, 5(10):1207–1213, 2007. → pages 71[182] M. Yin, L. Warner, L. Lerman, A. Manduca, and R. Ehman. Assessment ofkidney stiffness in a swine model of renal arterial stenosis with 3-d mrelastography. Proceedings of the International Society for MagneticResonance in Medicine. April 18Y24, 2009. → pages 84, 86, 87[183] M. Yin, K. Glaser, A. Kolipaka, L. Warner, J. Talwalkar, A. Manduca, andR. Ehman. Influence of perfusion on tissue stiffness assessed with mrelastography. In Proceedings of the International Society for MagneticResonance in Medicine, volume 18, 2010. → pages 84, 86, 87159[184] P. Yohannes, P. Rotariu, P. Pinto, A. Smith, and B. Lee. Comparison ofrobotic versus laparoscopic skills: Is there a difference in the learningcurve? Urology, 60(1):39–45, 2002. → pages 21[185] M. Zaffanello and C. Bruno. Clinical perspective on renal elasticityquantification by acoustic radiation force. World Journal of ClinicalUrology, 4(3):100–103, 2015. → pages 78[186] M. Zaffanello, G. Piacentini, C. Bruno, M. Brugnara, and V. Fanos. Renalelasticity quantification by acoustic radiation force impulse applied to theevaluation of kidney diseases. Journal of Investigative Medicine, 63(4):605–612, 2015. → pages 78[187] R. Zahiri-Azar and S. Salcudean. Motion estimation in ultrasound imagesusing time domain cross correlation with prior estimates. BiomedicalEngineering, IEEE Transactions on, 53(10):1990–2000, 2006. → pages18, 60160Appendix ASupporting MaterialsA.1 dVRK and its AccuracyThe DVRK [75] controllers are an “open source mechatronics platform consistingof hardware, firmware and software components [33], [79] and are being installedat multiple centers for telerobotics in medicine research. With the expansion of theuse of these controllers and software, is imperative to understand the accuracy andprecision of the system as compared to the da Vinci classic and S (and Si) systemsthat are used clinically. Previously, Kwartowitz et. al has investigated the accuracyand precision of the clinical API associated with the da Vinci classic and S robots[89].These controllers are being used to implement new haptic interfaces [114] andeven an implementation of a control system to compensate for the motion of a beat-ing heart [140]. Future research could also include augmented reality [70]. Eachof these applications requires that the location of the tools be known accurately inthe framework of the robot.In our lab, the teleoperation system is based on the components from the opensource CISST/SAW libraries [33], a common implementation for the controllers.The controllers are implemented using a da Vinci Classic robot.In these experiments, the Certus OptoTrak was used to validate the dVRK posi-tion. The OptoTrak Certus Motion Capture System (Northern Digital Inc., Ontario,Canada), a tracking system that uses active IR LEDs has an accuracy of 0.1 mm161and resolution of 0.01 mm over its working volume [138]. An OptoTrak styluswas used to localize the points in the OptoTrak frame. The position of the robottool control point is reported by the dVRK with respect to the base of the activatedjoints of the arm.Legos were used to create a repeatable and stable platform for the target local-ization. A custom Lego was designed and 3D printed for these experiments. Thecustom Lego was designed to have a semi circle hole for the OptoTrak stylus tip,3mm in diameter, and a divot, 1mm × 1.5mm to fit the tip of the Black DiamondMirco ForcepsTM, the da Vinci tool used in this experiment. The tool was cho-sen for its very small tip which can be the most accurately localized. Because thedVRK reports the control point of the tool, discussed later, other, similar tools areexpected to have the same accuracy results, given that the tip offset is known orcalibrated.The difference in position between the dVRK divot point and the OptoTrakpivot point needs to be corrected using the known geometry of the custom builtLego. A coordinate frame was defined for the Lego blocks using the axis of theLego locations (Lego). The OptoTrak points (Optopoints) were transformed into theLego frame and the difference in positions between the dVRK and the OptoTrak(Legoo f f set) were then applied to the OptoTrak positions. The transformed Opto-Trak points were then transformed back into their original coordinate frame. Thepositions of the dVRK and OptoTrak are then correlated.O = ((Lego∗Optopoints)∗Legoo f f set)∗ (Lego−1)Another offset must be applied in order to locate the tip of the tool exactly. Thetool location, or control point, as defined by the dVRK is not located at the tip ofthe tool, but near the base of the jaws [Figure A.2]. The offset between the controlpoint and the tool tip is known from the tool geometry. Each position reported bythe dVRK (Reported) is multiplied by this offset to find the 3D location of theactual tool tip.dV RK points = Reported ∗Toolo f f setThe custom Lego was placed in 36 locations in a rectangle with dimensions162Figure A.1: An image of the experimental setup. The OptoTrak stylus wasplaced in a divot in the center of the custom Lego, while the MicroForceps were placed in a divot to the left.163Figure A.2: A sketch of the tip of a da Vinci Instrument and the transformsthat describe the transforms that describe the orientation of the tool.The control point is located at O6. This image is taken from the DVRKuser guide.16cm × 6.3cm × 2cm which mimics the general workspace of the robot. At eachlocation, the tool tip was placed in the divot five times. The OptoTrak stylus wasplaced into the semi-circular divot three times.The mean location for both the DVRK locations and OptoTrak were calculatedand used in future registrations. The Target Localization Error (TLE) was calculatedas the Cartesian distance between the mean location for each Lego location andeach of the points for that location. Outliers were defined to be points that lay morethan 2.5 mm from the mean location of each target. These points were discardedand mean location of each target was recalculated with the remaining points. Let itbe noted that only 2 points were removed out of the 180 collected points from thecalculations.The OptoTrak points and the dVRK points were then registered. The OptoTrakpoints are used as ground truth in this registration. 50 registrations were completedusing Horns method [65]. For each registration 30 of the 36 points were used asinput, and then the remaining 6 points were used as testing data. The 36 Lego164Figure A.3: Point distributions of the locations defined by the DVRK.locations are randomly assigned to the training or testing data for each round ofregistration.dVRK Accuracy Results and Discussion:The resulting TLE for each location in the dVRK and the OptoTrak is shown inTable A.1 and A.2. An example of the points from the dVRK is shown in FigureA.3. The variance in the locations of the OptoTrak points was very small, with amean TLE of 0.085 mm.The mean and standard deviation of the training and testing data for these 50registrations is shown in Table A.3. The error is defined as the mean Cartesiandistance between the OptoTrak point and the resulting dVRK location. These areaveraged over the 30 training points or 6 testing points and then averaged over the50 registration rounds. Figure 2.4 shows the average resulting registration. The165Table A.1: TLE for Lego locations 1-10Mean and Standard Deviation of TLETarget Number dVRK Targets OptoTrak Targets1 0.71 ± 0.30 0.127 ± 0.0352 1.21 ± 0.31 0.123 ± 0.0513 1.29 ± 0.73 0.113 ± 0.0644 1.12 ± 0.73 0.309 ± 0.1385 1.10 ± 0.58 0.146 ± 0.0626 0.54 ± 0.17 0.055 ± 0.0137 1.09 ± 0.73 0.031 ± 0.0238 1.48 ± 0.46 0.088 ± 0.0389 1.02 ± 0.23 0.055 ± 0.02210 1.01 ± 0.51 0.128 ± 0.051transform from each of the 50 registration rounds was saved and then averaged us-ing dual quaternions [78]. The red stars in Figure A.4 represent the Lego locationsdefined by the OptoTrak and the black diamonds represent the DVRK locations afterbeing multiplied by this average transform.We have completed a test of the dVRKs precision and accuracy when comparedto the OptoTrak motion capture system. Using Legos as repeatable and stablelocalization platform, we showed that the errors in the dVRK are on par with thosedescribed by Kwartowitz et al. [89] and with other tracking systems used withina surgical environment. Kwartowitz reports that the Standard robot has a meanTLE of 1.31 mm and 1.35 mm using two calculation methods. They also report anexpected localization error of 1.02 mm, assuming infinite number of localizationtrials. In an update, Kwartowitz updates the results for the S robot [90]. The Srobot has an updated robot structure and they report an expected localization errorof 1.05 mm, which is not significantly different than what was found for the originalsystem.Using 36 points and 5 repetitions, we found a mean TLE of 1.16 mm. Whenregistering the DVRK data to the OptoTrak mean TRE of 1.24 mm for the testingdata was calculated. This is very similar to what has been found in previous ex-166Table A.2: TLE for Lego locations 11-3611 1.32 ± 0.66 0.056 ± 0.02112 1.00 ± 0.65 0.019 ± 0.01113 1.61 ± 0.80 0.093 ± 0.04814 0.56 ± 0.22 0.080 ± 0.03415 0.90 ± 0.33 0.108 ± 0.03116 1.25 ± 0.27 0.118 ± 0.04317 0.94 ± 0.39 0.075 ± 0.02418 1.46 ± 0.69 0.082 ± 0.02919 1.61 ± 0.41 0.070 ± 0.02320 1.59 ± 0.20 0.131 ± 0.04521 0.67 ± 0.28 0.106 ± 0.05722 1.61 ± 0.78 0.104 ± 0.04023 1.26 ± 0.68 0.098 ± 0.05024 1.54 ± 0.69 0.109 ± 0.05825 1.27 ± 0.35 0.046 ± 0.02226 1.38 ± 0.55 0.029 ± 0.01927 1.39 ± 0.41 0.077 ± 0.04828 0.85 ± 0.29 0.083 ± 0.01929 1.43 ± 0.62 0.042 ± 0.00530 0.87 ± 0.55 0.038 ± 0.00231 0.96 ± 0.32 0.065 ± 0.02732 1.36 ± 0.52 0.055 ± 0.01733 1.25 ± 0.17 0.054 ± 0.03234 0.90 ± 0.20 0.063 ± 0.02835 1.09 ± 0.40 0.073 ± 0.03336 1.17 ± 0.41 0.019 ± 0.007Mean Mean1.16 ± 0.46 0.085 ± 0.035Table A.3: The mean and standard deviation of the training and testing datafor these 50 registrationsSize ofTrainingdataResults ofTrainingDataSize of TestDataResults ofTest Data30 1.18 ± 0.51 6 1.24 ± 0.52167Figure A.4: Resulting registration, in the OptoTrak coordinate frame. Thered stars represent the Lego locations defined by the OptoTrak and theblack diamonds represent the DVRK locations after being multiplied bythis average transform.periments. The additional points collected in this experiment could have led to thesmaller TLE.The control point that is reported by the DVRK, and located near the pitch andyaw joints of the tool. The offset between the control point and the tool tip, whichis most often the location that desired, can be found from the geometry of any toolwith that general joint kinematics.Because of the distance between the control point and the tool tip there is some168lever arm effect. This will increase the error of the calculated tool tip location.Most of the da Vinci tools that use this style kinematics and thus the estimatederrors found here can be applied to other tools such as the Large Needle Driver, acommonly used tool for both research and clinical purposes.A.2 Individual Model Fitting ResultsFigure A.5: Individual model fitting results for each of the ex-vivo kidneysused in the study.169Figure A.6: Individual model fitting results for each of the ex-vivo kidneysused in the study when viscous components are set to zero.170Figure A.7: Individual model fitting results for each 10 perturbation trials.171Figure A.8: Individual model fitting results for each patient in the study (First8).172Figure A.9: Individual model fitting results for each patient in the study (Sec-ond 8).173


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items