UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Towards liver shear wave vibro-elastography : method repeatability and image registration technique Hemily, Julie 2017

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

Item Metadata


24-ubc_2017_november_hemily_julie.pdf [ 14.75MB ]
JSON: 24-1.0355830.json
JSON-LD: 24-1.0355830-ld.json
RDF/XML (Pretty): 24-1.0355830-rdf.xml
RDF/JSON: 24-1.0355830-rdf.json
Turtle: 24-1.0355830-turtle.txt
N-Triples: 24-1.0355830-rdf-ntriples.txt
Original Record: 24-1.0355830-source.json
Full Text

Full Text

Towards Liver Shear Wave Vibro-elastography: MethodRepeatability and Image Registration TechniquesbyJulie HemilyB.Eng, McMaster University, 2012A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFMaster of Applied ScienceinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Biomedical Engineering)The University of British Columbia(Vancouver)August 2017c© Julie Hemily, 2017AbstractLiver fibrosis is a largely prevalent concern in Canada and world-wide, due to high ratesof Hepatitis, fatty liver disease, alcoholism, as well as several other possible causes. It iscurrently diagnosed and staged by performing biopsies or by tissue elasticity measurementsreferred to as elastography. Elastography methods are a relatively new means of measuringthe mechanical properties of soft tissue non-invasively by measuring and processing thepropagation of shear waves through the body.The Robotics and Control Laboratory at the University of British Columbia has devel-oped an elastography technique, Vibro-elastography, that can quantitatively measure softtissue stiffness in real time. It has previously been applied to prostate and breast patholo-gies. It is now being developed and optimized for liver applications.To validate Vibro-elastography as a new diagnostic tool, a comparison study should beperformed on a clinical population. This work sets out to lay out the prerequisites needed toimplement a full clinical study. It starts out with a repeatability study using a tissue phantomto ensure repeatable results and compare our results to manufacturer stiffness values. In thiswork, we compare the precision of several different implementations of Vibro-elastographyincluding the placement of the excitation source, data acquisition techniques and singleversus multi-frequency excitation. Most of the implementations resulted in good, repeati-ble results, regardless of exciter placement. The quality of wave propagation deterioratedwith depth as expected, but elasticity results remained repeatable even at deeper regions ofinterest. The parameters are selected and designed for the use on the liver.Finally, a registration pipeline and initial case trial has been presented as a suggestedmeans of comparing the elastography data obtained using Vibro-elastography and any elas-tography measures that can be obtained from a magnetic resonance system. Using manualfiducial vessel markers and applying an Iterative Closest Point registration process resultsin a quick alignment of the ultrasound and MRI volumes with registration error less than 20mm.iiLay SummaryWhen tissue becomes diseased, its stiffness and viscosity will change, so identifying theproperty changes of the soft tissue is useful for diagnosis. Elastography is a technologythat uses mechanical vibrations to determine tissue stiffness. A new type of ultrasoundelastography, Vibro-elastography (VE), developed in the robotics and control laboratoryat the University of British Columbia, provides useful real-time information about tissuestiffness. This thesis presents a preliminary work necessary to carry out a clinical studyapplying VE to the liver to diagnose liver disease. It compares repeatability and accuracyof different implementations of the technology using a manufactured sample. We thenpresent a technique that can compare the ultrasound results with MRI results for a futurestudy where MRI Elastography will be available.iiiPrefaceThe author designed the repeatability study presented in Chapter 3 and collected the data.The vibro-elastography technology was developed by Salcudean et al. and was built inhouse by members of the Robotics and Control Laboratory at the University of BritishColumbia. The author carried out the acquisition and processing as well as the statisticalanalysis. The author was the lead on this with the help of Julio Lobo, while Dr. Salcudeanwas the primary supervisor. Julio Lobo acquired and processed the 4DL14 data presentedin Chapter 2, section 1.5.1-4DL14 Results.The data collected and used for registration in Chapter 4 was acquired at the 3D MRICentre following a protocol entitled ”Liver MR Elastography”, which was for a Elastog-raphy Comparison Study on Healthy Volunteers, which was approved by the Clinical Re-search Ethics Board of the University of British Columbia (H14-01964). The study wasdesigned to compare ultrasound and MRI elastography results from healthy volunteer liv-ers. The author collected the data and carried out the processing and registration.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xGlossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xivAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 General Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.1 Project Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.2 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 General Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2.1 Liver Disease in Canada . . . . . . . . . . . . . . . . . . . . . . . Liver Anatomy/function . . . . . . . . . . . . . . . . . . Fibrosis . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2.2 Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Staging . . . . . . . . . . . . . . . . . . . . . . . . . . . Serum Tests . . . . . . . . . . . . . . . . . . . . . . . . 71.3 Elastography Techniques Applied to the Liver . . . . . . . . . . . . . . . . 8v1.3.1 Transient Elastography (FibroScan) . . . . . . . . . . . . . . . . . 81.3.2 Shear Wave Elastography (SWE) . . . . . . . . . . . . . . . . . . 91.3.3 Acoustic Radiation Force Impulse (ARFI) Elastography . . . . . . 101.3.4 Supersonic Elastography . . . . . . . . . . . . . . . . . . . . . . . 111.3.5 Static Elastography - Real Time Elastography . . . . . . . . . . . . 111.3.6 Magnetic Resonance Elastography (MRE) . . . . . . . . . . . . . . 121.4 Other Applications of Liver Elastography . . . . . . . . . . . . . . . . . . 131.5 Current State of the Field . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 Vibro-Elastography Repeatability Study . . . . . . . . . . . . . . . . . . . . . 152.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.1.1 Waves in Tissue . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.1.2 Vibro-Elastography . . . . . . . . . . . . . . . . . . . . . . . . . . Sector based Imaging . . . . . . . . . . . . . . . . . . . Band-pass Imaging . . . . . . . . . . . . . . . . . . . . 232.1.3 Equipment Used/Experiment Setup . . . . . . . . . . . . . . . . . 252.1.4 Software Development Kits (SDKs) . . . . . . . . . . . . . . . . . Porta . . . . . . . . . . . . . . . . . . . . . . . . . . . . Texo . . . . . . . . . . . . . . . . . . . . . . . . . . . . Plane Triggering . . . . . . . . . . . . . . . . . . . . . . 262.1.5 Local Frequency Estimation . . . . . . . . . . . . . . . . . . . . . 272.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.2.1 Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.2.2 Parameter Selection . . . . . . . . . . . . . . . . . . . . . . . . . Frequency and Frame Rate Selection . . . . . . . . . . . Location of Excitation . . . . . . . . . . . . . . . . . . . 312.3 Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.4 Repeatability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.4.1 Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 352.4.1.1 ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . 352.4.1.2 Agreement . . . . . . . . . . . . . . . . . . . . . . . . . 352.4.1.3 Intraclass Correlation . . . . . . . . . . . . . . . . . . . 362.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.5.1 4DL14 Probe Results . . . . . . . . . . . . . . . . . . . . . . . . . 362.5.2 4DC7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38vi2.5.2.1 Single Frequency Results . . . . . . . . . . . . . . . . . 392.5.2.2 Multi-Frequency Results . . . . . . . . . . . . . . . . . 412.5.2.3 Quantifying the Quality of Elasticity Results . . . . . . . 452.5.2.4 Dependence of Exciter Location . . . . . . . . . . . . . 482.5.3 Other 4DC7 Results . . . . . . . . . . . . . . . . . . . . . . . . . 492.5.3.1 Low Amplitude Results . . . . . . . . . . . . . . . . . . 492.5.3.2 Effects of Focal Depth . . . . . . . . . . . . . . . . . . . 502.5.4 Elasticity Processing . . . . . . . . . . . . . . . . . . . . . . . . . 512.5.4.1 LFE . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523 Multi-modal Liver Registration . . . . . . . . . . . . . . . . . . . . . . . . . . 553.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.1.0.1 Prior Multimodal Liver Registration . . . . . . . . . . . 563.1.1 VE/MRI Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 563.1.1.1 Pre-processing of Ultrasound Images . . . . . . . . . . . 573.1.2 Liver Registration and Challenges . . . . . . . . . . . . . . . . . . 583.2 Method: 3D Multi-modal Liver Registration . . . . . . . . . . . . . . . . . 583.2.1 Introduction/Pipeline . . . . . . . . . . . . . . . . . . . . . . . . . 583.2.2 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.2.3 Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.2.3.1 Ultrasound Pre-processing . . . . . . . . . . . . . . . . . 603.2.3.2 MRI Pre-processing . . . . . . . . . . . . . . . . . . . . 613.2.4 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613.2.5 Initial Registration: Manual Alignment . . . . . . . . . . . . . . . 623.2.5.1 Vessel ICP . . . . . . . . . . . . . . . . . . . . . . . . . 633.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633.3.1 MR Inspiration to B-Mode Inspiration Registration . . . . . . . . . 633.3.2 MR Expiration to MR Inspiration Registration . . . . . . . . . . . 653.4 Conclusion and Recommendations . . . . . . . . . . . . . . . . . . . . . . 684 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 714.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73viiA Repeatability Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80A.1 4DC7 Elastograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80A.1.1 Near-Lateral Excitation Images - Single Frequency . . . . . . . . . 81A.1.2 Near-Elevation Excitation Images - Single Frequency . . . . . . . . 82A.1.3 Far Excitation Images - Single Frequency . . . . . . . . . . . . . . 83A.1.4 Near-Lateral Excitation Images - Multi-frequency . . . . . . . . . . 84A.1.5 Near-Elevation Excitation Images - Multi-frequency . . . . . . . . 85A.1.6 Far Excitation Images - Multi-frequency . . . . . . . . . . . . . . . 86A.2 Correlation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87A.2.1 Averaged Correlation Images: Near-Lateral Excitation . . . . . . . 88A.2.2 Averaged Correlation Images: Near-Elevation Excitation . . . . . . 89A.2.3 Averaged Correlation Images: Far Excitation . . . . . . . . . . . . 90viiiList of TablesTable 2.1 Data Acquisition Sheet. NE: Near-Elevation exciter position; NL: Near-Lateral exciter position; FB: Far-Both exciter position; PT: Plane Trig-gering used; NPT: No Plane Triggering . . . . . . . . . . . . . . . . . . 32Table 2.2 4DL14 Difference Table, single frequency . . . . . . . . . . . . . . . . 38Table 2.3 4DL14 variance, single frequency . . . . . . . . . . . . . . . . . . . . . 38Table 2.4 Average Elasticities (kPa) and 95% Confidence Interval limits for sin-gle frequency excitation at 120 Hz. Manufacturer stiffness values weregiven as 14 kPa for the inclusion, and 26.4 kPa for the background. . . . 42Table 2.5 Difference table (kPa) for single frequency excitation at 120 Hz. Manu-facturer stiffness values were given as 14 kPa for the inclusion, and 26.4kPa for the background. . . . . . . . . . . . . . . . . . . . . . . . . . . 42Table 2.6 Average Elasticities (kPa) and 95% Confidence Interval limits for multi-frequency excitation. Manufacturer stiffness values were given as 14kPa for the inclusion, and 26.4 kPa for the background. . . . . . . . . . 44Table 2.7 Difference table (kPa) for multi frequency excitation. Manufacturerstiffness values were given as 14 kPa for the inclusion, and 26.4 kPafor the background. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45Table 2.8 Median Quality Factor Values . . . . . . . . . . . . . . . . . . . . . . . 48ixList of FiguresFigure 1.1 Liver Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4Figure 1.2 Portal Triad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5Figure 1.3 Fibrosis Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . 5Figure 1.4 FibroScan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8Figure 2.1 Vibro-elastography equipment setup . . . . . . . . . . . . . . . . . . . 17Figure 2.2 Diagram of Elastogram Processing from B-mode Acquisition . . . . . 18Figure 2.3 Sector-based Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Figure 2.4 Sector based Scanning with a reduced number of lines . . . . . . . . . 22Figure 2.5 Bandpass sampling: undersampling . . . . . . . . . . . . . . . . . . . 24Figure 2.6 Bandpass Sampling of a band-limited signal . . . . . . . . . . . . . . . 25Figure 2.7 Placement of Exciter relative to the transducer placement . . . . . . . . 31Figure 2.8 Example of a B-mode, phasor image, and resulting elastogram from acentral plane of one of the Porta Near-lateral Excitation. The 5 kPainclusion is to the right of the main 14 kPa inclusion. . . . . . . . . . . 33Figure 2.9 Locations of phantom inclusions and areas of measurement on the B-mode image, phasor image, and elastograms. This particular scan wasusing the far exciter placement and band-pass, not plane triggered se-quence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Figure 2.10 Locations of inclusions and areas of measurement on the b-mode imageand elastograms for the 4DL14 probe. This particular scan was usingthe far exciter placement the porta Software Development Kits (SDK). . 37Figure 2.11 4DL14 repeatability results for the inclusion, using single frequencyacquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Figure 2.12 4DL14 repeatability results for the background, using single frequencyacquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39xFigure 2.13 The excitation was sufficient to create visible waves, shown as phasorimages, through the phantom. These images are from the Near Lateral,plane triggered, band-pass acquisition. . . . . . . . . . . . . . . . . . . 40Figure 2.14 Single frequency Acquisition Results . . . . . . . . . . . . . . . . . . 41Figure 2.15 Multifrequency Acquisition Results for each individual frequency . . . 43Figure 2.16 Averaged multifrequency Results . . . . . . . . . . . . . . . . . . . . 44Figure 2.17 Typical correlation image before scan conversion to transducer geome-try. The correlation falls mostly above 0.95 indicating a good fit alongscan lines. This scan used the Far-both excitation position and a sectorbased acquisition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46Figure 2.18 Correlation values plotted against the scan depth, using a central regionof the scan. This scan used the Far-both excitation position with noplane triggering and a sector based acquisition. . . . . . . . . . . . . . 46Figure 2.19 Correlation plots showing the trends across the scan width at variousdepths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47Figure 2.20 This plot shows the negative correlation between the average correla-tion value within the inclusion measurement areas and the accuracy ofthe elasticity value within that same measurement area, based on thedifference of the calculated elasticity using single frequency excitationand the manufacturer’s value of 14 kPa. . . . . . . . . . . . . . . . . . 49Figure 2.21 This plot shows the negative correlation between the average correla-tion value within the inclusion measurement areas and the accuracy ofthe elasticity value within that same measurement area, based on thedifference of the calculated elasticity using multi-frequency acquisitionand the manufacturer’s value of 14 kPa. . . . . . . . . . . . . . . . . . 50Figure 2.22 Inclusion measurement area elasticity results using low amplitude, sin-gle frequency excitation . . . . . . . . . . . . . . . . . . . . . . . . . 51Figure 2.23 Measurement area placements corresponding to the three different set-tings of focus. The top is located at a depth of 3.25cm, the middle mea-surement area is located at 7.5 cm deep, and the bottom measurementarea is 11.75 cm deep. . . . . . . . . . . . . . . . . . . . . . . . . . . 52Figure 2.24 Results of three measurement areas at different programmed focal values 53Figure 2.25 Inclusion Elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . 54Figure 3.1 Registration process . . . . . . . . . . . . . . . . . . . . . . . . . . . 59xiFigure 3.7 Multimodal registration results . . . . . . . . . . . . . . . . . . . . . . 66Figure A.1 Single-frequency Elastogram - NL, npt, bp . . . . . . . . . . . . . . . 81Figure A.2 Single-frequency Elastogram - NL, npt, sec . . . . . . . . . . . . . . . 81Figure A.3 Single-frequency Elastogram - NL, pt, bp . . . . . . . . . . . . . . . . 81Figure A.4 Single-frequency Elastogram - NL, pt, sec . . . . . . . . . . . . . . . . 81Figure A.5 Single-frequency Elastogram - NL, porta . . . . . . . . . . . . . . . . 81Figure A.6 Single-frequency Elastogram - NE, npt, bp . . . . . . . . . . . . . . . 82Figure A.7 Single-frequency Elastogram - NE, npt, sec . . . . . . . . . . . . . . . 82Figure A.8 Single-frequency Elastogram - NE, pt, bp . . . . . . . . . . . . . . . . 82Figure A.9 Single-frequency Elastogram - NE, pt, sec . . . . . . . . . . . . . . . . 82Figure A.10 Single-frequency Elastogram - NE, porta . . . . . . . . . . . . . . . . 82Figure A.11 Single-frequency Elastogram - FB, npt, bp . . . . . . . . . . . . . . . . 83Figure A.12 Single-frequency Elastogram - FB, npt, sec . . . . . . . . . . . . . . . 83Figure A.13 Single-frequency Elastogram - FB, pt, bp . . . . . . . . . . . . . . . . 83Figure A.14 Single-frequency Elastogram - FB, pt, sec . . . . . . . . . . . . . . . . 83Figure A.15 Single-frequency Elastogram - FB, porta . . . . . . . . . . . . . . . . . 83Figure A.16 Multi-frequency Elastogram - NL, npt, bp . . . . . . . . . . . . . . . . 84Figure A.17 Multi-frequency Elastogram - NL, npt, sec . . . . . . . . . . . . . . . 84Figure A.18 Multi-frequency Elastogram - NL, pt, bp . . . . . . . . . . . . . . . . . 84Figure A.19 Multi-frequency Elastogram - NL, pt, sec . . . . . . . . . . . . . . . . 84Figure A.20 Multi-frequency Elastogram - NL, porta . . . . . . . . . . . . . . . . . 84Figure A.21 Multi-frequency Elastogram - NE, npt, bp . . . . . . . . . . . . . . . . 85Figure A.22 Multi-frequency Elastogram - NE, npt, sec . . . . . . . . . . . . . . . 85Figure A.23 Multi-frequency Elastogram - NE, pt, bp . . . . . . . . . . . . . . . . . 85Figure A.24 Multi-frequency Elastogram - NE, pt, sec . . . . . . . . . . . . . . . . 85Figure A.25 Multi-frequency Elastogram - NE, porta . . . . . . . . . . . . . . . . . 85Figure A.26 Multi-frequency Elastogram - FB, npt, bp . . . . . . . . . . . . . . . . 86Figure A.27 Multi-frequency Elastogram - FB, npt, sec . . . . . . . . . . . . . . . . 86Figure A.28 Multi-frequency Elastogram - FB, pt, bp . . . . . . . . . . . . . . . . . 86Figure A.29 Multi-frequency Elastogram - FB, pt, sec . . . . . . . . . . . . . . . . 86Figure A.30 Multi-frequency Elastogram - FB, porta . . . . . . . . . . . . . . . . . 86Figure A.31 Correlation - NL, npt, bp . . . . . . . . . . . . . . . . . . . . . . . . . 88Figure A.32 Correlation - NL, npt, sec . . . . . . . . . . . . . . . . . . . . . . . . . 88Figure A.33 Correlation - NL, pt, bp . . . . . . . . . . . . . . . . . . . . . . . . . . 88xiiFigure A.34 Correlation - NL, pt, sec . . . . . . . . . . . . . . . . . . . . . . . . . 88Figure A.35 Correlation - NL, porta . . . . . . . . . . . . . . . . . . . . . . . . . . 88Figure A.36 Correlation - NE, npt, bp . . . . . . . . . . . . . . . . . . . . . . . . . 89Figure A.37 Correlation - NE, npt, sec . . . . . . . . . . . . . . . . . . . . . . . . . 89Figure A.38 Correlation - NE, pt, bp . . . . . . . . . . . . . . . . . . . . . . . . . . 89Figure A.39 Correlation - NE, pt, sec . . . . . . . . . . . . . . . . . . . . . . . . . 89Figure A.40 Correlation - NE, porta . . . . . . . . . . . . . . . . . . . . . . . . . . 89Figure A.41 Correlation - FB, npt, bp . . . . . . . . . . . . . . . . . . . . . . . . . 90Figure A.42 Correlation - FB, npt, sec . . . . . . . . . . . . . . . . . . . . . . . . . 90Figure A.43 Correlation - FB, pt, bp . . . . . . . . . . . . . . . . . . . . . . . . . . 90Figure A.44 Correlation - FB, pt, sec . . . . . . . . . . . . . . . . . . . . . . . . . 90Figure A.45 Correlation - FB, porta . . . . . . . . . . . . . . . . . . . . . . . . . . 90xiiiGlossaryANOVA Analysis of VarianceARFI Acoustic Radiation Force Impulse ImagingBPS Band-pass SamplingCT Computed TomographyFB Far BothFEM Finite-element MethodFOV Field of ViewHBV Hepatitis B VirusHCV Hepatitis C VirusHCC Hepatocellular CarcinomaICC Intra-class CorrelationICP Iterative Closest PointIVC Inferior Vena CavaLHV Left Hepatic VeinLFE Local Frequency EstimationLOA Limits of AgreementMHV Middle Hepatic VeinxivMI Mutual InformationMRE Magnetic Resonance ElastographyMRI Magnetic Resonance ImagingMHV Middle Hepatic VeinNAFLD Nonalcoholic Fatty Liver DiseaseNASH Nonalcoholic SteatohepatitisNE Near ElevationNL Near LateralPRF Point Repetition FrequencyQF Quality FactorRF Radio FrequencyRTE Real Time ElastographyRHV Right Hepatic VeinROI Region of InterestSDK Software Development KitsSNR Signal to Noise RatioSWE Shear-wave ElastographyTDE Time Domain Cross Correlation EstimationTDPE Time Domain Cross Correlation with Prior EstimatesUS UltrasoundVE Vibro-elastographyxvAcknowledgmentsI will first thank all the people at UBC who have provided help while I’ve undertaken thisthesis including my supervisor, Dr. Salcudean, who offered me this opportunity to explorethis fascinating field and work in the lab. It has been a huge learning experience. I alsowould like to highlight the help of Dr. Rohling and Julio Lobo who were both critical to mycompleting this project, especially when it was at its most challenging.This is also a thank you to the many friends and family both in Vancouver and afar whocannot know how much their love and support have sustained me over the past few years.xviChapter 1Introduction1.1 General BackgroundThe Canadian Liver Foundation estimates that one in 10 Canadians has some form of liverdisease, and these rates are increasing [58]. Liver related deaths are primarily due to viralHepatitis: Hepatitis B Virus (HBV) Hepatitis C Virus (HCV), liver cancer, Alcoholic LiverDisease, and Nonalcoholic Fatty Liver Disease (NAFLD) due to the prevalence of obesity inCanadians [58]. Many of these diseases present with few symptoms until the damage to theliver is irreversible and the patient has cirrhosis or liver cancer, inevitably leading to death.Therefore, screening tools that diagnose and monitor disease progression are necessary.Biopsy is the current gold standard for monitoring liver fibrosis, but due to the associatedrisks of pain and complications and the high cost of procedure, non-invasive methods arepreferable [45]. Transient elastography is an option but it is unable to investigate theheterogeneity of the damage, and it is inaccurate at distinguishing between less advancedfibrosis stages [45]. For this reason, various elastography methods are being developed andvalidated for the application of soft tissue pathologies.1.1.1 Project ObjectivesThe goal of this thesis is to develop a testing protocol for Vibro-elastography (VE) on theliver, and investigate its repeatability as a measure of diagnostic utility.• investigate different Vibro-elastography techniques and their applicability to liverimaging1• optimize parameters for Vibro-elastography using the three-dimensional curvilinearprobe• Registration of multi-modal liver images, with the final goal of comparing elasticityvalues between VE and MRECurrent methods of imaging liver fibrosis have several drawbacks. The depth at whichthey can perform a measurement is limited, and the acquisition is at a single test point,requiring repeated measurements and a very long acquisition time. Therefore an alternative,which is the subject of this thesis, is needed.1.1.2 Thesis Overview• Chapter 1 Introduction: Introduces the objectives and overview of the thesis. Pro-vides a general background of the importance of tissue characterization of the liverand relevant existing elastography work, and discusses challenges encountered in thisproject.• Chapter 2 Vibro-Elastography Set-up for Deep Tissue: Discusses the optimal proto-col and set-up for VE use on the liver at depths greater than 12 cm.• Chapter 3 Multi-modal Liver Registration to compare ultrasound and MRI liver vol-umes• Chapter 4 Conclusions and Future Work1.2 General Background1.2.1 Liver Disease in CanadaThere are many causes of liver fibrosis, including viral, autoimmune, hereditary, and metabolicdiseases. Liver fibrosis is a reversible inflammatory response of the liver that leads to achange in its tissue architecture. If left untreated, liver fibrosis becomes irreversible, andincreases the likelihood of developing Hepatocellular Carcinoma (HCC). The prevalence ofcirrhosis in the general population is hard to estimate since cirrhosis is asymptomatic duringits early stages. However in 2008, it was estimated that the prevalence of cirrhosis in thegeneral population was up to 1% [57]. Once it develops symptoms, cirrhosis is called de-compensated. Decompensated cirrhosis was ranked 6th in a list of the most common causes2of adult deaths in the developed world [42]. Worldwide, some of the common conditionsresulting in liver fibrosis include HBV, HCV, alcoholism, and NAFLD.Globally, HBV is the leading cause of liver disease in developing countries [42]. Anestimated 240 million people are chronically infected with HBV, leading to more than686,000 people dying every year from consequences of HBV, primarily cirrhosis and hep-atocellular carcinoma, a rapidly fatal form of primary liver cancer [42]. The World HealthOrganization has declared HBV a world health problem [68].While HBV is the leading cause of hepatic cirrhosis in developing countries, in the de-veloped world HCV, alcoholism and Nonalcoholic Steatohepatitis (NASH) lead the way [42].HCV is another common virus known to affect the liver and having severe consequences.It is estimated that more than 130 million people have chronic HCV. Of these, 15-45% willdevelop cirrhosis within 20 years. Approximately 700 000 people die every year from HCVrelated causes [68].Alcoholism is currently considered a leading risk factor for liver fibrosis in the UnitedStates and parts of Europe where cirrhosis related deaths have been linked to the numberof drinks consumed per capita by country. Rising alcoholism rates in the UK has led to anincreased prevalence of liver fibrosis [42].Finally, a newly emerging concern across the world, but particularly pronounced inNorth America, is the increase in NAFLD rates. NASH, a precursor to NAFLD, showsa wide range of liver damage ranging from steatosis, an infiltration of liver cells with fat,to cirrhosis. Increased risk factors for NAFLD and NASH include obesity, diabetes, andhyperlipidemia. Obesity rates have doubled in the past 25 years and do not seem to beslowing down. It is estimated that approximately 34% of the American population hashepatic steatosis and NASH-induced cirrhosis will become the most common indication forliver transplantation in the future [19]. Those suffering from hepatic steatosis are also at anincreased risk for HCC through the fibrosis-cirrhosis-HCC pathway. Liver Anatomy/functionThe liver (figure 1.1), found in the abdominopelvic cavity, is the second largest organ in thehuman body after the skin. The liver is responsible for a wide range of critical functions.These functions include carbohydrate metabolism, lipid metabolism, protein metabolism,processing of drugs and hormones, excretion of bilirubin, synthesis of bile salts, storage,phagocytosis, and activation of vitamin D [65]. The liver consists of two lobes (right andleft) separated by the falciform ligament. It has two main inputs: the portal vein from theintestines and the hepatic artery from the heart. One output from the liver, the hepatic vein,3Figure 1.1: Liver Anatomy. Used with permission from Nature Publishing Group[60]transports blood back to the heart via the vena cava.At a histological scale the liver is made up of hepatic lobules, which are hexagonaldivisions each having a central vein and three to six portal triads at their vertices [41].Each portal triad, seen in figure 1.2, consists of a branch of the hepatic artery, a branch ofthe hepatic vein, and a bile duct, which travel in portal tracts through the connective tissue.The portal tracts also include branches of the vagus nerve and some lymphatic vessels.Radiating out from the central vein are rows of hepatocytes and hepatic sinusoids. Thefunctional units of the liver have been described as oval shaped hepatic acini that includetwo portions of hepatic lobules with the branches of the portal triads running through thecentre [65]. FibrosisFibrosis is a result of oxidative stress on the liver. This stress can be caused by alcohol,bile acids, or viral proteins. It is a slow process resulting in hepatocellular dysfunction,expansion of extracellular matrix, distortion of hepatic architecture, and the formation ofscar tissue [6].The mechanism of fibrosis (Figure 1.3) starts with hepatocellular damage resulting inactivation of the macrophages in the liver, the Kupffer cells. Damaged hepatocytes, throm-bocytes, and endothelial cells are also activated. This results in the activation of hepaticstellate cells, the liver pericytes, and the recruitment of inflammatory cells from other partsof the body to the liver. The hepatic stellate cells are differentiated into myofibroblasts,4Figure 1.2: Portal Triad. With Permission by User:Reytan, Department of Histology,Jagiellonian University Medical College, CC BY-SA 3.0Figure 1.3: Fibrosis Mechanism. Used with permission [34]which proliferate. These myofibroblasts synthesise the extracellular matrix, resulting in anaccumulation of extracellular matrix material [15]. The result of cirrhosis is a replacementof the low density type IV collagen by interstitial collagens (types I and III).If fibrosis results in cirrhosis, the result is an alteration of microvasculature structure,an increase in portal hypertension, and the decline in liver function. Complications arisingfrom cirrhosis include ascites, encephalopathy, synthetic dysfunction, and an increased riskof HCC, a common liver cancer [6].51.2.2 DiagnosisStaging of fibrosis helps with decisions of when and what treatments to apply, assessingdisease progression, and screening for complications. Liver biopsy is useful in evaluatingliver disease because fibrosis is rooted in tissue morphological change that can be evaluatedwith histopathological analysis [41]. . Liver biopsy is used for staging, prognosis, andmanagement of liver disease [13]. Samples of tissue can be used to identify fibrosis stage,inflammatory grade, etiology, and degree of fat and iron deposition [30].Obtaining a tissue sample is usually performed through percutaneous biopsy, but withspecific symptoms such as ascites, slow blood clotting, during liver surgery, a transjugularor laparoscopic biopsy may also be used. Liver biopsies can be guided by ultrasound orby computer tomography. There is no official minimal adequacy for biopsy samples, butmost groups set their own standards involving length and number of portal triads includedin the sample. The portal triads are the initial location of fibrotic activity for viral hepatitis.Standards for biopsy often include a minimum core length of 1.5 cm. Longer samples areprefered for improving accuracy [13], while the number of portal tracts in the sample hasto exceed 6-8.Liver biopsy is still considered the reference standard despite considerable drawbacks[13]. Fibrosis is heterogeneously distributed in the liver and because the biopsy is only1/50,000 of the liver volume, it is possible that up to 30% of cirrhotic cases will be mis-classified [69]. In one study, there was a discordance in staging between biopsies from theleft and right liver lobes in 33% of patients [53]. Biopsy has repeatedly shown significantinterobserver and intraobserver variation [13]. The bleeding risk is 1% and mortality rateis 0.01%, with up to 25% of patients citing pain, and 40% citing discomfort associated withbiopsy [30]. Given these results, there is a need for less invasive and useful fibrosis staging. StagingIn general biopsies are fixed and stained before being scored by a hepatopathologist. Dif-ferent staging systems have been created for different purposes. Scoring fibrosis is oftendivided into two main categories. The first category is of necroinflammation, which is ameasure of severity and ongoing disease activity. It is most likely to show a change dueto therapy. Necroinflammation activity receives a grade [15]. The second score is givena stage. It represents the lesions of fibrosis and remodeling. The fibrosis stage remainsrelatively constant compared to the grade, which fluctuates with progression [15].The Metavir Score, developed by the METAVIR cooperative group in France, and the6Ishak score have been the most validated and applied systems [59]. METAVIR is com-monly used as a reference standard for most elastography research. It incorporates bothpiecemeal and lobular necrosis into their histological activity score, rated from 0 to 3. Theyuse a descriptive value for their fibrosis scoring, given a rating of 0 to 4 [7]. Ishak’s systemis another commonly used scoring system. To assess the degree of activity, qualitative rat-ings are given to portal area inflammation, piecemeal necrosis, spotty necrosis, and bridgingand/or multi-acinar necrosis. Their fibrosis scoring is based on fibrous expansion of portalareas, fibrosis, and bridging with nodules (cirrhosis) [15].The Ishak Modified HAI System is based on descriptions of piecemeal necrosis, conflu-ent necrosis, focal inflammation, and portal inflammation. A summed score is given out ofa possible 18 points. Their staging of fibrosis results in a score of 0 to 6 based on descriptivequalities [15].The concern about these grading systems is that they are only an indication of relativeseverity and not real quantitative values. Many of the systems were designed for viral hep-atitis and are now being applied to a variety of etiologies. Finally, the global score numbersmust be examined in conjunction with the individual components of clinical significance[15].Computerized histomorphometry (CH) scores have been produced to attempt to auto-matically stage fibrosis based on the percentage of the specimen area made up of collagenin three zones: the portal-periportal area, the pericellular space, and around the centrolob-ular venous area [69]. Overall, results show good correlation with fibrosis stages, withsome mixed results for the lower levels [30]. In one study, computerized histomorphome-try showed a better correlation with the hepatic venous pressure gradient, a mark of fibrosisextent, more than the ISHAK fibrosis staging measures [16]. Serum TestsSerum tests have been a popular solution to noninvasively diagnosing fibrosis because theyare repeatable, reproducible, and have an associated low risk. However, they are unable todiscriminate between most intermediate stages of fibrosis and they vary largely based onetiology, of which some etiologies, NAFLD for example, have limited number of biomark-ers available to assess the disease progression [18]. As well, because they test blood thatis circulating through the body, blood markers may reflect fibrogenesis in other organs[32]. Serum markers demonstrate low specificity. Patients with alcoholic liver disease haveshown completely normal levels of liver enzymes despite having clinically significant fibro-sis or cirrhosis on biopsy [15].7Figure 1.4: FibroScan. Used with permission from Elsevier [56]Despite the initial drawbacks, successful results have been shown through classificationschemes that incorporate multiple serum markers. Because of the success and practicalnature of these tests, they have been recommended as a first line assessment method [17]The best serum fibrosis marker panel available commercially is the Fibrotest, which uses sixblood serum tests and has shown a sensitivity of 75% and specificity of 85% [27]. Othersinclude Hepascore (Australia), which uses 4 serum markers, age, and sex to create a model,and Fibrometre (EchoSens, France), which uses 6 blood features, and scores involvingAspartate Aminotransferase and Alanine Aminotransferase ratios [17].1.3 Elastography Techniques Applied to the Liver1.3.1 Transient Elastography (FibroScan)Transient Elastography such as FibroScan (EchoSens, Paris, France) is a commonly usedmethod of obtaining a score for liver stiffness non-invasively. It has become standard prac-tice in some countries. The FibroScan consists of an electro-dynamic transducer with aunidimensional probe at 3.5 MHz, to measure the velocity of shear waves in the liverparenchyma as a response to vibration. It measures at depths of 2.5 to 6.5 cm, resultingin a cylindrical Region of Interest (ROI) with a depth of 4 cm and a diameter of 1 cm,starting at 2.5 cm to avoid surface layers of skin, fat, and muscle (Figure1.4) [56] .During the transient elastography examination, the patient lies in a supine position withtheir right arm in maximal abduction. Ten measurements are obtained on the right lobe ofthe liver through the intercostal space. The device displays the median stiffness in kPa as8well as the interquartile range and the success rate. One of the most significant advantagesof the FibroScan is that it provides immediate results following a painless 5-10 minute scantime and is easy to carry out by non-physicians, making it a convenient method of diagnosis.Unlike biopsy results, it has high intra-operator and inter-operator reproducibility, as wellas having good diagnostic accuracy, correlating well with worsening portal hypertensionmeasures [18].Like all of the diagnostic methods, FibroScan has its own limitations. Measurementsfor certain body types have proven to be difficult using the FibroScan system resulting inunreliable measurements in up to 15% of cases. These body types include high body massindex (BMI), thick chest walls, and the presence of ascites. Although an extra large (XL)probe has been developed to try and resolve some of these issues, it has yet to be provensuccessful in multiple trials. Its initial use showed that the XL probe obtained reliableresults in 59% of the patients presenting unreliable results with the standard probe [22].The challenge associated with aplication of FibroScan to individuals with high BMI couldbe a significant issue for the diagnosis of NAFLD. Another major limitation is the lackof morphological examination. The area of the liver cannot be visualized, so there is noprecise knowledge about the tissue being measured. Finally, the measurement is limited toa line through the right lobe and limits its use in diagnosing various parts of the liver [28].The stiffness cut-off values chosen depend on etiology and population [28]. As well,gender, a BMI greater than 30, and metabolic syndrome affect liver stiffness values, makingit harder to identify a universal cut-off value that can be generalizable to the population[55].1.3.2 Shear Wave Elastography (SWE)Shear Wave Elastography was introduced as a means to map the morphology of the liver.Shear waves are induced in the tissue by the application of Acoustic Radiation Force fromthe piezoelectric elements of the ultrasound scanner. Tissue displacements resulting fromthe induced shear wave are found using a pre-pulse reference image. From these displace-ments, the shear wave speed is calculated, which can be applied to an inversion algorithmto estimate tissue elasticity values [9]. Clinical shear wave elastography has high spatialresolution - around 1 millimeter [8] - and real-time capabilities [52], with some systemsachieving sub-millimeter resolution in both the axial and lateral directions [26].91.3.3 Acoustic Radiation Force Impulse (ARFI) ElastographyAcoustic Radiation Force Impulse Imaging (ARFI) elastography is a shear wave elastogra-phy method that is available on Siemens Acuson S2000 and S3000 (Siemens AG, Erlangen,Germany) and Philips iU22 (Philips Healthcare Solutions, Bothell, WA, USA) ultrasounddevices. Using an acoustic push pulse, shear waves are generated in the tissue that propa-gate perpendicularly to the push force. The shear wave propagation speed near the focusedpush pulse can be measured to produce a localized measurement of liver tissue. ARFI isessentially a point measurement. During the ARFI exam, the patient lies in the supine po-sition with their right arm maximally abducted to increase the intercostal space [27], or thepatient can be placed in the left lateral decubitus position [12]. The patient is instructedto hold their breath during the scan. A study by Karlas et al. investigated the effect ofrespiration phase on variance of shear-wave velocities by applying ARFI during a restingrespiratory position as well as following a deep inspiration. They found that values at dif-ferent points of respiration can vary by as much as 13%, and that variances were lowest inthe resting respiratory position [36]. Regular B-mode ultrasound is used to find the ideallocation to apply the pushing beam and the ROI. An area where the liver is at least 6 cmthick and free of large blood vessels is usually chosen [27]. The ROI is a 1 cm (axial) x0.5 cm (width) rectangular region that can be moved on the image to a maximum depth of 8cm. The excitations can be impulsive, harmonic or steady state [48]. After a few seconds,the velocity of the propagating waves in the ROI is shown on the screen based on time offlight wavespeed estimation [12].ARFI is easy, fast, painless, and provides the operator with quick results. It has beenshown to be less operator dependent than transient elastography [28]. The reproducibilityis good with an inter-operator Intra-class Correlation (ICC) of 0.81, and an intra-operatorICC of 0.9, where the ICC represent a measure of the ratio of variance within groups tothe variance between groups. Compared to transient elastography, it has the advantage ofshowing architecture, which provides information about necrosis, and steatosis in one exam.The same devices could also be used to perform multi-parametric ultrasound analysis usingDoppler and contrast enhanced methods. Despite the advantages, ARFI only shows elasticvalues in a small, predetermined region for each acquisition. Although obtaining measure-ments from obese patients is less of a challenge with ARFI [27], it is still a significantfactor [12].101.3.4 Supersonic ElastographySupersonic imaging is a type of Shear-wave Elastography (SWE) imaging that can imagea region of interest in an imaging plane. In this method, multiple ARFI push forces areapplied in quick succession at different depths along a line. The push forces are appliedat different depths so the shear force is moving faster than the shear waves, thus creating aMach cone wave front that propagates away from the push line [10]. Supersonic imagingcan be found on the Aixplorer (SuperSonic Imagine, Aix-en-Provence, France). Supersonicelastography uses short transient focused acoustic excitations resulting in shear-wave band-widths typically ranging from 60 to 600 Hz [5]. Ultrafast imaging, achieved through aparallel receive architecture, creates a movie at 3,000-5,000 frames/s of the entire imagingplane, at 50 MHz sampling frequency [10] [5].The patient is placed in a supine position with maximal abduction of the arm. As doneduring TE, the probe is placed parallel to the intercostal muscles. Pressure is applied toget better acoustic coupling and decrease tissue thickness. The 1 cm diameter region ofinterest is usually selected to focus in the parenchyma, 2 cm beneath the Glisson capsule,while avoiding vessels [61]. The patient should be in a fasting condition and is instructedto hold their breath but not after a large inspiration [61]. Often the procedure is performedmultiple times in three different directions to get three propagation movies then a time offlight algorithm is used to compute shear wave speed in the region of interest of the imagingplane [5].Supersonic imaging is fast enough that patient movement has much less impact onthe results than other techniques, which tend to be motion sensitive [28]. It is the onlymethod to provide a map with quantitative measures of local elastic information in realtime [52]. It is an easy, painless, and quick method providing a shear map result in under 30ms [10]. The reproducibility is good with intercorrelation coefficients and intracorrelationcoefficients between 0.84 and 0.95 respectively [28]. Some limitations of the technologyinclude inaccurate elasticity measurements in the area along which the radiation force isapplied [10], as well as unreliable measurements resulting from high BMI or older age[61].1.3.5 Static Elastography - Real Time ElastographyReal Time Elastography (RTE) is a method developed by Hitachi (EUB-8500 and EUB-900) used to measure the strain in soft tissue and creating a qualitative colour map super-imposed on a B-mode image [28].11The patient is placed in a supine position with their right arm extended above and behindtheir heads. The original method involved manual compression of the ultrasound transducerof an amplitude based on an arbitrary scale. A qualitative elasticity map was displayed orquantitative elasticity ratios could be found based on the deformation of the liver tissuecompared to the deformation of intercostal muscle [35]. Newer versions use the internalcompression of the heart as mechanical stimulation instead of manual compression of theultrasound transducer [21]. The display shows a histogram of elasticity scores in the ROIand can be used to identify many parameters including average relative stiffness value,standard deviation of relative strain value, complexity, kurtosis, and skewness among others[67] [21].RTE for assessing liver fibrosis is fast, painless, and reproducible, displaying quickresults. Breathing does not cause artifacts unlike some of the other competing methods[67]. This method is more successful with patients with ascites and so has a broader targetgroup [67] [35]. The quantitative version of RTE has not been used in as many studies asthe other elastography methods and so is less proven and less standardized [28]. In a directcomparison by Colombo et al. TE and ARFI had the highest diagnostic accuracy whencompared alongside RTE [21].1.3.6 Magnetic Resonance Elastography (MRE)Magnetic Resonance Elastography (MRE) has had great success in staging and imagingliver fibrosis. It has become a standard method of diagnosis at the Mayo Clinic [19].The patient lies in the MRI in a supine position with a drum-like driver attached to theright side of their chest near the liver [19] or on their backside [32]. It is preferable that theexamination occur after fasting because prandial state has shown to affect stiffness results[30]. The shear waves are induced in the patients body using an acoustic speaker systemattached via a PVC tube [19] to an exciter outside the examination room, an electromag-netic device that uses the static field of the magnet, or using pneumatic devices [30] [32].Vibrations are produced between 40 Hz and 80 Hz [19]. A phase-contrast MRI sequenceincorporating motion-encoding gradients is typically used to acquire images of the wavepropagation, where the phase is locked to the mechanical excitation [32]. Many sequenceshave been attempted including gradient echo [19], spin echo with sinusoidal displacementencoding gradients [32], and echo planar imaging [33]. Encoding is repeated in threedifferent directions to obtain three separate elasticity maps. This process is motion sensi-tive so respiratory gating is used with a navigator on the right hemidiaphragm, and onlyend-expiration data is accepted [33].12In direct comparisons, MRE offered the best diagnostic performance of non-invasiveimaging techniques of fibrosis staging [33]. MRE showed great repeatability with bothintraclass and interclass correlation coefficients and was capable of imaging obese patientswho returned unreliable ultrasound scores [32]. It has a great capacity for being combinedwith other functional imaging types, such as perfusion or diffusion weighted imaging. How-ever, there are many limitations involved with using MRI as an imaging technique. MRIis an expensive and time consuming procedure, typically taking around 20 minutes for astandard procedure [33]. The patients must fit inside the bore of the machine with theshaker and must not be claustrophobic. Patients with high iron levels, such as those suffer-ing from hemochromatosis show low signals [32]. This excludes an affected population.These reasons make the technology much less accessible than ultrasound.1.4 Other Applications of Liver ElastographyElastography has been shown to be a good predictor of fibrosis progression. It also hasother uses, including predicting cirrhosis-related complications and monitoring therapiesand treatments including fibrosis development in liver transplant recipients [63] [63]. Stud-ies have shown its success in characterizing tumours where elasticity values have beenshown to differentiate benign from the stiffer malignant tumours [63]. In general, MREoutperforms the various ultrasound elastography methods because of the tissue heterogene-ity and small ROIs. Studies have correlated the risk of developing HCC with elasticityvalues of the liver and have shown elasticity is an independent risk factor. Most recently,trials are emerging that show elastography use for monitoring ablation.1.5 Current State of the FieldBoth MRE and ultrasound elastography are currently clinically available in North America.Transient Elastography is marketed as FibroScan (EchoSens) and has been used by gas-troenterology clinics for several years. Current developments are geared towards improvingstiffness results where obesity is a factor. This is facilitated by a new XL probe, which hasshown promising results. There are two point Shear Wave Elastography systems: VirtualTouch Quantification (Siemens Healthcare, Mountain View, Calif), and ElastPQ (PhilipsHealthcare, Bothell, Wash). These both return a stiffness value in kP based on measuringthe shear wave speed from a single A-line. Four 2D Shear Wave Elastography systemsare used clinically. They include SuperSonic (Imagine, Aix-en-Provence, France), SiemensHealthcare (Mountain View, Calif), Toshiba Medical Systems (Tochigi Otawara, Japan),13and GE (Waukesha, Wisconsin). The 2D systems allow the user to choose a region of inter-est and often display elasticities or shear wave speed superimposed on a B-mode image.Recent studies show that Transient Elastography and Shear Wave Elastography are com-parable to histological findings for determining whether there is a presence of severe fibro-sis. Current work is focusing on determining appropriate cutoff values to separate fibroticstages, as well as investigating different etiologies of fibrosis and how they each affect stiff-ness values. Although liver stiffness for HCV has been well documented, other etiologiesincluding HBV and NAFLD still require more research.Originally approved by the FDA in 2009, MRE is now available at over 100 clinicsacross five continents [31]. Several MRE systems are commercially available includingthe MR Touch from GE Healthcare, and MAGNETOM Aera and Skyra MR from Siemens.MRE diagnostics has both a high sensitivity and specificity, but there are still some chal-lenges. More recently, pulse sequences that can reduce the influence of motion artifacts dueto breath holding are being developed.14Chapter 2Vibro-Elastography RepeatabilityStudy2.1 BackgroundThe availability of diagnostic equipment has become an essential part of medical practice.New technologies have greatly improved what medical practitioners are able to see and as-sess in the human body. When novel devices are being developed, it is essential that theyprove their reliability, through comparison with existing techniques, and by showing theirrepeatable nature between scans. To accomplish this validation, scans are repeated severaltimes and the results are compared against each other and against results obtained with otherdiagnostic tools. These repeatability studies allow the developer to assess the types of mea-surement error that occur, and whether they are acceptable for the intended purpose. Thischapter presents the results of a repeatability study applying Vibro-elastography (VE) on aphantom, using different scanning protocols, ultrasound probes, and processing techniques.2.1.1 Waves in TissueElastography techniques have been developed to investigate the mechanical properties ofsoft tissues. Some of these properties include Young’s modulus, the shear modulus, vis-cosity, and relaxation-time. A mechanical excitation mechanism is used to send waves intoa tissue of interest. To extract mechanical properties from the wave information, certainassumptions must be made. A central assumption is that linear time invariant systems canmodel the stress-strain behaviour in tissue [2]. As well, soft tissue in vivo is nearly incom-pressible, resulting in the following assumptions regarding the mechanical properties of the15tissue:υ ≈ 0.5 (2.1)λ  µ (2.2)E ≈ 3µ (2.3)where υ is poisson’s ratio, λ is Lame’s first parameter, µ is Lame’s second parameter,or the shear modulus, and E is Young’s modulus. µ and E determine the material stiffness[3]. Waves have different characteristics depending on the medium of propagation, so tissueproperties can be determined by investigating wave behaviour.By the Helmholtz equation, and following the behaviour of waves in solids, there aretwo resulting wave systems in the tissue: the irrotational part, which describes the longitu-dinal wave due to tissue compressibility; and the equivoluminal part, which describes thewave due to tissue shearing. Both of these parts can be described by wave equations. Theirpropagation speeds are given as:cp =√λ +µρ(2.4)andcs =√µρ(2.5)respectively, where ρ is the tissue density. Using these equations, the shear modulus, µ ,can be estimated from wave speed since the tissue density, ρ , is assumed to be 1,000 kg/m3,and the wave speed can be found from the product of the frequency and the wavelength.The longitudinal wave speed in tissue does not change much; in order to be used inproducing ultrasound images, it is usually assumed to be constant and equal to 1,540 m/s,which is the speed of sound in soft tissue at body temperature. The shear wave speeds aremuch smaller than the longitudinal waves: approximately 1 to 10 m/s.2.1.2 Vibro-ElastographyVibro-elastography was developed in the Robotics and Control Laboratory at the Universityof British Columbia. It is another shear wave elastography technique that uses constantsingle or multi-frequency excitation applied to the tissue and standard ultrasound hardware16Figure 2.1: Ultrasound repeatability setup showing the Ultrasonix ultrasound ma-chine, the function generator used to create the excitation, the speaker used totransmit the excitation, the 4DC7 convex transducer both held in place on theCIRS phantomto acquire the Radio Frequency (RF) lines. It is able to produce real-time elasticity results,which could be beneficial for techniques that require constant monitoring, and it is beingoptimized for greater depths than are currently capable with other elastography systems.VE has previously been used for scanning prostate and breast within a depth of 5 to 10cm, using the 4DEC9 (Ultrasonix Medical Corp., Richmond, British Columbia, Canada)and the linear 4DL14 (Ultrasonix Medical Corp., Richmond, British Columbia, Canada)probes. It is now being adjusted for use with the convex 4DC7 (Ultrasonix Medical Corp.,Richmond, British Columbia, Canada) abdominal probe to be applied in liver and kidneystudies. Different acquisition paradigms have been developed, each with associated benefitsand drawbacks.The following is a description of the general set-up of VE, and is shown in Figure 2.1.The ultrasound acquisition can last several seconds, so the object or the subject being im-aged should be placed in a position where it can remain motionless. A wave exciter isapplied on the object to deform the tissue. The exciter is powered by a function generatorthat is controlled by the ultrasound machine computer. For this project, a small acousticspeaker with a diameter of 6.5 cm was used. When compared against other exciters (speak-ers of different diameters, or a structure shaker that was used in other projects), it provided17Figure 2.2: Vibro-elastography process for each elastogram. 20 frames are used tocreate one phasor image. A set of 9 phasor images are used to obtain eachelastogram.good wave propagation with little skin irritation on the subjects. While the excitation isbeing applied, raw echo data or RF lines are acquired from the volume of interest beingimaged.The volume is acquired by imaging multiple planes in sequence, and for each plane,multiple RF lines are acquired using sequencing that is described in sections and2.1.4.2. Using the RF lines, displacements resulting from the mechanical excitation canbe found between the frames at each plane. This motion estimation method is a TimeDomain Cross Correlation with Prior Estimates (TDPE) as presented in [71]. It uses thewell-documented Time Domain Cross Correlation Estimation (TDE) method, which findsdelays based on maximizing the normalized cross-correlation between data windows in RFlines acquired at different times, but it incorporates information from prior estimates ofneighbouring windows to increase the computational speed, thus making it useful for realtime applications. Based on excitation of the tissue and with knowledge of the applied fre-quency, the phase and amplitude at each point can be computed resulting in phasor images.The 2-D phasor images were interpolated onto a 3-D grid in order to find the local 3-Dshear wavelength, which is then used to calculate elasticity. A general diagram of the dataacquisition and processing is presented in Figure 2.2.Elastography relies on the capabilities of the imaging system to detect small motions inthe tissue at high frequencies. Higher frequencies are necessary for better resolution and todiscern smaller inclusions and possible lesions since several wavelengths should fit in theregion of interest for accurate visualization of inclusions [1]. Typical wavelengths are in therange of a few centimetres. Assuming a density, ρ , of 1,000 kg/m3 and given an exampleshear modulus of 14 kPa and an excitation frequency of 120 Hz, the expected wavelengthof 3.11 cm can be found following a rearrangement of equation (1.5):18wavelength=√µρfexc(2.6)The imaging framerate is limited by the speed of sound in the tissue, which is assumedto be 1,540 m/s and the depth of imaging in the Field of View (FOV) being imaged, whichtogether result in a minimum time span between samples.Timeline =2Dc(2.7)where D is the depth between the transducer and the desired echo depth, and c is thespeed of sound in the tissue. Standard ultrasound captures each line sequentially, so themaximum framerate is the rate that full frames of N lines can be obtained.FRmax =1NTimeline(2.8)Assuming an imaging depth of 15 cm, which was selected to capture the liver depthand was used for all measurements unless otherwise stated, each line needs 195 µs to besampled. The transducer has 128 elements, so the maximum framerate would be 40 Hz.Any motion in the tissue would be sampled at this frequency. Because of the Nyquistcriterion, the tissue motion would be band limited to 20 Hz. This frequency is not a practicalfrequency for elastography purposes. Typically, frequencies of 50 to 200 Hz are used forultrasound elastography.As previously mentioned, high frequency excitation results in better image resolution.However, higher frequency waves attenuate at a faster rate than lower frequency waves, sofrequency values are therefore selected to satisfy resolution requirements, while being lowenough to adequately penetrate the tissue to the desired depth.To detect and measure rapid wave motion, several strategies have been developed toincrease the framerate. One of these strategies involves using parallel receive beamformingas in the Aixplorer machine (SuperSonic Imaging, Aix-en-Provence, France). However thisapproach requires specialized hardware and so cannot be used on conventional ultrasoundequipment.Two approaches developed to address the low frame rate challenges were implementedin this project: sector-based elastography and band-pass sampling elastography. Sector-based imaging was designed to use conventional ultrasound imaging equipment while main-taining the original methods of beam-forming, and sending and receiving of the RF pulses.It optimizes the scan-line sequence to capture smaller sectors of the FOV at higher frame19rates. The second method, bandpass imaging, can be applied to band-limited signals wherea low sampling rate can capture high frequency motion as long as the sampling frequencyis at least twice the bandwidth of the applied excitation. A detailed description of these twomethods are detailed in section and section RF data acquired using bandpass or sector based imaging is used to obtain tissuedisplacements using TDPE for the full FOV. From this displacement data and resultingdisplacement phasors, mechanical properties are found using several methods. The Mayoclinic developed a Local Frequency Estimation (LFE) algorithm that estimates spatial fre-quency using local filter banks [44]. Unless otherwise indicated, all elasticity measure-ments obtained in this chapter were found using a 3D LFE algorithm. An alternative toLFE is a Finite-element Method (FEM) inversion approach, where unknown parameters ofthe finite element model are found using an iterative technique that minimizes the error be-tween the model and measured displacement. A further description of LFE as used in thisexperiment can be found in section Sector based ImagingSector based imaging is an imaging method that was developed to increase ultrasound fram-erate using standard ultrasound equipment.For many years, Doppler imaging has used the method of capturing a smaller ROIwithin the FOV, to speed up imaging rates. By choosing a smaller area to image, fewer linescan be collected, so the repetition of each point in the ROI is captured at a faster frequency,the Point Repetition Frequency (PRF). Figure 2.3 shows a typical imaging sequence fora FOV of 12 lines. A typical ultrasound frame scans a region of interest of 128 or 256lines. Figure 2.4 shows the line sequence of a sector-based paradigm, similar to the processdeveloped for Doppler imaging. From this figure, we can see that any point at a depth, d, ona given line will be sampled at a higher frequency using a sector-based sequence than thestandard line sequence, so the time between scans, T , will decrease. Each smaller sectoris collected for a pre-determined number of frames before acquisition moves on to the nextsector. The full ROI is covered sector by sector this way.One drawback to sector based imaging is that the joining edges of adjacent sectors arenot aligned continuously because of the time that elapses while all the frames of each sectorare being collected. This causes a significant phase shift between each sector. To align thedata from adjoining sectors, a phase compensation scheme was developed and applied tothe sector-based vibro-elastography system. This method, as outlined by Baghani et al. [1],assumes that the tissue being deformed follows a continuous-time dynamic system model.20Figure 2.3: Normal ultrasound acquisition order for 12 scan lines and two frames.Each line is scanned sequentially for each frame. Used with permission, 2010IEEE[1]Ideally, when investigating tissue displacement, all points would be sampled at the sametime. However this is impossible since the ultrasound wave can only sample one point at atime along individual lines, across each line, then through each sector. Because the speedof the ultrasound is assumed to be 1,540 m/s, the distance traveled can be calculated fromthe image depth, and the number of lines and the lines per sector are known, the phase lagbetween the different points can be estimated.In Dr. Baghani’s method, compensation followed three stages. At a given frequency,fe, the phase delay can be found based on the time, the distance travelled along each line,d, and the speed of the ultrasound wave, c.φ(d) = 2pi fet = 2pi fe2dc(2.9)With reference to Figure 2.4, the first phase compensates for the phase delay along anindividual line:Ucomp1(d) =Umeas(d)exp(−φ) (2.10)The second phase adds a factor that compensates for the delay between adjacent lines:21Figure 2.4: Scanning timeline for sector-based imaging. The FOV is split into sectors.The lines in each sector are scanned sequentially for all the frames before movingon to the next sector. Used with permission, 2010 IEEE[1]Ucomp2(d;m) =Ucomp1(d)exp(− j2pi fe(m−1)T ) (2.11)where T is the time interval between acquisitions of any given point, and m is the givenline. The final phase accounts for the time delay between a sector, n, for M lines per sectorand K sectors:Ucomp3(d;m,n) =Ucomp2(d)exp(− j2pi fe(n−1)MKT ) (2.12)Compensation is applied between points on each line, between neighbouring lines, andbetween neighbouring sectors, simulating acquisition of all data points at the same time.This method can also be extended between planes by either using timestamps or throughtriggering. Band-pass ImagingBandpass Imaging uses the theory of bandpass sampling to be able to detect motion fromsingle frequency or narrowband excitation that is applied at a higher frequency than thesampling rate of a conventional ultrasound system. Typically, to reconstruct a basebandsignal, the Nyquist criterion states that the sampling frequency must be greater than twicethe upper frequency of the signal. However, for band-limited signals at higher frequencies,a sampling frequency of twice the highest frequency component is often unattainable dueto the limited framerate of conventional ultrasound systems. Using bandpass sampling, asampling frequency that exceeds twice the bandwidth of the signal can be used instead. Iffu is the upper frequency limit of the applied excitation, and fL is the lower frequency limit,then the sampling frequency, fs must follow: fs > 2B, where B = fu− fL, and B is thepositive bandwidth of the excitation frequency band [24].Figure 2.5 shows two possible wave reconstructions resulting from the undersamplingof a 110 Hz wave: one at 50 Hz and the other at 60 Hz. Since the phase and amplitude arefound from the data, and the applied excitation frequency is known, the original signal canbe reconstructed.When undersampling a signal, the resulting spectrum in the frequency domain is theoriginal spectrum in addition to several frequency-shifted copies of the original spectrum.These copies are all shifted by an integer multiple of the sampling frequency. Because ofthis, it is important to select sampling frequencies that do not cause overlap, or aliasing, ofthe different spectral bands. An acceptable range for fs has been defined by Vaughan et alas [66]:2 fc+Bm+1≤ fs ≤ 2 fc−Bm (2.13)where fc is the center frequency of the signals specrum, and m is any positive integer.Figure 2.6 shows two examples of frequency domain representations of signals andsome possible results using bandpass sampling. The first signal is located at the baseband,where the positive and negative lobes of the signal are shown in Figure 2.6(a). When fs isselected so it is twice the bandwidth of the signal, where the bandwidth starts from 0 Hz,the signal is properly shifted and no aliasing can be seen (Figure 2.6(b)). In the figure,2.6(c) demonstrates the results of sampling a signal where fs does not meet the Nyquistcriteria, resulting in overlap of the aliased bands. With the proper selection of fs, where it isgreater than the bandwidth of the signal, figure 2.6(d), the resulting bands show no aliasing(Figure 2.6(e)). Spectral inversion, a phase shift of pi , in the baseband copy of the signal,23Figure 2.5: A demonstration of undersampling at 50 Hz (solid line) and 60 Hz (dashedline) of a 110 Hz waveform. The resulting waveform is a 10 Hz signal. The 60Hz sampling frequency results in a phase inversion. Used with permission, 2011IEEE[24]can occur if m is an odd number, as seen in Figure 2.6(f).When sampling the signal, only a finite number of samples can be taken over a periodof time, resulting in a windowing effect of the signal. With windowing, there is a trade-off between the bandwidth of the window and spectral leakage, so even a single frequencysinusoidal excitation has a bandwidth that must be considered when selecting the samplingfrequencies to avoid overlap [24].When using bandpass sampling, the Signal to Noise Ratio (SNR) is reduced comparedto other ultrasound techniques since the out-of-band white noise is amplified m times andoverlayed with the passband spectrum. For this reason, higher frequencies may be asso-ciated with increased noise, so parameters should be selected to minimize the number ofspectral shifts, m.For multi-frequency excitation, the different excitation frequencies selected must not re-sult in the same baseband frequency. Two signals with the same baseband frequency wouldcause spectral overlap resulting in errors when obtaining phasors from the displacements.24Figure 2.6: Bandpass Sampling of a band-limited signal. For a baseband signal (a)a sampling rate can be selected such that the original signal is reproduced andshifted (b). If the sampling freqency does not meet the Nyquist criteria, thenoverlap of the shifted signal may occur (c). For a higher frequency band-limitedsignal, a sampling frequency can be selected to properly reproduce the originalsignal (e). Sometimes, a phase-shift occurs (f). Used with permission, 2011IEEE[24]2.1.3 Equipment Used/Experiment SetupThe ultrasound device used for this experiment was a SonixTouch Ultrasound from Ultra-sonix Medical Corporation (Richmond, BC). Their Sonix software, used for clinical ultra-sound exams, was used to identify the location of the inclusion in the phantom. The phan-tom scanned was a custom made CIRS phantom with 8 inclusions, which were designedto have similar echogenicity as the background thus requiring elastography techniques todistinguish them. The background material, made of the elastic substance Zerdine, had astiffness of 26.4 kPa, while there were two inclusions of different sizes for each of the fol-lowing four stiffness values: 5 kPa, 14 kPa, 32 kPa, and 45.2 kPa. The 14 kPa inclusionwas selected for the experiments because of its location towards the mid-line of the phan-tom and its similarity to typical liver stiffness values, which should fall below 7 kPa forhealthy individuals [55]. The captured frame is roughly centered on the 14 kPa inclusion.25The VE software, eScan, was developed in house, and it triggers the mechanical excitation,runs the data acquisition, and processes elasticity values using an LFE algorithm.2.1.4 Software Development Kits (SDKs)Vibro-elastography was implemented using two Software Development Kits (SDK)s pro-vided by Ultrasonix: Ultrasonix Porta, and Texo. PortaThe Porta development toolkit uses the standard ultrasound acquisition available on theirclinical system, but allows users to build their own software around these settings. Portadoes not allow the user to change the sequence of lines fired so sector based imaging wasnot possible using Porta. TexoIn contrast, Texo allows lower level control of the ultrasound data acquisition, includingcontrol over beam sequencing, and transmit/receive control. Texo also provides a capabilityto enable frame triggering. Because of this researcher control, custom firing sequences canbe implemented on Texo, so both sector based and band pass elastography systems wereavailable using this SDK.Although Texo allowed more control in the data acquisition stage, a disadvantage ofworking with the Texo system is the poor quality of the B-mode image. As well, the Texoimage is not a clinically approved B-mode image, which may be needed for future humanstudies. The Porta B-mode images benefit from the post-processing and filtering techniquesthat Ultrasonix has implemented for their clinical images. When collecting phantom data,this was not an issue since the Sonix software was first used to identify the inclusion andset the transducer location, but during data acquisition of the liver a clear clinical B-modeimage is necessary to select appropriate volumes and visualize the anatomy being scanned.This would be a requirement for both clear visibility as well as for the approval and use inclinical studies. Plane TriggeringUsing the Texo SDK and its frame triggering, one of the methods we tested involved planetriggering, whereby each new imaging plane was triggered to collect RF echo data at thesame phase of excitation. The alternative and default setting uses the timestamp of collec-26tion to determine the time delay between planes and compensate for any phase differencesin the waves. The hypothesis is that by collecting each plane at the same phase using planetriggering, the system is less dependent on the timestamp of the ultrasound software, whichmay include inaccuracies that could affect results. Inaccuracies in the timestamp valuescould cause fitting challenges in the elevational direction, leading to inaccurate wavelengthestimates. Ideally all the points in the volume should be captured instantaneously. Thedrawback to plane triggered frameworks is a much slower acquisition time since there isan increased delay between the acquisition of each plane in the volume, because an entireperiod might be waited out before triggering the next plane. This method would be a chal-lenge for real-time scanning on a patient, particularly when breath holding is required. Thebandpass and sector-based schemes were each repeated twice: once using plane triggering,and once without2.1.5 Local Frequency EstimationLFE is a method, first described by Knutsson et al. in 1994 [38], that is used to obtain themechanical properties of the tissue from phasor information. Given that the wavelength, λis found and the excitation frequency, fex, is known, the shear modulus, µ , can be calculatedby:µ = f 2exλ2ρ (2.14)In this method, filters set at different central frequencies and applied in different direc-tions and different scales are combined to find the local spatial wavelengths of the phasors.By combining the output of these filters, a local frequency is found and applied across theimage. 3-D LFE produces more uniform stiffness maps than 2-D LFE because 2D LFE isaffected by the 3D variation of the phasor field, and the complexity of the wave field aroundedges, resulting in an overestimation of the stiffness values [70].Although LFE is quick to run, it has associated drawbacks including lower resolutionresults compared to FEM results. As well, in regions with strong refraction and reflection,artifacts and inaccuracies can develop near boundaries or objects [43]. For this reason,inversion algorithms and more robust methods have been developed, such as Manduca etal.’s inverse approach [43]. The VE system described in this chapter uses Manduca’s LFE.272.2 Method2.2.1 Set-upThe phantom was placed on a solid table with the transducer and shakers each fixed in placewith a stand and clamp. A small foam pad was placed under the speaker to ensure propercontact with the phantom surface. All measurements were taken by the same operator toavoid any inter-observer discrepancies. To properly assess repeatability instead of repro-ducibility, measurements were repeated on the same phantom sequentially, under identicalconditions. For each scan, five individual volumes were collected immediately one after theother.For the purpose of this repeatability study, the following five methods were compared,incorporating the two SDKs, the two acquisition methods, and plane triggering:1. Porta, using the bandpass acquisition scheme (plane-triggering and sector-based op-tions were not available). For more information, refer to sections and Texo, plane triggered, sector based method ( ( Texo, plane triggered, bandpass method ( ( Texo, non-plane triggered, sector based method ( ( Texo, non-plane triggered), bandpass method ( ( Parameter SelectionFor each volume, 11 planes were collected in the elevational direction with a motor step of1.09◦ between each plane. To obtain elasticity results, a three-dimensional LFE algorithmwas used. The user selects the number of phasor planes used to obtain each 3D elastogram.If too few planes are selected, there will not be enough of an elevational span to properlyestimate the wavelength using LFE; however, if all planes - or too many - are used, thecomputation speed of the elasticity will be very slow. In this study, 9 planes were used toestimate each elastogram. Since there were 11 planes in total, we obtained three elasticityimages centered in each sweep located on the central planes: 5, 6, and 7. The first elas-togram, displayed on plane 5, used phasor planes 1 through 9 to calculate elasticities, thesecond (on plane 6) used planes 2 through 10, and the final (plane 7 elastogram) used planes3 through 11.28Some parameters were constant regardless of the method of acquisition. Some of theseinclude:• frames per plane: 20• number of planes: 11• number of sweeps (volumes): 5• focus depth: 7.5 cm• ultrasound depth: 15 cm• transmit frequency: 3.33 MHz• sampling frequency: 20 MHzThe ultrasound depth of 15 cm and a focus depth of 7.5 cm were selected to simulatetypical parameters that would be required for future data collection on the liver. One ex-periment (section varied the focal depth to investigate its influence on elasticityresults. For all other sets with the 4DC7 curvilinear probe, the focus remained at 7.5 cm.The 2 cm - diameter inclusions were located approximately 4 cm deep from the surface ofthe phantom. Both the transmit frequency and sampling frequency were fairly low values toallow for deeper wave penetration and to capture a large volume. Increasing either of thesevalues resulted in memory issues or errors in the program.The frame rates were determined based on the acquisition method. It is detailed in thefollowing section. Frequency and Frame Rate SelectionThe effects of single frequency and multi-frequency excitation were tested by repeatingeach test twice by first using individual frequency followed by a multi-frequency vibration.Wave attenuation in media increases with frequency; this is true of both the dilatationalwaves used to generate B-mode images and for shear waves [49]. So frequencies wereselected to ensure a short enough wavelength to be able to discern the small, 2cm inclusionfor the given stiffness values, while keeping a low enough frequency for adequate wavepropagation through the depth of the capture range. To ensure a proper comparison betweenthe five methods, the frequencies selected had to be the same for both bandpass and sector-based imaging, and so had to satisfy each of their requirements.29Sector-based acquisition results in a limited amount of time for which each segmentof tissue is being sampled, so the observation timespan is limited and therefore some re-strictions must be added to the excitation frequencies. For a given point in a sector, it willbe sampled K times, where K is the number of frames chosen. So with a fixed samplingfrequency, the observed time frame for each point is:Tobs =Kfs(M)(2.15)This results in a fundamental frequency of:f f und =1Tobs(2.16)A limited time duration signal can be evaluated using the Fourier series resulting inenergy found at integer multiples of the fundamental frequencyfe = l f f und =l fs(M)K(2.17)Because of the Nyquist criterion, the integer is limited by:− K2< l <K2(2.18)resulting in a set of possible excitation frequencies [1].The maximum frame rate for each sector was limited by the line depth of 15 cm, theultrasound speed of 1,540 m/s, and the number of lines in each sector. With 8 lines persector, the maximum frame rate was 641 Hz. The number of frames collected was 20, so aframe rate of 400 Hz was chosen, so possible frequencies could be integer multiples of 20Hz. A good range of excitation frequencies for elastography is between 100 and 200 Hz,but with the relatively deep FOV, frequencies at the lower end of this range are preferred,so 100 Hz, 120 Hz, and 140 Hz were selected.For bandpass sampling, the main requirement in selecting excitation frequencies is thatthey should not be mapped to the same baseband frequency. Recall that the frequency rangecan be a shifted version of the baseband. The base-band frequencies were found for 100 Hz,120 Hz, and 140 Hz, using a 32 Hz frame rate. The resulting baseband values were 4 Hz,8 Hz (with inversion), and 12 Hz respectively, thus satisfying the criteria for both methods.302.2.2.2 Location of ExcitationThe shaker was placed at three different locations relative to the ultrasound probe to mea-sure the wave propagation and the effects on the repeatability of the resulting elasticityvalues. The first position, the Near-Lateral (NL) position (Figure 2.7a), placed the speakeras close to the transducer as possible on its lateral side, while avoiding any direct con-tact. The second position, the Near-Elevation (NE) position (Figure 2.7b), had the speakerplaced just above the axis of the transducer avoiding direct contact. The final position,the Far-Both (FB) position represented a speaker location at the furthest corner from thetransducer (Figure 2.7c). This last location was selected to simulate situations in whichthe applied vibration is not directly adjacent to the ultrasound transducer and to see howthis may impact the outcome. In the case of Far Both (FB) exciter placement, there was adistance of approximately 13 cm between the excitation speaker and the ultrasound probe.(a) Near-Lateral (NL) (b) Near-Elevational (NE) (c) Far Both (FB)Figure 2.7: Placement of Exciter relative to the transducer placementEach of the acquisition methods were used to collect data using both single frequencyexcitation and multi-frequency excitation. Table 2.1 shows a template of the different meth-ods investigated: the three exciter locations, the five acquisition methods (including band-pass, sector and plane triggering), and the single versus multi-frequency options. Five con-secutive ultrasound volumes were collected for each of the combinations displayed in thetable.2.3 ProcessingThe raw RF data was processed offline using the eScan software script runElastography.cpp.The information extracted includes the B-mode images, the correlation coefficients at each31NE NL FBsingle multi single multi single multiTexo, PTsectorbpTexo, NPTsectorbpPorta bpTable 2.1: Data Acquisition Sheet. NE: Near-Elevation exciter position; NL: Near-Lateral exciter position; FB: Far-Both exciter position; PT: Plane Triggering used;NPT: No Plane Triggeringpoint of the image, and the phasor data, which include the real, imaginary, and absolutephasor parts. Elasticities were found using a three dimensional LFE algorithm, unless oth-erwise indicated. The 3DLFE technique fits a wave at the excitation frequency to the phasordata, along the three volume dimensions. When applying the phasors to a cartesian grid,the largest rectangular region that included the inclusions was selected to avoid padding theedges with zeros. The LFE algorithm would fit the waves in the elevational direction ofthe planes to obtain the elastograms. With a total of 11 planes acquired for each sweep,this resulted in three elasticity images per volume. The quality factor for each plane wascalculated based on the goodness of fit of the Fourier transform from plane to plane, andthe correlation coefficient represents the effectiveness of measuring displacement from theframes. The phasor, B-mode, and elastograms were scan converted to the geometry ofacquisition before further analysis. Figure 2.8 presents an example of a these images.For multi-frequency datasets, the three frequencies were processed individually by ex-tracting phasor information for each component: 100 Hz, 120 Hz, and 140 Hz. The elas-tograms were created for each frequency and these were averaged together for final results.2.4 Repeatability AnalysisTo analyze the data and investigate the repeatability of the data, MATLAB scripts comparedand displayed the calculated elasticity results. Using the elastograms, the B-mode images,and any relevant information from the initialization.xml file. The user selects appropriateareas of measurement using the B-mode and phasor images. These regions were chosensuch that they were big enough and had adequate wave coverage. To determine the wavecoverage, the phasor images were used. The average elasticity within each area of mea-surement was calculated for each elastogram. These values were plotted for each sweep32(a) B-mode image(b) Real Phasor Image(c) ElastogramFigure 2.8: Example of a B-mode, phasor image, and resulting elastogram from acentral plane of one of the Porta Near-lateral Excitation. The 5 kPa inclusion isto the right of the main 14 kPa inclusion.33(a) Measurement Areas, B-Mode (b) Measurement Areas, Phasors(c) Measurement Areas, ElastogramFigure 2.9: Locations of phantom inclusions and areas of measurement on the B-modeimage, phasor image, and elastograms. This particular scan was using the farexciter placement and band-pass, not plane triggered sequence.to compare how repeatable the elasticity values were from sweep to sweep. Figure 2.13shows the outlines of the inclusions as well as the two measurement areas superimposed ona B-mode, a phasor image, and an elastogram.342.4.1 Statistical AnalysisTo quantify the repeatability of the VE method, three areas were considered. To start, sev-eral Analysis of Variance (ANOVA) tests were performed to assess the accuracy of eachmethod and to investigate which of the methods contributes to the variation in elasticityresults. These used the placement of the exciter, the sector-based and bandpass acquisi-tion schemes, and the plane triggering as parameters. Secondly, Confidence Intervals werefound to assess the agreement - an indication of data precision - of the elasticity values.Thirdly, difference tables show reliability, whereby the closeness of experimental elasticityvalues to the true value of the phantom being scanned were calculated. The ICC values forthe different protocols were found to discuss the source of variance in the data, contributingto the estimate of reliability of the methods. ANOVAThe ANOVA test is performed to assess whether factors influence repeatability while ac-counting for the influence of the other factors. After performing an ANOVA between obser-vations assigned to different groups, the results indicate whether the differences seen in themeasured observations can be attributed to the grouping. For example, in this repeatabilitystudy, two-way ANOVAs were carried out, investigating the effects of 1) Exciter Placementand 2) Scanning method (plane triggering, sector vs band-pass sampling, etc...) on resultingelasticity results. Each of these tests also gives an indication of the interaction of the twolevels. Specifically, three two-way ANOVA protocols were run for each dataset.1. The first investigated exciter placement (rows) and the five different methods as de-scribed in section 2.1.2 individually (columns)2. The second looked into exciter placement (rows) and sampling method: Band-passSampling (BPS) versus sector based (columns)3. The last ANOVA looked into exciter placement (rows) and plane triggering (columns)Significance was determined based on a p-value of AgreementAgreement, which is an assessment of the measurement error of the device following mul-tiple tests, can be defined using 95% Limits of Agreement (LOA)s. LOAs are the values35found, where it is expected that 95% of future measurements will fall within these 95%limits. This value gives a measure of the precision of the data.LOA= mean±1.96(std.dev) Intraclass CorrelationReliability considers the true value of the measured object [4]. It contrasts the measurementerror of one method (within variance) with the difference in values between the differentmethods (between variance). The result is a value from 0 to 1, where 0 would indicate thatall variance is due to the measurement error, and 1 is a high reliability measure representingzero measurement error. It is also known as the intraclass correlation (ICC).ICC =(s)2(b)((s)2(b)+(s)2(w))(2.19)where s(b) is the variance between subjects, and s(w) is the variance within subjects.As well as finding the ICC for each frequency and both measurement areas, the differencesof the data from their true stiffness values were calculated for each method.2.5 Results2.5.1 4DL14 Probe ResultsVibro-elastography results have previously been successful using the linear 4DL14 probe.It is presented here to provide comparison and verify the results using the 4DC7 curvilinearprobe found in section 2.5.2. The initial repeatability experiment was performed on thelinear 4DL14 probe and the shallow CIRS049 phantom. The parameters were selectedin accordance with the transducer properties. The transducer is designed for abdominal,musculoskeletal, nerve block, or vascular imaging, with a depth range of 2 to 9 cm, and abandwidth of 14 - 5 MHz. The depth was set to 5 cm, with a focus of 3 cm. Parametersincluded a transmit frequency of 5 MHz and a frame rate of 72 Hz, values higher thansettings for the curvilinear probe. 25 frames per plane were collected, with 9 planes total,and 7 used for each elasticity image, resulting in three elasticity images centered on thefourth plane of each sweep. The frequencies applied were 175 Hz, 200 Hz, and 225 Hz.These frequencies were higher due to the shallower imaging depth required. The largeinclusion can clearly be seen in Figure 2.10b, which is the multi-frequency elastogram36(a) ROI placement, B-Mode(b) ROI placement, ElastogramFigure 2.10: Locations of inclusions and areas of measurement on the b-mode imageand elastograms for the 4DL14 probe. This particular scan was using the farexciter placement the porta SDK.resulting from averaging the individual elastograms of the three different frequencies ofexcitation. Similarly to the experimental set-up for the curvilinear repeatability experiment,data was collected using the three different excitation locations relative to the transducer:NL, NE, and FB.37MethodPT,bp PT,Sec NPT,bp NPT,sec PortaNEInclusion 0.26 -0.55 0.67 -1.0 3.5Background 12.27 11.21 9.37 11.29 15.83NLInclusion -2.55 -3.0 -2.2 -2.7 1.78Background -5.29 -7.12 -5.38 -7.88 -0.67FBInclusion -2.44 -2.99 -2.71 -2.85 -0.05Background -0.41 -1.05 -0.08 -2.07 3.21Table 2.2: 4DL14 Difference Table, single frequencyMethodPT,bp PT,Sec NPT,bp NPT,sec PortaNEInclusion 0.01 0.00 0.08 0.03 0.33Background 0.11 0.08 2.59 0.04 0.13NLInclusion 0.02 0.01 0.02 0.08 0.99Background 0.04 0.01 0.10 0.09 0.74FBInclusion 0.00 0.00 0.01 0.09 0.59Background 0.03 0.01 0.41 0.21 1.73Table 2.3: 4DL14 variance, single frequencyThe resulting elasticities, as shown in Figure 2.11 and Figure 2.12 for inclusion andbackground elasticities respectively, show repeatable results for all five imaging methodsand all three exciter locations. A difference table (Table 2.2) shows a fairly consistent un-derestimation of the elasticity, except for excitation from the elevation side. The near lateralposition generally underestimates the elasticity values. Data collected using Porta returnedhigher and less precise results for all three exciter locations (Table 2.3). These trends wereall exaggerated in the background measurement areas compared to the inclusion.2.5.2 4DC7 ResultsThe above 4DL14 linear probe results are the existing results previously obtained. The goalof this chapter is to try to reach similar repeatability results using the 4DC7 curvilinearprobe that can obtain greater FOVs and is appropriate for liver imaging. Using this probeintroduces new challenges because of the increased depth, which can lead to issues with38Figure 2.11: 4DL14 repeatability results for the inclusion, using single frequency ac-quisitionFigure 2.12: 4DL14 repeatability results for the background, using single frequencyacquisitionwave propagation and noise, as well as the curvilinear geometry. The greater FOV intro-duces limitations in terms of memory usage since more memory is needed for each volumeof data. The following are results obtained using the 4DC7 probe. Single Frequency ResultsFigure 2.14 presents a visualization of single frequency results for all of the acquisitionmethods and exciter locations. Calibration based on the excitation frequencies may beneeded to reduce the differences between obtained and true elasticity values. To do this,39(a) Real Phasors, 100 Hz Excitation(b) Real Phasors, 120 Hz Excitation(c) Real Phasors, 140 Hz ExcitationFigure 2.13: The excitation was sufficient to create visible waves, shown as phasorimages, through the phantom. These images are from the Near Lateral, planetriggered, band-pass acquisition.40(a) Inclusion measurement area (b) Background measurement areaFigure 2.14: Single frequency Acquisition Resultsa closer investigation into the effect of excitation frequency on resulting elasticity shouldbe done. Table 2.4 shows the average value and the LOA for each method. The rangeof the LOAs were under 8% of the average value for all the cases except one, showinggood precision and therefore good repeatability. Table 2.5 is a difference table presentingthe results of subtracting the manufacturer stiffness values of 14 kPa for the inclusion and26.4 kPa for the background from the calculated elasticity values. This table shows goodaccuracy in the results, particularly for the inclusion.The ICC value for single frequency acquisition for the inclusion and background were0.986 and 0.999 respectively. A large part of the variance in the data can be explained by themethods themselves, which would be a type of systematic inaccuracy, rather than randomerror, resulting in high ICC values in both cases. The ANOVA results investigating the rolesof exciter placement, plane triggering, and BPS versus sector-based acquisition indicate thatthe difference in results between bandpass and sector based imaging is significant. However,the influence of plane-triggering is not significant. Multi-Frequency ResultsFigure 2.16 displays the resulting elasticities from multi-frequency acquisition, where theresults of the individual frequency acquisitions are averaged together. The individual fre-quency elasticity components can be seen in Figure 2.15. Table 2.6 shows the average41MethodPT,bp PT,Sec NPT,bp NPT,sec PortaNEInclusion 13.37[12.09,14.66]13.01[12.87,13.14]15.45[14.85,16.04]13.00[12.63,13.35]12.78[12.50,13.07]Background 37.61[36.92,38.30]32.18[31.27,33.10]35.42[35.13,35.71]31.49[31.30,31.67]31.33[30.56,32.10]NLInclusion 11.34[11.22,11.47]11.01[10.92,11.10]11.47[11.33,11.62]10.81[10.69,10.92]10.82[10.63,11.02]Background 20.59[20.43,20.75]20.39[19.96,20.15]21.03[20.80,21.26]19.99[19.83,0.15]19.68[9.56,19.79]FBInclusion 8.95[8.90,9.00]9.02[8.95,9.00]9.36[9.18,9.53]8.81[8.74,8.88]9.96[9.91,10.01]Background 33.72[33.63,33.80]31.58[31.37,31.80]34.58[34.18,34.99]31.63[31.53,31.73]31.58[31.42,31.74]Table 2.4: Average Elasticities (kPa) and 95% Confidence Interval limits for singlefrequency excitation at 120 Hz. Manufacturer stiffness values were given as 14kPa for the inclusion, and 26.4 kPa for the background.MethodPT,bp PT,Sec NPT,bp NPT,sec PortaNEInclusion -0.63 -0.99 1.45 -1.01 -1.22Background 11.21 5.78 9.02 5.09 4.93NLInclusion -2.66 -3.00 -2.53 -3.19 -3.18Background -5.81 -6.01 -5.37 -6.41 -6.72FBInclusion -5.05 -4.98 -4.64 -5.19 -4.04Background 7.32 5.18 8.18 5.23 5.18Table 2.5: Difference table (kPa) for single frequency excitation at 120 Hz. Manufac-turer stiffness values were given as 14 kPa for the inclusion, and 26.4 kPa for thebackground.value and the LOA for each method. The range of the LOAs were much lower for the back-ground areas, with the highest value as a percentage of the average elasticity value being6.7%, with one exception (Not Plane Triggered, bandpass, far excitation). Compared tosingle-frequency acquisition, multi-frequency had a larger range of LOA indicating morevaried results. Similar to single frequency results, a difference table (Table 2.7) showsthat the calculated elasticity results often underestimates the stiffness values relative to themanufacturer’s values.The ICC value for multi frequency acquisition for the inclusion and background were42(a) 100 Hz, Inclusion re-gion(b) 120 Hz, Inclusion re-gion(c) 140 Hz, Inclusion re-gion(d) 100 Hz, Backgroundregion(e) 120 Hz, Background re-gion(f) 140 Hz, Background re-gionFigure 2.15: Multifrequency Acquisition Results for each individual frequency0.9393 and 0.9587 respectively. These values are still very high showing that the type ofmethod used is the dominant source of variance. However, they are slightly lower thansingle frequency showing that random error plays a larger role. The ANOVA results in-vestigating the roles of exciter placement, plane triggering, and BPS versus sector-basedacquisition suggested that exciter placement played a significant role in the results.Overall, the results of the repeated acquisition show very little variation within the meth-ods suggesting the random error of the system is very low. However, the results showsystematic error because the results differ from expected values in a consistent way. Ifmanufacturer values are true, there is a consistent underestimation of elasticity values. Thephantom used was several years old so the stiffness values should be verified using MagneticResonance Elastography (MRE). Calibration may be needed to adjust elasticity estimates.43(a) Inclusion (b) BackgroundFigure 2.16: Averaged multifrequency ResultsMethodPT,bp PT,Sec NPT,bp NPT,sec PortaNEInclusion 17.96[17.38,18.54]13.59[13.02,14.15]13.38[12.59,14.17]13.90[13.30,14.50]15.20[14.74,15.65]Background 32.95[32.40,33.49]24.81[24.21,25.42]24.26[23.54,24.98]24.58[24.07,25.08]25.76[25.25,26.27]NLInclusion 8.09[7.71,8.48]9.72[8.51,10.93]9.59[6.83,12.35]7.72[4.97,10.48]8.51[8.28,8.75]Background 18.31[18.12,18.49]18.95[18.34,19.57]18.47[17.86,19.09]17.93[17.66,18.19]21.64[21.49,21.80]FBInclusion 10.75[10.22,11.28]9.40[8.38,10.43]10.31[8.73,11.90]11.23[9.82,12.65]12.04[11.73,12.35]Background 24.66[24.47,24.86]24.93[24.38,25.48]27.49[21.79,33.20]25.69[25.02,26.36]28.93[28.72,29.14]Table 2.6: Average Elasticities (kPa) and 95% Confidence Interval limits for multi-frequency excitation. Manufacturer stiffness values were given as 14 kPa for theinclusion, and 26.4 kPa for the background.44MethodPT,bp PT,Sec NPT,bp NPT,sec PortaNEInclusion 3.96 -0.41 -0.62 -0.10 1.20Background 6.55 -1.59 -2.14 -1.82 -0.64NLInclusion -5.91 -4.28 -4.41 -6.28 -5.49Background -8.09 -7.45 -7.93 -8.48 -4.76FBInclusion -3.25 -4.60 -3.69 -2.77 -1.96Background -1.74 -1.47 1.09 -0.71 2.53Table 2.7: Difference table (kPa) for multi frequency excitation. Manufacturer stiff-ness values were given as 14 kPa for the inclusion, and 26.4 kPa for the back-ground. Quantifying the Quality of Elasticity ResultsTwo metrics have been implemented to display the quality of elasticity results: the correla-tion coefficient and the Quality Factor (QF). The first, the normalized correlation betweentwo signals, is a measure of the accuracy of the displacement estimate as calculated usingTDPE. The displacements are found using pairs of sequential frames. Correlation valuesare found for each line and each pair of frames at the peak position using [71]:∑t=i+Wt=i a(t)b(t)√∑t=i+Wt=i a(t)2√∑t=i+Wt=i b(t)2(2.20)The correlation falls between zero and one. A typical correlation image shows the cor-relation at each point in the image (Figure 2.17). The correlation values generally becomemore consistent at deeper depths. This is particularly true through the centre of the scanwidth, as displayed in Figure 2.18. Correlation values typically remain fairly high, but aremore consistent at greater depths than more superficial regions (Figure 2.19).The QF is a representation of how well the elasticity can be estimated between consec-utive frames. It uses a least squares method to quantify the fit of the applied wave based onthe excitation frequency. One QF value is saved for every plane acquired. Values are usu-ally greater than 90% and are relatively consistent for each plane throughout the volume.Table 2.8 shows median QF results values for each acquisition method.To investigate the relationship between the correlation coefficient and the accuracy ofelasticity results, the correlations for each elastogram were plotted against the differencesbetween estimated elasticity and manufacturer’s values for each method. Figure 2.20 andFigure 2.21 show example results for all the data points collected for single and multi-45Figure 2.17: Typical correlation image before scan conversion to transducer geometry.The correlation falls mostly above 0.95 indicating a good fit along scan lines.This scan used the Far-both excitation position and a sector based acquisition.Figure 2.18: Correlation values plotted against the scan depth, using a central regionof the scan. This scan used the Far-both excitation position with no plane trig-gering and a sector based acquisition.46(a) Correlation across scan width: top section(b) Correlation across scan width: middle sec-tion(c) Correlation across scan width: deep sectionFigure 2.19: Correlation plots showing the trends across the scan width at variousdepths47MethodPT,bp PT,Sec NPT,bp NPT,sec PortaNESingle Frequency 98.50 98.71 98.68 98.37 98.92Multi Frequency 99.57 94.07 99.14 93.91 99.10NLSingle Frequency 98.79 98.50 98.86 98.63 98.82Multi Frequency 99.35 96.41 99.25 95.69 99.22FBSingle Frequency 98.52 98.70 98.80 98.65 98.59Multi Frequency 99.33 82.43 99.43 80.80 99.36Table 2.8: Median Quality Factor Valuesfrequency acquisition. The negative relationship indicate that as correlation values increase,the estimated elasticities are closer to the true values, suggesting that correlation is a goodindicator of measurement reliability.A comparison was also carried out using the quality factor. No significant relation-ship was found between the quality factor values and elasticity average values or elasticityvariance. Dependence of Exciter LocationFrom the results, it is evident that the location of the excitation has a large influence onresults compared to the acquisition method and plane triggering. An excitation placed lat-erally to the transducer produces the most repeatable results, while the further away thetransducer is, the larger the range of results. This outcome is expected because the near-lateral position creates waves that propagate laterally across each plane where the fit of thewaves and thus the tracking of displacements and resulting elasticity estimation would bemore consistent due to a higher resolution. For the position on the elevation side of thetransducer, the wave fitting goes through the elevational direction, and the fit must crossthrough the various ultrasound planes where the resolution is lower, creating a less accu-rate elasticity estimation. Despite being the most repeatable across all acquisition method,excitation in a position lateral to the probe consistently gave a lower elasticity estimatethan the elevational position and the further excitation location, which usually had elastic-ity estimates in the middle. In general, the elevational elasticity values are higher than theother two exciter locations for both measurement areas and both single and multi-frequencyscanning. This could be because the resolution across the 2-D ultrasound plane in the lateral48Figure 2.20: This plot shows the negative correlation between the average correlationvalue within the inclusion measurement areas and the accuracy of the elasticityvalue within that same measurement area, based on the difference of the calcu-lated elasticity using single frequency excitation and the manufacturer’s valueof 14 kPa.direction is much higher than the resolution in the elevation direction, so the wavelengthsacross the plane from the lateral position can be accurately estimated, whereas the elevationwaves are likely to be projected and appear longer resulting in a longer elasticity estimate.2.5.3 Other 4DC7 Results2.5.3.1 Low Amplitude ResultsThe excitation amplitude affects the elasticity results. If the amplitude is too low, there isa risk of wave attenuation as the distance in the medium increases. On the other hand, ifthe amplitude is too high, time domain cross-correlation estimates may fail because theyare ineffective at high strains due to de-correlation noise [71]. To investigate the effect ofexcitation amplitude, a dataset was collected using a low setting, where the excitation am-plitude was a quarter of what was used for the other sets. Figure 2.22 shows the results ofsingle frequency acquisition on the inclusion. The performance for data collected when the49Figure 2.21: This plot shows the negative correlation between the average correlationvalue within the inclusion measurement areas and the accuracy of the elasticityvalue within that same measurement area, based on the difference of the calcu-lated elasticity using multi-frequency acquisition and the manufacturer’s valueof 14 kPa.shaker was near the transducer is fairly repeatable, if the excitation is further away, resultsbecome more variable (blue squares in Figure 2.22). At low amplitudes, elasticity was con-sistently underestimated regardless of exciter placement or method. These results indicatethe excitation currently being applied using /acVE on phantoms all create consistent resultsand the system is not very sensitive to moderate amplitude changes. Effects of Focal DepthThe focal depth was set to three different settings, representing each interquartile depth ofthe ultrasound span. Because the depth of the ultrasound is 15 cm, the foci were set to: 3.75cm, 7.5 cm, and 11.25 cm deep. This dataset was collected on a solid portion of the phantomwith no inclusions. Three measurement areas were defined surrounding each focal centre(Figure 2.23). The single frequency results are displayed in Figure 2.24. Results were bestfor a shallow measurement region and a shallow focus. The elasticity results for a shallowmeasurement region was also closest to manufacturer’s values, regardless of focal depth.50Figure 2.22: Inclusion measurement area elasticity results using low amplitude, singlefrequency excitationFor deeper measurement areas, the accuracy of the results deteriorated.For the shallow and deep focal depths, the shallow and deep measurement areas returnedthe best results suggesting that setting the focus to the approximate area of interest returnsmore repeatable elasticity results.2.5.4 Elasticity Processing2.5.4.1 LFETo test the effectiveness of the elasticity processing algorithm, both a two dimensional off-line LFE processing technique was used, as well as the three dimensional LFE method.Figure 2.25 shows that the range of results for 2D LFE is much higher than 3D LFE. Aswell, 2D LFE tends to overestimate elasticity results. This could be because it is gatheringthe wavelength estimate from the projection of the 3D wave onto the 2D planes, naturallyresulting in an overestimate and therefore a higher stiffness value. So 3D LFE should beused when possible.51(a) Measurement area placement on a b-mode image(b) Measurement area placement on an elastogramFigure 2.23: Measurement area placements corresponding to the three different set-tings of focus. The top is located at a depth of 3.25cm, the middle measurementarea is located at 7.5 cm deep, and the bottom measurement area is 11.75 cmdeep.2.6 ConclusionThis chapter has presented a repeatability experiment for the VE technology developed inthe Robotics and Control Laboratory at UBC. It has described the various techniques em-ployed and has looked at results for sector-based and bandpass acquisition, the influenceof single and multi-frequency excitation, the impact of changing the location of the waveexcitation, and the results of using plane-triggering. The outcome has shown that the lo-cation of excitation is the biggest determinate of elasticity results. In terms of precision,results of elasticity values obtained from measurement areas ranging from 3 cm to 11 cm52(a) Measurement area centred at a depth of 3.25cm(b) Measurement area centred at a depth of 7.5cm(c) Measurement area centred at a depth of11.25 cmFigure 2.24: Results of three measurement areas at different programmed focal values53(a) 2D LFE box plot (b) 3D LFE box plotFigure 2.25: Inclusion Elasticitiesall show consistent results. Overall all the techniques show good repeatability, but there isstill discrepancy between calculated results and manufacturer values of stiffness using thisphantom.54Chapter 3Multi-modal Liver Registration3.1 IntroductionImage registration is a well established area of image processing that involves the alignmentof two separate images or volumes into one coordinate system. The goal of image registra-tion in medical imaging is to gain more information than would have been obtained fromonly one image. More recently, real-time registration has been used during surgical pro-cedures whereby real-time ultrasound images are overlaid with a pre-operative ComputedTomography (CT) or Magnetic Resonance Imaging (MRI) scan to obtain more detail wherethe procedure is taking place. Another potential use of registration could be the blending offunctional imaging modalities with anatomical ones to better assess the precise locations ofactivity in the body. Images from individuals can be compared to a generic atlas to observeany differences that may be present, or registration can provide information on pathologyprogression if data is collected following a period of time or a treatment program. Lastly,registration is also the alignment of images that may have been taken from different loca-tions and directions. In essence, image registration is the application of a transformation onone moving target so that it most closely aligns with a fixed target.When choosing registration methods, considerations include speed of computation,complexity of computation, and accuracy [37]. The ideal method depends on the intendedpurpose. Registration was performed for this project with the intent of setting up a methodof comparison between elastography values obtained from the VE system and results froma commercial MRE system. For this application, 3-D VE scans are registered to a clin-ical MRI scan, so a registration method that works well for multimodal cases was cho-sen. Feature-based methods are often chosen for multimodal applications, whereas inten-55sity based methods may be more efficient for unimodal applications. Although a non-rigidregistration would be ideal, a rigid registration should be sufficient to compare elasticityvalues at different regions of the liver. Prior Multimodal Liver RegistrationUltrasound and MR or CT scans have usually been registered in the context of pre-operativeplanning for liver resection or surgery. Usually MR or CT images are obtained in advancewhere the lesion or surgical area can be identified, and ultrasound images obtained duringthe procedure can be registered to this volume. Because this is a multi-modal applicationwith a whole volume for the pre-operative image, and a smaller segment obtained duringthe procedure, the outcome is similar to the goal of this project. Vessels are a useful land-mark in the liver and can be identified in most clinical scans, so vessel based registration isoften performed for multi-modal liver registration [25]. Surface based methods were morechallenging given that only small segments of the liver surface were apparent at any giventime.Iterative Closest Point (ICP) is a common technique that uses either vessel contours orcenterlines to match points from the ultrasound and MR or CT volumes. The differencesbetween several methods presented in the literature typically lie in how the vessel structuresare segmented and the variation of ICP algorithm used. Centerlines are considered a bettermarker than vessel surfaces because of the change in vessel size with the pulsatile motion ofthe heart and so have been used by many groups [39] [51] [40] [62] [25] [54]. The originalICP algorithm proposed by Besl [11] looked for individual point matches, but since thensome groups have refined the method by matching points to lines [62], or by looking forsimilar directions of vessel segment matches [25]. Since ICP is sensitive to incorrect pair-ings from noisy or only partially overlapping volumes, some methods incorporate pruningof data points to remove outliers [54]3.1.1 VE/MRI ChallengesOne of the challenges that arises with multimodal image registration is that the relationshipbetween intensity information from one set to the other may not be strong, whereas withunimodal registration, the intensity values between matching sections of each set are verysimilar, so cost functions can take direct measures comparing these intensity values [50].Depending on the imaging method chosen, imaging geometries and orientations could bean added challenge.56Ultrasound is a very commonly used imaging modality due to its quick imaging time,low cost, safety, and portability. Rapid registration of real-time ultrasound scans with pre-operative images is a growing field to increase information available to operating surgeonsand thus improve outcomes.Despite the benefits of ultrasound imaging, it has some drawbacks. Due to the freedomand flexibility of obtaining ultrasound images, the initial spatial relationship between ultra-sound and other images is never known unless the probe is being tracked. There could be asignificant rotational transformation between two ultrasound images or an ultrasound imageand an MR image. The types of coordinate systems could also be complicated by the useof a curvilinear transducer. Additionally, ultrasound has a limited view and can easily beobstructed, so many volumes collected using ultrasound contain shadows where bones orgas are present, or only partial organ volumes. The liver images can also be obtruded by theribs or in some cases the liver can be hard to image if the bowel is full of air. Sometimes,the depth of imaging is limited, the volumes will only be partial or the quality of image maynot be adequate. In the case of liver imaging with ultrasound, only partial volumes can everbe seen and are limited by the size of the participant, the imaging depth, and the angle stepsize of the volume. Ultrasound also can suffer from poor contrast or resolution, especiallyin the lateral direction, and low image quality. For this reason, preprocessing is necessary. Pre-processing of Ultrasound ImagesRelative to other imaging modalities, ultrasound can be quite noisy so it is necessary todo some image processing before the registration [14]. The noise comes from speckle andfrom challenges imaging at the depths required for abdominal imaging, where an increase inthe gain at deeper depths also amplifies any noise present. Some techniques that have beenused to overcome these challenges include the application of a depth gain compensator,which emphasizes the deeper layers of tissue and understates the surface layers, or specklereduction through the use of linear homogeneous filters. Nazem et al. found that the depthgain compensators were more effective [47]. Despeckling techniques in general can includemedian filtering, Wiener filtering, adaptive weighted median filtering, bilateral filtering,anisotropic diffusion filtering, and wavelet soft threshold filtering [23]. A sticks filter isa speckle-reduction algorithm that can accentuate edges or linear features, which can beuseful for automatic vessel segmentation [23][47].573.1.2 Liver Registration and ChallengesThere are several challenges associated with performing registration on the liver. To start,the liver is not a rigid organ and instead shows elastic properties causing it to change shapebased on its surroundings. It is affected by the beat of the heart, primarily in the left lobe,resulting in segmenting challenges. It also moves up to 5.5 cm with breathing motion[64] [20], so its position in the body and shape depends on the phase of respiration. This isa particular challenge for MRI-Ultrasound (US) registration as breath-hold MRI is typicallytaken on expiration because the position of the liver is most repeatable at this time, and itreduces artifacts in the resulting images. However, ultrasound frames are generally takenupon inhalation because when the lungs are full, the liver is pushed down below the ribcageand becomes more visible from the typical abdominal ultrasound scanning position inferiorto the bottom rib.Despite these challenges, the presence of vessels creates obvious landmarks for regis-tration and the surfaces can be used for surface based registration methods. However, due toits large size, only partial ultrasound volume acquisitions are available so surface based reg-istration algorithms that use the liver capsule will have to manage with only partial pieces.Several different registration methods have been applied to the liver in the past decade,utilizing both feature-based and intensity based algorithms. All use multiple stages of regis-tration, starting with a rough alignment at first, followed by refining the transform. Nazemet al. used a point-based registration, ICP, of segmented vessel points and an unscentedKalman filter of surface information on a liver phantom [47]. Similarly, Nam et al. used thegeometry of the vessels to implement an initial rough alignment and refined the matchingwith surface information [46].3.2 Method: 3D Multi-modal Liver Registration3.2.1 Introduction/PipelineFigure!3.1 shows a general pipeline for the multi-modal registration.3.2.2 Data AcquisitionFor the study presented in this chapter, the Sonix Touch ultrasound machine and the 4DC7-4D convex transducer (Ultrasonix Corp, Richmond, BC), designed for abdominal scanswere used. The volumes selected were primarily transverse volumes obtained just inferiorto the sternum and angled superiorly and laterally. Some images included few vessels58Figure 3.1: Sequence of steps for multi-modal registration59(a) A slice of the 3D B-mode volumewhere the liver surface can clearly beseen, but few vessels. The heart is alsovisible in the bottom right of the image.(b) A region of the liver where the ivc!can be seen with the LHV and MHVbranching off.Figure 3.2: Sample ultrasound imageswith the distal surface of the liver in sight (Figure 3.2a), whereas others showed the threehepatic veins clearly branching from the Inferior Vena Cava (IVC) (Figure 3.2b). Thesethree hepatic veins, the Left Hepatic Vein (LHV), Middle Hepatic Vein (MHV), and RightHepatic Vein (RHV), were segmented and used for registration.The MRI data was acquired using the Philips Achieva 3.0T MRI Scanner at the UBCMRI Research Centre. Anatomy volumes were collected using a T2-weighted Spin-Echosequence. The parameters for each dataset varied slightly, but all used under 1mm resolu-tion on each plane and a slice thickness of 4 to 5 mm. The volume used for multi-modalregistration taken at inspiration had 40 slices resulting in a volume depth of 16 cm, whichcovered most of the liver with the exception of the inferior right lobe. For better image qual-ity, a breath-hold was performed instead of breathing-gated imaging. Because of the motionof the heart, the outline of the left lobe was blurry and could not always be distinguishedfrom the boundary of the heart.3.2.3 Pre-processing3.2.3.1 Ultrasound Pre-processingThe first step after collecting both data sets was to scan convert the ultrasound data into its3D positions then interpolate this data onto a rectangular grid where it would become themoving object for the registration. The MR data, already set on a rectangular coordinatesystem was treated as the fixed object. The ultrasound volume was the moving object60(a) A slice of the raw B-mode volume be-fore processing.(b) A B-mode slice following histogramequalization to brighten image and im-prove contrastFigure 3.3: Impact of pre-processing on ultrasoundbecause of its limited view and feature points, whereas the MRI images contained the entireliver volume.Once the coordinate system was established, pre-processing was required to smooth theultrasound images and display the vessels clearly before segmentation. The images mostlysuffered from low brightness, so contrast was improved by equalizing the image histograms.Following the image processing, the data was scan converted to interpolate the planes ontoa cartesian grid. At this stage the resolution was selected and was chosen to include enoughdetails but not be too computationally heavy. A resolution of 0.5 mm for each of the x,y,and z directions was used for the ultrasound volumes. MRI Pre-processingThe MRI image quality was mostly determined based on MR parameters set during acqui-sition. The priority when choosing parameters was to ensure good contrast between thevessels and liver parenchyma. No offline processing was performed on these scans.3.2.4 SegmentationThere are three different software systems for image segmentation in our laboratory: Strad-win, 3D Slicer and VTK. 3D Slicer was used for image segmentation because of its versa-tility and wide array of modules for editing, placing fiducials, and smoothing. 3D slicer isprimarily designed for MRI and CT scans and can handle a wide variety of medical imag-ing formats but is best used with DICOM images. The B-mode volumes acquired using the61(a) The 3D segmentation of the liver sur-face taken from the MRI data. Therange spans from the top dome of theliver and down through most, but notall, of the right liver lobe.(b) The 3D segmentations of the liverfrom the ultrasound data. The surfacerepresents the field of view of the ul-trasound but only contains a small seg-ment of the outer boundary of the liveritself.eScan software and processed in MATLAB must then be converted into a readable formatfor 3D Slicer. After encountering some issues with writing the volumes to DICOM, theywere saved in the meta-data format using imageJ, a program used for image processingfor the National Institutes of Health (Bethesda, MD). At this stage, resolution and size re-quirements were set based on the original dimensions and resolution from acquisition andinterpolation.The liver vessels and surfaces were segmented using the segmentation tool for both theMRI and Ultrasound datasets, and surface models were created. Once the layers had beenidentified and smoothed in all three views, a surface model was created of the anatomy. TheMRI volume showed the most of the surface of the liver (Figure 3.4a), while the shape ofthe B-mode surface volume consisted of an outer boundary of the right liver lobe with theboundaries of the ultrasound field of view on all the other sides (Figure (3.4b)).To identify the main vessels visible in the ultrasound volume, the markups tool wasused in Slicer and fiducials were visually placed and labeled for the IVC, LHV, MHV, andRHV, when visible. This was done for both the ultrasound and MRI volumes. The savedfiducials were exported from 3D Slicer in a .fcsv file, and the surfaces were saved as .objfiles. These were all loaded in MATLAB and used for the registration itself or for visualvalidation.3.2.5 Initial Registration: Manual AlignmentOnce the fiducials were loaded into MATLAB, an initial rough registration was performed.The ultrasound images were generally acquired in a subcostal position, lateral to the volun-62teer’s midline, so an initial orientation relative to the MRI could be calculated. As well, thecentre of mass of the group of points - the manually selected vessel fiducials - was foundfor both the MRI and ultrasound volumes. The transform required to align the two masscenters was calculated and applied to the ultrasound points so both images were in the samegeneral area. Vessel ICPThe ICP algorithm is designed to best align two point clouds obtained from separate vol-umes. It finds a transformation that when applied to a moving set of data to best align itto a fixed set by minimizing the distance between two point clouds. The transformation,consisting of a rotational component Tr, and a translation, Tt can be applied to the movingpoint cloud through:pt = Trpi+TtIf the moving points are represented by pi, and the fixed points are represented by qi,then after applying the transformation, τ , the error, E, should be minimized, where E:E =N∑n=1τ((pi)−qi)2Following the rough alignment, a point to point ICP algorithm was applied to the land-marked vessel points to find the best rigid transformation to align the point clouds. Theresulting transform was applied to the ultrasound and segmented surfaces in Slicer to visu-ally check the registration success.3.3 Results3.3.1 MR Inspiration to B-Mode Inspiration RegistrationFive points were selected as target registration points based on bifurcations of the majorhepatic vessels. They were selected by an untrained researcher using the planar viewsin 3D Slicer (Figure 3.5) and were confirmed in the 3D view of the segmented vessels(Figure 3.6). The following are the locations of the targets used:1. Target 1: bifurcation of IVC and LHV2. Target 2: bifurcation of IVC and RHV63(a) Target selection using plane view(b) Target selection using the elevation view(c) Target selection using depth viewFigure 3.5: Three different views are used to place each target fiducial that will laterbe used for estimating registration error.64Figure 3.6: The 3D view along with vessel segments are used to check target fiducials3. Target 3: bifurcation of IVC and MHV4. Target 4: first bifurcation off LHV5. Target 5: first bifurcation off MHVThis process was repeated for both the ultrasound and MRI volume.The points initially had a registration error of 24.4 mm following rough alignment,which following the ICP registration was reduced to 18.5 mm, which is a reasonable valuefor the sake of roughly aligning the liver and comparing elasticity values, but it could be fur-ther refined in the future. The registration error suffered from human error in the placementof the fiducials. Approximations were made because the slice thickness of the MRI vol-ume was 4 mm thick, and the centre of the vessels at the locations of the bifurcations wereroughly estimated, so the initial placement of the target points could have varied by severalmm from the true anatomical landmark. Figure 3.8 shows the results of the registration ofthe vessel centerlines from the B-mode (Figure 3.7a) and the MRI vessels (Figure 3.7b).For the fiducial points used for registration, the final error was 9.3 mm, which is acceptablefor these conditions.3.3.2 MR Expiration to MR Inspiration RegistrationOne of the challenges of registering MRE and VE data through their correspondence withanatomy images is due to the drastic movement of the organ during respiration. BecauseMRE must be acquired on expiration, and VE is acquired after inspiration, the liver hasmoved within the body. To investigate the effects of respiration on the movement of theliver a registration was performed between MRI T2-weighted images on expiration and65(a) Ultrasound vessels before registration(b) MRI vessels before registrationFigure 3.7: Multimodal registration results66Figure 3.8: Alignment of vessel centerlines following ICP registrationinspiration. The inspiration image incorporated a nearly-full volume of the liver, whilethe expiration data was a reduced set of 16 slices that aligned more closely to the narrowvolume of eXpresso MRE [29], an MRE method being developed in-house. Both volumeswere loaded into MATLAB. The volumes were already roughly lined up since they camefrom the same MRI scan.The volumes were first registered using a Mutual Information (MI) algorithm. MI wasselected because both volumes came from the same imaging modality, and the algorithmused was well established and did not require user input. The resulting transform was ap-plied to the expiration volume, the moving volume, since it was smaller. A mid-axial slicecan be seen in Figure 3.9, showing a good alignment of the body outline before (Figure 3.9a)and after (Figure 3.9b) registration. However, the vessels are not identically matching dueto the large deformation of the liver during the breathing cycle and weak results in the z di-rection. Although the Mutual Information method required less user input, the computationitself and adjustment of the initial parameters took time.67(a) The two MRI volumes before registra-tion(b) The two MRI volumes after registra-tionFigure 3.9: Registration of two MRI volumes collected on inspiration (purple) andexpiration (green).The same ICP process described in section 3.3 was used for the two MRI scans to com-pare results, and the registration error decreased significantly. This was true even with arelatively small number of fiducials acquired due to the thin expiration slab. The final regis-tration error between the target points was 8.10 mm, while the error between the fiducials ofthe centre lines was 3.7 mm. This is a very good result. The final alignment of the vesselscan be seen in Figure Conclusion and RecommendationsRegistration results are heavily dependent on the circumstances of data acquisition. Be-cause of this, it is important that the volumes collected are able to promote better results.Ultrasound scans should be collected showing clear vessels that can be easily recognizedand labeled. Appropriate filtering and post-processing is required to obtain the best seg-mentation results. To improve outcomes, longer vessel segments should be identified or theliver surface could be incorporated into the registration. However, this is associated withlonger acquisition times if more planes are collected, thus demanding a lengthy breath-holdfrom the subject. Alternatively, the planes can be separated by a greater distance, resultingin poor resolution. Anatomical MR images should clearly show the outline of the liver andvessels, which is optimized with greater resolution and increased contrast. T2 Weightedimages are ideal for showing good contrast between vessels and liver tissue. Blurring at theboundary between the heart and the liver can be mitigated by using a cardiac gated scan,68Figure 3.10: Alignment of vessel centerlines following ICP registration of Inspirationand Expiration MRI volumesbut this greatly increases acquisition time. If possible the slice thickness or space betweenslices should be reduced for better resolution in the z direction.Further improvements would include automatic segmentation, which could be moreeasily performed with improved quality of the B-mode images such as further image pro-cessing and denoising. As well, more accurate results could be obtained if a final non-rigidregistration were performed, accounting for the deformation of the liver.69Chapter 4ConclusionElastography is a growing field in the realm of medical imaging as it offers additionalinformation about the health of soft tissue that standard imaging does not provide. Becausethe mechanical properties of tissue change according to pathological state, being able toidentify the stiffness and viscosity of tissue is useful for diagnosing and tracking of diseaseprogression. Ultrasound is an affordable, versatile, and accessible technology that providesadvantages over MRI imaging. Therefore the utility of ultrasound elastography is endless.Liver fibrosis is a condition that affects millions of people world-wide and is the result ofmany different disorders. It is a precursor to more severe and life threatening liver states,and so is a good marker of the severity of liver disease. Ultrasound elastography of the liveris a useful and non-invasive tool.VE is the elastography technique developed in the Robotics and Control Laboratory atthe University of British Columbia. It is currently being tested and developed for severaldifferent tissue types, pathologies, and hardware options. This thesis tests different methodsof implementing VE, its repeatability as a diagnostic tool, and it sets up the foundationfor a liver fibrosis elastography comparison study by presenting a registration method thataligns MRI and ultrasound volumes. It finds that Vibro-elastograhy is reliable on phantomsubjects, but some calibration may be required to align stiffness values with gold standardsor other elastography techniques.The repeatability study presented in this thesis is a thorough experiment that involvedperforming data acquisition many times. Based on the large number of samples, and thelow number of outliers, it shows the technology is reliable in acquiring data on a phantom.A next step would be to choose the best methods and carry out acquisition on in-vivo tissueto investigate the impact of human motion and the transfer of the excitation waves into the70liver. Some limitations of the registration included limited MRI data. Only one MRI scanwas available for each of the inspiration and expiration protocols, so the registration couldonly be validated on one case.The results of this thesis can be applied to various uses of Vibro-elastography in thefuture including work on other soft tissues, especially those that would use a curvilinearultrasound probe, require a greater FOV, or have a large number of vessels. An example ofsuch a use would be for elastography of the kidney.4.1 Contributions• A thorough repeatability study was developed and carried out to investigate the con-sistency of VE measurements based on different acquisition schemes including focaldepth, and excitation amplitude• A comparison of bandpass and sector based imaging, two methods developed andused with VE• A comparison of exciter placement relative to the transducer and its influence onelasticity measurements• A statistical analysis on the application of VE to a phantom using the parameters andequipment necessary for a liver study• A registration method that could compare the ultrasound volumes selected from theeScan VE software to an MRI gold standard if clinical MRE is available4.2 Future WorkFuture work would be carrying out a full patient trial to compare and validate VE as adiagnostic tool for staging liver fibrosis. This could be compared to one of several otherelastography methods including: fibroscan, clinically available MRE, or an MRE methoddeveloped in-house. To do this, the repeatability information found in this thesis will beuseful to do a power analysis and determine the size of the sample needed. The parameterschosen for the repeatability study were selected specifically for liver fibrosis analysis andcan be used for the implementation of a patient trial. The correlation between VE and otherelastography methods should be investigated along with any sources of bias.To improve VE results, several modifications can be implemented. These include:71• ECG gating to better dilineate the left lobe of the liver and to reduce any noise thatmay be associated with the motion of the heart• Adjusting parameters to reduce the imaging time for patients who have trouble hold-ing their breath for the extent of the VE scan• Investigating the influence of excitation frequency on elasticity results and weightingtheir contribution to multi-frequency results to improve the accuracy of results• Adapting FEM to the inverse problem of obtaining elasticity results from phasors forthe curvilinear probe and comparing FEM and LFE results• Further image processing and denoising of eScan B-mode images to allow for auto-matic vessel segmentation and extraction to automate the registration process• Identifying the best MRE method for use of comparison with VEThe above improvements are suggested before carrying out the patient study to improveVE outcomes. The registration can be further automated if it’s helpful for the sake ofmethod comparison.72Bibliography[1] Baghani, A., Brant, A., Salcudean, S., and Rohling, R. (2010). A high-frame-rateultrasound system for the study of tissue motions. Ultrasonics, Ferroelectrics, andFrequency Control, IEEE Transactions on, 57(7):1535–1547. → pages 18, 20, 21, 22,30[2] Baghani, A., Eskandari, H., Salcudean, S., and Rohling, R. (2009a). Measurement ofviscoelastic properties of tissue-mimicking material using longitudinal wave excitation.Ultrasonics, Ferroelectrics, and Frequency Control, IEEE Transactions on,56(7):1405–1418. → pages 15[3] Baghani, A., Salcudean, S., and Rohling, R. (2009b). Theoretical limitations of theelastic wave equation inversion for tissue elastography. The Journal of the AcousticalSociety of America, 126(3):1541–1551. → pages 16[4] Bartlett, J. and Frost, C. (2008). Reliability, repeatability and reproducibility: analysisof measurement errors in continuous variables. Ultrasound in Obstetrics &Gynecology, 31(4):466–475. → pages 36[5] Bavu, E´., Gennisson, J.-L., Couade, M., Bercoff, J., Mallet, V., Fink, M., Badel, A.,Vallet-Pichard, A., Nalpas, B., Tanter, M., et al. (2011). Noninvasive in vivo liverfibrosis evaluation using supersonic shear imaging: a clinical study on 113 hepatitis cvirus patients. Ultrasound in medicine & biology, 37(9):1361–1373. → pages 11[6] Bechmann, L. P. and Friedman, S. L. (2011). Fibrosis as a major mechanism ofchronic liver disease. In Chronic Liver Failure, pages 91–107. Springer. → pages 4, 5[7] Bedossa, P. and Poynard, T. (1996). An algorithm for the grading of activity in chronichepatitis c. Hepatology, 24(2):289–293. → pages 7[8] Bercoff, J. (2008). Shearwave elastography. Supersonic Imagine. → pages 9[9] Bercoff, J., Tanter, M., and Fink, M. (2004a). Supersonic shear imaging: a newtechnique for soft tissue elasticity mapping. Ultrasonics, Ferroelectrics, and FrequencyControl, IEEE Transactions on, 51(4):396–409. → pages 973[10] Bercoff, J., Tanter, M., and Fink, M. (2004b). Supersonic shear imaging: a newtechnique for soft tissue elasticity mapping. IEEE transactions on ultrasonics,ferroelectrics, and frequency control, 51(4):396–409. → pages 11[11] Besl, P. J. and McKay, N. D. (1992). Method for registration of 3-d shapes. InRobotics-DL tentative, pages 586–606. International Society for Optics and Photonics.→ pages 56[12] Bota, S., Sporea, I., Sirli, R., Popescu, A., Danila, M., and Sendroiu, M. (2011).Factors that influence the correlation of acoustic radiation force impulse (arfi)elastography with liver fibrosis. Medical ultrasonography, 13(2):135. → pages 10[13] Bravo, A. A., Sheth, S. G., and Chopra, S. (2001). Liver biopsy. New EnglandJournal of Medicine, 344(7):495–500. → pages 6[14] Brendel, B., Winter, S., Rick, A., Stockheim, M., and Ermert, H. (2003). Boneregistration with 3d ct and ultrasound data sets. In International Congress Series,volume 1256, pages 426–432. Elsevier. → pages 57[15] Brunt, E. M. (2000). Grading and staging the histopathological lesions of chronichepatitis: the knodell histology activity index and beyond. Hepatology, 31(1):241–246.→ pages 5, 6, 7[16] Calvaruso, V., Burroughs, A. K., Standish, R., Manousou, P., Grillo, F., Leandro, G.,Maimone, S., Pleguezuelo, M., Xirouchakis, I., Piero Guerrini, G., et al. (2009).Computer-assisted image analysis of liver collagen: Relationship to ishak scoring andhepatic venous pressure gradient. Hepatology, 49(4):1236–1244. → pages 7[17] CarEy, E. and Carey, W. D. (2010). Noninvasive tests for liver disease, fibrosis, andcirrhosis: Is liver biopsy obsolete? Cleveland Clinic journal of medicine,77(8):519–527. → pages 8[18] Castera, L. and Pinzani, M. (2010). Biopsy and non-invasive methods for thediagnosis of liver fibrosis: does it take two to tango? Gut, 59(7):861–866. → pages 7, 9[19] Chen, J., Yin, M., Glaser, K. J., Talwalkar, J. A., and Ehman, R. L. (2013). Mrelastography of liver disease: State of the art. APPLIED RADIOLOGY. → pages 3, 12[20] Clifford, M. A., Banovac, F., Levy, E., and Cleary, K. (2002). Assessment of hepaticmotion secondary to respiration for computer assisted interventions. Computer AidedSurgery, 7(5):291–299. → pages 58[21] Colombo, S., Buonocore, M., Del Poggio, A., Jamoletti, C., Elia, S., Mattiello, M.,Zabbialini, D., and Del Poggio, P. (2012). Head-to-head comparison of transientelastography (te), real-time tissue elastography (rte), and acoustic radiation forceimpulse (arfi) imaging in the diagnosis of liver fibrosis. Journal of gastroenterology,47(4):461–469. → pages 1274[22] De Le´dinghen, V., Vergniol, J., Foucher, J., El-Hajbi, F., Merrouche, W., andRigalleau, V. (2010). Feasibility of liver transient elastography with fibroscan R© using anew probe for obese patients. Liver international, 30(7):1043–1048. → pages 9[23] Eramian, M. G., Adams, G. P., and Pierson, R. A. (2007). Enhancing ultrasoundtexture differences for developing an in vivo’virtual histology’approach to bovineovarian imaging. Reproduction, Fertility and Development, 19(8):910–924. → pages 57[24] Eskandari, H., Goksel, O., Salcudean, S. E., and Rohling, R. (2011). Bandpasssampling of high-frequency tissue motion. Ultrasonics, Ferroelectrics, and FrequencyControl, IEEE Transactions on, 58(7):1332–1343. → pages 23, 24, 25[25] Eulenstein, S., Lange, T., Hu¨nerbein, M., Schlag, P.-M., and Lamecker, H. (2004).Ultrasound-based navigation system incorporating preoperative planning for liversurgery. In International Congress Series, volume 1268, pages 758–763. Elsevier. →pages 56[26] Evans, A., Whelehan, P., Thomson, K., McLean, D., Brauer, K., Purdie, C., Jordan,L., Baker, L., and Thompson, A. (2010). Quantitative shear wave ultrasoundelastography: initial experience in solid breast masses. Breast cancer research, 12(6):1.→ pages 9[27] Friedrich-Rust, M., Wunder, K., Kriener, S., Sotoudeh, F., Richter, S., Bojunga, J.,Herrmann, E., Poynard, T., Dietrich, C. F., Vermehren, J., et al. (2009). Liver fibrosis inviral hepatitis: noninvasive assessment with acoustic radiation force impulse imagingversus transient elastography 1. Radiology, 252(2):595–604. → pages 8, 10[28] Frulio, N. and Trillaud, H. (2013). Ultrasound elastography in liver. Diagnostic andinterventional imaging, 94(5):515–534. → pages 9, 10, 11, 12[29] Garteiser, P., Sahebjavaher, R. S., Ter Beek, L. C., Salcudean, S., Vilgrain, V.,Van Beers, B. E., and Sinkus, R. (2013). Rapid acquisition of multifrequency,multislice and multidirectional mr elastography data with a fractionally encodedgradient echo sequence. NMR in Biomedicine, 26(10):1326–1335. → pages 67[30] Godfrey, E. M., Mannelli, L., Griffin, N., and Lomas, D. J. (2013). Magneticresonance elastography in the diagnosis of hepatic fibrosis. In Seminars in Ultrasound,CT and MRI, volume 34, pages 81–88. Elsevier. → pages 6, 7, 12[31] Howell, W. Mr elastography growing as preferred modality for liver diagnosis. →pages 14[32] Huwart, L., Peeters, F., Sinkus, R., Annet, L., Salameh, N., ter Beek, L. C.,Horsmans, Y., and Van Beers, B. E. (2006). Liver fibrosis: non-invasive assessmentwith mr elastography. NMR in Biomedicine, 19(2):173–179. → pages 7, 12, 1375[33] Huwart, L., Salameh, N., Ter Beek, L., Vicaut, E., Peeters, F., Sinkus, R., andVan Beers, B. E. (2008). Mr elastography of liver fibrosis: preliminary resultscomparing spin-echo and echo-planar imaging. European radiology,18(11):2535–2541. → pages 12, 13[34] Ismail, M. H. and Pinzani, M. (2011). Reversal of hepatic fibrosis:pathophysiological basis of antifibrotic therapies. Hepat Med, 3:69–80. → pages 5[35] Kanamoto, M., Shimada, M., Ikegami, T., Uchiyama, H., Imura, S., Morine, Y.,Kanemura, H., Arakawa, Y., and Nii, A. (2009). Real time elastography for noninvasivediagnosis of liver fibrosis. Journal of hepato-biliary-pancreatic surgery,16(4):463–467. → pages 12[36] Karlas, T., Pfrepper, C., Wiegand, J., Wittekind, C., Neuschulz, M., Mo¨ssner, J.,Berg, T., Tro¨ltzsch, M., and Keim, V. (2011). Acoustic radiation force impulse imaging(arfi) for non-invasive detection of liver fibrosis: examination standards and evaluationof interlobe differences in healthy subjects and chronic liver disease. Scandinavianjournal of gastroenterology, 46(12):1458–1467. → pages 10[37] Keyvanpour, M. R. and Alehojat, S. (1963). Analytical classification of multimodalimage registration based on medical application. Int. J. Adv. Eng. Technol. v1, pages138–147. → pages 55[38] Knutsson, H., Westin, C.-F., and Granlund, G. (1994). Local multiscale frequencyand bandwidth estimation. In Image Processing, 1994. Proceedings. ICIP-94., IEEEInternational Conference, volume 1, pages 36–40. IEEE. → pages 27[39] Lange, T., Eulenstein, S., Hu¨nerbein, M., and Schlag, P.-M. (2003). Vessel-basednon-rigid registration of mr/ct and 3d ultrasound for navigation in liver surgery.Computer Aided Surgery, 8(5):228–240. → pages 56[40] Lange, T., Papenberg, N., Heldmann, S., Modersitzki, J., Fischer, B., Lamecker, H.,and Schlag, P. M. (2009). 3d ultrasound-ct registration of the liver using combinedlandmark-intensity information. International journal of computer assisted radiologyand surgery, 4(1):79–88. → pages 56[41] Lefkowitch, J. H. (2015). Scheuer’s liver biopsy interpretation. Elsevier HealthSciences. → pages 4, 6[42] Lim, Y.-S. and Kim, W. (2008). The global impact of hepatic fibrosis and end-stageliver disease. Clinics in liver disease, 12(4):733–746. → pages 3[43] Manduca, A., Dutt, V., Borup, D. T., Muthupillai, R., Ehman, R. L., and Greenleaf,J. F. (1998). Reconstruction of elasticity and attenuation maps in shear wave imaging:An inverse approach. In Medical Image Computing and Computer-AssistedInterventationMICCAI98, pages 606–613. Springer. → pages 2776[44] Manduca, A., Muthupillai, R., Rossman, P., Greenleaf, J. F., and Ehman, R. L.(1996). Local wavelength estimation for magnetic resonance elastography. In ImageProcessing, 1996. Proceedings., International Conference on, volume 3, pages527–530. IEEE. → pages 20[45] Murtagh, J. and Foerster, V. (2006). Transient elastography (fibroscan) fornon-invasive assessment of liver fibrosis. Issues in emerging health technologies,(90):1–4. → pages 1[46] Nam, W. H., Kang, D.-G., Lee, D., and Ra, J. B. (2010). Robust registration of 3-dultrasound and ct images of the liver for image-guided intervention. In BiomedicalImaging: From Nano to Macro, 2010 IEEE International Symposium on, pages1201–1204. IEEE. → pages 58[47] Nazem, F., Ahmadian, A., Seraj, N. D., and Giti, M. (2014). Two-stage point-basedregistration method between ultrasound and ct imaging of the liver based on icp andunscented kalman filter: a phantom study. International journal of computer assistedradiology and surgery, 9(1):39–48. → pages 57, 58[48] Nightingale, K. (2011). Acoustic radiation force impulse (arfi) imaging: a review.Current medical imaging reviews, 7(4):328. → pages 10[49] Nightingale, K., McAleavey, S., and Trahey, G. (2003). Shear-wave generation usingacoustic radiation force: in vivo and ex vivo results. Ultrasound in medicine & biology,29(12):1715–1723. → pages 29[50] Orchard, J. (2008). Multimodal image registration using floating regressors in thejoint intensity scatter plot. Medical image analysis, 12(4):385–396. → pages 56[51] Penney, G. P., Blackall, J. M., Hamady, M., Sabharwal, T., Adam, A., and Hawkes,D. J. (2004). Registration of freehand 3d ultrasound and magnetic resonance liverimages. Medical image analysis, 8(1):81–91. → pages 56[52] Popescu, A. and Sporea, I. Shearwave elastography (swe). Hepatic ElastographyUsing Ultrasound Waves, page 96. → pages 9, 11[53] Regev, A., Berho, M., Jeffers, L. J., Milikowski, C., Molina, E. G., Pyrsopoulos,N. T., Feng, Z.-Z., Reddy, K. R., and Schiff, E. R. (2002). Sampling error andintraobserver variation in liver biopsy in patients with chronic hcv infection. TheAmerican journal of gastroenterology, 97(10):2614–2618. → pages 6[54] Reinertsen, I., Descoteaux, M., Siddiqi, K., and Collins, D. L. (2007). Validation ofvessel-based registration for correction of brain shift. Medical image analysis,11(4):374–388. → pages 5677[55] Roulot, D., Czernichow, S., Le Cle´siau, H., Costes, J.-L., Vergnaud, A.-C., andBeaugrand, M. (2008). Liver stiffness values in apparently healthy subjects: influenceof gender and metabolic syndrome. Journal of hepatology, 48(4):606–613. → pages 9,25[56] Sandrin, L., Fourquet, B., Hasquenoph, J.-M., Yon, S., Fournier, C., Mal, F.,Christidis, C., Ziol, M., Poulet, B., Kazemi, F., et al. (2003). Transient elastography: anew noninvasive method for assessment of hepatic fibrosis. Ultrasound in medicine &biology, 29(12):1705–1713. → pages 8[57] Schuppan, D. and Afdhal, N. H. (2008). Liver cirrhosis. The Lancet,371(9615):838–851. → pages 2[58] Sherman, M. (2013). Liver disease in canada: A crisis in the making. → pages 1[59] Shiha, G. and Zalata, K. (2011). Ishak Versus METAVIR: Terminology, Convertibilityand Correlation with Laboratory Changes in Chronic Hepatitis C. INTECH OpenAccess Publisher. → pages 7[60] Siriwardena, A. K., Mason, J. M., Mullamitha, S., Hancock, H. C., andJegatheeswaran, S. (2014). Management of colorectal cancer presenting withsynchronous liver metastases. Nature Reviews Clinical Oncology, 11(8):446–459. →pages 4[61] Sirli, R., Bota, S., Sporea, I., Jurchis, A., Popescu, A., Gradinaru-Tasca˘u, O., andSzilaski, M. (2013). Liver stiffness measurements by means of supersonic shearimaging in patients without known liver pathology. Ultrasound in medicine & biology,39(8):1362–1367. → pages 11[62] Song, Y., Totz, J., Thompson, S., Johnsen, S., Barratt, D., Schneider, C., Gurusamy,K., Davidson, B., Ourselin, S., Hawkes, D., et al. (2015). Locally rigid, vessel-basedregistration for laparoscopic liver surgery. International journal of computer assistedradiology and surgery, 10(12):1951–1961. → pages 56[63] Sudharshan Parthasarathy, M., Bolster Jr, B. D., and Miller, F. H. Magneticresonance elastography: Proven indications, challenges and future considerations. liver(elastograms), 23:34–36. → pages 13[64] Suramo, I., Pa¨iva¨nsalo, M., and Myllyla¨, V. (1983). Cranio-caudal movements of theliver, pancreas and kidneys in respiration. Acta radiologica: diagnosis, 25(2):129–131.→ pages 58[65] Tortora, G. J. and Derrickson, B. H. (2008). Principles of anatomy and physiology.Wiley. com. → pages 3, 4[66] Vaughan, R. G., Scott, N. L., and White, D. (1991). The theory of bandpasssampling. Signal Processing, IEEE Transactions on, 39(9):1973–1984. → pages 2378[67] Wang, Q.-B., Zhu, H., Liu, H.-L., and Zhang, B. (2012). Performance of magneticresonance elastography and diffusion-weighted imaging for the staging of hepaticfibrosis: A meta-analysis. Hepatology, 56(1):239–247. → pages 12[68] WHO (2016). Hepatitis b fact sheet. → pages 3[69] Wong, G. L.-H., Wong, V. W.-S., Choi, P. C.-L., Chan, A. W.-H., Chum, R. H.-L.,Chan, H. K.-W., Lau, K. K.-K., Chim, A. M.-L., Yiu, K. K.-L., Chan, F. K.-L., et al.(2008). Assessment of fibrosis by transient elastography compared with liver biopsyand morphometry in chronic liver diseases. Clinical gastroenterology and hepatology,6(9):1027–1035. → pages 6, 7[70] Yin, M., Manduca, A., Romano, A., Glaser, K., Drapaca, C., Lake, D., and Ehman,R. (2007). 3-d local frequency estimation inversion for abdominal mr elastography. InProc Intl Soc Magn Reson Med, volume 15, page 960. → pages 27[71] Zahiri-Azar, R. and Salcudean, S. E. (2006). Motion estimation in ultrasound imagesusing time domain cross correlation with prior estimates. Biomedical Engineering,IEEE Transactions on, 53(10):1990–2000. → pages 18, 45, 4979Appendix ARepeatability ResultsA.1 4DC7 ElastogramsThe following elastograms are the results of the main 4DC7 repeatability experiments. Theyshow a single plane taken from the middle of the first volume acquired. This sweep isrepresentative of what could be seen for the rest of the planes and the rest of the volumes.The two inclusions can be roughly identified at the top of the phantom.80A.1.1 Near-Lateral Excitation Images - Single FrequencyFigure A.1: Single-frequency Elas-togram, Central PlaneFigure A.2: Single-frequency Elas-togram, Central PlaneFigure A.3: Single-frequency Elas-togram, Central PlaneFigure A.4: Single-frequency Elas-togram, Central PlaneFigure A.5: Single-frequency Elas-togram, Central Plane81A.1.2 Near-Elevation Excitation Images - Single FrequencyFigure A.6: Single-frequency Elas-togram, Central PlaneFigure A.7: Single-frequency Elas-togram, Central PlaneFigure A.8: Single-frequency Elas-togram, Central PlaneFigure A.9: Single-frequency Elas-togram, Central PlaneFigure A.10: Single-frequency Elas-togram, Central Plane82A.1.3 Far Excitation Images - Single FrequencyFigure A.11: Single-frequency Elas-togram, Central PlaneFigure A.12: Single-frequency Elas-togram, Central PlaneFigure A.13: Single-frequency Elas-togram, Central PlaneFigure A.14: Single-frequency Elas-togram, Central PlaneFigure A.15: Single-frequency Elas-togram, Central Plane83A.1.4 Near-Lateral Excitation Images - Multi-frequencyFigure A.16: Multi-frequency Elas-togram, Central PlaneFigure A.17: Multi-frequency Elas-togram, Central PlaneFigure A.18: Multi-frequency Elas-togram, Central PlaneFigure A.19: Multi-frequency Elas-togram, Central PlaneFigure A.20: Multi-frequency Elas-togram, Central Plane84A.1.5 Near-Elevation Excitation Images - Multi-frequencyFigure A.21: Multi-frequency Elas-togram, Central PlaneFigure A.22: Multi-frequency Elas-togram, Central PlaneFigure A.23: Multi-frequency Elas-togram, Central PlaneFigure A.24: Multi-frequency Elas-togram, Central PlaneFigure A.25: Multi-frequency Elas-togram, Central Plane85A.1.6 Far Excitation Images - Multi-frequencyFigure A.26: Multi-frequency Elas-togram, Central PlaneFigure A.27: Multi-frequency Elas-togram, Central PlaneFigure A.28: Multi-frequency Elas-togram, Central PlaneFigure A.29: Multi-frequency Elas-togram, Central PlaneFigure A.30: Multi-frequency Elas-togram, Central Plane86A.2 Correlation ResultsThe following correlation results were acquired from correlation data using the eScan soft-ware. The first set of images is an average correlation image for all the frames of a cen-tral plane of a volume. Width and Depth plots are then presented to show the correlationchanges at various acquired depths and across the ultrasound field of view. Finally, plots ofthe correlation at various depths are presented.87A.2.1 Averaged Correlation Images: Near-Lateral ExcitationFigure A.31: Average Correlation forCentral Plane of First VolumeFigure A.32: Average Correlation forCentral Plane of First VolumeFigure A.33: Average Correlation forCentral Plane of First VolumeFigure A.34: Average Correlation forCentral Plane of First VolumeFigure A.35: Average Correlation forCentral Plane of First Volume88A.2.2 Averaged Correlation Images: Near-Elevation ExcitationFigure A.36: Average Correlation forCentral Plane of First VolumeFigure A.37: Average Correlation forCentral Plane of First VolumeFigure A.38: Average Correlation forCentral Plane of First VolumeFigure A.39: Average Correlation forCentral Plane of First VolumeFigure A.40: Average Correlation forCentral Plane of First Volume89A.2.3 Averaged Correlation Images: Far ExcitationFigure A.41: Average Correlation forCentral Plane of First VolumeFigure A.42: Average Correlation forCentral Plane of First VolumeFigure A.43: Average Correlation forCentral Plane of First VolumeFigure A.44: Average Correlation forCentral Plane of First VolumeFigure A.45: Average Correlation forCentral Plane of First Volume90


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