UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

A research platform for ultrasonic elastograpy based targeted prostate biopsy Schiro, Arthur Leland 2013

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

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

Item Metadata


24-ubc_2013_fall_schiro_arthur.pdf [ 6.48MB ]
JSON: 24-1.0071989.json
JSON-LD: 24-1.0071989-ld.json
RDF/XML (Pretty): 24-1.0071989-rdf.xml
RDF/JSON: 24-1.0071989-rdf.json
Turtle: 24-1.0071989-turtle.txt
N-Triples: 24-1.0071989-rdf-ntriples.txt
Original Record: 24-1.0071989-source.json
Full Text

Full Text

A Research Platform for Ultrasonic Elastograpy Based Targeted Prostate Biopsy by Arthur Leland Schiro Bachelor of Science, University of California, Santa Cruz, 2009 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF Master of Applied Science in THE FACULTY OF GRADUATE STUDIES (Electrical and Computer Engineering) The University Of British Columbia (Vancouver) May 2013 c© Arthur Leland Schiro, 2013 Abstract Prostate cancer has been identified as a ubiquitous threat to the well being of North American men living past their fourth decade. An accurate diagnosis, enabling the selection of an appropriate treatment regime, is a key component to disease management. The current gold standard for diagnosis is the transrectal ultrasound guided prostate biopsy procedure, where predefined templates are used to select tis- sue sample sites. Unfortunately, this procedure is incapable of producing reliable results; causing multiple repeat biopsies to become common practice and motivat- ing many men, in the face of uncertainty, to choose unnecessarily sever treatment options with undesirable side effects. Elastography has shown great potential in its ability to detect cancerous tissue, and may enable for the current systematic sample site selection biopsy procedure technique to be replaced by a targeted biopsy approach. This transition promises higher cancer yields and improved diagnostic reliability, while at the same time de- creasing procedural side effects by reducing the number of required core samples. The objective of this thesis has been to integrate the elastography imaging modality into a standard prostate biopsy system. This system may then act as a research platform for determining the feasibility of an elastography based, targeted prostate biopsy procedure. Realization of this objective required the development of four primary compo- nents. First, a tissue excitation mechanism, developed outside of this thesis, was both analyzed for performance capability and augmented in order to improve a design limitation necessitating frequent maintenance. Second, a plastic bracket, enabling an unobtrusive rigid coupling between the excitation mechanism and ul- trasound probe, was designed so that excitation forces may be transmitted into the ii prostate tissue during the standard biopsy procedure. Third, a sensor was designed which is capable of accurately detecting biopsy needle insertion depth by tracking biopsy gun position using an optical absolute position sensor. And fourth, code was developed for enabling an evaluation of system performance by performing elasticity measurements. This code was used to process elastography data, col- lected from phantom and human subjects, in order to obtain a preliminary system validation. iii Preface The text in this thesis is original and has not been included in any previous pub- lications. An initial design for the shaker device described in Chapter 3 was pro- vided by Dr. Hani Eskandari. Data for the prostate biopsy experiment described in Chapter 6 was acquired under the permission of UBC Ethics Certificate number H11-01125. The graphical user interface used to orchestrate the prostate biopsy procedure was designed and created by Siavash Khallaghi, Samira Sojoudi, and Saman Nouranian. The motion tracking and least squares strain estimation code used to perform the data processing sequence described in Chapter 6 was provided by Dr. Reza Zahiri Azar. All additional contributions to this project were made by myself under the supervision of my adviser, Dr. Tim Salcudean, and my co- supervisor, Dr. Purang Abolmaesumi. iv Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.3.1 Imaging Technique Selection . . . . . . . . . . . . . . . . 4 1.3.2 System Components . . . . . . . . . . . . . . . . . . . . 5 1.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.5 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.1 Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.2 Prostate Cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3 Detection and Diagnosis . . . . . . . . . . . . . . . . . . . . . . 11 2.3.1 Disease Symptoms . . . . . . . . . . . . . . . . . . . . . 11 2.3.2 Screening Tests . . . . . . . . . . . . . . . . . . . . . . . 12 v 2.3.3 Epidemiological Indicators . . . . . . . . . . . . . . . . . 12 2.4 The Current Gold Standard for Prostate Biopsy . . . . . . . . . . 13 2.4.1 Transrectal Ultrasound Guidance . . . . . . . . . . . . . . 13 2.4.2 Anatomical Tumor Distributions . . . . . . . . . . . . . . 14 2.4.3 Tissue Extraction Mechanism . . . . . . . . . . . . . . . 16 2.4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.5 Tissue Sampling Techniques . . . . . . . . . . . . . . . . . . . . 17 2.6 Advanced Ultrasound Imaging Techniques . . . . . . . . . . . . . 19 2.6.1 Standard Ultrasound Imaging . . . . . . . . . . . . . . . 19 2.6.2 Imaging With Tissue Scatterer Characteristics . . . . . . . 20 2.6.3 Imaging From Tissue Vascularity . . . . . . . . . . . . . 22 2.7 Imaging From Tissue Mechanical Characteristics . . . . . . . . . 22 2.7.1 Tissue Excitation . . . . . . . . . . . . . . . . . . . . . . 22 2.7.2 Motion Estimation . . . . . . . . . . . . . . . . . . . . . 23 Doppler Based Methods . . . . . . . . . . . . . 23 Signature Tracking Methods . . . . . . . . . . 23 2.7.3 Parameter Estimation . . . . . . . . . . . . . . . . . . . . 27 Strain Imaging . . . . . . . . . . . . . . . . . . 27 Relative Parameter Estimation . . . . . . . . . 28 Absolute Parameter Estimation . . . . . . . . . 29 2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3 Materials and Methods: Shaker Development . . . . . . . . . . . . . 32 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.2 Functional Overview . . . . . . . . . . . . . . . . . . . . . . . . 33 3.2.1 Shaker Operation . . . . . . . . . . . . . . . . . . . . . . 33 3.2.2 Control Hardware . . . . . . . . . . . . . . . . . . . . . . 34 3.3 Shaker Performance Evaluation . . . . . . . . . . . . . . . . . . . 35 3.3.1 Analytical Model . . . . . . . . . . . . . . . . . . . . . . 35 Model Creation and Parameter Identification . . 36 Parameter Estimation . . . . . . . . . . . . . . 38 3.3.2 Quantitative Experiment . . . . . . . . . . . . . . . . . . 42 3.3.3 Model Validation . . . . . . . . . . . . . . . . . . . . . . 45 vi 3.3.4 Evaluation of Displacement Results . . . . . . . . . . . . 46 3.4 Flexure Modification . . . . . . . . . . . . . . . . . . . . . . . . 49 3.4.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . 49 3.4.2 The Fix . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 3.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 4 Materials and Methods: Sleeve Development . . . . . . . . . . . . . 55 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.2 Design Constraints . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.3 Fabrication Techniques . . . . . . . . . . . . . . . . . . . . . . . 57 4.4 Design Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . 57 4.4.1 Clamping Force Mechanism . . . . . . . . . . . . . . . . 58 4.4.2 Contact Point Selection . . . . . . . . . . . . . . . . . . . 59 4.5 Final Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 4.6 Core Sample Depth Sensor Feature . . . . . . . . . . . . . . . . . 65 4.7 Modifications for Clinical Trial . . . . . . . . . . . . . . . . . . . 68 5 Materials and Methods: Core Sample Depth Sensor Development . 70 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 5.2 Design Constraints . . . . . . . . . . . . . . . . . . . . . . . . . 72 5.3 System Description . . . . . . . . . . . . . . . . . . . . . . . . . 73 5.4 Optical Address Pattern Design . . . . . . . . . . . . . . . . . . . 75 5.4.1 Resolution Limitations . . . . . . . . . . . . . . . . . . . 75 5.4.2 Improving Resolution . . . . . . . . . . . . . . . . . . . 76 5.4.3 Implemented Optical Pattern Configuration . . . . . . . . 78 5.5 Implementation Details . . . . . . . . . . . . . . . . . . . . . . . 80 5.5.1 Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . 80 5.5.2 Hardware and Software Design . . . . . . . . . . . . . . 81 5.6 Error Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 6 Results: System Performance Evaluation . . . . . . . . . . . . . . . 85 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 6.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . 85 6.2.1 Specimen Configuration . . . . . . . . . . . . . . . . . . 85 vii 6.2.2 Hardware Setup . . . . . . . . . . . . . . . . . . . . . . . 86 6.2.3 Data Set Components . . . . . . . . . . . . . . . . . . . . 87 6.3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 6.3.1 Speckle Tracking Block . . . . . . . . . . . . . . . . . . 89 6.3.2 Strain Estimation Block . . . . . . . . . . . . . . . . . . 90 6.3.3 Spectral Analysis Block . . . . . . . . . . . . . . . . . . 91 6.3.4 Wave Velocity and Elasticity Estimation Block . . . . . . 92 6.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 6.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 6.5.1 Figures 6.4 and 6.8: B-mode, Point Strain and Probe Mo- tion Images . . . . . . . . . . . . . . . . . . . . . . . . . 102 6.5.2 Figures 6.5 and 6.9: Strain Phasor Amplitude and SNR Images . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 6.5.3 Figures 6.6 and 6.10: Real Strain Value Images . . . . . . 103 6.5.4 Figures 6.11 and 6.11: Elasticity Computation Images . . 104 6.5.5 System Performance Conclusions . . . . . . . . . . . . . 105 7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 7.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 7.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 viii List of Figures Figure 1.1 System setup. . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Figure 2.1 Prostate gland and local anatomical structures. . . . . . . . . 10 Figure 2.2 Configuration of standard biopsy procedure. . . . . . . . . . . 14 Figure 2.3 Zonal prostate anatomy . . . . . . . . . . . . . . . . . . . . . 15 Figure 2.4 B-mode image creation . . . . . . . . . . . . . . . . . . . . . 21 Figure 2.5 Segments of RF lines taken from consecutive frames. . . . . . 25 Figure 3.1 Shaker components. . . . . . . . . . . . . . . . . . . . . . . 33 Figure 3.2 Shaker control box. . . . . . . . . . . . . . . . . . . . . . . . 33 Figure 3.3 Production of Lorentz force. . . . . . . . . . . . . . . . . . . 34 Figure 3.4 Shaker control box components. . . . . . . . . . . . . . . . . 35 Figure 3.5 Analytical model subsystems. . . . . . . . . . . . . . . . . . 36 Figure 3.6 Electrical model of shaker and cable assembly. . . . . . . . . 37 Figure 3.7 Dynamic model for the analytical mechanical subsystem. . . . 38 Figure 3.8 Shaker mechanical model in the Laplace domain. . . . . . . . 38 Figure 3.9 Shaker step response. . . . . . . . . . . . . . . . . . . . . . . 41 Figure 3.10 Experimental configurations for shaker model verification. . . 43 Figure 3.11 Displacement results for experimental procedure. . . . . . . . 44 Figure 3.12 Analytical model displacement predictions. . . . . . . . . . . 46 Figure 3.13 Strain decay. . . . . . . . . . . . . . . . . . . . . . . . . . . 48 Figure 3.14 Increased shaker displacement threshold. . . . . . . . . . . . 49 Figure 3.15 Flexure functionality. . . . . . . . . . . . . . . . . . . . . . . 50 Figure 3.16 Second flexure redesign attempt. . . . . . . . . . . . . . . . . 51 Figure 3.17 Modified flexure implementation. . . . . . . . . . . . . . . . 53 ix Figure 3.18 Modified flexure mount. . . . . . . . . . . . . . . . . . . . . 53 Figure 3.19 Displacement results for modified flexure design. . . . . . . . 54 Figure 3.20 Experimental setup for modified flexure design test. . . . . . . 54 Figure 4.1 startingSystem. . . . . . . . . . . . . . . . . . . . . . . . . . 56 Figure 4.2 Sleeve time-lapse development. . . . . . . . . . . . . . . . . 58 Figure 4.3 Initial clamping force mechanism. . . . . . . . . . . . . . . . 59 Figure 4.4 Kelvin clamp. . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Figure 4.5 Probe constraint features. . . . . . . . . . . . . . . . . . . . . 61 Figure 4.6 Final sleeve design. . . . . . . . . . . . . . . . . . . . . . . . 62 Figure 4.7 Sleeve clamping force. . . . . . . . . . . . . . . . . . . . . . 63 Figure 4.8 Lower sleeve constraints. . . . . . . . . . . . . . . . . . . . . 63 Figure 4.9 Upper sleeve constraints. . . . . . . . . . . . . . . . . . . . . 64 Figure 4.10 Biopsy gun alterations. . . . . . . . . . . . . . . . . . . . . . 66 Figure 4.11 CSDS circuitry and mounting mechanism. . . . . . . . . . . . 67 Figure 4.12 NTPS circuitry protection. . . . . . . . . . . . . . . . . . . . 67 Figure 4.13 probeSleeveRedesign . . . . . . . . . . . . . . . . . . . . . . 69 Figure 4.14 shakerMagnetMount . . . . . . . . . . . . . . . . . . . . . . 69 Figure 5.1 Needle path constraints. . . . . . . . . . . . . . . . . . . . . 71 Figure 5.2 Biopsy needle appearance in B-mode image. . . . . . . . . . 72 Figure 5.3 Position sensor system components. . . . . . . . . . . . . . . 74 Figure 5.4 Reflective sensor. . . . . . . . . . . . . . . . . . . . . . . . . 74 Figure 5.5 Reflective sensor optical isolation. . . . . . . . . . . . . . . . 75 Figure 5.6 Initial sensor configuration. . . . . . . . . . . . . . . . . . . . 76 Figure 5.7 Quadrature encoder configuration. . . . . . . . . . . . . . . . 77 Figure 5.8 Alternative quadrature configuration. . . . . . . . . . . . . . 77 Figure 5.9 Symbolic representation of quadrature configuration. . . . . . 78 Figure 5.10 Symbolic tri-phase tracking representation. . . . . . . . . . . 78 Figure 5.11 Optical pattern and sensor array detail. . . . . . . . . . . . . . 79 Figure 5.12 PCB Gerber file output. . . . . . . . . . . . . . . . . . . . . . 80 Figure 5.13 Reflective sensor circuit. . . . . . . . . . . . . . . . . . . . . 82 Figure 5.14 Error measurement experimental setup. . . . . . . . . . . . . 83 x Figure 5.15 Error measurement experimental results . . . . . . . . . . . . 83 Figure 6.1 Phantom experimental setup. . . . . . . . . . . . . . . . . . . 86 Figure 6.2 Shaker driver voltage source. . . . . . . . . . . . . . . . . . . 88 Figure 6.3 Data processing and parameter estimation steps. . . . . . . . . 90 Figure 6.4 Phantom experiment operating conditions. . . . . . . . . . . . 94 Figure 6.5 Phantom experiment strain phasor amplitude and SNR values. 95 Figure 6.6 Phantom experiment strain phasor real tissue values. . . . . . 96 Figure 6.7 Phantom experiment elasticity computation images. . . . . . . 97 Figure 6.8 Patient experiment operating conditions. . . . . . . . . . . . . 98 Figure 6.9 Patient experiment strain phasor amplitude and SNR values. . 99 Figure 6.10 Patient experiment strain phasor real tissue values. . . . . . . 100 Figure 6.11 Patient experiment elasticity computation images. . . . . . . . 101 xi Glossary 1D one dimensional 3D three dimensional ADC analog to digital conversion BPH Benign Prostatic Hyperplasia CAD computer aided design CNB Core Needle Biopsy CNC computer numerical control CSDS Core Sample Depth Sensor DC zero cycles per second DOF degree of freedom DFT Discrete Fourier transform DRE Digital Rectal Exam DTFT Discrete-time Fourier transform FE Finite Element FDM fused deposition modeling FNA Fine Needle Aspiration xii GUI graphical user interface LED light emitting diode LFE Local Frequency Estimation LVR linear voltage regulator MRI Magnetic Resonance Imaging NDA non-disclosure agreement PC personal computer PCa Prostate Cancer PCB printed circuit board PSA Prostate Specific Antigen RF Radio Frequency ROI region of interest SNR Signal to Noise Ratio SOS Second Order System TRUS Transrectal Ultrasound UART universal asynchronous receiver/transmitter USB universal serial bus VCP virtual communication port VGH Vancouver General Hospital xiii Acknowledgments Firstly, I would like to thank my thesis supervisor, Dr. Tim Salcudean, and co- supervisor, Dr. Purang Abolmaesumi, for providing valuable advice and support throughout the development of this thesis. I would also like to thank Dr. Hani Eskandari for his guidance in the many areas of engineering design necessitated for completion of this project and for contributing many of the system compo- nents used by this thesis. I would like to thank Dr. Reza Zahiri Azar for provid- ing well written speckle tracking and strain estimation code. A special thanks to the machine shop gurus, Mark Finniss and Donald Dawson, who are a source of invaluable mechanical design and fabrication knowledge. I would like to thank Siavash Khallaghi and Saman and Samira Nouranian for their tremendous code development contributions which enabled the system produced by this thesis to be integrated into the clinical setting. I would also like to thank my fellow lab mates for their friendship and inspiration. And finally, I would like to thank my family, Michael, Mettie and Stephanie for contributing the core elements of that indestructible existential concept called home, which is an active element of my life; wherever and in whatever state I have settled. xiv Chapter 1 Introduction 1.1 Motivation Prostate Cancer (PCa) is the most commonly diagnosed non-cutaneous cancer in American men and the second most deadly, with 238,590 new cases and 29,720 deaths estimated in 2013 [101]. The disease exhibits a broad array of clinical be- havior and, consequentially, an equally broad array of treatment options. Virulently malignant prostate tumors may require treatment routes such as radical prostatec- tomy or radiation treatment, which can be accompanied by undesirable side effects imparting a great toll on a patient’s quality of life. Whereas, active surveillance, where tumor growth is simply monitored by periodic testing, may be the best treat- ment route for prostate tumors with relatively benign behavior. Misdiagnoses of the more malignant prostate tumors could result in an un- der treatment with fatal consequences, whereas misdiagnoses of the more benign prostate tumors could result in an over treatment with an undesirable set of treat- ment side effects. Therefore, the initial and pivotal challenge faced by urologists is to obtain an accurate diagnosis so that an appropriate treatment regime may be chosen. The only way to accurately diagnose and predict prostate cancer malignancy is to examine tissue samples obtained from the cancerous tumors. A range of tech- niques exist for obtaining prostate tissue samples, however the majority of samples are obtained with a Transrectal Ultrasound (TRUS) guided prostate biopsy proce- 1 dure. Ideally, standard ultrasound imaging techniques would enable the identifica- tion of specific prostate locations containing suspicious cell growth so that these locations could be targeted by a biopsy device. Unfortunately, this technique is not a feasible solution, since the standard B-mode ultrasound imaging modality fails to detect up to 40% of prostate tumors of appreciable size [86]. The low cancer detection rate of standard ultrasound imaging techniques has lead researchers to develop systematic sampling practices, where tissue is obtained from predefined anatomical locations. TRUS guided prostate biopsies using sys- tematic sampling techniques have become the current gold standard for prostate biopsy procedures; however, the multifocal nature of prostate cancer [27] compli- cates the task of adequate tumor detection and results in a disconcerting number of false negative diagnoses, with detection rates averaging around 40% [44, 86]. The cancer yield of systematic sampling procedures may be increased by in- creasing the number of anatomical sample sites. However, this increased sensitiv- ity is accompanied by the undesirable consequence of increased patient morbidity. Additionally, PCa sensitivity reaches a plateau of about 58% for 18 core extended biopsy patterns [22]. Much research effort has been devoted to determining the most ideal location and quantity of sample sites employed by the systematic sampling prostate biopsy technique. However, a promising alternative to this research direction may involve returning to the targeted biopsy approach, aided by an advanced imaging technique capable of reliably detecting cancerous lesions. The benefits of this approach in- clude increased cancer yield, enabling improved detection rates, and decreased sample counts, resulting in lowered patient morbidity. 1.2 Objective A number of techniques have been proposed for the purpose of targeted prostate biopsy guidance. In particular, imaging techniques labeled under the category of ‘elastography’ have been successfully applied to a number of clinical trials where they demonstrated promising results. In a study with 404 individuals, targeted prostate biopsy under the guidance of elastography imaging was found to have a 2 cancer sensitivity of 84.1% [47]. Another study found it to be 2.9 times more ef- fective at detecting PCa than the standard systematic sampling biopsy scheme [74]. A number of other studies have been conducted, suggesting an average detection rate of around 80% for this technique [68, 74, 75, 78, 92]. The primary objective of this thesis is to integrate the necessary elastography imaging system components into a standard prostate biopsy system so that elastog- raphy imaging data may be collected during a prostate biopsy procedure. We intend to use this system for a subsequent study which will compare the elastography im- ages collected over a set of clinical trials to the histological results for prostate tissue samples obtained during those trials. The trials will be orchestrated so that the tissue sample sites will correspond to regions in the elastography images. This correspondence will enable the degree of correlation between elastography imag- ing parameters and prostate tissue malignancy to be determined. If elastography images prove capable of identifying prospectively cancerous prostate regions then this system may be employed as an elastography image guidance based targeted prostate biopsy system. 1.3 System Overview Elastography, described in detail in Section 2.7, is an imaging technique where tissue mechanical characteristics are used to create images. Creating such images involves three steps. First, an exciter mechanism is used to induce tissue motion. Second, an imaging modality, such as ultrasound or Magnetic Resonance Imaging (MRI), is used to measure tissue motion. And third, information on tissue motion, boundary conditions and excitation signal features are processed by an algorithm that estimates various parameters of the exited tissue, such as tissue elasticity or viscosity, and creates images from these parameters. Elastography imaging techniques may be divided into the three main categories of: strain imaging, relative parameter estimation and absolute parameter estima- tion. Each imaging technique has a different set of requirements for induced tissue motion. Therefore, before the mechanical components of an elastography system may be implemented, one must make some decisions about which imaging tech- niques the system will be compatible with. The following sections describe, first, 3 the imaging technique implemented by our system, and second, the system com- ponents required to realize this system. 1.3.1 Imaging Technique Selection Of the three imaging technique categories previously listed, strain imaging tech- niques are the least computationally expensive and require the simplest mechanical implementations. Mechanical simplicity comes from the fact that tissue excitation may be produced by using one’s hand to manually apply a pseudo-static force to the imaging probe, thereby requiring no physical alterations to the prostate biopsy system. Simplified mechanical system and computer hardware implementations may explain the reliance on strain imaging as the primary technique used for previous elastography clinical trials [68, 74, 75, 78, 92]. However, this imaging technique has two primary drawbacks. First, strain imaging relies on an assumption of tissue stress uniformity, which would be invalidated by our choice to induce tissue motion by applying a force to an imaging probe with a small cross sectional area and by the inhomogeneous nature of the tissue under investigation [73]. And second, the low frequency freehand tissue excitation implemented by previous strain imaging systems [68, 74, 75, 78, 92] is operator dependent, requiring a learning period which could affect clinical trial results. In light of these factors, we have chosen to implement a system capable of en- abling an absolute elastography imaging technique [7, 8, 62]. Absolute imaging techniques are less dependent on tissue stress distribution assumptions and are im- plemented with operator independent hardware controlled tissue excitation mech- anisms. Hardware controlled tissue excitation and advanced parameter estimation algorithms come at the cost of increased mechanical and computational system complexity, but may potentially enable higher cancer detection rates than those, already progressively optimistic results, provided by strain imaging based systems. The absolute imaging technique intended to be implemented by this system will estimate tissue elastic parameters by inducing steady state sinusoidal tissue motion for a variety of excitation frequencies [7]. Tissue excitation energy propa- gates through the tissue as longitudinally polarized shear waves with spatial wave- 4 lengths proportional to both excitation temporal frequency and tissue elastic pa- rameters. Tissue elastic parameters may then be estimated by measuring the local spatial wavelengths of the known excitation frequencies. Therefore, the primary system components required for this form of elastography include a tissue excita- tion mechanism capable of inducing dynamic tissue motion and a means of cou- pling this excitation energy into the imaged tissue. 1.3.2 System Components Figure 1.1 shows the components of our system and their interconnections. The ultrasound machine, TRUS probe, needle guide and biopsy gun components are the primary components of any current prostate biopsy system and may be viewed as an initial starting point for this thesis. The rest of the components in this figure were integrated into the initial system in order to realize the overall thesis objective, by enabling the collection of elastography imaging data during a standard prostate biopsy procedure. The tissue excitation mechanism, labeled ‘shaker’ in Figure 1.1, is coupled to the TRUS probe via a rigid connection to an intermediary device labeled as the ‘sleeve’, which utilizes minimum constraint design to achieve a repeatable zero degree of freedom (DOF) coupling with the probe. The shaker device is concealed within the rear sleeve compartment and is driven by signals from the ‘shaker control box’. The shaker control box is connected to the ultrasound machine personal computer (PC) via a universal serial bus (USB) connection and is controlled by a graphical user interface (GUI) software application running on the PC. The GUI software application orchestrates elastography data collection and prostate biopsy sampling during prostate biopsy procedures and receives commands from system operators via a USB foot pedal. An additional component, labeled as the ‘Core Sample Depth Sensor (CSDS)’, measures biopsy needle insertion depth. This sensor is mounted to the sleeve and reports needle depth information the ultrasound machine PC via a USB connection. Needle insertion depth information is used to determine the imaging plane regions from which the biopsy samples were obtained. 5 Figure 1.1: System setup and component connections. 1.4 Contributions The following contributions were made in order to realize the primary objective of this thesis. 1. An initial shaker system, developed outside this thesis, was integrated into our system by performing two steps. First, an analytical model for the shaker system was developed so that we could determine its theoretical impulse re- sponse. This model would enable us to predict if the shaker was capable of producing sufficient tissue motion. We could also use this model to de- termine alterations to shaker component parameters which would result in 6 more favorable operation. The development and testing of this model are described in Section 3.3. The second shaker integration step required the development of an alternate solution for the mechanism constraining shaker internal motion. This devel- opment step was necessary because the previous solution was not sufficiently robust and required frequent maintenance. 2. The sleeve device was engineered, using minimum constraint design, so that the TRUS probe could be robustly, yet temporarily, attached to the system in such a way that it would obtain a relatively identical orientation after multiple attachment-detachment cycles. The sleeve was also designed for compatibility with our in-house stereolithography three dimensional (3D) printing machines and these machines were used to print multiple prototype revisions. 3. The CSDS device was designed so that it could report biopsy needle inser- tion depth information back to applications running on the ultrasound PC. Needle depth information was determined by tracking biopsy gun position using an absolute position sensor. The position sensor was implemented by mounting an optical array to the sleeve so that it could read an optically en- coded address pattern attached to the biopsy gun. The design and testing of this sensor are described in Chapter 5. 4. Code was developed for enabling an evaluation of system performance by performing spectral analysis on tissue strain measurements and using the re- sultant information to perform course elasticity measurements. This code was used to process elastography data, collected from phantom and human subjects, in order to obtain a preliminary system validation. The develop- ment of this code and a discussion on system performance are covered in Chapter 6. 1.5 Thesis Outline The remaining chapters of this thesis are structured as follows. Chapter 2 provides a comprehensive overview of prostate cancer, the TRUS guided prostate biopsy 7 procedure and advanced ultrasound imaging modalities used for tissue classifica- tion and identification. Chapters 3, 4 and 5 provide detailed accounts of the devel- opment of the shaker, the sleeve and the CSDS, respectively. Chapter 6 describes the series of experiments used to evaluate system performance. Finally, Chapter 7 concludes this thesis by summarizing the covered topics and suggesting improve- ments for future system revisions. 8 Chapter 2 Background 2.1 Anatomy The prostate is a tubuloalveolar gland of the male reproductive system located inferior to the urinary bladder and anterior to the rectum. It encompasses the upper portion of the urethra, and the ejaculatory ducts and is shown in relation to nearby structures of the lower abdomen in Figure 2.1. Its glandular excretions consist of a milky fluid that enhances sperm motility, constitutes approximately 30% of the volume of semen and is necessary for sexual reproduction. The prostate gland is small during childhood, weighing about 2 g. It undergoes a dramatic, hormone driven growth during puberty, reaching a mature weight of approximately 20g and approximate volume of a chestnut [40]. 2.2 Prostate Cancer The prostate gland is a dependable source of trouble for men, especially those liv- ing past their fourth decade. It is estimated that nearly 15% of men will suffer from prostatitis at one point in their lives [50] and that 50% of men between 51 and 60 years of age will endure symptomatic Benign Prostatic Hyperplasia (BPH). These diseases may impart an undesirable degree of discomfort but pale in comparison with the severity of their pathological colleague, PCa, which is currently ranked as the third leading cause of non-cutanious cancer deaths in Canadian men [72] and 9 Figure 2.1: Prostate gland and local anatomical structures. (Marieb, Elaine N.; Mallatt, Jon B.; Wilhelm, Patricia Brady, Human Anatomy, 5th Edi- tion, c©2008. Reprinted with permission of Pearson Education, Inc., Upper Saddle River, NJ.) the second most deadly in American men [99]. In 2012 there were an estimated 241,740 new cases diagnosed and 28,170 deaths in American men [99], and 26,500 new cases diagnosed and 4,000 deaths in Canadian men [72]. As daunting as this disease may seem, an early stage diagnosis may have a broad range of prognoses, for the disease exhibits clinical behavior ranging from biologically benign, small, slow growing tumors, which would have caused no symptoms in the lifetime of their host had they never been detected, to virulently malignant tumors capable of metastasizing to the bone, lymph nodes and other organs, with little chance of successful treatment. The broad spectrum of PCa malignancy has resulted in an equally broad se- lection of treatment options, which range from watchful waiting for the more be- nign tumors, to radiation treatments or prostatectomy for the more malignant, pre- metastatic tumors, to hormone therapy treatments recommended for post metastatic carcinoma. In order to choose the treatment option which best matches a patient’s tu- 10 mor malignancy, urologists must obtain an accurate measurement of PCa severity. Luckily, the diverse clinical manifestations of PCa are accompanied by histological differences in the microscopic appearance of the cancerous tissue. In 1966 Gleason proposed a tumor grading scale based on an observation of the significant correlation between PCa tissue histological appearances and degree of biological malignancy [2, 37]. This grading system is used by today’s practi- tioners to predict the aggressiveness of cancerous tissue and thereby recommend an appropriate treatment strategy. Unfortunately, the multi-focal nature of PCa [25] complicates the task of obtaining tumor samples which adequately represent glandular tumor distributions. The resulting uncertainty may result in unnecessary over treatments, with undesirable morbidity and decreased quality of life [11], or under treatments, enabling stage progression of the disease, thereby decreasing the chances of survival [54]. The occult nature of PCa is therefore one of the greatest challenges faced by the opponents of this disease. 2.3 Detection and Diagnosis Accurate and early diagnosis of PCa are essential components to successful man- agement of the disease. The only way to make this diagnosis is by obtaining prostate tissue samples which contain histological evidence of PCa. Understand- ably, the level of discomfort and side effects associated with this procedure delays its application until indicators of the disease have been found. These indicators include identification of disease symptoms, abnormal results from screening tests and suggestive epidemiological correlations and will be discussed in the following paragraphs. 2.3.1 Disease Symptoms PCa is commonly asymptomatic in its earlier stages due to its likely development in the posterior portion of the prostate, away from the prostatic urethra where it could cause urinary track complications [56]. Symptoms, therefore, usually indi- cate advanced or metastatic disease and include, in addition to urinary track com- plications, edema in the lower extremities, pain in the hips, thighs, or ribs and weight loss [56]. 11 2.3.2 Screening Tests The Digital Rectal Exam (DRE) procedure is a slight improvement over allowing symptoms to appear before suspecting the presence of PCa since it is capable of detecting cancer before it metastasizes beyond the prostatic capsule. In this test, the practitioner palpates the posterior prostate through the rectum in an attempt to de- tect any morphological indicators of the disease. The drawback to this test is that it is limited to the detection of tumors in the posterior portion of the prostate, thereby missing approximately 40% of anteriorly located tumors [59]. Additionally, this test has low sensitivity to lesions less than 1.5cm and therefore has a limited ability to result in a favorable prognosis [57]. Introduced in 1986, the Prostate Specific Antigen (PSA) test has been respon- sible for a dramatic increase in the diagnosis of PCa [119]. In the PSA test, a blood sample is analyzed in an attempt to determine the amount of PSA in the blood. PSA is a protein produced in healthy prostate tissue which may be released into the blood stream in increased amounts in the presence of PCa [105]. The useful- ness of this test is limited, however, since increased PSA levels may also be caused by physical trauma to the prostate, or the presence of BPH or prostatitis, resulting in 60-80% false positive PSA test results [105]. In addition to the false positive results, one study found that, among men with normal PSA levels, who were also suspected of PCa, 19% were found to have cancer [102]. 2.3.3 Epidemiological Indicators In conjunction with the DRE procedure and the PSA test, the application of epi- demiological observations enables our medical community to focus its efforts in a more productive manner. For example, 97% of prostate cancer cases occur in men over the age of 50 [99]; and as a result, the American Cancer Society recommends that practitioners not consider PCa to be a concern for the average American male until he reaches his fifth decade of life [100]. Other factors contributing to the probability of an individual developing PCa include ethnicity, family history and diet [20]. If the compounded evidence from an individual’s DRE procedure, PSA test and epidemiological background are indicative of PCa, practitioners will recom- 12 mend that a prostate tissue sample be obtained for the purpose of identifying the presence of carcinoma and grading its corresponding malignancy. Diagnosis and treatment options will be made based on this evaluation. The precise methods used to obtain these tissue samples have undergone considerable evolution time and will be discussed in the following section. 2.4 The Current Gold Standard for Prostate Biopsy As previously mentioned, a PCa diagnosis requires a tissue sample analysis. The predominant method for obtaining these samples is by performing a prostate biopsy procedure. The gold standard for the current prostate biopsy operation has become the TRUS guided procedure where a spring-driven Core Needle Biopsy (CNB) de- vice is used to obtain tissue samples from a combination of systematic and targeted anatomical locations. This procedure, shown in Figure 2.2, has obtained its current form as a result of developments in three primary areas: TRUS image guidance, knowledge of anatomical tumor distributions, and tissue extraction mechanisms. These three areas of development will be discussed in further detail in the follow- ing paragraphs. 2.4.1 Transrectal Ultrasound Guidance The first area of development leading to the current gold standard for prostate biopsy procedures consists of advances in TRUS imaging of the prostate and sur- rounding anatomy. This technology was pioneered in Japan in the late 1960’s, with the first English language publication of the ground breaking work by Watanabe in 1971 [118]. These early systems consisted of chair mounted, low frequency (3.5 MHz) rotational array transducers, producing black and white images with low spatial resolution and limited diagnostic potential [83]. Significant develop- ments which enabled this technology to step into its current role include both the introduction of high frequency systems (5 MHz to 10 MHz), that produce images with higher axial resolution, and the development of handheld, contact enabled, end firing probes that allow practitioners to easily maneuver to any imaging plane and to integrate a biopsy needle guidance mechanism onto the transducer shaft [55, 84]. 13 Figure 2.2: Configuration of TRUS guided biopsy procedure. (Reprinted with permission from MayoClinic.com http://www.mayoclinic.com/health/medical/IM02640) 2.4.2 Anatomical Tumor Distributions The second area of development contributing to the current day biopsy procedure consists of McNeal’s definition of prostate anatomy in terms of histological and functionally unique zones, in the late 1960s [65, 85]. This development is signif- icant due to its accompaniment by a statistical mapping of tumor distributions in relation to the newly defined anatomy [66]. This information provides practitioners with a better understanding of where to look for malignant prostatic neoplasms and a means of developing improved sampling practices. McNeal’s model divides the prostate into four zones: three glandular zones, shown in Figure 2.3, and one nonglandular zone [65]. The peripheral zone con- stitutes about 70% of prostate glandular mass, forming the posterior portion of the prostate, and is located anterior to the rectum. The central zone constitutes 14 about 25% of prostate glandular mass and is located superior to the peripheral zone and posterior to the prostatic urethra. The transition zone constitutes about 5% of prostate glandular mass and is located proximal to the bladder neck, strad- dling the upper prostatic urethra. The fibromuscular stroma is the one nonglandular zone, and it occupies the anterior portion of the prostate and is responsible for its convex appearance. Approximately 70% of prostate cancers originate in the pe- ripheral zone, 20% in the transition zone and 10% in the central zone [66]. The fibromuscular stroma is not susceptible to PCa due to its nonglandular nature [55]. Figure 2.3: Zonal prostate anatomy depicting the peripheral zone (PZ), cen- tral zone (CZ) and transition zone (TZ). (Image reproduced from [117], with permission from Saunders Publishing and Baylor College of Medicine) With the zonal mapping of cancer distribution in hand, Hodge developed a TRUS guided, systematic, anatomically defined template for selecting biopsy sam- ple sites and demonstrated its significant contribution to techniques which had pre- viously limited their sample sites to those areas displaying suspicious echogenic- ity [42]. Named the sextant pattern, three laterally directed biopsies, equally dis- tributed between the base and apex of the prostate, were taken bilaterally from the 15 posterior portion of the prostate. In the late 1980s, Hodge went on to demonstrate the superiority of the combined systematic and targeted sampling method over the previously dominant digitally guided method, establishing it as the predominant sampling regime and predecessor of the range of sampling practices used today [41]. 2.4.3 Tissue Extraction Mechanism The third area of development contributing to the current day biopsy procedure relates to the physical mechanism used to extract tissue from the prostate. Nee- dle extraction of tissue for diagnostic purposes was first documented in 1847 [53], as discussed by Frable [32], and has since then diverged into the two primary ar- eas of Fine Needle Aspiration (FNA) and CNB [24]. These methods differ in the mechanism used for specimen extraction, the nature of the collected specimen, the process used to prepare the specimen for examination and the practice employed for evaluating the prepared tissue. The FNA method was introduced in 1926 [63] and first applied to the prostate in 1930 [31] via the transperineal rout, as referenced in Frable [33]. FNA was popularized in 1960 when Franzen introduced a mechanism for enabling digital guidance through the transrectal route [35]. In FNA biopsy, a thin, hollow needle fixed to the end of a syringe is inserted into the tissue and a vacuum, created by drawing on the syringe plunger, causes cells to move into the needle bore. In order to collect sufficient material for diagnosis it is necessary for the needle to be moved back and forth multiple times, along the same track, while maintaining the vacuum pressure. The disordinate mass of collected tissue is then placed on a glass slide and smeared, pressed and dried in preparation for investigation by a cytologist. CNB was introduced with the Silverman type needle, in 1938 [96], as discussed by Crile and Vickery [21], and was first used to perform prostate biopsy in 1943 [77]. CNB consists of a three step process. First, a stylet, specially designed for the purpose of isolating a small segment of tissue, is inserted into a tissue mass. Second, a cannula is slid over the stylet, thereby securing the isolated tissue sample within. And, finally, the two bodies are simultaneously extracted from the body, producing a coherent tissue sample. The sample is divided into microscopic slices 16 and died in preparation for a histological evaluation by a pathologist. The pros and cons attributed to FNA and CNB presented a challenging state of equilibrium for much of the 20th century [32]. FNA held an advantage because the high gauge needles used in this procedure (22-gauge and higher [33]) enabled the procedure to be carried out with local or no anesthesia, whereas the larger 14-gauge CNB needles [24] required general anesthesia, access to an operating room and hospital recovery time [26]. On the other hand, CNB had an advantage over FNA due to the loss of diagnostically significant histological tissue characteristics when preparing FNA samples [24], and the high dependency of FNA diagnosis quality on the adeptness of the individuals performing the clinical procedure, specimen preparation, and cytological evaluation [49]. The FNA and CNB debate was resolved with the introduction of three signif- icant developments. First, the finer, 18-gauge tru cut CNB needle, first appearing in literature in 1974 [123], enabled a less traumatic CNB procedure [82]. Second, Gleason’s 1966 introduction of a reproducible and accurate grading scale for histo- logical evaluation of PCa tumor malignancy resulted in a preference of histological over cytological evaluation of tissue. And third, the 1986 introduction of a spring- driven, tru cut needle biopsy gun enabled the CNB procedure to be performed with greatly reduced trauma on an out-patient basis [82]. 2.4.4 Summary In summary, the current gold standard for the prostate biopsy procedure was es- tablished in the late 1980’s through developments in TRUS imaging technology, knowledge of prostate anatomy, and tissue extraction mechanisms. Since then, developments in the practice of prostate biopsy have consisted of a series of dis- heartening findings on the inability of standard biopsy protocol to accurately map clinically significant tumors of the prostate and our attempts to remedy this situa- tion. 2.5 Tissue Sampling Techniques The primary challenge faced by a prostate biopsy system is the multi-focal nature of PCa, where up to 45% of clinically significant tumors are under 1 cm3 [27]. With 17 the introduction of TRUS imaging, practitioners were enthusiastically optimistic for the capability of B-mode ultrasound imaging to detect these scattered tumors, thereby enabling effective targeted sampling. However, it was soon discovered that approximately 40% of malignant lesions of significant volume were not detectable in B-mode ultrasound images [86, 95]. In addition to this, the percentage of le- sions which posses both suspicious echogenicity and are also malignant has been found to be as low as 25% [85]. These findings motivated a migration towards the systematic sextant sampling template mentioned in Section 2.4.2. However, the in- troduction of the PSA test, enabling detection of PCa in less malignant forms and in earlier stages of progression, necessitated biopsy system sensitivities beyond that which could be provided by the sextant pattern. Extended biopsy schemes were introduced in an attempt to increase cancer yield. These schemes consist of the standard sextant pattern plus additional ante- riorly directed biopsy sites [107], resulting in a total of 8 to 18 core samples, and are the current standard for initial biopsy [100]. Extended biopsy procedures are an improvement over the sextant pattern, detecting 20% to 31% more malignant tumors than the sextant pattern [80, 107]. Saturation biopsy schemes, consisting of up to 45 samples, were introduced in 2001 for men with persisting indicators of PCa after initial negative biopsies [106]. These protocols attempt to sample regions which are not addressed with sextant or extended biopsy templates. One study found a 34% cancer detection rate in men undergoing saturation repeat biopsy after a previous negative sextant biopsy [106]. Autopsy studies have shown that systematic sampling protocols result in can- cer detection rates that increase with core count, starting at approximately 30% for sextant patterns and reaching a plateau of approximately 58% for 18 core extended biopsy patterns [22]. The undesirably low biopsy sensitivity shown in such studies, in addition to the increased cost and morbidity associated with today’s high sample counts, has generated great interest in developing methods capable of improving detection rates while decreasing overall core counts. These methods involve appli- cation of advanced imaging modes for effectively targeting specific lesions. 18 2.6 Advanced Ultrasound Imaging Techniques As previously mentioned, standard ultrasound imaging techniques are insufficient for use in targeted biopsy systems. Advanced ultrasound imaging techniques at- tempt to improve cancer detention by measuring tissue parameters which demon- strate pathological correlations associated with the three main categories: tissue scatterer characteristics, tissue vascularity and tissue mechanical characteristics. The standard ultrasound imaging technique employed during a prostate biopsy session will be described in the following section, followed by descriptions of the imaging techniques reliant on tissue scatterer characteristics, and tissue vascularity. Imaging techniques based on tissue mechanical characteristics will be discussed in greater detail in Section 2.7, as these techniques apply directly to the system pro- posed in this thesis. 2.6.1 Standard Ultrasound Imaging Ultrasound images are created by extracting information from the recorded echoes of ultrasonic pulse injected into tissue by transducer arrays held against tissue sur- faces. Echoes are created when the injected pulses encounter boundaries between materials of different acoustic properties. Echo amplitude and phase are deter- mined by the relationship between the acoustic properties of these boundary ma- terials [17]. The tissue depth associated with each temporal echo signal sample is determined by the time delay between echo pulse transmission and sample acqui- sition time, in addition to prior knowledge ultrasonic signal wave speed [17]. High transmit pulse carrier frequencies, between 5 MHz and 10 MHz, enable sub-millimeter axial resolution and result in Gaussian shaped system transfer func- tions centered at these carrier frequencies [70]. Recorded ultrasonic echo signals may be represented by a convolution between these system transfer functions and tissue echogenicity signals, which have the acoustic appearance of a field of ran- domly distributed, closely spaced, varying amplitude reflector impulses. Conse- quently, ultrasound system recordings are referred to as Radio Frequency (RF) signals and closely resemble the result of a high frequency amplitude modulation, with spectral energy concentration around the system transfer function center, or ‘carrier,’ frequency. Images used in clinical settings, referred to as brightness mode 19 or ‘B-mode’ images, provide information on anatomical relations and tissue acous- tic properties. These images are created by sampling the demodulated RF signal envelopes. The process of creating a B-mode image from an RF echo signal is illustrated in Figure 2.4. 2.6.2 Imaging With Tissue Scatterer Characteristics Where the noisy, speckled pattern inherent in B-mode images, like the one found in Figure 2.4a, may be seen as an impediment to image interpretation, imaging tech- niques based on tissue scatterer characteristics see this signal feature as convey- ing valuable information on tissue microarchitecture [60]. Since the noisy signal appearance is caused by acoustically reflective scatterers with histologically corre- lated semiperiodic distributions, scatterer distribution information contained within ultrasonic echo signals may be used to identify histologic abnormalities indicative of malignant tissue [60]. Texture analysis techniques include the determination of statistical parameters, such as mean, standard deviation and kurtosis, from envelope signal grey level histograms [43]. Texture parameters may also be derived from a second-order histogram analysis using the cooccurrence matrix [39, 81]. Alternatively, spectral analysis techniques may use the RF signal phase or mag- nitude spectrums to determine tissue scatterer size, concentrations and acoustic impedances in an attempt to identify pathological signatures [61, 111]. RF signal analysis advantageously enables ultrasound system transfer function decorrelation from echo signal spectra, thereby removing system dependent artifacts and extend- ing the usable signal bandwidth. In clinical application, systems that utilize texture analysis may implement a large collection of tissue classifiers, utilizing parameters from both texture and spectral techniques, to predict tissue pathology [108]. One such study used a Kohonen-map classifier with Mahalanobis-distance measure on a set of 10 param- eters to identify cancerous regions in a group of 33 patients with local prostatic cancer, yielding 82% sensitivity and 88% specificity [93]. 20 (a) (b) (c) Figure 2.4: A typical B-mode image from a TRUS prostate exam is shown in (a). Image (b) shows the RF (blue) and envelope (red) signals used to create the single B-mode image line within the highlighted section of (a). The anatomy corresponding to each echo segment are labeled in (b) and color coded in (b) and (a). Image (c) shows a close-up of the prostate echo signals. 21 2.6.3 Imaging From Tissue Vascularity The tendency for malignant neoplasms to develop capillary neovascularization has motivated attempts to utilize tissue vascularity for identification of prostate cancer [10]. Tissue vascularity may be detected using color or power doppler techniques with or without contrast enhancing agents for detecting abnormal blood flow in areas with increased vascularity [109]. A 2007 study on 690 patients with ele- vated PSA level found that contrast enhanced color doppler targeted biopsy de- tected cancer in 2% more patients than the standard 10 core extended systematic biopsy scheme [67]. 2.7 Imaging From Tissue Mechanical Characteristics Imaging techniques which investigate and display tissue mechanical characteris- tics, such as elasticity and viscosity, are generally referred to as elastography. Mo- tivation for these techniques may be understood as an intuitive extension to the driving principles behind the DRE procedure, which attributes abnormal, digitally excited, tissue motion to the presence of cancer. The significance of elasticity in the detection of cancer may be seen in the results of Krouskop, Wheeler, Kallel, Garra, and Hall [52], where ex-vivo prostate tissue, compressed at 2 and 4% and excited between 0.1 Hz and 4 Hz, was found to have an elastic modulus between 55 and 71 kPa, for normal tissue, and a significantly different elastic modulus, be- tween 96 kPa and 240 kPa, for cancerous tissue. The significance of viscosity as an indicator of cancerous breast tissue has also been shown in the research of Sinkus et al. [98]. Creating elastography images involves three steps: first, the tissue is excited, second, tissue motion is measured, and third, the motion measurements are processed to estimate various tissue mechanical parameters. These steps will be discussed in the following sections. 2.7.1 Tissue Excitation Induced tissue motion may be quasi-static [73], transient [18] or dynamic [58]. Tissue excitation may be implemented by exciting the transducer itself or by a me- chanical, pneumatic [122] or hydraulic [90] coupling of exciter motion to nearby tissue. Exciter mechanisms may consist of piezoelectric bimorphs [120], mechani- 22 cally converted rotational motion from electric motors [91], manual freehand com- pression [75], guided sound waves from speakers [6] or linear motion electromag- netic devices [3]. 2.7.2 Motion Estimation Doppler Based Methods Initially, tissue motion estimation systems were implemented using the readily available pulsed doppler imaging hardware for measuring dynamically excited tis- sue velocity amplitude [51]. By vibrating tissue with a single sinusoid, echo signal spectrums took the form of a frequency modulated pure tone, exhibiting harmon- ics centered at the carrier frequency, spaced by integer multiples of the vibration frequency, with a Bessel function envelope linearly parameterized by the tissue vibration amplitude [58]. Tissue vibration amplitude was determined by segment- ing the RF signal into many overlapping windows and then estimating the Bessel envelope parameterization within each window [121]. Signature Tracking Methods In contrast to doppler based methods, which require single echo signal acquisitions for motion detection, the majority of current motion estimation systems use the differences between sequentially captured echo signals to track tissue signatures between frames. Where doppler methods are limited to detecting motion in only the axial direction (along the scan line), with minimum vibration frequencies in the hundreds of cycles per second, the tracking based methods advantageously enable motion estimation in both axial and azimuthal (perpendicular to the scan line) directions, with minimum frequency detection approaching zero cycles per second (DC) and maximum frequency detection limited only by the system frame rate [116]. Tracking based methods may be divided into the three main categories of: fea- ture tracking methods, phase measurement methods and cross-correlation methods. These methods all rely on the concept that, as tissue is stretched or compressed along the axial direction, the echo signal returned from that tissue is stretched or 23 compressed in a similar manner. Therefore, if an echo signal peak from an ini- tial frame, plotted against depth, appears at depth d1 and a deformation causes the tissue at d1 to move to depth d2, in a subsequent frame acquisition, then the echo signal peak at d1 in the initial plot will appear at d2 in a plot of the corresponding echo signal from the subsequent frame. Feature Tracking Methods Feature tracking methods attempt to estimate tissue motion by a one to one map- ping of signal features between frames. The following features have been pro- posed: hardware detected zero crossings [103], wavelet transform extracted RF signal peaks [30] and individual sample amplitudes mapped from the samples in one RF frame to continuous interpolations of the signals a subsequent frame [5]. The drawback of using these methods is that tissue deformation between frames must be small. The advantages of these methods include a low computational cost, a high immunity to local compressions and a dense output displacement map. Cross Correlation Methods Cross-correlation and phase measurement based tissue tracking methods operate under the assumption that windowed sections of precompressed frames appear as time scaled and shifted versions of themselves in postcompression frames: xpost = xpre(a · t+ t0) (2.1) Additionally, since the windows are small, the time scaling is ignored so that the relationship becomes: xpost = xpre(t+ t0) (2.2) An intuitive appreciation for this assumption may be obtained by considering the pre- and postcompression signals plotted in Figure 2.5. Under this assumption, tissue tracking is performed by dividing precompression frames into many overlap- ping windows, which, due to tissue deformation, will appear at displaced locations in subsequent postcompression frames. Phase measurement and cross-correlation based methods differ in how they determine window displacement between frames. 24 Figure 2.5: Segments of RF lines taken from consecutive frames. Data ob- tained from a human prostate under dynamic excitation. Cross-correlation methods determine window displacement by finding the peak of the cross-correlation function operating on the windowed precompression signal and the prospective window displacement path in the postcompression frame [73]. The computational cost of this operation may be reduced by using neighboring window displacement estimations to decrease search region size [4] or by using the sum of absolute differences operation in place of the cross-correlation sum of products operation [12]. Subsample displacements are found by using cosine in- terpolation [15] or a parabolic fit [13] to the discrete cross-correlation function for locating its peak. Alternatively, tissue displacements may be determined analyti- cally by providing a continuous-time spline approximation of the postcompression signal, along with the precompression window samples, to a pattern matching func- tion [115]. Decorrelation errors due to large tissue displacements may be reduced by logarithmic compression [16] or temporal stretching with global [114] or local [1] time scaling. Phase Measurement Methods Phase measurement techniques use the in phase and quadrature components of 25 either RF or envelope signals to create their analytic counterparts, whose phase relationships are used to estimate window time shifts [70]. These analytic signals are modeled as: x̂pre(t) = A(t)e jwmt , x̂post(t) = A(t− τ)e jwm(t−τ), (2.3) where wm is the ultrasound system transfer function spectral centroid, A(t) is the real RF signal envelope and τ is the window time delay. Complex conjugate multiplication of these signals yields their phase difference, at every point in time, which is in linear proportion to the window time delay between pre and postcompression frames and the center frequency wm: x̂(t) = x̂∗post(t)x̂pre(t) = A(t)A(t− τ)e jwmτ , ϕ(t) = angle{x̂(t)}= wmτ. (2.4) The time delay for each N-length window is found by applying the complex inner product for averaging the phase differences of each discrete time sample: τ = angle{ N ∑ n=1 x̂∗post [n]x̂pre[n]} wm . (2.5) By enabling time delay estimation with only one sum of products operation, these techniques enable a computationally inexpensive approach to speckle track- ing. However, two primary disadvantages present themselves: first, the approx- imation, that dϕdτ = wm, results in decreased time delay estimation accuracy, and second, due to the nature of phase measurements, tissue displacement amplitudes must be small to prevent aliasing errors. Delay estimation accuracy limitations may be addressed, at an increased com- putational cost, by using initial delay estimations to improve window alignment before performing subsequent phase calculations [79]. 26 Aliasing errors result from the implicit ϕ(t) (mod 2pi) behavior of phase mea- surements, which limits the detection of tissue displacements to amplitudes result- ing in phase differences of less than ±pi . This corresponds to tissue displacement amplitudes of ±λ/2, where λ = 2pi/wm, for RF analytic signal analysis, and am- plitudes of±λ/4, for baseband analytic signal analysis, as a result of the nonlinear demodulation process. Phase unwrapping methods enable detection of greater tissue displacement am- plitudes, without accumulating aliasing errors. Shiina et al. [94] calculated cross- correlation coefficients, in addition to phase measurements, for a set of ±pi phase regions; unaliased phase measurements were then selected from regions yielding the largest cross-correlation coefficients. Pesavento et al. [79] used the time shift of preceding windows to determine the unaliased ±pi phase region. Alternatively, phase calculation may be performed in the Fourier domain, where a time delay appears as a linear phase shift whose slope may be averaged across the transform spectrum, enabling the amplitude of unaliased displacement measurements to be limited by an adjustable window size [69]. 2.7.3 Parameter Estimation A number of parameter estimation methods have been developed, each one re- quiring a specific form of tissue excitation and relying on a specific set of tissue mechanical characteristic, boundary conditions or displacement field symmetry as- sumptions. The majority of these techniques model tissue as a linear, isotropic, incompressible medium, which enables a great simplification of elastic medium modeling parameters [29]. However, since the linearity assumption has been shown to be reasonable only for small strains, exciter mechanisms are limited to induc- ing tissue displacement amplitudes around 100 µm [97]. The collection of these Elasticity techniques diverge past these assumptions and will be explored over the continuation of this section. Strain Imaging With the additional assumption of stress uniformity and axial symmetry, images displaying tissue strain measurements, through the application of Hook’s law, may 27 be interpreted as portraying the arithmetic inverse of the relative elastic modulus. This assumption holds true when an infinitely large compressor is applied to an in- finite, homogeneous medium [73]. However, since this is not the case in practice, stress decay, stress concentrations and the generally complex stress distributions in 3D elastic materials lead to image artifacts which limit the applicability of such im- ages [23]. Nevertheless, application of strain image guidance for targeted prostate biopsy has shown great potential in clinical trials. In a study of 404 individuals, this technique demonstrated a cancer sensitivity of 84.1% [47]. Strain image guid- ance was also found to be 2.9 times more effective at detecting PCa than systematic ultrasound guided biopsy [74]. Tissue strain may be estimated by finding the gradient of tissue displacements between pre- and post-compression frames [73]. However, the noise sensitive gra- dient operation requires application of techniques such as median filtering [87] and multicompression averaging [113] to produce reliable images. Alternatively, the noisy displacement derivatives may be avoided by considering the linear rela- tionship between tissue strain and spectral shifts resulting from the compression of acoustic reflectors. Methods for computing these shifts include comparison of spectral centroid locations [48] and frequency domain cross-correlation of pre- and post-compression spectrums [112]. Additionally, direct strain estimation may be performed by iteratively searching for stretch factors which maximize the cross- correlation between pre- and post-compression RF lines [1]. Relative Parameter Estimation Improved relative parameter estimation may be achieved by methods that do not rely on the assumption of stress uniformity. These methods use tissue models, which define the interdependence of tissue units, so that unknown excitation pa- rameters, such as force or stress, may be driven to the model boundaries. In the research of Turgay et al. [110], Newton’s second law is applied to a dis- crete one dimensional (1D) tissue model, consisting of a series of mass, spring, damper elements in a Voigt configuration. By accumulating measurements over a sequence of time intervals, the resultant linear system of equations becomes over constrained and may be solved in the usual least squares sense, where the un- 28 known force boundary conditions are overcome by assuming the value of the first stiffness element whereby all parameters become relative to this assumed value. In the research of Eskandari et al. [28], the necessity for accurate modeling assump- tions is bypassed by representing tissue as a series of transfer functions whose low frequency asymptotes and phase slope were used to obtain robust estimations for relative elastic modulus and the higher order tissue parameter, relaxation time. Absolute Parameter Estimation Finite Element Modeling Methods Absolute viscoelastic parameters may be estimated by the inverse problem ap- proach where Finite Element (FE) methods are used to solve the wave equation thereby creating forward models capable of predicting tissue displacements for given parameter distributions [46]. Various iterative non-linear optimization meth- ods may be employed for estimating the model parameters which minimize a cost function that measures the similarity between measured and predicted tissue dis- placements [23]. Alternatively, the forward problem may be solved so that a linear relationship between model parameters and tissue displacements enables direct in- version [76]. Wave Velocity Estimation Methods Absolute viscoelastic parameters may also be estimated by exciting tissue with a slow moving shear wave of known temporal frequency and then measuring the resultant spatial frequency or attenuation [14]. Analytical solutions to the wave equation may then be used to relate these measurements to desired tissue param- eters [7]. Alternatively, a direct inversion could be carried out, once spatial fre- quency is determined, by application of Equation 2.6 where E represents Young’s modulus, or elasticity, ρ represents tissue density, which is approximately equal to the density of water, and c represents the wave speed of the longitudinally polarized sheer wave [7]. E = c2ρ (2.6) 29 A number of techniques have been developed for measuring shear wave spa- tial frequency. Local Frequency Estimation (LFE) achieves this task by combining the outputs from a bank of logarithmically separated Gaussian filters, operating on windowed displacement data [62]. Other techniques use the direct algebraic in- version of Helmholtz wave equation to predict local spatial frequency [14]. This process is summarized by Equation 2.7, where U stands for the temporal Fourier transform of spatial displacements at various tissue locations, ∇2 is the Laplacian operator and k is the local angular spatial frequency. Problems associated with find- ing second order derivatives of the noisy displacement data have been addressed by first applying a linear filter to the noisy data and then taking derivatives on poly- nomials fit to that data [71]. A third approach to finding local spatial frequency involves using an iterative, non-linear optimization approach to fit a parameterized model to the data and then finding local spatial frequency by taking derivatives on the analytical model rather than the noisy measurement data [8]. ∇2U =−k2U =⇒ k2 =−∇ 2U U (2.7) 2.8 Summary In summary, the prostate is a lime sized gland, central to the male urogenital sys- tem, which has a tendency to develop malignant lesions in men living past their fourth decade. Evidence of PCa may be detected by performing a DRE test, a PSA test or by observations of suspicious epidemiological indicators. An accurate diagnosis is necessary for the selection of an appropriate treatment plan and re- quires the acquisition of prostate tissue containing malignant growth. The standard means of obtaining tissue samples consists of a TRUS guided prostate biopsy pro- cedure using a combination of systematic and B-mode targeted sampling, however, this system is incapable of providing disease diagnosis with sufficient reliability. Consequently, various advanced imaging techniques have been presented for the purpose of enabling an improved targeted biopsy procedure. Elastography is a promising imaging technique which creates images based on tissue mechanical properties. Tissue mechanical properties are obtained by processing tissue motion information which is extracted from RF ultrasound frames obtained during tissue 30 excitation. Tissue excitation is, therefore, and important component of any elastography system. The excitation mechanism employed by our system consists of an elec- tromechanical device creating linear proof mass displacements through the interac- tion of fixed and alternating magnetic fields and will be discussed in the following chapter. 31 Chapter 3 Materials and Methods: Shaker Development 3.1 Introduction As described in Section 1.3.2, the shaker is a software controlled excitation mech- anism used to induce dynamic tissue motion, thereby enabling the elastography imaging modality, with the shaker box providing the shaker drive signal, in accor- dance with commands sent from the ultrasound machine PC. Both the shaker and the shaker box were designed outside of this thesis and are shown in their initial form in Figure 3.1 and Figure 3.2, respectively. The shaker box was incorporated into the system without any changes made; however, two primary design steps were necessary before the shaker device could be integrated. The first shaker integration step involved an investigation into the range of tissue motion achievable by the shaker. This investigation, discussed in Section 3.3, involved both a series of experiments where shaker induced probe motion was measured and the development of an analytical model, which could act as a tool to guide the augmentation of shaker mechanical parameters in order to achieve the desired range of functionality. The second shaker integration step involved a modification to the mechanical solution used to enable shaker linear motion. This step was necessary because the initial design was not sufficiently robust and required frequent maintenance. This 32 step is discussed in Section 3.4. A functional overview of the shaker and shaker box is provided in the follow- ing section so that these operating principals may be referenced in the subsequent sections. Figure 3.1: Shaker components. Figure 3.2: Shaker control box. 3.2 Functional Overview 3.2.1 Shaker Operation The shaker consists of two main bodies: a voice coil and a proof mass, shown labeled in Figure 3.1. Magnets attached to the proof mass induce a Lorentz force 33 on the current carrying wire contained within the voice coil. The thin metal flexures shown in Figure 3.1 elastically couple the two bodies so that their relative motion is roughly linear and coaxial with the Lorentz force. Magnet polarity and field as well as current density and the resultant Lorentz force are depicted in Figure 3.3. Application of the Lorentz force governing equa- tion (Force = Current×Magnetic f ield), using the right hand rule, may be used to find the relationship between the current density, magnetic field, and forces de- picted in this figure. Figure 3.3: Production of Lorentz force. 3.2.2 Control Hardware The shaker control box is shown, with removed lid and labeled components, in Fig- ure 3.4. A USB connection with the ultrasound machine PC enables applications to send driving waveforms to the control box sound card. The sound card produces an analog signal which is buffered by the audio amplifier before transmission to the shaker voice coil. An accelerometer has been placed on the voice coil so that probe motion may be monitored during operation. The accelerometer controller board, shown in Fig- 34 ure 3.4, provides power to the accelerometer and conditions the returned analog accelerometer output signal so that it may be passed to the sound card microphone, digitized and sent to the ultrasound machine PC over the USB connection. Figure 3.4: Shaker control box components. 3.3 Shaker Performance Evaluation This section describes an investigation into the range of tissue motion achievable by the shaker. First, the derivation of an analytical model for the system will be discussed. Next, a series of quantitative experiments, used to determine shaker in- duced probe motion, will be presented. This will be followed by a validation of the analytical model by comparing model results to quantitative experimental results. And finally, a discussion on shaker motion capabilities and system requirements will be provided. 3.3.1 Analytical Model Presentation of the analytical model development will be broken into two parts. Section describes the mathematical structure of the analytical model and 35 identifies its parameters. Section describes the procedures used to estimate the value of these parameters. Model Creation and Parameter Identification Since a voltage source is used to drive the system, our model will consist of a trans- fer function that takes a voltage signal as its input and provides a probe displace- ment signal as its output. This system can be broken into three main components, as shown in Figure 3.5. The total transfer function, T (s), may then be represented as the product of these three subsystems. Figure 3.5: Analytical model subsystems. The subsystem, E(s), represents the lumped parameter electrical model of the shaker and cable assembly. The input to this model is a driving voltage provided by the shaker control box and the output is the shaker voice coil current which interacts with the proof mass magnets to produce the Lorentz force. The electrical model for E(s) is shown in Figure 3.6. The desired voice coil current is represented by Im and the current through the capacitive leg, which does not contribute to the Lorentz force is represented by Ic. This transfer function for this model is shown in Equation 3.1. E(s) = Im(s) V (s) = 1 Ls+R (3.1) The subsystem, L(s), shown in Figure 3.5, converts voice coil current to Lorentz force and has the transfer function shown in Equation 3.2. We have made the simplifying assumption that the voice coil’s force producing currents experience a 36 Figure 3.6: Electrical model of shaker and cable assembly. magnetic field that does not vary with variations in the voice coil’s location. We have also made the simplifying assumption that the magnetic field produced by the permanent magnets is always perpendicular to the voice coil current loop surface. With these assumptions, the transfer function, L(s), becomes simply a constant gain. L(s) = G (3.2) The subsystem, M(s), shown in relation to the total system transfer function in Figure 3.5, represents the mechanical system shown in Figure 1.1 and consists of the shaker, sleeve, TRUS probe, biopsy gun and tissue mimicking phantom. The model does not include a system operator’s hand, which will be responsible for holding the entire assembly against the phantom. This decision was made by ob- serving that experimentally measured TRUS probe displacements were relatively immune to variations in grip strength and arm rigidity. The dynamic model for this mechanical subsystem can be found in Figure 3.7. The phantom tissue is modeled as the spring and dashpot with parameters kh and bh. The voice coil, sleeve, TRUS probe and biopsy gun are modeled by the mass labeled mp. The flexures responsible for coupling the proof mass to the voice coil are modeled with the parameters k f and b f . And the proof mass is modeled by the mass ma. When transformed into the Laplace domain the system shown in Figure 3.7 can be redrawn as shown in Figure 3.8a, where displacement, x, represents probe motion. The parameters may be combined and a simplified system can be drawn, as shown in Figure 3.8b. The transfer function from the input force to output TRUS 37 Figure 3.7: Dynamic model for the analytical mechanical subsystem. probe displacement is shown in Equation 3.3. (a) Full model. (b) Simplified model. Figure 3.8: Shaker mechanical model in the Laplace domain. M(s)= X(s) F(s) = c ab+ac+bc where a= kh+bhS+mpS2 , b= k f +b f S and c=maS2 (3.3) Parameter Estimation L, C, R, ma and mp Electrical model parameters L, C, and R were found by applying an LCR meter to the end of the cable assembly and are listed in Table 3.1. A scale was used to find 38 the mechanical model mass parameter values, ma and mp, listed in Table 3.1. G The constant gain value that defines subsystem L(s), which converts current to Lorentz force, as described by Equation 3.2, was determined by the following pro- cedure. A DC current of 325 mA was applied to the shaker-cable assembly. The displacement that resulted between the proof mass and the voice coil was measured and found to have a value of 1 mm. The Lorentz force responsible for this displace- ment was found to be approximately 3.1N and was calculated by multiplying the measured displacement by the flexure spring constant, determined experimentally to be approximately 3100 Nm . The value for G was found by dividing the Lorentz force by the applied current and is listed in Table 3.1. kh The spring constant of the phantom tissue was determined by measuring the depth that the TRUS probe tip depressed the phantom tissue when a known mass was ap- plied to the probe tip. The applied mass was 432 g, which applies a force of 4.24 N, and resulted in a TRUS probe displacement of about 6 mm. The spring constant was calculated by dividing the applied force by the measured displacement and is listed in Table 3.1. bh The damping coefficient at the phantom-tissue/probe-tip interface was calculated by modeling the behavior of this interaction as a mass (m) spring (k) dashpot (b) Second Order System (SOS) with transfer function denominator polynomial shown in equation 3.4. However, this polynomial is commonly rewritten, as shown in equation 3.5, where ς represents the system damping ratio and wn is the natural frequency. The equivalence of these two polynomials enables the derivation of Equation 3.6. The damping ratio, bh, was found by substituting typical muscle tissue values for damping ratio, ς = 0.44, and natural frequency, wn = 20 Hz, into equation 3.6, along with the phantom spring constant value previously determined to be Kh = 706 Nm . The resultant value may be found in Table 3.1. 39 DEN(s) = s2+ b m s+ k m (3.4) DEN(s) = s2+2ςwns+w2n (3.5) b = 2ςk wn (3.6) b f An equation for determining the voice-coil/proof-mass coupling damping coeffi- cient was found by combining the common definition for SOS settling time, ts, shown in equation 3.7 [34], with equation eqn:shakerModeldenomequiv, derived from the equivalence of equation 3.4 and 3.5, and is shown in equation 3.9. The settling time was found to be approximately 500 ms and was determined by taking measurements from a recording of the voice coil accelerometer output, shown in Figure 3.9, for a step input. The value for b f is listed in Table 3.1. ts = 4.6 ςwn (3.7) 2ςwn = b m (3.8) b f = 9.2m ts (3.9) k f Initially, the flexure spring constant was analytically derived by modeling the two flexures as series cantilever beams. Equation 3.10, as presented by Beer et al. [9], was used to calculate the spring constant for each flexure, where E is the modulus of elasticity for yellow brass, 105 GPa, I is the moment of inertia of the flexure, L is flexure length, 4.5 mm, b is flexure width, 19.5 mm, and h is flexure height, 76 µm. The resultant spring constant had a value of 5 Nm . 40 Figure 3.9: Shaker step response. k = 3EI L3 where I = bh3 12 (3.10) With the intention of confirming the accuracy of our flexure spring model, the spring constant was determined experimentally by measuring the displacement be- tween the proof mass and the voice coil that resulted when a known force was applied to the proof mass while the voice coil was immobilized in a clamp. The spring constant was then calculated by dividing the applied force by the resultant displacement. This experiment yielded a spring constant value of approximately 3000 Nm , which was nearly three orders of magnitude larger than the value deter- mined analytically. A second experiment was performed to provide resolution for the situation presented by these conflicting measurements. For this experiment, the proof mass was immobilized and the output of an accelerometer, attached to the voice coil, was recorded while the voice coil experienced a manually induced step displacement input. The accelerometer output, shown in Figure 3.10, indicates that the system is significantly under damped; which means that the natural frequency could be approximated by measuring the damped frequency present in the captured signal. The spring constant was then calculated using the relationship: k = mw2n, which can be derived for a simple second order, spring, mass, dashpot system. The voice coil was weighed, thereby determining that m = 30.2 g. The parameter wn was determined, from observations of the plot in Figure 3.10, to be 52 Hz. Then the spring constant was calculated and found to have the value of 3200 Nm , which is 41 L C R ma mp k f b f G kh bh 2.4 5.5 10 78.6 570 3100 0.55 9.5 706 31 mH uF ohm g g Nm Ns m N A N m Ns m Table 3.1: Shaker model parameter values. very close to the result of the first experiment, thereby confirming the accuracy of the experimental results. The value for the spring constant used in our model was the average of the two values determined experimentally and is listed in Table 3.1. The flexure spring constant analytical calculation error indicates the degree to which the flexures do not behave as the cantilever beams they were intended to implement. Instead, they behave more similarly to an over constrained beam operating beyond its yield point. This is one of the reasons that the flexures break so frequently and, require the modification discussed in Section 3.4. 3.3.2 Quantitative Experiment An experiment was conducted to determine the system displacement frequency response under a range of conditions. Matlab code was generated to drive the shaker with 2V-peak sinusoidal signals and to record the output of the voice coil accelerometer. The accelerometer output was converted to displacement amplitude by finding the signal spectral energy about the driving signal frequency and then converting this value to acceleration units which could be multiplied by the square of the driving signal radian frequency. Displacement frequency response was recorded for a variety of experimental conditions in order to determine the extent to which variations in certain system components altered the overall system performance. These variations included the presence or absence of the biopsy gun, the operator’s hand, and tissue phantom, as well as variations in the operators probe grip strength and level of static force ap- plied to the tissue phantom. Therefore, for each experimental setup that involved manually holding the probe against the tissue phantom, separate measurements were found for high and mild grip strength and high and mild static applied phan- tom force. The various system experimental configurations are shown in Figure 3.10. System displacement frequency response is shown in Figure 3.11. 42 (a) Experimental configuration ‘A’ (b) Experimental configuration ‘B’ (c) Experimental configuration ‘C’ (d) Experimental configuration ‘D’ (e) Experimental configuration ‘E’ Figure 3.10: Experimental configurations for shaker analytical model verifi- cation. 43 Figure 3.11: Displacement results for experimental procedures. 44 3.3.3 Model Validation Validation of the analytical model was achieved by comparing the theoretical probe displacement frequency response data to the experimentally collected data. A Mat- lab script was written to generate a theoretical displacement frequency response for the analytical model developed in Section 3.3.1. This response, driven by sinusoids of 2V-peak amplitude, is shown in Figure 3.12 and labeled as ‘Analytical model’ in the legend. Displacement data from experimental configuration ‘D’, Figure 3.10, is also plotted on this figure and is labeled ‘Experimental data’ in the legend. The analytical and experimental data follow a similar trend, however, three primary discrepancies between the two data sets indicate errors in model parameter estimations. First, the two data sets exhibit different system damping levels; which may be seen by comparing the large analytical model resonance peak, indicating an under damped system, to the experimental data results, which appear to be only slightly under damped. Second, the analytical system indicates a natural frequency of approximately 35 Hz, whereas the experimental data shows a natural frequency of about 25 Hz. And third, the analytical system appears to have a slightly higher gain than the system indicated by the experimental data. The system damping level discrepancy may be due to an underestimation of the shaker damping coefficient, b f . The discrepancy in system natural frequency may be caused by an overestimation of shaker spring constant, k f . Finally, the difference between analytical and experimental system gain values may be due to estimation errors in the coil current to Lorentz force conversion constant, G, where our simplifying assumptions may have lead us to an overestimation. The parameter values indicated by the row labeled ‘Adjusted’ in Table 3.2 were used, in place of the initial parameter estimations, to produce the probe displace- ment frequency response labeled in Figure 3.12 as ‘Analytical model with adjusted parameters’. The response from this adjusted system accurately matches the ex- perimental data, suggesting the degree to which the measured system parameters may be in error. 45 k f b f G Initial 3100 0.55 9.5 Adjusted 2000 5 5 Unit Nm Ns m N A Table 3.2: Shaker model parameter value adjustments. Figure 3.12: Model and experimental probe displacement frequency re- sponse. 3.3.4 Evaluation of Displacement Results Deciding on a target tissue strain level is the first step in determining the threshold for satisfactory probe displacement amplitudes. If tissue strain is too small, the strain Signal to Noise Ratio (SNR) becomes insufficient to produce reliable images. If the strain is too large (above 1% [97]) the tissue no longer behaves in a linear fashion. Since tissue strain of 0.3% was used to validate the primary elastography method utilized by our system, described by Eskandari et al. [28], let us start our 46 analysis with this value as our target strain. An estimation for the probe displacement amplitude required to obtain the tar- get strain value may be obtained by assuming axial strain symmetry and then in- tegrating axial strain along the axial direction. A relationship between axial strain and axial depth may be obtained by modeling the tissue as a semi-infinite, homo- geneous, elastic medium and by modeling the probe head, used to compress the tissue, as a circular compressor 14 mm in diameter, which is equal to the larger of its cross-sectional dimensions. These assumptions enable us to use the solution to the Boussinewq problem, derived by Saada [88], to define the desired relationship. Equation 3.11 shows this relationship, where z represents axial depth, a represents compressor radius, and ε(0) is the strain at the surface of the tissue. ε(z) = ε(0) [ z3 (a2+ z2)3/2 −1 ] (3.11) Equation 3.11 indicates that, for small compressor diameters, tissue strain de- cays rapidly with tissue depth. Application of this relationship indicates that, in order to obtain the targeted 0.3% strain at a depth equivalent to the far perimeter of a prostate (approximately 30 mm) a surface strain of 3.9% is required. This large surface strain would result in undesirable non-linear tissue behavior. If, instead, we constrain the surface strain to be 1%, which is the maximum value enabling tissue linearity assumptions, we find that probe displacement must be 140 µ , close to the measured maximum probe displacement threshold, and the strain at the prostate perimeter is 0.008%. The extreme strain decay resulting from small compressor diameters is illus- trated in Figure 3.13. This figure also illustrates that we could achieve both the 1% surface strain and the 0.3% prostate perimeter strain if our compressor head were to be increased to a diameter of 31 mm; however this is not a practical solution since the compressor (TRUS probe head) must be capable of passing through the anal sphincter. Additionally, this tissue strain decay model may not be accurate for our situation due to the fact that we are exciting the tissue with a relatively high frequency dynamic source, producing propagating waves, whereas the model was intended to be applicable for systems using psudostatic forces. Luckily, the investigated shaker system has been used to validate previous dy- 47 Figure 3.13: Strain decay parameterized by compressor radius. namic elastography systems, which achieved similar probe displacements capabil- ities to the ones found in Section 3.3.2. Therefore, the current probe displacement threshold was deemed adequate and no shaker alterations were implemented. How- ever, since clinical trials could indicate the need for an increased probe displace- ment threshold, the analytical model was used to predict shaker augmentations which could achieve this goal. The primary limiting factor establishing the maximum probe displacement am- plitude is the limited proof mass range of motion of about plus-minus 1.5 mm. As a result of this limitation, the shaker must be driven at low voltages. However, we could increase probe displacements, without increasing proof mass displace- ments, by increasing the mass of the proof mass, which also enables the shaker to be driven at a higher voltage. An additional result of this alteration would be to shift the system’s natural frequency to a lower value, a consequence that may be countered by increasing the flexure spring constant. Figure 3.14 compares the the- oretical displacement frequency response for the initial system to that of a system where the previously mentioned shaker parameters have been altered. The theoret- ical displacement results for the initial system, driven at 2.1Vp, are in blue and the results for the system with altered parameters are in red. For the altered system, the proof mass has been increased to 500 g, approximately the weight of a coffee mug, the spring constant has been increase to a value of 10000 Nm and the system is driven 48 at 9.5Vp, resulting in the same maximum proof mass displacement amplitudes as the initial system but with increased probe displacement amplitudes. Figure 3.14: Predicted increased shaker displacement threshold resulting from parameter augmentation. 3.4 Flexure Modification One of the main shaker mechanical design challenges is to constrain the relative proof-mass/voice-coil motion to be roughly linear while, at the same time, provid- ing a restoring force which establishes an equilibrium displacement between the two bodies that is roughly immune to the force of gravity and establishes a reason- able natural frequency. The initial shaker hardware used a pair of thin metal flex- ures to accomplish this task. Unfortunately, the geometry of this implementation requires the flexures to undergo very high levels of strain, resulting in flexure fail- ure after an undesirably short period of operation. This section describes the devel- opment of a revised flexure mechanism which enables an extended, maintenance- free, period of shaker operation. 3.4.1 The Problem The initial flexure geometry, shown in Figure 3.1, consisted of thin, rectangular, metal sheets, mounted to the proof mass and the voice coil via the four circular, 49 2-56 machine screw clearance holes cut into their faces. Figure 3.15 illustrates that the roughly linear motion, resulting from the flexure constraints, is actually a portion of a circular path. Although the principle of this elastic linkage is a commonly employed and ro- bustly implemented mechanism, its dimensions are not. Standard practice is to design this kind of elastic linkage so that it undergoes a maximum deflection that is no more than 10% of the flexure height [45]. The maximum shaker deflection is about 1.5 mm, which would require for the height of its flexures to be no less than 150 mm, if the 10% rule were to be followed. However, such large dimensions would result in a device of burdensome dimensions, thereby disabling discrete in- tegration into current medical imaging devices. Therefore, the flexure height was reduced to about 5 mm. For material that undergoes repeated loading, as in the case of our flexures, rupture will occur for stresses that are within the elastic limit of the material [9]. Therefore, a robust design requires for material stresses to be no greater than their endurance limit, which is the stress at which infinitely many cycles may be applied before rupture and which is much lower than the materials elastic limit. By so drastically reducing flexure height, the flexure material experienced stresses that were greater than its elastic limit, resulting in rapid fatigue and inevitable rup- ture. Evidence that the stresses were greater than the material elastic limit was given by the creases that formed along the flexures shortly after initiation of shaker motion. Figure 3.15: Flexure functionality. 50 3.4.2 The Fix Our first attempt to solve the flexure failure problem was to fabricate the flexures out of materials with high endurance limits in an attempt to increase the maximum allowable flexure stress to a value greater than those experienced during operation. We first used brass, then beryllium copper and then Nitinol. The flexures imple- mented out of these materials all experienced failure after still undesirably short periods of time. Our second approach to solve the flexure failure problem was to eliminate the mounting hole features from the flexures and to control their minimum radius of curvature. We were motivated to remove the mounting holes because failure is often times due to crack propagation initiated by small cracks or imperfections in the material [9], both of which were provided by the presence of the mounting hole features. Control of the minimum radius of curvature was motivated by the desire to decrease the maximum stress experienced by the flexures in hopes of bringing this stress below the endurance limit of the material. The mechanism used to implement these changes is shown in Figure 3.16, where the grey bodies, used to mount the flexure, were printed on an in-house 3D printer and the red body is the flexure. The flexure mounts were attached to the voice coil and proof mass via the same mounting holes used for the previous flexure geometry and the flexure was held in place by the clamping action between the two bodies of each mount. Flexures made of beryllium copper and Nitinol were implemented with this configuration but both materials ruptured after still undesirably short periods of operation. (a) Modified flexure geometry. (b) Flexure and mount configuration. Figure 3.16: Second flexure redesign attempt. 51 Our third and final approach was based on an observation that the flexures were attempting to solve two separate problems which could be delegated to separate mechanisms, each ideally suited for its specific task. These two tasks were first, to constrain the proof-mass/voice-coil relative motion in a roughly linear fashion and second, to provide enough stiffness to the coupling between the voice coil and proof mass that their relative equilibrium position is roughly immune to the DC force of gravity and shaker orientation. With this division of labor, the mechanism implementing the linear motion con- straint could now be made out of materials with very low elastic modulus values. As a result, material strain levels may be proportional to those experienced by the flexures of previous designs, except that now, since stress is equal to the product of strain and elastic modulus, the induced material stresses would be decreased to values below the material endurance limit. The mechanism designed to implement the roughly linear motion task is shown in Figure 3.17a, where the metallic flexures were replaced by ones made out of Kapton, with an elastic limit roughly 50 times lower than the brass flexures ini- tially used. The device shown in Figure 3.18 was used to attach the flexures and enforced a minimum flexure curvature constraint as well as enabling the flexures to be mounted without the introduction of holes or other imperfections. This mount- ing device is different than the one implemented on our second design approach because it has the additional benefit of thicker flexures with increased torsional stiffness. The majority of the mounting device bulk is a result of an attempt to adapt it to the current shaker without drilling new mounting holes in the voice coil, putting the internal coil at risk of being shorted. If implemented on future designs, however, correct placement of mounting holes would eliminate the necessity of the grey body, shown in Figure 3.18, and the features of the green body could be incor- porated into the voice coil structure, greatly reducing the height of this adaptation. The second flexure task, to supply a restoring force that minimizes the DC dis- placement caused by gravity, was accomplished by placing rubber bands about the mechanism in such a way that they resisted proof-mass/voice-coil displacements, as seen in Figure 3.17b. 52 (a) Kapton flexure and mount. (b) Kapton flexure and elastic band. Figure 3.17: Modified flexure implementation shown on shaker. Figure 3.18: Modified flexure mount. 3.4.3 Results A series of experiments were run in order to determine the probe displacement capabilities for the flexure modified system. The results of these experiments are shown in Figure 3.19 and the experimental setups associated with each of the plot- ted data sets are shown in Figure 3.20. Displacement data was collected by reading the shaker accelerometer while driving the shaker with a sequence of signals, each containing a sinusoid of a known frequency with an amplitude adjusted to maxi- mize probe displacement at that frequency. By comparing these results to those obtained in Section 3.3.2, where a similar experiment was conducted on the system with the initial flexure configuration, one can see that the system displacement capabilities have not been significantly altered as a result of the flexure modification. 53 Figure 3.19: Displacement results for modified flexure design. (a) Total system configuration. (b) Handheld configuration. (c) Free hanging configuration. Figure 3.20: Experimental setup for modified flexure design test. 54 Chapter 4 Materials and Methods: Sleeve Development 4.1 Introduction The objective of this thesis was to develop a mechanism which enables ultrasound elastography image collection during prostate biopsy exams. Elastography imag- ing requires that the imaged tissue experience small strains during data collection. Our system induces this tissue strain by coupling vibrations from the software con- trolled shaker mechanism, described in Chapter 3, to the ultrasound probe which is held against the prostate during the examination procedure. This chapter de- scribes the device, referred to as the ‘sleeve’, which enables shaker vibrations to be coupled to the ultrasound probe. The task of rigidly coupling the shaker and probe could have been solved, in a simplified manner, by drilling mounting holes into the TRUS probe, enabling for the shaker to be attached by means of an intermediary bracket. This was not a realistic solution, however, since the possibility of damaging fragile internal probe components was a prohibitively expensive deterrent, in addition to the fact that such alterations would jeopardize the probe sterilization procedure approval ob- tained by the manufacturer. Instead, the sleeve body was designed so that it could be rigidly coupled to the TRUS probe without requiring any probe alterations. The shaker could then be mounted to the sleeve by incorporating the necessary mount- 55 ing features into the sleeve design. Sleeve design constraints and fabrication methods will be discussed in Sections 4.2 and 4.3, respectively. The sleeve design evolution will be discussed in Section 4.4 and the final sleeve design will be described in Section 4.5. Additional features associated with the CSDS, described in Chapter 5, are described in Section 4.6. And clinical trial compatibility issues are discussed in Section 4.7. 4.2 Design Constraints The majority of sleeve design constraints originated from a requirement for sys- tem compatibility with the clinical environment in which it was deployed. These clinical trials, as described in Chapter 6, consisted of TRUS guided prostate biopsy procedures, using systematic sampling, where elastography image data was col- lected from the same imaging planes from which the biopsy samples were taken. Clinical trial compatibility requires that the sleeve not interfere with biopsy gun travel or the biopsy needle or needle guide locations shown in Figure 4.1. It would need to have a slim profile, enabling for comfortable handheld operation with a single hand. It would need to possess sufficient durability for sustaining the wear associated with its operation. Finally, the sleeve would need to be easily cleanable so that a sanitization procedure could be developed to satisfy hospital standards. Figure 4.1: startingSystem. 56 4.3 Fabrication Techniques The sleeve was fabricated using 3D printing technologies implementing both stere- olithography and fused deposition modeling (FDM) techniques. Sheet metal parts were cut with a water jet cutter and formed with a metal break and spot welder. Larger metal bodies were fabricated with a computer numerical control (CNC) mill. Patterns were etched into powder coated stainless steel sheets using a laser cutter. The computer aided design (CAD) program, Solidworks, was used to de- sign system components and for producing control files required by the 3D printers, water jet cutter, CNC mill and laser cutter. The design process was initiated by obtaining a CAD file defining the TRUS probe shape. This definition would enable for the utilization of probe features which would otherwise have been difficult to accurately recreate. The probe CAD file was acquired, in a vender neutral format, after an non-disclosure agreement (NDA) was negotiated with the manufacturer. Once imported into Solidworks, the vender neutral probe file produced a patchwork of poorly defined surfaces, requir- ing for a robustly defined surface to be formed by interpolation onto the available probe geometry. This surface interpolation was the primary building block used to design the probe sleeve. 4.4 Design Evolution The sleeve underwent six major design revisions during its development, assem- bled in chronological order in Figure 4.2, with additional features introduced through- out the design process. The two primary design tasks for enabling a rigid probe coupling involved a strategic selection of contact points between the probe and sleeve and the implementation of a mechanism which produced a force, referred to as the ‘clamping force’, for holding the probe against the selected contact points. The implementation of these two tasks will be discussed in the following sections. 57 Figure 4.2: Sleeve time-lapse development. 4.4.1 Clamping Force Mechanism The first five sleeve revisions created a clamping force by designing the sleeve so that a slight deformation was required to insert or remove the probe. This defor- mation produced a clamping force in proportion to the sleeve stiffness along that bending mode. The deformation mechanism was implemented by placing contacts along the length of the roughly cylindrical probe body so that the sleeve encompassed slightly more than 180 degrees of probe cross section. Figure 4.3 shows a cross section of this mechanism which illustrates the described functionality. Although this design resulted in a relatively effective sleeve, it had the undesir- able downfall that the clamping force magnitude was not easily adjustable due to its dependence the complex sleeve geometry. Additionally, sleeve fabrication was limited by its dependency on materials with high resistance to fracture and spe- cific elastic modulus values. This limitation prevented us from using our in-house printers, which used materials with inadequate mechanical properties. The final sleeve revision overcame these downfalls by creating the clamping force in an alternative manner. The sleeve was designed to be rigid and the defor- mations responsible for producing the clamping force were delegated to a commer- cial grade plastic latch. Since the latch was fabricated with an injection molding 58 Figure 4.3: Initial mechanism for producing clamping force. process, it could be made out of materials with desirable mechanical properties. Additionally, the amount of latch deformation and resultant clamping force mag- nitude could be easily configured by adjusting the distance between the two latch bodies. 4.4.2 Contact Point Selection The first four sleeve revisions attempted to mimic the behavior of a grasping hand, which makes contact with the probe over a large surface area. Therefore, the inner surface of the sleeve was either defined to match the external probe surface exactly or to be an offset of this surface. The consequence of this design is that it requires for a nearly perfect fabrication of both probe and sleeve so that there is a small dif- ference (or low error) between their planned surfaces and manufactured surfaces. Since the probe manufacturing error was unknown and the 3D printing error of our sleeve was not negligible, an alternative approach was necessary. The principles of kinematic design were used to resolve this dilemma. Kinematic, or minimum constraint design, was first described by James Clerk Maxwell in 1890 [64] and defines a method for coupling two bodies together in a definable and repeatable way. Kinematic design states that, when appropriately selected, the number of contact points between two bodies is consistent with the 59 number of free DOF between them, where there are six DOF in total: three trans- lational and three rotational. The number of contact points must be minimized so that any constraint redundancy is eliminated [36]. If a kinematic coupling is correctly implemented, the removal of any one constraint would increase the un- restrained DOF of that coupling, enabling motion in the direction of the constraint point tangent surface normal [64]. A zero DOF coupling must, therefore, consist of six contact points; any more than six and the coupling is over constrained, any less than six and the coupling is not complete. The Kelvin clamp, introduced by Lord Kelvin, provides an example of these principles and is shown in Figure 4.4 [38]. This zero DOF coupling is accom- plished with three pairs of constrains consisting of the mating of hemispherical surfaces with a trihedral hollow, a vee, and a plane, providing three, two and one contact point, respectively. Gravity provides the clamping force in this case. Figure 4.4: The Kelvin clamp coupling with zero DOF. Unfortunately, since we do not have control over the available probe mating features, such as conveniently located vees and trihedral hollows, a semikinematic design was implemented rather than a purely kinematic one. The probe features provided for this purpose are shown in Figure 4.5. The design process followed similar lines as a kinematic design would, where the initial design consisted of totally free bodies and proceeded with the addition of constraints until all undesired DOF were eliminated. 60 Figure 4.5: Probe features enabling semikinematic constraint. 4.5 Final Design The final sleeve design consists of three main components: the upper sleeve, the lower sleeve and a pair of latches, shown in Figure 4.6. The shaker is attached to the upper sleeve, which is constrained to the probe with zero DOF. The lower sleeve is constrained to the probe so that it has a single DOF in the y direction, where the coordinate axes are defined in Figure 4.6. And the upper and lower sleeve bodies are constrained with respect to each other, along the y direction. The latches provide a vertical clamping force component by pulling the upper and lower sleeve bodies together with the probe sandwiched between them. An additional clamping force component is created along the positive y direction by utilizing the conical shape of the rear probe surface. This conical shape enables lower sleeve constraints to have surface normals with components in the positive y direction so that part of the vertical clamping force may be converted into a force in this direction, as illustrated by Figure 4.7. Since the lower sleeve is not constrained along this axis, it attempts to move along this direction. However, the y axis constraint, between the lower sleeve and the fully constrained upper sleeve, restricts its motion, resulting in a positive y direction clamping force between the probe and upper sleeve. The zero DOF coupling between probe and upper sleeve is enabled by the two clamping force components, which hold the probe against the upper sleeve constraints. Coupling constraints between the probe and lower sleeve are shown in Figure 4.8 with the coordinate directions defined in Figure 4.6. It may be intuitively seen that constraint pair B eliminates the yrotation and zrotation DOF. That constraint pair A eliminates the ztranslation and xrotation DOF. And that constraint pair C eliminates 61 (a) Top View. (b) Side View. (c) Bottom View. (d) Isometric View. Figure 4.6: Final sleeve design with the upper sleeve colored yellow, the lower sleeve colored green, the latches colored black and the probe in grey. the xtranslation DOF. Constraint pairs D makes contact with the upper sleeve, shown in Figure 4.9, and constraint pairs A and C aid in the production of the y direction clamping force, as illustrated by Figure 4.7. Coupling constraints between the probe and upper sleeve are shown in Figure 4.9. Constraint pairs G and E remove the zrotation and xtranslation DOF. Constraint F and H remove the xrotation and ztranslation DOF. Constraint H with constraint pair G or E removes the yrotation DOF. Constraint H removes the ytranslation DOF. And con- straint pair D, mating with the lower sleeve body induces the positive y direction clamping force. 62 (a) Side view. (b) Bottom view. (c) Rear view. (d) Isometric view. Figure 4.7: Image illustrating production of clamping force in positive y di- rection. The lower sleeve is shown in yellow, the probe is shown in gray, the contact point constraints, between the probe and lower sleeve which enable this horizontal clamping force, are shown in red, the constraint surface normals are shown in green and the positive y components of the surface normals are show in blue. (a) With transparent probe. (b) With lower sleeve body. Figure 4.8: Lower sleeve constraints. Constraint pair A is colored blue, con- straint pair B is colored green, constraint pair C is colored red and con- straint pair D is colored yellow. 63 (a) Constraint surfaces shown on probe. (b) Upper constraint surfaces shown on sleeve body. Figure 4.9: Upper sleeve surface constraints. Constraint pair E is colored dark gray, constraint pair G is colored green, constraint F is colored blue, constraint G is colored red and constraint pair D is colored yellow and shown in Figure 4.8 as well. 64 4.6 Core Sample Depth Sensor Feature The CSDS, described in Chapter 5, requires for the biopsy gun motion to be con- strained to the probe axial direction so that the optically encoded pattern mounted to it may be read, by a sensor mounted to the sleeve, as the biopsy gun is moved for- ward during needle insertion. The biopsy gun motion and address pattern mounting requirements were implemented by replacing the portion of the biopsy gun, shown in Figure 4.10a, with one that contained an insert for the address pattern and also provided small tabs which would enable for its motion to be constrained in the desired manner, shown in Figure 4.10b, with the mating mechanism between the sleeve and biopsy gun shown in Figure 4.10c. A CNC mill was used to fabricate the modified biopsy gun component out of aluminum. A laser cutter was used to etch the optical address pattern into a black powder coated piece of polished stainless steel. The circuitry for reading the biopsy gun address pattern is secured to a mount which is attached to the sleeve in the manner shown in Figure 4.11. A cover is secured to the mount to protect the circuit during use. Sanitization concerns are addressed by placing the circuitry in a sterile plastic covering during operation, as shown in Figure 4.12. Screws pass through this covering in order to secure the circuit to the mount and must be sanitized before and after use in a procedure. 65 (a) Initial biopsy gun. (b) Altered biopsy gun. (c) mating mechanism between sleeve and biopsy gun. Figure 4.10: Biopsy gun alterations. 66 (a) Side view. (b) Top view. Figure 4.11: NTPS circuitry and mounting mechanism. (a) Side view. (b) Top view. Figure 4.12: NTPS circuitry protection. 67 4.7 Modifications for Clinical Trial As described in the previous chapters, the mechanical system components were designed so that they could undergo the same sanitation procedures as those used for preparing the TRUS probe prior to each biopsy session. Unfortunately, upon requesting permission to use our system in the hospital setting, we learned of a newly enforced sterilization protocol which would require our system to undergo extensive testing and evaluation by a cross disciplinary sterilization board before the previously established TRUS probe sanitation procedure could be approved for use with our device. Alternatively, we could design the system for one time use, where it would undergo an initial ‘terminal sterilization’ procedure and then be disposed of after exposure to biological contaminants. In order to avoid the cost and time prohibitive testing and sterilization board evaluation, we choose to redesign our system to be compatible with the one time use item protocol. Typical radiation tolerances for ordinary silicon chips are around 10 JKg , how- ever the terminal sterilization process would up to 50000 Jkg of ionizing radiation in our parts. Consequentially, the core sample depth sensor, consisting of a micro- controller and other silicon components, would most likely not survive the terminal sterilization procedure, motivating a redesign of the probe sleeve with this feature removed, as shown in Figure 4.13. Additionally, magnets were attached to the shaker, with corresponding metal plates attached to the sleeve, so that the shaker could be mounted to the sleeve via this magnetic link. With this adjustment, the shaker could be encased within a sterile plastic bag during the procedure and easily mounted to the sleeve without puncturing the sterile bag. The mechanism enabling this magnetic coupling is shown in Figure 4.14. 68 Figure 4.13: probeSleeveRedesign Figure 4.14: shakerMagnetMount 69 Chapter 5 Materials and Methods: Core Sample Depth Sensor Development 5.1 Introduction Chapters 3 and 4 describe the system components necessary for enabling elastog- raphy image collection during a prostate biopsy exam. The system was used in a series of clinical trials where each biopsy sample was preceded by elastography im- age collection. The purpose of this thesis is to prepare a system which will enable a subsequent study to determine the relationship between the histological results for these biopsy samples and the elastography image values corresponding to the tis- sue locations from which the samples were taken. If a positive correlation is found between elastography image values and pathological histology results, a targeted prostate biopsy procedure, using elastography image guidance, may be proposed. The task of associating elastography image values to biopsy sample sites is simplified by the fact that biopsy needle guides are used to constrain prospective sample sites to lie along a linear path which may be found on the TRUS probe imaging plane, as shown in Figure 5.1. Biopsy core samples are 1 mm in diameter and up to 15 mm in length and correspond to a rectangular image section lying 70 somewhere along the constrained sample path, as shown in Figure 5.1. Core sample depth is, therefore, the unknown variable which must be determined if one is to associate biopsy samples with image values. Figure 5.1: Dotted blue lines showing constrained path of biopsy needle. Biopsy needle design and the mechanical coupling between biopsy gun and needle enable core sample depth to be determined if either the distance between the biopsy gun and the TRUS probe transducer array is known or if the biopsy needle tip depth, before or after sample collection, is known. Core depth may then be found as an offset from the measured value. The task of identifying prostate biopsy needle tip depth using a TRUS image segmentation algorithm is described by Cool et al. [19]. The experimental results of this system yielded a mean error of 2.4 mm and an error standard of deviation of 4.0 mm. These results would not have been adequate for our purposes, so an alternative method for determining core depth was pursued. The system proposed by this thesis determines biopsy sample core depth by measuring biopsy gun location prior to a gun trigger, sample collection event. Biopsy gun location is measured with a position sensor fixed to the probe sleeve and is recorded for each TRUS image frame capture. An offset from the biopsy gun location measurement, recorded for the image frame captured prior to the gun trigger event, is used to calculate core depth. The image frame recorded prior to the gun trigger event is determined by visually inspecting the TRUS images, which display a biopsy needle as a rectangular, hyper-echoic body, as may be seen in Figure 5.2. 71 Figure 5.2: Appearance of biopsy needle in B-mode image collected during prostate biopsy exam. An absolute, optically encoded, reflective based sensor is used to determine biopsy gun location. The position sensor has a resolution of 1.27 mm and an ex- perimentally determined error mean of 0.15 mm and error standard deviation of 0.12 mm. The development and testing of this sensor will be discussed during the remainder of this chapter. 5.2 Design Constraints Position sensors may be implemented by detecting changes in various measurable phenomena in response to changes in a specified system dimension. Common measurable phenomena used for this purpose include changes in capacitance, in- ductance, optical reflectivity, optical translucency, electromagnetic phase change or field intensity or direction, and reflected pressure wave time of flight or location of a reflected light beam. The following design constraints helped to select the specific sensor design implemented by this system. • The sensor would need to be capable of measuring biopsy gun position with a reasonable resolution. Clinically significant tumors are generally larger than 0.5 cm3 [104] but are commonly less than 1 cm3 [27], so a resolution between 1 mm and 2 mm was deemed adequate. • The biopsy needle is capable of being inserted up to 4.5 cm past the end of the needle guide, requiring the position sensor range to be at least 4.5 cm. 72 • The sensor must be capable of reporting biopsy gun position information to a master PC upon request. • The sensor would be mounted to a handheld device, requiring for it to be small and not obstructive to the movement of a system operator. • The sensor would be used in a hospital setting, requiring compatibility with a sanitization procedure. • The sensor would be used in close proximity to directional, magnetic field based sensors, requiring for it to be magnetically quiet. • The sensor would be integrated into a complex system, already requiring the use of many cables and parts, so the need for external hardware and additional cable should be minimized. • Any additions to the biopsy gun should be passive so that there is no need for batteries or additional cables to obstruct its operation. 5.3 System Description Taking the design constraints into consideration, an absolute, optically encoded, reflective based sensor was selected for implementation. The resultant system is shown in Figure 5.3. The optically readable pattern, labeled ‘optical address pat- tern’, is mounted to the biopsy gun and the pattern reading circuit, labeled ‘sensor circuit’, is attached to a plastic mount body, labeled ‘sensor mount’, which is at- tached to the sleeve. This reflective based optical system enables the sensor circuit to be enclosed in a sterile bag and does not produce any magnetic field interference. The USB protocol was chosen for enabling communication between the master PC and sensor circuit for the following reasons. First, the digital interface elimi- nated the need for any additional analog to digital conversion (ADC) hardware at the PC end of the communication channel. Second, the digital interface enabled a low power communication channel with high noise immunity. Third, USB hard- ware is a common addition to PC systems, ensuring compatibility with a broad 73 Figure 5.3: Position sensor system components. range of machines. And fourth, the USB protocol includes a regulated power sup- ply, which eliminates the need for additional external hardware and cabling or an increase in circuit size for a battery mount. In order to minimize sensor circuit size, the system was built around the small- est reflective sensor available. The GP2S60 reflective sensor, made by Sharp, was selected for this purpose and is shown in Figure 5.4. The reflective sensor foot- print measures 3.7 mm by 1.7 mm, is sold for less than one dollar per unit and is designed to be mounted about 1 mm from the reflecting surface. It consists of an infrared emitter and detector facing the same direction. Figure 5.4: Reflective sensor GP2S60, made by Sharp. Through experimentation with the sensor array, it was found that the reflective sensors must be optically isolated when mounted in close proximity to other re- flective sensors. When this precaution is not taken, the reflected infrared light from each sensor may be detected by its neighboring sensors, causing interference. Optical isolation was achieved by mounting the sensor circuit to an opaque 74 body so that each reflective sensor is seated within a separate window, as illustrated in Figure 5.5. When mounted in this fashion, the output of each reflective sensor corresponds to the average reflectivity of the surface in front of the window in which it is seated. (a) Sensor circuit and mount front. (b) Sensor circuit and mount back. Figure 5.5: Reflective sensor optical isolation mount. 5.4 Optical Address Pattern Design 5.4.1 Resolution Limitations An intuitive starting point for designing this optical sensor is shown in Figure 5.6, where a laser printer was used to create a binary optical pattern in black and white and the sensors are mounted in an optically isolated vertical array. The difference in reflectivity between the black ink and white paper used by laser printers was found to be adequate for producing a readable optical pattern. However, the minimum optical pattern feature width was found to be quite large due to the fact that the reflective sensor output was roughly proportional to the average reflectivity over the entire surface in front of its viewing window. Therefore, any optical track with a feature width less than half of the reflective sensor window width produced a constant reflective sensor output while the sensor moved over the pattern track. In the perspective of the reflective sensor, any such optical track appeared as a solid gray color. Optical tracks with feature widths larger than half the sensor window width produced outputs with discernible maximum and minimum values enabling for sensor phase location, along the track, to be estimated. The difference between 75 Figure 5.6: Initial absolute position sensor pattern and reflective sensor array configuration. maximum and minimum sensor outputs increased with increased optical track fea- ture width and reached a maximum when the optical track feature width was equal to the reflective sensor window width. This behavior suggested that the minimum position sensor resolution would be equal to the reflective sensor window width, which was approximately 4 mm, and would not satisfy the specified design criteria. 5.4.2 Improving Resolution The resolution problem was solved by applying an extension to the principles be- hind the operation of a quadrature encoder. Quadrature encoders provide two out- puts, each consisting of the same frequency with a 90◦ phase shift between them. This functionality may be achieved by creating an optical pattern with two tracks of the same period, offset from each other by 90◦, as shown in Figure 5.7. Alternatively, the same functionality could be accomplished by mounting the reflective sensors along the same pattern track with the sensors displaced from each other by 90◦ instead of the pattern tracks over which they traverse. This setup is shown in Figure 5.8. This feature would be utilized in the position sensor design to decrease the height of the address pattern and sensor circuit. In order to understand the quadrature encoding conceptual extension it is help- ful to view the problem with a more symbolic representation of the relationship between optical sensor array configuration and optical pattern. This representation is shown in Figure 5.9, depicting the quadrature encoder configuration in the previ- ous examples, with a 12 mm optical track period. Locations along the optical track 76 Figure 5.7: Quadrature encoder configuration. Figure 5.8: Space saving quadrature encoder configuration. are mapped to phase angles and sensor arrays are represented by groups of vectors which rotate together around the optical track phase circle. Notice that the reflec- tive sensor array may be adjusted by translating any sensor by an integer multiple of 360◦ = optical pattern period without changing the sensor output and that an additional 180◦ translation would result in an inversion of the sensor output but would produce a logical transition at the same location. Therefore, ‘sensor 2’ could be located at 3 mm, 9 mm, 15 mm or 21 mm, etc., without affecting the sensor array output behavior. The extension made to the quadrature encoding technique is to use three sen- sors, where the first sensor is located at 0◦ pattern phase plus an integer multi- ple of 180◦ pattern phase, the second sensor is located at 360◦/6 = 60◦ pattern phase plus an integer multiple of 180◦ pattern phase, and the third sensor is lo- cated at (360◦/6)∗2 = 120◦ pattern phase plus an integer multiple of 180◦ pattern phase. This configuration is shown in Figure 5.10 and is referred to as a ‘tri-phase tracking’ configuration. For an optical pattern period of 12 mm, the quadrature configuration provided a resolution of 12/4 = 3 mm but for the tri-phase tracking 77 Figure 5.9: Symbolic representation for quadrature encoder configuration. configuration an increased resolution of 12/6 = 2 mm is achieved. Figure 5.10: Symbolic representation for tri-phase tracking encoder configu- ration. 5.4.3 Implemented Optical Pattern Configuration Imperial units will be used for describing the specific dimensions of the optical sensor array and pattern due to their convenient alignment with optical sensor di- mensions. The unit, mil will be used, where 1 mil = 1/1000 f oot = 25.4 µm. The minimum optical pattern feature width was set to be equal to the optical sensor window width in order to minimize feature width while maximizing the difference between optical sensor high and low output values. The optical sensor array configuration was chosen so that it minimized the number of optical sensors and pattern tracks while achieving the specified sensor range and resolution con- straints. The resultant optical pattern and sensor array configuration are shown in Figure 5.11. 78 (a) (b) (c) Figure 5.11: Optical address pattern and sensor array configuration detail. The reflective sensor windows were designed to be 150 mil, resulting in a max- imum optical pattern track resolution with feature width of 150 mil and period of 300 mil. This pattern track was traversed by three reflective sensors in a tri- phase tracking configuration, resulting in a combined sensor array resolution of 300 mil/6 = 50 mil = 1.27 mm. The phase difference between the three sensors would need to be 60◦ = 50 mil plus a multiple of 180◦ = 150 mil. This was conve- niently implemented by displacing each sensor by 200 mil = 60◦+ 180◦ from its neighboring sensor, enabling a 50 mil wall width between sensor windows. The sensor array achieves its absolute positioning capability by the addition of 79 two additional tracks, each with a resolution equal to the pattern track period of the preceding, higher resolution track. Therefore, the second track has a resolution of 300 mil, with two sensors in a quadrature configuration, traversing over a pattern track with a period of 300 mil ∗4 = 1200 mil, with a 90◦ = 300 mil displacement between them. And the third track has a resolution of 1200 mil, with one sensor traversing over a pattern track with period of 1200 mil ∗ 2 = 2400 mil, producing the position sensor range of 2400 mil ≈ 6 cm. Therefore, the specified sensor range requirements are meet and only three optical tracks are required, whereas a sensor array which did not use quadrature or tri-phase tracking would require six tracks in order to achieve the same range, assuming it’s sensors were capable of detecting 50 mil = 1.27 mm track features. 5.5 Implementation Details 5.5.1 Fabrication The position sensor printed circuit board (PCB) was designed using Altium De- signer CAD software. Gerber files were generated, shown in Figure 5.12 and used to fabricate the board with an in-house PCB router. A stereo microscope was used to solder the surface mount parts. The PCB router was not capable of producing re- liable vias, so routing complication were reduced by careful component placement and the use of resister arrays. (a) Top copper. (b) Bottom copper. Figure 5.12: Sensor circuit Gerber file output. The biopsy gun was modified so that the optical address pattern could be mounted to it and so that its motion could be constrained in such a way that the mounted address pattern translated past the position sensor circuit in a controllable 80 fashion as the biopsy gun was advanced. These modifications are described in Section 4.6. 5.5.2 Hardware and Software Design A Future Technology Devices International Ltd. FT232R chip enabled the USB communication channel to be viewed as a universal asynchronous receiver/trans- mitter (UART) interface by the sensor circuit microcontroller and allowed the mas- ter PC to communicate with the position sensor over a virtual communication port (VCP). This master PC requests position sensor location data by sending a specified character (‘p’) over a VCP connection with the position sensor. The sen- sor responds with the requested information, transmitted at 115200 baud. A tactile switch and red-green light emitting diode (LED) were used to imple- ment a user interface for enabling the position sensor to boot into one of two debug modes in addition to the default clinical boot mode. The user interface is also used to calibrate the optical sensor output thresholding algorithm so that its functionality may be adjusted to match any changes in optical address pattern material reflectiv- ity or changes in the translucency of the sterile bag through which the sensors must function. The large current draw, required by the six optical sensor infrared emitter LEDs, was pulled directly from the USB power line, enabling for components with power supply level sensitivities to be driven by a small footprint linear voltage regu- lator (LVR) with low heat dissipation capabilities. An N-channel MOSFET switch, placed between ground and a node shared by the optical sensor emitter LED cath- odes, further alleviated the high reflective sensor current demand by enabling the optical sensor emitter LED current to be turned off between sensor readings. Figure 5.13 shows the circuit used to implement the optical sensors. Optical sensor analog outputs are sampled by multiplexed microcontroller ADC channels. The discretized sensor outputs are converted to logic values using a software state machine implementing a hysteresis thresholding algorithm. Adjustments to the tilt sensitive state machine thresholding values are made by a state machine, running in the background ADC driver, which detects reflective sensor output offsets caused by optical address pattern angular displacements and adjusts the logic conversion 81 state machine thresholds accordingly. The optical sensor logic values are converted to an optical address pattern position using a lookup table. The position is then passed to an error control state machine which filters erroneous position reports and ensures reliable position sensor performance. Figure 5.13: Reflective sensor circuit implementation. 5.6 Error Results As long as every address location is detected, which is usually the case, the posi- tion sensor resolution is exactly 150 mil ≈ 1.27 mm. However, conservative sensor threshold value selection, necessitated by reflective based optical sensor tilt sensi- tivities, in addition to the fluctuation optical sensor output values, prevent position sensor state transitions from occurring precisely at their intended locations. Con- sequentially, an experiment was needed to determine a realistic value for position sensor error. By combining sensor error and resolution information, we may char- acterize the position sensor performance in an applicable manner. A linear translation stage with 10 µm resolution was used to translate the mod- ified biopsy gun component (with its embedded optical address pattern) past the position sensor circuit. The circuit was placed inside a section of sterile bag and attached to the sleeve body, which was mounted to a block and bolted to an ex- perimentation table, enabling a square alignment with the linear translation stage bolted to the same table. This experimental apparatus is shown in Figure 5.14. The position sensor was booted into a debug mode where it is programmed to report any change in optical address pattern position. This instantaneously updated position information was displayed on a PC communication terminal application. 82 Figure 5.14: Apparatus configuration for error measurement experimental. The translation stage was advanced until a position update command was displayed on the terminal and the location at which the position transition occurred was man- ually recorded. Figure 5.15a shows the reported position sensor position plotted against the actual position. Figure 5.15b the measurement error, calculated as the difference between the measured and expected position sensor transition locations. And Figure 5.15c shows a histogram of the absolute value of the error plotted in Figure 5.15b. (a) (b) (c) Figure 5.15: Error measurement experimental results. The mean of the absolute value of the position sensor error was found to be 0.15 mm and the standard of deviation was found to be 0.12 mm. If we de- fine ‘continuous resolution’ to be the sum of the resolution, mean error and er- ror standard of deviation, then the continuous resolution of this position sensor is 1.27+ 0.15+ 0.12 = 1.54 mm. This resolution falls within the initial project 83 specifications and is an improvement over the error results associated with the ul- trasound image needle segmentation technique proposed by Cool et al. [19], which reported a mean error of 2.4 mm and an error standard of deviation of 4.0 mm, resulting in a continuous resolution of 2.4+4 = 6.4 mm 84 Chapter 6 Results: System Performance Evaluation 6.1 Introduction There are many different types of elastography imaging techniques, as discussed in Section 2.7, where different techniques are compatible with different types of tissue excitation. Our system was designed with the intention of performing absolute elasticity estimation using dynamic excitation, for the reasons described in Section 1.3. This chapter describes the set of experiments carried out in order to determine whether our system was successful in achieving this goal. Experimental setup is discussed in Section 6.2. Data processing techniques used to extract the desired tissue parameter information are described in Section 6.3. Experimental results are presented in Section 6.4. And a discussion of system performance, as indicated by the results, is presented in Section 6.5. 6.2 Experimental Setup 6.2.1 Specimen Configuration Two data sets are presented in this chapter. The first data set was collected from a standard quality assurance elastography phantom CIRS Model 049 (CIRS Inc., 85 Norfolk, VA) with the experimental configuration shown in Figure 6.1. The second data set was collected during a prostate biopsy procedure carried out at Vancouver General Hospital (VGH) as part of a clinical study for which we were granted permission to gather data from up to 20 patients. Figure 6.1: Phantom experimental setup. VGH biopsy procedures are carried out by interventional radiologists using an extended, 10 to 12 core systematic biopsy pattern, as described in Section 2.5. Patients are positioned on their left sides, with their knees bent towards their chests. Patient prostates are accessed transrectally, using a TRUS probe, with attached needle guide, to guide the procedure. Images are collected from the transverse plane, with medial and lateral biopsy samples taken from both the left and right halves of the base, midgland and apex prostate regions. 6.2.2 Hardware Setup Ultrasound data was collected using a ‘SonixTouch’ ultrasound machine (Ultra- sonix Medical Corporation, Richmond, BC, Canada) with an EC9-5/10GPS curvi- linear, 128 element TRUS probe equipped with an internal ‘GPS’ magnetic track- ing sensor (Ascension Technology Corporation, Milton, VT, USA). The TRUS probe elements were driven by a 6 MHz pulse and were sampled at 40 MHz with an imaging depth of 60 mm. Each captured RF frame consists of 192 scan lines. Each scan line is captured individually and in sequential order with a delay of 101.5 µs between the initiation of each line acquisition. Image line density and 86 line acquisition time result in an RF frame rate of 1192 lines ∗ 101.5 µs = 51.3 Hz. During the phantom experiment, the excitation mechanism, referred to as the ‘shaker’ and described in Chapter 3, was coupled to the probe via the magnetic link and ‘probe sleeve’ mechanisms described in Section 4.7. During the patient study the shaker was attached to the sleeve with machine screws. The shaker was driven by the ‘shaker driver box’ described in Section 3.2.2, which was in turn controlled by a GUI software application, designed by a group of researchers working on a parallel project, running on the Sonix Touch ultrasound machine PC. The GUI software application enables integration of our system into the stan- dard prostate biopsy exam so that tissue excitation and RF data collection may take place simultaneously and just prior to each biopsy sample collection. This oper- ation sequence will enable a future study to determine any correlation that may exist between elastography imaging parameters and prostate tissue malignancy, since the tissue samples are obtained from the same prostate region as the recorded elastography data. 6.2.3 Data Set Components Elastography data sets consist of four seconds of recorded RF ultrasound frames. Frames are captured during dynamic tissue excitation by sequentially acquiring in- dividual scan lines. Additional probe motion information was obtained by using the microphone jack on an external sound card to record the output of an accelerometer rigidly coupled to the probe sleeve at a sample rate of 4200 Hz. The shaker was driven by a voltage source consisting of three sinusoidal wave- forms, at 110 Hz, 117 Hz and 22 Hz, with amplitudes of approximately 120 mV . However, since RF ultrasound frames were recorded at a rate of 51.3 Hz, the 110 Hz and 117 Hz excitation frequencies are aliased down to 110− 2 ∗ 51.3 = 7.4 Hz and 117−2∗51.3 = 14.4 Hz. This shaker driving voltage source signal, sampled at the output of the shaker driver box, is shown in Figure 6.2, in both the time and frequency domain. Figure 6.2a shows a sufficiently oversampled version of the shaker driving signal, which may be considered as a true representation of the signal, and Figure 6.2b shows the signal once it has been resampled at the ultrasound machine frame rate of 51.3 Hz, 87 and shortened to an acquisition time of four seconds. Since the latter signal is re- sampled at the same sample rate and duration as the recorded RF ultrasound data, it will contain similar artifacts as those found in our experimental RF ultrasound data sets. Specifically, Figure 6.2b shows the frequency shifted 110 Hz and 117 Hz signals, in addition to the spectral leakage resulting from the finite duration sam- pling period of 4 seconds. A periodic hamming window was used to reduce the spectral leakage so that the majority of signal energy occupies a bandwidth of ap- proximately 1 Hz around each excitation signal. (a) Oversampled shaker driver signal represents true signal. (b) Shaker driver signal, sampled under similar conditions as the recorded ultrasound RF data. The aliased 110 Hz and 117 Hz signals undergo frequency translation due to the 51.3 Hz sample rate. And the 4 second sampling period results in the spectral leakage shown as occu- pying a 1 Hz bandwidth around the excitation signals. Figure 6.2: Shaker driver voltage source. 88 6.3 Data Analysis Theoretically, system boundary conditions would cause the externally applied exci- tation energy to travel through the tissue as a slow moving, longitudinally polarized shear wave with propagation speeds between 1 ms and 10 m s . Since ultrasonic pulses travel through tissue at much greater speeds, around 1540 ms , our ultrasound sys- tem may be used to track propagation of these waves. Dynamic excitation results in a steady state wave pattern which may be characterized by creating a temporal phasor map of the excited tissue. The phasor map can then be used to determine tissue wave speed, from which elasticity is estimated. As indicated by the data processing sequence block in Figure 6.3, the input to the experimental processing sequence is a 3D RF data set, where each data point contains one sample from scan line signal consisting of a beam formed, ultrasonic pulse. One data set dimension corresponds to RF frame line index, one dimension corresponds to RF frame tissue depth, and one dimension corresponds to frame acquisition time. The data processing sequence output is a set of 2D tissue strain phasor images, where one phasor image is produced for each tissue excitation frequency. One phasor image dimension corresponds to RF frame scan line index and the other dimension corresponds to scan line tissue depth. The intention of the data pro- cessing sequence is, therefore, to determine the temporal phasor amplitude and phase of tissue strain, at a variety of tissue locations, for each individual excitation frequency. Next, the phasor data sets are passed to the parameter estimation sequence shown in Figure 6.3, where they are used to determine wave speed, from which tissue elasticity is estimated. Explanation of the implemented data processing and parameter estimation sequences will follow the order shown in Figure 6.3, where each processing block will be sequentially explained. 6.3.1 Speckle Tracking Block The first processing block applies a time domain cross correlation based speckle tracking algorithm, as described in Section, to the 3D RF data set, thereby converting it to a 3D tissue displacement data set. Decorrelation noise resulting 89 Figure 6.3: Data processing and parameter estimation steps. from out of plane motion prevented application of an absolute tissue tracking im- plementation, where an initial frame is used as reference. Instead, relative tissue motion, between sequential RF frames, was measured and a temporal cumulative sum was performed to convert these relative displacements to absolute displace- ments. 6.3.2 Strain Estimation Block Tissue displacement measurements from the speckle tracking block are inherently made relative to the ultrasound probe used to collect the RF data set. However, since the ultrasound probe is also used to induce tissue motion, and therefore un- dergoing its own dynamic motion, the previously measured tissue displacements, when viewed from a fixed coordinate system, consist of the difference between true tissue displacements and ultrasound probe displacements, each measured relative to the fixed coordinate system. Additionally, the angles between each scan line axis and the probe motion axis are unique, resulting in unique probe motion phasor amplitude offsets for each scan line. Phasor image artifacts associated with the displacement measurements greatly 90 obscure wave propagation information and must somehow be removed. One means of removing probe motion from tissue displacement data is to convert displace- ments, measured relative to the moving probe, to tissue strain. Strain measurements are independent of probe motion since measurements for each tissue location are made relative to local displacements containing acquisition coordinate system mo- tion artifacts that are common amongst them and therefore undetectable between them. Tissue strain measurements were estimated by the slope of scan line displace- ment curve samples. Strain noise was reduced by filtering the displacement data with, first, a median filter and, second, a low pass filter. Displacement curve slope measurements were then estimated by a linear fit to displacement curve samples in a region of interest (ROI) around each scan line strain estimation location. The slope of displacement curves, at each scan line strain estimation location, was then estimated by fitting a line to displacement samples in a ROI around each strain estimation location. 6.3.3 Spectral Analysis Block The 3D tissue strain data set is converted to a set of three temporal phasor images by finding the amplitude and phase of tissue strain for each of the three excitation frequencies, at every tissue strain measurement spatial location. Phasor amplitude and phase measurements are extracted from the frequency domain transformed time domain signal and then undergo phase compensation to correct for errors caused by RF frame acquisition time delay. Phasor RMS amplitude is determined as the square root of the average power within the time windowed signal spectral bandwidth. Phasor angle is determined as the phase of the frequency domain signal peak located closest to the excitation frequency spectral index. Phasor calculations are complicated by the fact that excitation signal spectral locations are shifted from their expected values by varying amounts across data set spatial dimensions. These shifts are small, with a maximum range of about 0.2 Hz, and do not affect the accuracy of phasor amplitude measurements. However, phase measurements are very sensitive to these small shifts, since phase measurement error, under our sampling conditions, increases linearly at a rate of pi [radians]0.25 [Hz] , as 91 the spectral distance increases, between Discrete Fourier transform (DFT) spectral sample peak locations and the true excitation frequency bandwidth centers. Phase error is reduced by zero padding time domain signals before application of the DFT. Zero padding increases the DFT sample resolution, of the underlying Discrete-time Fourier transform (DTFT), thereby decreasing the maximum spec- tral shift between complex DFT spectral samples and true excitation frequency locations. Periodic Hamming windows are also applied to the time domain signals before frequency domain transformation in order to decrease spectral leakage of both signal and contaminating noise components. Phasor image phase compensation is necessitated by the fact that, for each RF frame, every spatial location is sampled at a different temporal offset from the first acquired frame sample time. These time delays result in phase offset errors propor- tional to the product of phasor excitation frequency and tissue location acquisition time delay. Compensation for this phase error is performed by phase shifting each phasor by 2pi f ∗∆t, where f represents phasor frequency and ∆t represents phasor acquisition time delay. 6.3.4 Wave Velocity and Elasticity Estimation Block Wave velocity and tissue elasticity computations were simplified by performing these measurements along scan lines oriented roughly parallel to the wave prop- agation direction and by assuming a constant wave velocity and elasticity for the region under investigation. Under these circumstances, we may assume that the stress wave travel time between two phasor measurement locations is ∆t = ∆θ2pi f , where ∆t corresponds to the travel time, ∆θ corresponds to the phasor angular dif- ference and f is the phasor frequency. This is true as long as the distance between phasor locations is less than the spatial wavelength. Wave speed may then be cal- culated as c = ∆d∆t , where c is the wave speed, ∆d is the spatial difference between the two phasor locations and ∆t is the wave travel time between the two phasor locations. In an attempt to improve wave speed estimation, phasors were arranged in order of descending SNR values and the wave speed between sequentially ranked phasor locations was computed for all phasors with a SNR value greater than the selected 92 tissue region phasor SNR mean. These values were then averaged to find an overall wave speed. Tissue wave speed values were converted to elasticity estimations via Equation 2.6, with ρ = 1000 kgm2 set to equal the density of water. 6.4 Results The two experimental datasets are explored through the presentation of four sets of images for each dataset. Figures 6.4 and 6.8 provide information on experi- mental procedure operating conditions such as the test specimen B-mode appear- ance, probe displacement amplitudes during elastography data collection, and tis- sue strain spectrum appearance. Figures 6.5 and 6.9 display tissue strain phasor amplitudes, for each of the three excitation frequencies, in addition to the SNR values associated with the data from which the strain measurements were obtained. Figures 6.6 and 6.10 provide a view of strain wave propagation by displaying tis- sue strain amplitudes at two points in time. Figures 6.7 and 6.11 provide a more comprehensive view of wave prorogation by showing tissue strain amplitudes for a one dimensional ROI, at many points in time. These last figures also demonstrate the relationship between wave speed, temporal frequency, spatial frequency, and material elasticity. Image values for the 22 Hz excitation frequency were consistently higher than those for the 110 Hz and 117 Hz frequencies. Consequently, in order to display sufficient image contrast, image value color mappings for the 22 Hz excitation frequency are slightly different than those used for the 110 Hz and 117 Hz fre- quencies. All image value color mappings are indicated on the images. 93 (a) (b) (c) Figure 6.4: Phantom experiment operating conditions. (a) B-mode image of investigated tissue. The cyan outline shows tissue displacement and strain computation ROI. The red line is referenced in Figure 6.7. (b) Tissue strain spectrum for tissue location indicated by the yellow dot in (a). (c) Probe motion spectrum during data collection. Values were de- rived from accelerometer output. Excitation frequencies are color coded in images (c) and (b) but appear at different spectral locations due to dif- fering sample frequencies, as discussed in Section 6.2.3. 94 (a) (b) (c) Figure 6.5: Phantom experiment strain phasor amplitude and SNR values. Three pairs of images are presented, one for each excitation frequency. Each image pair consists of an image showing tissue strain phasor am- plitude (on the right) and an image showing the SNR associated with strain phasor amplitude measurements (on the left). SNR values were calculated as the ratio of signal average power to average noise power. 95 (a) (b) (c) Figure 6.6: Phantom experiment strain phasor real tissue values. These tem- poral tissue motion snapshots were constructed by projecting phasor images onto their real axes. Two images are provided for each excita- tion frequency, where the temporal difference between the two images is equal to a 90◦ phase shift in the imaged excitation frequency. 96 (a) (b) (c) Figure 6.7: Phantom experiment elasticity computation images. These im- ages provide a better view of tissue strain spatial variations over the course of time and provide insight into the relationship between com- puted elasticity values and wave propagation. The images were created by finding the real axis projections of phasor image components from the three excitation frequency phasor maps, at the scan line locations indicated by the red line in Figure 6.4a. The bottom edge of the im- age contains real phasor values at an initial point in time and as time progresses phasor real axis projections are stacked vertically. The same phasor image values used to create these images were also used to cal- culate tissue wave speed and elasticity, displayed on the images as ‘c’ and ‘E’, respectively, using the method discussed in Section 6.3.4. The slope of the black line on each figure is the algebraic inverse of the computed wave speed and should theoretically lie parallel to the image texture. 97 (a) (b) (c) Figure 6.8: Patient experiment operating conditions. (a) B-mode image of in- vestigated tissue. The cyan outline shows tissue displacement and strain computation ROI. The red line is referenced in Figure 6.11. (b) Tissue strain spectrum for tissue location indicated by the yellow dot in (a). (c) Probe motion spectrum during data collection. Values were derived from accelerometer output. Excitation frequencies are color coded in images (c) and (b) but appear at different spectral locations due to dif- fering sample frequencies, as discussed in Section 6.2.3. 98 (a) (b) (c) Figure 6.9: Patient experiment strain phasor amplitude and SNR values. Three pairs of images are presented, one for each excitation frequency. Each image pair consists of an image showing tissue strain phasor am- plitude (on the right) and an image showing the SNR associated with strain phasor amplitude measurements (on the left). SNR values were calculated as the ratio of signal average power to average noise power. 99 (a) (b) (c) Figure 6.10: Patient experiment strain phasor real tissue values. These tem- poral tissue motion snapshots were constructed by projecting phasor images onto their real axes. Two images are provided for each excita- tion frequency, where the temporal difference between the two images is equal to a 90◦ phase shift in the imaged excitation frequency. 100 (a) (b) (c) Figure 6.11: Patient experiment elasticity computation images. These images provide a better view of tissue strain spatial variations over the course of time and provide insight into the relationship between computed elasticity values and wave propagation. The images were created by finding the real axis projections of phasor image components from the three excitation frequency phasor maps, at the scan line locations in- dicated by the red line in Figure 6.8a. The bottom edge of the image contains real phasor values at an initial point in time and as time pro- gresses phasor real axis projections are stacked vertically. The same phasor image values used to create these images were also used to calculate tissue wave speed and elasticity, displayed on the images as ‘c’ and ‘E’, respectively, using the method discussed in Section 6.3.4. The slope of the black line on each figure is the algebraic inverse of the computed wave speed and should theoretically lie parallel to the image texture. 101 6.5 Discussion The two datasets presented in Section 6.4 will be discussed in parallel, so that con- trasts in their characteristics may be compared. The discussion will be completed by an evaluation of system performance. 6.5.1 Figures 6.4 and 6.8: B-mode, Point Strain and Probe Motion Images The phantom data set B-mode image shows a circular inclusion with a diameter of about 15 mm, centered at a depth of about 40 mm. The sleeve accelerometer spectrum image reports probe motion to be approximately 1.2 mm, for the 22 Hz signal and approximately 40 µm for the 110 Hz and 117 Hz signals. The point strain spectrum image illustrates that, even though there is a large discrepancy be- tween the low frequency excitation signal probe displacement amplitude and the probe displacement amplitudes of the high frequency excitation signals, the resul- tant tissue strain is relatively similar. This is due to the fact that tissue strain is proportional to a spatial derivative, and high frequency temporal excitations pro- duce high frequency spatial waves resulting in large spatial derivatives. The patient B-mode image shows a hypoechoic region, between about 5 mm and 20 mm which may be used to identify the prostate boundary. The bladder may be identified as the hypoechoic region at the top right of the image, separated from the prostate by a hyperechoic section of abdominal tissue. Excitation signal probe displacements for this data set have similar amplitudes to those found in the phantom data set, however, the patient probe displacement spectrum appears to be much less noisy than the phantom spectrum. The patient point strain spectrum image also conveys this observation of noise amplitude contrast, where the patient tissue strain SNR values are higher than those found for the phantom experiment. One explanation for this difference could be due to the fact that a different shaker-sleeve coupling was used for these two data sets. The shaker was bolted to the sleeve for the phantom experiment but coupled to the sleeve through the magnetic link described in Section 4.7 for the phantom experiment. Upon investigation, it was found that the magnets which enable this coupling, and are intended to be rigidly attached to the shaker, exhibit a slight 102 amount of play. The poor shaker magnetic coupling could be responsible for an in- crease in signal noise and should be remedied before future experiments are carried out. 6.5.2 Figures 6.5 and 6.9: Strain Phasor Amplitude and SNR Images Tissue strain amplitude is a key factor in our system’s ability to accurately track shear wave propagation. If the strain levels are too high, above about 1.0% [97], the emersion of tissue nonlinear behavior would invalidate necessary assumptions made by absolute elasticity estimation techniques and reduce phasor measurement accuracy. If strain levels are too low, signal submersion into the noise floor would prevent accurate wave speed measurements. In the work of Eskandari et al. [28], tissue strain amplitudes were limited to be no more than 0.3%, therefore a strain amplitude range between 0.1% and 0.3% may be seen as a desired system target. Tissue strain amplitudes for the two data sets do appear to stay below or close to the upper limit of our tissue strain target range, however, the strain SNR values are quite low for the two higher frequency excitation signals. Strain waves do not appear to penetrate the phantom inclusion and deteriorate to unusable amplitudes in a variety of regions in both phantom and patient phasor maps. Patient tissue strain amplitudes appear to be greatly attenuated past the prostate boundary, which is to be expected since this boundary would reflect a good por- tion of the shear waves due to the elastic contrast between prostate and abdominal tissue. Patient tissue strain amplitude and SNR values also seem to be less con- sistent than measurements made from the phantom data set, resulting in a more speckled appearance. This phenomenon may be due to the fact that the phantom material contains a homogeneous distribution of acoustic scatterers, whereas the patient tissue scatterer distribution would be more sporadic. 6.5.3 Figures 6.6 and 6.10: Real Strain Value Images These figures provide a view of spatial wave propagation over the course of time by presenting two images of the phasor real axis projections, where the two pha- sor images are separated by a phase difference of 90◦. Shear wave reflections of sufficient amplitude would result in standing waves, which would have a station- 103 ary appearance in space and would also be visible in the phasor amplitude images. However, a lack of standing wave features in the phasor amplitude images indicates that such reflections are not present in the data sets, and that wave propagation pat- terns would therefore consist of features that appear to travel away from the probe head as time is advanced. Shear wave spatial wavelength is proportional to tissue wave speed and in- versely proportional to its temporal frequency, following the relationship: c = fλ , where c represents wave speed, f represents temporal frequency and λ represents spatial wavelength. Also, tissue wave speed, under the assumptions made by abso- lute elastography techniques, is proportional to the square root of tissue elasticity, given by Equation 2.6. Therefore, we may predict excitation signal spatial wave- length if tissue elasticity is known. A previous study, by Baghani et al. [8], used four separate absolute elasticity estimation techniques to find elasticity values for the same phantom used by our experiment, resulting in an averaged background phantom material elasticity value of 20.3 kPa. Therefore, we would expect the 22 Hz excitation signal to produce a longitudinally polarized shear wave with wavelength of approximately 203 mm, the 110 Hz signal to produce a 40 mm wave and the 117 Hz signal to produce a 38 mm wave. Phantom data set images for the 110 Hz excitation signal provide the best view of traveling wave propagation. The peak to trough distance for this wave is approx- imately 20 mm, which means that it would have a wavelength of about 40 mm, as predicted. The 22 Hz excitation signal also exhibits an expected behavior, since the spatial wavelength for this frequency appears to be large in comparison to image depth. 6.5.4 Figures 6.11 and 6.11: Elasticity Computation Images In order to provide a better view of wave propagation over the course of time, images have been constructed from the real axis projections of 1D phasor map sections, as the phasor map is advanced through time, as described in Section 6.4. These images convey a texture with negative slope, where following the pattern will cause your eyes to move vertically and to the right. This pattern is created 104 by the translation of wave features, towards an increased tissue depth, with the progression of time. The slope of this pattern should, therefore, be the algebraic inverse of the wave speed. Wave speeds were computed, in the fashion described in Section 6.3.4, and used to plot the black line appearing on the bottom half of the images. As predicted, the slope of this line does match the general pattern found in the images. Wave speed computations were also used to find elasticity estimations and both of these values were added to a legend displayed on the figures. The average phan- tom elasticity estimation was 20.9 kPa, which is satisfyingly close to the value of 20.3 kpa found by the study mentioned in the previous section. The average prostate elasticity estimation was 20.7 kPa. It is difficult to find a gold standard to compare this value to since absolute prostate elasticity mea- surements reported in the literature vary by two orders of magnitude [89] between 2 kPa and 500 kPa. However, since the measurement obtained by our experiment lies squarely in the middle of this range, it is fair to say that this is not an unrea- sonable measurement, to say the least. 6.5.5 System Performance Conclusions The experiments presented in this chapter have demonstrated that the system under investigation is capable of performing rudimentary tissue elasticity measurements and therefore shows potential as an effective tool for researching the use of elastog- raphy imaging techniques to guide targeted prostate biopsy procedures. The data sets presented above, however, indicate that parameter estimation is performed on signals with undesirably low SNR values. One explanation for these low SNR values is that strain information, which is computed from displacement data at the cost of high frequency noise amplification, was used to perform parameter estimations. Strain signals were chosen due to their ability to remove probe motion artifacts from the data sets that are inherently mea- sured relative to the moving probe, however, an alternative, less noisy technique could prove to be effective in increasing SNR values. One such solution could be to apply a high pass filter to the spatial domain of tissue displacements. This could be effective due to the fact that probe mo- 105 tion information is transmitted through tissue at the speed of the ultrasonic pulses, approximately 1540 ms , resulting in much lower spatial frequencies than those of longitudinally polarized shear waves of the same temporal frequency which travel at much lower speeds. In fact, conversion of tissue displacement signals into tissue strain information is performed by application of a similar operation. The difference, however, is that strain computation is equivalent to application of a high pass filter with very poor passband signal constraints, whereas a custom built high pass filter could reduce high frequency noise amplification. 106 Chapter 7 Conclusion The design of system components for an elastography based targeted prostate biopsy research platform have been presented in the previous chapters. This chapter de- scribes specific thesis contributions enabling the realization of this system and pos- sible directions for future work. 7.1 Contributions Tissue Excitation Device: The Shaker A previously developed shaker device was integrated into our system by perform- ing two steps. First, an analytical model was developed for the shaker electrical and mechanical systems. This model was used to predict both tissue excitation performance and the consequences of modifications to specific shaker components. Second, the shaker flexure system was redesigned to enable durable performance under its challenging operating constraints. Shaker-Probe Mechanical Coupling Device: The Sleeve A sleeve device was designed and fabricated for rigidly coupling the shaker to the ultrasound probe. Sleeve design followed minimum constraint design principles to enable reliable and consistent operation in the presence of fabrication errors. The restorative clamping force, necessitated by the zero DOF coupling, was enabled through the integration of a commercial grade plastic latch capable of enduring the 107 required force-generating deformations. Delegation of these deformations to the latch mechanism enabled sleeve fabrication from a wide range of materials. Core Sample Depth Sensor: The CSDS A sensor for detecting biopsy needle tissue depth, the CSDS, was designed for use with the sleeve device. This sensor uses an optical receiver array to read a biopsy gun mounted, absolute position address pattern. Biopsy gun position information is then used to calculate needle depth with a resolution of 1.27 mm and error of 0.15± 0.12 mm. This is an improvement over needle segmentation based algo- rithms which report needle depth measurement errors of 2.4±4.0 mm. Data Processing Steps Code was developed for enabling an evaluation of system performance by perform- ing spectral analysis on tissue strain measurements and using the resultant infor- mation to perform course elasticity measurements. This code was used to process elastography data, collected from phantom and human subjects, in order to obtain a preliminary system validation. 7.2 Future Work Tissue Excitation Device: The Shaker One concern over the current shaker design is the lack of a safety mechanism en- abling drive current shutdown under dangerous operating conditions that might result in coil overheating. Such a mechanism could be implemented by including a thermal fuse, such as the Cantherm SDF DF192S, in the shaker coil circuit. Shaker-Probe Mechanical Coupling Device: The Sleeve Bioburden and mechanical testing by private, certified companies are still needed in order to obtain VGH approval for a sterilization procedure which would enable multiple use operation of the sleeve device. This approval may also enable for the CSDS device to be re-integrated into the sleeve mechanism. 108 Core Sample Depth Sensor: The CSDS The CSDS sensor is capable of providing accurate needle depth measurements. However, this functionality is enabled at the cost of requiring the biopsy gun lo- cation to be tightly constrained. The necessity for this constraint is due to the fact that reflective based optical sensors are highly sensitive to address pattern tilt er- rors. This is much less the case for transmissive based optical sensors which are, for this reason, more frequently implemented. Stringent requirements for biopsy gun motion limitations necessitated the al- teration of biopsy gun components in order to enable better kinematic mating con- straints. A system requiring such alterations may have difficulty obtaining approval for use from the cross disciplinary sterilization board assembled by VGH. Conse- quently, I would recommend for future attempts at implementing such a sensor to design their system around a color sensor such as the TCS3200D-TR, manufac- tured by AMS-TAOS USA Inc. Color sensor readings, which report intensities from three color bands, may be interpreted as three dimensional vectors. This approach enables sensor outputs to be defined in terms of angular location measurements which may be far more immune to tilt errors. With decreased sensor tilt sensitivity, biopsy gun motion constraints may be loosened, enabling effective sensor integration in the absence of biopsy gun alterations. Absolute position measurements may be obtained by using a single color sen- sor to read the patterns printed onto temporary stickers placed on the biopsy gun. Since the sensors can detect three colors, one color may be set at a constant level for the entirety of the track, a second color may be varied by one cycle across the dura- tion of the track and the third color could be made to vary by many cycles over the length of the track. From this configuration, two vectors may be constructed from every pattern position, where one provides a course location measurement and the second enables a fine scale measurement. The sensor electronics may be placed in a 3D printed mold and encased in epoxy in order to accommodate a liberal range of sterilization procedures. Data Processing Steps The parameter estimation sequence developed by this thesis was intended to pro- 109 vide a rough estimate of system efficacy. Additional work is required for the se- lection and integration of an absolute elasticity estimation technique such as the LFE or FE modeling methods described in Section These methods may use the phasor generation code produced by this thesis. However, improved pa- rameter estimation may be enabled by the use of tissue displacement information, rather than the noise sensitive tissue strain information, for phasor map generation. This improvement will require the development of a technique for removing probe motion artifacts from tissue displacement information. 110 Bibliography [1] S.K. Alam, J. Ophir, and E.E. Konofagou. An adaptive strain estimator for elastography. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 45(2):461–472, 1998. → pages 25, 28 [2] W.C. Allsbrook Jr, K.A. Mangold, M.H. Johnson, R.B. Lane, C.G. Lane, and J.I. Epstein. Interobserver reproducibility of gleason grading of prostatic carcinoma: general pathologist. Human Pathology, 32(1):81–88, 2001. → pages 11 [3] R.Z. Azar, K. Dickie, and L. Pelissier. Real-time 1-d/2-d transient elastography on a standard ultrasound scanner using mechanically induced vibration. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 59(10):2167–2177, 2012. → pages 23 [4] R.Z. Azar and S.E. Salcudean. Motion estimation in ultrasound images using time domain cross correlation with prior estimates. Biomedical Engineering, IEEE Transactions on, 53(10):1990–2000, 2006. → pages 25 [5] R.Z. Azar and S.E. Salcudean. Time-delay estimation in ultrasound echo signals using individual sample tracking. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 55(12):2640–2650, 2008. → pages 24 [6] A. Baghani, A. Brant, S. Salcudean, and R. Rohling. A high-frame-rate ultrasound system for the study of tissue motions. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 57(7):1535–1547, 2010. → pages 23 [7] A. Baghani, H. Eskandari, S.E. Salcudean, and R. Rohling. Measurement of viscoelastic properties of tissue-mimicking material using longitudinal wave excitation. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 56(7):1405–1418, 2009. → pages 4, 29 111 [8] A. Baghani, S.E. Salcudean, M. Honarvar, R.S. Sahebjavaher, R. Rohling, and R. Sinkus. Travelling wave expansion: a model fitting approach to the inverse problem of elasticity reconstruction. Medical Imaging, IEEE Transactions on, 30(8):1555–1565, 2011. → pages 4, 30, 104 [9] F.P. Beer, E.R. Johnston, and J.T. Dewolf. Mechanics of Materials. McGraw-Hill, 3 edition, 2002. → pages 40, 50, 51 [10] S.A. Bigler, R.E. Deering, and M.K. Brawer. Comparison of microscopic vascularity in benign and malignant prostate tissue. Human Pathology, 24(2):220–226, 1993. → pages 22 [11] R. Blum and M. Scholz. Invasion of the Prostate Snatchers: No More Unnecessary Biopsies, Radical Treatment Or Loss of Sexual Potency. Other Press, 2010. → pages 11 [12] L.N. Bohs and G.E. Trahey. A novel method for angle independent ultrasonic imaging of blood flow and tissue motion. Biomedical Engineering, IEEE Transactions on, 38(3):280–286, 1991. → pages 25 [13] R. Boucher and J. Hassab. Analysis of discrete implementation of generalized cross correlator. Acoustics, Speech and Signal Processing, IEEE Transactions on, 29(3):609–611, 1981. → pages 25 [14] S. Catheline, J.L. Gennisson, G. Delon, M. Fink, R. Sinkus, S. Abouelkaram, and J. Culioli. Measurement of viscoelastic properties of homogeneous soft solid using transient elastography: An inverse problem approach. The Journal of the Acoustical Society of America, 116(6):3734–3741, 2004. → pages 29, 30 [15] I. Cespedes, Y. Huang, J. Ophir, and S. Spratt. Methods for estimation of subsample time delays of digitized echo signals. Ultrasonic Imaging, 17(2):142–171, 1995. → pages 25 [16] I. Céspedes and J. Ophir. Reduction of image noise in elastography. Ultrasonic Imaging, 15(2):89–102, 1993. → pages 25 [17] D.A. Christensen. Ultrasonic Bioinstrumentation. John Wiley and Sons, Inc., 1988. → pages 19 [18] D.L. Cochlin, R.H. Ganatra, and D.F.R. Griffiths. Elastography in the detection of prostatic cancer. Clinical Radiology, 57(11):1014–1020, 2002. → pages 22 112 [19] D.W. Cool, L. Gardi, C. Romagnoli, M. Saikaly, J.I. Izawa, and A. Fenster. Temporal-based needle segmentation algorithm for transrectal ultrasound prostate biopsy procedures. Medical physics, 37(4):1660–1673, 2010. → pages 71, 84 [20] E.D. Crawford. Epidemiology of prostate cancer. Urology, 62(6):3–12, 2003. → pages 12 [21] G. Crile and A.L. Vickery. Special uses of the silverman biopsy needle in office practice and at operation. The American Journal of Surgery, 83(1):83–85, 1952. → pages 16 [22] N.B. Delongchamps, G. De La Roza, R. Jones, M. Jumbelic, and G.P. Haas. Saturation biopsies on autopsied prostates for detecting and characterizing prostate cancer. British Journal of Urology International, 103(1):49–54, 2008. → pages 2, 18 [23] M.M. Doyley, P.M. Meaney, and J.C. Bamber. Evaluation of an iterative reconstruction method for quantitative elastography. Physics in Medicine and Biology, 45(6):1521–1540, 2000. → pages 28, 29 [24] J.N. Eble and P.A. Angermeier. The roles of fine needle aspiration and needle core biopsies in the diagnosis of primary prostatic cancer. Human Pathology, 23(3):249–257, 1992. → pages 16, 17 [25] C.N. Edwards, E. Steinthorsson, and D. Nicholson. An autopsy study of latent prostatic cancer. Cancer, 6(3):531–554, 1953. → pages 11 [26] H. Ekman, K. Hedberg, and P.S. Person. Cytological versus histological examination of needle biopsy specimens in the diagnosis of prostatic cancer. British Journal of Urology International, 39(5):544–548, 1967. → pages 17 [27] J.I. Epstein, M.J. Carmichael, A.W. Partin, and P.C. Walsh. Small high grade adenocarcinoma of the prostate in radical prostatectomy specimens performed for nonpalpable disease: pathogenetic and clinical implications. The Journal of Urology, 151(6):1587–1592, 1994. → pages 2, 17, 72 [28] H. Eskandari, S.E. Salcudean, and R. Rohling. Viscoelastic parameter estimation based on spectral analysis. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 55(7):1611–1625, 2008. → pages 29, 46, 103 113 [29] H. Eskandari, S.E. Salcudean, R. Rohling, and J. Ohayon. Viscoelastic characterization of soft tissue from dynamic finite element models. Physics in Medicine and Biology, 53(22):6569–6590, 2008. → pages 27 [30] H. Eskandari, S.E. Salcudean, and R. Rohlirig. Tissue strain imaging using a wavelet transform-based peak search algorithm. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 54(6):1118–1130, 2007. → pages 24 [31] R.S. Ferguson. Prostatic neoplasms: their diagnosis by needle puncture and aspiration. The American Journal of Surgery, 9(3):507–511, 1930. → pages 16 [32] W.J. Frable. Fine-needle aspiration biopsy: a review. Human Pathology, 14(1):9–28, 1983. → pages 16, 17 [33] W.J. Frable. Needle aspiration biopsy: past, present, and future. Human Pathology, 20(6):504–517, 1989. → pages 16, 17 [34] G.F. Franklin, J.D. Powell, A. Emami-Naeini, and J.D. Powell. Feedback control of dynamic systems. Addison-Wesley, 5 edition, 2006. → pages 40 [35] S. Franzen, G. Giertz, and J. Zajicek. Cytological diagnosis of prostatic tumours by transrectal aspiration biopsy: a preliminary report. British Journal of Urology, 32(2):193–196, 1960. → pages 16 [36] J.E. Furse. Kinematic design of fine mechanisms in instruments. Journal of Physics E: Scientific Instruments, 14(3):264–272, 2000. → pages 60 [37] D.F. Gleason. Histologic grading of prostate cancer: a perspective. Human Pathology, 23(3):273–279, 1992. → pages 11 [38] L.C. Hale and A.H. Slocum. Optimal design techniques for kinematic couplings. Precision Engineering, 25(2):114–127, 2001. → pages 60 [39] R.M. Haralick, K. Shanmugam, and I.H. Dinstein. Textural features for image classification. Systems, Man and Cybernetics, IEEE Transactions on, 6:610–621, 1973. → pages 20 [40] S.W. Hayward and G.R. Cunha. The prostate: development and physiology. Radiologic Clinics of North America, 38(1):1–14, 2000. → pages 9 [41] K.K. Hodge, J.E. McNeal, and T.A. Stamey. Ultrasound guided transrectal core biopsies of the palpably abnormal prostate. The Journal of Urology, 142(1):66–70, 1989. → pages 16 114 [42] K.K. Hodge, J.E. McNeal, M.K. Terris, and T.A. Stamey. Random systematic versus directed ultrasound guided transrectal core biopsies of the prostate. The Journal of Urology, 142(1):71–74, 1989. → pages 15 [43] A.G. Houston, S.B. Premkumar, D.E. Pitts, and R.J. Babaian. Prostate ultrasound image analysis: localization of cancer lesions to assist biopsy. In Computer-Based Medical Systems, Proceedings of the Eighth IEEE Symposium on, pages 94–101, 1995. → pages 20 [44] J.S. Jones. Saturation biopsy for detecting and characterizing prostate cancer. British Journal of Urology International, 99(6):1340–1344, 2007. → pages 2 [45] R.V. Jones. Some uses of elasticity in instrument design. Journal of Scientific Instruments, 39(5):193–203, 2002. → pages 50 [46] F. Kallel and M. Bertrand. Tissue elasticity reconstruction using linear perturbation method. Medical Imaging, IEEE Transactions on, 15(3):299–313, 1996. → pages 29 [47] K. KÖNIG, U. Scheipers, A. Pesavento, A. Lorenz, H. Ermert, and T. Senge. Initial experiences with real-time elastography guided biopsies of the prostate. The Journal of Urology, 174(1):115–117, 2005. → pages 3, 28 [48] E.E. Konofagou, T. Varghese, J. Ophir, and S.K. Alam. Power spectral strain estimators in elastography. Ultrasound in Medicine and Biology, 25(7):1115–1129, 1999. → pages 28 [49] L.G. Koss and M.R. Melamed. Koss’ diagnostic cytology and its histopathologic bases, volume 1. Lippincott Williams & Wilkins, 2005. → pages 17 [50] J.N. Krieger, D.E. Riley, P.Y. Cheah, M.L. Liong, and K.H. Yuen. Epidemiology of prostatitis: new evidence for a world-wide problem. World Journal of Urology, 21(2):70–74, 2003. → pages 9 [51] T.A. Krouskop, D.R. Dougherty, and F.S. Vinson. A pulsed doppler ultrasonic system for making noninvasive measurements of the mechanical properties of soft tissue. Journal of Rehabilitation Research and Development, 24(2):1–8, 1987. → pages 23 [52] T.A. Krouskop, T.M. Wheeler, F. Kallel, B.S. Garra, and T. Hall. Elastic moduli of breast and prostate tissues under compression. Ultrasonic Imaging, 20(4):260–274, 1998. → pages 22 115 [53] M. Kun. A new instrument for the diagnosis of tumors. Monthly Journal of Medical Sciences, 7:853–854, 1847. → pages 16 [54] F. Labrie, B. Candas, A. Dupont, L. Cusan, J.L. Gomez, R.E. Suburu, P. Diamond, J. Lévesque, and A. Belanger. Screening decreases prostate cancer death: first analysis of the 1988 quebec prospective randomized controlled trial. The Prostate, 38(2):83–91, 1999. → pages 11 [55] J.E. Langer. The current role of transrectal ultrasonography in the evaluation of prostate carcinoma. In Seminars in Roentgenology, volume 34, pages 284–294, 1999. → pages 13, 15 [56] E.H. Lee. Assessment, screening and diagnosis of prostate cancer. In M. Wallace and L.L. Powel, editors, Prostate Cancer: Nursing Assessment Management, and Care, pages 33–46. Springer, 2002. → pages 11 [57] F. Lee, P.J. Littrup, S.T. Torp-Pedersen, C. Mettlin, T.A. McHugh, J.M. Gray, G.H. Kumasaka, and R.D. McLeary. Prostate cancer: comparison of transrectal us and digital rectal examination for screening. Radiology, 168(2):389–394, 1988. → pages 12 [58] R.M. Lerner, S.R. Huang, and K.J. Parker. sonoelasticity images derived from ultrasound signals in mechanically vibrated tissues. Ultrasound in Medicine and Biology, 16(3):231–239, 1990. → pages 22, 23 [59] P.J. Littrup, F. Lee, and C. Mettlin. Prostate cancer screening: current trends and future implications. American Cancer Society: a Cancer Journal for Clinicians, 42(4):198–209, 1992. → pages 12 [60] F.L. Lizzi, E.J. Feleppa, S. Kaisar Alam, and C.X. Deng. Ultrasonic spectrum analysis for tissue evaluation. Pattern Recognition Letters, 24(4):637–658, 2003. → pages 20 [61] F.L. Lizzi, M. Ostromogilsky, E.J. Feleppa, M.C. Rorke, and M.M. Yaremko. Relationship of ultrasonic spectral parameters to features of tissue microstructure. Ultrasonics, Ferroelectrics, and Frequency Control, IEEE transactions on, 34(3):319–329, 1987. → pages 20 [62] A. Manduca, R. Muthupillai, P.J. Rossman, J.F. Greenleaf, and R.L. Ehman. Local wavelength estimation for magnetic resonance elastography. In Image Processing, IEEE International Conference on, volume 3, pages 527–530, 1996. → pages 4, 30 116 [63] H.E. Martin and E.B. Ellis. Biopsy by needle puncture and aspiration. Annals of Surgery, 92(2):169–182, 1930. → pages 16 [64] J.C. Maxwell. The Scientific Papers of James Clerk Maxwell..., volume 2. University Press, 1890. → pages 59, 60 [65] J.E. McNeal. Normal histology of the prostate. The American Journal of Surgical Pathology, 12(8):619–633, 1988. → pages 14 [66] J.E. McNeal, E.A. Redwine, F.S. Freiha, T.A. Stamey, et al. Zonal distribution of prostatic adenocarcinoma. correlation with histologic pattern and direction of spread. The American Journal of Surgical Pathology, 12(12):897–906, 1988. → pages 14, 15 [67] M. Mitterberger, G.M. Pinggera, W. Horninger, G. Bartsch, H. Strasser, G. Schafer, A. Brunner, E.J. Halpern, J. Gradl, and L. Pallwein. Comparison of contrast enhanced color doppler targeted biopsy to conventional systematic biology: Impact on gleason score. Jefferson Department of Radiology Faculty Papers, 2007. → pages 22 [68] N. Miyanaga, H. Akaza, M. Yamakawa, T. Oikawa, N. Sekido, S. Hinotsu, K. Kawai, T. Shimazui, and T. Shina. Tissue elasticity imaging for diagnosis of prostate cancer: a preliminary report. International Journal of Urology, 13(12):1514–1518, 2006. → pages 3, 4 [69] M. O’Donnell, A.R. Skovorada, and B.M. Shapo. Measurement of arterial wall motion using fourier based speckle tracking algorithms. In Ultrasonics Symposium. IEEE Proceedings on, pages 1101–1104. IEEE, 1991. → pages 27 [70] M. O’Donnell, A.R. Skovoroda, B.M. Shapo, and S.Y. Emelianov. Internal displacement and strain imaging using ultrasonic speckle tracking. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 41(3):314–325, 1994. → pages 19, 26 [71] T.E. Oliphant, A. Manduca, R.L. Ehman, and J.F. Greenleaf. Complex-valued stiffness reconstruction for magnetic resonance elastography by algebraic inversion of the differential equation. Magnetic Resonance in Medicine, 45(2):299–310, 2001. → pages 30 [72] Canadian Cancer Societys Steering Committee on Cancer Statistics. Canadian cancer statistics 2012. Technical report, Canadian Cancer Society, Toronto, ON, 2012. → pages 9, 10 117 [73] J. Ophir, I. Cespedes, H. Ponnekanti, Y. Yazdi, and X. Li. Elastography: a quantitative method for imaging the elasticity of biological tissues. Ultrasonic Imaging, 13(2):111–134, 1991. → pages 4, 22, 25, 28 [74] L. Pallwein, M. Mitterberger, J. Gradl, F. Aigner, W. Horninger, H. Strasser, G. Bartsch, D. Nedden, and F. Frauscher. Value of contrast-enhanced ultrasound and elastography in imaging of prostate cancer. Current Opinion in Urology, 17(1):39–47, 2007. → pages 3, 4, 28 [75] L. Pallwein, M. Mitterberger, P. Struve, G. Pinggera, W. Horninger, G. Bartsch, F. Aigner, A. Lorenz, F. Pedross, and F. Frauscher. Real-time elastography for detecting prostate cancer: preliminary experience. British Journal of Urology International, 100(1):42–46, 2007. → pages 3, 4, 23 [76] E. Park and A.M. Maniatty. Shear modulus reconstruction in dynamic elastography: time harmonic case. Physics in Medicine and Biology, 51(15):3697, 2006. → pages 29 [77] E.L. Peirson and D.A. Nickerson. Biopsy of the prostate with the silverman needle. New England Journal of Medicine, 228(21):675–678, 1943. → pages 16 [78] A. Pesavento and A. Lorenz. Real time strain imaging and in-vivo applications in prostate cancer. In Ultrasonics Symposium, IEEE Proceedings on, volume 2, pages 1647–1652. IEEE, 2001. → pages 3, 4 [79] A. Pesavento, C. Perrey, M. Krueger, and H. Ermert. A time-efficient and accurate strain estimation concept for ultrasonic elastography using iterative phase zero estimation. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 46(5):1057–1067, 1999. → pages 26, 27 [80] J.C. Presti Jr, J.J. Chang, V. Bhargava, and K. Shinohara. The optimal systematic prostate biopsy scheme should include 8 rather than 6 biopsies: results of a prospective clinical trial. The Journal of Urology, 163(1):163–166, 2000. → pages 18 [81] U. Raeth, D. Schlaps, B. Limberg, I. Zuna, A. Lorenz, G. Van Kaick, W.J. Lorenz, and B. Kommerell. Diagnostic accuracy of computerized b-scan texture analysis and conventional ultrasonography in diffuse parenchymal and malignant liver disease. Journal of Clinical Ultrasound, 13(2):87–99, 2005. → pages 20 118 [82] H. Ragde, H.C. Aldape, and C.M. Bagley. Ultrasound-guided prostate biopsy biopty gun superior to aspiration. Urology, 32(6):503–506, 1988. → pages 17 [83] M.I. Resnick. Prostatic ultrasonography. Pmph Bc Decker, 1990. → pages 13 [84] M.D. Rifkin. Transrectal prostatic ultrasonography: comparison of linear array and radial scanners. Journal of Ultrasound in Medicine, 4(1):1–5, 1985. → pages 13 [85] M.D. Rifkin, W. Dähnert, and A.B. Kurtz. State of the art: endorectal sonography of the prostate gland. American Journal of Roentgenology, 154(4):691–700, 1990. → pages 14, 18 [86] M.D. Rifkin, E.A. Zerhouni, C.A. Gatsonis, L.E. Quint, D.M. Paushter, J.I. Epstein, U. Hamper, P.C. Walsh, and B.J. McNeil. Comparison of magnetic resonance imaging and ultrasonography in staging early prostate cancer. New England Journal of Medicine, 323(10):621–626, 1990. → pages 2, 18 [87] R. Righetti, J. Ophir, and P. Ktonas. Axial resolution in elastography. Ultrasound in Medicine and Biology, 28(1):101–113, 2002. → pages 28 [88] A.S. Saada. Elasticity: theory and applications. J Ross Pub, 2009. → pages 47 [89] R. S. Sahebjavaher, A. Baghani, M. Honarvar, R.h Sinkus, and S. E. Salcudean. Transperineal prostate mr elastography: initial in vivo results. Magnetic Resonance in Medicine, 69(2):411–420, 2013. → pages 105 [90] R.S. Sahebjavaher, A. Baghani, R. Sinkus, and S.E. Salcudean. Prostate mre at 3t: Trans-perineal wave propagation. In Proceedings of the 19th Annual Meeting of International Society for Magnetic Resonance in Medicine, page 1477, 2011. → pages 22 [91] S. Salcudean, D. French, S. Bachmann, R. Zahiri-Azar, X. Wen, and W. Morris. Viscoelasticity modeling of the prostate region using vibro-elastography. Medical Image Computing and Computer-Assisted Intervention, pages 389–396, 2006. → pages 23 [92] G. Salomon, J. Köllerman, I. Thederan, F.K.H. Chun, L. Budäus, T. Schlomm, H. Isbarn, H. Heinzer, H. Huland, and M. Graefen. Evaluation of prostate cancer detection with ultrasound real-time elastography: a 119 comparison with step section pathological analysis after radical prostatectomy. European Urology, 54(6):1354–1362, 2008. → pages 3, 4 [93] G. Schmitz, H. Ermert, and T. Senge. Tissue-characterization of the prostate using radio frequency ultrasonic signals. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 46(1):126–138, 1999. → pages 20 [94] T. Shiina, M.M. Doyley, and J.C. Bamber. Strain imaging using combined rf and envelope autocorrelation processing. In Ultrasonics Symposium. IEEE Proceedings on, volume 2, pages 1331–1336. IEEE, 1996. → pages 27 [95] K. Shinohara, T.M. Wheeler, and P.T. Scardino. The appearance of prostate cancer on transrectal ultrasonography: correlation of imaging and pathological examinations. The Journal of Urology, 142(1):76–82, 1989. → pages 18 [96] I. Silverman. A new biopsy needle. The American Journal of Surgery, 40(3):671–672, 1938. → pages 16 [97] R. Sinkus, J. Bercoff, M. Tanter, J.L. Gennisson, C. El Khoury, V. Servois, A. Tardivon, and M. Fink. Nonlinear viscoelastic properties of tissue assessed by ultrasound. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 53(11):2009–2018, 2006. → pages 27, 46, 103 [98] R. Sinkus, M. Tanter, S. Catheline, J. Lorenzen, C. Kuhl, E. Sondermann, and M. Fink. Imaging anisotropic and viscous properties of breast tissue by magnetic resonance-elastography. Magnetic Resonance in Medicine, 53(2):372–387, 2005. → pages 22 [99] American Cancer Society. Cancer facts & figures 2012. Technical report, American Cancer Society, Atlanta, 2012. → pages 10, 12 [100] American Cancer Society. Prostate cancer: Early detection. Technical report, American Cancer Society, Atlanta, 2012. → pages 12, 18 [101] American Cancer Society. Cancer facts & figures 2013. Technical report, American Cancer Society, Atlanta, 2013. → pages 1 [102] J.A. Spencer, A.A. Alexander, L. Gomella, T. Matteucci, and B.B. Goldberg. Clinical and us findings in prostate cancer: patients with normal prostate-specific antigen levels. Radiology, 189(2):389–393, 1993. → pages 12 120 [103] S. Srinivasan and J. Ophir. A zero-crossing strain estimator for elastography. Ultrasound in Medicine and Biology, 29(2):227–238, 2003. → pages 24 [104] T.A. Stamey, F.S. Freiha, J.E. McNeal, E.A. Redwine, A.S. Whittemore, and H.P. Schmid. Localized prostate cancer. Cancer, 71(S3):933–38, 1993. → pages 72 [105] C. Stephan, K. Jung, M. Lein, and E.P. Diamandis. Psa and other tissue kallikreins for prostate cancer detection. European Journal of Cancer, 43(13):1918–1926, 2007. → pages 12 [106] C.S. Stewart, B.C. Leibovich, A.L. Weaver, and M.M. Lieber. Prostate cancer diagnosis using a saturation needle biopsy technique after previous negative sextant biopsies. The Journal of Urology, 166(1):86–91, 2001. → pages 18 [107] M.K. Terris. Prostate biopsy strategies: past, present, and future. The Urologic Clinics of North America, 29(1):205–212, 2002. → pages 18 [108] J.M. Thijssen. Spectroscopy and image texture analysis. Ultrasound in Medicine and Biology, 26(1):S41–S44, 2000. → pages 20 [109] E.J. Trabulsi, D. Sackett, L.G. Gomella, and E.J. Halpern. Enhanced transrectal ultrasound modalities in the diagnosis of prostate cancer. Urology, 76(5):1025–1033, 2010. → pages 22 [110] E. Turgay, S. Salcudean, and R. Rohling. Identifying the mechanical properties of tissue by ultrasound strain imaging. Ultrasound in Medicine and Biology, 32(2):221–235, 2006. → pages 28 [111] T. Varghese and K.D. Donohue. Estimating mean scatterer spacing with the frequency-smoothed spectral autocorrelation function. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 42(3):451–463, 1995. → pages 20 [112] T. Varghese, E.E. Konofagou, J. Ophir, S.K. Alam, and M. Bilgen. Direct strain estimation in elastography using spectral cross-correlation. Ultrasound in Medicine and Biology, 26(9):1525–1538, 2000. → pages 28 [113] T. Varghese and J. Ophir. Performance optimization in elastography: Multicompression with temporal stretching. Ultrasonic Imaging, 18(3):193–214, 1996. → pages 28 121 [114] T. Varghese and J. Ophir. Enhancement of echo-signal correlation in elastography using temporal stretching. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 44(1):173–180, 1997. → pages 25 [115] F. Viola and W.F. Walker. A spline-based algorithm for continuous time-delay estimation using sampled data. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 52(1):80–93, 2005. → pages 25 [116] J.U. Voigt. Quantification of left ventricular function and synchrony using tissue doppler, strain imaging, and speckle tracking. In L. Galiuto, L. Badano, L. Fox, R. Sicari, and J.L. Zamorano, editors, The European Association of Echocardiography Textbook of Echocardiography, pages 81–97. OUP Oxford, 2011. → pages 23 [117] P.C. Walsh, A.B. Retik, Jr. Vaughan, E.D., and A.J. Wein. Campbell’s Urology. Saunders, 8 edition, 2002. → pages 15 [118] H. Watanabe, H. Kaiho, M. Tanaka, and Y. Terasawa. Diagnostic application of ultrasonotomography to the prostate. Investigative Urology, 8(5):548–559, 1971. → pages 13 [119] H.G. Welch and P.C. Albertsen. Prostate cancer diagnosis and treatment after the introduction of prostate-specific antigen screening: 1986–2005. Journal of the National Cancer Institute, 101(19):1325–1329, 2009. → pages 12 [120] Z. Wu, L.S. Taylor, D.J. Rubens, and K.J. Parker. Sonoelastographic imaging of interference patterns for estimation of the shear velocity of homogeneous biomaterials. Physics in Medicine and Biology, 49(6):911–922, 2004. → pages 22 [121] Y. Yamakoshi, J. Sato, and T. Sato. Ultrasonic imaging of internal vibration of soft tissue under forced vibration. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, 37(2):45–53, 1990. → pages 23 [122] Y. Zheng, G. Li, M. Chen, Q.C.C. Chan, S.G. Hu, X.N. Zhao, R.L. Ehman, E.Y. Lam, and E.S. Yang. Magnetic resonance elastography with twin pneumatic drivers for wave compensation. In Engineering in Medicine and Biology Society. 29th Annual International Conference of the IEEE on, pages 2611–2613. IEEE, 2007. → pages 22 122 [123] H. Ziegler, D. Völter, and G.E. Schubert. Morphological criteria for the control of carcinoma of the prostate with estrogen therapy. International Urology and Nephrology, 6(3):195–200, 1974. → pages 17 123


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