THESIS. the Graduate School of The Ohio State University. Kimberly Mulligan, BS, DVM. The Ohio State University. Master's Examination Committee:

Transcription

1 Elastography Characterization and Repeatability in 14 Normal Canine Spleens THESIS Presented in Partial Fulfillment of the Requirements for the Degree Master of Science in the Graduate School of The Ohio State University By Kimberly Mulligan, BS, DVM Graduate Program in Comparative and Veterinary Medicine The Ohio State University 2015 Master's Examination Committee: Wm Tod Drost, DVM, DACVR, Advisor Mary Jo Burkhard, DVM, ACVP (Clinical Pathology) John Bonagura, DVM, MS, DACVIM (Cardiology)

2 Copyrighted by Kimberly Mulligan, DVM 2015

3 Abstract Ultrasound is commonly used to evaluate the canine spleen. There is poor correlation between ultrasonographic findings and cytological or histopathological diagnosis. Ultrasound elastography is a relatively new technology that can measure the elasticity of tissues, and has shown usefulness in differentiating benign versus malignant conditions in the parenchymal organs in people. In animals elastography imaging is in its earliest stages of application, and has been applied mainly to musculoskeletal tissues. Only limited studies have investigated elastography characteristics of the normal canine spleen. The purpose of this study was to describe elastography characteristics of normal canine spleens and assess intra-and interobserver repeatability of the method. Elastography (Toshiba Aplio500, Elasto program) was performed by two experienced ultrasonographers in 14 clinically healthy, non-sedated dogs. Each spleen was considered normal based on B-mode ultrasound examinations. Color map images depicting the strain distributionacross splenic regions of interest were recorded and were measured at minimum, neutral, and maximum transducer compression points as indicated by a simultaneously recorded applied pressure waveform. Regions of interest (ROIs) encompassing as much of the spleen as possible were drawnand pixel color percentages for each region of interest were determined. A strain score was then calculated for each ii

4 spleen, using a weighted average of the percentages. Three neutral values per spleen were averaged for the first observer to see if repeatability could be improved. Elastography was repeated by the same observers using the same dogs approximately two weeks later. Statistical analysis included Friedman s test (followed by Bonferroni-corrected signed-rank tests) to compare strain scores over the three compression points. Intra-and interobserver repeatability were assessed using Spearman s rank-order correlation. Waveform height was evaluated to assess if differences in operator probe pressure might contribute to the data and this variable was compared between examiners by calculation of the Mann-Whitney U statistic (i.e., Wilcoxon rank-sum test). Strain scores obtained at neutral compression were significantly lower (harder) than maximum compression for both observers (P<0.05), but only observer one found a significant difference in strain score compared with minimum compression. Intraobserver 2 correlation was moderate and statistically significant (repeatable) for observer one (R s = , P=0.043), but there was marked variation among individual dogs. For observer 2 two, there was poor to moderate correlation which was not statistically significant (R s = 0.433, P=0.122), indicating poor intraobserver repeatability. Interobserver correlation for the first time point was weak to moderate but did not achieve statistical significance 2 (R s = , P=0.095), and even lower correlation was found at the second time point 2 (R s = 0.301, P=0.122). Overall these correlation coefficients indicate poor interobserver repeatability with the methods used in this study. When three neutral values per spleen were averaged for observer one, the correlation coefficient was lower but not statistically iii

5 significant (R 2 s = 0.209, p=0.747). Observer two exhibited significantly more compression than observer one (P=0.002). This study demonstrates that abdominal compression influences strain scores in the canine spleen. Based on these results, we recommend obtaining strain scores at neutral compression, which is theoretically the most accurate approach and differs significantly from maximum compression. Both inter- and intraobserver repeatability were poor to moderate, findings related in part to differences in operator compression.it is likely that patient dependent factors, such as variable compliance and lack of sedation, also contributed to the less-than-ideal repeatability in this study. Accordingly, thorough training and practice as well as other standardizations are needed if this elastography technique is to become useful in canine splenic imaging in the clinical or research settings. iv

6 Acknowledgments A sincere thank you to Dr. Meghann Lustgarten, DVM, DACVR, and Michael Morgan, DVM, for providing the pixel analysis program used in this project. I would also like to thank my Master s Committee for all of their help and advice, as well as my radiology service faculty and resident-mates. Special acknowledgement should be given to Danelle Auld for her infinite patience, excellent instruction, and enormous help with this project. Finally, thank you to the ACVR and the Veterinary Ultrasound Society Resident Research Award for funding this project. v

7 Vita Owensboro Catholic High School B.S. Animal Science, University of...kentucky D.V.M Auburn University 2012 to present...graduate Student, Department of Veterinary Clinical Sciences, The Ohio State University Fields of Study Major Field: Comparative and Veterinary Medicine vi

8 Table of Contents Abstract ii Acknowledgements..v Vita..vi Fields of Study vi Table of Contents...vii List of Tables..ix List of Figures..x CHAPTER 1: INTRODUCTION The Reason for Elastography Physics of Elasticity Physics of Elasticity Imaging Utility of Ultrasound Elastography Based on the Human Literature Elastography in Veterinary Medicine Specific Aims/Hypotheses CHAPTER 2: MATERIALS AND METHODS Inclusion and Exclusion Criteria Ultrasound and Elastography Procedure Image Analysis and Strain Value Calculation Height Analysis vii

9 2.5 Statistics Methods CHAPTER 3: RESULTS Subject Demographics Descriptive Statistics and Comparing Compression Types Inter-observer Results Intra-observer Results Height Analysis Results...33 CHAPTER 4: DISCUSSION Comparing Compression Types Repeatability Effect of Compression Strength Limitations Future Research Conclusions..49 REFERENCES..52 APPENDIX A: Subjects 57 APPENDIX B: Descriptive Statistics and Normality Tests..58 APPENDIX C: Comparing Compression Levels.. 63 APPENDIX D: Repeatability Testing...65 APPENDIX E: Comparing Waveform Height.. 68 viii

10 List of Tables Table 1.Ultrasound machine settings for each patient Table 2.Example of pixel percentages and strain value calculation Table 3. Subject characteristics..57 Table 4. Normality test for Observer 1, initial time point.58 Table 5. Descriptive statistics for Observer 1, initial time point...60 Table 6. Normality tests for Observer 2, initial time point 60 Table 7. Descriptive statistics for Observer 2, initial time point...62 Table 8. Friedman's test for Observer 1, initial time point 63 Table 9. Post-hoc Wilcoxon signed-rank test for Observer 1, initial time point...63 Table 10. Friedman's test for Observer 2, initial time point..64 Table 11. Post-hoc Wilcoxon signed-rank test for Observer 2, initial time point.64 Table 12. Inter-observer repeatability, initial time point...65 Table 13. Inter-observer repeatability, recheck time point 65 Table 14. Intra-observer repeatability for Observer Table 15. Intra-observer repeatability for Observer Table 16. Intra-observer repeatability for Observer 1, with three data points averaged...67 Table 17. Normality assessment for the height data..68 Table 18.Comparison of maximum peak waveform for each observer.69 ix

11 Table 19. Comparison of strain scores for each observer..69 x

12 List of Figures Figure 1. Digital palpation of an abnormality Figure 2.Concept of elasticity using a simple spring analogy Figure 3.Measurement of strain using a one-dimensional spring model Figure 4. Estimation of tissue displacement using cross-correlation Figure 5.Simplified way of estimating strain from tissue displacement for hard and soft masses Figure 6.Depiction of tissue Doppler imaging Figure 7.Example of an elastogram Figure 8.Elastogram demonstrating maximum compression Figure 9.Region of interest Figure 10.Approximate cut-off regions for determining weighting values Figure 11.Example of waveform height calculation Figure 12.Summary of strain scores for Observer 1, initial time point Figure 13. Summary of strain scores for Observer 2, initial time point Figure 14. Before-and-after graph demonstrating inter-observer repeatability at the initial time point xi

13 Figure 15. Before-and-after graph demonstrating inter-observer repeatability at the recheck time point Figure 16. Before-and-after graph demonstrating intra-observer repeatability for Observer Figure 17. Before-and-after graph demonstrating intra-observer repeatability for Observer Figure 18. Intra-observer repeatability for Observer 1, with three compression types averaged together...40 Figure 19. Comparison of the amount of probe pressure exerted by each observer..41 Figure 20. Example of an elastogram obtained with a MyLab 70 XVG machine, ElaXto software...50 Figure 21. Elastogram comparison...51 xii

14 Chapter 1: INTRODUCTION 1.1 The Reason for Elastography Physical palpation has been a vital component of diagnostic testing since the beginnings of medicine. Pathology such as neoplasia or fibrosis can render a tissue softer or firmer than surrounding tissues, allowing it to be diagnosed as an abnormality. 1 In most casesdigital pressure reveals a region of stiffer tissue or reduced compressibility compared to surrounding tissues (Figure 1). Indeed it has been shown that scirrhous carcinoma of the human breast is stiffer and less mobile than benign fibroadenomas. 2 In some situations the shape of abnormal tissue can also be determined by palpation, as with subcutaneous nodules. However, palpation is not always practical or is limited for a variety of reasons; for example, when the tissue of interest is too deep within the body or the patient is obese. Ultrasound imaging provides a method to evaluate internal structures from the outside. Conventional B-mode ultrasound utilizes differences in acoustic impedance to provide the characteristic black and white speckles in an image. Acoustic impedance is the product of tissue density and velocity of sound through the tissue and is characteristic for different tissue types (e.g., adipose tissue). 2 Although ultrasound can provide a variety of morphological (and in some cases functional) information, the findings are often non specific and do not correlate with a precise histopathological diagnosis. 3, 4, 5-89 In 1

15 addition, if the acoustic impedance of two tissues does not differ, these tissues will not appear different on a B-mode ultrasound image, even when a true pathological difference exists. Ultrasound elastography was developed as a method to aid in the diagnosis of abnormal tissues. Elastography combines structural and impendence information obtained with B-mode ultrasound and also provides information about tissue stiffness. It is often considered a complement to conventional B-mode imaging, and is not necessarily designed as a stand-alone test. 10 Numerous studies have explored the characteristics and utility of elastography in human patients (see Section 1.4 for more detail). In contrast there is a paucity of information about elastography in veterinary medicine. 1.2 Physics of Elasticity Elastography measures tissue stiffness, or elasticity. There are two basic variables when considering elasticity stress and strain. Stress is the force applied to an object and strain is the resulting displacement. The amount of displacement relates to the elasticity of the object a more elastic object is more deformed or displaced for a given amount of force. 12 The two-dimensional elastic property of an object is simplifiedusing a spring analogy (Figure 2). A known mass is suspended from a spring and a known force is applied to it. The elongation of the spring can be measured, and the amount of elongation depends on the stiffness of the spring and the weight of the mass. If two springs of equal stiffness are used with different masses suspended, the more massive object will create 2

16 greater elongation of the spring. 11 The stiffness of the spring can be determined via Hooke s law, ΔF=kΔx. This equation relates the difference in elongation (Δx) with the difference in applied force (ΔF) to produce k, which is a constant and characterizes the stiffness of a spring. 11 This principle is also demonstrated in Figure 3, where the spring constant is defined as Δl/l. The strain is constant for all springs in this image. This also demonstrates that the strain profile is dependent on the number and stiffness of all the springs. 12 This becomes more relevant in clinical imaging when the number of tissues in the path of the ultrasound beam is considered. Additionally, the absolute value of the strain along the compression axis is proportional to the magnitude of the initial compression. This is important to consider in the live patient when operators attempt to control the amount of pressure applied to the ultrasound probe. When a three-dimensional structure is considered, these concepts become more complex, involving stress, strain, and Young s modulus. 13 Consider the compression of a cube made of an ideal material that is both homogeneous and isotropic (ignore gravitational forces). Now consider a second cube of similar material, which we will use to compress the other object. Each object is three-dimensional, and each object has a surface upon which forces will be placed one surface will be doing the compression and the other surface will be compressed. The force of compression is known as stress. There is stress imparted by the compressing object, and there is also a counter-stress exerted by the object being compressed. 3

17 Strain is the result of stress. The object being compressed undergoes strain, and the degree of strain in the object depends upon the force of the stress and the compliance of the material being stressed. The more compliant the material, the more strain (or deformation) will occur. The measurement of stress and strain requires knowledge of tensors. A tensor is a generalized concept of a vector. A vector of a three-dimensional object (such as in the cube analogy) uses a 3D coordinate system. A stress tensor σ ij is a 3 x 3 matrix comprised of the nine combinations that result from combining two independent 3D coordinate systems a force and the surface element on which that force acts. 11 A strain tensor, ε kl, uses a similar 3 x 3 matrix, but this corresponds to the rate of deformational change (another way to describe strain) of the object. Combining the strain tensor and the stress tensor yields a constitutive equation. In a purely elastic, lossless deformation (an ideal solid), the stress only depends on the strain. This yields the equation σ ij = C ijkl ε kl. C ijkl is the modulus tensor, which is equivalent to k described above. To reiterate, this value is intrinsic to the object and characterizes its stiffness. It is also known as Young s modulus. 1 Accurate measurement of the elastic modulus of an object requires knowledge of the exact amount of force applied (stress) and the ability to measure the amount of deformation (strain). This is impractical in all but the strictest of experimental conditions. Additionally, when in vivo experiments are considered, we no longer have an ideal solid with lossless deformation. Most biological tissues exhibit a non-linear stress- 4

18 strain relationship that further reduces the ability to accurately determine the intrinsic elasticity of a tissue. 1.3 Physics of Elasticity Imaging While there has been some interest in imaging the stress distribution, strain imaging has gained the most attention. Initially, M-mode imaging was used to track tissue movements 13 and tissue elasticity. 14 Through development of cross-correlation techniques, static compression elastography was developed. Other types of elastography are shear-wave and acoustic radiation force impulse. However, static elastography was used in this thesis experiment, and so will receive the most attention here. In the initial quantitative study about static elastography, the absolute method for elastic modulus imaging was described. 12 However, this method of imaging was under tight experimental conditions in phantoms and bacon slabs where the amount of compression was controlled using a mechanical arm. This does not mimic a clinical situation. While using a mechanical arm to deliver compression was more accurate, a technique that will allow for freehand scanning was desirable. The first real-time system for imaging elasticity was developed and reported for prostate imaging 15 using an endocavity transducer. Numerous other studies followed, further refining the technology. The basic mechanism of static compression elastography (strain elastography) is measurement of the relative tissue displacement. Relative displacement is emphasized here because we are estimating elasticity, not measuring the intrinsic elastic modulus. The imaging system obtains a pre-compression image (map of anatomy) of the tissue. 5

19 A deformation force is then applied, either mechanically by the operator using the ultrasound probe or by using the breathing motion of the patient. A post-compression image is then obtained. The strain is estimated by calculating the rate of spatial change of the image field. 11 This rate of change is measured using cross-correlation technology (Figure 4). The radiofrequency backscatter waveform becomes compressed when the tissue is compressed, and is also shifted in position. The position shift is determined using a cross-correlation function. The process is repeated at multiple depths and at each position on the transducer face, resulting in a 2D map of the tissue displacement versus depth. 1 Softer structures exhibit greater strain, and harder tissues exhibit less strain (Figure 5). Measuring tissue displacement in this manner is also known as Tissue Doppler Imaging (TDI), which is used by the Toshiba Aplio 500 Elasto program. TDI uses the same principle that spectral Doppler imaging and color Doppler imaging use to image the movement and velocity of blood. The movement of blood changes the frequency of the returning echoes, and this information is used to determine direction and velocity of blood flow. In TDI, high velocity, low amplitude signals from blood are filtered out, leaving tissue motion information visible (Figure 6). 16 Using TDI the following information can be derived: velocity, displacement, strain, and strain rate. When elastography is performed with the Toshiba Aplio 500, three basic components are seen (Figure 7). One is the gray-scale B-mode image used for reference, one is the elastogram color map, and the other is the green waveform at the bottom. With the color map, the settings can be adjusted, but the purposes of this study, blue 6

20 indicates firmer, or stiffer tissue, and red indicates softer tissue. The green waveform is a plot of the strain rate over time, which is derived from the temporal derivative of displacement. Taller waves mean greater tissue velocity. In order to achieve accurate results, the waveform must be smooth and even. This requires the operator to make smooth and even compressions with the ultrasound probe. 1.4 Utility of Ultrasound Elastography Based on the Human Literature In the first six weeks of 2015, a literature search ( [search terms = st elastography not MR not shear]) found 52 articles regarding static ultrasound elastography. At least 3600 articles have been published since the early 1990s. In the interest of brevity, only a selection of articles (focusing on abdominal organs) will be presented here. Normal stiffness of the human spleen was reported in The main focus of splenic elastography in humans has related to estimating splenic stiffness for the purposes of identifying portal hypertension. More specifically, it was hoped that elastography could provide a non-invasive way to determine the presence of esophageal varices in the face of hepatic cirrhosis. One study determined that spleen stiffness could predict the presence, but not the grade, of esophageal varices. 18 However, they also determined that the presence of esophageal varices was better predicted when both spleen and liver measurements were used, but this was not found in a similar study. 19 More generally, splenic stiffness could predict clinical decompensation in patients with cirrhosis. 20 Other 7

21 publications corroborate these findings, and have shown promising results that elastography will be useful in humans with cirrhosis and portal hypertension. 21,22 Substantial attention has also been paid to differentiating malignant from benign tumors using elastography. Elastography has been able to demonstrate an 8 mm cancer nodule within a breast 23 and differentiate between malignant and benign breast nodules. 24 Elastography is also useful for distinguishing malignant from benign hepatic nodules Additional areas of research with elastography include elastography of the prostate, 15 thyroid gland, 31 lymph nodes, 32 and musculoskeletal structures. 33 While many studies are preliminary in scope, the utility of elastography to differentiate various diseases within these tissues has been documented. In particular, musculoskeletal structures elastography can be used to measure dynamic conditions such as tendon motion and strain Elastography in Veterinary Medicine Elastography has only recently been applied to veterinary medicine. An in vitro study of the normal canine prostate was conducted which described normal elasticity findings. 35 Several similar studies have used animals as models for humans, including a study of canine hepatic stiffness. 36 To the author s knowledge, the first report of elastography in veterinary patients not used for biomedical research purposes was a study that documented normal elastography values for the liver, spleen, and kidneys in cats. 37 Another study 8

22 investigated whether elastography could differentiate between benign and malignant nodules in the canine spleen, but determined that is elastography was not helpful. 38 Normal values for elastography in the canine liver, spleen, kidneys and prostate were recently reported. 39,46 Using acoustic radiation force impulse elastography the characteristics of the spleen in healthy cats 40 and canine mammary neoplasia were described. 41 In the latter study, elastographic values for malignant tumors were significantly higher than for benign tissues, indicating this may be a useful technology for determining benign from malignant conditions in mammary masses. 41,42 In the musculoskeletal system, the elastographic characteristics of metacarpal tendons in horses without clinical evidence of tendon injury were reported, 43 and another study evaluated reproducibility and feasibility of elastography in the superficial digital flexor tendons of clinically normal horses. 44 Both of these studies showed elastography to be repeatable and feasible. In dogs, elastography in the gastrocnemius tendon was also reported to be feasible and repeatable, and a different study showed elastography useful 45, 47 to monitor healing of the common calcaneal tendon following surgery. 1.6 Specific Aims/Hypotheses The aim of our study was to describe elastography characteristics and repeatability in normal canine spleens using a Toshiba Aplio 500 ultrasound machine with Elasto technology. 9

23 The first goal was to determine if there was a difference in overall strain between minimum compression, neutral compression, and maximum compression. We hypothesized that neutral compression would be significantly different than maximum or minimum compression. The second goal was to determine inter-observer and intra-observer repeatability. Our hypothesis was that there would be good intra-observer repeatability but poor interobserver repeatability. The third goal was to determine if there was a significant difference in compression strength by each observer. We hypothesized that there would be a significant difference in compression strength between the two observers. 10

24 Figure 1: Digital palpation of an abnormality. The black oval provides greater resistance (red arrows) to physical deformation than the surrounding tissue (orange and pink arrows) when acted on by an outside force (black arrow). In this case, a finger provides the deformational force

25 Figure 2: Concept of elasticity using a simple spring analogy. Two weights of known but different mass are each suspended from a spring. The applied force (acceleration) is gravity. The difference in the stretch of the spring is a function of the different mass of the weights. k is a proportionality constant that characterizes the stiffnesss of the spring

26 Figure 3: Measurement of strain using a one-dimensional spring model. (a) Three springs in a pre-compression state. (b) Post-compression state, with an applied force F. Note that the very hard spring does not compress as much as the soft springs. (c) Strain profile

27 Figure 4: Estimation of tissue displacement using cross-correlation. 1 14

28 Figure 5: Simplified way of estimating strain from tissue displacement for hard and soft masses. 1 15

29 Figure 6: Depiction of tissue Doppler imaging. High frequency, low amplitude echoes from blood are filtered out, leaving low frequency, high amplitude echoes from tissues visible

30 Figure 7: Example of an elastogram. The grey scale B-mode is on the right, the color map is on the left, and the velocity waveform is on the bottom. 17

31 Chapter 2: MATERIALS AND METHODS 2.1 Inclusion and Exclusion Criteria Dogs were recruited from privately owned pets of faculty members, residents, and technicians employed by The Ohio State Veterinary Medical Center, and fourth year students enrolled in The Ohio State University College of Veterinary Medicine. Dogs were included in the study if they were systemically healthy (based on history and physical examination) and did not have any underlying condition that would affect the spleen or the cardiovascular system. All animals were current on vaccinations. Dogs were excluded from the study if the spleen was sonographically abnormal or another abnormality was detected on full abdominal ultrasound that might affect the spleen directly. 2.2 Ultrasound and Elastography Procedure None of the dogs were sedated for the procedure. Animals were placed in dorsal recumbency and manually restrained. The hair was parted if possible and alcohol +/- coupling gel was applied to the skin as needed to allow good ultrasound transducer contact. If good contact could not be achieved in this fashion, a small amount of hair was clipped from the animal. 18

32 A full abdominal ultrasound (Aplio 500, Toshiba Medical Systems, Tochigi-Ken, Japan) was performed on each dog to rule out gross abnormalities. Transducer selection for the initial full-abdomen scan varied from a 5-8 MHz curvilinear, 8-11 MHz microconvex, and MHz linear probe, depending on the anatomical region. The ultrasound was performed by either the author (Observer 1) or an ultrasound technician with over 20 years of sonographic experience (Observer 2). Particular attention was paid to the spleen of each patient, which was thoroughly evaluated for evidence of mottling, nodules, size changes, or shape changes. Unanimous agreement among the author, the sonographer, and one radiology faculty member was needed before the spleen was deemed sonographically normal. Elastograms were obtained in each dog by both observers in random succession, and neither observer was blinded to the results of the other. Prior to beginning the study, both operators received one training session by a Toshiba representative in the correct usage of the Elasto program. To standardize the region of measurement, the hilus of the spleen was located and measurements obtained just off-axis from the hilus. The high-frequency linear probe was used at 18 MHz and the Elasto button on the ultrasound system was activated producing a display with the gray-scale B-mode image on the operator s right, and a color map elastogram on the operator s left (Figure 8). Table 1 lists the system settings that were used. All settings were the same for each patient and were set according to the Toshiba recommendations. 19

33 Next, mild rhythmic compressions with the ultrasound transducer were begun to deform the spleen and complete the elastographic procedure. A waveform at the bottom of the elastogram provided real-time feedback on the quality of compressions (see Figure 8). Each observer monitored the compressions using the waveform and strived to make the waveforms as even and consistent as possible. When the observer was satisfied with the consistency of the compressions, the image was frozen. The color bar in the upper left of the image (see Fig. 8) shows the scale of elasticity and the associated color. This was set via the Color Map setting on the controls (Table 1) and was the same for each animal. In this study, harder tissues were denoted blue and softer tissues were denoted red. Yellow, cyan, and green were also used to provide a range from hardest (blue) to softest (red). The track ball was used to move the cursor to the peak of a representative waveform, henceforth referred to as maximum compression. The cursor was then placed where the waveform crossed the baseline point neutral compression. Finally, the cursor was placed at the trough of the waveform minimum compression. All three still frames (maximum, neutral, and minimum compression) were then saved to the PACS system. A short (approximately 6 seconds) cine loop was saved for Observer 1. This elastography procedure was repeated in the same subjects fourteen days after the initial examination. The landmarks and machine settings were all the same. Each animal remained healthy during the interim period. 20

34 2.3 Image Analysis and Strain Value Calculation All images were viewed using a DICOM viewer (efilm Workstation 3.3, Merge Healthcare, Milwaukee, WI) and all analysis was performed by one author (K.M). Three images per animal per observer were then saved in bitmap form (for a total of 6 images per animal). Bitmap form was a requirement of the pixel analysis program. Each image was opened in Microsoft Paint and a rectangular region of interest (ROI) was drawn to encompass as much of the spleen in the image as possible (Figure 9). The operating procedures of the pixel analysis program indicated that the rectangle must be the thinnest line and true red in color. Once every image had a region of interest drawn, the images were analyzed by a proprietary DOS based program (created by Dr. Michael Morgan, DVM, North Carolina State University). 43 This program counted the total number of pixels in the given ROI, and provided the number of pixels of each color and the percentage of total of each color. The five colors recognized by the software are blue, green, cyan, yellow, and red. The output of the pixel analysis was imported into a spreadsheet (Microsoft Excel, Microsoft Corp, Redmond, WA). The pixel analysis output was five numbers for each ROI, yielding a total of 30 numbers for each patient. For statistical analysis, it was desirable to create a single number for each ROI this number was meant to reflect the relative overall strain. To obtain this number, a weighted average approach was used. First, the color bar was used to determine a weighting value for each color. Figure 10 shows the color bar and the approximate levels of demarcation between colors. 21

35 The entire bar was assumed to equal 100, and the percentage of each color based on visual inspection was used to determine the relative amount of each color contained within the bar. This yielded the following weighting factors: blue = 5, cyan = 13, green = 35, yellow = 65, red = 87. Since the sizes of the ROIs were not standardized (i.e., the number of pixels in each ROI were different), the percentage of each color was used for analysis (as opposed to the absolute number of pixels of each color). Each color percentage was multiplied by its appropriate weighting factor and then averaged. An example calculation is provided in Table 2. All calculations were performed using a spreadsheet (Microsoft Excel 2011, Microsoft Corp, Redmond, WA). This pixel analysis was performed for each of the six ROIs per patient, per observer, at the initial time point. However at the recheck time point, only data for the neutral compression was used. To identify the best possible repeatability for these study conditions and operators, strain scores at maximum compression for three different regions of interest per animal, for Observer 1 only, were averaged and this data was analyzed separately. 2.4 Height Analysis To determine if a real difference existed, a waveform analysis was pursued. The initial time point at maximum compression was used for each animal, and for each observer. The most consistent, smooth waveforms in each image were chosen for analysis. Using a DICOM viewer (OsiriX, The OsiriX Foundation, Geneva, 22

36 Switzerland), a line (line A) was drawn from the baseline to the peak of three consecutive waveforms. A second line (line B) was drawn from the baseline to the top of the waveform graphing area. The three A values were averaged and then divided by B to obtain a single percentage number indicating how high the waveform peak reached. Units were in pixels. A higher percentage value inferred harder probe pressure was used (Figure 11). 2.5 StatisticalMethods Statistical analysis was performed by two of the authors (K.M and J.B.) using commercial software (SPSS Software, Version , IBM Corp, Armony, NY). The initial goal was to determine if there was a difference in strain scores obtained during minimum compression, neutral compression, and maximum compression. Descriptive statistics were calculated by operator and for the three levels of compression, and the data were tested for normality using inspection of scatterplots and the Kolmogorov-Smirnov test. Inasmuch as the distribution of the overall data set was non-normal a nonparametric Friedman s test for repeated measures was performed to identify differences in strain related to compression. When a significant difference was identified by the omnibus Friedman s test, paired comparisons were performed using Bonferroni-corrected, Wilcoxon signed-rank tests (p=0.017). The second goal was to determine the interobserver and intraobserver repeatability. Before-and-after graphs were generated to compare strain scores 23

37 obtained between and within observers. Scatterplots were used to graphically display the strain scores by spleen between the two observers. Spearman s rank-order correlation coefficient was used to assess the strength of any correlation between the two observers for examinations of the same organ. The third and final goal was to determine if one observer used a firmer compression technique than the other observer. This was assessed using a Mann- Whitney U statistic (i.e., the Wilcoxon rank sum test). Statistical significance was set at p<0.05. A correlation coefficient of was considered strong, was moderate, and <0.35 was poor. 45 Inasmuch as this is a new technology and this study represents a preliminary evaluation of this analysis in dogs, an estimate standard deviation of measurements within healthy dogs was unavailable. As an alternative, a number of potential effect sizes and the related statistic η 2 were calculated using the average correlation for among repeated measurements over the three study periods (0.41); the desired statistical power of 0.8, an α error of 0.05, and a nonsphericity correction of ε of 0.9. The estimated sample size needed for a large effect under these conditions (η 2 =0.50 was 4 dogs). The estimated sample size for a moderate effect (η 2 =0.25) was 8 dogs. The estimated sample size for a smaller effect (η 2 =0.125) was 17 dogs. Based on this approach, the study could detect an effect (difference) between two of the study periods of approximately 14.5% with a statistical power of 0.81 and α error maintained at If the lower bound correction for nonsphericity was set at the minimal value to adjust degrees of freedom (ε=0.5) the partial η 2 increased to an effect of 0.21 (21%). 24

38 Figure 8:Elastogram demonstrating maximum compression. The color map elastogram is on the lefthand side, and the B-mode display is on the right for positioning reference. The green waveform at the bottom of the image is used to provide real-time feedback on the quality of compressions. The waveform should be as smooth and even as possible. This figure is also a representation of maximum compression, where the cursor is at the peak of the waveform. 25

39 Figure 9: A red, rectangular region of interest was drawn to encompass as much of the spleen as possible, without incorporating any other structures. This image was obtained at the maximimum compression point (see white cursor on green waveform). 26

40 Figure 10: Approximate weighting factor for blue cut-off regions for determining weighting values. The = 5, cyan = 13, green = 35, yellow = 65, red = 87. Figure 11: Example of waveform height calculation. The three A values were averaged and then dividedd by B to provide the average percentage of maximum the waveforms reach. 27

41 Number Switch Function Setting in study 1 Normalization Used to select the color map normalization method Max: The calculated upper limit value is set to the maximum (soft) of the color map 2 Color Map Used for setting the 0: Soft (red), hard color map 3 Map Range Used for adjusting the range of the color map 4 Fusion Used for setting the weighting between the Elastography image and the 2D-mode image 5 Smoothing Used for adjusting the smoothing level of Elastography image 6 Frame Rate Used to adjust the frame rate 7 C Focus Used to adjust the color focus position 8 Scale Used to adjust the velocity range for acquiring Elastography data 9 Target Size Used to select the target size for Elastography 10 C Freq Used to select the transmission/reception frequency for acquiring Elastography data 11 Persistence Used to adjust the image smoothness (blue) 0.9: Displays the image with a color map that is 90% of the reference display 0.3: 2D mode image and Elastography image are displayed superimposed M: Medium-level smoothing is performed 4 50% 1 cm/s 0.6 mm Up 0 Table 1: Ultrasound machine settings for each patient. These settings were recommended by Toshiba. (Table adapted from Toshiba Aplio 500 user manual). 28

42 Blue Cyan Green Yellow Red % pixels Strain value = 20(5) + 20(13) + 20(35) + 20(65) + 20(87) = = 4100 / 5 Strain value = 820 Table 2: Example of pixel percentages and strain value calculation, shown in long-hand format for demonstration. It was performed with project data using an Excel spreadsheet. 29

43 Chapter 3: RESULTS 3.1 Dogs Fourteen dogs met the inclusion criteria. The dogs ranged in age from one year to 9 years and in weight from 6.8kg to 41.6kg. The following breeds were represented: three pit bull terrier crosses, two Labrador crosses, two Australian shepherds, two Sealyham terriers, two German shepherds, one French bulldog, one Mastiff terrier, and one wirehair terrier cross. There were 5 spayed females and 9 neutered males. Patient characteristics are summarized in Table 3 (Appendix A). 3.2 Descriptive Statistics and Comparing Compression Types The data for Observer 1 was normally distributed (Table 4, Appendix B). For Observer 2, the data were normally distributed for minimum and neutral compression, but maximum compression was not normally distributed (Table 6, Appendix B). Because all data sets were not normally distributed and could not be transformed to a normal distribution, non-parametric statistical tests were chosen. All descriptive data can be found in Tables 5 and 7, Appendix B. At minimum compression, the median strain score and interquartile rangefor Observer 1, initial time 30

44 point, was 560 ( ). For Observer 2, the median strain score and interquartile range was 644 ( ). At neutral compression, the median strain score and interquartile range for Observer 1 was (93 406). For Observer 2, the median strain score was ( ). At maximum compression, the median strain score and interquartile range for Observer 1, initial time point, was ( ). For Observer 2, the median strain score was ( ). Figure 12 summarizes data obtained for Observer 1 at the initial time point. 11/14 spleens had lower strain scores at neutral compression than at minimum or maximum compression. Friedman s test statistic was 7.000, p = (Table 8, Appendix C). Posthoc analysis indicated that strain scores at neutral compression were significantly lower than during minimum compression (p = 0.016) and maximum compression (p = 0.011) (Table 9, Appendix C). Figure 13 summarizes data obtained for Observer 2 at the initial time point. 9/14 spleens had lower strain scores at neutral compression than at minimum or maximum compression. Friedman s test statistic was 7.673, p = 0.02 (Table 10, Appendix C). Post hoc analysis indicated that strain scores at neutral compression were significantly lower than those obtained at maximum compression (p=0.005), but the differences in scores did not achieve statistical significance at minimum compression (p=0.055) (Table 11, Appendix C). 31

45 3.3 Inter-observer Results For repeatability testing, only the data obtained at neutral compression were used. This was because neutral compression was consistently lower than the other compression types, whereas the other compression types were not statistically different from one another. Repeatability between the two observers for neutral compression is shown in Figures 14 and 15. The correlation coefficient at the initial time point was , p=0.095 (Table 12, Appendix D). This is a moderate correlationbut did not achieve statistical significance. At the recheck time point, the correlation coefficient was 0.301, p=0.295 (Table 13, Appendix D). This is a poor correlation and was again not statistically significant. 3.4 Intra-observer Results Figures 16 and 17 show before-and-after plots to demonstrate intra-observer repeatability. For Observer 1, the correlation coefficient was 0.548, p=0.043, which is moderate and statistically significant (Table 14, Appendix D). For Observer 2, the correlation coefficient was 0.433, p=0.122, which wasmoderate but not statistically significant (Table 15, Appendix D). When values for Observer 1 were averaged, the correlation coefficient was poor at 0.209, p=0.747 (Table 16, Appendix D). This is shown in Figure

46 3.5 Compression Wave Height Analysis Results The height data were assessed for normality (Table 17, Appendix E). Using the Mann-Whitney U test, Observer 2 had significantly more probe pressure than Observer 1 (p=0.002) (Table 18, Appendix E). This is demonstrated in Figure 19. Observer 2 had significantly higher strain scores at maximum compression than Observer 1 (p=0.03). There was no statistically significant difference in strain score at minimum compression (0.272) or neutral compression (0.177) (Table 19, Appendix E). 33

47 Figure12: Summary of strain scores for Observer 1, initial time point. 34

48 Figure 13: Summary of strain scores for Observer 2, initial time point 35

49 Figure 14: Before-and-after graph demonstrating inter-observer repeatability at the initial time point. 36

50 Figure15: Before-and-after graph demonstrating inter-observer repeatability at the recheck time point. 37

51 Figure16: Before-and-after graph demonstrating intra-observer repeatability for Observer 1. 38

52 Figure17: Before-and-after graph demonstrating intra-observer repeatability for Observer 2. 39

53 Figure18: Intra-observerr repeatability for Observer 1, with three compression types averaged together. 40

54 Figure19: Comparison of the amount of probe pressure exerted by each observer. Observer 2 used significantly more pressure than Observer1 (p=0.002). 41

55 Chapter 4: DISCUSSION 4.1 Comparing Compression Types The results from this study suggest that elastography images made at neutral compression have significantly more regions of blue or cyan color than those made at maximum compression. Blue indicates harder or stiffer tissue. For both observers, there was no significant difference between values obtained at maximum or minimum compression. Therefore, it can be concluded that when selecting a time point to analyze elastography images on the normal canine spleen, one should be consistent with the compression level (either neutral, or minimum/maximum) and should not interchange the two. Examining the theory behind the waveform feedback can help determine if measuring strain at neutral compression is preferable to measuring at maximum or minimum. As discussed, when using the Toshiba Aplio 500, the waveforms at the bottom of the image are used to provide real-time feedback on the quality of compressions. Another way to look at the elastogram and associated waveform is as a series of alternating pressure phases and relaxation phases. Above the baseline is Pressure Phase and below the baseline is Relaxation Phase. In other words, as the sinusoidal wave goes up, this is compression. As the sinusoidal wave goes down, this is 42

56 relaxation. According to the Toshiba Aplio 500 user manual, 16 relaxation phase is the theoretically most accurate time to measure strain ratios because it is less affected by unbalanced pressure and the tissue will demonstrate its true elastic properties. Based on these ideas, measuring at baseline (on the downslope of the curve) might be the most correct. 4.2 Repeatability Interobserver and intraobserver repeatability were moderate at best in this study and insufficient for clinical decision-making. In the before-and-after graphs presented in Chapter 3, perfect repeatability would result in each case being a horizontal line. While some of the lines were close to horizontal, many were diagonal, indicating very different values were obtained from one time-point to the next or from one observer to the next. Onereason for the poor repeatability could be that only one moment on the waveform was chosen to analyze. To investigate whether or not this may have played a role, the analysis was repeated after averaging three consecutive points on the waveform (i.e., three neutral compression points in a row). When this was performed the before-andafter graph showed an improvement in the repeatability. However, there was no statistical improvement in the correlation coefficient. Again, this may have been due to poor power. So in this case it is reasonable to conclude that the graphical depiction is more reliable than the statistical number, and that averaging at least three compression points together will yield a more repeatable result. 43

57 Our repeatability is different compared toa study performed in the canine gastrocnemius tendon where repeatability was good to excellent. 46 However, this study used acoustoelasticity technology, which measures the speed of sound in tissue instead of amount of displacement. Therefore, it is difficult to compare repeatability between these two differing technologies. A study using strain elastography (as used here) in normal equine tendons showed moderate to good repeatability. 43 However, the feedback mechanism for monitoring the quality of manual compressions was different than the one used here. The machine used in that study was a MyLab 70 XVG machine with ElaXto software. With this software an image of a coil is used to provide feedback on the quality of compressions; the more turns of the coil that were filled-in during a compression indicated a more reliable the compression strength (Figure 20). It seems reasonable that filling in a coil would be more objective than monitoring the evenness of a waveform. Another reason for the better repeatability in the equine tendonstudy may have been related to the nature of musculoskeletal imaging compared to abdominal organ imaging. When performing strain elastography, one should minimize the lateral movement of the organ being compressed as much as possible. There will be less potential for lateral motion of a somewhat rigid tendon with compression than with a relatively mobile spleen inside an abdominal cavity. A study using strain elastography (as was used here) in normal canine spleens showed moderate to excellent repeatability. 47 This study also uses a type of waveform to provide real-time quality control of the compressions. Therefore, this may be a more 44

58 appropriate comparison to our study. However, it was our experience that our waveform was quite sensitive to small variations in compression pressure, and it is possible that the machine used in the other study was less sensitive to variation and therefore more repeatable. Controlled studies comparing machine types would need to validate our conjecture. Each observer received limited training in performing elastography prior to the onset of this study. In people, elastography training improved the repeatability of all observers, even an experienced observer. 48 Similarly, in a study that examined results of liver stiffness measurements over 5 years and had 13,369 examinations, better results were obtained after users performed at least 500 studies. 49 Even with the training and several practice sessions, neither operator in this study reached nearly 500 examinations. Therefore, lack of extensive training and experience likely played the largest role in the less than ideal repeatability. 4.3 Effect of Compression Strength As mentioned earlier, a subjective waveform was used to assess quality of compressions. Each observer attempted to make the waveform as smooth and even as possible, with each peak reaching a similar height. While each observer was not blinded to the compressions of the other observer, and so theoretically the second observer could have modeled their compression after the first observer, the waveforms still proved difficult to reproduce between observers. Figure 21 shows an example of a waveform obtained by Observer 1, and a waveform obtained by Observer 2 in the same patient. 45

59 Thewaveforms are visually different from one another, even though it is in the same patient with the same machine settings. The waveform height is associated with compression strength and speed. Harder external compressions will result in higher waveforms. The waveform analysis portion of this study sought to determine if there was a significant difference between compression strength between each observer. We determined that the waveforms from Observer 2 were significantly higher than Observer 1, indicating that Observer 2 used firmer compressions. Comparing strain scores obtained by Observer 1 with strain scores obtained by Observer 2, values obtained by Observer 2 were significantly higher (p = 0.03) than those obtained by Observer 1 at maximum compression only. This indicates that the significantly increased probe pressure used by observer 2 resulted in significantly higher strain scores. It would seem that if one uses firmer compression, the tissue being compressed would be more likely to undergo greater deformation and therefore appear more elastic. However, it is possible that the quick, firm compression would not allow time for the compression wave to propagate through the tissue, resulting in a falsely elevated strain score. Also, there is a risk of more lateral motion with firmer compressions, which could alter the elastogram. 46

60 4.4 Limitations There were several limitations to this study. First, the amount of probe pressure was not standardized between observers. One observer used significantly different pressure than the other observer. Becausetransducer pressure can influence strain measurement, 50 and training plus experience increase repeatability, we recommend that users practice elastography as much as possible before data collection. Additionally, there were other factors we did not or could not control that may have influenced the results. We could not fully control the amount of lateral displacement that occurred in the spleen when manual compressions were begun although we did attempt to keep the same portion of the spleen in the image during compressions. While strain elastography measures axial displacement much more accurately than lateral displacement, it is possible the some lateral displacement or slippage of the ultrasound probe occurred and this may have changed the elastogram. For more superficial abdominal structures, it would be possible to manually stabilize a structure to minimize lateral displacement of the structure. Patient factors likely influenced our results. None of the animals were sedated, and so there were variations in patient compliance. Some animals had very tense abdomens, which may have altered the compressibility. A variety of breeds and patient sizes were used, therefore patient conformation varied. Some deeper-chested dogs had a spleen that was located more within the costal arch, which limited compressibility. Our scenario mimics clinical situations, illustrating the need for more standardization if elastography is to be used clinically. 47

61 Finally, cytology and histopathology were not performed on the spleens. Because of this, it is possible that some spleens were not actually normal. 4.5 Future Research There are several experiments that could come from this pilot study. One is investigating the effects of sedation on the elasticity profiles of canine spleens. Similarly, measuring body wall elasticity in addition to splenic elasticity could be considered to evaluate the effect (if any) that body wall tension has on elasticity measurements. Further refining the method of image analysis is needed, preferably with a larger number of subjects. In this study a region of interest (ROI) was drawn to encompass as much of the spleen as possible; further research could investigate ROI size and placement, as well as inclusion of outside structures. The effect of obtaining multiple elastograms throughout a single spleen and seeing if this changes overall strain scores for an individual patient could be considered as a next step. Additionally, averaging more than three phases of the elastogram may also influence strain scores. Finally, the effect of operator experience cannot be overlooked. As previously mentioned, operator training greatly improves reliability and repeatability of elastography. Providing more intensive training and/or performing a similar study again after each observer has had the opportunity to practice on many more animals would allow assessment of the effect of operator training in veterinary patients. 48

62 4.6 Conclusions In conclusion, ultrasound elastography of the canine spleen is possible, and results from this study suggest that measurements should be obtained at the neutral compression point when using a Toshiba Aplio 500 machine. However, interobserver and intraobserver repeatability is poor to moderate, which is likely due to operator and patient variables. Further studies are needed to try to standardize these variables and assess their impact on repeatability before splenic elastography in dogs can be deemed clinically useful. 49

63 Figure 20: Example of an elastogram obtained with a MyLab 70 XVG machine, ElaXto software. The green coil in the bottom right-hand corner of each image was used to provide feedback on compression quality the more green the coil, the better the compression

64 A B Figure 21:Elastogram comparison. The elastogram in A was performed by observer 1 and the elastogram in B was performed by observer 2. It can be seen that, while this is the same patient with the same machine settings, the waveform in B is quite different in shape than the waveform in A. 51