American Institute of Ultrasound in Medicine <strong>Proceedings</strong> J Ultrasound Med 32(suppl):S1–S134, 2013Transplant ImagingModerator: Susan Ackerman, MDSonographic Evaluation of Liver TransplantsSusan Ackerman Medical University of South Carolina,Charleston, South Carolina USAThe purpose of this lecture is to discuss the use of ultrasound toevaluate liver transplants. In addition to normal or expected findings inthe post-transplant patient, complications will also will be discussed.SPECIAL INTEREST SESSIONSTUESDAY, APRIL 9, 2013, 4:00 PM–5:30 PMAbdominal and Lower Extremity Arterial Imaging:Pitfalls and MisdiagnosesModerator: Jennifer McDowell, MM, RDMS, RT, RVTThis session will review case studies in abdominal and lowerextremity vascular imaging and demonstrate examples of technical pitfalls,artifacts, and misdiagnoses to learn how to prevent them.Hands-on Ultrasound-Guided Vascular AccessModerator: Jason Nomura, MD, RDMSIn this session, participants will be given a short didactic lectureon patient preparation, sterile technique, and basics of ultrasound-guidedvascular access and fluid aspiration. This will be immediately followedby hands-on practice and interactions at various stations with expert facultyto learn and improve on techniques.S59
American Institute of Ultrasound in Medicine <strong>Proceedings</strong> J Ultrasound Med 32(suppl):S1–S134, 2013SCIENTIFIC SESSIONSTUESDAY, APRIL 9, 2013, 4:00 PM–5:30 PMBasic Science: Tissue Characterization, Part 2Moderators: Michael Oleze, PhD, James Miller, PhD1540933 Three-Dimensional In Vivo Prostate Shear Wave ElasticityImage ReconstructionStephen Rosenzweig, 1 * Mark Palmeri, 1 Samantha Lipman, 1Ned Rouze, 2 Evan Kulbacki, 2 John Madden, 2 Thomas Polascik,2 Kathryn Nightingale 1 1 Biomedical Engineering, DukeUniversity, Durham, North Carolina USA; 2 Duke UniversityMedical Center, Durham, North Carolina USAObjectives—Shear wave elasticity imaging (SWEI) andacoustic radiation force impulse (ARFI) imaging techniques have beenreported to portray cancer and other pathologies as stiffer than the surroundingtissue. 1,2 Previous work has shown artifacts in reconstructingSWEI images due to reflected waves. 3,4 In this work, methods for reconstructingSWEI images designed to reduce these artifacts were validatedin phantoms, applied to in vivo data, and compared to concurrently acquiredARFI data.Methods—Data were collected using a Siemens AcusonSC2000 and an ER7B transducer (Mountain View, CA) and a transducerrotation stage. The pulse sequence consisted of rapidly pushing at multiplefoci (SSI-type push 4 ) and tracking the resulting displacement using 16parallel receive beams. The beam sequence was then translated laterally0.7 mm and repeated across the field of view. The resulting SWEI datawere spatially and temporally aligned to generate an image using highspatial sampling of the data. Separate left and right wave propagation imageswere generated along with combining the data via mean and maximumvalue approaches; these were compared to matched ARFI images inboth calibrated CIRS (Norfolk, VA) phantoms and radical prostatectomypatients from an ongoing Institutional Review Board–approved study.Results—The contrast to noise ratios (CNRs) in the phantomdata for the different combined SWEI methods were higher than those forthe individual propagating waves. For example, a 10-mm cylindrical targetwith a 4:1 stiffness ratio had SWEI image CNR values of 1.65 (left),1.47 (right), 2.59 (mean), and 3.74 (maximum). We will present data fromall methods in various phantoms in addition to results from prostatectomypatients, after the whole-mount pathology is registered in 3D to the SWEIand ARFI volumes.Conclusions—The high spatial sampling of SWEI data obtainedfrom concurrent acquisition with ARFI data affords opportunitiesfor reducing SWEI image artifacts and improving the CNR. We are nowapplying the algorithms to data from an ongoing in vivo study to detectpathologies in the prostate.References1. Zhai L, et al. Ultrasound Med Biol 2012; 38:50–61.2. Barr RG, et al. Ultrasound Q 2012; 28:13–20.3. Rouze N, et al. IEEE Trans Ultrason Ferroelectr Freq Control 2012;59:1729–1740.4. Deffieux T, et al. IEEE Trans Ultrason Ferroelectr Freq Control 2011;58:2032–2035.1511996 Hashimoto’s Thyroiditis Tissue Characterization and PixelClassification Using UltrasoundAgnieszka Witkowska, 1 U. Rajendra Acharya, 2,3 RatnaYantri, 2 Filippo Molinari, 4 Witold Zieleznik, 5 Justyna Tumidajewicz,5 Beata Stepien, 5 Ricardo Bardales, 6 Jasjit Suri 7, 8 *1Internal Medicine, Diabetology, and Nephrology, MedicalUniversity of Silesia, Katowice, Poland; 2 Electronics andComputer Engineering, Ngee Ann Polytechnic, Singapore;3Biomedical Engineering, University of Malaya, Kuala Lumpur,Malaysia; 4 Electronics and Telecommunications, PolitecnicoTorino, Torino, Italy; 5 Internal Medicine Practice, Silesia,Poland; 6 Outpatient Pathology Associates, Sacramento,California USA; 7 Global Biomedical Technologies, Roseville,California USA; 8 Biomedical Engineering, Idaho State University,Pocatello, Idaho USAObjectives—Hashimoto’s thyroiditis (HT) is the most commontype of inflammation of the thyroid gland, and accurate diagnosis of HTwould be helpful to better manage the disease process and predict thyroidfailure. This paper presents a computer-aided diagnostic (CAD) techniquethat uses grayscale features and classifiers to provide a more objective andreproducible classification of normal and HT-affected cases.Methods—Thyroid images were obtained from 68 normal and82 patients affected by HT (a total of 150 patients). In this paradigm, weextracted grayscale features based on entropy, Gabor wavelet, moments,image texture, and higher-order spectra from the 100 normal and 100 HTaffectedthyroid ultrasound images. Significant features were selectedusing the t test. The resulting feature vectors were used to build the following3 classifiers using a 10-fold stratified cross-validation technique:support vector machine (SVM), K-nearest neighbor (KNN), and radialbasis probabilistic neural network (RBPNN).Results—Our results show that a combination of 12 featurescoupled with the SVM classifier with the polynomial kernel of order 1and linear kernel gives the highest accuracy of 80%, sensitivity of 76%,specificity of 84%, and positive predictive value (PPV) of 83.3% for thedetection of HT.Conclusions—The proposed CAD system uses novel featuresthat have not yet been explored for HT diagnosis. The technique is noninvasive,cost-effective, fast, and automatic and provides a more objectiveand reproducible classification of the thyroid in normal and HT-affectedpatients. Even though the accuracy is only 80%, the presented preliminaryresults are encouraging to warrant analysis of more such powerfulfeatures on larger databases.Table 1. Classifier Performance MeasuresAccuracy, PPV, Sensitivity, Specificity,TN FN TP FP % % % %SVM linear8 2 8 2 80 83.3 76 84SVM poly 18 2 8 2 80 83.3 76 84SVM RBF8 3 7 2 78.5 82.3 74 83KNN 7 2 8 3 75.5 75.6 77 74RBPNN 8 4 6 2 74 80.3 64 84FN indicates false-negative; FP, false-positive; RBF, radial basis function; TN,true-negative; and TP, true-positive.S60