Skip to Main Content
The selection of carotid atherosclerosis patients for surgery or stenting is a crucial task in atherosclerosis disease management. In order to select only those symptomatic cases who need surgery, we have, in this work, presented a computer-aided diagnostic technique to effectively classify symptomatic and asymptomatic plaques from B-mode ultrasound carotid images. We extracted several grayscale features that quantify the textural differences inherent in the manually delineated plaque regions and selected the most significant among these extracted features. These features, along with the degree of stenosis (DoS), were used to train and test a support vector machine (SVM) classifier using threefold stratified cross-validation using a data set consisting of 160 (50 symptomatic and 110 asymptomatic) images. Using 32 features in an SVM classifier with a polynomial kernel of order 1, we obtained the best accuracy of 90.66%, sensitivity of 83.33%, and specificity of 95.39%. The DoS was found to be a valuable feature in addition to other texture-based features. We have also proposed the plaque risk index (PRI) made up of a combination of significant features such that the PRI has unique ranges for both plaque classes. PRI can be used in monitoring the variations in features over a period of time which will provide evidence on how and which features change as asymptomatic plaques become symptomatic.