Classification of carotid artery abnormalities in ultrasound images using an artificial neural classifier

This work presents a computer-aided system for the identification of plaques and atherosclerosis of carotid abnormalities and the individuals at risk of stroke. Intima Media Thickness (IMT) of carotid artery is the standard biomarker of subclinical atherosclerosis and plaques. Conventional IMT measurement by expert sonologist is time consuming, associated with subjectivity and the process becomes difficult when the number of patients is very large. This paper proposes a standard protocol to diagnose patients efficiently and the process is made extremely fast. In this paper, the decision making ability of an artificial learning machine is investigated in carotid ultrasound artery image classification. Architecture with multilayer Back Propagation Network (BPN) using Levenberg-Marquardt training with good generalization capabilities and extremely fast learning capacity that overcomes the local minima problem of generalized BPN has been proposed. Carotid images are preprocessed, normalized and segmented to extract eighteen different feature sets and given as input to Artificial Neural Network (ANN). The selected features are found to be the good choice of feature vectors and have the ability to discriminate between normal and abnormal image. The proposed system is robust to any ultrasound image artifact. ANN classifier is evaluated using 361 ultrasound images. The efficiency is measured by validating the outputs of this decision support system with that of medical experts. This system improves the classification rate, reaching the diagnostic yield of 89.43%. The simulation results depicts that ANN achieves good classification accuracies with less implementation complexity when compared with manual operation.