Multiple Feature Extraction for Content-Based Image Retrieval of Carotid Plaque Ultrasound Images

The extraction of multiple features from highresolution ultrasound images of atherosclerotic carotid plaques characterizing the plaque morphology and structure can be used for the retrieval of similar plaques and the identification of individuals with asymptomatic carotid stenosis at risk of stroke. The objective of this work was to develop a computer aided system that will facilitate the automated retrieval of similar carotid plaque ultrasound images based on texture, shape, morphological, histogram and correlogram features, and the neural self organising map (SOM) and the statistical Knearest neighbour (KNN) classifiers. The results in this work show that content-based image retrieval for carotid plaque image is feasible reaching a correct retrieval rate of 76%. II. MATERIAL Ultrasound scans of carotid plaques were performed using duplex scanning and color flow imaging. A total of 336 carotid plaque ultrasound images (137 symptomatic and 199 asymptomatic) were analysed. For training the system 90 symptomatic and 90 asymptomatic plaques were used, whereas for evaluation of the system the remaining 109 symptomatic and 47 asymptomatic plaques were used. The carotid plaques were labeled as symptomatic after one of the following symptoms was identified: Stroke, transient ischemic attack or amaurosis fugax.