Ultrasonographic characterization and identification of symptomatic carotid plaques

Carotid plaques are the main cause of neurological symptoms due to distal embolization or flow reduction. An objective classification of such lesions into symptomatic or asymptomatic is crucial for optimal treatment planning. The paper proposes a diagnostic framework to tackle this problem which consists of image processing, plaque detection, feature extraction and classification using AdaBoost in B-mode ultrasound (BUS) images. Image processing includes greylevel normalization, envelope Radio-Frequency (eRF) image retrieval, de-speckling and speckle extraction early proposed by the authors. The estimated images are used to extract a set of echo-morphology and texture features which are fused with clinical information provided by the physician. The classification performance, assessed by means of the Leave-One-Patient-Out (LOPO) cross-validation technique applied to a population of 44 symptomatic and 102 asymptomatic plaques, yields 99.2% overall accuracy and 100% sensitivity in classifying symptomatic vs. asymptomatic plaques. Feature analysis and comparison of classification results obtained with different feature sets suggest the usefulness of an extended feature set here proposed for the identification of symptomatic plaques among the traditional ones used in the literature.