AM-FM representations for the characterization of carotid plaque ultrasound Images

Abstract

Stroke is the third leading cause of death in the western world and a major cause of disability in adults. The objective of this work was to investigate the use of AM-FM representations for the characterization of carotid plaques ultrasound images for the identification of individuals with asymptomatic carotid stenosis at risk of stroke. To characterize the plaques using AM-FM features, we compute (i) the instantaneous amplitude, (ii) the instantaneous frequency magnitude and (iii) the instantaneous frequency angle in order to capture directional information. For each AM-FM feature, we compute the histograms over the plaque regions. The statistical K-nearest neighbour (KNN) classifier was implemented for the classification of the carotid plaques into symptomatic or asymptomatic using the leave-one-out methodology. Tests were carried out on a dataset of 274 carotid plaque ultrasound images (137 symptomatic + 137 asymptomatic), which showed that the AM-FM features performed slightly better than the traditional texture features and gave better results than simple histogram. Best results were obtained when a combination of the three AM-FM representations was used reaching a classification success rate of 71.5%. © 2009 Springer Berlin Heidelberg.