Texture-driven coronary artery plaque characterization using wavelet packet signatures

Abstract

High-frequency ultrasound transducers are being widely used to generate high resolution, real time, cross-sectional images of the coronary arteries. In this paper, we present a robust unsupervised texture-derived technique based on multi-channel wavelet frames to delineate atherosclerotic plaque compositions. The intravascular ultrasound (IVUS) signals were acquired from coronary arteries dissected from 32 diseased cadaver hearts employing 40 MHz mechanically rotating, single-element transducers. The wavelet packet representations were classified using a K- means clustering algorithm to generate IVUS-histology color maps (IV-HCMs) and categorize tissues in lipidic, fibrotic and calcified. Finally, two independent observers evaluated the results contrasting the histology images corresponding to the IV-HCMs. Our results show that the proposed algorithm may have great potential as an alternative to existing spectrum-based classification techniques.