Session SA2.2

Estimation of Coronary Atherosclerotic Plaque Composition Based Only on Grey-Scale Intravascular Ultrasound Images

FJR Sales*, JLAA Falcao, BAA Falcao, SS Furuie, PA Lemos

University of São Paulo
Sao Paulo, Brazil

Intravascular ultrasound (IVUS) has been widely used in cardiovascular interventions due to its unique capability to provide in vivo visualization of the lumen and arterial wall of coronary arteries. Several studies have shown that atherosclerotic plaque composition has been increasingly credited as a major factor modulating the clinical course and the risk of future coronary events. Some techniques based on intravascular ultrasound (IVUS) have recently been developed with the main objective of assessing plaque composition. In addition, these approaches are derived from the spectral analysis of backscattered IVUS signals, which were processed before image formation. Conversely, conventional grey-scale IVUS images are based only on amplitude envelope of ultrasonic signals, discarding some frequency content. This way, tissue characterization by spectral features, as previously described, has been disabled. In this work, a computational tool was developed for evaluation of coronary atherosclerotic plaque composition, as assessed by grey-scale IVUS, but without using backscattered radiofrequency attributes. Textural analysis from atheromatous lesions were combined with pattern recognition techniques in order to solve this problem.The new analytic method was compared with the results of coronary tissue characterization by intravascular ultrasound with virtual histology (VH-IVUS) imaging (one of the spectral analysis of backscattered IVUS signals). According to VH-IVUS, there are four distinct types of tissue: fibrous, fibro-fatty, necrotic-core and dense calcium. Haralick co-occurrence matrix and modified Hu moment invariants have been used for feature extraction from regions of interest. In the classification step, a k-nearest neighbor decision rule with Euclidean distance has been designed for performance evaluation. A preliminary test sample with 10 coronary arteries from 5 different patients, totaling 95923 regions of interest, resulted into an average error-rate of 4.00%. These findings are encouraging and further analyses are currently ongoing to obtain more robust and more reliable results, especially concerning the choice of regions of interest and features selection.

(Abstract Control Number: 216)