Comparative study of clustering algorithms in order to virtual histology (VH) image segmentation

Abstract

Atherosclerosis is the deadliest type of heart disease caused by soft or “vulnerable” plaque (VP) formation in the coronary arteries. Recently, Virtual Histology (VH) has been proposed based on spectral analysis of Intravascular Ultrasound (IVUS) provides color code of coronary tissue maps. Based on pathophysiological studies, obtaining information about existence and extension of confluent pool’s component inside plaque is important. In addition, plaque components’ localization respect to the luminal border has major role in determining plaque vulnerability and plaque–stent interaction. Computational methods were applied to prognostic the pattern's structure of each component inside the plaque. The first step for post-processing of VH methodology to get further information of geometrical features is segmentation or decomposition. The medical imaging segmentation field has developed to assist cardiologist and radiologists and reduce human error in recent years as well. To perform color image clustering, several strategies can be applied which include traditional hierarchical and nonhierarchical. In this paper, we applied and compared four nonhierarchical clustering methods consists of Fuzzy C-means (FCM), Intuitionistic Fuzzy C-means (IFCM), K-means and SOM artificial neural networks in order to automate segmentation of the VH-IVUS images.