Automatic plaque segmentation based on hybrid fuzzy clustering and k nearest neighborhood using virtual histology intravascular ultrasound images

Abstract

Thin cap fibroatheroma (TCFA) or “vulnerable plaque” is responsible for the majority of coronary artery death. Virtual Histology Intravascular Ultrasound (VH-IVUS) image is a clinically available method for visualizing color coded tissue maps. However, this technique has considerable limitations in providing medical relevant information for identifying vulnerable plaque. The aim of this paper is to improve the identification of TCFA in VH-IVUS image. Therefore, this paper proposes a set of algorithms for segmentation, feature extraction, and plaque type classification to accurately identify TCFA. A hybrid model using the FCM and kNN (HFCM-kNN) is proposed to accurately segment the VH-IVUS image. The proposed technique is capable of eliminating outliers and detecting clusters with different densities in VH-IVUS image. The next process is extracting plaque features to provide an accurate definition of the unstable (vulnerable) plaque. To achieve the above contribution, five algorithms are proposed to extract significant features from VH-IVUS images. Machine learning approaches are applied for training 440 in-vivo images obtained from 8 patients. Results proved the dominance of the proposed method for TCFA detection with accuracy rate of 98.02% compared with the 76.5% obtained by the cardiologist decision. Moreover, by validation of VH-IVUS images and their corresponding Optical Coherence Tomography (OCT) images, accuracy of 92.85% is achieved.