Improved Vulnerable Plaque Detection System using Virtual Histology-Intravascular Ultrasound Image Based on an Empirical Study

Abstract A life-threatening atherosclerosis plaque has been termed as Thin Cap Fibroatheroma (TCFA). Atherosclerotic plaque located between two borders recognized by Virtual Histology-Intravascular Ultrasound (VH-IVUS) images. In order to improve the reliability of plaque classification and TCFA detection, two approaches are suggested which are firstly based on the feature extraction technique and secondly by employing a set of ensemble classification techniques using Support Vector Machine (SVM), Radial Basis Function (RBF), and Extreme Learning Machine (ELM) as base classifiers. Plaque Burden Assessment by Local Search (PBALS) is proposed for extracting the plaque features. The geometric features are extracted from the plaque region and combined with IVUS features. In the classification part, different types of ensemble methods have been proposed and employed to identify the non-TCFA plaques from TCFA plaques with the expected reliability and robustness. 599 in-vivo IVUS along with their matching VH-IVUS images which are gathered from 10 patients are used for the experiment. According to the results, the combination of VH-IVUS with IVUS features performed better than standalone VH-IVUS features in terms of accuracy (22), sensitivity (17), and specificity (21) out of 23 different proposed methods. Furthermore, M2 model which only used 5 features (GFC3) with the combination of SVM, RFB, and ELM for the ensemble classifier performed well using either both VH-IVUS with IVUS features or single alone VH-IVUS features. Finally, prediction models which were built using HDCT-DWT features did not perform as well as the proposed set of features and ensemble prediction model.