Computer aided decision support system for mitral valve diagnosis and classification using depthwise separable convolution neural network

Abstract

The significance of mitral valve (MV) treatment is increasing recently because of an aging population. The computer vision-based acquisition and quantification of the valve anatomy becomes helpful for surgical and intercessional planning. The right option of common treatment and implantation is pertinent for the most favorable results. Several studies reported that the decision support system (DSS) could offer decisions based on the virtual involvement planning and prediction models. Generally, the segmentation and classification of MV from the computed tomography (CT) images are highly complicated, owing to the variations in appearance and visibility. In this paper, an efficient automated DSS model is introduced using watershed segmentation with Xception model for the MV classification. It incorporates four modules: bilateral filtering (BF) based preprocessing, watershed segmentation, Xception based feature extraction and random forest (RF) classification. A watershed algorithm with channel separation is used to segment the MV images. The Xception model with random forest (RF) model is utilized for training and classifying images. A detailed simulation is performed on the CT images collected from hospitals. The presented WS-X model is tested and a comparative study is made with the relevant works to highlight its superior nature. The obtained results stressed out that the WS-X model is an appropriate model for the MV problem under various aspects.