Efficient Image Segmentation of Cardiac Conditions after Basketball Using a Deep Neural Network

Abstract

The evaluation of heart health status is the reference standard for measuring the intensity of exercise performed by different individuals. Thus, the effective analysis of heart conditions is an important research topic. In this study, we propose a system designed to segment images of the right ventricle. In this system, the right ventricle of the heart is segmented using an improved model called RAU-Net. The sensitivity and specificity of the network are enhanced by improving the loss function. We adopted an extended convolution rather than ordinary convolution to increase the receptive field of the network. In the network-sampling phase, we introduce an attention module to improve the accuracy of network segmentation. In the encoding and decoding stages, we also introduce three residual modules to solve the gradient explosion problem. The results of experiments are provided to show that the proposed algorithm exhibited better segmentation accuracy than an existing algorithm. Moreover, the algorithm can also be trained more rapidly and efficiently.