Automatic segmentation of the Epicardium and Endocardium using convolutional neural network

Automatic segmentation of the Epicardium and Endocardium plays an important role in the diagnosis of myocardial ischemia. Due to the speckle noise of ultrasound images and the complexity of cardiac tissue, the segmentation is still manual or semi-automatic. A fully automatic segmentation method based on Convolutional Neural Network (CNN) is proposed in this paper: localization segmentation method, which can obtain the position of the Epicardium and Endocardium based on the output of the network, then the segmentation is completed according to reprocessing based on the position. We concentrate the solutions of the following problems to supervised learning models: 1) the need of training images which have obvious features; 2) the design of network; 3) correlation between group images. The performance of our approach is evaluated using four different datasets containing 800 ultrasound images of Sprague-Dawley rats. The quality grades of the manual and automatic segmentation methods have been calculated, and the proportion of poor segmentation quality is decreased from 3.48% to 2.39%, which shows that our proposed method has improved segmentation accuracy and stability. The efficiency of our fully automatic segmentation method is higher than traditional segmentation strategies. The proposed method can optimize the clinical procedure and decrease the variability of manual segmentations.