Fully Automatic Pediatric Echocardiography Segmentation Using Deep Convolutional Networks Based on BiSeNet

Accurate segmentation of pediatric echocardiography is an essential preprocessing step for a wide range of analysis tasks. Currently, it highly relies on sonographer's manual segmentation, which is time-consuming and redundant, and therefore might lead to mistakes. In this paper, we present a deep learning method based on Bilateral Segmentation Network (BiSeNet) to fully automatic segment pediatric echocardiography images in 4 chamber view. BiSeNet consists of two paths, a spatial path for capturing low-level spatial features, and a context path for exploiting high-level context semantic features. In addition, a feature fusion module is used to fuse features learned by both the two paths. Experiments based on our selfcollected dataset shows that our method achieves 0.932, and 0.908 in term of Dice index in the left ventricle and left atrium segmentation task, which outperforms different state-of-the-art U-Net architectures.