Dual Network Generative Adversarial Networks for Pediatric Echocardiography Segmentation

Abstract

Pediatric echocardiography is a commonly used medical imaging method for examining congenital heart disease (CHD). Accurate segmentation of pediatric echocardiography is usually used to derive quantitative measurements or biomarkers for subsequent CHD diagnosis and treatment planning. In order to achieve quality segmentation results, clinical pediatric echocardiography segmentation now is mainly performed by sonographers manually, which is time-consuming, labor-intensive, and highly dependent on the professional level of the sonographers. To address these issues, in this paper, we propose a novel convolutional neural network (CNN) architecture, called dual network generative adversarial networks (DNGAN). DNGAN consists of one generator and two discriminators, the generator uses parallel dual networks to extract more useful features to improve its performance. We use a dual discriminator to force the generator to learn more spatial features and segment the edges of the left heart more accurately. Experiments on the self-collected dataset shows that our proposed method achieves superior results over the state-of-the-art approaches and may help sonographers segment the left heart area faster and more accurately.

This work was supported partly by National Natural Science Foundation of China (Nos. 61871274, 61801305 and 81571758), National Natural Science Foundation of Guangdong Province (No. 2017A030313377), Guangdong Pearl River Talents Plan (2016ZT06S220), Shenzhen Peacock Plan (Nos. KQTD2016053112051497 and KQTD2015033016 104926), and Shenzhen Key Basic Research Project (Nos. JCYJ20170413152804728, JCYJ20180507184647636, JCYJ20170818142347251 and JCYJ20170818094109846).