Right Ventricle Segmentation of Cine MRI Using Residual U-net Convolutinal Networks

Abstract

Right ventricle (RV) segmentation is difficult due to the variable shape and ill-defined borders of the RV. In this paper, we propose a method to segment RV using a residual U-net convolutional network. A U-net shaped network structure is employed in our method to extract RV features in the encoding layers and make end-to-end decisions in the decoding layers. In the encoding layers, several residual blocks are cascaded extract RV features. In the decoding layers, convolutional layers are employed to make the RV predication. Our network is light with less parameters compared with state-of-art networks. Experiments on public datasets demonstrate that our network outperforms most existed automated segmentation method in respect of several commonly used evaluation measures.