Inter-Slice Context Residual Learning for 3D Medical Image Segmentation

Automated and accurate 3D medical image segmentation plays an essential role in assisting professionals to evaluate disease progresses make fast therapeutic schedules. Although deep convolutional neural networks (DCNNs) have widely applied this task, the accuracy of these models still need be further improved mainly due their limited ability context perception. In paper, we propose residual network (ConResNet) for images. This model consists encoder, a decoder, decoder. We design module use it bridge both decoders at each scale. Each contains mapping attention mapping, formal aims explicitly learn inter-slice information latter uses such as kind boost accuracy. evaluated on MICCAI 2018 Brain Tumor Segmentation (BraTS) dataset NIH Pancreas (Pancreas-CT) dataset. Our results not only demonstrate effectiveness proposed learning scheme but also indicate that ConResNet is more than six top-ranking methods brain tumor seven pancreas segmentation. Code available https://git.io/ConResNet