Medical Image Segmentation Based on Recurrent Residual Convolution Neural Network

Abstract

In order to solve the problem of network degradation when using deep learning for medical image segmentation, and to achieve more accurate image segmentation effect by extracting larger scale feature mapping, an image segmentation model based on recurrent residual convolutional neural network is proposed. Firstly, the recurrent convolution unit is introduced to realize the feature extraction on discrete step size, improves the utilization of image context semantic information, and realizes more extensive feature mapping extraction. Then, combined with the residual learning unit and the recurrent convolution unit, the recurrent residual convolution unit is formed to replace the feed-forward convolution unit of the ordinary convolution neural network to solve the network degradation problem faced by the deep network model. Finally, the full-scale skip connection is introduced to fuse the feature images under different scales to generate the segmented image. The experimental results of three datasets compared with the other four algorithms in PyTorch environment show that the proposed algorithm has better segmentation performance and higher image segmentation accuracy.