FCTRANSNET: A FULL-SCALE CHANNEL TRANSFORMER NETWORK FOR MEDICAL IMAGE SEGMENTATION

There is still incompatibility between codecs of the traditional medical image segmentation model. To solve this problem, this paper proposes an efficient and powerful full-scale channel transformer network (FCTransNet). FCTransNet followed the successful U-shaped architecture design and had two appealing designs: 1) A full-scale channel transformer (FCTrans) was utilized instead of skip connections. It explored a rich global context at full scale by learning sequence representations of encoder features to bridge the semantic gap between codecs. 2) Multiple convolution sequences with different receptive fields were used to enhance the encoder and extract more diverse local semantic features at each scale. Experimental results on the Synapse multi-organ segmentation dataset and ACDC (Automated Cardiac Diagnosis Challenge) dataset show that the performance of FCTransNet is better than other latest methods.