DCTNet: A Hybrid Model of CNN and Dilated Contextual Transformer for Medical Image Segmentation

Abstract

Medical image segmentation is a prerequisite for the development of medical systems, especially for disease diagnosis and treatment planning. Due to the inherent limitations of convolutional operations, convolutional neural networks (CNNs), although they have become the consensus for various medical image segmentation tasks, show limitations in extracting remote image features. transformer shows superior performance in extracting remote image features, but it cannot capture low-level features. Existing studies have shown that combining CNN and Transformer can give better results. In this paper, we propose a Transformer module that can effectively combine low-level and high-level features while maintaining feature consistency, leveraging the rich context between adjacent keys, and combining it with the classical model of convolutional neural networks to form a new approach to medical image segmentation that effectively connects CNN and Transformer. We conducted experiments on the dataset. The experimental results show that our proposed algorithm is the best in all segmentation metrics.