CTransNet: Convolutional Neural Network Combined with Transformer for Medical Image Segmentation

The Transformer has been widely used for many tasks in NLP before, but there is still much room to explore the application of image domain. In this paper, we propose a simple and efficient hybrid framework, CTransNet, which combines self-attention CNN improve medical segmentation performance. Capturing long-range dependencies at different scales. To end, paper proposes an effective mechanism incorporating relative position information encoding, can reduce time complexity from O(n2) O(n), new decoder that recover fine-grained features encoder skip connection. This aims address current dilemma applications: i.e., need learn induction bias large amounts training data. layer CTransNet allows be initialized as without pre-training. We have evaluated performance on several datasets. shows superior performance, robustness, great promise generalization other tasks.