FAFuse: A Four-Axis Fusion framework of CNN and Transformer for medical image segmentation

Shoukun Xu, Dehao Xiao, Baohua Yuan, Yi Liu, Xueyuan Wang, Ning Li, Lin Shi, Jialu Chen, Ju-Xiao Zhang, Yanhao Wang, Jianfeng Cao, Yeqin Shao, Mingjie Jiang

Abstract: Highlights•Based on CNN and Transformer, the FAFuse framework can effectively integrate fine-grained local features and high-level semantic context information through FAF.•Four-Axis attention can effectively capture non-local and multi-directional contextual information. At the same time, long-range dependencies between feature maps can be captured.