SGBTransNet: Bridging the semantic gap in medical image segmentation models using Transformers

Yunlai Zhou, Bing Wang, Jie Yang, Ying Yang, Xuedong Tian

Abstract: Highlights•Analyzed semantic gaps in U-shaped segmentation models and summarized into sub-problems of semantic inconsistency and spatial misalignment.•The model attentively refines shallow feature maps in a global manner with the high-level semantic guidance from deep feature maps.•The model performs channel-attention synergistically on refined shallow and deep feature maps to alleviate semantic gaps from the channel perspective.