IMedSeg: Towards efficient interactive medical segmentation

Zidi Shi, Ting Chen, Qian Zhou, Hua Zou, Xuan Xiao

Abstract: Highlights•A two-stage framework facilitates user interaction and precise segmentation refinement.•The memory-optimized backbone of ICS-Unet improves efficiency and segmentation accuracy.•Local mask segmentation leverages user clicks while preserving mask integrity.•CSA computes feature correlations for accurate pixel-level classification.•Compatibility with existing IMS architectures enhances its utility.