Enhancing semi-supervised medical image segmentation with boundary awareness and Competitive Contrastive Learning

Zhikang Lin, Zhenteng Li, Jiawei Su, Lei Li, Sheng Lian

Abstract: Highlights•Novel SS-MIS framework enhances boundary precision and inter-class discrimination.•BMT module integrates multi-task learning for reduced edge segmentation errors.•DCCL enhances model discrimination by promoting compact class distributions.•Experiments on ACDC, PROMISE12, and BUSI show notable performance gains.