To address the problem of insufficient annotation data in the medical field, the proposed semi-supervised learning-based medical image segmentation model C3S3 focuses on accurately identifying the boundary contour by combining competitive and contrastive learning. The result-driven contrastive learning module improves the boundary location accuracy, and the dynamic complementary competitive module generates pseudo labels with two high-performance sub-networks to improve the segmentation quality. Experimental results using MRI and CT scan datasets show that it outperforms existing state-of-the-art models, and achieves at least 6% performance improvement in 95HD and ASD metrics. The source code is available on GitHub.
