AstMatch: Adversarial self-training consistency framework for semi-supervised medical image segmentation

Guanghao Zhu, Jing Zhang, Juanxiu Liu, Xiaohui Du, Ruqian Hao, Yong Liu, Lin Liu

Abstract: Highlights•AstMatch is a novel method for semi-supervised medical image segmentation.•AstMatch focuses on high-level consistency regularization and pseudo-label quality.•Discriminator scores reflect the quality of segmentation predictions.•AstMatch demonstrates state-of-the-art performance over existing methods.•The method achieves excellent results with only 5 % or 10 % of labeled data.