Labeled-to-unlabeled distribution alignment for partially-supervised multi-organ medical image segmentation

Xixi Jiang, Dong Zhang, Xiang Li, Kangyi Liu, Kwang-Ting Cheng, Xin Yang

Abstract: Highlights•This paper highlights the distribution mismatch issue in partially-supervised segmentation.•The proposed LTUDA reduces distribution discrepancy and in turn produces unbiased pseudo-labels.•To bridge the distribution of labeled data and unlabeled data, we propose a cross-set data augmentation strategy.•To facilitate learning aligned and compact feature representations, we design a prototype-based distribution alignment method.