Semi-supervised Context-aware Multi-Organ Segmentation

jiacheng wang, Yongkang Ma

Abstract:

The automatic segmentation of organs at risk is extremely dedicated to the clinical assistance that can significantly reduce the clinical resource cost. However, training such a good enough model usually requires a large amount of labeled data, or the model performance is
likely to meet heavy drop. Semi-supervised training strategies are proved to be an effective solution to reduce the reliance of labeled data. In this paper, we develop a powerful semi-supervised learning framework to address the label-efficient multi-organ segmentation. The experiments are conducted on the MICCAI FLARE 2022 dataset, where the results show that the semi-supervised learning strategy has significant performance boost.