Abstract: Since medical image pixel-level annotations are expensive and difficult to obtain, semi-supervised learning methods could reduce the reliance of deep learning models on annotated data.