Self-supervised Mean Teacher for Semi-supervised Chest X-ray Classification

• DOI: 10.48550/arxiv.2103.03629
• Publication Date: 2021-01-01
• Authors: Fengbei Liu; Yu Tian; Filipe R. Cordeiro; Vasileios Belagia...; Ian Reid; Gustavo Carneiro
• Keywords: Leverage (statistics); Margin (machine learning); Supervised Learning; Labeled data

The training of deep learning models generally requires a large amount annotated data for effective convergence and generalisation. However, obtaining high-quality annotations is laboursome expensive process due to the need expert radiologists labelling task. study semi-supervised in medical image analysis then crucial importance given that it much less obtain unlabelled images than acquire labelled by radiologists. Essentially, methods leverage sets enable better generalisation using only small set images. In this paper, we propose Self-supervised Mean Teacher Semi-supervised (S$^2$MTS$^2$) combines self-supervised mean-teacher pre-training with fine-tuning. main innovation S$^2$MTS$^2$ based on joint contrastive learning, which uses an infinite number pairs positive query key features improve representation. model fine-tuned exponential moving average teacher framework trained learning. We validate multi-label classification problems from Chest X-ray14 CheXpert, multi-class ISIC2018, where show outperforms previous SOTA margin.