Uncertainty-aware temporal self-learning (UATS): Semi-supervised learning for segmentation of prostate zones and beyond

作者：

Highlights：

• We propose a semi-supervised CNN method named uncertainty-aware temporal self-learning (UATS) for medical segmentation.

• UATS leverages performance gains from temporal ensembling and uncertainty-guided self-learning.

• UATS surpasses fully supervised performance on prostate zone segmentation and achieves human observer results quality.

• Experiments demonstrate its generalizability on multiple biomedical segmentation tasks.

摘要

•We propose a semi-supervised CNN method named uncertainty-aware temporal self-learning (UATS) for medical segmentation.•UATS leverages performance gains from temporal ensembling and uncertainty-guided self-learning.•UATS surpasses fully supervised performance on prostate zone segmentation and achieves human observer results quality.•Experiments demonstrate its generalizability on multiple biomedical segmentation tasks.

论文关键词：Semi-supervised deep learning,Biomedical segmentation,Prostate zones

论文评审过程：Received 30 October 2020, Revised 9 February 2021, Accepted 7 April 2021, Available online 10 April 2021, Version of Record 24 April 2021.

论文官网地址：https://doi.org/10.1016/j.artmed.2021.102073