Weak label based Bayesian U-Net for optic disc segmentation in fundus images

作者：

Highlights：

• We propose a Bayesian model for optic disc segmentation without manual annotation.

• The proposed Bayesian U-Net considers uncertainty introduced by noisy labels.

• We explore our Bayesian U-Net under the Expectation-Maximization framework.

摘要

•We propose a Bayesian model for optic disc segmentation without manual annotation.•The proposed Bayesian U-Net considers uncertainty introduced by noisy labels.•We explore our Bayesian U-Net under the Expectation-Maximization framework.

论文关键词：Optic disc segmentation,Bayesian U-Net,Expectation-maximization,Weak labels,Fundus image

论文评审过程：Received 3 June 2021, Revised 18 January 2022, Accepted 20 February 2022, Available online 26 February 2022, Version of Record 12 March 2022.

论文官网地址：https://doi.org/10.1016/j.artmed.2022.102261