Label noise in segmentation networks : mitigation must deal with bias
This work addresses label noise in medical image segmentation, which is crucial for improving diagnostic accuracy in healthcare, but it is incremental as it builds on prior noise mitigation research.
The study investigated the impact of biased and unbiased label errors on brain tumor segmentation in MRI data, finding that supervised and semi-supervised methods are robust to unbiased errors but sensitive to biased ones, highlighting the need for targeted mitigation strategies.
Imperfect labels limit the quality of predictions learned by deep neural networks. This is particularly relevant in medical image segmentation, where reference annotations are difficult to collect and vary significantly even across expert annotators. Prior work on mitigating label noise focused on simple models of mostly uniform noise. In this work, we explore biased and unbiased errors artificially introduced to brain tumour annotations on MRI data. We found that supervised and semi-supervised segmentation methods are robust or fairly robust to unbiased errors but sensitive to biased errors. It is therefore important to identify the sorts of errors expected in medical image labels and especially mitigate the biased errors.