Towards Fairness under Label Bias in Image Segmentation: Impact, Measurement and Mitigation
This work provides a method for detecting and mitigating label bias in image segmentation without requiring clean annotations, which is important for fairness in medical or other sensitive applications where biased annotations are common.
The paper addresses label bias in image segmentation, where group-conditional label errors cause performance disparities across demographic subgroups. They adapt Confident Learning to segmentation to detect bias without clean labels and mitigate it by leveraging feature space artifacts, achieving equitable performance across three datasets.
Labeled datasets reflect the biases of their annotation pipelines, which sometimes introduce label bias: group-conditional label errors that cause systematic performance disparities across demographic subgroups. Label bias in image segmentation remains underexplored, as even detecting it typically requires clean, unbiased annotations, which are not readily available. We present a data-centric adaptation of Confident Learning to segmentation, allowing detection of label bias directly in the training data without a clean, unbiased ground truth. By comparing the provided training labels to the model's confident predictions, we isolate directional errors that quantify the presence and nature of bias, where standard overlap metrics like Dice fail. We further show that label bias influences subgroup separability in the encoder's feature space, an artifact we leverage for bias mitigation rather than suppressing it. We evaluate three datasets, spanning from synthetic to real-life bias, showing how our framework reliably detects and mitigates bias without access to clean labels, achieving equitable performance across experimental conditions.