The Role of Subgroup Separability in Group-Fair Medical Image Classification
This addresses fairness issues in medical imaging AI by explaining how models become biased, which is important for developers and practitioners in healthcare.
The paper investigates how the ability of classifiers to separate individuals into subgroups (subgroup separability) varies across medical imaging modalities and protected characteristics, and shows this property predicts algorithmic bias, with findings linking it to subgroup disparities and performance degradation in biased data.
We investigate performance disparities in deep classifiers. We find that the ability of classifiers to separate individuals into subgroups varies substantially across medical imaging modalities and protected characteristics; crucially, we show that this property is predictive of algorithmic bias. Through theoretical analysis and extensive empirical evaluation, we find a relationship between subgroup separability, subgroup disparities, and performance degradation when models are trained on data with systematic bias such as underdiagnosis. Our findings shed new light on the question of how models become biased, providing important insights for the development of fair medical imaging AI.