The Role of Subgroup Separability in Group-Fair Medical Image Classification

TL;DR

This paper explores how the ability of medical image classifiers to distinguish subgroups influences algorithmic bias, revealing that subgroup separability predicts disparities and performance issues, guiding fair AI development.

Contribution

It introduces the concept that subgroup separability is predictive of bias and provides theoretical and empirical evidence linking these factors in medical imaging classifiers.

Findings

Subgroup separability varies across modalities and characteristics.

Higher subgroup separability correlates with increased bias.

Models trained on biased data show performance degradation related to subgroup separability.

Abstract

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.