Unsupervised bias discovery in medical image segmentation

TL;DR

This paper introduces an unsupervised method to detect biases in medical image segmentation models without needing ground-truth labels, helping ensure fairness in biomedical applications.

Contribution

The proposed approach is the first to anticipate biases in medical segmentation models without relying on annotated ground-truth data, using reverse classification accuracy.

Findings

Successfully anticipates fairness issues without ground-truth labels

Effective in synthetic and real-world scenarios

Provides a new tool for bias detection in medical imaging

Abstract

It has recently been shown that deep learning models for anatomical segmentation in medical images can exhibit biases against certain sub-populations defined in terms of protected attributes like sex or ethnicity. In this context, auditing fairness of deep segmentation models becomes crucial. However, such audit process generally requires access to ground-truth segmentation masks for the target population, which may not always be available, especially when going from development to deployment. Here we propose a new method to anticipate model biases in biomedical image segmentation in the absence of ground-truth annotations. Our unsupervised bias discovery method leverages the reverse classification accuracy framework to estimate segmentation quality. Through numerical experiments in synthetic and realistic scenarios we show how our method is able to successfully anticipate fairness issues in the absence of ground-truth labels, constituting a novel and valuable tool in this field.