Algorithmic encoding of protected characteristics in image-based models for disease detection

TL;DR

This study investigates how protected characteristics like race and sex influence disease detection models from chest X-ray images, revealing biases and proposing methods to analyze and mitigate such disparities in AI healthcare tools.

Contribution

The paper introduces a new methodology combining test-set resampling, multitask learning, and model inspection to analyze and understand encoding of protected characteristics in image-based disease detection models.

Findings

Subgroup disparities in true and false positive rates are confirmed.

Biases are partially mitigated by correcting population and prevalence shifts.

Transfer learning alone is insufficient to determine if protected characteristics influence predictions.

Abstract

It has been rightfully emphasized that the use of AI for clinical decision making could amplify health disparities. An algorithm may encode protected characteristics, and then use this information for making predictions due to undesirable correlations in the (historical) training data. It remains unclear how we can establish whether such information is actually used. Besides the scarcity of data from underserved populations, very little is known about how dataset biases manifest in predictive models and how this may result in disparate performance. This article aims to shed some light on these issues by exploring new methodology for subgroup analysis in image-based disease detection models. We utilize two publicly available chest X-ray datasets, CheXpert and MIMIC-CXR, to study performance disparities across race and biological sex in deep learning models. We explore test set resampling, transfer learning, multitask learning, and model inspection to assess the relationship between the encoding of protected characteristics and disease detection performance across subgroups. We confirm subgroup disparities in terms of shifted true and false positive rates which are partially removed after correcting for population and prevalence shifts in the test sets. We further find a previously used transfer learning method to be insufficient for establishing whether specific patient information is used for making predictions. The proposed combination of test-set resampling, multitask learning, and model inspection reveals valuable new insights about the way protected characteristics are encoded in the feature representations of deep neural networks.