Benchmarking Dependence Measures to Prevent Shortcut Learning in Medical Imaging

TL;DR

This paper benchmarks various dependence measures like mutual information and adversarial classifiers to prevent shortcut learning caused by confounding factors in medical imaging, aiming to improve model generalization.

Contribution

It systematically evaluates dependence measures for reducing spurious correlations in medical imaging models, providing practical insights for better confound mitigation.

Findings

Dependence measures vary in effectiveness for preventing shortcut learning.

Adversarial classifiers show promise in reducing confounding effects.

Insights applicable to real-world medical imaging scenarios.

Abstract

Medical imaging cohorts are often confounded by factors such as acquisition devices, hospital sites, patient backgrounds, and many more. As a result, deep learning models tend to learn spurious correlations instead of causally related features, limiting their generalizability to new and unseen data. This problem can be addressed by minimizing dependence measures between intermediate representations of task-related and non-task-related variables. These measures include mutual information, distance correlation, and the performance of adversarial classifiers. Here, we benchmark such dependence measures for the task of preventing shortcut learning. We study a simplified setting using Morpho-MNIST and a medical imaging task with CheXpert chest radiographs. Our results provide insights into how to mitigate confounding factors in medical imaging.