Source Matters: Source Dataset Impact on Model Robustness in Medical Imaging

TL;DR

This study investigates how the choice of source dataset in transfer learning affects model robustness in medical imaging, revealing that ImageNet models overfit confounders more than RadImageNet models.

Contribution

The paper introduces the MICCAT taxonomy to analyze confounders and demonstrates that source dataset impacts overfitting and robustness in medical imaging models.

Findings

ImageNet models overfit confounders more than RadImageNet models.

Performance is similar between ImageNet and RadImageNet, but robustness differs.

Recommends reexamining model robustness when using ImageNet-pretrained models.

Abstract

Transfer learning has become an essential part of medical imaging classification algorithms, often leveraging ImageNet weights. The domain shift from natural to medical images has prompted alternatives such as RadImageNet, often showing comparable classification performance. However, it remains unclear whether the performance gains from transfer learning stem from improved generalization or shortcut learning. To address this, we conceptualize confounders by introducing the Medical Imaging Contextualized Confounder Taxonomy (MICCAT) and investigate a range of confounders across it -- whether synthetic or sampled from the data -- using two public chest X-ray and CT datasets. We show that ImageNet and RadImageNet achieve comparable classification performance, yet ImageNet is much more prone to overfitting to confounders. We recommend that researchers using ImageNet-pretrained models reexamine their model robustness by conducting similar experiments. Our code and experiments are available at https://github.com/DovileDo/source-matters.