On Evaluating Adversarial Robustness of Chest X-ray Classification: Pitfalls and Best Practices

TL;DR

This paper critically examines the challenges and best practices in evaluating the adversarial robustness of chest X-ray classification models, highlighting dataset, architecture, and metric dependencies, and proposing methodological improvements.

Contribution

It provides the first comprehensive analysis of robustness evaluation pitfalls in medical imaging, with extensive experiments across multiple datasets, models, and diseases.

Findings

Robustness assessments vary significantly with dataset and model choice.

Medical diagnosis peculiarities affect robustness evaluation outcomes.

Current evaluation methods may not accurately reflect real-world robustness.

Abstract

Vulnerability to adversarial attacks is a well-known weakness of Deep Neural Networks. While most of the studies focus on natural images with standardized benchmarks like ImageNet and CIFAR, little research has considered real world applications, in particular in the medical domain. Our research shows that, contrary to previous claims, robustness of chest x-ray classification is much harder to evaluate and leads to very different assessments based on the dataset, the architecture and robustness metric. We argue that previous studies did not take into account the peculiarity of medical diagnosis, like the co-occurrence of diseases, the disagreement of labellers (domain experts), the threat model of the attacks and the risk implications for each successful attack. In this paper, we discuss the methodological foundations, review the pitfalls and best practices, and suggest new methodological considerations for evaluating the robustness of chest xray classification models. Our evaluation on 3 datasets, 7 models, and 18 diseases is the largest evaluation of robustness of chest x-ray classification models.