Game-Theoretic Defenses for Robust Conformal Prediction Against Adversarial Attacks in Medical Imaging

TL;DR

This paper proposes a game-theoretic framework integrating conformal prediction to improve the robustness and reliability of medical imaging models against adversarial attacks, ensuring high coverage and minimal prediction sets.

Contribution

It introduces a novel game-theoretic approach combining conformal prediction with defense strategies to enhance robustness against adversarial attacks in medical imaging.

Findings

Maintains high coverage guarantees under adversarial attacks.

Optimal defenses often converge to a single robust model.

Outperforms baseline strategies across multiple datasets.

Abstract

Adversarial attacks pose significant threats to the reliability and safety of deep learning models, especially in critical domains such as medical imaging. This paper introduces a novel framework that integrates conformal prediction with game-theoretic defensive strategies to enhance model robustness against both known and unknown adversarial perturbations. We address three primary research questions: constructing valid and efficient conformal prediction sets under known attacks (RQ1), ensuring coverage under unknown attacks through conservative thresholding (RQ2), and determining optimal defensive strategies within a zero-sum game framework (RQ3). Our methodology involves training specialized defensive models against specific attack types and employing maximum and minimum classifiers to aggregate defenses effectively. Extensive experiments conducted on the MedMNIST datasets, including PathMNIST, OrganAMNIST, and TissueMNIST, demonstrate that our approach maintains high coverage guarantees while minimizing prediction set sizes. The game-theoretic analysis reveals that the optimal defensive strategy often converges to a singular robust model, outperforming uniform and simple strategies across all evaluated datasets. This work advances the state-of-the-art in uncertainty quantification and adversarial robustness, providing a reliable mechanism for deploying deep learning models in adversarial environments.