Is Mamba Reliable for Medical Imaging?

2026-02-16

AI Summary

This study presents the first systematic evaluation of the robustness of state-space models—specifically Mamba—against combined software and hardware threats in medical imaging tasks. Using the MedMNIST dataset, the authors simulate diverse input distortions through FGSM and PGD adversarial attacks, PatchDrop occlusions, Gaussian noise, and defocus blur, while modeling hardware faults via targeted and random bit-flip injections. Experimental results reveal a significant drop in Mamba’s accuracy under these perturbations, highlighting critical vulnerabilities that could compromise its reliability in clinical deployment. The findings underscore the fragility of current models in real-world medical settings and emphasize the urgent need for robustness-enhancing mechanisms tailored to the unique demands of safe and trustworthy medical AI applications.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Security and Privacy: Security and privacy of machine learning and AI applications

Abstract

State-space models like Mamba offer linear-time sequence processing and low memory, making them attractive for medical imaging. However, their robustness under realistic software and hardware threat models remains underexplored. This paper evaluates Mamba on multiple MedM-NIST classification benchmarks under input-level attacks, including white-box adversarial perturbations (FGSM/PGD), occlusion-based PatchDrop, and common acquisition corruptions (Gaussian noise and defocus blur) as well as hardware-inspired fault attacks emulated in software via targeted and random bit-flip injections into weights and activations. We profile vulnerabilities and quantify impacts on accuracy indicating that defenses are needed for deployment.

Problem

• Mamba
• medical imaging
• robustness
• adversarial attacks
• hardware faults

Innovation

• Mamba
• robustness evaluation
• medical imaging
• adversarial attacks
• hardware fault emulation