Is Mamba Reliable for Medical Imaging?

Banafsheh Saber Latibari1, Najmeh Nazari2, Daniel Brignac1, Hossein Sayadi3, Houman Homayoun4, Abhijit Mahalanobis1
1University of Arizona, 2UC Davis, 3California State University Long Baech, 4University of California Davis


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 MedMNIST classification benchmarks under input-level attacks, including white-box adversarial perturbations (FGSM/PGD), occlusion-based Patch- Drop, 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.