Beyond Attack Success Rate: A Multi-Metric Evaluation of Adversarial Transferability in Medical Imaging Models

Emily Curl * , Kofi Ampomah * , Md Erfan * , and Sayanton Dibbo † \dagger · 2026

Read Beyond Attack Success Rate: A Multi-Metric Evaluation of Adversarial Transferability in Medical Imaging Models by Emily Curl * , Kofi Ampomah * , Md Erfan * , and Sayanton Dibbo † \dagger (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-16532.txt.

Description

Curl and colleagues argue that attack success rate alone misleads medical imaging robustness tests. The binary metric records only whether an attack succeeds, ignoring strength, image quality and cross-architecture transfer. Their study covers four medical datasets, seven models spanning convolutional and transformer models, and seven attacks at five budgets. It measures success rate plus signal-to-noise ratio, structural similarity and perturbation size. Perceptual and distortion metrics tracked each other closely but barely correlated with success rate, so risk assessment needs multiple metrics.

Catalog record

CREATOREmily Curl * , Kofi Ampomah * , Md Erfan * , and Sayanton Dibbo † \dagger
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34188
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.16532

Browse the full library · About the Tower of Babel