Med-SegLens: Latent-Level Model Diffing for Interpretable Medical Image Segmentation
Salma J. Ahmed, Emad A. Mohammed, Azam Asilian Bidgoli · 2026
Read Med-SegLens: Latent-Level Model Diffing for Interpretable Medical Image Segmentation by Salma J. Ahmed, Emad A. Mohammed, Azam Asilian Bidgoli (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-2602-10508.txt.
Description
Salma J. Ahmed and two co-authors make medical segmentation models easier to diagnose in a 2026 paper. Modern models segment scans well but remain opaque, so failures are hard to explain and dataset shifts are hard to trace. Clinicians and teams need to know what a model learned, not just how it scores.
Their system, Med-SegLens, decomposes activations from two segmentation systems into clear latent traits using sparse autoencoders. Across healthy, adult, pediatric, and sub-Saharan African glioma cohorts, a stable shared backbone emerges, while dataset shift traces to population-exact traits. The paper shows those traits act as causal bottlenecks for failures, and targeted latent interventions can correct errors without retraining.
Catalog record
| CREATOR | Salma J. Ahmed, Emad A. Mohammed, Azam Asilian Bidgoli |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-10508