PRISM: High-Resolution & Precise Counterfactual Medical Image Generation using Language-guided Stable Diffusion
Amar Kumar 1,2, Anita Kriz 1,2, Mohammad Havaei 3, Tal Arbel 1,2 · 2025
Read PRISM: High-Resolution & Precise Counterfactual Medical Image Generation using Language-guided Stable Diffusion by Amar Kumar 1,2, Anita Kriz 1,2, Mohammad Havaei 3, Tal Arbel 1,2 (2025) 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-2503-00196.txt.
Description
Amar Kumar, Anita Kriz, Mohammad Havaei, and Tal Arbel (2025) present PRISM, a system that fabricates medical images with edited features. Deep learning for medical imaging is hampered by spurious correlations, imbalanced data, and scarce text labels: a model may learn to spot the scanner's marker instead of the disease. Counterfactual images, versions of a scan with a feature changed and the rest intact, can expose and correct shortcuts. PRISM makes them with language-guided Stable Diffusion.
Text prompts steer editing: a user can ask for a device removed or a disease sign added at high resolution. Control is precise: attributes change while unrelated anatomy stays put. Evaluation shows how the counterfactuals can audit models for reliance on spurious cues and balance skewed datasets. The work adapts vision-language foundation models, built for natural images, to such pictures.
Catalog record
| CREATOR | Amar Kumar 1,2, Anita Kriz 1,2, Mohammad Havaei 3, Tal Arbel 1,2 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-00196