Research map / Image generation / Text-to-image diffusion
RadEdit: stress-testing biomedical vision models via diffusion image editing
Fernando Pérez-García, Sam Bond-Taylor, Pedro Sanchez, Boris van Breugel, Daniel Coelho de Castro, Harshita Sharma, Valentina Salvatelli, Maria Teodora A Wetscherek, Hannah CM Richardson, Lungren Matthew, Aditya Nori, Javier Alvarez-Valle and 2 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train segmentation on larger, abnormality-rich CheXmask instead of MIMIC-Seg | Lung segmentation model · Healthy->pacemaker (RadEdit synthetic) | Dice drop vs real data: 12.4 -> 8.2 | Improved |
| Train segmentation on CheXmask instead of MIMIC-Seg | Lung segmentation model · Healthy->consolidation (RadEdit synthetic) | Dice drop vs real data: 11.5 -> 7.4 | Improved |
| Train segmentation on CheXmask instead of MIMIC-Seg | Lung segmentation model · Healthy->edema (RadEdit synthetic) | Dice drop vs real data: 3.6 -> 1.6 | Improved |
| Use masks and DDPM inversion in RadEdit rather than LANCE or DDIM inversion (ablation) | RadEdit editing · Lung border preservation after edema edit | Dice / AHD of strong predictor vs original masks | Improved |
| Use RadEdit keep/edit masks to remove COVID-19 features instead of unmasked editing (LANCE) | COVID-19 acquisition shift editing · BIMCV+ edits | Preservation of laterality markers and contrast | Improved |
| Add a chest drain keep mask when editing out pneumothorax | RadEdit · CANDID-PTX edits | Chest drain preserved after edit | Improved |
| Use DiffEdit automatic masks instead of ground-truth masks | Pneumothorax mask prediction · Pneumothorax segmentation | Dice / AHD | Hurt |
| Strong predictor (trained on BIMCV+ and BIMCV-) tested on real vs synthetic edited set | COVID-19 classifier · BIMCV+/MIMIC-CXR vs RadEdit synthetic | Accuracy | No clear effect |
| Weak COVID-19 classifier evaluated on RadEdit synthetic negatives | Weak predictor (BIMCV+ vs MIMIC-CXR) · Synthetic COVID-negative images | Accuracy: 95% drop | Hurt |
Closest papers by meaning
- Trade-Offs in Fine-Tuned Diffusion Models between Accuracy and Interpretability (AAAI 2024 · similarity 0.85)
- Rethinking Visual Attribution for Chest X-ray Reasoning in Large Vision Language Models (NeurIPS 2026 · similarity 0.85)
- A Textbook Remedy for Domain Shifts: Knowledge Priors for Medical Image Analysis (NeurIPS 2024 · similarity 0.84)
- BenchX: A Unified Benchmark Framework for Medical Vision-Language Pretraining on Chest X-Rays (NeurIPS 2024 · similarity 0.84)
- Learning General-purpose Biomedical Volume Representations using Randomized Synthesis (ICLR 2025 · similarity 0.84)
- ImgEdit: A Unified Image Editing Dataset and Benchmark (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
