Spatial gating improves localized editing of chest X ray images
Spatially Gated Diffusion for Localized Counterfactual Chest Radiograph Editing
Computer Vision and Pattern Recognition
Summary
Making specific changes to chest X ray images without messing up the rest of the image is hard. The authors studied a new method that carefully edits only the targeted parts, using a combination of two processes and a learned gate to decide how to mix edits. Their method was tested on thousands of chest X rays and worked well, keeping changes accurate while preserving other details. This approach helps ensure edits look realistic and accurate in medical images.
What this means in practice
- •For medical imaging teams: Improve automated editing of chest X ray images to add or remove features while keeping other image areas intact, enhancing clinical image correction workflows.
- •For radiology software developers: Implement spatially gated diffusion techniques to create tools that enable localized, high-fidelity chest radiograph editing with better preservation of unrelated image regions.
Authors
Kamran Ullah Afaq, Basit Raza
Abstract
Editing a chest radiograph requires completing the requested change while preserving unrelated content. We study a latent diffusion editor with an instruction-independent source trajectory and an instruction-conditioned editing trajectory. A learned gate mixes their post-sampler candidates at each executed step, and a separate image-space mask composites the decoded proposal with the source. On 2,400 MIMIC-derived requests, 2,244 outputs met the joint target, preservation, quality, and coverage rubric (93.5%), and target completion was 97.4%. Joint validity exceeded a matched composition-only control by 1.7 percentage points (paired patient-cluster 95% interval, 0.6--2.8). At a fixed learned mask, the learned-gate proposal improved joint validity by 1.6 points (0.8--2.4); at a fixed learned proposal, the learned mask improved it by 2.6 points (1.7--3.5). Across three training seeds, mean joint validity was 93.5% with a 0.3-point sample standard deviation. A blinded 240-request assessment yielded adjudicated joint validity of 93.3% for the full editor and 91.7% for composition only. Protected-region mean absolute error decreased from 0.0190 in the raw proposal to 0.0075 after composition. These findings distinguish recurrent-gating effects on the proposal from preservation through final composition in the assessed cohort.