{"slug": "beyond-target-scores-measuring-off-target-drift-in-diffusion-based-medical-image", "title": "Beyond Target Scores: Measuring Off-Target Drift in Diffusion-Based Medical Image Editing", "summary": "A new benchmark, CIB-Med-1, reveals that diffusion-based medical image editors can appear successful by altering non-target findings rather than isolating intended pathology, a failure mode termed reward-hacking. Researchers introduced constrained diffusion guidance that reduces median off-target drift from 0.46 to 0.20 while preserving target progression, and a blinded validation with radiology trainees showed stronger agreement with intended progression orderings (τ=0.61 vs. 0.29 for Pix2Pix). The findings argue for evaluating medical image editing as trajectory-level semantic control rather than endpoint score maximization.", "body_md": "arXiv:2607.16291v1 Announce Type: new\nAbstract: Diffusion models can now edit medical images in visually plausible ways, but the standard evaluation question is too narrow: did the target score increase? In clinical imaging, target findings are entangled with co-morbidities, acquisition effects, and selection bias, so a model can appear successful by changing correlated non-target findings rather than isolating the intended pathology. We introduce CIB-Med-1, a trajectory-level benchmark for controlled biomarker editing in chest radiography. CIB-Med-1 evaluates directional pleural effusion editing through calibrated target progression, inversion rate, and off-target semantic drift over 14 clinically motivated nuisance axes. The benchmark exposes a reward-hacking failure mode in which diffusion editors increase effusion scores while simultaneously altering parenchymal, cardiomediastinal, pleural, chronic, or artifact-related findings. We further present a constrained diffusion guidance baseline that optimizes target progression subject to bounded off-target change. Across held-out radiographs, the constrained editor preserves target progression ($\\rho_{\\mathrm{trend}}=0.88$ vs. $0.90$ for unconstrained guidance) while reducing median off-target drift from $0.46$ to $0.20$ and 90th-percentile drift from $0.98$ to $0.33$. Drift magnitude tracks empirical target--off-target association, supporting the view that semantic instability is structured rather than incidental. A blinded human validation probe with radiology trainees further shows stronger agreement with intended progression orderings ($\\tau=0.61$ vs.\\ $0.29$ for Pix2Pix). These results argue that medical image editing should be evaluated as trajectory-level semantic control, not as endpoint score maximization.", "url": "https://wpnews.pro/news/beyond-target-scores-measuring-off-target-drift-in-diffusion-based-medical-image", "canonical_source": "https://arxiv.org/abs/2607.16291", "published_at": "2026-07-21 04:00:00+00:00", "updated_at": "2026-07-21 04:08:59.331336+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "generative-ai", "ai-safety"], "entities": ["CIB-Med-1", "Pix2Pix"], "alternates": {"html": "https://wpnews.pro/news/beyond-target-scores-measuring-off-target-drift-in-diffusion-based-medical-image", "markdown": "https://wpnews.pro/news/beyond-target-scores-measuring-off-target-drift-in-diffusion-based-medical-image.md", "text": "https://wpnews.pro/news/beyond-target-scores-measuring-off-target-drift-in-diffusion-based-medical-image.txt", "jsonld": "https://wpnews.pro/news/beyond-target-scores-measuring-off-target-drift-in-diffusion-based-medical-image.jsonld"}}