CRAFT: Clinical Reward-Aligned Finetuning for Medical Image Synthesis

Yunsung Chung, Alex El Darzi, Carlo El Khoury, Han Feng, Nassir Marrouche, Jihun Hamm

arXiv:2605.12650 · 2026-05-14 공개 · arXiv · PDF

diffusion-models vision-language fine-tuning medical-imaging reward-modeling image-synthesis checklist-auditing clinical-alignment

Abstract

Foundation diffusion models can generate photorealistic natural images, but adapting them to medical imaging remains challenging. In medical adaptation, limited labeled data can exacerbate hallucination-like and clinically implausible synthesis, while existing metrics such as FID or Inception Score do not quantify per-image alignment with pathology-relevant criteria. We introduce the Clinical Alignment Score (CAS), a foundation-model-based proxy for clinical alignment that evaluates generated images along four complementary dimensions beyond visual fidelity. Building on CAS, we propose Clinical Reward-Aligned Finetuning (CRAFT), a reward-based adaptation framework that transfers medical knowledge from multimodal large language models and vision-language models through label-conditioned prompt enrichment, clinical checklists, and differentiable reward optimization. Across four diverse modalities, CRAFT improves CAS and downstream classification performance over strong adaptation baselines. Beyond average CAS gains, CRAFT reduces the empirical low-alignment tail below a real-image reference threshold by 5.5-34.7% points relative to the strongest baseline, corresponding to a 20.4% average relative reduction across datasets. These results indicate fewer hallucination-like generations under CAS, and are corroborated by out-of-family evaluator evaluation, structured checklist auditing, memorization analysis, and a blinded physician preference study on CheXpert.

한국어 요약

📋 한 줄 요약

**[의료 영상 생성 / 정렬]** 자연 이미지용 파운데이션 확산 모델을 의료 영역에 임상 정합적으로 적응시키기 위한 보상 정렬 파인튜닝 프레임워크 CRAFT와 평가 지표 CAS를 제안.

🎯 핵심 기여도

💡 핵심 아이디어

의료 영상 생성 모델의 품질은 평균 FID가 아니라 **임상적 저정합 꼬리(low-alignment tail)**가 얼마나 두꺼운가에 의해 결정된다. CAS와 CRAFT는 이 꼬리를 직접 측정·최적화하는 도구다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 의료 영상 생성 모델 평가·정렬 분야에서 평균 지표에 가려진 임상 위험을 표면화하고, 보상 정렬로 직접 줄이는 표준 레시피를 제공. **한계**: CAS가 결국 파운데이션 모델 프록시이므로 평가-학습 모델 간 상관 편향 가능성, 평가 모달리티 다양화는 후속 과제.

🚀 실용적 활용