MedReCo: Vision-Language Framework for Comparative Reasoning in Radiology

arXiv·low signalXBlueskyLinkedInCopy link

The authors formulate radiological diagnosis as an entity-aware cross-image reasoning problem and build MedReCo-DB to support both reference-case retrieval and temporal comparison across prior studies — closing the gap between isolated image AI and real radiological practice where diagnosis depends on comparing across exams. It moves medical imaging AI from single-image interpretation toward the longitudinal comparison clinicians actually perform. Notable for multimodal practitioners in regulated domains.