CARDEA: Auditable Reasoning Grounded in Spatial Evidence for End-to-End Coronary Angiography Interpretation
CARDEA pairs coronary angiography diagnoses with bounding-box evidence meant to make its reasoning auditable.
The model runs from raw multi-view CAG videos through keyframe selection to study-level diagnosis. In tests, it matched a dedicated classifier under domain shift for dominance classification and was comparable to two interventional cardiologists on complexity assessment. Its RLVR stage also improved zero-shot report generation on an external cohort, while supervised imitation did not. The paper says clinical use still needs prospective validation against expert cardiologists. HF Daily Papers' note
The model runs from raw multi-view CAG videos through keyframe selection to study-level diagnosis. In tests, it matched a dedicated classifier under domain shift for dominance classification and was comparable to two interventional cardiologists on complexity assessment. Its RLVR stage also improved zero-shot report generation on an external cohort, while supervised imitation did not. The paper says clinical use still needs prospective validation against expert cardiologists. HF Daily Papers' note
score 5