CARDEA: Auditable Vision-Language Model for Coronary Angiography
September 6, 2026
CARDEA is a vision-language model trained using Chain-of-Box (CoB) distillation and RLVR for coronary angiography interpretation. It generates auditable reasoning traces grounded in spatial bounding boxes to increase clinician trust.
HOW THIS AFFECTS YOU
●
builderYou can implement RLVR with verifiable spatial rewards to improve medical AI transparency.
●
healthThis provides an auditable decision process for high-stakes cardiac diagnosis.