CARDEA: chain-of-box reasoning makes coronary angiography AI diagnoses auditable

Jia-Jen Lee · hf · 2026-09-11

CARDEA is an end-to-end vision-language model for coronary angiography that grounds diagnoses in spatial evidence via chain-of-box reasoning, trained with RL using verifiable rewards. It produces auditable diagnoses and improves zero-shot report generation.

Original post →

More from Research

Research channel →