ARIADNE: A Perception-Reasoning Synergy Framework for Trustworthy Coronary Angiography Analysis
Zhan Jin, Yu Luo, Yizhou Zhang, et al.
ARIADNE is a new AI system for analyzing heart vessel images that combines two approaches: first, it uses preference-based learning to train a vision model to recognize complete vessel structures rather than just matching pixels, and second, it uses a decision-making algorithm to identify blockages while intelligently avoiding mistakes on tricky cases like vessel intersections. Tested on 1,400 real patient scans, the system outperforms existing methods and works reliably across different hospitals and imaging equipment.
medical imagingvessel segmentationreinforcement learningvision-language models