J Imaging Inform Med. 2026 Jun 1. doi: 10.1007/s10278-026-02010-1. Online ahead of print.
ABSTRACT
Conventional pixel-wise loss functions fail to enforce topological consistency in coronary vessel segmentation, producing fragmented vascular trees despite high pixel-level accuracy. We present ARIADNE, a two-stage framework coupling preference-aligned perception with RL-based diagnostic reasoning for topologically consistent stenosis detection through an explicit Perception-Reasoning Synergy in which topology-aware segmentation serves as the structural prerequisite for reliable downstream diagnosis. The perception module employs DPO to fine-tune the Sa2VA vision-language foundation model using Betti number constraints as preference signals, aligning the policy toward topologically consistent vessel structures rather than pixel-wise overlap metrics. The reasoning module formulates stenosis localization as a Markov Decision Process with an explicit rejection mechanism that autonomously defers ambiguous anatomical candidates such as bifurcations and vessel crossings, shifting from coverage maximization to reliability optimization and thereby mitigating the clinical alert fatigue that has historically constrained automated decision support. Validated through a conservative patient-level statistical design (n = 35), ARIADNE achieves state-of-the-art Dice of 0.8034 and centerline Dice (clDice) of 0.8378, significantly outperforming generic foundation models including MedSAM3, while attaining a True Positive Rate of 0.867 and reducing False Positives Per Image to 0.85 in stenosis detection. External validation on the public XCAD benchmark confirms generalization across acquisition protocols. This represents the first application of DPO for topological alignment in medical imaging, demonstrating that preference-based learning over structural constraints mitigates topological violations while maintaining diagnostic sensitivity in interventional cardiology workflows. The code is available at https://github.com/qimingfan10/ARIADNE .
PMID:42225888 | DOI:10.1007/s10278-026-02010-1
