An Imaging Informatics Workflow for Angiography-Derived Coronary Physiology Profiling and Virtual PCI Simulation

Coronary angiography is routinely acquired during invasive assessment of coronary artery disease, but its interpretation remains fragmented across visual stenosis assessment, quantitative anatomy, physiology estimation, and procedural planning. Existing angiography-derived physiology tools typically provide vessel-level scalar outputs and are not tightly integrated with automated image analysis, longitudinal functional profiling, and interactive simulation within a unified imaging informatics workflow. We developed AngioAI-QFR, an end-to-end imaging informatics pipeline for coronary angiography that integrates automated stenosis localisation, lumen segmentation, centreline-based geometric reconstruction, per-millimetre relative flow capacity profiling, angiography-derived physiology estimation, graph-image co-registration, and virtual PCI simulation. The system was retrospectively evaluated in 100 coronary vessels from 92 patients with invasive fractional flow reserve as the reference standard. Evaluation included computer-vision performance, agreement with invasive FFR, diagnostic performance for FFR ≤ 0.80, automation rate, processing time, and exploratory analysis of focal versus diffuse functional patterns. The pipeline achieved high stenosis detection performance and feasible lumen segmentation, with detection precision of 0.966, mean average precision at an intersection over union (IoU) threshold of 0.50 (mAP@50) of 0.973, segmentation IoU of 0.757, and Dice coefficient of 0.861. AngioAI-QFR demonstrated strong correlation and good overall agreement with invasive FFR, with Pearson r = 0.917, mean absolute error of 0.035, root mean squared error of 0.044, mean bias of - 0.004, and 95% limits of agreement from - 0.091 to 0.083. Diagnostic discrimination for invasive FFR ≤ 0.80 was high, with an AUROC of 0.943, sensitivity of 0.881, and specificity of 0.897. The workflow completed fully automatically in 93% of vessels, with a median time to result of 41 s. Longitudinal RFC profiling separated focal from diffuse impairment patterns and enabled exploratory virtual PCI simulations with immediate recomputation of post-simulation physiology estimates. AngioAI-QFR demonstrates the feasibility of a unified imaging informatics workflow that transforms routine coronary angiography into co-registered anatomical, functional, and simulation-ready representations. The system integrates automated image interpretation, longitudinal physiology visualisation, and virtual intervention simulation within a sub-minute workflow. Prospective multicentre validation, external dataset testing, and post-PCI physiological validation are required before clinical deployment.