Artificial Intelligence and Machine Learning for Coronary Artery Stenosis Assessment: A Rapid Overview of Systematic Reviews

Background: Coronary artery disease (CAD) remains a leading global cause of mortality. We hypothesized that artificial intelligence (AI) integrated with coronary computed tomography angiography (CCTA) may improve non-invasive diagnostic precision and significantly reduce unnecessary invasive procedures. This study aimed to systematically evaluate the clinical maturity, diagnostic accuracy, and methodological quality of these emerging AI models. Methods: A rapid review and thematic synthesis were conducted. We systematically analyzed ۶ high-impact systematic reviews and meta-analyses (encompassing over ۱۸۰ primary studies and tens of thousands of patients) evaluating deep learning and machine learning-based fractional flow reserve (ML-FFRCT) against invasive angiography and reference standards. Results: Compiled data showed patient-level diagnostic AUCs ranging from ۰.۹۲ to ۰.۹۳۲, with pooled sensitivities of ۰.۸۶ to ۰.۸۹. AI diagnostic accuracy was notably higher at severe (≥۷۰%) stenosis thresholds (۰.۹۰) compared to obstructive (≥۵۰%) thresholds (۰.۸۵). Non-invasive ML-FFRCT demonstrated potential to defer up to ۴۹% of unnecessary invasive angiographies. However, diagnostic specificity fell significantly to ۰.۵۹ in patient subgroups with prior coronary interventions (PCI/CABG) due to metal artifacts. Methodological appraisal revealed high retrospective design bias and critically low radiomics quality scores (RQS) of ۲۷.۸% to ۵۰.۰%. Conclusion: AI-assisted CCTA and ML-FFRCT show strong potential to act as non-invasive gatekeepers for catheterization labs, but their current clinical utility is limited by coronary calcification, metal blooming artifacts, and domain shift. To facilitate translational clinical adoption, we propose establishing standardized, multi-vendor prospective randomized controlled trials that focus primarily on long-term major adverse cardiac events (MACE) and cost-effectiveness.

کلیدواژه ها:

Coronary Artery Disease ، Computed Tomography Angiography ، Artificial Intelligence ، Fractional Flow Reserve ، Myocardial ، Deep Learning