Artificial Intelligence in Percutaneous Coronary Intervention: A Scoping Review of Pre-procedural, Intra-procedural, and Prognostic Applications

Artificial intelligence (AI) has emerged as a transformative force in interventional cardiology, fundamentally reshaping the practice of percutaneous coronary intervention (PCI). This scoping review synthesizes and critically evaluates the advancements in AI applications across the entire PCI workflow, focusing on PubMed-indexed studies published between 2020 and 2025. The integration of AI, particularly machine learning (ML) and deep learning (DL) algorithms, has catalyzed a paradigm shift from subjective, qualitative assessments to objective, quantitative, and highly reproducible analyses. In pre-procedural planning, AI has enabled the rapid and accurate segmentation of coronary anatomy from various imaging modalities, including coronary computed tomography angiography (CCTA), intravascular ultrasound (IVUS), and optical coherence tomography (OCT). This foundational step has unlocked automated quantification of plaque burden, characterization of high-risk vulnerable plaque features, and near-instantaneous, non-invasive prediction of fractional flow reserve (FFR), thereby enhancing patient selection and procedural strategy. Intra-procedurally, AI-powered software integrated into IVUS and OCT systems provides real-time guidance on stent sizing and landing zone selection, aiming to optimize deployment and reduce complications. Furthermore, emerging applications in real-time fluoroscopy analysis and tool navigation promise to enhance procedural efficiency and safety. In post-procedural care, AI models consistently outperform traditional risk scores in predicting a spectrum of outcomes, including acute complications like contrast-induced nephropathy, as well as long-term events such as in-stent restenosis and major adverse cardiovascular events (MACE). Despite this remarkable progress, significant challenges persist. The reliance on limited, often homogenous datasets raises concerns about model generalizability and algorithmic bias. The "black box" nature of many sophisticated DL models hinders clinical trust and adoption, underscoring the critical need for explainable AI (XAI). Most importantly, a deficit of prospective, randomized trials demonstrating tangible improvements in clinical outcomes remains the largest barrier to widespread implementation. The continued maturation of AI in PCI will depend on collaborative efforts to build diverse datasets, develop transparent and trustworthy models, and rigorously validate these technologies in clinical trials designed to prove their ultimate value in improving patient care.