Title : A scoping review of the role of artificial intelligence in left atrial appendage closure
Abstract:
Atrial fibrillation (AF), the most common sustained arrhythmia globally, significantly increases stroke risk, particularly in older individuals. The left atrial appendage (LAA) is the primary site for clot formation in non-valvular AF, making Left Atrial Appendage Closure (LAAC) a crucial alternative to long-term oral anticoagulation for stroke prevention, especially for patients unsuitable for anticoagulants. However, optimizing patient selection, procedural planning, and post-procedural care for LAAC remains complex. The integration of Artificial Intelligence (AI) across the LAAC care pathway offers significant potential to enhance decision-making, procedural safety, and long-term patient outcomes.
This review systematically examines the current and prospective applications of AI in LAAC, categorized into five key areas: patient selection, pre-procedural planning, intra-procedural guidance, post-procedural monitoring, and antithrombotic therapy optimization.
In patient selection, AI-driven causal machine learning models leverage large datasets to compare LAAC outcomes with direct oral anticoagulants (DOACs). These models effectively identify patient subgroups—such as older adults with comorbidities—who are more likely to benefit from LAAC. For pre-procedural planning, AI-enhanced analysis of cardiac CT and 3D echocardiography significantly improves the accuracy of anatomical measurements and device sizing. Tools like FEops HEARTguide™ and automated 3D LAA measurement algorithms show strong correlations with intra-procedural findings, reducing planning time and improving efficiency. AI integration with computational modelling and virtual reality further refines personalized procedural strategies.
While direct AI applications for intra-procedural guidance in LAAC are still emerging, advancements in other cardiac interventions suggest considerable potential. Real-time image analysis and the fusion of pre-operative data with live imaging could enhance catheter navigation and device deployment precision. In post-procedural monitoring, AI-driven analysis of transesophageal echocardiography (TEE) and cardiac CT angiography facilitates early and accurate detection of complications like peri-device leaks and device-related thrombus (DRT). Furthermore, machine learning models can predict short- and long-term adverse events post-LAAC, aiding risk stratification and personalized follow-up.
Optimizing antithrombotic therapy after LAAC is a persistent challenge. AI holds promise in personalizing treatment regimens by balancing thromboembolic and bleeding risks based on complex patient data. However, current predictive models for DRT, even those using machine learning, have shown only modest accuracy, underscoring the need for further refinement.
Despite these promising developments, significant challenges persist. These include the necessity for robust clinical validation through prospective trials, addressing concerns regarding data quality and potential biases, clarifying regulatory frameworks, and developing more explainable AI systems to foster trust and adoption. Future directions for AI in LAAC involve integrating multimodal patient data, advancing explainable AI methodologies, and conducting rigorous clinical trials to solidify its role.
In conclusion, AI is poised to revolutionize the LAAC workflow by enhancing precision, safety, and personalization. Continued research and validation will be crucial to fully realize its clinical impact and improve stroke prevention outcomes for atrial fibrillation patients.
