Abstract 4370347: Comparative Performance Of Al-driven CCTA Models For Pre-TAVR CAD Detection: A Systematic Review

Introduction: Accurate assessment of coronary artery disease (CAD) before Transcatheter Aortic Valve Replacement (TAVR) is crucial for optimal outcomes. Al- enhanced coronary a andiocraphy (CCA shows promise for CAD detection, but its vertormance in TAVR patients lacks systemaic eva uation. his reviev analyzes Al-based CCTA models for pre-TAVR CAD detection focusing on both anatomic and physiologic criteria. Methods: We searched Pubiied and Embase tor studies on Al-based CAD detection via CCIA in pre-TAVR patients. Five studies met inclusion criteria, each using different Al models. Key metrics analyzed included sensitivity, specificity, accuracy, negative predictive value (NPV), positive predictive value (PPV), and Net reclassification index (NRI) for CAD+ve and CAD-ve populations. Results: The studies showed significant variability in Al model performance for CAD detection. Sensitivity ranged from 94.9% to 100%, specificity from 40.0% to 100%, and accuracy from 60.0% to 91.9%. CAD+ve NRI: -0.28 to 0.66 (e.g., Brandt et al. 2022 showed strong positive reclassification). CAD-Ve NAl: 005 to 0.13 (most models showed postive reclassication) Models integrating anatomic and physiological criteria (e.g., Brendel et al. 2024 with Al-based CAD-RADS + Al-derived FFR) achieved superior accuracy (96.8% sensitivity, 100% specificity, 91.9% accuracy, CAD+ve NRI 0.11, CAD-ve NRI 0.07). In contrast, purely anatomic models (e.g., Gohman et al. 2022) had lower performance (52.0% specificity, 67.3% accuracy, CAD +ve NRI -0.05, CAD-ve NRI 0.13). Conclusions: Al-based CCTA models show high sensitivity for CAD detection in pre-TAVR assessment, with varying specificity and accuracy. Combining anatomic and physiological parameters improves diagnostic accuracy and NRI. This review underscores the potential of integrated Al models for more precise pre-TAVR CAD screening, with further research needed for clinical validation.