AI-Assisted CT Planning for Transcatheter Aortic Valve Replacement

Introduction: TAVR relies on precise CT planning for annular sizing and complication prediction, but conventional expert assessment is time-consuming and variable. Artificial intelligence (AI) offers automated, rapid analyses. However, the comparative impact of AI-assisted versus conventional CT planning on valve sizing accuracy, procedural efficiency, and clinical outcomes remains undefined. This meta-analysis addresses that gap. Methods: We systematically searched databases (inception-June 2026) for studies comparing AI-assisted versus conventional CT planning for TAVR. Primary outcome: correct valve sizing. Random-effects meta-analysis used Stata 18.0. Risk of bias assessed with ROB 2.0. Heterogeneity via I 2 . Results: Eleven studies (n=9,842 patients) met inclusion criteria. AI-assisted CT planning significantly improved correct valve sizing accuracy compared with conventional expert planning (OR 1.62; 95% CI 1.38-1.90; p<0.001; I 2 =34%). Rates of moderate-to-severe paravalvular leak were lower with AI assistance (OR 0.71; 95% CI 0.59-0.86; p=0.001; I 2 =28%). Patient-prosthesis mismatch did not differ between groups (OR 0.94; 95% CI 0.78-1.14; p=0.52; I 2 =0%). CT analysis time was significantly shorter with AI (mean difference -9.4 minutes; 95% CI -12.1 to -6.7; p<0.001; I 2 =89%). Procedural duration also favored AI (MD -6.2 minutes; 95% CI -9.8 to -2.6; p=0.001; I 2 =56%). Coronary obstruction prediction accuracy was comparable (OR 1.08; 95% CI 0.82-1.42; p=0.58). No significant differences were observed for pacemaker implantation (OR 0.92; 95% CI 0.76-1.11) or 30-day mortality (OR 0.89; 95% CI 0.68-1.16). Heterogeneity was low to moderate for most outcomes. Conclusion: AI-assisted CT planning for TAVR improves valve sizing accuracy, reduces paravalvular leak, and shortens both CT analysis and procedural duration compared with conventional expert planning, without increasing adverse clinical events. These findings support AI integration into routine TAVR workflows to enhance efficiency and precision. Risk of bias was low.