AI-based imaging software showed a moderate correlation (r=0.7) with implanted device size for LAAO but only 49% concordance, often underestimating device size.
Does AI-driven imaging software accurately predict the optimal device size for LAAO compared to manual TEE assessment?
AI-driven imaging software for LAAO sizing shows moderate concordance (49%) with final implanted device size and frequently underestimates LAA length, indicating that manual confirmation remains essential.
Abstract Background Left atrial appendage occlusion (LAAO) is a well-established procedure for stroke prevention in atrial fibrillation patients with contraindications to anticoagulation. Preprocedural imaging is critical in determining the appropriate device size. Recent advancements in artificial intelligence (AI) have introduced software for automated sizing based on transesophageal echocardiography (TEE). However, its accuracy compared to manual measurements remains unclear. Objective To evaluate the reliability of AI-based imaging software in predicting the optimal device size for LAAO by comparing its measurements with manual TEE assessments. Methods This prospective analysis included 47 patients (mean age 64 ± 17 years, 50% male) undergoing LAAO from January 2022 to February 2025. Landing Zone (LZ) diameter and Left Atrial Appendage (LAA) length were measured at 45°, 60°, 90°, and 135° views using both AI-based software and TEE. Device size, which can be compressed by 10-30%, was selected based on the Approximate Implant Length and Approximate Maximum LAA Implant Diameter. The AI-recommended device size was compared with the final implanted device size. Discrepancies were analyzed using paired t-tests or Wilcoxon matched-pairs signed rank tests, as appropriate. The concordance and correlation between AI-recommended device size and final device size were evaluated. Results 47 patients (57% male) with a mean age of 75.3 ± 8.4 years were enrolled. The mean implanted device diameter was 25.8 ± 3.4 mm, while the software suggested 24.9 ± 3.6 mm (p = 0.7). A significant correlation was found between software predictions and implanted device size (r = 0.7, p 0.001). However, concordance between software suggestion and implanted device was only 49%, with the software underestimating the size in 36% of cases and overestimating it in 15%. AI-based measurements consistently underestimated LAA length compared to TEE, with mean differences of -6.3 mm at 45°, -4.1 mm at 60°, -6.8 mm at 90°, and -7.7 mm at 135° (p 0.001). A significant discrepancy of -1.5 mm (p = 0.03) was also found in LZ diameter at 135°. A significant difference was observed between AI-based and manual measurements of maximum LAA implant diameter (p 0.0001). The average difference between the suggested measurements and final implanted sizes was statistically significant (p = 0.034). The logistic regression model coefficient for AI-suggested size was 0.035, suggesting a marginal positive effect of the suggested size on concordance probability. Conclusions While AI-driven imaging software serves as a valuable tool for preprocedural planning, it does not perfectly align with operator-derived measurements, particularly in assessing LAA length. The moderate concordance indicates that manual confirmation is essential for accurate device selection. Future improvements incorporating 3D TEE and AI-enhanced algorithms may enhance predictive accuracy and optimize LAAO planning.
Gatto et al. (2025) studied this question. AI-based imaging software showed a moderate correlation (r=0.7) with implanted device size for LAAO but only 49% concordance, often underestimating device size.