Abstract 4370649: AI-Generated Semiautomated Echocardiographic-Derived Measurements Show Strong Agreement with CT for Tricuspid Annular Sizing

Abstract

Background: Accurate tricuspid annular sizing is essential for procedural planning in transcatheter tricuspid valve intervention (TTVI). While contrast-enhanced, ECG-gated cardiac computed tomography (CT) remains the gold standard, it has limitations such as radiation exposure, contrast-related risks, and procedural delays. Artificial intelligence (AI)-powered echocardiographic tools may offer a safer, contrast-free alternative. Objective: To evaluate the accuracy and reproducibility of tricuspid annular measurements generated from an AI-based semiautomated software from transesophageal echocardiography (TEE) compared with gated cardiac CT. Methods: We retrospectively analyzed 19 patients with severe tricuspid regurgitation undergoing evaluation for TTVI who underwent both contrast-enhanced cardiac CT and TEE, with TEE performed a median of 93 days after CT (range: 0–247 days). Tricuspid annular dimensions were generated using an AI-powered 3D semiautomated analysis (3D Auto TV; Philips Ultrasound). CT annular measurements were independently performed in duplicate using the Aquarius iNtuition software. AI-generated echo measurements were visually reviewed and confirmed to ensure accuracy. Annular dimensions were measured in both diastole and systole using 2D and 3D perimeter-based diameter dimensions (perimeter/π) for both TEE and CT-derived measurements. Agreement between TEE and CT-derived dimensions was assessed with Pearson correlation coefficient. Inter-observer variability was assessed for CT-derived measurements of10 patients. Results: There was strong correlation between diastolic 2D TEE and CT dimensions (r = 0.564, p = 0.036). Intraclass correlation coefficient (ICC) analysis showed excellent reliability for 2D diastolic (ICC = 0.925) and 3D diastolic (ICC = 0.935) measurements. Systolic measurements showed slightly lower but strong agreement (2D ICC = 0.824; 3D ICC = 0.810). All ICC values were significant (p < 0.001). Interobserver variability for the CT scan measurements showed excellent agreement across all 2D and 3D tricuspid annulus measurements (ICC range: 0.916–0.969; all p < 0.001). Conclusion: AI-based TEE semiautomated annular measurements show strong correlation and excellent reproducibility compared with CT-derived sizing, particularly in diastole. This approach may support preprocedural planning and intraprocedural guidance in TTVI, though further studies are needed.