Artificial intelligence-analyzed computed tomography in patients undergoing transcatheter tricuspid valve repair.

Background: Baseline right ventricular (RV) function derived from 3-dimensional analyses has been demonstrated to be predictive in patients undergoing transcatheter tricuspid valve repair (TTVR). The complex nature of these cumbersome analyses makes patient selection based on established imaging methods challenging. Artificial intelligence (AI)-driven computed tomography (CT) segmentation of the RV might serve as a fast and predictive tool for evaluating patients prior to TTVR.

Methods: Patients suffering from severe tricuspid regurgitation underwent full cycle cardiac CT. AI-driven analyses were compared to conventional CT analyses. Outcome measures were correlated with survival free of rehospitalization for heart-failure or death after TTVR as the primary endpoint.

Results: Automated AI-based image CT-analysis from 100 patients (mean age 77 ± 8 years, 63% female) showed excellent correlation for chamber quantification compared to conventional, core-lab evaluated CT analysis (R 0.963-0.966; p < 0.001). At 1 year (mean follow-up 229 ± 134 days) the primary endpoint occurred significantly more frequently in patients with reduced RV ejection fraction (EF) <50% (36.6% vs. 13.7%; HR 2.864, CI 1.212-6.763; p = 0.016). Furthermore, patients with dysfunctional RVs defined as end-diastolic RV volume > 210 ml and RV EF <50% demonstrated worse outcome than patients with functional RVs (43.7% vs. 12.2%; HR 3.753, CI 1.621-8.693; p = 0.002).

Conclusions: Derived RVEF and dysfunctional RV were predictors for death and hospitalization after TTVR. AI-facilitated CT analysis serves as an inter- and intra-observer independent and time-effective tool which may thus aid in optimizing patient selection prior to TTVR in clinical routine and in trials.