Abstract 10891: Artificial Intelligence Echocardiographic Detection of Right Ventricular Dysfunction in Patients With High-Risk Pulmonary Embolism

Introduction: Patients (pts) with pulmonary embolism (PE) and right ventricular (RV) dysfunction have a worse prognosis. We previously validated a real-time artificial intelligence software (AI, LVivoRV®) which calculates RV fractional area changes (FAC), free wall strain (FWS), and tricuspid annular planar systolic excursion (TAPSE) from a single unedited, non-ultrasound agent-enhanced apical 4-chamber (A4C) TTE view. Hypothesis: AI-calculated TTE parameters of RV function will accurately identify pts with intermediate-high and high-risk PE that would otherwise have been identified by comprehensive physician assessment of all TTE parameters, physical exam and biomarkers (“physician assessment”). Methods: We retrospectively identified pts in whom both a TTE (60.7% with ultrasound contrast) and chest CTA were performed for PE evaluation (median 1 day between studies). Based on comprehensive physician assessments, pts were stratified by the 2019 ESC guidelines from low-risk to high-risk for PE mortality. The accuracy of AI-TTE thresholds for RV dysfunction (previously defined) to identify physician-assessed high-risk PE was examined. Results: Of the 107 pts, 66 (61.7%) had confirmed PE on CTA. By physician assessment, 28 of these 66 cases were classified as intermediate-high/high-risk PE. Across all AI-TTE parameters, the sensitivities and negative predictive values for intermediate-high/high-risk PE (n=28) ranged from 79-86% and 83-87% respectively (Table). In contrast, the specificities and positive predictive values ranged from 30-56% and 29-38%. Conclusions: A simple to use, fully automated, AI-based TTE assessment of RV dysfunction at the bedside identified ~85% of all cases that would otherwise have been identified as intermediate-high and high-risk PE by comprehensive physician assessment (although the false positive rate was high). Further studies are warranted to examine how best to integrate this AI into clinical care pathways.