Abstract 15341: Multi-center Validation of a Novel Echocardiography Artificial Intelligence System to Predict Post-operative Right Ventricular Failure in LVAD Patients

Abstract

Introduction: Post-operative right ventricular failure (RV failure) is the single largest contributor to short-term mortality in patients with left ventricular assist devices (LVAD); yet predicting which patient is at risk of developing this complication in the pre-operative setting has remained beyond the abilities of experts in the field. We hypothesized that deep artificial intelligence (AI) driven characterization of subtle pre-operative myocardial motion features, could predict post-operative RV failure in LVAD patients. Methods: We developed a novel echocardiography AI system using an improved dense trajectory algorithm that tracks motion vectors in an unsupervised fashion, and a 3-dimensional convolutional neural network that tracks spatiotemporal features from videos. We used pre-operative ECHO videos from a 536 patient multicenter echocardiographic and clinical dataset, and via a standard 10-fold cross validation, trained and validated the AI system to predict severe or higher grades of post-operative RV failure at the time of index hospitalization. Finally, we independently benchmarked our AI system against clinicians equipped with contemporary clinical risk scores to predict post-operative RV failure (Penn and CRITT score) and manually derived echocardiographic metrics of RV function. Results: 173 (32%) patients were adjudicated to have severe post-operative RV failure. The area under the receiver operator characteristic curve (AUC ROC) for the AI system was 0.86 (95% CI 0.824-0.896). The performance of our AI system exceeded that of both the best performing clinical risk score and manually derived echocardiographic metric by a significant margin (ΔAUC +0.30, 95% CI 0.27 - 0.33 and ΔAUC +0.24, 95% CI 0.21 - 0.27; both p &lt 0.0001). Conclusions: A novel ECHO AI system trained and validated on a multicenter dataset outperformed clinical experts equipped with both contemporary risk scores and manually calculated echocardiographic metrics.