AbstractSignificant aortic regurgitation (AR) is a common complication after continuous-flow left ventricular assist device (LVAD) implantation. Using machine-learning algorithms, this study was designed to examine valuable predictors obtained from LVAD sound and to provide models for identifying AR. During a 2-year follow-up period of 13 patients with Jarvik2000 LVAD, sound signals were serially obtained from the chest wall above the LVAD using an electronic stethoscope for 1 min at 40,000 Hz, and echocardiography was simultaneously performed to confirm the presence of AR. Among the 245 echocardiographic and acoustic data collected, we found 26 episodes of significant AR, which we categorized as “present”; the other 219 episodes were characterized as “none”. Wavelet (time–frequency) analysis was applied to the LVAD sound and 19 feature vectors of instantaneous spectral components were extracted. Important variables for predicting AR were searched using an iterative forward selection method. Seventy-five percent of 245 episodes were randomly assigned as training data and the remaining as test data. Supervised machine learning for predicting concomitant AR involved an ensemble classifier and tenfold stratified cross-validation. Of the 19 features, the most useful variables for predicting concomitant AR were the amplitude of the first harmonic, LVAD rotational speed during intermittent low speed (ILS), and the variation in the amplitude during normal rotation and ILS. The predictive accuracy and area under the curve were 91% and 0.73, respectively. Machine learning, trained on the time–frequency acoustic spectra, provides a novel modality for detecting concomitant AR during follow-up after LVAD.
Prediction of aortic valve regurgitation after continuous-flow left ventricular assist device implantation using artificial intelligence trained on acoustic spectra
