Heart Sound Analysis in Individuals Supported With Left Ventricular Assist Devices: A First Look

Abstract

The left ventricular assist device (LVAD) has emerged as a bridge or alternative to heart transplant in individuals with advanced heart failure. However, the LVAD recipient population currently faces high rehospitalization rates. Remote analysis of precordial sounds in LVAD recipients may improve aftercare through early detection of complications. Prior work on analyses of precordial sounds in LVAD recipients focused on identifying pump thrombosis. Here, we focus on analyzing intrinsic precordial sounds to provide insight into intrinsic cardiac function. We analyzed a dataset of patient-acquired recordings of precordial sounds in LVAD recipients. We developed a signal processing pipeline to separate LVAD-generated sounds from other precordial sounds, making heart sound analysis in LVAD recipients feasible. Unsupervised clustering of features extracted from LVAD-mitigated sounds revealed subgroups of subjects possessing heart sounds with distinct frequency characteristics. The results provide preliminary evidence for the potential utility of exploring heart sound analysis in LVAD recipients for remote monitoring.

Multiview Outlier Filtered Pediatric Heart Sound Classification

“…UMAP has been used extensively in all domains to reduce dimensionality for various purposes including interpretability, primarily for feature selection in context of ECG [28], [19], [15], [35] and heart sound classification [4], [5]. While various anomaly detection techniques have been used, there is limited use of Local Outlier Filtering (LOF) [37], [33], [16].…”