Cross-Domain Detection of Pulmonary Hypertension in Human and Porcine Heart Sounds

Abstract

Detection of Pulmonary Hypertension (PH) via the automated analysis of cardiac auscultation may offer a non-invasive, accurate, and reliable solution with low resource requirements. We detect PH in human and in porcine datasets and demonstrate domain generalization across the two datasets. Extending our previous work, we train a deep network on a representation of segmented second heart sounds (S2). The human dataset contains digital stethoscope (PCG) recordings of 42 patients. The porcine dataset contains 110 samples of PCG and seismocardiography (SCG) recordings obtained from pigs with chemically induced PH. In both datasets, ground truth reference indicators of PH were obtained via right heart catheterization (RHC). The area under the ROC curve (auROC) and area under the Precision-Recall curve (AP) on human data are 0.92 and 0.97, respectively. On the porcine dataset, leave-one-out cross-validation gives 0.84 auROC and 0.85 AP. Moreover, we demonstrate transferability across domains, where training on the porcine dataset and evaluating on the human dataset gives 0.702 auROC and 0.848 AP. Results show that it is possible to use porcine data for developing human AI models, and that Phonocardiogram (PCG) and Seismocardiogram (SCG) training data can be used to evaluate PCG data.