AI-Enhanced Pulmonary Auscultation for Heart Failure Detection

Despite a decline in mortality from common cardiovascular diseases (CVDs), the global incidence of heart failure (HF) is on the rise. Traditional methods for early HF detection, such as electrocardiograms (ECGs), are often impractical for home use. This study investigates the feasibility of utilizing artificial intelligence (AI) and lung auscultation for early HF detection in a home setting. Pulmonary sounds were recorded from 15 HF patients and 15 healthy subjects. Two deep learning (DL) models - a novel, compact convolutional neural network (CNN) and a pre-trained transformer - were trained on this dataset. The optimal model achieved a specificity of 71.4% and a sensitivity of 83.3% with subject-independent training and testing. Analysis of time and frequency domain signal power revealed that HF patients typically exhibited louder pulmonary sounds between 750 and 1800 Hz range. These findings indicate that auscultation could be an affordable, efficient, and rapid screening method for detecting HF at home. This study highlights the potential for developing an AI stethoscope based on nonclinical auscultation. By combining signal processing and machine learning (ML), we can overcome subject-dependent variations and gain new insights into the pulmonary sounds associated with HF.