Toward the Development of a Low-Cost AI-Enabled Digital Stethoscope

Auscultation plays a significant role in the screening and clinical diagnosis of cardiovascular and respiratory diseases (CVDs). Experience and domain expertise knowledge are required for accurate diagnosis. However, many resource-constrained countries may not have sufficient expertise. This study, therefore, focuses on the preliminary development and evaluation of a low-cost digital stethoscope which leverages machine learning to distinguish between normal and cardiovascular disease (CVD) related heart sounds. A prototype of a battery-powered digital stethoscope was developed, and the preliminary results showed that the system could capture heart sounds with good quality. Using a PhysioNet Dataset, Machine Learning models (Fine-K-Nearest neighbour FKNN, Gaussian kernel support vector machines GSVMs and Deep Convolutional Neural Networks CNNs) were trained on temporal-frequency derived features of phonocardiogram signals (PCG) by Wavelet scattering and Short-Time-Fast-Fourier Transform (SFFT). The overall validation accuracy range was 89.9%-98.5% and the models could effectively classify normal and abnormal heart sounds. These promising results are essential for screening and the early detection of cardiovascular diseases, enabling timely intervention and disease management in low-resource countries. The work contributes to AI research aimed at enhancing CVD diagnosis and screening.