Deep learning-driven automated detection of canine cardiac murmurs via digital wireless stethoscope auscultation

Abstract

A heart murmur is a key indicator of cardiovascular disease, making auscultation essential. Nonetheless, its accuracy is influenced by the clinician’s experience and subjective interpretation. This study developed a deep learning–based algorithm (CNN6) for automated detection of heart murmurs using phonocardiogram (PCG) data acquired via a digital wireless stethoscope. A total of 2,269 recordings (over 20 h) from 406 dogs were used for model development and validation, and 297 recordings from 60 dogs were reserved for independent testing. The model achieved 89.9% sensitivity, 92.7% specificity, and 90.9% accuracy, demonstrating diagnostic performance comparable to that of experienced veterinarians. This AI-assisted approach provides a consistent and objective murmur assessment and represents a clinically applicable screening tool for myxomatous mitral valve disease (MMVD), enhancing diagnostic precision, facilitating telemedicine, and promoting the integration of artificial intelligence into veterinary cardiology.