A new machine-learning algorithm that listens to digital heartbeat data could help veterinarians diagnose murmurs and early-stage heart disease in dogs. Developed by a team of researchers from the University of Cambridge, the study analyzes electronic stethoscope recordings to grade murmur intensity and diagnose the stage of myxomatous mitral valve disease (MMVD)—the most common form of heart disease. It could alleviate the need for medical imaging, making diagnosis faster, less costly, and less stressful.
“Our study lays the groundwork for more accessible and affordable detection and treatment of valvular heart disease in dogs. With this machine-learning algorithm, veterinarians could prescribe medication based on a simple stethoscope check, enabling earlier intervention and potentially improving outcomes,” said study senior author Anurag Agarwal, a professor of Acoustics and Biomedical Technology at the University of Cambridge.
About 10% of dogs and up to 75% of senior dogs in the US develop heart disease. Vets typically find and assess heart murmurs—a critical indicator of heart disease—during physical exams, listening for irregularities in how the heart chambers pump.
Echocardiograms, or heart ultrasounds, provide a detailed view of heart structure and functions, helping detect abnormalities and determine the severity or stage of the disease. This imaging is valuable for ongoing monitoring and assessing treatments. However, it’s costly, requires a specialist, and often means the dog must stay with the vet for the day—adding stress to an already taxed heart.
While human heart sound analysis has progressed with the help of large datasets, similar resources for dogs are scarce. They could help pave the way for automated murmur detection.
To bridge this gap, the researchers used electronic stethoscopes, which are increasingly available to veterinarians, to make digital recordings of heart sounds. They collected audio recordings from 756 dogs with and without heart disease, each of which underwent a full physical exam and echocardiogram. Using a recurrent neural network originally trained for heart murmur detection in people, they fine-tuned the model on dog-specific data to predict murmur grade from the recordings.
The models were trained using PyTorch and NVIDIA CUDA on NVIDIA GeForce 10 Series GPUs, enabling efficient data processing.
In testing, the algorithm achieved an 87.9% sensitivity rate in detecting murmurs, matching an expert cardiologist’s assessment with higher sensitivity for louder, more severe murmurs. It also showed strong accuracy in grading heart murmurs, matching expert assessment in 57% of cases, which the researchers suggest is impressive due to the high variability across vets when grading murmurs.
According to the study, the researchers plan to deploy the algorithm in general veterinary practices to assess its performance in real-world settings. They plan to expand the dataset to improve accuracy and reliability, particularly for dogs in the early stages of heart disease.
Early detection and correct grading of heart disease helps veterinarians decide the best treatment options, giving dogs with heart disease healthier, longer lives.
“By making advanced diagnostic capabilities available in general practice, this technology may lower costs and make high-quality cardiac care more widely available to dog owners,” said Agarwal.
Read the study A machine-learning algorithm to grade heart murmurs and stage preclinical myxomatous mitral valve disease in dogs.
Featured image courtesy of www.freepik.com
