Researchers have created a machine learning algorithm that can effectively identify heart murmurs in dogs, which are significant indicators of heart disease, particularly prevalent in many smaller breeds like King Charles Spaniels.
Researchers have created a machine learning algorithm that can effectively identify heart murmurs in dogs, which are significant indicators of heart disease, particularly prevalent in many smaller breeds like King Charles Spaniels.
The University of Cambridge-led research team modified an algorithm initially intended for humans and successfully used it to automatically recognize and assess heart murmurs in dogs by analyzing audio recordings taken from digital stethoscopes. During testing, the algorithm achieved a 90% sensitivity rate in detecting heart murmurs, matching the accuracy level of experienced cardiologists.
Heart murmurs are a major sign of mitral valve disease, the most common cardiac ailment in adult dogs. Approximately one in every 30 dogs that visit a veterinarian is diagnosed with a heart murmur, with a higher occurrence in small breeds and older dogs.
Given the high incidence of mitral valve disease and other heart issues in dogs, early detection is essential because timely treatment can enhance their lifespan. The technology developed by the Cambridge team may serve as an affordable and effective screening method for primary care veterinarians, potentially improving the quality of life for canines. Details of the findings are published in the Journal of Veterinary Internal Medicine.
“Heart disease in humans is a significant health concern, but it’s an even greater issue for dogs,” commented Dr. Andrew McDonald, the study’s first author from Cambridge’s Department of Engineering. “Most small dog breeds will develop heart disease as they age, but dogs cannot express their symptoms as humans do. Hence, it falls upon primary care vets to identify heart disease early enough for treatment.”
Leading the research, Professor Anurag Agarwal, an expert in acoustics and bioengineering, noted, “To our knowledge, there aren’t any existing databases of heart sounds specific to dogs, which is why we used a database of human heart sounds as our starting point. Mammalian heart structures are quite similar, and when problems arise, they often manifest in comparable ways.”
The researchers began with a dataset of heart sounds from around 1,000 human patients and developed a machine learning algorithm that could mimic a cardiologist’s ability to detect heart murmurs. They then modified this algorithm to analyze heart sounds from dogs.
The team collected data from nearly 800 dogs undergoing standard heart evaluations at four specialist veterinary centers in the UK. Each dog had a comprehensive physical examination and an echocardiogram performed by a cardiologist to classify heart murmurs and identify any cardiac issues, with heart sounds recorded using an electronic stethoscope. This is, by far, the largest collection of dog heart sounds ever compiled.
“Mitral valve disease primarily impacts smaller dogs, but to refine our algorithm, we aimed to gather data from dogs of various breeds, sizes, and ages,” explained co-author Professor Jose Novo Matos from Cambridge’s Department of Veterinary Medicine, who specializes in small animal cardiology. “The more data we can collect, the more valuable our algorithm will be for veterinarians and dog owners.”
The researchers fine-tuned their algorithm to not only identify but also classify heart murmurs from the audio recordings, distinguishing between those linked to mild conditions and those indicative of severe heart disease necessitating further intervention.
“Assessing a heart murmur’s grade and determining if treatment is required demands extensive experience, referral to a veterinary cardiologist, and costly specialized heart scans,” said Novo Matos. “We aim to empower general practitioners to recognize heart disease and evaluate its seriousness, aiding owners in making informed choices for their dogs.”
Analysis showed that the algorithm’s evaluations aligned with cardiologists in more than half of the cases, and it was within one grading level in 90% of instances. The researchers consider this a promising outcome, especially considering the frequent inconsistencies in how different veterinarians grade heart murmurs.
“The grading of heart murmurs is crucial for deciding subsequent actions and treatments, and we’ve automated that process,” stated McDonald. “For veterinarians and nurses with varying levels of stethoscope skills, and even the most skilled among them, we believe this algorithm will be an immensely helpful resource.”
Unlike humans with valve disease, where surgery is the only solution, there are effective medications available for dogs. “Recognizing when to medicate is key to ensuring dogs enjoy the best possible quality of life for as long as feasible,” Agarwal commented. “Our goal is to empower vets in making those decisions.”
“While many believe AI poses a threat to jobs, I view it as a tool to enhance my capabilities as a cardiologist,” stated Novo Matos. “We can’t conduct heart scans on every dog in the country due to time and specialist limitations. However, tools like this could assist both vets and pet owners in swiftly identifying those dogs most in need of care.”
This research received support from the Kennel Club Charitable Trust, the Medical Research Council, and Emmanuel College, Cambridge.
