How AI can help veterinarians detect heart murmurs earlier

In a conversation with dvm360, Brian Scansen, DVM, MS, DACVIM (Cardiology), explores why heart murmurs may be missed in clinical practice — and what's at stake when they are. He also shares his perspective on EKO Vet+ CANINEBEAT® AI, a recently launched tool with the potential to sharpen detection, support clinical decision-making, and help bridge the gap between what a specialist hears and what a general practitioner might miss.

dvm360: Starting with your name, can you tell me a little bit about yourself?

Brian Scansen, DVM, MS, DACVIM (Cardiology): I'm a veterinary cardiologist and head of the Cardiology and Cardiac surgery section at Colorado State University.

dvm360:At what stage of heart disease are murmurs most likely to go undetected, and how does earlier detection change the treatment approach?

Scansen: We know that heart disease can be missed. In some cases, it's because a murmur is not heard early in the course of the disease. And the reason that a murmur might be missed relates to several factors. One, in a busy veterinary practice, having adequate time and an optimal environment to listen to a patient can sometimes be a challenge. In addition, in a very busy practice, being able to listen carefully over all cardiac valves can sometimes be a struggle. And then there are human related factors. We know that sometimes with aging hearing loss can occur, and that can potentially impact our ability to detect a heart murmur by auscultation.

dvm360: What are the consequences of a heart murmur going undetected in its early stages?

Scansen: In certain forms of heart disease, the only sign may be a soft heart murmur. As an example, dilated cardiomyopathy in [dogs] tends to have a softer heart murmur, and the animal may not show outward signs of disease. In fact, many forms of heart disease don't manifest outward signs until much later in the course of the disease.

If we can detect the disease early by making sure we hear abnormalities in the heart, such as a heart murmur, then that allows us to initiate therapy, and for several diseases, we do know that earlier therapy, before the onset of clinical signs, results in an improved survival or outcome for the patient.

dvm360: What is the value in clinical terms of the AI algorithm detecting heart murmurs with 95% sensitivity and specificity? Is that threshold sufficient for a tool like this?

Scansen: When we look at the sensitivity and specificity of any diagnostic test, that tells us how we might want to think about using this test in practice.

To put sensitivity and specificity into more clinical terms, sensitivity is something that helps us to screen for a disease. So, if we have a tool that has high sensitivity, then we're likely to capture all the patients, or most of the patients with that disease, and a sensitivity of over 95% as with this tool is robust and useful as a good screening test.

Now, specificity is helpful because it helps us to rule in the presence of the disease, so that when we have high specificity, we can be relatively confident that a positive result is real or true. And having both high sensitivity and high specificity is useful to us to know that this tool can help us to screen patients and when it detects something, particularly if there's a moderate to high prevalence of disease, to be confident in that result.

dvm360: How was the Eko Vet+^TM | CANINEBEAT®AI trained, and what made a dataset of 4,000+ canine heart recordings sufficient for validation?

Scansen: The Eko Vet+^TM | CANINEBEAT®AI’s algorithm was developed using a very large number of heart sounds.We know, though I am by no means an AI expert, that these models require and are most effective when they have a large amount of input data to help train them. This data for the CANINEBEAT® AI algorithm was acquired across numerous countries, all correlated to an echocardiographic diagnosis and the auscultation of a cardiologist on the patient at that time.

I think what that tells us is that training for this device occurred across international boundaries and in a large and diverse population, making this device more likely to be broadly applicable in the field as well.

dvm360:If you could guess, how accurate is the average general practitioner when detecting heart murmurs compared to a specialist, and where does AI fit in closing that gap?

Scansen: We don't have a large amount of literature on how accurate a veterinarian is when auscultating. But there are a few papers that we can point to. There was a study done in the Netherlands looking at puppies, and interestingly, that study showed that when primary care veterinarians listened to puppies, there were fewer heart murmurs detected then when a cardiologist listened to that same population of puppies. In fact, in around 300 puppies, the cardiologist heard soft murmurs in nearly 100 and the primary care veterinarians only heard a heart murmur in 1 of the 300 puppies.

So, there's a bit of a discrepancy there. The other studies that we have support the fact that about 70-80% accuracy can be found in veterinary students when auscultating. Veterinary students are obviously not the same as a practicing veterinarian with their level of experience, but if we use that analogy, what it would suggest is that about 70 to 80% of the time the veterinary student and by proxy, perhaps the practicing veterinarian, will hear a murmur correctly. But that does leave a gap of perhaps 20 to 30% of the time where a heart murmur may be missed – whether that applies to all patients, all situations, or the experience level of the primary care veterinarian is unknown. Unfortunately, we don't have as much data in those populations.

dvm360: What still needs to improve in canine cardiac care that technology alone cannot address?

Scansen:In cardiology, I think the biggest area that I see for improvement is helping to move the pet family from information (“your dog has a heart murmur”) to action (diagnostic tests are performed). I see this tool actually fostering that transition. For decades, if not centuries, listening to an animal's heart was a solitary event – the veterinarian listened, made a decision, and relayed that to the family. What I think AI tools like this can now do is reinforce those findings and move it from a solitary event of trusting whatever the veterinarian says – thankfully, people are often very trusting of veterinarians – to confirming things to the family that yes, there is an abnormality here. It also helps to confirm things for the veterinarian – what you are hearing is real.

What I hope that will do is support information moving to action. Instead of saying “there's a heart murmur”, and the client says, “are you sure? what does that mean?” Now this tool gives guidance and says, “Yes, in fact, we're quite certain there is a heart murmur. And knowing there's a heart murmur, we should probably move towards additional diagnostics because that may guide therapy that has improved outcome for the patient.”

dvm360:Is there anything I have not asked that you would like to add?

Scansen: This is not a tool in isolation, but rather a tool that can be used within the conversation that the veterinarian is having with the family and that's where I see the value. Well, I see the value in 2 ways. One, I think this helps us regain a focus on auscultation in the physical exam.Careful auscultation is critical to detecting heart disease, and this may bring auscultation back to its rightful place as a critical part of the veterinary exam. That's always been the case. But I think as technology has advanced and visit times have reduced, sometimes we miss things or even jump to diagnostic testing before we do a full comprehensive evaluation.This device reinforces the value of careful auscultation, and uses technology to support that critical piece, the physical examination.

I think that this device represents a nice interplay between technology and the history of medicine, where the stethoscope goes back 200 plus years.Those two facets come together in this tool, which will hopefully improve our ability to detect disease, and therefore improve our ability to treat those patients that need treatment, and therefore extend and improve quality of life for dogs.

Want to learn more from Scansen? Join him for his upcoming keynote address at Fetch Kansas City!