Mayo Clinic experts in AI, Francisco Lopez-Jimenez, M.D., M.S., and Zachi Attia, Ph.D., M.B.A., share how artificial intelligence (AI) is reshaping cardiovascular care by unlocking new insights from the electrocardiogram (ECG).
Over the past decade, Mayo Clinic researchers have reimagined the ECG as a rich physiological signal rather than a limited diagnostic checklist. By applying AI models to routine ECGs, clinicians can now identify patients at risk of conditions such as heart failure, atrial fibrillation and cardiomyopathy — often before symptoms appear.
The conversation highlights how AI-enabled ECG tools support earlier diagnosis, guide more-appropriate use of electrocardiography and help clinicians make more-consistent, informed decisions without replacing clinical judgment.
Hi, I'm Doctor Francisco Lopez Jimenez, a clinical cardiologist at Mayo Clinic. And I'm Doctor Saa Tia, AI scientist at the Mayo Clinic, and together we co-direct the AI program in cardiology, and we're here today to tell you our story. That when we look back 10 years, it is easy to forget how different clinical practice felt. AI wasn't part of everyday cardiology. We were focused on doing the best we could with the tools we had, but many diseases were still being detected late. Yeah, exactly. A lot of the patients were being diagnosed only after symptoms appeared, heart failure after decompensation, atrial fibrillation after a stroke, and cardiomyopathy after years of sy progression. The challenge wasn't lack of expertise, it was a lack of early signal. An ECG or electrocardiography was everywhere, but it knows what, the role of ECG was fully narrow. Uh, it will tell us the rhythm, the conduction. Um, acute heart attacks, etc. but it wasn't something we used to think proactively about risk. Yeah, and I think that changed when we started treating the electrocardiogram as a rich physiological signal rather than a checklist. With AI, we weren't replacing clinical interpretation. We were adding. Nails on top of it. The early low ejection fraction work we did was a good example. Patients felt fine. The electrocardiogram looked remarkable to the eye of the cardiologist, but the model was flagging high risk for low ejection fraction. Yes, and what mattered most was that Not only the algorithm result, but how clinicians use that in their practice. They started sending patients who need an echocardiogram to an echocardiogram, and maybe more importantly, not sending ones that would have a normal echocardiogram just based on the AI ECG. That is true, and that's when the conversation shift from, is this accurate to is this helpful? Does it really lead to better decisions for patients? And I think the answer depended on how we integrate AI in the practice, because AI by itself doesn't change care, workflows do, how you as a clinician use the AI to help your patients. That is true, and we saw that again with the atrial fibrillation algorithm. Detecting atrial fibrillation risk when the electrocardiogram was completely normal, showing normal sinus rhythm, really was a paradigm shift. We started thinking more about monitoring, thinking about risk for a particular condition, and for patients, it meant less surprises, less strokes as the first sign of a disease, and more continuity in their care. Yeah, that's true. And, and one of the most important shifts also with this AI movement was equity. Uh, ECGs are done everywhere in the world, are done in rural clinics, urgent care, emergency departments, and, uh, this system allows us to screen in settings that will not have immediate access to advanced specialty imaging. And that's critical. And maybe the nicest thing about AI ECG, it did not require a new hardware. ECG machines exist everywhere, and we can just use that to add another layer of information and add value to existing medical tests. Yeah, indeed. And we also learned humility. Some of the early assumptions we uh had didn't halt. Models behave different across populations. Some alerts didn't lead to action. Some clinicians were reluctant to believe that the output of the AI was real or was true or was reliable. So we have to move, we have to iterate. And that's maybe the most important thing, how clinicians and AI scientists working together to create this feedback loop that allows us to improve the model to see when we should alert, when we should send the patient to an echocardiogram, and having clinicians plus algorithms working together to reduce miss diseases. Yes, and today we see how the impact has expanded. For example, single lead ECGs promise a walk, remote monitoring, home-based care. Patients will never come in for a screening that now are coming are being followed. And for patients with chronic disease it's even more important because it's not longer about diagnosis, it's about being able to monitor them and make sure that the intervention actually helps them. That is true. So if we step back, the biggest change is not technological, right? It's about how clinicians are using the AI to find diseases earlier and having fewer patients that meet us for the first time at their worst moment. Yeah, I don't think AI is really making cardiologists smarter. I think it's helping us to make diagnosis earlier, to be more consistent, and to be more equitable. That's the, the real story of the last decade, using AI not to replace clinical judgment, but to augment it. That is true, and that is truly the outcome that matters.
