# Q&A: UM's Venkatesh Murthy on how AI can help cardiovascular patients - The Detroit News

*Источник: The Detroit News*
*Дата: 2026-08-02*
*Язык: en*

**Кратко:** Q&A: UM's Venkatesh Murthy on how AI can help cardiovascular patients
A University of Michigan professor and his team have created an AI model that can detect if a person may have a certain kind of cardiovascular disease, which could improve access to care for patients with the disease. Venkatesh Murthy, the Melvyn Rubenfire Professor of Preventive Cardiology at the University of Michigan, said the AI tool uses data from electrocardiograms, a common medical test, to determine whether people could have microvascular dysfunction.

Q&A: UM's Venkatesh Murthy on how AI can help cardiovascular patients
A University of Michigan professor and his team have created an AI model that can detect if a person may have a certain kind of cardiovascular disease, which could improve access to care for patients with the disease.
Dr. Venkatesh Murthy, the Melvyn Rubenfire Professor of Preventive Cardiology at the University of Michigan, said the AI tool uses data from electrocardiograms, a common medical test, to determine whether people could have microvascular dysfunction.
Microvascular dysfunction refers to problems with small blood vessels in the body, which cause people chest pain and other health issues, Murthy said in a recent interview. The disease is believed to be particularly common in women.
He said the cardiac PET scan is "the gold standard approach" for diagnosing microvascular dysfunction in the heart, but it isn't available in many communities. Even where it is available, there is limited capacity, he said. In many centers, patients have to wait three or more months to get the tests, so many people go undiagnosed, Murthy said. His team's method, which is not yet cleared by the Food and Drug Administration, could help with this issue.
"We believe this method will help identify the patients who would most benefit from gold-standard testing and would also help identify those who don't have microvascular dysfunction," Murthy said in an email.
The AI tool was accurate around 75% of the time in Murthy's team's study, which was published in November 2025. Murthy thinks that with further development, it could exceed 90%.
He said his team will soon be evaluating the tools in data from patients around the country and the world as part of the Wellcome Leap VISIBLE program.
Murthy discussed his research with The Detroit News in an interview that has been edited lightly for brevity and clarity. Murthy used the terms "ECG" and "EKG" interchangeably. They both refer to an electrocardiogram.
Question: Tell me about ... the AI model that you're using and how it's helping your department with the work that you do?
Answer: ... It's incredible what we're able to do with AI, but ... the challenge is that historically you need a ton of data with ... what we call gold standards paired with it in order to train AI models. So if you want an AI model that knows how to identify a person who has heart disease, you'd need a lot of both inputs, but also people with and without heart disease that have been very carefully tested and proven to have it or not, and you'd need large, large numbers ― hundreds of thousands, maybe, in some cases.
And so that is really challenging to get that kind of high quality data, so ... we started to take the approach of: we actually want this to be useful not just for common diseases where we can find hundreds of thousands of patient data sets, but like also for things that either are relatively common, but require advanced testing that might be rare, or are rare diseases outright. ... We want the tools to be available broadly and ... easily for all of that. ...
The work we focused on initially has been on identifying microvascular dysfunction. So people can get abnormalities of the large vessels of the heart called coronary arteries, and ... that's what you evaluate on a traditional stress test or a CT scan, but there are also small vessels that ... affect the heart, and (there) also are small vessels in all of your organs, and ... these small vessels are really important for a lot of diseases, and they can cause people chest pain if they're not working properly or they're too few of them or they're not able to carry enough blood supply. It can cause heart failure over time. The similar disease process in the brain can contribute to dementia, in the kidneys can contribute to heart failure. ...
And the way we test for it traditionally at Michigan has been PET scans. PET scans are super expensive. ... They're only maybe a half a dozen ... centers in the state of Michigan ... that offer this kind of testing, maybe fewer than that. It's a moving target, of course. And so the wait times are long. It takes three or four months for a patient to be scheduled for a PET scan sometimes.
And so we thought, can we help develop a tool using commonly available data ― EKG data ― to identify the people who might have an abnormal PET scan? And so that's what we set out to do, and we used ... some kind of tricks to be able to do this with a lot less data.
Q: Can you explain to me what an ECG is?
A: Oh, an electrocardiogram .... An ECG is an electrocardiogram. .... It's a very commonly done electrical test of the heart. Many people have it done in their primary care doctor's office. It involves putting some stickers on the chest, which are connected to ... electrodes, and you can use that test to look for your heart rhythm. You can look for evidence of whether you're having a heart attack or not. ...
If we could use that really inexpensive information to figure out, like, how we could do more complex testing, like to get the information that you get from more expensive testing, even if it's not perfect, it would improve access in a big way.
Q: What ended up happening once you looked through all that data, and what did you find?
A: The goal is to find people who are likely to ... either be very normal ― they don't need a PET scan, tell the doctor that ― or ... potentially abnormal, and therefore worth investigating further and for them to make a trip to wait, to go through the process of getting more complicated testing.
And so we found that we can actually do this using not hundreds of thousands of data sets, but actually a few thousand. We think more data will be still helpful. And we're in the process of kind of trying to extend it, but the initial study, we looked at around 4,000 ... scans with ECGs as well, but we first pre-trained our AI on 800,000 ECGs, which didn't have PET scans. So the AI is first taught: here's all this plentiful data without PET scans, so it understands the picture of what an ECG should look like. ... It doesn't necessarily know normal or abnormal, but it knows the range of what something should be.
Q: When you were looking at that, was that like people who had just undergone the testing, or is that historical information?
A: Yeah, exactly. It's all historical data that people who had ... come to the University of Michigan for care, and ... we handled the process to reuse that data under IRB. ... IRB is called Institutional Review Board supervision, where we work on this data, protecting people's identity and privacy, but use it to learn from.
Q: AI is being used to help doctors diagnose and identify some medical conditions. What do you expect for the future of this?
A: The role of AI as a diagnostic is only going to grow. We have already seen many examples where AI has enabled diagnoses earlier in the course of disease, for example.
There are now FDA-cleared AI tools in virtually every aspect of medicine, many in medical imaging and cardiovascular arenas. They are used to identify disease, generate quantitative measures of disease which doctors can track over time, reduce errors, lower costs, or improve turn around/access. The number of new tools is growing exponentially.
asnabes@detroitnews.com

[Оригинал](https://eu.detroitnews.com/story/news/local/michigan/2026/08/02/qa-ums-venkatesh-murthy-on-how-ai-can-help-cardiovascular-patients/90985751007/)