Will Algorithms See What Clinicians Cannot? AI Application for ECG and MRI

Discussion With Bradley Knight, MD, and Han Feng, PhD

© 2026 HMP Global. All Rights Reserved.

Any views and opinions expressed are those of the author(s) and/or participants and do not necessarily reflect the views, policy, or position of EP Lab Digest or HMP Global, their employees, and affiliates. 

Edited by Jodie Elrod

Bradley Knight, MD, talks with Han Feng, PhD, about his presentation entitled "Will Algorithms See What Clinicians Cannot? AI Application for ECG and MRI" at Western AFib 2026.

Bradley Knight, MD: Hi, I'm Brad Knight, Editor-in-Chief of EP Lab Digest. We're here at the Western Atrial Fibrillation (AFib) meeting in Salt Lake City. I'm excited to have Han Feng, PhD, from Tulane. He's a biostatistician and has done a lot of work with artificial intelligence (AI). I attended your session and I have a few questions for you, but maybe we can start with the fact that you welcomed us to the Chinese New Year.

Han Feng, PhD: Yes. Thank you. 

Bradley Knight, MD: So when was that?

Han Feng, PhD: It was last week, Tuesday.

Bradley Knight, MD: It's the year of the horse, and you made a point that horses have a lot of AFib.

Han Feng, PhD: Yes. I happen to know that. I feel like it was a coincidence.

Bradley Knight, MD: I think the whale has the most AFib because it has the largest atrium, but horses have a lot of AFib.

Han Feng, PhD: Yes, as well.

Bradley Knight, MD: So the title of your presentation was “Will Algorithms See What Clinicians Cannot? AI Application for Electrocardiograms (ECGs) and Magnetic Resonance Imaging.” You initially and frequently referred to the term SMURDEN. Can we start with that? What is SMURDEN?

Han Feng, PhD: Because in clinical practice, AF monitoring can feel limited. For example, a patch can only be used for about 2 weeks, and an insertable cardiac monitor (ICM) cannot be provided to every patient. So what do we do for longer-term monitoring? This is why we started thinking about using AI with a portable, smartphone-based device that allows patients to record their own ECG at home. They can perform the ECG themselves and send the data to us. This idea led to SMURDEN. We collect the data from the patient side and developed a concept similar to AF burden, but based on data from these smart portable devices.

Bradley Knight, MD: So SMURDEN is smartphone-based AF burden.

Han Feng, PhD: Yes.

Bradley Knight, MD: All right. So before we talk about the specifics of your presentation, just in general, we hear a lot about AI. It's on the news every day about all these potential applications in health care. But honestly, when I go to work every day, very few things are touched by AI and health care. Where are the current applications of AI in cardiology?

Han Feng, PhD: I think there are 2 sides to this. One side is within the lab, in the experimental setting. We are doing research to see whether it will work, and if it does, we plan to make it scalable and apply it more broadly. The other side is that AI is already part of our clinical practice, even if we were not fully aware of it. As I mentioned, devices like ECG patches and ICMs have both hardware and software components, and their software is based on relatively simple machine learning.

Bradley Knight, MD: For example, like the Zio patch (iRhythm), people don't recognize that they are using already AI or machine learning as part of an algorithm.

Han Feng, PhD: Yes. Also, in the clinic, systems such as Epic already use AI. For example, when you are entering diagnoses or working with insurance information, AI is assisting in the background. So this is already happening. In addition, some nurses are using AI tools to help translate conversations with patients, which allows them to create better notes afterward.

Bradley Knight, MD: Yes, it’s an audible way to document your clinic notes.

Han Feng, PhD: Yes.

Bradley Knight, MD: I think there are other examples too. For example, one of the mapping systems has automated left atrial detection with ultrasound. So it's an ICE-based automatic detection and it's AI based. But in general, I would say, compared to the rest of the world, health care is a little behind in terms of applications directly. What are the obstacles to bringing AI into the health care field? Is it regulation? Who's paying for it? What are your feelings about that?

Han Feng, PhD: My feeling is that there are 2 major obstacles. One is FDA approval. The FDA acts as a gatekeeper to ensure that any new technology or advancement in health care does not harm patients. That is why progress in health care is relatively slower compared to other fields. For example, companies like Google or Amazon can collect data directly, run trials, and if an algorithm does not perform well, they can quickly adjust it and continue. This allows them to adapt very quickly. In health care, however, everything we apply directly affects patient health. We want to avoid causing any harm, and we care about every individual patient. Because of this, we tend to be more conservative, which can slow progress—not only in AI, but in medicine overall.

Bradley Knight, MD: When I talk to colleagues outside of medicine, law firms can bring this stuff in pretty quickly. Consulting firms can take Claude and form a relationship where their stuff remains proprietary to them. They don't get the information, so they don't have these FDA hurdles. Are those obstacles true around the world? In the rest of the world, are there similar approval obstacles for AI and health care, or is that too hard to know currently?

Han Feng, PhD: From my understanding, in China the regulations are relatively less strict than in the United States, and this is both good and bad. In the US, we apply stricter rules to protect patients and to ensure that new advancements truly benefit people. Much of the major innovation in health care happens here. In China, the regulatory approach can be more relaxed, partly because many of the innovations are more marginal and not as critical. Because of that, they may apply less strict oversight.

Bradley Knight, MD: Let's get back to your presentation. You talked a lot about the difference between photoplethysmography (PPG), which is a device that counts pulses and provides a different wave form, versus an electrocardiogram or rhythm recording. What are the pros and cons of these 2 technologies?

Han Feng, PhD: PPG is mainly based on a light signal to measure blood flow, so it indirectly measures the pulse. One advantage is that, because it can emit light continuously, it doesn’t require the patient to stay still and can automatically collect data.

Bradley Knight, MD: Right. And you talked about how people have to hold still to get a clean ECG.

Han Feng, PhD: Yes. But that's not the case with PPG. You can wear a watch and go to sleep, and it will continue collecting your PPG signal.

Bradley Knight, MD: So if a patient comes to see me and says their Fitbit told them they have AFib, compared with a patient who bought a cardio app and shows me an ECG recording, which is more reliable? If someone tells me their Fitbit detected AFib, should I consider that accurate?

Han Feng, PhD: No, I don't think so. Not at this stage. But that one is more about informing, because they are not FDA approved. But the device is more accurate. So my presentation is not saying PPG is replacing all clinical settings right now. It may in the future, but right now we still rely heavily on the patch and ICM to do the diagnosis. But PPG and wristbands are also good for a couple of situations. One is your patient is cleared by you, but they may still be worried and want to continue to monitor.

Bradley Knight, MD: Cleared, meaning we think you're doing okay, but they still want to use it to look.

Han Feng, PhD: Yes. Also, the most important part is for people who have not yet come to your clinic. For patients with diabetes or coronary artery disease who may not be aware they have AFib, wearing a watch that provides this kind of information can prompt them to consider whether they should see an electrophysiologist to check their AFib status. That's the most important aspect right now.

Bradley Knight, MD: A lot of your work seems to be taking PPG, which has some limitations, and maybe applying AI to that to get it closer to the ECG because of the limitations of the ECG. Is that fair?

Han Feng, PhD: Yes. PPG is a surrogate signal for ECG, because ECG is measured directly. When you compare it to ECG, which directly reflects the cardiac electrical activity and pulse, PPG does not provide the same high-quality standard of signal. Because of that, it requires a lot of additional work. For example, we need signal quality checks and signal cleaning, and we have developed algorithms to determine whether a portion of the signal is usable and qualified for further classification. We want to avoid giving random or incorrect predictions. So there is a substantial amount of prerequisite work involved, and that is why we are very proud of the progress that has been made.

Bradley Knight, MD: What would you say you can accomplish now? You talked about some of what you've learned using the AI-based PPG as a predictor. What is it capable of predicting at this point?

Han Feng, PhD: It can be used to predict effort, and it can also predict certain features related to your blood. As we showed in the presentation, some blood biomarkers can be estimated using PPG data. Right now, the performance is around an AUC of 0.7, which is good. I expect that this will become more accurate in the future. With PPG, it may be possible to indicate that a patient’s glucose or LDL levels are high, potentially reducing the need for frequent blood tests in the future. As I mentioned, PPG provides a different type of signal compared to ECG. When the 2 are combined, they can offer more comprehensive information for prediction and a better understanding of the body.

Bradley Knight, MD: There may be changes in blood flow even though there's a regular rhythm. If you have a premature beat that doesn't generate a pulse, it's not going to be detected by the PPG. So there are different sources of information.

Han Feng, PhD: Sure.

Bradley Knight, MD: Is there anything else you want to share about what you presented?

Han Feng, PhD: I also want to emphasize that human beings have always used tools. In the past, we relied more on hardware and rule-based methods for measurement, and now we are gradually shifting toward AI. These are all tools. What matters is understanding how to use these tools effectively and how to apply them to make our lives easier—whether in clinical practice, research, engineering, or everyday life.

Bradley Knight, MD: Well, I really enjoyed your presentation. I appreciate your time now. I hope you enjoy the year of the horse.

Han Feng, PhD: Thank you.

The transcripts were edited for clarity and length.