Dr Brandon Fornwalt Discusses the Preventive Efficacy of AI in Monitoring for Atrial Fibrillation

November 30, 2019

Artificial intelligence has the potential to change the way physicians have monitored for conditions such as atrial fibrillation in the past 5 decades, said Brandon Fornwalt, MD, PhD, director of the Cardiac Imaging Technology Laboratory at Geisinger.

Artificial intelligence has the potential to change the way physicians have monitored for conditions such as atrial fibrillation in the past 5 decades, said Brandon Fornwalt, MD, PhD, director of the Cardiac Imaging Technology Laboratory at Geisinger.

Transcript

Can you discuss the initial intrigue in artificial intelligence for both electrocardiogram (ECG) trace studies? What must physicians and patients understand about its efficacy?

Electrocardiograms are a very widely used test in medicine. Most patients by the end of their lives have undergone an electrocardiogram at some point. We've spent a lot of time over the last many decades understanding how to do automated computer analysis of electrocardiogram data. Recently, artificial intelligence algorithms have improved, and data sets have gotten larger, and now we can use different approaches called deep learning to further analyze these signals that we collect from patients. So, the intrigue is, can we start to use deep learning to tell us insights about patients that are going to help us treat them better.

As far as efficacy is concerned, we have spent a lot of time building models where we know we can predict in a bunch of data fairly accurately what the likelihood of these events will be in the future; but what physicians and patients need to understand is that it's going to take a lot of prospective clinical trials to really demonstrate that when we know that prediction—we can actually impact an outcome that a patient cares about, for example, reduce the risk of stroke or reduce the risk of mortality.

Based on study results showing that AI predicted one-year risk of atrial fibrillation (AF) directly from ECG traces, can you discuss the preventive significance of the deep neural network as an accurate predictor of AF?

We had a couple of interesting findings in the studies. We’re trying to find ways in which these algorithms can add value above and beyond what a physician already does. Physicians do a ton of really good stuff with electrocardiogram data, and what we found is that we have prediction power to predict these future clinical events like atrial fibrillation and mortality in the broad population of patients that undergoes electrocardiogram. Then, building on that concept of doing something above and beyond what a physician does, we looked at cases where the physician called these ECGs normal. So, that means they said, look, there's nothing wrong here—move on next case, normal electrocardiogram. Now, the models still had prediction power inside of those ECGs, and that means that the model is latching on to things that right now as physicians, we either recognize them and think they're clinically insignificant, or we don't recognize them at all. That's what we think is really cool about AI in medicine is, it’s going to come in and potentially teach us things that can change the way we've been doing something for the last 5 decades.