Artificial Intelligence (AI) Enables Accurate Screening for Atrial Fibrillation

A new artificial intelligence (AI)-infused ECG algorithm was successful at identifying atrial fibrillation (AFib), according to a new Mayo Clinic study.

“When people come in with a stroke, we really want to know if they had AFib in the days before the stroke, because it guides the treatment,” lead author Paul Friedman, MD, chair of the Department of Cardiovascular Medicine at Mayo Clinic in Rochester, Minn., said in a press release. “Blood thinners are very effective for preventing another stroke in people with AFib. But for those without AFib, using blood thinners increases the risk of bleeding without substantial benefit. That’s important knowledge. We want to know if a patient has AFib.”

Deep learning #AI may identify atrial fibrillation from a normal rhythm #ECG: finding from study involving almost 181,000 patients & the first to use deep learning to identify patients with potentially undetected #AF with an overall accuracy of 83% https://t.co/YVweXpis0X pic.twitter.com/pqy6u9iVcj

— The Lancet (@TheLancet) August 1, 2019

To accomplish this, the authors conducted a retrospective analysis that included 180,922 patients with a total of 649, 931 normal sinus rhythm ECGs. Patients with at least one ECG that indicated AFib or atrial flutter were classified as “positive” for AFib. They were then separated into training (n=126,526), interval validation (n=18,116), and testing (n=36,280) datasets (7:1:2 ratio), and area under the curve (AUC) of the receiver operating characteristic curve was calculated for the internal validation set, and the resulting probability threshold was applied to the testing data set.

According to the study results, a single AI-enabled ECG identified AFib with an AUC of 0.87 (95% CI, 0.86 to 0.88), a sensitivity of 79%, specificity of 79.5%, an F1 score of 39.2%, and overall accuracy of 79.4. When all acquired ECGs from each patient’s window of interest were included, AUC rose to 0.90, sensitivity to 82.3%, specificity to 83.4%, F1 score to 45.5%, and overall accuracy to 83.3%.

Finding A Needle in a Haystack

“An ECG will always show the heart’s electrical activity at the time of the test, but this is like looking at the ocean now and being able to tell that there were big waves yesterday,” Dr. Friedman added. “AI can provide powerful information about the invisible electrical signals that our bodies give off with each heartbeat — signals that have been hidden in plain sight.”

Commentary by Jeroen Hendriks, PhD, of the University of Adelaide in Australia, and Larissa Fabritz, MD, of the University of Birmingham in the U.K., which accompanied the piece, noted that these results could reduce the time it takes to detect AFib.

“Rather than finding the needle in the haystack by prolonged monitoring, authors basically suggest that AI will be able to judge by looking at the haystack if it has a needle hidden in it,” they wrote.

Using ~650,000 electrocardiograms from over 180,000 patients, a neural network could predict future atrial fibrillation when patients were in normal sinus rhythm (AUC 0.87-90) #AI #deeplearninghttps://t.co/ISTR40WrD7 @zachia5 @noseworthypete @drpaulfriedman @MayoClinicCV pic.twitter.com/tGizLxPf54

— Eric Topol (@EricTopol) August 1, 2019

A potentially practice changing innovation by the HRS team @MayoClinicCV. Now @TheLancet, AI can identify PAF on surface EKG taken in sinus rhythm. @drpaulfriedman https://t.co/vkMLEu30iD

— Mohamad Alkhouli (@adnanalkhouli) August 1, 2019