Neural network predicts risk of atrial fibrillation

  • August 3, 2022
  • Steve Rogerson

Newly published research shows that a deep learning model can predict the short-term risk of atrial fibrillation (AFib) based on 24-hour Holter recordings that show normal sinus rhythm

AFib affects millions of people each year. However, the condition is often unrecognised and untreated. Patients are often subject to 24-hour ambulatory electrocardiograms (ECGs) to receive a diagnosis, but this short-duration recording is known to have a low diagnostic yield and misses many patients with infrequent AFib episodes.

A recent study published in European Heart Journal – Digital Health examined a learning model using a deep neural network from Cardiologs, part of Philips’ cardiac monitoring and diagnostics offering, to identify patients at risk of AFib in the two weeks following a 24-hour ambulatory ECG with no previously documented AFib.

Led by Jagmeet Singh, cardiologist at Massachusetts General Hospital (MGH) and professor of medicine at Harvard Medical School, the study consisted of training the deep neural network to predict the near-term presence or absence of AFib by only using the first 24 hours of an extended Holter recording.

Results showed that the network was able to predict whether AFib would occur in the near future with an area under the receiver operating curve, sensitivity and specificity of 79.4%, 76%, and 69%, respectively, and outperformed ECG features previously shown to be predictive of AFib. These results showed a ten-point improvement compared to a baseline model using age and sex.

“The Cardiologs study shows that 24-hour Holter data can be used to enhance current monitoring capabilities, bringing hope to high-risk patients who would benefit from proactive treatment and AFib mitigation strategies,” says Singh. “By getting patients the care they need sooner and potentially preventing more severe outcomes, we could help save lives.”

The Cardiologs study demonstrates the capability of artificial intelligence in predicting AFib in the short-term using 24-hour Holter compared to resting 12-lead ECGs. While 12-lead ECG gives access to a larger view of the hearts’ activity for a short period, 24-hour Holter provides longer-duration signals, therefore, offering additional inputs for predicting models.

The extension of AI capabilities towards predictions and digital biomarkers has the potential to bring improved health outcomes leading to new diagnostic paradigms. Predictive biomarkers may lead to early detection, enhanced patient monitoring and improved patient management in general.

Philips’ portfolio of cardiac care technology includes real-time patient monitoring, therapeutic devices, telehealth and informatics for the hospital, as well as ambulatory cardiac diagnostics and monitoring. Developed in partnership with physicians, Cardiologs’ technology accelerates diagnostic reporting, decreases the occurrence of reporting errors, and streamlines clinician workflow and patient care, empowering clinicians to deliver cardiac care faster and more efficiently.