AI filter helps heart-monitoring devices limit false-positive AFib findings

Researchers have found that using an AI-based analysis tool can make implantable loop recorders (ILRs) more effective at detecting atrial fibrillation (AFib) episodes.

The team’s findings, published in JACC: Clinical Electrophysiology, focused on data from 425 patients who underwent ILR implantation at a single facility. The study’s cohort included patients with cryptogenic stroke, patients with known AFib being monitored for medical management and patients who had already undergone catheter ablation for AFib.

The AI solution in question, developed by Cardiologs, includes two convolutional neural networks trained on more than one million ECGs from an anonymized dataset. It was employed with the use of a deep neural network (DNN) for AFib detection.

Overall, the AI-based analysis tool increased the positive predictive value (PPV) of ILR-detected AFib episodes from 53.9% to 74.5%. The biggest improvement was seen in the AI solution’s ability to filter out false-positive findings.

“Routine application of such an AI-based solution has the ability to significantly affect clinical practice by reducing the time and effort needed to adjudicate false-positive AFib events,” wrote lead author Suneet Mittal, MD, a cardiologist at the Snyder Center for Atrial Fibrillation in Ridgewood, New Jersey, and colleagues. “We can conceptualize three distinct use case scenarios for this DNN filter in the management of ILR patients. The DNN filter could be applied directly to the digital electrocardiograms received by the manufacturer’s online platform. Alternatively, it could be applied to the PDF reports on either the manufacturer’s online platform or one of a third-party vendor.” 

While Mittal reported no relevant relationships, the study’s other authors were either Cardiologs employees or consultants. Cardiologs also covered all costs “associated with institutional review board approval of the study.”

The full study can be read here.