A Deep Neural Network for Interpreting Wearable Electrocardiogram Data in Atrial Fibrillation: Prospective...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Devices

Background:

• Atrial fibrillation (AF) and atrial flutter (AFL) are common arrhythmias increasing ischemic stroke risk.
• Early diagnosis is crucial for effective anticoagulation therapy but challenging due to paroxysmal and asymptomatic nature.
• Current diagnostic methods are time-consuming and resource-intensive.

Purpose of the Study:

• To evaluate a mobile system with a wearable ECG and AI for detecting AF/AFL episodes and burden.
• To assess the system's ability to detect rhythm changes and estimate detection delay.
• To determine the rhythm classification performance of the AI algorithm.

Main Methods:

• 116 patients with recent-onset AF/AFL undergoing cardioversion were monitored.
• A wireless single-lead chest strap ECG system was used alongside a 3-lead Holter ECG as reference.
• A deep neural network (DNN)-based AI analyzed ECG data for AF/AFL detection, burden estimation, and rhythm change detection.

Main Results:

• The AI system achieved 91.9% sensitivity and 99.6% specificity for AF/AFL detection.
• Sensitivity for AF detection was 96.2%, while AFL detection sensitivity was 55.8%.
• High agreement (ICC=0.96) was found for AF/AFL burden estimation; rhythm changes were detected within 1 minute.

Conclusions:

• The AI-powered mobile ECG system shows strong performance in detecting AF, estimating burden, and recognizing rhythm changes.
• This automated, AI-driven approach supports clinical decision-making for arrhythmia management.
• Further validation in real-world ambulatory settings is recommended.