Reimagining atrial fibrillation screening beyond age-based thresholds using AI

Figures

• 

  Fig. 1. Transfer learning pipeline enabling AF risk prediction from handheld single-lead ECGs. A schematic overview of the AI-enabled transfer learning pipeline used to predict atrial fibrillation (AF) risk from handheld single-lead (1 L) ECG recordings. A convolutional neural network (CNN) is first pre-trained on over 450,000 clinical 12-lead ECGs (Steps 1–2). Transfer learning adapts the model for use with handheld 1 L ECGs (Step 3). A 30-second 1 L ECG is acquired (Step 4), the cleanest 10-second window is automatically identified (Step 5), and the segment is analyzed by the 1 L ECG-AI model (Step 6) to produce an AF risk estimate (Step 7).