uPVC-Net: A Universal Premature Ventricular Contraction Detection Deep Learning Algorithm.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Premature Ventricular Contractions (PVCs) are common arrhythmias.
• Accurate PVC detection from electrocardiogram (ECG) waveforms is challenging due to signal variability.

Purpose of the Study:

• To develop a universal deep learning model for PVC detection from any single-lead ECG.
• To evaluate the model's out-of-distribution generalization capabilities.

Main Methods:

• Developed uPVC-Net, a custom deep learning architecture.
• Trained the model on four independent ECG datasets (8.3 million beats) from Holter monitors and wearable patches.
• Employed a multi-source, multi-lead training strategy.

Main Results:

• uPVC-Net achieved an Area Under the Curve (AUC) between 97.8% and 99.1% on held-out datasets.
• Performance on wearable single-lead ECG data reached an AUC of 99.1%.

Conclusions:

• uPVC-Net demonstrates robust generalization across varied ECG lead configurations and patient demographics.
• The model shows significant potential for real-world clinical application in PVC detection.