Atrial Fibrillation Detection from Ambulatory ECG with Accelerometry Contextualisation: A Semi-Supervised Learning...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Atrial fibrillation (AF) is a common arrhythmia requiring early detection via ambulatory electrocardiogram (ECG) screening.
• Deep learning (DL) shows promise for automated AF detection but requires extensive labelled data.
• Acquiring diverse labelled ECG data for DL models is costly and time-consuming.

Purpose of the Study:

• To propose and evaluate a semi-supervised learning (SSL) model for AF detection using variational auto-encoders (VAEs).
• To assess the impact of incorporating accelerometry data for ambulatory context on model performance.
• To demonstrate the efficacy of SSL in optimizing AF detection with limited labelled ECG data.

Main Methods:

• Developed an SSL model using a VAE architecture for AF detection on ambulatory ECG.
• Incorporated accelerometry data to account for free-living ambulatory contexts.
• Trained the model on a large dataset of 72,003 unique patients, classifying sinus rhythm, AF, and other arrhythmias.

Main Results:

• The SSL model achieved over 91% accuracy on an unseen test dataset and the CACHET-CADB dataset.
• High performance was maintained even with only 20% of the training data being labelled.
• The model demonstrated strong generalisability across different datasets.

Conclusions:

• SSL effectively enhances AF detection from ambulatory ECG using minimal labelled data.
• The proposed VAE-based SSL model offers a cost-effective solution for large-scale ECG analysis.
• Incorporating contextual data like accelerometry can further refine arrhythmia detection in real-world settings.