A transformer-based deep neural network for arrhythmia detection using continuous ECG signals.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Machine learning methods are increasingly used for arrhythmia classification.
• Existing methods often overlook crucial inter-heartbeat dependencies, limiting performance.
• A novel approach is needed to leverage these dependencies for improved arrhythmia detection.

Purpose of the Study:

• To propose a novel transformer-based deep learning neural network, ECG DETR, for arrhythmia detection.
• To perform simultaneous prediction of heartbeat positions and categories within continuous single-lead ECG segments.
• To develop a compact, end-to-end algorithm that does not require explicit heartbeat segmentation.

Main Methods:

• Developed ECG DETR, a transformer-based deep learning model.
• Applied the model to continuous single-lead ECG segments for simultaneous position and category prediction.
• Validated performance on the MIT-BIH arrhythmia and MIT-BIH atrial fibrillation databases using 10-fold cross-validation.

Main Results:

• The ECG DETR model achieved high accuracy across three different arrhythmia detection tasks (8, 4, and 2 labels).
• Overall accuracies reached 99.12%, 99.49%, and 99.23% for the respective tasks.
• The proposed method demonstrated comparable performance to previous works in both segmentation and classification.

Conclusions:

• ECG DETR effectively utilizes inter-heartbeat dependencies for enhanced arrhythmia classification.
• The model offers a compact, end-to-end solution for arrhythmia detection without explicit beat segmentation.
• The validated performance confirms the generalizability and efficacy of the proposed deep learning approach.