Segment Origin Prediction: A Self-supervised Learning Method for Electrocardiogram Arrhythmia Classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Automatic arrhythmia classification systems are crucial for reducing cardiovascular disease mortality.
• Current deep learning models show promise but are limited by small dataset sizes.
• Improving ECG analysis accuracy is vital for effective arrhythmia detection.

Purpose of the Study:

• To introduce a novel self-supervised pre-training method, Segment Origin Prediction (SOP), to enhance arrhythmia classification performance.
• To address the limitations of small datasets in deep learning models for ECG analysis.
• To improve the accuracy and robustness of automatic arrhythmia classification systems.

Main Methods:

• Proposed a self-supervised pre-training method called Segment Origin Prediction (SOP).
• Designed a data reorganization module for predicting segment origins without annotations.
• Incorporated a feed-forward layer during pre-training to boost downstream classification performance.
• Applied SOP to six representative models and evaluated on the PhysioNet Challenge 2017 dataset.

Main Results:

• All baseline models demonstrated significant performance improvements after SOP pre-training.
• The proposed SOP method effectively enhances ECG feature learning.
• Experimental results confirmed the effectiveness of SOP in improving arrhythmia classification accuracy.

Conclusions:

• Segment Origin Prediction (SOP) is an effective self-supervised pre-training strategy for deep learning-based arrhythmia classification.
• SOP enhances model performance, particularly in scenarios with limited labeled ECG data.
• The method offers a promising approach to advance automatic arrhythmia detection and reduce cardiovascular disease mortality.