Specialized ECG data augmentation method: leveraging precordial lead positional variability.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning models achieve high performance in diverse fields, with data augmentation being a key technique.
• Data augmentation synthesizes new data while maintaining accurate labels, crucial for training robust models.
• Existing data augmentation methods lack optimization for the specific characteristics of electrocardiogram (ECG) data.

Purpose of the Study:

• To introduce a novel data augmentation technique tailored for 12-lead ECG data.
• To enhance the diagnostic accuracy of deep learning models for cardiovascular conditions using ECG signals.
• To validate the effectiveness of the proposed ECG data augmentation method.

Main Methods:

• Developed a data augmentation technique focusing on the angles between precordial leads in 12-lead ECG.
• Applied the proposed technique to train a deep learning model for diagnosing atrial fibrillation/flutter, supraventricular tachycardia, AV block, LBBB, and myocardial infarction.
• Evaluated the model's performance against other data augmentation methods across various datasets.

Main Results:

• The proposed data augmentation method demonstrated improved performance compared to existing techniques for most diagnostic tasks.
• The technique is simple to implement and resulted in reduced total training time.
• The method showed potential in enhancing diagnostic accuracy for ECG-based condition identification.

Conclusions:

• The novel ECG data augmentation technique effectively improves deep learning model performance for cardiovascular disease diagnosis.
• This approach offers a practical and efficient solution to the lack of optimized data augmentation for ECG data.
• The study contributes to advancing bio-signal processing and deep learning in clinical applications.