Generation and Selection: A Self-Iterative Two-Stage Data Augmentation Method for Automated ECG Classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Automated electrocardiogram (ECG) classification is vital for clinical diagnosis but hindered by limited labeled data.
• Existing ECG data augmentation techniques can introduce noise, degrading classification model performance.
• Generative models for ECG data augmentation often struggle with label accuracy and data quality.

Purpose of the Study:

• To develop a novel self-iterative two-stage data augmentation method (SiTs-ECG) for automated ECG classification.
• To address the challenges of data scarcity and noise in ECG data augmentation.
• To improve the performance of automated ECG classification models through enhanced data generation and selection.

Main Methods:

• Proposed SiTs-ECG, a self-iterative two-stage augmentation approach.
• Generation stage: Utilized an unconditional diffusion model guided by a Transformer encoder to generate high-quality ECG-like samples.
• Selection stage: Employed a base classification model to assign pseudo-labels and selected confidently predicted samples for iterative refinement.

Main Results:

• SiTs-ECG significantly improved classification metrics on three real-world datasets.
• On the Apnea-ECG dataset with ECG-Transformer, Precision, Recall, F1, and Accuracy increased by 7.9, 9.1, 9.2, and 7.3 percentage points, respectively.
• Demonstrated versatility and compatibility with various generative and classification models.

Conclusions:

• SiTs-ECG effectively enhances automated ECG classification by generating high-quality, confidently labeled data.
• The self-iterative approach continually improves classification model performance.
• SiTs-ECG shows significant promise for clinical applications in automated ECG analysis.