Data imbalance in cardiac health diagnostics using CECG-GAN

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Heart disease remains a leading global cause of mortality.
• Electrocardiogram (ECG) diagnostic models face limitations due to poor data quality and imbalance.
• Existing methods struggle with ECG waveform jitter and slow processing.

Purpose of the Study:

• To develop an advanced generative model for high-fidelity ECG data synthesis.
• To overcome challenges of data scarcity, imbalance, and waveform anomalies in ECG datasets.
• To enhance the performance of diagnostic models for heart disease detection.

Main Methods:

• Implementation of a conditional generative adversarial network (CECG-GAN).
• Integration of a transformer architecture to address waveform jitter and processing speed.
• Utilizing MIT-BIH and CSPC2020 datasets for model evaluation.

Main Results:

• The CECG-GAN demonstrated superior performance in generating ECG data.
• High similarity between generated and actual ECG waveforms was confirmed (PRD: 55.048).
• Low error metrics were achieved: Fréchet distance (1.139), RMSE (0.232), and MAE (0.166).

Conclusions:

• CECG-GAN effectively generates realistic ECG data, addressing key limitations in current diagnostic approaches.
• The proposed method shows significant potential for improving heart disease diagnostic models.
• This approach offers a robust solution for ECG data augmentation and quality enhancement.