Advancing electrocardiogram synthesis: Analyzing key metrics for enhanced evaluation.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• High-quality electrocardiogram (ECG) datasets are crucial for automated cardiac disease diagnosis.
• Existing challenges include data scarcity, small dataset sizes, and class imbalances.
• Deep generative models (DGMs) offer a solution by creating synthetic ECG data, enhancing privacy and addressing data limitations.

Purpose of the Study:

• To systematically review and evaluate metrics for assessing the quality of synthetic ECG data generated by DGMs.
• To identify limitations in current evaluation methodologies for synthetic ECGs.
• To propose a standardized framework for evaluating synthetic ECG data quality.

Main Methods:

• Comprehensive literature review of evaluation metrics for synthetic ECGs.
• Experimental analysis of metric strengths, weaknesses, and applicability.
• Critical assessment of existing evaluation frameworks.

Main Results:

• Current evaluation metrics for synthetic ECGs vary widely in application and effectiveness.
• Significant limitations exist in the current methodologies for assessing morphological and functional consistency.
• A need for a standardized, robust framework for synthetic ECG quality assessment is evident.

Conclusions:

• Deep generative models show promise for augmenting ECG datasets, but rigorous evaluation is essential.
• A standardized framework is proposed to ensure the reliability and fidelity of synthetic ECG data.
• This work provides a foundation for future research and downstream applications in automated cardiac diagnosis.