ECG synthesis for cardiac arrhythmias: Integrating self-supervised learning and generative adversarial networks

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Supervised deep learning for arrhythmia classification requires extensive labeled clinical data, often limited by disease prevalence and patient anonymization policies.
• Current models face performance limitations due to imbalanced data distributions and privacy concerns in electronic health records.

Purpose of the Study:

• To develop a conditional generative architecture for electrocardiography (ECG) time series that integrates self-supervision and generative adversarial principles.
• To generate realistic synthetic ECG data that can augment limited clinical datasets and improve arrhythmia classification.
• To address challenges related to data scarcity and patient anonymization in ECG analysis.

Main Methods:

• Introduced a novel conditional generative model named ECGAN, combining self-supervision and generative adversarial networks (GANs).
• Employed empirical validation to assess the morphological plausibility and realism of generated synthetic ECG signals.
• Evaluated the model's performance on various rhythm abnormalities, including congestive heart failure, myocardial infarction, sinus rhythm, and premature ventricular contractions.

Main Results:

• ECGAN demonstrated enhanced morphological plausibility and generated realistic synthetic ECG signals.
• The model proved capable of conditioning the probability distribution of ECG recordings.
• Achieved an average improvement of 2.4% in arrhythmia classification accuracy across the MIT-BIH, BIDMC, and PTB datasets compared to state-of-the-art models.

Conclusions:

• The proposed ECGAN methodology effectively generates realistic synthetic ECG time series data.
• This approach enhances arrhythmia classification accuracy and improves training stability.
• The workflow offers a competitive solution for augmenting ECG datasets and addressing data limitations in cardiovascular research.