Variational autoencoder-based neural electrocardiogram synthesis trained by FEM-based heart simulator

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• Traditional electrocardiogram (ECG) synthesis relies on complex heart models requiring supercomputers.
• This computational demand limits accessibility and efficiency in ECG synthesis.

Purpose of the Study:

• To develop an efficient method for synthesizing 12-lead ECG signals using cardiac parameters.
• To create a computationally inexpensive alternative to finite element method (FEM)-based heart simulators.

Main Methods:

• A variational autoencoder (VAE) model was developed, conditioned by cardiac parameters.
• The VAE encoder/decoder learns the relationship between ECG signals and cardiac parameters.
• Training data was generated using a comprehensive FEM-based heart simulator.

Main Results:

• The VAE model successfully synthesized adequate ECG signals, preserving key features and overall shape.
• Exploration of model architecture (layers, latent variables) optimized the balance between complexity and accuracy.
• Synthesized ECGs were generated within seconds on a laptop.

Conclusions:

• The proposed VAE method offers a computationally efficient alternative to FEM-based heart simulators.
• This approach enables rapid ECG synthesis from various cardiac parameters.
• The method demonstrates potential for widespread use in ECG analysis and research.