Electrocardiogram generation with a bidirectional LSTM-CNN generative adversarial network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Heart disease poses a significant global health risk.
• Electrocardiogram (ECG) is crucial for diagnosing heart conditions.
• Automated diagnosis requires large, privacy-protected datasets, a current challenge.

Purpose of the Study:

• To develop a method for generating synthetic ECG data that retains clinical features.
• To address the challenge of limited privacy-protected data for training diagnostic models.
• To improve automated heart disease diagnosis through data augmentation.

Main Methods:

• A generative adversarial network (GAN) model, termed BiLSTM-CNN GAN, was developed.
• The generator utilized bidirectional long short-term memory (BiLSTM) networks.
• The discriminator was based on convolutional neural networks (CNNs).
• The model was trained using 48 ECG records from the MIT-BIH database.

Main Results:

• The BiLSTM-CNN GAN demonstrated the fastest convergence of its loss function to zero compared to RNN-AE and RNN-VAE.
• Evaluations showed that the BiLSTM-CNN GAN generates synthetic ECG data with high morphological similarity to real recordings.
• The proposed GAN architecture outperformed other generative models in data synthesis.

Conclusions:

• The BiLSTM-CNN GAN is effective in generating high-fidelity synthetic ECG data.
• This approach can help overcome data limitations in automated heart disease diagnosis.
• The method holds promise for enhancing the development of AI-driven cardiovascular diagnostic tools.