Why the study?
Automated ECG analysis is limited by scarce and imbalanced data, and while synthetic ECGs can mitigate these issues, most existing methods rely on generative adversarial networks rather than variational autoencoders.
A novel conditional variational autoencoder (cNVAE-ECG) successfully generates high-resolution synthetic 12-lead ECGs, improving downstream classification performance compared to existing GAN-based methods.
Cardiovascular diseases remain the leading cause of mortality. Automated electrocardiogram (ECG) analysis can ease clinical workloads but is limited by scarce and imbalanced data. Synthetic ECGs can mitigate these issues, and while most methods use generative adversarial networks (GANs), recent work has shown that variational autoencoders (VAEs) perform comparably. This paper presents a cNVAE-ECG model, a conditional Nouveau VAE (NVAE) that generates high-resolution 10-s 12-lead ECGs with multiple pathologies. Leveraging a compact channel-generation scheme and class embeddings for multilabel conditioning, cNVAE-ECG improves downstream binary and multilabel classification, achieving an AUROC gain of up to 2% in transfer learning over GAN-based models. The model is publicly available at https://github.com/univanxx/cNVAE_ECG .
No takes yet. Share an insight, caveat, or question.
Sviridov et al. (2025) studied Cardiovascular diseases (ECG abnormalities) (n=67,086). cNVAE-ECG (conditional hierarchical variational autoencoder) vs. GAN-based models (WaveGAN*, Pulse2Pulse) was evaluated on AUROC in binary and multi-label classification tasks.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: