ellibsЭлектронные библиотекиRussian Digital Libraries Journal1562-5419Казанский (Приволжский) федеральный университет10.26907/1562-5419-2025-28-5-1186-1206ellibs-615Research ArticleСтатьиУсловная генерация электрокардиограмм с помощью иерархических вариационных автокодировщиковConditional Electrocardiogram Generation using Hierarchical Variation-al AutoencodersСвиридовИван АнатольевичSviridovIvan Anatolevichianatosviridov@sberbank.ruЕгоровКонстантин СергеевичEgorovKonstantin SergeevichEgorov.K.Ser@sberbank.ruSber AI LabSber AI Lab20251912202528511861206Copyright © Свиридов И.А., Егоров К.С., 20252025Свиридов И.А., Егоров К.С.Sviridov I.A., Egorov K.S.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://ellibs.elpub.ru/jour/article/view/615

Сердечно-сосудистые заболевания являются одной из основных причин смертности. Автоматический анализ электрокардиограмм (ЭКГ) может существенно облегчить работу врачей, но его эффективность ограничена нехваткой и несбалансированностью данных. Создание синтетических ЭКГ помогает частично решить эти проблемы. Хотя чаще всего для этого применяются генеративно-состязательные сети (GAN), но последние исследования показали, что вариационные автокодировщики (VAE) могут обеспечивать сопоставимое качество. В работе представлена модель cNVAE-ECG — модификация Nouveau VAE (NVAE), способная генерировать 12 отведений 10-секундных ЭКГ с различными патологиями. Используя компактную схему работы с каналами и встроенные представления классов для условной генерации, cNVAE-ECG улучшает результаты в задачах бинарной и multi-label классификации, обеспечивая прирост метрики AUROC до 2% по сравнению с моделями на основе GAN. Модель представлена в открытом доступе: https://github.com/univanxx/cNVAE_ECG.

Cardiovascular diseases remain the leading cause of mortality, and automated electrocardiogram (ECG) analysis can ease clinical workloads but is limited by scarce and imbalanced data. Synthetic ECG can mitigate these issues, and while most methods use Generative Adversarial Networks (GANs), recent work show variational autoencoders (VAEs) perform comparably. We introduce cNVAE-ECG, a conditional Nouveau VAE (NVAE) that generates high-resolution, 12-lead, 10-second ECGs with multiple pathologies. Leveraging a compact channel-generation scheme and class embeddings for multi-label conditioning, cNVAE-ECG improves downstream binary and multi-label classification, achieving up to a 2% AUROC gain in transfer learning over GAN-based models.

ЭКГвариационный автокодировщикусловная генерацияGANECGvariational autoencoderconditional generationGANReferencesTsao C.W., Aday A.W., Almarzooq Z.I., et al. Heart Disease and Stroke Statistics–2023 Update: A Report from the American Heart Association // Circulation. 2023. Vol. 147, No. 8. P. e93–e621.Tsao C.W., Aday A.W., Almarzooq Z.I., et al. Heart Disease and Stroke Statistics–2023 Update: A Report from the American Heart Association // Circulation. 2023. Vol. 147, No. 8. P. e93–e621.Liu X., Wang H., Li Z., Qin L. Deep learning in ECG diagnosis: A review // Knowledge-Based Systems. 2021. Vol. 227. P. 107187.Liu X., Wang H., Li Z., Qin L. Deep learning in ECG diagnosis: A review // Knowledge-Based Systems. 2021. Vol. 227. P. 107187.Gerke S., Minssen T., Cohen G. Ethical and legal challenges of artificial intelligence-driven healthcare // Artificial Intelligence in Healthcare / Eds. A. Bohr, K. Memarzadeh. Academic Press, 2020. P. 295–336.Gerke S., Minssen T., Cohen G. Ethical and legal challenges of artificial intelligence-driven healthcare // Artificial Intelligence in Healthcare / Eds. A. Bohr, K. Memarzadeh. Academic Press, 2020. P. 295–336.Reyna M.A., Sadr N., Alday E.A.P., et al. Will two do? Varying dimensions in electrocardiography: The PhysioNet/Computing in Cardiology Challenge 2021 // Computing in Cardiology. 2021. Vol. 48. P. 1–4.Reyna M.A., Sadr N., Alday E.A.P., et al. Will two do? Varying dimensions in electrocardiography: The PhysioNet/Computing in Cardiology Challenge 2021 // Computing in Cardiology. 2021. Vol. 48. P. 1–4.Friesen G., Jannett T., Jadallah M., Yates S., Quint S., Nagle H. A comparison of the noise sensitivity of nine QRS detection algorithms // IEEE Transactions on Biomedical Engineering. 1990. Vol. 37, No. 1. P. 85–98.Friesen G., Jannett T., Jadallah M., Yates S., Quint S., Nagle H. A comparison of the noise sensitivity of nine QRS detection algorithms // IEEE Transactions on Biomedical Engineering. 1990. Vol. 37, No. 1. P. 85–98.Maron B.J., Friedman R.A., Kligfield P., et al. Assessment of the 12-lead ECG as a screening test for detection of cardiovascular disease in healthy general populations of young people (12–25 years of age) // Circulation. 2014. Vol. 130, No. 15. P. 1303–1334.Maron B.J., Friedman R.A., Kligfield P., et al. Assessment of the 12-lead ECG as a screening test for detection of cardiovascular disease in healthy general populations of young people (12–25 years of age) // Circulation. 2014. Vol. 130, No. 15. P. 1303–1334.Kingma D.P., Welling M. Auto-encoding variational Bayes. 2022.Kingma D.P., Welling M. Auto-encoding variational Bayes. 2022.Vahdat A., Kautz J. NVAE: A deep hierarchical variational autoencoder. 2020.Vahdat A., Kautz J. NVAE: A deep hierarchical variational autoencoder. 2020.Golany T., Radinsky K. PGANs: Personalized generative adversarial networks for ECG synthesis to improve patient-specific deep ECG classification // Proceedings of the AAAI Conference on Artificial Intelligence. 2019. Vol. 33, No. 1. P. 557–564.Golany T., Radinsky K. PGANs: Personalized generative adversarial networks for ECG synthesis to improve patient-specific deep ECG classification // Proceedings of the AAAI Conference on Artificial Intelligence. 2019. Vol. 33, No. 1. P. 557–564.Yang H., Liu J., Zhang L., Li Y., Zhang H. ProEGAN-MS: A progressive growing generative adversarial networks for electrocardiogram generation // IEEE Access. 2021. Vol. 9. P. 52089–52100.Yang H., Liu J., Zhang L., Li Y., Zhang H. ProEGAN-MS: A progressive growing generative adversarial networks for electrocardiogram generation // IEEE Access. 2021. Vol. 9. P. 52089–52100.Golany T., Radinsky K., Freedman D. SimGANs: Simulator-based generative adversarial networks for ECG synthesis to improve deep ECG classification // Proceedings of the 37th International Conference on Machine Learning (PMLR). 2020. Vol. 119. P. 3597–3606.Golany T., Radinsky K., Freedman D. SimGANs: Simulator-based generative adversarial networks for ECG synthesis to improve deep ECG classification // Proceedings of the 37th International Conference on Machine Learning (PMLR). 2020. Vol. 119. P. 3597–3606.Nankani D., Baruah R.D. Investigating deep convolution conditional GANs for electrocardiogram generation // 2020 International Joint Conference on Neural Networks (IJCNN). 2020. P. 1–8.Nankani D., Baruah R.D. Investigating deep convolution conditional GANs for electrocardiogram generation // 2020 International Joint Conference on Neural Networks (IJCNN). 2020. P. 1–8.Thambawita V., Isaksen J.L., Hicks S.A., et al. DeepFake electrocardiograms using generative adversarial networks are the beginning of the end for privacy issues in medicine // Scientific Reports. 2021. Vol. 11. P. 21896.Thambawita V., Isaksen J.L., Hicks S.A., et al. DeepFake electrocardiograms using generative adversarial networks are the beginning of the end for privacy issues in medicine // Scientific Reports. 2021. Vol. 11. P. 21896.Wu J., Wang L., Pan H., Wang B. MLCGAN: Multi-lead ECG synthesis with multi label conditional generative adversarial network // ICASSP 2023 – IEEE International Conference on Acoustics, Speech and Signal Processing. 2023. P. 1–5.Wu J., Wang L., Pan H., Wang B. MLCGAN: Multi-lead ECG synthesis with multi label conditional generative adversarial network // ICASSP 2023 – IEEE International Conference on Acoustics, Speech and Signal Processing. 2023. P. 1–5.Alcaraz J.M.L., Strodthoff N. Diffusion-based conditional ECG generation with structured state space models. 2023.Alcaraz J.M.L., Strodthoff N. Diffusion-based conditional ECG generation with structured state space models. 2023.Ho J., Jain A., Abbeel P. Denoising diffusion probabilistic models // CoRR. 2020. abs/2006.11239.Ho J., Jain A., Abbeel P. Denoising diffusion probabilistic models // CoRR. 2020. abs/2006.11239.Dhariwal P., Nichol A. Diffusion models beat GANs on image synthesis. 2021.Dhariwal P., Nichol A. Diffusion models beat GANs on image synthesis. 2021.Xia Y., Wang W., Wang K. ECG signal generation based on conditional generative models // Biomedical Signal Processing and Control. 2023. Vol. 82. P. 104587.Xia Y., Wang W., Wang K. ECG signal generation based on conditional generative models // Biomedical Signal Processing and Control. 2023. Vol. 82. P. 104587.El-Kaddoury M., Mahmoudi A., Himmi M.M. Deep generative models for image generation: A practical comparison between variational autoencoders and generative adversarial networks // Mobile, Secure, and Programmable Networking. Cham: Springer, 2019. P. 1–8.El-Kaddoury M., Mahmoudi A., Himmi M.M. Deep generative models for image generation: A practical comparison between variational autoencoders and generative adversarial networks // Mobile, Secure, and Programmable Networking. Cham: Springer, 2019. P. 1–8.Kuznetsov V., Moskalenko V., Gribanov D., Zolotykh N. Interpretable feature generation in ECG using a variational autoencoder // Frontiers in Genetics. 2021. Vol. 12. P. 638191.Kuznetsov V., Moskalenko V., Gribanov D., Zolotykh N. Interpretable feature generation in ECG using a variational autoencoder // Frontiers in Genetics. 2021. Vol. 12. P. 638191.Sang Y., Beetz M., Grau V. Generation of 12-lead electrocardiogram with subject-specific, image-derived characteristics using a conditional variational autoencoder // 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI). 2022. P. 1–5.Sang Y., Beetz M., Grau V. Generation of 12-lead electrocardiogram with subject-specific, image-derived characteristics using a conditional variational autoencoder // 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI). 2022. P. 1–5.Beetz M., Banerjee A., Sang Y., Grau V. Combined generation of electrocardiogram and cardiac anatomy models using multi-modal variational autoencoders // 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI). 2022. P. 1–4.Beetz M., Banerjee A., Sang Y., Grau V. Combined generation of electrocardiogram and cardiac anatomy models using multi-modal variational autoencoders // 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI). 2022. P. 1–4.Salimans T., Karpathy A., Chen X., Kingma D.P. PixelCNN++: Improving the PixelCNN with discretized logistic mixture likelihood and other modifications. 2017.Salimans T., Karpathy A., Chen X., Kingma D.P. PixelCNN++: Improving the PixelCNN with discretized logistic mixture likelihood and other modifications. 2017.Deng L. The MNIST database of handwritten digit images for machine learning research [Best of the Web] // IEEE Signal Processing Magazine. 2012. Vol. 29, No. 6. P. 141–142.Deng L. The MNIST database of handwritten digit images for machine learning research [Best of the Web] // IEEE Signal Processing Magazine. 2012. Vol. 29, No. 6. P. 141–142.Krizhevsky A. Learning multiple layers of features from tiny images. 2009.Krizhevsky A. Learning multiple layers of features from tiny images. 2009.Larsen A.B.L., Sønderby S.K., Larochelle H., Winther O. Autoencoding beyond pixels using a learned similarity metric. 2016.Larsen A.B.L., Sønderby S.K., Larochelle H., Winther O. Autoencoding beyond pixels using a learned similarity metric. 2016.Karras T., Aila T., Laine S., Lehtinen J. Progressive growing of GANs for improved quality, stability, and variation. 2018.Karras T., Aila T., Laine S., Lehtinen J. Progressive growing of GANs for improved quality, stability, and variation. 2018.Malmivuo J., Plonsey R. Bioelectromagnetism. 15. 12-Lead ECG System. 1975. P. 277–289.Malmivuo J., Plonsey R. Bioelectromagnetism. 15. 12-Lead ECG System. 1975. P. 277–289.Griffin D., Lim J. Signal estimation from modified short-time Fourier transform // IEEE Transactions on Acoustics, Speech, and Signal Processing. 1984. Vol. 32, No. 2. P. 236–243.Griffin D., Lim J. Signal estimation from modified short-time Fourier transform // IEEE Transactions on Acoustics, Speech, and Signal Processing. 1984. Vol. 32, No. 2. P. 236–243.Strodthoff N., Wagner P., Schaeffter T., Samek W. Deep learning for ECG analysis: Benchmarks and insights from PTB-XL // IEEE Journal of Biomedical and Health Informatics. 2021. Vol. 25, No. 5. P. 1519–1528.Strodthoff N., Wagner P., Schaeffter T., Samek W. Deep learning for ECG analysis: Benchmarks and insights from PTB-XL // IEEE Journal of Biomedical and Health Informatics. 2021. Vol. 25, No. 5. P. 1519–1528.Donahue C., McAuley J., Puckette M. Adversarial audio synthesis. 2019.Donahue C., McAuley J., Puckette M. Adversarial audio synthesis. 2019.Wu C.-J., Raghavendra R., Gupta U., et al. Sustainable AI: Environmental implications, challenges and opportunities // ArXiv. 2021. abs/2111.00364.Wu C.-J., Raghavendra R., Gupta U., et al. Sustainable AI: Environmental implications, challenges and opportunities // ArXiv. 2021. abs/2111.00364.Wagner P., Strodthoff N., Bousseljot R.-D., et al. PTB-XL, a large publicly available electrocardiography dataset // Scientific Data. 2020. Vol. 7. P. 154.Wagner P., Strodthoff N., Bousseljot R.-D., et al. PTB-XL, a large publicly available electrocardiography dataset // Scientific Data. 2020. Vol. 7. P. 154.Goldberger A., Amaral L., Glass L., et al. PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals // Circulation. 2000. Vol. 101. P. e215–e220.Goldberger A., Amaral L., Glass L., et al. PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals // Circulation. 2000. Vol. 101. P. e215–e220.Zheng J., Chu H., Struppa D., et al. Optimal multi-stage arrhythmia classification approach // Scientific Reports. 2020. Vol. 10.Zheng J., Chu H., Struppa D., et al. Optimal multi-stage arrhythmia classification approach // Scientific Reports. 2020. Vol. 10.El-Sappagh S., Franda F., Ali F., Kwak K.-S. SNOMED CT standard ontology based on the ontology for general medical science // BMC Medical Informatics and Decision Making. 2018. Vol. 18, No. 1. P. 76.El-Sappagh S., Franda F., Ali F., Kwak K.-S. SNOMED CT standard ontology based on the ontology for general medical science // BMC Medical Informatics and Decision Making. 2018. Vol. 18, No. 1. P. 76.Berkaya S.K., Uysal A.K., Gunal E.S., et al. A survey on ECG analysis // Biomedical Signal Processing and Control. 2018. Vol. 43. P. 216–235.Berkaya S.K., Uysal A.K., Gunal E.S., et al. A survey on ECG analysis // Biomedical Signal Processing and Control. 2018. Vol. 43. P. 216–235.Zhang M., Wang Y., Luo T. Federated learning for arrhythmia detection of non-IID ECG // 2020 IEEE 6th International Conference on Computer and Communications (ICCC). 2020. P. 1176–1180.Zhang M., Wang Y., Luo T. Federated learning for arrhythmia detection of non-IID ECG // 2020 IEEE 6th International Conference on Computer and Communications (ICCC). 2020. P. 1176–1180.

The authors declare that there are no conflicts of interest present.