DIFFUSION NEURAL NETWORKS IN THE PROBLEMS OF CLASSIFICATION OF MOTOR IMAGES BY ELECTROENCEPHALOGRAM SIGNALS

Authors

  • Andrey A. Tolstykh OOO «RTK»
  • Andrey N. Golubinsky A. A. Kharkevich Institute of Information Transmission Problems of the Russian Academy of Sciences
  • Dmitry V. Zhuravlev Voronezh State Technical University image/svg+xml
  • Andrey A. Reznichenko Voronezh State Technical University image/svg+xml

DOI:

https://doi.org/10.17308/sait/1995-5499/2026/2/121-143

Keywords:

convolutional neural network, diffusion neural network, TimeGrad, ScoreGrad, D-Va, dataset, classifier, model

Abstract

The classification of electroencephalogram signal patterns in brain-computer interfaces is an urgent scientific and technical task. An analysis of previous research shows that most of the work in this area was carried out using convolutional neural networks. The use of diffusion neural networks in the classification of electroencephalogram signals is not found in the literature, although this is a promising and poorly studied area. The aim of the work was to study the degree of effectiveness of models for recognizing real and imagеry motor skills based on electroencephalogram signals and to evaluate the possibility of using diffusion artificial neural networks as neuroclassifiers. A classification of known diffusion neural networks has been proposed. Three models of building diffusion neural networks (TimeGrad, ScoreGrad, and D-Va) with the best performance were selected for the study. As part of the computational experiment, the selected models were trained and validated both on third-party datasets (open source materials were used) and on own sets of different types: dictor-dependent and dictor-independent. The influence of real motor skills and imaginary, as well as frequency filtering on classification results is investigated. The evaluation of the quality of the classifiers, based on the studied models of diffusion networks, was carried out using such evaluation criteria as the classical classification metric «Accuracy» and the ROC AUC method, using the metric variance estimation where possible. In the course of the work, a single classifier architecture was developed for the three models. The analysis of the results of the experiment allows us to speak about the statistical indistinguishability of the results of the classification of real and imagery motor skills based on the signals of electroencephalograms obtained using the TimeGrad, ScoreGrad and D-Va models. The highest classification accuracy of more than 60 % (for three classes) was obtained when classifying imaginary movements. Empirical evidence was also obtained for the effectiveness of using diffusion neural networks in working with dictor-independent datasets, in comparison with a classifier based on the standard convolutional neural network apparatus. Based on the results of the computational experiment, a number of conclusions and recommendations were made that make it possible to use the apparatus of diffusion neural networks in the tasks of classifying operator motor skills based on electroencephalogram signals.

Author Biographies

  • Andrey A. Tolstykh, OOO «RTK»

    PhD. (Eng), Software engineer

  • Andrey N. Golubinsky, A. A. Kharkevich Institute of Information Transmission Problems of the Russian Academy of Sciences

    PhD., Dr. Sci. (Eng.), Associate Professor, Acting Deputy Director for Research

  • Dmitry V. Zhuravlev, Voronezh State Technical University

    PhD. (Eng), Associate Professor, Acting Head of the Department of Radio electronic Devices and Systems

  • Andrey A. Reznichenko, Voronezh State Technical University

    is a post-graduate student of the Department of Radioelectronic Devices and Systems

References

1. Попов Е. Ю. Детектирование событий движения руки в сигналах ЭЭГ головного мозга / Е. Ю. Попов, С. А. Фоменков // Известия Волгоградского государственного технического университета. – 2015. – № 4 (32). – C. 131–140.

2. Попов Е. Ю. Детектирование событий движения руки в сигналах ЭЭГ головного мозга с помощью сверточных нейронных сетей / Е. Ю. Попов, С. А. Фоменков // Известия Волгоградского государственного технического университета. – 2016. – № 1 (33). – C. 62–67.

3. Popov E. Classification of hand motions in EEG signals using recurrent neural networks / E. Popov, S. Fomenkov // 2016 2nd International Conference on Industrial Engineering, Applications and Manufacturing (ICIEAM), Chelyabinsk, Russia. – 2016. – No 47. – P. 1–4. DOI: 10.1109/icieam.2016.7911620.

4. EEG Classification of Motor Imagery Using a Novel Deep Learning Framework / M. Dai [et al.] // Sensors. – 2019. – Vol. 19, No 3. – P. 551. DOI:10.3390/s19030551.

5. Гудфеллоу Я. Глубокое обучение / Я. Гудфеллоу, И. Бенджио И., А. Курвилль. – Москва: «ДМК Пресс», 2017. – 652 c.

6. Kingma D. P. An Introduction to Variational Autoencoders / D.P. Kingma, W. Max. – Boston Delft: Now Publisher, 2019. – 102 p.

7. Almutairi H. Improving Sleep Health with Deep Learning: Automated Classidication of Sleep Stages and Detection of Sleep Disorders : PhD thesis, University of Western Australia, 2024.

8. Tosato G. EEG Synthetic Data Generation Using Probabilistic Diffusion Models / G. Tosato, C.-M. Dalbagno, F. Fumagalli // Synapsium 2023 conference. – 2023. – 16 p. https://doi. org/10.48550/arXiv.2303.06068

9. Ho J. Denoising Diffusion Probabilistic Models / J. Ho, A. Jain, P. Abbeel // NIPS’20: 34th International Conference on Neural Information Processing Systems, Vancouver, Canada. – 2020. – No 574. – P. 6840–6852. DOI:10.1145/3583780.3614844.

10. Dhariwal P. Diffusion Models Beat GANs on Image Synthesis / P. Dhariwal, A. Nichol // NIPS ‘21: Proceedings of the 35th International Conference on Neural Information Processing Systems, NY, United States. – 2021. – No 672. – P. 8780–8794.

11. GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models / A. Nichol [et al.] // International Conference on Machine Learning, ICML 2022, Proceedings of Machine Learning Research, Baltimore, Maryland, USA. – 2022. – Vol. 162. – P. 16784–16804.

12. Takagi Y. High-resolution image reconstruction with latent diffusion models from human brain activity / Y. Takagi, S. Nishimoto // IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, BC, Canada. – 2023. – P. 14453-14463.

13. Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding / C. Saharia [et al.] // NIPS’22: 36th International Conference on Neural Information Processing Systems, New Orleans, LA, USA. – 2022. – No 2643. – P. 36479–36494.

14. Scheider F. Archi Sound: audio generation with diffusion : Master’s thesis, ETH Zurich, 2023.

15. Audio LDM: Text-to-Audio Generation with Latent Diffusion Models / Liu H. [et al.] // ICML’23: Proceedings of the 40th International Conference on Machine Learning, Honolulu, Hawaii, USA. – 2023. – No 886. – P. 21450–21474.

16. Autoregressive Denoising Diffusion Models for Multivariate Probabilistic Time Series Forecasting / K. Rasul [et al.] // Proceedings of the 38th International Conference on Machine Learning (ICML 2021), Virtual. – 2021. – P. 8857–8868.

17. ScoreGrad: Multivariate Probabilistic Time Series Forecasting with Continuous Energy-based Generative Models / Yan T. [et al.] // CoRR.Arxiv. – 2021. – 12 p. https://doi. org/10.48550/arXiv.2106.10121

18. Li Y. Generative Time Series Forecasting with Diffusion, Denoise, and Disentanglement / Y. Li, X. Lu, Y. Wang, D. Dou // Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems (NeurIPS 2022), New Orleans, LA, USA. – 2022. – P. 23009–23022.

19. Modeling Temporal Data as Continuous Functions with Stochastic Process Diffusion / M. Biloš [et al.] // ICML’23: Proceedings of the 40th International Conference on Machine Learning, Honolulu, Hawaii, USA. – 2023. – No 104. – P. 2452–2470.

20. Shen L. Non-autoregressive Conditional Diffusion Models for Time Series Prediction / L. Shen, J. Kwok // ICML’23: Proceedings of the 40th International Conference on Machine Learning, Honolulu, Hawaii, USA. – 2023. – No 1284. – P. 31016–31029.

21. DiffSTG: Probabilistic Spatio-Temporal Graph Forecasting with Denoising Diffusion Models / H. Wen [et al.] // Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems, Hamburg, Germany. – 2023. – No 60. – P. 1–12.

22. Graph Convolution Recurrent Denoising Diffusion Model for Multivariate Probabilistic Temporal Forecasting / R. Li [et al.] // Advanced Data Mining and Applications, ADMA 2023, Lecture Notes in Computer Science, Springer, Cham. – 2023. – Vol. 14176. – P. 661–676. DOI:10.1007/978-3-031-46661-8_44.

23. CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation / Y. Tashiro [et al.] // NIPS ‘21: Proceedings of the 35th International Conference on Neural Information Processing Systems, Virtual. – 2021. – No 1900. – P. 24804–24816.

24. Diffusion models for time-series applications: a survey / L. Lin [et al.] // Frontiers of Information Technology & Electronic Engineering. – 2024. – No 25. – P. 19–41.

25. Diffusion Variational Autoencoder for Tackling Stochasticity in Multi-Step Regression Stock Price Prediction / K.-J.-L. Koa [et al.] // CIKM ‘23: Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, Birmingham United Kingdom. – 2023. – P. 1087–1096. DOI:10.1145/3583780.3614844.

26. BCI2000: a general-purpose brain-computer interface (BCI) system / G. Schalk [et al.] // IEEE Trans Biomed Eng, USA. – 2004. – Vol. 51, No 6. – P. 1034–1043. DOI:10.1109/TBME.2004.827072.

27. Журавлёв Д. В. Аппаратура для электроэнцефалографических исследований: монография / Д. В. Журавлёв. – Воронеж : ФГБОУ ВО «ВГТУ», 2021. – 259 c.

28. Kingma D. P. Auto-Encoding Variational Bayes / D. P. Kingma, M. Welling // Arxiv. – 2022. – Режим доступа: https://arxiv.org/ abs/1312.6114. – (Дата обращения: 15.10.2024).

29. Attention is All you Need / A. Vaswani [et al.] // Advances in Neural Information Processing Systems 30 (NeurIPS 2017), Long Beach, California, USA. – 2017. – P. 1–11.

30. WaveNet: A Generative Model for Raw Audio / A. Oord van den [et al.] // The 9th ISCA Speech Synthesis Workshop, SSW 2016, Sunnyvale, CA, USA. – 2016. – No 125. – P. 1–15. DOI:10.48550/arXiv. 1609.03499.

31. DiffWave: A Versatile Diffusion Model for Audio Synthesis / Z. Kong [et al.] // ICLR 2021 Conference, Virtual. – 2021. – No 1087. – P. 1–17. DOI:10.48550/arXiv. 2009.09761.

32. Kingma D. P. Adam: A Method for Stochastic Optimization / D. P. Kingma, J. Ba // International Conference on Learning Representations, ICLR 2014, San Diego, USA. – 2014. – P. 1–15. DOI:10.48550/arXiv. 1412.6980.

33. Vahdat A. NVAE: A Deep Hierarchical Variational Autoencoder / A. Vahdat, J. Kautz // Advances in Neural Information Processing Systems 33 (NeurIPS 2020), Virtual. – 2020. – P. 19667– 19679. DOI: 10.48550 /arXiv.2007.03898.

34. How transferable are features in deep neural networks? / J. Yosinski [et al.] // Advances in Neural Information Processing Systems 27 (NeurIPS 2014), Montreal, Canada. – 2014. – P. 1–9.

35. He K. Deep residual learning for image recognition / K. He, X. Zhang, S. Ren, J. Sun // 2016 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, Nevada. – 2016. – P. 770–778.

36. Ho J. Denoising diffusion probabilistic models / J. Ho, A. Jain, P. Abbeel // Advances in Neural Information Processing Systems 33 (NeurIPS 2020), Virtual. – 2020. – P. 1–12.

37. Хайкин Л. А. Нейронные сети. Полный курс / Л. А. Хайкин. – 2-е изд. – Москва : Издательский дом «Вильямс», 2006. – 1104 c.

38. Куркин С. А. Использование искусственных нейронных сетей для классификации электрической активности головного мозга в процессе воображения движений у нетренированных испытуемых / С. А. Куркин, Е. Н. Пицик, А. Е. Храмов // Информационно-управляющие системы. – 2020. – № 6. – C. 77–84.

Published

2026-06-30

Issue

Section

Intelligent Information Systems, Data Analysis and Machine Learning

How to Cite

Tolstykh, A. A., Golubinsky, A. N., Zhuravlev, D. V., & Reznichenko, A. A. (2026). DIFFUSION NEURAL NETWORKS IN THE PROBLEMS OF CLASSIFICATION OF MOTOR IMAGES BY ELECTROENCEPHALOGRAM SIGNALS. Proceedings of Voronezh State University. Series: Systems Analysis and Information Technologies, 2, 121-143. https://doi.org/10.17308/sait/1995-5499/2026/2/121-143