Tensor decomposition of human narrowband oscillatory brain activity in frequency, space and time
Investigators: Roman Rosipal, Zuzana Rošťáková
The detection and changes of oscillatory rhythms in the EEG signal are important for understanding the processes taking place in the human brain during experiments aimed at modulating brain activity, as well as for the proper functioning of the brain-computer interface. In , we created a set of simulated EEG data copying the basic properties of a real EEG signal with the aim of rigorously demonstrating the advantages of tensor decomposition – PARAFAC and Tucker’s model – compared to standard methods in the time-space domain. We also showed that the limitations of the decomposition in the spatial and frequency domain (unimodality and bimodality) lead to a better physiological interpretation of the detected latent components representing oscillatory rhythms. Simulated EEG data helped us in  determine the suitability, or inappropriateness of applying the existing methods of determining the number of latent components F in the PARAFAC model to EEG data. We also confirmed the relevance of our proposed heuristic approach to determining F, which achieved comparable results with the best method based on Bayesian statistics. The created simulated EEG data can also serve in the future as a standardized set with known properties for further rigorous comparison of different EEG signal analysis approaches.
Fig.: Left: Human brain modeled using 2004 dipoles in 3D space while simulating 1 minute of EEG data. Broadband brain activity was generated using fractal Brownian motion. Seven dipoles were selected as sources of oscillatory activity at 5 Hz (blue), 8 Hz (green), 11 Hz (yellow), and 14 Hz (red). Right: Graphic diagram of the PARAFAC decomposition of the simulated EEG signal in the time-space-frequency domain. The PARAFAC model detected four latent components representing oscillatory activity at 5 Hz in the frontal region (component 1), 8 Hz in the central region (component 2), 11 Hz in the occipital region (component 3), and 14 Hz in the central region (component 4).
Related projects: Solved within projects APVV-16-0202, APVV-21-0105 and VEGA 2/0023/22.
 ROSIPAL, Roman – ROŠŤÁKOVÁ, Zuzana – TREJO, L.J. Tensor decomposition of human narrowband oscillatory brain activity in frequency, space and time. In Biological Psychology, 2022, vol. 169, art. no. 108287. (2021: 3.111 – IF, Q2 – JCR, 1.023 – SJR, Q1 – SJR). ISSN 0301-0511. Available at: https://doi.org/10.1016/j.biopsycho.2022.108287 Type: ADCA
 ROŠŤÁKOVÁ, Zuzana – ROSIPAL, Roman. Determination of the number of components in the PARAFAC model with a nonnegative tensor structure: A simulated EEG data study. In Neural Computing & Applications, 2022, vol. 34, p. 14793-14805. (2021: 5.102 – IF, Q2 – JCR, 1.072 – SJR, Q1 – SJR). ISSN 0941-0643. Available at: https://doi.org/10.100/s00521-022-07318-x Type: ADCA