EEG-based digital twin fMRI

Currently, electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) are the two most commonly used methods for noninvasive recording of brain activity. Based on very different principles, these two noninvasive functional mapping technologies have complementary properties of spatial and temporal resolution.  However, as recent studies suggest, the relationship between the BOLD signal, which registers a localized functional hemodynamic response, and multichannel EEG measurements of brain electrical activity is amenable to modeling using machine learning techniques.

Fig.1. Basal ganglia modeling of different brain regions. (Source: Volkmann, J., Daniels, C., & Witt, K. (2010). Neuropsychiatric effects of subthalamic neurostimulation in Parkinson's disease. Nature Reviews Neurology, 6(9), 487-498.)

This project aims to create a new technology for mapping the function of deep brain structures based on a compact and accessible electroencephalograph (EEG) combined with machine learning algorithms and neural network models. So far, a model has been developed to convert multichannel electrical activity into a functional hemodynamic signal (BOLD-signal) of the activity of several subcortical structures, including the thalamic, amygdala, pale globe, caudate nucleus, adjoining nucleus and the shell. Details of the model architecture are given below.

Figure 2 shows the results with the topographies of the key sources found as a result of model training.

Fig. 2. Topographies of found sources using the developed model. Taking into account the rapid development of fMRI neurofeedback technology, as well as the positive effect of feedback delay reduction we demonstrated earlier (Belinskaya et. al., 2021), we can conclude that the use of the EEG→BOLD solution will allow to control the hemodynamics of individual cortical or subcortical areas ahead of time.
 
In the future it will allow to fully implement the principle of PID (Proportional Integral & Derivative) control when developing self-regulation skills, and will find application in forming a priori spatial distributions to improve the quality of EEG inverse problem solving. 

Project article:
1. Kovalev, A., Mikheev, I., & Ossadtchi, A. (2022). fMRI from EEG is only Deep Learning away: the use of interpretable DL to unravel EEG-fMRI relationships. arXiv preprint arXiv:2211.02024. https://arxiv.org/abs/2211.02024