A New Multi-Resolution Approach to EEG Brain Modeling Using Local-Global Graphs and Stochastic Petri-Nets.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Brain activity modeling increasingly involves fusing multiple data modalities.
• Combining electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) offers enhanced localization and interpretation of brain activity.
• Existing methods face challenges in efficient and compatible signal representation for multimodal fusion.

Purpose of the Study:

• To develop a novel methodology for abstract visual representation of EEG signals.
• To enable efficient modeling and fusion of EEG data with other neuroimaging modalities.
• To demonstrate the utility of the Bar-LG method using an epileptic seizure example.

Main Methods:

• EEG signals are treated as 2D string images for feature extraction.
• A new methodology, Bar-LG, discretizes EEG signals into selected minima/maxima tokens.
• Formal context-free language represents extracted tokens for active brain regions.
• Generalized Stochastic Petri-Nets (GSPN) model functional associations of EEG signal regions.

Main Results:

• The Bar-LG method provides a reduced, token-based representation of EEG signals.
• This abstract representation is suitable for EEG signal modeling.
• The methodology facilitates future fusion with modalities like fMRI.
• An illustrative example of epileptic seizure analysis demonstrates the method's capabilities.

Conclusions:

• The Bar-LG methodology offers a novel approach to abstractly represent EEG signals as visual tokens.
• This method enhances EEG data analysis and opens possibilities for multimodal brain imaging fusion.
• The approach shows promise for analyzing complex brain activities, such as epileptic seizures.