Finer-Grained Dynamic Functional Graph Structure Learning for EEG Sequence Modeling.

Area of Science:

• Neuroscience
• Computational Neuroscience

Background:

• Electroencephalography (EEG) is crucial for understanding brain dynamics, but modeling its dynamic functional connectivity is challenging.
• Existing methods struggle with spatio-temporal specificity and fine-grained dynamic changes in brain interactions.
• Current state space models lack functional connectivity modeling for EEG data.

Purpose of the Study:

• To propose the Dynamic Functional Graph Structure Learning (DFGSL) framework for capturing dynamic functional connectivity in EEG signals at a finer-grained level.
• To address limitations of existing methods in reflecting spatio-temporal specificity and rapid dynamic reorganization of brain functions.

Main Methods:

• DFGSL constructs dynamic similarity probability maps to reveal information exchange between brain regions.
• A selective state space model simulates the dynamic evolution of functional connectivity.
• Dynamic similarity probabilities between states yield compact representations of brain state evolution.

Main Results:

• The DFGSL framework demonstrates superior performance across three benchmark EEG datasets.
• It consistently outperforms state-of-the-art methods in functional modeling capabilities.
• The method effectively captures fine-grained dynamic functional connectivity.

Conclusions:

• DFGSL offers a robust approach for modeling dynamic functional connectivity in EEG.
• The framework provides insights into neural mechanisms by revealing dynamic brain state evolution.
• This method advances the analysis of complex brain dynamics using neuroimaging data.