Inferring effective connectivity in the brain from EEG time series using dynamic Bayesian networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging

Background:

• Effective brain connectivity, the influence between neuronal populations, is crucial for understanding brain organization.
• Altered effective connectivity patterns are observed in neurological and psychopathological diseases, necessitating robust modeling techniques.
• Non-invasive neuroimaging data, such as electroencephalogram (EEG), offers a valuable window into brain activity.

Purpose of the Study:

• To propose and evaluate dynamic Bayesian networks (DBN) for learning effective brain connectivity from EEG data.
• To investigate differences in effective connectivity between healthy subjects and individuals with schizophrenia.
• To identify statistically significant neural interactions using a robust statistical method.

Main Methods:

• Utilized dynamic Bayesian networks (DBN), a probabilistic graphical model, to represent EEG time series.
• Employed a first-order Markov chain within the DBN framework to model temporal dependencies in multi-electrode EEG data.
• Applied the Fourier bootstrapping technique to determine the statistical significance of interactions between different electrodes.

Main Results:

• Successfully applied DBNs to model effective brain connectivity from EEG data.
• Identified distinct patterns of effective connectivity in healthy versus schizophrenic subjects.
• The Fourier bootstrapping technique effectively pinpointed statistically significant neural interactions.

Conclusions:

• Dynamic Bayesian networks provide a powerful framework for modeling effective brain connectivity using EEG.
• This approach can reveal neural connectivity alterations associated with psychiatric conditions like schizophrenia.
• The findings highlight the potential of DBNs for advancing our understanding of brain function and dysfunction.