EEG dipole source localization with information criteria for multiple particle filters

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Electroencephalography (EEG) provides high temporal resolution for neural activity but requires accurate source localization for interpretation.
• Current dipole source localization (DSL) methods often rely on pre-determined dipole numbers, limiting accuracy.
• Conventional information criteria like AIC and BIC may not be suitable for nonparametric particle filtering in EEG analysis.

Purpose of the Study:

• To develop and validate a novel dipole source localization (DSL) method for electroencephalography (EEG).
• To introduce an adaptive method for estimating the number of neural dipoles using a tailored information criterion.
• To improve the accuracy and reliability of interpreting EEG signals for clinical and functional applications.

Main Methods:

• Proposed a DSL method employing multiple parallel particle filters, each assuming a fixed dipole number.
• Introduced a novel information criterion to select the most predictive particle filter at each time step.
• Validated the method using artificial datasets and real human EEG data from an auditory short-term memory task.

Main Results:

• The proposed information criterion demonstrated superior performance compared to Akaike's Information Criterion (AIC) and Bayesian Information Criterion (BIC) on artificial datasets.
• Analysis of human EEG data revealed accurate localization of alpha-band dipoles in auditory areas during a memory task.
• The method successfully identified neural activity consistent with known physiological findings.

Conclusions:

• The novel information criterion effectively estimates dipole numbers in EEG source localization.
• The proposed DSL method enhances the accuracy of interpreting complex neural activity.
• This approach offers a promising tool for both research and clinical applications of EEG.