Neural stochastic differential equations network as uncertainty quantification method for EEG source localization.

Area of Science:

• Neuroscience
• Biophysics
• Computational Biology

Background:

• Electroencephalography (EEG) source localization is hindered by uncertain conductivity values in volume conductor models (VCMs).
• These conductivity variations across individuals significantly impact EEG forward and inverse solutions, increasing localization errors and misdiagnosis risks.
• Uncertainty quantification (UQ) techniques offer a promising avenue for calibrating conductivity values and reducing localization errors.

Purpose of the Study:

• To introduce and evaluate the Neural Stochastic Differential equations Network (SDE-Net) for minimizing conductivity uncertainties in VCMs for EEG source localization.
• To compare the performance of SDE-Net against traditional Bayesian UQ methods in improving EEG inverse problem solutions.

Main Methods:

• The study employed SDE-Net, integrating dynamical systems and deep learning with a Wiener process to address conductivity uncertainties.
• This approach aimed to minimize uncertainties within the VCM and enhance the accuracy of the EEG inverse problem.

Main Results:

• SDE-Net demonstrated a lower localization error rate in the EEG inverse problem compared to conventional Bayesian techniques.
• The findings indicate SDE-Net's efficacy in improving the precision of EEG source localization.

Conclusions:

• SDE-Net presents a novel and effective deep learning-based UQ technique for EEG source localization.
• Future research should explore advanced stochastic dynamical systems for addressing residual uncertainties in EEG analysis.