Inverse modeling on decomposed electroencephalographic data: a way forward?

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Traditional inverse modeling of electroencephalogram (EEG) signals struggles with multiple, deep, or noisy sources.
• Existing methods lack sensitivity, stability, and precision in dipole estimation.
• Need for advanced signal decomposition techniques to improve accuracy in non-invasive brain source localization.

Purpose of the Study:

• To validate and compare different dipole estimation techniques for electroencephalogram (EEG) inverse modeling.
• To identify the optimal combination of analysis principles for accurate source localization.
• To assess the performance of signal decomposition methods prior to inverse modeling.

Main Methods:

• Simulated EEG data with known superficial, deep, and simultaneous sources.
• Recorded somatosensory-evoked potentials from human subjects.
• Comparison of Independent Component Analysis (ICA), Second-Order Blind Identification (SOBI), and Multichannel Matching Pursuit (MMP) for signal decomposition, followed by DIPFIT inverse modeling.

Main Results:

• MMP successfully separated all simulated source types (superficial, deep, simultaneous), unlike ICA and SOBI which only identified superficial sources.
• Inverse modeling using MMP components achieved significantly higher accuracy (99.2% dipole localization) compared to ICA (35%) and SOBI (39.6%).
• MMP-based dipole modeling on real evoked potentials provided results more consistent with physiological knowledge.

Conclusions:

• Multichannel Matching Pursuit (MMP) is a superior method for decomposing EEG signals before inverse modeling.
• Employing MMP prior to inverse modeling offers a reliable approach for non-invasive estimation of cortical activation.
• This optimized approach enhances the sensitivity, stability, and precision of dipole source localization.