Influence of personalized human head modeling and resolution on EEG source localization for rapid brain mapping

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• Electroencephalography (EEG) source imaging is crucial for understanding brain activity.
• Accurate source localization requires detailed head models, often involving complex segmentation.
• Computational cost and model resolution present trade-offs in current EEG source imaging techniques.

Purpose of the Study:

• To evaluate segmentation-free EEG source imaging using personalized head models.
• To assess the impact of model resolution, anatomical fidelity, and computational cost on localization accuracy.
• To explore sparse inverse analysis for rapid brain mapping.

Main Methods:

• Utilized a scalar-potential finite-difference method and orthogonal matching pursuit for inverse source localization.
• Employed machine learning models (CondNet, CondNet-TART) for direct estimation of tissue conductivity from MRI, creating continuous conductivity distributions.
• Compared segmentation-free models against a finite-element method (FEM) across various resolutions.

Main Results:

• Segmentation-free models achieved localization errors under 23 mm with stable accuracy at coarser resolutions.
• Computational time was reduced by 87.9% compared to FEM at 2.0-mm resolution.
• CondNet-TART demonstrated superior localization accuracy and stability among segmentation-free models.

Conclusions:

• The proposed segmentation-free framework offers efficient and accurate EEG source localization with significantly lower computational demands.
• This approach is suitable for time-sensitive applications like presurgical functional mapping and real-time neuroimaging.