Unsupervised Segmentation of Head Tissues from Multi-modal MR Images for EEG Source Localization

Area of Science:

• Medical Imaging
• Biophysics
• Computational Neuroscience

Background:

• Accurate patient-specific head models are crucial for electroencephalography (EEG) source localization.
• Current segmentation methods can be limited in accuracy and robustness.

Purpose of the Study:

• To present and evaluate an automatic unsupervised segmentation method, HSA-BAMS, for constructing patient-specific head conductivity models.
• To assess the segmentation and source localization accuracy of HSA-BAMS.

Main Methods:

• The hierarchical segmentation approach (HSA) combined with Bayesian-based adaptive mean shift (BAMS) was used for tissue segmentation from multi-modal MR head images.
• Direct evaluation used Dice index and Hausdorff distance on synthetic and real data.
• Indirect evaluation assessed source localization accuracy using synthetic EEG data.

Main Results:

• HSA-BAMS demonstrated superior segmentation accuracy compared to Brain Extraction Tool (BET)-FMRIB's Automated Segmentation Tool (FAST) and HSA variants.
• The method showed robustness to noise and bias fields.
• HSA-BAMS resulted in more accurate EEG source localization than the reference method.

Conclusions:

• HSA-BAMS is an effective method for segmenting head tissues for EEG source localization.
• The approach offers improved accuracy and robustness over existing methods.
• HSA-BAMS shows potential as a replacement for manual segmentation in EEG applications.