Cortical parcellation optimized for magnetoencephalography with a clustering technique.

Area of Science:

• Neuroscience
• Biophysics
• Computational Neuroscience

Background:

• Current magnetoencephalography (MEG) analysis methods often suffer from signal cancellation and crosstalk when combining source activities within anatomical brain parcels.
• This limits the accuracy of functional connectivity estimation and source-space visualization.

Purpose of the Study:

• To develop a novel method for parceling the cortex in MEG data that minimizes inter-parcel crosstalk and signal cancellation.
• To create a compact, anatomically-informed source-level representation of MEG data.

Main Methods:

• Utilized unsupervised clustering of MEG leadfields to define parcels based on source activity representation.
• Incorporated spatial distances between sources to ensure parcel contiguity and minimize crosstalk, using k-nearest neighbor memberships.
• Optimized parceling by assigning specific weights to spatial distances, resulting in 60-120 parcels.

Main Results:

• The proposed method successfully divides the cortex into parcels that can be represented by single dipolar sources.
• Demonstrated minimization of inter-parcel crosstalk through optimized spatial weighting in the clustering process.
• Achieved a compact source-level representation of MEG data with dimensionality comparable to sensor-level data.

Conclusions:

• The new parceling approach, implemented in the 'megicparc' Python package, offers a more faithful source-level representation of MEG data.
• This method is expected to significantly enhance the visualization of MEG features and improve the accuracy of functional connectivity estimation.
• The approach provides a robust alternative to traditional parceling strategies in MEG analysis.