Non-invasive laminar inference with MEG: Comparison of methods and source inversion algorithms

Area of Science:

• Neuroscience
• Biophysics
• Computational Neuroscience

Background:

• Magnetoencephalography (MEG) offers direct measurement of neuronal activity, with anatomical resolution limited by data quality and modeling.
• Previous simulations suggested laminar-specific MEG signal discrimination is possible with precise sensor-model knowledge.
• Prior work primarily used a single inversion scheme (multiple sparse priors) and a global metric (free energy).

Purpose of the Study:

• To assess the robustness of laminar discrimination in MEG using diverse source inversion algorithms and fit metrics.
• To validate the ability of MEG to differentiate signals originating from deep versus superficial cortical layers.
• To identify factors influencing the accuracy of laminar discrimination in MEG data.

Main Methods:

• Employed multiple source inversion algorithms, including those with sparsity constraints.
• Utilized various local and global, parametric and non-parametric fit metrics (e.g., t-statistics, cross-validation, free energy).
• Investigated the impact of patch size, cortical features, and lead field strength on discrimination accuracy.

Main Results:

• Only source inversion algorithms incorporating sparsity constraints successfully achieved laminar discrimination.
• Local t-statistics, global cross-validation, and free energy metrics provided consistent and reliable measures of fit.
• Discrimination accuracy was sensitive to estimates of patch size, cortical surface characteristics, and lead field strength.

Conclusions:

• Demonstrated the feasibility of determining the laminar origin of MEG signals.
• This capability enables direct empirical testing of cognitive theories involving laminar- and frequency-specific neural mechanisms.
• Recent advancements in high-precision MEG, including subject-specific head-casts, enhance data quality for anatomically precise recordings.