Characterization of network structure in stereoEEG data using consensus-based partial coherence.

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Coherence measures frequency-resolved functional connectivity but is confounded by shared inputs.
• Partial coherence computation involves matrix inversion, posing challenges for large-scale intracranial recordings.
• Existing pseudoinverse methods for partial coherence yield high false positives with numerous channels.

Purpose of the Study:

• To develop a novel method for time-resolved partial coherence analysis in large-scale neural recordings.
• To overcome limitations of pseudoinverse methods in identifying true functional connectivity.
• To improve the accuracy and interpretability of brain network dynamics.

Main Methods:

• Developed a consensus partial coherence (cPCOH) method by aggregating channels into effective units.
• Applied random channel aggregation (permutations) and consensus thresholding for robust analysis.
• Validated the cPCOH method using model data and human stereotactic EEG recordings.

Main Results:

• cPCOH effectively filters shared inputs, outperforming the pseudoinverse method on model data.
• cPCOH significantly reduces false positives compared to the pseudoinverse method.
• The method allows adjustable trade-offs between false positives and false negatives via a consensus threshold.
• cPCOH yields sparser brain networks than standard coherence, revealing clearer spatial patterns.

Conclusions:

• The cPCOH method offers a more accurate and reliable approach to analyzing functional connectivity in large-scale neural data.
• It provides a tunable parameter for balancing false positives and negatives, enhancing analytical flexibility.
• The resulting sparser networks facilitate the study of brain dynamics during cognitive processes and generalize to EEG/MEG data.