Multichannel matching pursuit validation and clustering--a simulation and empirical study.

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Multichannel matching pursuit (MMP) is an emerging technique for analyzing electroencephalogram (EEG) signals with inverse modeling.
• The performance limitations of MMP, particularly with increasing complexity and noise, require thorough investigation.
• A modified K-means clustering algorithm was developed to aggregate similar MMP-derived brain activity patterns.

Purpose of the Study:

• To evaluate how the accuracy of the MMP algorithm is affected by an increasing number of simulated brain sources.
• To assess the robustness of the MMP algorithm to varying levels of noise in EEG signals.
• To implement and validate a K-means clustering approach for grouping similar spatiotemporal brain activity patterns across subjects.

Main Methods:

• Simulated 20 EEG signals across four groups (5, 10, 15, 20 sources) to test MMP accuracy with increasing source count and noise.
• Applied a modified K-means clustering algorithm to datasets from 10 simulated subjects per group (5-20 sources).
• Validated the clustering method on empirical somatosensory evoked potential (SEP) and brainstem evoked potential (BAEP) data.

Main Results:

• MMP algorithm accuracy showed a decline with a higher number of brain sources.
• MMP algorithm demonstrated significant robustness against increasing levels of noise.
• The K-means clustering successfully grouped similar MMP-derived brain activity components across subjects.

Conclusions:

• The combined MMP and K-means clustering approach is effective for identifying and grouping similar brain activity.
• This methodology facilitates the study of inter-subject differences in brain activation sequences in response to sensory stimuli.