Robust k-means-based Clustering of Independent Components Estimated from the EEG data

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Traumatic brain injury (TBI) often impairs anticipatory postural control.
• Accurate identification of neural sources is crucial for understanding TBI-related balance dysfunction.
• Electroencephalography (EEG) is a key tool for analyzing brain activity.

Purpose of the Study:

• To evaluate and compare different k-means clustering algorithms for robustly clustering independent components (ICs) from EEG data.
• To identify the most accurate and computationally efficient clustering method for localizing cortical sources related to balance deficits in TBI.
• To assess the performance of clustering algorithms on noisy, real-world EEG datasets.

Main Methods:

• EEG data from individuals with TBI performing postural control tasks were analyzed.
• Independent Component Analysis (ICA) was used to extract neural sources.
• Performance of traditional k-means, repeated k-means, and global k-means algorithms was evaluated.
• Clustering quality and computational runtime were key metrics for comparison.

Main Results:

• Global k-means demonstrated superior clustering quality compared to traditional and repeated k-means.
• Global k-means offered a significantly improved computational runtime.
• The global k-means algorithm proved effective in a noisy, real-world TBI EEG dataset.

Conclusions:

• Global k-means is the optimal algorithm for robustly clustering ICs in TBI research using EEG.
• This method accurately identifies cortical sources linked to balance dysfunction.
• The findings support the use of global k-means for efficient and reliable neural source localization in clinical neuroscience.