Hybrid Genetic Algorithm for Clustering IC Topographies of EEGs.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Clustering independent component (IC) topographies from electroencephalograms (EEG) aids in identifying brain processes, especially when event-related potential (ERP) features are absent.
• Current clustering algorithms may not be optimal for analyzing EEG IC topographies.

Purpose of the Study:

• To propose a novel algorithm for clustering EEG IC topographies.
• To compare the performance of the proposed algorithm against existing clustering methods.

Main Methods:

• A hybrid clustering approach combining spectral clustering and genetic algorithms for centroid refinement.
• Automatic selection of the optimal number of clusters using a fitness function based on local density, compactness, and separation.
• Validation using specific internal metrics adapted for the absolute correlation coefficient as a similarity measure.

Main Results:

• The proposed algorithm significantly outperforms baseline clustering algorithms (EEGLAB's CORRMAP) in clustering EEG IC topographies.
• The algorithm demonstrates robust performance across different independent component analysis (ICA) decompositions and subject groups.

Conclusions:

• The novel hybrid clustering algorithm offers a superior method for analyzing EEG IC topographies.
• This advancement can improve the identification of brain-generated processes in EEG data.