EEG Microstate Sequences From Different Clustering Algorithms Are Information-Theoretically Invariant.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Information Theory

Background:

• Electroencephalography (EEG) microstates represent fundamental, transient states of large-scale brain activity.
• Analyzing microstate sequences provides insights into brain dynamics during rest.
• Different clustering algorithms can yield varying microstate representations.

Purpose of the Study:

• To investigate the statistical and information-theoretical properties of EEG microstate sequences.
• To compare the impact of five distinct clustering algorithms on these properties.
• To determine if key microstate sequence characteristics are algorithm-dependent.

Main Methods:

• Computed microstate sequences from 20 resting-state EEG recordings.
• Applied five clustering algorithms: AAHC, K-means, K-medoids, PCA, and Fast-ICA.
• Analyzed static (e.g., Shannon entropy) and dynamic (e.g., entropy rate) properties, alongside Markov property and long-range correlations.

Main Results:

• Static properties showed some dependence on clustering algorithm choice.
• Information-theoretical quantities (e.g., entropy rate) were largely invariant across algorithms.
• Algorithm-independent properties suggest inherent features of resting-state EEG microstate sequences.

Conclusions:

• The choice of clustering algorithm does not significantly affect core information-theoretical measures of EEG microstate sequences.
• This invariance supports the comparability of results across different analytical approaches.
• Key dynamic properties of brain activity, as reflected in microstate sequences, are robust to methodological variations.