EEG microstate repertoire & dynamics captured by linear statistics or Gaussian stationary twin

1675 

Abstract Submission 

Jaroslav Hlinka¹, Nikola Jajcay²

¹Institute of Computer Science of the Czech Academy of Sciences, Prague, Czech Republic, ²National Institute of Mental Health, Klecany, Czech Republic

Jaroslav Hlinka  
Institute of Computer Science of the Czech Academy of Sciences
Prague, Czech Republic

Nikola Jajcay  
National Institute of Mental Health
Klecany, Czech Republic

One of the interesting aspects of EEG data is the presence of temporally stable and spatially coherent patterns of activity, known as microstates (Pascual-Marqui et al., 1995), which have been linked to various cognitive and clinical phenomena. However, various clustering algorithms have been used for microstate computation, and there is still no general agreement on the interpretation of microstate analysis.

This study addresses two gaps in the literature. Firstly, by applying several state-of-the-art microstate algorithms to a large dataset of EEG recordings, we aim to characterize and describe relations of various microstate algorithms. Secondly, we aim to test the hypothesis that dynamical microstate properties might be, to a large extent, determined by the linear characteristics of the underlying EEG signal, in particular, by the cross-covariance and autocorrelation structure of the EEG data, paving way to more efficient estimation (Pascual-Marqui, 2022; Jajcay, 2023).

Data: For the experimental part of our study, we used the publicly available EEG data that are part of the Max Planck Institut Leipzig Mind-Brain-Body Dataset (LEMON) dataset (Babayan et al., 2019). The dataset consists of 228 healthy participants comprising a young (N = 154, 25.1 +/-3.1 years, range 20–35 years, 45 female) and an elderly group (N = 74, 67.6 +/- 4.7 years, range 59–77 years, 37 female) acquired cross-sectionally in Leipzig, Germany, between 2013 and 2015. Among other neuroimaging modalities, participants completed a 62-channel EEG experiment at rest (rsEEG hereafter) using two paradigms: eyes open and eyes closed. We used directly the preprocessed EEG data (total N = 204) provided as EEGLAB .set and .fdt files. The complete description can be found in (Babayan et al., 2019). Briefly, all EEG data have a sampling frequency of 250 Hz, are low-pass-filtered with a 125 Hz cutoff frequency, and are ~8 min long.

Simulations: We generated a Fourier transform surrogate of the EEG signal to compare microstate properties. Alternatively, we treated the EEG data as a vector autoregression process, estimated its parameters, and generated surrogate stationary and linear data from fitted VAR.

Microstate algorithms: We compared 6 different algorithms that can be used for the clustering stage of the microstate analysis: (Topographic) atomize and agglomerate hierarchical clustering, Modified K-means, Principal component analysis, Independent component analysis, and Hidden Markov model.

Microstate measures: apart from comparing directly the extracted microstates topographies, we assess a number of standard summary statistics to describe the temporal characteristics of inferred sequences from various clustering algorithms: average lifespan, coverage, occurrence; as well as selected dynamic statistics: mixing time, entropy, entropy rate, and the first peak of auto-mutual information function.

In terms of relations of the algorithms, we show theoretically why the three ''classically'' used algorithms ((T)AAHC and modified K-Means) yield virtually the same results, while HMM algorithm gives the most dissimilar results.

By simulations, we show that microstate statistics of data (Fig. 1) and its Fourier surrogates (Fig. 2) are largely similar, hinting that microstate properties depend to a very high degree on the linear covariance and autocorrelation structure of the underlying EEG data.

Similarly, for a linear VAR data model we observed that it generates microstates highly comparable to those estimated from real EEG data, potentially providing even higher reliability of microstate repertoire and dynamics estimation due to robustness of the linear estimates.

The observation of high reproducibility of the microstates properties from the linear models, particularly Fourier surrogates and VAR model, support the conclusion that a linear EEG model can help with methodological and clinical interpretation of both static and dynamic human brain microstate properties.

Connectivity (eg. functional, effective, structural)

EEG/MEG Modeling and Analysis ¹

Methods Development ²

Multivariate Approaches

Task-Independent and Resting-State Analysis

Computational Neuroscience

Data analysis

Electroencephaolography (EEG)

Machine Learning

Modeling

Multivariate

NORMAL HUMAN

Open Data

Statistical Methods

Other - microstate