Inappropriate assumptions about EEG state changes and their impact on the quantification of EEG state dynamics

Abstract

It was with great interest that we read the recently published paper by Gärtner and colleagues (Gärtner et al., 2015) in Neuroimage regarding EEG microstates and background processes. The parcellation of multichannel continuous EEG data into periods of quasi-stable spatial configuration of oscillatory activity (microstate analysis) is becoming increasingly recognized as an innovative analysis for the spontaneous organization of brain function, and has frequently been covered by this journal Brodbeck et al., 2012;Custo et al., 2014;Gärtner et al., 2015;Katayama et al., 2007;Koenig et al., 2002;Musso et al., 2010;Yuan et al., 2012). This interest is timely, because conceptually very similar approaches (i.e., focusing on temporally coherent network activity) have become a powerful tool to analyze fMRI resting state data.The paper by Gärtner and colleagues addresses an important technical issue in the analysis of transiently stable EEG microstates. Since there is predominantly oscillatory activity in EEG, the signal is small and noisy around the moments of polarity reversal of the EEG field, which complicates the attribution of those EEG spatial configurations to a particular microstate. Many EEG studies have dealt with this problem by analyzing and attributing only time-points where the EEG field amplitude has a momentary maximum (the so-called Global Field Power (GFP) peak), and interpolating the assignment of these periods in between those maxima, thereby addressing the topographic stability across GFP peaks. These studies have quite consistently yielded microstate durations of approximately 60-100 msec, assuming 4 classes of microstates (Koenig et al., 2002).Gärtner and colleagues rightfully criticize the above described methodology for being incomplete, since a considerable part of the results is based on interpolation. Given that an EEG is dominated by alphaband activity, one would expect around 20 GFP peaks per second, which at a standard sampling rate of 250Hz, amounts to less than 10% of the data points. They propose a theoretical model that assumes that the sequence of EEG microstates may be explained as the result of a sparse sampling of an unobservable underlying background process. Based on the assumptions discussed in more detail below, but still relying only on observations made at GFP peaks, they estimate the duration of the states of this background process to be around 10 msec, which is much shorter than what the literature has reported on microstates. Although their model still reproduces longer microstate durations when considering only the stability across GFP peaks, it implies that the observation that microstate durations are much longer than state durations of the background process is a result of the sparse sampling of the background process.

Cognitive manipulation of brain electric microstates

“…Future studies should investigate this matter further without an a priori hypothesis for the number of microstates. A new method of analysis may be beneficial towards this end, such as that developed by Betzel and colleagues (Betzel et al, 2012) or Gartner and colleagues (Gärtner et al, 2015); however, the latter has engendered debate within the field recently (Koenig and Brandeis, 2016). Additional future directions include replicating the experiment in a population with schizophrenia and moving from a cognitive manipulation to a more robust sensory manipulation of microstates (e.g., EEG alpha photic driving).…”

EEG microstates as a tool for studying the temporal dynamics of whole-brain neuronal networks: A review

“…Importantly, it should be noted that the transition probabilities between microstate class assignments of maps at GFP peaks cannot simply be taken as proxies for the transition probabilities among microstate assignments in general. This remains a topic of controversy in current research (Gartner et al, 2015;Gschwind et al, 2015;Koenig and Brandeis, 2016).…”

Capturing the spatiotemporal dynamics of self-generated, task-initiated thoughts with EEG and fMRI

“…A modified k-means cluster analysis (Pascual-Marqui et al, 1995) was applied to the data of each subject and condition. Only maps at local maxima of the Global Field Power (GFP) entered the cluster analysis, as they represent time points of highest signal-to-noise ratio (Koenig and Brandeis, 2016;Koenig et al, 2002). Polarity of the maps was ignored in the clustering.…”