Infancy is a key period of human development during which the neurobiological foundations of emerging social, emotional, and cognitive skills are shaped through the interaction of child-specific factors and environmental experiences. Therefore, reliably characterizing the functional properties of brain activity and organization during infancy provides a unique opportunity for understanding the impact of early life experiences on brain development and associated behavior. Toward this end, electroencephalography (EEG) has been routinely used as a direct, non-invasive, and low-cost measure of brain activity that can be collected from infants during various states of arousal and/or activity. Most commonly, given its ease of acquisition and tolerance of head and body movement, EEG is readily and frequently acquired from infants while they sit on their caregiver’s lap and watch relaxing videos. Prior research using EEG data from awake infants during video-watching have tended to focus on single metrics of global (e.g., total power or power spectral density analyses) or local (e.g., region-of-interest power or event-related potential analyses) features. However, recent data indicates that understanding the associations between functional brain networks (i.e., whole-brain dynamics; Xie et al., 2022) and behavior are crucial for advancing understanding of brain development during the first years of life. One promising but relatively unexplored method for characterizing whole-brain dynamics collected from high-density EEG during infancy is microstate analysis.
Microstate analysis is a data-driven approach for identifying patterns of scalp potential topographies, or microstates, that reflect very short periods (i.e., typically less than ~ 150 ms) of synchronized neural activity (i.e., large-scale functional networks) evolving dynamically over time (Khanna et al., 2015; Michel & Koenig, 2018). A small number of four to seven canonical topographies have been replicated and consistently shown to explain the majority of topographic variance in the entire EEG signal recorded during rest (i.e., the absence of external task demands) in both children and adults. Several temporal properties are also frequently calculated for each microstate and have been reported to show unique variation in their values and associations with individual differences in behavior. Temporal measures routinely used in studies include 1) global explained variance (GEV), the total variance in the data explained by a microstate, 2) duration, the average time in milliseconds (ms) that a microstate was present before transitioning to another microstate, 3) coverage, the percentage of time for which a microstate was present, 4) occurrence, the frequency with which a microstate was present per second, and 5) transition probabilities, the probability of one microstate coming after another in the sequence. Importantly, the neural generators for each microstate can be identified with source localization techniques, a critical step in understanding their functional significance and potential relevance to developing behavior.
While microstate analysis has proven to be a highly informative method for studying brain function and organization at the millisecond-level in adults, very few studies using this approach in infants have been published. More specifically, of the six publications that we identified, four used the microstate analytic approach to examine event-related data (Bucsea et al., 2023; Gui et al., 2021; Maitre et al., 2020; Rupawala et al., 2023), one examined microstates during sleep (Khazaei et al., 2021), and one used microstate analysis to examine spontaneous EEG data collected from infants (i.e., 34, 6-10-month-olds) during video-watching (Brown & Gartstein, 2023). Unfortunately, none of this prior work investigated the reliability of microstate-related measures at this age, a critical step in understanding the potential use of microstates to study individual differences in behavior and development (Lopez et al., 2023). However, previous work has demonstrated the reliability of microstate analysis in adults and suggests it is likely present in younger age groups as well. More specifically, in adults, prior studies have indicated good-to-excellent internal consistency (i.e., stability of temporal properties within the same session) and short- and long-term test-retest reliability (i.e., stability of temporal properties between multiple sessions recorded in the same week) for each temporal property of each identified microstate (Antonova et al., 2022; Khanna et al., 2014; Kleinert et al., 2023; Liu et al., 2020; Popov et al., 2023). Notably, Liu et al. (2020) demonstrated that as little as 1–2 minutes of data showed sufficient psychometric properties for GEV, duration, coverage, and occurrence values (i.e., intraclass correlations (ICCs) > .60). Recently, Kleinert et al. (2023) demonstrated good-to-excellent short-term (ICCs = .87-.92) and long-term (ICCs = .67-.85) test-retest reliability of duration, coverage, and occurrence. Transition probabilities, however, have been shown to be much less reliable than GEV, duration, coverage, and occurrence values (Antonova et al., 2022; Kleinert et al., 2023; Liu et al., 2020). Critically, strong reliability has been demonstrated across microstate clustering algorithms (Khanna et al., 2014), recording lengths (two vs. three minutes; Kleinert et al., 2023), and EEG channel densities (Khanna et al., 2014; Kleinert et al., 2023; Zhang et al., 2021); though Zhang et al. (2021) demonstrated 8- and 19-channel arrays to have significantly lower reliability than higher density arrays. While previous research has indicated high reliability of resting-state EEG source localization with approximately 1.5-2 minutes of data (Cannon et al., 2012), no research exists examining the reliability of microstate sources at any age. Taken together, while studies in adults indicate strong promise for the reliability of microstate analysis in EEG data collected from infants, equal reliability cannot be assumed across developmental stages as shown in Popov et al. (2023), and must be individually examined for each population.
One barrier that may be contributing to the dearth of published studies using the microstate analytic approach for characterizing infant EEG data is the lack of comprehensive, step-by-step methodological resources for infant researchers to employ this approach in their own work. Resources that specifically use examples from infant EEG data are more likely to be adopted by infant EEG researchers than resources that focus on other populations. Production of resources for analyzing EEG data in ways that inform understanding of brain function and organization (such as microstate analysis) are especially important as large-scale, multi-site, longitudinal infant EEG studies such as the HEALthy Brain and Child Development (HBCD) study (Jordan et al., 2020) are amidst data collection, with opportunities for data access and analysis in the near future. And, in line with the open science movement, sharing of data and analytic methods will be critical for the replication of findings.
As a first step toward validating the use of EEG microstates for investigating infant brain development, the current study explored the feasibility of identifying microstates during video-watching resting-state and examined their psychometric reliability in 55, 5-10-month-old infants using high-density EEG. Specifically, we assessed 1) the stability of microstate topographies, their temporal properties, their transition probabilities, and their neural sources with increasing EEG data durations (i.e., 1–5 minutes), and 2) the internal consistency (i.e., split-half reliability) of the temporal properties, transition probabilities, and neural sources at each data duration. Given the lack of studies examining resting-state microstates during infancy, we did not make specific predictions about microstate characteristics (i.e., topographies, temporal properties, transition probabilities, neural sources) or their reliability. In order to facilitate methodological access to microstate analysis, the current study also provides resources for analyzing microstates during infancy in line with recent efforts to maximize the potential of EEG as a developmental neuroscience tool (Buzzell et al., 2023). Toward this end, we have provided a step-by-step tutorial and accompanying website for performing microstate analysis and microstate source localization of EEG data using Cartool software (Brunet et al., 2011). We also shared our EEG data in the Brain Imaging Data Structure (BIDS; Pernet et al., 2019) format on OpenNeuro (Markiewicz et al., 2021).