Stroke is widely acknowledged as the second leading cause of death and a prominent contributor to disability globally (Feigin et al., 2017, 2021), which results in severe behavioral impairments and widespread structural and functional network disruptions (Alkhachroum et al., 2022). Post-stroke, patients commonly experience symptoms such as disorder of consciousness (DOC) or coma, which contribute to increasing in-hospital mortality and unfavorable outcomes upon discharge for stroke patients (Li et al., 2016). Therefore, it is crucial to diagnose post-stroke disorder of consciousness (PS-DOC) promptly and accurately, while gaining a comprehensive understanding of the neural mechanisms underlying brain injury. Traditionally, clinical rating scales like the Glasgow Coma Scale (GCS) and Coma Recovery Scale Revision (CRS-R) have been used to assess patients with DOC. Although clinical behavioral assessment remains the gold standard (Hermann et al., 2020), these scoring systems exhibit high inter-rater and inter-examiner variability and lack objective evidence of central nervous system damage following brain injury (Claassen et al., 2016; Giacino et al., 2018; Song et al., 2018).

Currently, the utilization of electrophysiological methods, specifically electroencephalography (EEG), to measure neurological function in patients has been demonstrated as an effective method for rapidly assisting in the diagnosis of DOC (Bai et al., 2021b; Ballanti et al., 2022; Duszyk-Bogorodzka et al., 2022). Extensive research utilizing EEG-based spectral analysis, source imaging analysis, and graph theory analysis has improved neurophysiological assessments in the fields of stroke and DOC rehabilitation and diagnosis (Finnigan et al., 2016; Bai et al., 2021a; Zhuang et al., 2022; Bouchereau et al., 2023; Colombo et al., 2023). Regarding spectral patterns, previous studies have documented a notable reduction in alpha power after brain injury, including stroke (Edlow et al., 2021). Consequently, stroke leads to a marked elevation in the delta/alpha ratio (DAR), which quantifies the ratio of delta band power to alpha band power (Finnigan and van Putten, 2013). Likewise, distinguishing between patients with DOC and healthy controls often relies on the analysis of delta and alpha frequency bands, where increased delta rhythms and diminished alpha rhythms serve as prominent indicators of reduced consciousness levels (Rossi Sebastiano et al., 2015). Specifically, DOC patients demonstrate higher delta power than healthy controls (Sitt et al., 2014), while an augmentation in alpha power is observed during the recovery of consciousness in these individuals (Stefan et al., 2018). These studies seem to suggest that we can observe the relationship between the spectral feature DAR and reduced consciousness in patients with PS-DOC.

Although traditional spectral analysis of resting-state EEG integrates brain activity over several seconds in different frequency bands, this method fails to capture the spatial and temporal characteristics of resting-state brain networks occurring at shorter time scales (e.g., tens of milliseconds; Li et al., 2022). In contrast, multi-channel fusion of EEG microstate analysis can capture the spatiotemporal dynamics of activity in different brain regions at a sub-second time scale (Bréchet et al., 2019). Microstates represent specific topological patterns of electrical potentials and are typically classified into four distinct classes (Michel and Koenig, 2018), and the microstates persist for a transient period of approximately 60–120 ms in a quasi-stable state before rapidly transitioning to another microstate category (Lehmann et al., 1987). The swift transitions between microstates reflect rapid changes in brain dynamics, revealing the interconnectedness between cognitive function, information processing, and neural regulation in the brain (Khanna et al., 2014; Von Wegner et al., 2018; Liu et al., 2020). Furthermore, different microstate classes exhibit strong associations with specific resting-state networks (RSNs) in the brain, including the auditory network, visual network, salience network, and attention network, among others (Britz et al., 2010; Michel and Koenig, 2018). Increasing evidence suggests that abnormal alterations in temporal characteristics (such as mean duration, coverage, and occurrence) of microstates are observed in various neuropsychiatric disorders, including post-traumatic stress disorder (Terpou et al., 2022), schizophrenia (Rieger et al., 2016; Lin et al., 2022), Alzheimer’s disease (Tait et al., 2020), Parkinson’s disease (Pal et al., 2021), and depression (Zhao et al., 2022). However, microstate analysis research related to DOC primarily focuses on patients with diverse etiologies, including brain trauma, intracranial bleeding, hypoxic–ischemic, and other conditions (Guo et al., 2022; Toplutaş et al., 2023; Zhang et al., 2023). In contrast, there is limited research on microstate analysis in DOC patients with a single etiology, such as ischemic stroke, and our understanding of the temporal dynamics and spatiotemporal interaction effects in their brains remains insufficient.

Moreover, substantial evidence suggests that microstate time sequences display dynamic and nonlinear characteristics, including non-Markovian transition behaviors, where the transition to the next microstate class is independent of the current microstate class (Gschwind et al., 2015; Von Wegner et al., 2017). Increasing studies have introduced nonlinear measures applied to microstate sequences. In particular, Tait et al. pioneered the utilization of the Lempel-Ziv complexity (LZC) algorithm to investigate microstate transition patterns (Tait et al., 2020), revealing a reduction in microstate LZC among individuals with Alzheimer’s disease in comparison to their healthy counterparts. Subsequently, Zhang et al. explored the alterations in the LZC of microstate sequences in patients with brain diseases (Zhang et al., 2021), and Zhao et al. discovered an increase in the LZC of microstates in adolescents with depression (Zhao et al., 2022). Nonlinear analysis of EEG microstate sequences quantifies the persistent characteristics of brain electrical activity, revealing complex dynamic changes at very small time scales (Von Wegner et al., 2023). We suggest that the microstate LZC in PS-DOC patients may also exhibit some degree of abnormality, providing new insights into the neuro anomalies associated with DOC.

The aforementioned analysis indicates that the current understanding of EEG microstates in PS-DOC remains limited. To this end, the innovations and contributions of our study are summarized as follows.

Firstly, to the best of our knowledge, this paper is the first work to investigate the differences in EEG microstates between PS-DOC patients and post-stroke awake (PS-AW) conscious state patients. Comparison results show that there exist differences in microstate topographies between the two groups and especially exhibit significant alterations in temporal features among them.

Secondly, we analyze the Lempel-Ziv complexity of the microstate time sequences and find that, there exhibits higher repetitiveness and slower transition trends in the microstates of PS-DOC patients than that of PS-AW patients. Additionally, to supplement the spectral information in resting-state EEG, we calculate the DAR of spectral features in both groups. We find that DAR is significantly higher in PS-DOC patients.

Finally, we explore the potential of the aforementioned extracted features that are sensitive to intergroup variability in the classification of DOC. In particular, we fuse these features and feed the combined sets into an SVM classifier to identify the DOC among stroke patients. The outcome demonstrates that our work could accurately identify 92.86% of DOC patients.

In summary, our study contributes to a better understanding of resting-state EEG microstate features in patients with DOC post-stroke, helps us to identify potentially valid electrophysiological biomarkers, and provides important insights and neurological evidence into the causative mechanisms of decreased levels of consciousness post-stroke.

Over the past decade, the integration of electrocorticography, functional magnetic resonance imaging, and EEG source imaging techniques in the analysis of EEG microstates has revealed a close relationship between microstates and RSNs along with their corresponding cortical regions (Yuan et al., 2012; Khanna et al., 2015; Gschwind et al., 2016; Bréchet et al., 2019; Zoubi et al., 2020; Rajkumar et al., 2021; Mikutta et al., 2023), demonstrating their capacity to evaluate the large-scale electrical activity in cortical networks (Tarailis et al., 2023). In this study, both the PS-AW group and the PS-DOC group exhibit four typical microstate topographies, with noticeable differences observed in the topological structure of microstate C. Regarding the source localization and specific association of microstate C with RSNs, a consensus has not yet been reached. Previous studies suggested an association between microstate C and the salience network (Britz et al., 2010), while recent research indicated its potential origins from the posterior cingulate cortex and precuneus (Custo et al., 2017; Zanesco et al., 2021), regions known for their critical role in the default mode network (DMN) (Fransson and Marrelec, 2008; Hagmann et al., 2008). An investigation on the consciousness level of brain-injured patients demonstrated a significant peak in DMN connectivity within the posterior cingulate cortex, highlighting the strong correlation between activity in this region and patients’ consciousness levels (Cavanna and Trimble, 2006; Vanhaudenhuyse et al., 2010). Croce et al. found that transcranial magnetic stimulation to the two pivotal regions of the DMN significantly altered the microstate C topography (Croce et al., 2018), suggesting the potential of microstates in elucidating local inhibition within brain regions. These findings imply that the decreased consciousness level in patients with stroke-induced brain damage may be associated with the inhibition of key areas within the DMN. This inhibition is likely to result in alterations in the microstate C topography, reflecting changes in neural activity and connectivity.

Furthermore, we observe a noticeable increase in the mean duration and coverage of microstate C in the PS-DOC patient group, which is in line with recent research (Toplutaş et al., 2023), although the DOC patients in their study cohort were of different aetiologies and their control group consisted of healthy subjects. These findings imply that the temporal features of microstate C may not be sufficiently sensitive to differentiate DOC patients caused by stroke and those with other etiologies, however, it still holds potential as a biomarker for DOC diagnosis. Another study reported a substantial reduction in the mean duration and coverage of microstate C in improving DOC patients during the consciousness recovery process following high-definition transcranial direct current stimulation (Guo et al., 2022), which further supports this viewpoint.

Regarding microstate D, we find a noticeable decrease in the mean duration, coverage, and occurrence in the PS-DOC group, which is opposite to the increased temporal parameters of microstates B and C. Microstate D is negatively correlated with the BOLD signals of the frontal and parietal lobes’ right posterior dorsal regions, which correspond to the dorsal attention network (DAN) activity (Britz et al., 2010; Michel and Koenig, 2018). Indeed, the opposite behavior of microstates C and D has been found in previous cognitive manipulation tasks, where the temporal measures of microstate D showed a significant augmentation in tasks involving the DAN, while the metrics of microstate C decreased (Seitzman et al., 2017). These research findings imply dysfunction of the DAN network in PS-DOC patients. Moreover, microstate D is suggested to represent certain aspects of attention and focus switching, and its temporal parameters show a decrease in response to reduced alpha power (Croce et al., 2020). The increase in DAR caused by the decreased alpha power observed in the PS-DOC group appears to explain the reduced time features of microstate D, indicating a dysregulation of attention in PS-DOC patients compared to PS-AW patients.

Moreover, The PS-DOC patient group shows an extended time parameter for microstate B, consistent with the results of Toplutaş et al. (2023) who found a remarkable increase in the mean duration and coverage of microstate B in DOC patients. Comparable results have been found in other neurological disorders, such as increased microstate B occurrence and coverage in migraine patients (Li et al., 2022), and higher mean duration, occurrence, and coverage of microstate B in adolescent depression patients (Zhao et al., 2022). Furthermore, microstate B has a negative correlation with bilateral occipital lobe BOLD activation (Michel and Koenig, 2018), and it has a 36% overlap with the DAN, followed by the visual network and the ventral attention network (36 and 31%, respectively; Zoubi et al., 2020). This suggests a relationship between the attenuated indicators of microstate D and the enhanced indicators of microstate B in PS-DOC patients, but further explanation is needed using more precise brain imaging techniques. Lastly, similar to microstates B and C, there is a significant increase in the mean duration of microstate A in the PS-DOC patient group. Guo et al. (2022) found a negative correlation between the CRS-R scores and the mean duration of microstate A in patients with DOC, which indicates that the mean duration of microstate A decreases gradually during the treatment process of consciousness recovery. Similarly, a longer mean duration of microstates in patients with DOC was associated with lower brain dynamical activity (Zhang et al., 2023). Moreover, the sleep or anesthesia state showed an increased mean duration of microstates in comparison to the awake state (Bréchet and Michel, 2022). In our study, the PS-AW group exhibits a shorter mean duration of microstates compared to the PS-DOC group, indicating that PS-DOC patients require more time for brain information transmission and processing under the same microstates and RSNs, which may affect the patient’s consciousness level.

The quantitative analysis of the frequency domain on resting-state EEG is indispensable. As mentioned earlier, several investigations have reported augmented delta-wave activity coupled with reduced alpha-wave activity in both stroke patients (Schleiger et al., 2014) and DOC patients following brain injury (Varotto et al., 2014). In this study, our results show an appreciable increase of DAR in the PS-DOC patient group, consistent with previous reports in brain injury and DOC patients. DAR correlated strongly negatively with functional recovery in acquired brain injury (Leon-Carrion et al., 2009), and Finnigan et al. found that a DAR ratio of 3.7 accurately distinguished ischemic stroke patients (Finnigan et al., 2016). Furthermore, Lechinger et al. (2013) found a high correlation between spectral peak frequency and CRS-R scores, and delta wave power exhibited a significant negative correlation with CRS-R scores (Bareham et al., 2018). These findings provide robust evidence for the increased DAR in PS-DOC patients and further emphasize the importance of using DAR as a quantitative EEG measure for DOC diagnosis and assessment.

Furthermore, there is a theoretical proposition that neural complexity is one of the important foundations of consciousness formation (Tononi, 2005, 2008), and the brain exhibits higher complexity in information integration and processing in states of higher consciousness levels (Dehaene and Changeux, 2011). These viewpoints have been validated in studies of EEG complexity in DOC patients, such as significantly lower complexity measured by LZC values in DOC patients (Wu et al., 2011), and a non-monotonic increase in LZC during the consciousness recovery process (Lei et al., 2022). In recent years, researchers have utilized the LZC feature of microstate time series to study the neurodynamics of patients with neurological disorders at a smaller time scale (Tait et al., 2020; Artoni et al., 2022). In our study, we apply the microstate LZC to DOC patients for the first time, and the results display lower microstate LZC values in the PS-DOC patient group, indicating a tendency for more repetitive and slower transitions of the microstates in the PS-DOC group, which may reflect a decrease in complexity during information integration and distributed processing. Additionally, a recent study found lower EEG LZC and permutation LZC to be associated with residual consciousness in DOC patients (Liu et al., 2023), which supports our results.

In recent years, significant progress has been made in the classification methods for DOC. This progress can largely be attributed to the application of electrophysiological methods in measuring brain neural function with the need to supplement the criteria for DOC classification with information from EEG (Bayne et al., 2017). Recent research proposed a multiple-scale convolutional few-shot learning network that can extract more information from EEG and applied it to evaluate DOC patients with residual consciousness in a P300-based brain-computer interface (BCIs) (Pan et al., 2023). Although there have been preliminary results in the assistance of diagnosing DOC patients using BCIs, BCI tasks often require active cooperation from patients, which remains a significant challenge for DOC patients (Galiotta et al., 2022). Another research attempted to maximally utilize functional connectivity information in brain networks to evaluate DOC, using convolutional neural networks and network reconfiguration techniques to classify patients with different consciousness states, achieving a classification accuracy of 87.2% (Cai et al., 2023). However, a recent study demonstrated that compared to convolutional neural networks or deep learning algorithms, SVM classifier showed stability in classifying various EEG signal preprocessing patterns for tasks such as DOC classification and had better interpretability, making them more suitable for doctors and patients to adopt (Rosenfelder et al., 2023). Our research explores the potential value of EEG microstates in DOC classification, which has not been explored in previous studies according to the available literature. We employ an SVM classifier to identify DOC patients among stroke patients using only the temporal features of the four microstate classes (accuracy = 77.59%), and obtain accuracies of 84.48 and 81.03% when incorporating the spectral feature DAR and the microstate complexity feature LZC, respectively. When combining DAR and microstates’ temporal and LZC features together, the highest accuracy and sensitivity are achieved (accuracy = 91.07%, sensitivity = 92.86%). The microstate temporal sequences and topographies have higher temporal and spatial resolution compared to EEG sequences. The features extracted from microstates can be complemented with frequency-domain information to improve the performance of classification, and higher sensitivity in classification is crucial for diagnosing patients with PS-DOC in a clinical setting.

Despite providing potential electrophysiological biomarkers for the post-stroke disorder of consciousness, this study has several limitations to consider. Firstly, the study only includes DOC patients caused by subacute ischemic stroke, which, although avoiding potential influences from other etiologies, still has high heterogeneity among ischemic stroke patients, such as lesion location and size, stroke severity, etc. These factors need to be considered in future studies for their impact on group analysis. Secondly, due to missing clinical scale scores in some DOC patients in the study cohort, the correlation analysis between different features and clinical scales is not conducted. Therefore, future studies need to supplement the correlation information between the features and scales. Finally, this study highlights the potential role of these features in the diagnosis and assessment of PS-DOC. However, to delve deeper into the observed uniqueness, future research should include longitudinal studies to investigate the effectiveness of these indicators in predicting stroke patient prognosis.

In this study, we conduct EEG microstate analysis on patients with DOC resulting from a single etiology, specifically ischemic stroke. Our findings reveal distinct differences in the topography of microstate C between PS-DOC patients and PS-AW patients, indicating localized suppression of DMN activity in PS-DOC. Microstates B and C exhibit significant increases in temporal features in the PS-DOC group compared to the PS-AW group, while microstate D shows the opposite pattern. These novel microstate findings provide valuable insights and additional information on abnormal RSN changes in DOC. Moreover, integrating EEG frequency domain analysis and nonlinear analysis of microstates, we observe notable increases in the DAR and reductions in the microstate LZC within the PS-DOC group. Lastly, utilizing these distinctive features as a comprehensive set, we achieve high accuracy in classifying PS-DOC patients (Accuracy = 91.07%). Our findings emphasize the potential of EEG microstates as promising electrophysiological biomarkers for the assessment of DOC among stroke patients, with further enhancements possible through the integration of spectral features.

The datasets presented in this article are not readily available because individual patient data cannot be made available under Chinese law because we did not obtain patient approval for sharing individual patient data, even in the form of electroencephalogram (EEG) data. Requests to access the datasets should be directed to lifenglian@tyut.edu.cn.

The studies involving humans were approved by the Ethics Committee of Shanxi Provincial People’s Hospital. The studies were conducted in accordance with the local legislation and institutional requirements. The ethics committee/institutional review board waived the requirement of written informed consent for participation from the participants or the participants’ legal guardians/next of kin because written informed consent for participation was not required for this retrospective study in accordance with the national legislation and the institutional requirements.

FY: Conceptualization, Methodology, Formal analysis, Investigation, Software, Writing—original draft, Writing—review and editing. YG: Methodology, Formal analysis, Writing—original draft, Writing—review and editing. FL: Conceptualization, Methodology, Formal analysis, Data collection, Writing–original draft, Writing–review and editing, Funding acquisition, Supervision. XZ: Data collection, Funding acquisition, Supervision. FH: Data collection, Investigation, Software. WJ: Data collection, Investigation, Funding acquisition. XL: Investigation, Writing–review and editing.

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. The study leading to these results received funding from the National Natural Science Foundation of China, the Youth Fund of Shanxi Health Commission, Provincial Special Supporting Fund Project of Shanxi Provincial People’s Hospital, under Grant Agreement Nos. 62171307, 201301029, and sj20019007.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.