Keywords are a high-level summary of the topic and content of the article. An analysis of keyword co-occurrence can reflect the hotspot and trend of research in the field (25). This study analyzes the keyword co-occurrence patterns across two distinct categories of EEG research in stroke. Table 5 shows the top 20 keywords from both Dataset 2 and Dataset 3 ranked by TLS. In Dataset 2 which includes 1,207 articles focused on “EEG in acute stroke and its early complications,” the keyword co-occurrence network (Figure 5A) highlights terms like “stroke,” “EEG,” and “epilepsy” as central nodes emphasizing the focus on EEG applications for managing acute stroke complications such as seizures and non-convulsive status epilepticus. Figure 1 showcases a total of 83 keywords with a minimum co-occurrence frequency of seven. Other significant terms such as “ischemic stroke,” “MRI,” and “intracerebral hemorrhage” reflect the intersection of EEG studies with imaging and other neurological complications. This network illustrates how EEG is used to monitor and manage the early consequences of stroke indicating the importance of EEG in understanding and addressing stroke-induced brain changes during the acute phase. Dataset 3 which includes 1,724 articles focusing on “EEG in neurological rehabilitation,” shows a different pattern in its keyword co-occurrence network (Figure 5B). This figure displays a network of 137 keywords all of which appear with a minimum frequency of seven. These central terms like “brain-computer interface (BCI),” “motor imagery (MI),” and “neurorehabilitation” are prominent reflecting the shift in focus toward using EEG in stroke recovery particularly in enhancing rehabilitation strategies through BCI systems. Keywords such as “functional connectivity,” “training,” and “rehabilitation” are tightly linked indicating the growing interest in leveraging EEG to promote motor recovery and brain plasticity in the chronic phase of stroke. Moreover terms like “virtual reality (VR)” suggest an expanding interest in integrating advanced technologies with EEG-based rehabilitation. Comparing the keyword co-occurrence networks of the two categories reveals clear differences: acute stroke research is primarily concerned with monitoring and managing immediate stroke-related complications while the rehabilitation category emphasizes long-term recovery and functional improvement through EEG-based interventions.

Analysis of keywords clustering is to categorize closely related keywords, which can reveal the hotspot of research in the field (26). The collected data were imported into CiteSpace for keyword clustering analysis, the smaller the cluster number, the more keywords the cluster contains. Modularity Q is a measure of the efficacy of clustering, with a range from 0 to 1. A value approaching 1 indicates a high degree of connectivity within clusters (14). For Dataset 2, which focuses on acute stroke and its early complications, the analysis revealed a Modularity Q of 0.7094, indicating substantial network modularity and high clustering quality, as shown in Figure 6A. The LLR clustering method identified 19 distinct clusters, each representing a distinct area of research, which were subsequently labeled with descriptive terms, including #0 carotid endarterectomy, #1 animal models, #2 cardiac surgery, #3 antiepileptic drug, #4 cerebrovascular disease, #5 stroke, #6 functional connectivity, #7 status epilepticus, #8 delayed cerebral ischemia, #9 spreading depression, #10 cognition, #11 biomedical signal processing, #12 stroke-related seizures, #13 temporal lobe epilepsy, #14 cerebral blood flow, #15 cortical excitability, #16 medulla-oblongata, #17 stroke-like episodes, #18 cortical infarction. Figure 6B presents the clustering results for Dataset 3, which is centered around EEG research in neurorehabilitation of stroke, also exhibited strong clustering results with a Modularity Q of 0.7791. The LLR clustering method was employed to identify distinct 19 clusters, including #0 quantitative electroencephalography, #1 stroke, #2 transcranial magnetic stimulation, #3 brain-computer interface, #4 functional connectivity, #5 ischemic stroke, #6 motor imagery, #7 carotid endarterectomy, #8 upper extremity, #9 feature extraction, #10 traumatic brain injury, #11 case report, #12 sensorimotor integration, #13 brain activity, #14 transcranial direct current stimulation, #15 cerebrovascular accident, #16 corticomuscular coherence, #17 brain plasticity, #18 seizures.

Keyword burst analysis has been demonstrated to reveal the areas that have received the most attention within a specific timeframe thereby identifying the emerging research frontiers (26). Figure 7 illustrates the top 25 burst keywords. The “Begin” and “End” columns indicate the timeframe of the keyword burst while “strength” denotes the intensity of the burst. For Dataset 2 (analyzed with a 1-year duration parameter) Figure 7 displays the top 25 burst keywords. The earliest detected burst corresponds to “blood-flow,” which also exhibits the longest sustained burst period. Notably “therapeutic hypothermia” demonstrates the highest burst strength. The keywords “functional connectivity,” “guidelines,” “acute symptomatic seizures,” “patterns,” “score,” “connectivity,” “acute ischemic stroke,” and “outcm” are still experiencing bursts. For Dataset 3 Figure 8 reveals distinct patterns: “activation” emerges as the earliest burst keyword while “cortex” maintains the most prolonged burst duration and the keyword “task analysis” registers the strongest burst intensity. The following keywords are still experiencing bursts: “upper limb,” “feature extraction,” “stroke,” “task analysis,” “machine learning,” “deep learning,” “network,” and “stimulation.”

EEG research in acute stroke primarily focuses on the early identification and management of complications such as epilepsy and disorders of consciousness. The keyword co-occurrence analysis for Dataset 2 reveals a dominant emphasis on using EEG for monitoring post-stroke complications, especially seizures and non-convulsive status epilepticus, which are crucial for early intervention. Terms such as “stroke,” “EEG,” and “epilepsy” prominently feature in the research landscape, reflecting the clinical focus on utilizing EEG to understand and manage stroke-induced brain changes.

The analysis of keywords highlights epilepsy and seizures as central themes in acute stroke research. Post-stroke epilepsy is significantly associated with adverse outcomes and elevated mortality rates (40). QEEG can assist in identifying the typical EEG patterns associated with stroke (41). Non-convulsive seizures are frequently unrecognized clinically, as standard observations may prove inadequate for detecting these anomalies. Nevertheless, EEG monitoring is capable of capturing essential electrophysiological changes. Non-convulsive seizures are frequently unrecognized clinically, as standard observations may prove inadequate for detecting these anomalies (42). However, EEG monitoring can effectively capture essential electrophysiological changes. Post-stroke epilepsy in EEG typically manifests as focal or generalized slowing, with some cases also showing lateralized periodic discharges. Bentes et al. (43) conducted a study of long-term follow-up of patients who had experienced an anterior ischemic circulation stroke, finding that 25.2% experienced seizures within the first year, with 22.7% of these acute symptomatic seizures detected only by EEG. For instance, Bentes et al. (43) demonstrated that 64-channel EEG with synchronized video-polysomnography during the first post-stroke week captured electrographic seizures in over 20% of anterior ischemic stroke patients, with 62% of these events occurring during sleep. These findings emphasize the necessity of prolonged monitoring, as 22.7% of acute symptomatic seizures are identifiable only through EEG. Beyond seizure detection, EEG’s predictive capacity extends to acute neurological deficits. Vanderschelden et al. (44) conducted a prospective study of 50 acute stroke patients evaluated by recording of EEG at rest state. The delta-theta/alpha-beta ratio (DTABR) was calculated. Multivariable modeling revealed that while age, diabetes status, and infarct volume explained 47% of NIHSS score variance, adding contralesional DTABR enhanced prediction, achieving 60% explanatory power. These seizures can contribute to worsened neurological outcomes, increased risk of mortality, and prolonged recovery time, highlighting the importance of EEG in reducing these adverse effects by enabling timely treatment (45).

Expanding beyond epilepsy, EEG provides objective biomarkers for post-stroke consciousness disorders. Reduced prefrontal-to-motor cortical information flow, measurable via transcranial magnetic stimulation coupled with high-density EEG (TMS-EEG), correlates with impaired arousal states. Bai et al. (46) reported that patients with unresponsive wakefulness syndrome exhibit ≥40% reductions in gamma band connectivity between prefrontal and motor regions, while minimally conscious patients show disrupted prefrontal-parietal alpha coherence predictive of 6-month recovery. Such electrophysiological signatures align with spectral shifts observed in consciousness research: elevated low-frequency oscillations (delta/theta) and attenuated cross-frequency coupling reflect diminished cortical integration, whereas preserved theta-gamma phase-amplitude interactions may signify recovery potential (47, 48). While EEG excels in detecting electrophysiological anomalies, its integration with advanced neuroimaging techniques enables deeper insights into structure–function relationships In a targeted investigation of thalamic stroke, researchers employed DTI-derived fractional anisotropy (FA) maps alongside qEEG to probe microstructural and functional connectivity disruptions. Correlational analyses linked theta-band EEG power reductions to FA decreases in the cingulum bundle and corpus callosum-key components of the default mode network known to modulate resting-state theta activity. Alpha-band power further correlated with FA in cortico-thalamic circuits, supporting the “thalamocortical dysrhythmia” model of stroke-induced network dysfunction. This multimodal approach also bridged behavioral deficits with neural markers: FA reductions in the right cingulum predicted impaired spatial memory, while splenium of the corpus callosum correlated with facial recognition deficits (49). Thus, EEG in acute stroke research is not only focused on identifying early stroke-related complications but also on understanding the underlying neurological processes that influence stroke recovery, especially in critical conditions like seizures and consciousness disorders. While the current study primarily focuses on EEG applications in neural monitoring and rehabilitation, we acknowledge the importance of exploring broader physiological mechanisms, including inflammation and oxidative stress (ROS), in stroke pathology (50–53). Although EEG itself does not directly measure inflammatory markers or ROS levels, emerging research highlights indirect correlations between EEG patterns and these mechanisms. For example, post-stroke neuroinflammation can disrupt cortical excitability and functional connectivity, which may manifest as altered EEG spectral power or coherence (54–57). Additionally, oxidative stress has been linked to impaired neurovascular coupling (58–60), potentially affecting EEG-derived metrics. Future studies could integrate EEG with biomarkers (e.g., serum cytokines, ROS assays) to investigate these relationships. The increasing emphasis on real-time monitoring and the integration of EEG with other diagnostic tools, such as MRI or PET scans, has great potential in enhancing the clinical management of acute stroke patients, reducing mortality and improving recovery rates.

EEG research in neurological rehabilitation (Dataset 3) shifts its focus from immediate stroke complications to the long-term recovery process. The analysis of keywords for this dataset highlights the growing prominence of BCI technology, MI, and neurorehabilitation, reflecting the increased integration of EEG into rehabilitation efforts aimed at improving motor function and cognitive recovery. Terms such as functional connectivity, training, and rehabilitation emphasize the growing recognition of EEG’s potential in promoting neural plasticity during stroke recovery, particularly in enhancing motor recovery through non-invasive brain-computer technologies. One of the most profound advancements in neurological rehabilitation is the use of BCI systems. MI enables patients to engage motor-related brain regions by imagining limb movements without actual physical execution, thereby promoting neural plasticity, particularly in the recovery of upper limb and hand function. BCI technology further enhances the effectiveness of MI by decoding patients’ motor intentions and translating them into commands for external devices, thereby enabling hemiplegic patients to achieve indirect motor control (61). This not only improves motor function but also helps patients maintain active engagement during rehabilitation training. Benzy et al. (62) analyzed cortical activity during MI and successfully decoded the imagined hand movement direction (left/right) in stroke patients. The patients used the phase-locking value of EEG signals to decode the direction of imagined hand movement, which then controlled a motorized arm assistive device, allowing patients to move their impaired arms in the intended direction. The combination of EEG with EMG (electromyography) has also gained attention in recent years, particularly in enhancing the precision of motor control during rehabilitation. Li et al. (63) introduced EEG–EMG hybrid systems, which combine the advantages of both EEG for motor intention detection and EMG for muscle activity detection. This dual approach improves the accuracy of rehabilitation training, providing more personalized feedback to patients and potentially accelerating recovery. The hybrid system allows for more accurate decoding of patients’ movements and enhances their ability to perform motor tasks during rehabilitation. Another notable advancement in EEG-based rehabilitation is the application of VR in conjunction with EEG, offering patients immersive and interactive environments that stimulate motor recovery. As research progresses, the integration of EEG with VR systems is showing promise in fostering neuroplasticity by creating engaging and tailored rehabilitation experiences (64, 65). The development of signal processing technologies has greatly optimized EEG preprocessing, feature extraction, and classification methods. Traditional feature extraction frequently employs time-domain, frequency-domain, and time-frequency analyses, such as the utilization of power spectral density to examine patterns of brain activity across diverse frequency bands in patients (16, 66). The damaged regions of the brain in patients with stoke frequently exhibit augmented low-frequency bands and diminished high-frequency bands. The advent of deep learning, encompassing convolutional neural networks and generative adversarial networks, has facilitated the automated extraction of features. These neural network models allow for complex pattern recognition directly from raw EEG signals, facilitating a deeper understanding of post-stroke brain functions (67, 68). Tong et al. (69) constructed a deep learning model based on EEG signals for rapid detection of ischemic stroke. They gathered EEG data from 20 acute ischemic stroke patients and 19 healthy controls and introduced a fusion feature combining correlation-weighted phase lag index and sample entropy to explore inter-channel synchronization and functional connectivity. Recent studies have demonstrated that complex network analysis of EEG data can provide insights into the reorganization of brain networks after stroke. For example, one study (70) analyzing resting-state and task-state functional connectivity identified a “cognitive network” comprising nodes in the subcortical, frontoparietal, visual, and cerebellar networks. This network shows differential effective connectivity patterns that are sensitive to post-stroke cognitive impairment and improvement. Moreover, another study (71) focused on mild stroke patients compared EEG-based functional connectivity during cognitive tasks across groups with cortical infarctions, subcortical infarctions, and healthy controls. Their graph theory analysis revealed significantly reduced global and local efficiencies in patient groups, along with distinct nodal strength distributions that differed by lesion location. A systematic review (72) compared EEG-derived complex network parameters between stroke patients and healthy subjects. Although the effect sizes for parameters such as path length, clustering coefficient, and cohesion were modest, the review highlighted both structural differences and certain overlapping features between the groups. Additionally, multimodal data fusion techniques are increasingly applied to stroke EEG studies. By combining EEG with fMRI or near-infrared spectroscopy, researchers can obtain more comprehensive brain activity data, valuable for early prognosis prediction and evaluating the effectiveness of different rehabilitation interventions (73, 74). Stroke may not only damage local neural structures but also disrupt large-scale brain networks, affecting both structural and functional connectivity. Stroke-induced lesions may impair the integrity of the default mode network and the cortico-thalamic circuits, leading to reduced global efficiency and altered modular organization. Such disruptions may contribute to deficits in cognitive and motor function by impairing inter-hemispheric communication and reducing the integration of distributed neural systems (71, 75–77).