Neural mechanisms underlying loss of functional independence after stroke: EEG evidence of impaired motor-to-sensory directed connectivity on the affected side

Abstract

Background

Post-stroke reorganization of the prefrontal-sensorimotor network is critical for functional recovery, yet single connectivity metrics fail to capture its multidimensional characteristics. This exploratory study investigates network reorganization patterns and their clinical relevance to motor function and daily independence using multimodal electroencephalography (EEG) connectivity analysis, including Coherence (COH), phase lag index (PLI), and Granger causality (GC).

Methods

Resting-state EEG from stroke patients (n = 22) and healthy controls (n = 22) was analyzed. COH, PLI, and GC metrics were computed for key regional connections involving the prefrontal (PFC), motor (MC), sensory (SC), and parietal (PC) cortices and correlated with Fugl-Meyer Assessment for Upper Extremity (FMA-UE) and Modified Barthel Index(MBI) scores. For stroke patients, hemispheres were classified as ipsilesional (i, affected) and contralesional (c, unaffected) based on the lesion side.

Results

Compared to healthy controls, stroke patients exhibited distinct reorganization patterns. Preliminary evidence suggests enhanced parietal-motor coherence (COH_cPC-cMC↑) in the unaffected hemisphere alongside reduced motor-to-prefrontal directed connectivity (GC_cMC→cPFC↓). Interhemispherically, the findings may indicate weakened information flow from the unaffected to affected motor/sensory areas (GC_cMC→iMC↓, GC_cMC→iSC↓), accompanied by reduced bilateral motor synchronization (PLI_iMC-cMC↓). Within the affected hemisphere, there was evidence of enhanced bidirectional parietal-motor phase transfer (PLI_iPC-iMC↑), whereas directed pathways were significantly impaired (GC_iMC→iSC↓, GC_iPC→iSC↓, GC_iPFC→iMC↓). Clinically, the affected motor-to-sensory pathway (GC_iMC→iSC) showed a strong negative correlation with the MBI (r = -0.61, P = 0.002), and both interhemispheric motor regulation (GC_cMC→iMC) and synchronization (PLI_iMC-cMC) were significantly correlated with functional deficits (P < 0.05).

Conclusion

Post-stroke networks may exhibit a “local compensation-global impairment” reorganization pattern. The affected motor-sensory pathway (GC_iMC→iSC) may be a biomarker for functional independence and thus allow for guided sensory integration-based precision rehabilitation.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12984-026-02089-1.

Background

Stroke is a leading cause of mortality and disability worldwide [1]. Although advancements in medical technology have significantly reduced stroke-related mortality, approximately 80% of stroke survivors still experience severe limb motor dysfunction, substantially impairing their ability to perform daily living activities [2]. Meanwhile, against the backdrop of a rapidly aging global population, the social and economic burden attributed to stroke is expected to continue rising in the coming years [3]. Therefore, gaining a better understanding of the neuropathological mechanisms underlying post-stroke motor impairments and developing effective intervention strategies remain essential goals in stroke research.

Electroencephalogram (EEG) signals, which originate from the electrical activity of discharging cortical neurons, can provide insights into the synchronization and regulatory state of cortical population neuronal activity. They are considered a valuable tool for investigating abnormal neural mechanisms following stroke [4]. A recent systematic review has further highlighted the utility of high-density EEG in detecting biomarkers for diagnosis and prognosis in stroke, particularly in investigating motor and cognitive network dynamics [5]. Numerous EEG studies have revealed various aberrant electrophysiological patterns post-stroke, including suppressed neural activity in local brain regions, compensatory activation in remote areas, and redistribution of interhemispheric functional connectivity [6]. However, current research has primarily focused on task-based EEG signals, while studies on resting-state EEG remain relatively limited. Resting-state EEG has the advantage of including patients with any neurological deficit and is easier to perform than task-based variants [7]. Important gaps remain in our understanding of information exchange between brain regions during rest in stroke patients. Resting-state synchronous neural activity, which influences subsequent task performance in stroke patients, is considered fundamental to intrinsic functional integration of the brain. Therefore, further investigation of resting-state cortical network characteristics in patients may offer new perspectives for understanding the pathological mechanisms of stroke.

Functional connectivity analysis has become an important tool in brain network research over the past few years. Coherence (COH) is a classical metric that measures the linear synchronization strength between signals from different brain regions within specific frequency bands, and is often used to investigate synchronization among brain areas. Existing studies have found that post-stroke COH can exhibit bidirectional changes—enhanced or weakened—demonstrating stage- and region-specific characteristics. For example, some studies have observed enhanced connectivity between the affected motor area and contralateral brain regions, interpreted as possible compensatory neural plasticity, while others have reported weakened connectivity, suggesting reduced network efficiency [8]. Recent work using machine learning approaches has demonstrated that resting-state EEG coherence features can distinguish chronic stroke survivors with severe upper limb impairment from those with mild/moderate impairment, achieving high classification accuracy [9]. Although COH reveals the strength of functional connectivity, it only reflects correlation and cannot provide information on the direction of information flow. Granger causality(GC) analysis has been introduced into brain network research to address this limitation and identify directed information flow between signals. Studies employing directed connectivity measures such as directed Coherence (DC) and generalized partial directed Coherence (gPDC) have shown that rehabilitation can increase intra-hemispheric coupling between pre-motor and primary motor areas in subacute stroke patients, with specific connections correlating with upper-limb motor recovery [10]. A limited number of studies have revealed altered information flow in motor-related brain regions after stroke [11]. However, how information flows between brain regions is reorganized during the resting state after stroke, and the relationship between this reorganization and motor and sensory functional recovery remains to be fully elucidated.

Currently, relevant studies often analyze COH and GC separately, which limits our ability to comprehensively reveal the coupling changes in network structure and function after stroke. More importantly, neural activity is inherently nonlinear, and relying solely on linear metrics (such as COH) may overlook important interactive features. In recent years, stroke research has increasingly used nonlinear connectivity measures, such as the phase lag index (PLI). By quantifying non-zero phase differences between brain regions, PLI is thought to reduce volume-conduction interference and may more reliably reflect true functional connectivity. For instance, the phase synchronization index (PSI), a measure related to PLI, has been shown to predict motor recovery in patients with chronic stroke before rehabilitation intervention [12]. Furthermore, network topology analysis has revealed that cortical network modifications, including changes in clustering coefficient and path length, may underlie clinical evolution in the acute phase of ischemic stroke [13]. Beyond pairwise connectivity, emerging evidence suggests that rehabilitation modulates high-order interactions among large-scale brain networks, including the motor, default mode, and executive control networks, with increased synergy among these networks associated with motor recovery [14].

Despite these advances, several important gaps remain: (1) linear (COH) and nonlinear (PLI) connectivity measures are typically examined in isolation, which may fail to capture the full complexity of neural interactions; (2) undirected (COH/PLI) and directed (GC) connectivity are rarely integrated, limiting simultaneous assessment of connection strength and information flow direction; and (3) the relationship between these connectivity changes and large-scale network reorganization remains poorly understood. These gaps suggest the need for a comprehensive approach that integrates multiple connectivity metrics to elucidate how stroke alters both the strength and directionality of neural communication and how these changes might collectively contribute to functional network reorganization. After confirming the presence of both linear and nonlinear connections between brain regions, further identification of the direction of information flow will contribute to a more comprehensive and accurate elucidation of neural signal-processing mechanisms in stroke patients [12].

To address these gaps, the present study aims to integrate COH, PLI, and GC to systematically investigate the synergistic changes in functional connectivity strength and information flow direction in stroke patients during the resting state. We propose a “connection enhancement - flow reorganization” compensation hypothesis: patients may exhibit widespread increases in COH and PLI, along with selective changes in GC flow direction, suggesting that post-stroke brain networks achieve functional reorganization by strengthening connections and optimizing information routing. This exploratory study seeks to provide new insights into the compensatory mechanisms of brain networks after stroke and may offer a theoretical basis for precise rehabilitation treatments (such as targeted transcranial magnetic stimulation).

Methods

Participant recruitment

This study aims to recruit stroke patients and healthy subjects for an EEG experiment to compare differences in brain functional connectivity and information flow characteristics between stroke patients and normal individuals, thereby elucidating the abnormal neural mechanisms in stroke patients.

All participants provided informed consent and signed the informed consent form. The study was approved by the Ethics Committee of the Rehabilitation Hospital Affiliated with Fujian University of Traditional Chinese Medicine (Ethics Approval No.: 2023KY-048-002) and complied with the principles of the Declaration of Helsinki.

Specific inclusion and exclusion criteria are as follows:

Inclusion and exclusion criteria for the patient group

Inclusion Criteria: (1) First-time unilateral stroke; (2) Limb hemiplegia resulting from stroke; (3) Bunnstrom stage between 3 and 6; (4) No significant cognitive impairment, with basic language comprehension and expression abilities, and able to cooperate to complete the experiment; (5) Age 28–78 years; (6) Time since stroke: 1–12 months.

Exclusion Criteria: (1) Recurrent stroke; (2) Peripheral nerve injury in the upper limb due to trauma, fracture, or tissue damage; (3) Severe cognitive impairment or history of neurological/psychiatric disorders, making it difficult to cooperate with the experiment; (4) History of traumatic brain injury or other neurological diseases such as multiple sclerosis.

Inclusion criteria for the control group

The control group consisted of healthy individuals with no history of neurological, muscular, or skeletal disorders, no diagnosed mental health conditions, and no long-term use of medications that may affect brain function or upper limb motor function. The control group was matched with the case group for age and gender.

Participant characteristics

Based on the above criteria, this study recruited 22 healthy subjects and 22 stroke patients with upper limb motor impairments. The demographic and clinical characteristics of all participants are shown in Table 1.

Table 1.

Demographic and clinical characteristics of participants

Group	Numbers	Age	Sex male/female	Disease Course (months)	Brunnstrom Stage(upper limb) ()	FMA-UE ()
Stroke group	22	56.0 ± 13.35	16/6	5.5 ± 4.40	4.7 ± 1.16	44.0 ± 17.55
Control group	22	55.68 ± 8.97	16/6	–	–	–
P value	–	0.80	1.00	–	–	–

FMA-UE: Fugl-Meyer Assessment for Upper Extremity

P values indicate group comparisons (independent t-test for age; chi-square test for sex)

It is worth noting that after comprehensive clinical evaluation and examination by expert physicians, all stroke patients were diagnosed with cortical stroke, indicating that the stroke occurred in the cerebral cortex region. Although there was some inter-subject heterogeneity within the stroke patient group, their demographic and clinical characteristics showed no significant differences, ensuring comparability within the patient cohort.

Data acquisition and preprocessing

Data acquisition

The EEG experiment was conducted in an electromagnetically shielded laboratory with appropriate lighting and minimal noise. Participants sat comfortably before a screen with their trunks leaning against the chair back to restrict anterior-posterior tilting. Before the experiment, a 64-channel Neuroscan saline electrode cap was immersed in 0.9% NaCl saline solution for 15 min. EEG data were acquired using a Neuroscan system with a 64-channel QuickCap, with electrodes distributed according to the 10–10 system—an extension of the standard 10–20 system providing higher spatial resolution while maintaining correspondence with conventional electrode positions. The cap was then positioned according to the International 10–20 system and allowed to stabilize for 10 min. Additional saline was applied to fine-tune the scalp electrode impedance to ≤ 10 kΩ. During the experiment, participants were instructed to close their eyes, minimize head movements, and refrain from performing motor or cognitive tasks. The entire resting-state recording lasted 3 min with a sampling frequency of 1 kHz. See Fig. 1.

Fig. 1.

Experimental design and data acquisition pipeline

Preprocessing

To account for inter-individual variability in lesion laterality, all stroke patient data were spatially normalized based on the sagittal plane. For patients with right-hemispheric lesions, the data were mirrored to the left side, ensuring that the lesioned hemisphere was consistently aligned on the left for all participants. This approach was implemented to control variables and enhance the rigor of the results.

EEG data (including mirrored data for patients with right-hemisphere lesions) were preprocessed using the EEGLAB toolbox (v2024.2) in MATLAB (R2021b). The data were acquired using a 64-channel Neuroscan QuickCap with electrodes arranged according to the 10–10 system, an extension of the standard 10–20 system. During preprocessing, electrodes were precisely localized by directly matching them to the standard 10–10 system positions available in EEGLAB; no additional spatial transformations or interpolations were applied. Non-functional channels (e.g., M2, CB1, CB2) were removed. A 0.5–80 Hz bandpass filter was applied to eliminate low-frequency drift and suppress high-frequency noise, followed by a 49–51 Hz notch filter to remove 50 Hz line noise. Both filters were implemented as linear finite impulse response (FIR) filters using the standard EEGLAB function pop_firws(), with filter orders automatically estimated by the filter design algorithm to achieve the minimum sufficient order for the specified transition bandwidths. The data were then downsampled to 250 Hz to reduce computational load.

Subsequently, independent component analysis (ICA) was performed using the pop_runica() function in EEGLAB. The ICLabel function automatically identified artifactual independent components based on their spatiotemporal features; components with a probability greater than 90% of being classified as vertical or horizontal eye movements or cardiac artifacts were manually verified and removed by an experienced researcher. The remaining components were back‑projected to reconstruct artifact‑corrected continuous EEG signals. This process removed physiological artifacts without discarding any time segments, thereby preserving the data continuity required by the HERMES toolbox. Whole‑brain average referencing was then applied. This order ensured that high‑amplitude artifacts did not bias the reference computation. Due to the automated and individualized nature of this process, the number of removed components varied across subjects and was not systematically recorded; however, all subjects retained sufficient neural components for subsequent analysis.

Brain region partitioning and feature extraction

This study aims to investigate the impact of stroke on brain networks by analyzing functional connectivity and information flow characteristics in resting-state EEG signals from stroke patients and healthy controls. Based on research objectives and standard electrode systems, 18 EEG channels were selected for targeted analysis: Fp1, Fp2, FC1, FC2, FC3, FC4, C1, C2, C3, C4, CP1, CP2, CP3, CP4, P1, P2, P3, and P4. The study focuses on motor, sensory, and higher-order integration networks that are frequently affected by stroke and are critical for rehabilitation. The selected channels cover key regions including the prefrontal cortex (Fp1/Fp2, involved in executive functions and attention), central areas (FC1-4 and C1-4, corresponding to the primary motor cortex and premotor areas), centroparietal regions (CP1-4, corresponding to the primary somatosensory cortex), and parietal areas (P1-4, involved in sensory integration and spatial attention) [15]. These regions play essential roles in resting-state networks (e.g., sensorimotor and dorsal attention networks) and are crucial for understanding post-stroke functional reorganization [16–18].

To facilitate subsequent functional analysis of brain regions, the 18 channels were symmetrically divided into four bilateral brain regions per hemisphere based on their anatomical positions and corresponding functional areas, as illustrated in Fig. 2. The Prefrontal Cortex (PFC), comprising Fp1 and Fp2 and located at the anteriormost portion of the prefrontal lobe, is critically involved in higher-order cognitive functions, including working memory, decision-making, and emotional regulation. The Motor Cortex (MC), encompassing FC1-4 and C1-4, features FC electrodes anterior to the central sulcus (covering premotor/supplementary motor areas) and C electrodes over the central sulcus crown (representing hand/arm regions of the primary motor cortex), serving as the core region for motor execution and planning. The Sensory Cortex (SC), consisting of CP1-4 with electrodes positioned posterior to the central sulcus (primary somatosensory cortex), primarily processes tactile and proprioceptive information, with sensory input essential for motor control. The Parietal Cortex (PC), including P1-4 electrodes covering the superior parietal lobe, participates in sensory integration, spatial attention, and motor intention formation, and is pivotal in sensorimotor integration and attention networks [19]. Following the spatial normalization procedure described in Sect. “Preprocessing” (where all lesions were aligned to the left hemisphere), the left hemisphere regions (PFC, MC, SC, PC) are hereafter designated as ipsilesional (prefixed with “i-”), and the right hemisphere regions as contralesional (prefixed with “c-”) in subsequent connectivity analyses.

Fig. 2.

Schematic of brain regions and electrode placement. The 18 EEG channels (Fp1, Fp2, FC1–FC4, C1–C4, CP1–CP4, P1–P4) are displayed on a standard 10–10 template. Colors denote functional regions: prefrontal (green), motor (purple), somatosensory (blue), and parietal (yellow). This parcellation was used for all subsequent functional connectivity analyses

Feature extraction was performed using the HERMES toolbox (v2020.4) in MATLAB (R2021b) to compute functional connectivity metrics between EEG channels [20]. Three complementary features were selected: COH, PLI, and GC. These metrics are widely used in neuroscience, and each captures distinct aspects of functional connectivity [21] .

All connectivity features were estimated directly from continuous 3-minute resting-state EEG recordings, without epoch segmentation, following the default implementation in the HERMES toolbox.

For COH and PLI, spectral estimation was performed using the Fast Fourier Transform (FFT) as implemented in HERMES default settings. COH is a frequency-domain measure that quantifies the linear correlation between two signals [21, 22] and was computed across the broadband range (0.5–80 Hz) to provide a global overview of functional coupling across frequencies. Broadband coherence estimates may be influenced by the 1/f spectral slope of EEG signals, which amplifies low‑frequency power. PLI measures the asymmetry of the phase difference distribution between two signals, offering resistance to volume conduction [23, 24]. To enhance stability, PLI was computed using a narrowband approach (center frequency 10–20 Hz, bandwidth 4 Hz) [23], covering mu/alpha and low beta rhythms, which are closely tied to sensorimotor function.

For GC, multivariate autoregressive (MVAR) models were fitted to the continuous data using the ARfit algorithm in HERMES [25]. GC reveals the directionality of information flow between signals [25–27]. GC analysis was performed on broadband signals (0.5–80 Hz) after bandpass filtering. Model order was automatically selected using information criteria (HERMES default). Stationarity was verified using the stability test embedded in HERMES; only models that passed the test were retained.

For each connectivity metric, all possible inter-regional pairs were computed. Specifically, for the undirected measures (COH and PLI), 28 pairs were calculated across the eight regions of interest (four per hemisphere). For the directed measure (GC), 56 directed connections (excluding self-connections) were computed. The statistical comparison of these features between groups is described in Sect.  “Statistical analysis”.

These features were subsequently compared between stroke patients and healthy controls. To further validate our findings, we also computed the weighted phase lag index (wPLI) and the imaginary part of Coherence (iCOH) as complementary measures. The results of these analyses are provided in Supplementary Table S2.

Statistical analysis

Functional connectivity metrics (COH, PLI, GC) were computed as described in Sect. “Brain region partitioning and feature extraction”. After obtaining feature matrices from the HERMES toolbox, feature values were averaged across channels within each predefined brain region to obtain regional representational signals. This approach—computing connectivity at the channel-pair level first and then averaging within regions—preserves the statistical dependencies inherent to functional connectivity metrics, which are defined at the level of signal pairs rather than averaged signals [21]. Averaging after connectivity calculation reduces random variability and enhances the stability of regional signal representation.

The number of inter-regional pairs (28 for undirected measures, 56 for directed measures) followed the same scheme described in Sect.  “Brain region partitioning and feature extraction”. Before statistical comparison, we assessed the normality and homogeneity of variance assumptions for each connectivity pair separately. Specifically, for each feature (e.g., COH_cP-cMC), the Lilliefors test (a Kolmogorov-Smirnov-based normality test suitable for small samples, n < 30) was applied to the patient and healthy control groups separately, with α = 0.05. The null hypothesis (H₀) was that the data were drawn from a normal distribution, and the null was accepted when P > 0.05.

If the normality assumption was satisfied in both groups (P > 0.05), an independent samples t-test was conducted to compare the feature between the patient group (n = 22) and the healthy control group (n = 22). Levene’s test was also performed to assess variance homogeneity. When Levene’s test indicated unequal variances (P < 0.05), the independent-samples t-test with Welch’s correction (ttest2 in MATLAB with ‘Vartype’, ‘unequal’) was used.

If the normality assumption was violated in either group (P < 0.05), the non-parametric Mann-Whitney U test (ranksum in MATLAB) was used as an alternative.

To quantify the direction of pathological changes, the mean difference (Δ = µ_patients - µ_controls) was calculated. A positive Δ (Δ > 0) indicated elevated connectivity in patients, suggesting potential neural compensatory enhancement, while a negative Δ (Δ < 0) showed reduced connectivity, reflecting neural impairment.

Given the exploratory nature of this study and the relatively small sample size, no correction for multiple comparisons (e.g., FDR) was applied. We acknowledge that this increases the risk of Type I error; however, this approach was chosen to avoid inflating Type II error and potentially obscuring biologically meaningful patterns. A total of 28 statistical tests were performed for each undirected measure (COH and PLI) and 56 for the directed measure (GC). Features showing nominally significant intergroup differences (uncorrected P < 0.05) were identified as potential pathology-sensitive indicators, providing preliminary hypotheses for future confirmatory studies. The complete statistical results for all connectivity pairs (including mean values, test statistics, and uncorrected P-values) are provided in Supplementary Table S1.

Correlation analysis

To further explore the clinical relevance of the identified connectivity abnormalities, this study collected two clinical scales from each stroke patient: Fugl-Meyer Assessment for Upper Extremity (FMA-UE) and the Modified Barthel Index (MBI). The FMA-UE was used to assess upper-limb motor impairment, and the MBI to evaluate activities of daily living. The scoring details of the MBI used in this study are provided in Supplementary File 1. This version adopts a non‑uniform scoring scheme (some items allow non‑multiples of 5, such as 1, 3, 4, 8, or 12 points), so the total score is not necessarily a multiple of 5.

Before correlation analysis, we assessed the normality of both the independent variables (EEG feature values) and the dependent variables (FMA-UE scores and MBI scores) using the Lilliefors test. The FMA-UE and MBI scores were approximately normally distributed (P > 0.05). For each EEG feature entered into the correlation analysis (i.e., those showing significant intergroup differences), normality was also verified (all P > 0.05). As all variables satisfied the normality assumption, Pearson correlation coefficients (r) were calculated to evaluate linear relationships between patients’ brain network feature values and their clinical scores (FMA-UE and MBI).

If either variable violated the normality assumption in future analyses, Spearman’s rank correlation coefficient would be used as a non-parametric alternative.

A “difference-correlation” cross-validation strategy was employed: only features that showed significant intergroup differences (patient vs. control) were entered into the correlation analysis. Within this subset, features demonstrating significant correlations with either FMA-UE or MBI were identified as potential biomarkers—those that were not only pathology-sensitive but also closely related to clinical function (upper limb motor function for FMA‑UE, and daily independence for MBI).

Consistent with the exploratory nature of this study as described in Sect.  “Statistical analysis”, no correction for multiple comparisons was applied to the correlation analyses to avoid inflating Type II error. The total number of correlation tests performed equals the number of features that show significant intergroup differences. All uncorrected P-values are reported, and the findings should be interpreted as preliminary and warrant validation in future confirmatory studies.

Results

COH results

Intergroup comparison of resting-state COH revealed a marked bidirectional enhancement in Coherence between the contralesional parietal cortex (cPC) and the contralesional motor cortex (cMC) in the patient group (P = 0.042). See Table 2; Fig. 3.

Table 2.

Functional connectivity pairs with significant differences

Functional Connectivity Pairs	Control group	Stroke group	P value
COH
cMC-cPC	0.460 ± 0.126	0.507 ± 0.136	0.042*
PLI
iMC-iPC	0.230 ± 0.075	0.260 ± 0.081	0.049*
iMC-cMC	0.266 ± 0.080	0.223 ± 0.060	0.028*
GC
iPFC→iMC	0.043 ± 0.014	0.034 ± 0.013	0.030*
iPFC→iSC	0.041 ± 0.015	0.033 ± 0.013	0.048*
iSC→iPFC	0.050 ± 0.029	0.035 ± 0.018	0.041*
iPFC→cMC	0.051 ± 0.022	0.038 ± 0.014	0.026*
iMC→iSC	0.044 ± 0.013	0.036 ± 0.012	0.041*
cMC→iMC	0.057 ± 0.025	0.042 ± 0.013	0.016*
iPC→iSC	0.052 ± 0.018	0.042 ± 0.013	0.027*
cMC→iSC	0.045 ± 0.014	0.036 ± 0.012	0.021*
cPC→iPC	0.050 ± 0.025	0.037 ± 0.013	0.036*
cMC→cPFC	0.060 ± 0.040	0.036 ± 0.018	0.015*

Data are presented as mean ± SD

*P<0.05

Fig. 3.

Resting‑state coherence (COH) differences between stroke patients and healthy controls. Red lines indicate connections with significantly enhanced Coherence in the patient group (P < 0.05). Brain regions are color‑coded as follows and labeled with their hemispheric laterality relative to the lesion side (ipsilesional [I] = left hemisphere; contralesional [C] = right hemisphere): PFC (prefrontal cortex) – yellow; MC (motor cortex) – purple; SC (sensory cortex) – blue; PC (parietal cortex) – yellow. Line thickness represents the strength of connectivity (coherence magnitude)

PLI results

Analysis of the resting-state PLI revealed the following findings: a bidirectional increase in functional connectivity between the ipsilesional parietal cortex (iPC) and the ipsilesional motor cortex (iMC) (P = 0.049), and a significant decrease in bidirectional phase synchronization between the ipsilesional motor cortex (iMC) and the contralesional motor cortex (cMC) (P = 0.028). See Table 2; Fig. 4.

Fig. 4.

Resting‑state phase lag index (PLI) differences between stroke patients and healthy controls. Red lines indicate connections with significantly enhanced phase synchronization in the patient group (P < 0.05); blue lines indicate connections with significantly reduced phase synchronization (P < 0.05). Brain regions are color‑coded as follows and labeled with their hemispheric laterality relative to the lesion side (ipsilesional [I] = left hemisphere; contralesional [C] = right hemisphere): PFC (prefrontal cortex) – green; MC (motor cortex) – purple; SC (sensory cortex) – blue; PC (parietal cortex) – yellow. Line thickness represents the strength of connectivity (magnitude of PLI)

GC results

Intergroup analysis of resting-state GC revealed widespread reductions in directed connectivity in stroke patients compared to healthy controls. The most prominent impairments involved pathways originating from or projecting to the core sensorimotor regions. Specifically:

Cross-hemispheric motor output from the contralesional motor cortex (cMC) was severely disrupted, with significantly weakened information flow to the ipsilesional motor cortex (GC_cMC→iMC, P = 0.016), ipsilesional sensory cortex (GC_cMC→iSC, P = 0.021), and contralesional prefrontal cortex (GC_cMC→cPFC, P = 0.015).

Intra-hemispheric sensorimotor integration within the ipsilesional hemisphere was extensively impaired, including reduced information flow from the ipsilesional motor cortex to sensory cortex (GC_iMC→iSC, P = 0.041), from the ipsilesional parietal cortex to sensory cortex (GC_iPC→iSC, P = 0.048), and bidirectional weakening between the ipsilesional sensory cortex and prefrontal cortex (GC_iSC→iPFC, P = 0.041; GC_iPFC→iSC, P = 0.027).

Additional significant reductions were observed in pathways from the ipsilesional prefrontal cortex to bilateral motor cortices (GC_iPFC→iMC, P = 0.030; GC_iPFC→cMC, P = 0.026) and from the contralesional parietal cortex to the ipsilesional parietal cortex (GC_cPC→iPC, P = 0.036). See Table 2; Fig. 5 for detailed results.

Fig. 5.

Resting‑state Granger causality (GC) differences between stroke patients and healthy controls. Blue lines indicate connections with significantly weakened information flow in the patient group (P < 0.05). Brain regions are color‑coded as follows and labeled with their hemispheric laterality relative to the lesion side (ipsilesional [I] = left hemisphere; contralesional [C] = right hemisphere): PFC (prefrontal cortex) – green; MC (motor cortex) – purple; SC (sensory cortex) – blue; PC (parietal cortex) – yellow. Line thickness represents the strength of information flow (magnitude of GC)

A complete list of all COH, PLI, and GC connections, including detailed statistics and P-values, is provided in Supplementary Table S1 and Supplementary Figure S1.

Correlation results

Correlation analysis revealed that specific connectivity alterations were significantly associated with clinical impairments. Reduced phase synchronization between the ipsilesional and contralesional motor cortices (PLI_iMC-cMC) showed a significant negative correlation with upper limb motor function (FMA-UE: r = -0.47, P = 0.027; Fig. 6). Regarding daily living ability, two directed connections were significantly correlated with the MBI: the weakened ipsilesional motor-to-sensory pathway (GC_iMC→iSC; r = -0.61, P = 0.002) and the impaired cross-hemispheric motor regulation (GC_cMC→iMC; r = -0.45, P = 0.036) (Figs. 7 and 8).

Fig. 6.

Correlation between interhemispheric phase synchronization of motor cortices (PLI_iMC-cMC) and FMA-UE scores

Fig. 7.

Correlation between information flow strength from the ipsilesional motor cortex to the sensory cortex (GC_iMC→iSC) and MBI scores

Fig. 8.

Correlation between information flow strength from the contralesional motor cortex to the ipsilesional motor cortex (GC_cMC→iMC) and MBI scores

Core hubs

Resting-state EEG analysis revealed that functional connectivity reorganization in stroke patients was highly concentrated in three core regions: the bilateral motor cortices (cMC, iMC) and the ipsilesional sensory cortex (iSC).

The contralesional motor cortex (cMC) showed bidirectional coherence enhancement with the ipsilateral parietal cortex (COH_cPC-cMC↑) but significantly reduced cross-hemispheric output to the ipsilesional motor and sensory cortices (GC_cMC→iMC↓, GC_cMC→iSC↓; all P < 0.05). The ipsilesional motor cortex (iMC) exhibited enhanced phase synchronization with the ipsilateral parietal cortex (PLI_iPC-iMC↑) yet severely disrupted output to the ipsilesional sensory cortex (GC_iMC→iSC↓) and impaired bilateral synchronization with cMC (PLI_iMC-cMC↓; all P < 0.05). The ipsilesional sensory cortex (iSC) emerged as a widely impaired information input hub, with reduced inputs from multiple regions, including iMC, cMC, iPC, and bidirectional disruption with iPFC (all P < 0.05).

Clinical correlations reinforced the functional relevance of these core hubs (see Fig. 9 for a radar chart summarizing all significant connectivity changes).

Fig. 9.

Radar chart summarizing functional connectivity features with significant intergroup differences between stroke patients and healthy controls

This radar chart displays the normalized mean values (0–1) for all connectivity features that showed statistically significant differences between groups (P < 0.05). Blue lines represent stroke patients; red lines represent healthy controls. Each radial axis corresponds to one connectivity feature, labeled with the specific metric and brain regions involved. Brain region abbreviations follow the lesion‑based nomenclature: ipsilesional (I, left hemisphere) and contralesional (C, right hemisphere). For directed connections (GC), arrows on the chart indicate the direction of information flow. For undirected connections (COH and PLI), no arrows are shown. Values were normalized to the maximum value across both groups for each feature, allowing direct comparison of relative differences.

Discussion

This exploratory study, integrating EEG functional connectivity analyses (COH, PLI, GC), investigated resting-state brain network reorganization in stroke patients. The findings revealed several interrelated patterns, including altered interhemispheric connectivity, compensatory changes in the contralesional hemisphere, and disruptions in ipsilesional higher-order networks. These observations offer preliminary insights into potential neural mechanisms underlying post-stroke dysfunction, which warrant confirmation in larger, well-powered studies.

Impaired interhemispheric connectivity

We observed a significant reduction in Granger-causal information flow from the contralesional motor cortex (cMC) to the ipsilesional motor cortex (iMC) (GC_cMC→iMC), which was negatively correlated with the MBI score. This finding suggests that impaired regulatory input from the contralesional to the ipsilesional motor network may contribute to reduced functional independence. Concurrently, phase synchronization between the bilateral motor cortices (PLI_iMC-cMC) was significantly reduced and negatively correlated with the FMA-UE score, indicating a potential link between interhemispheric synchronization and motor coordination. It should be noted, however, that the data points in this correlation were somewhat sparse (Fig. 7), suggesting that the result may be influenced by individual variability or by the relatively small sample size. Therefore, while this finding supports a possible association between interhemispheric connectivity and motor impairment, it requires confirmation in larger cohorts. The underlying pathology may involve structural and functional alterations mediated by the corpus callosum [15].

These observations align with previous studies highlighting the complexity of post-stroke network reorganization. Research has shown that after subcortical stroke, microstructural damage to the ipsilesional corticospinal tract can lead to compensatory enhancement of functional connectivity between the bilateral primary motor cortices; however, this increase is negatively correlated with motor recovery [28]. This suggests that increased synchronization alone may not compensate for the loss of specific directed information flow.

Such impairment in directed interhemispheric connectivity may extend beyond the motor network. In this study’s sensory/attention network, directed connectivity from the contralesional to the ipsilesional parietal cortex (GC_cPC→iPC↓) was also significantly impaired, consistent with multiple independent studies [29, 30]. These findings raise the possibility that impaired contralesional→ipsilesional connectivity in the sensory-attention network could contribute to sensory deficits and exacerbate secondary motor symptoms by disrupting sensory-motor integration. Therefore, rehabilitation strategies might benefit from addressing both motor and sensory networks.

Compensatory reorganization in the contralesional hemisphere

This study revealed a paradoxical reorganization pattern in the contralesional hemisphere, characterized by enhanced local sensorimotor network synchronization alongside impaired motor-cognitive regulatory pathways.

Specifically, we observed enhanced local sensorimotor network synchronization within the contralesional hemisphere. COH analysis showed a significant increase in bidirectional synchronization between the contralesional parietal cortex (cPC) and the contralesional motor cortex (cMC) (COH_cPC-cMC↑), which may reflect a compensatory concentration of neural resources toward primary sensorimotor areas. This phenomenon is supported by multimodal evidence from previous studies [31–33], suggesting that enhanced local synchronization could facilitate sensorimotor integration and support basic motor function recovery.

However, this local compensatory enhancement was accompanied by impaired higher-order regulatory pathways. GC analysis revealed a substantial reduction in directed information flow from the contralesional motor cortex (cMC) to the ipsilateral prefrontal cortex (cPFC) (GC_cMC→cPFC↓), suggesting a potential disruption of regulatory communication from motor-execution regions to cognitive-control areas. This pattern is consistent with evidence from healthy individuals [34] and stroke patients [35, 36], suggesting that damage to this pathway may contribute to motor control deficits.

Impaired higher-order functions in the ipsilesional network

The ipsilesional hemisphere exhibited a complex reorganization pattern, with compensatory enhancement of local sensorimotor network synchronization alongside impaired directed connectivity in higher-order cognitive control pathways.

We observed compensatory enhancement of local phase synchronization in the sensorimotor network. PLI analysis showed a significant increase in bidirectional neural oscillation phase synchronization between the ipsilesional parietal cortex (iPC) and the ipsilesional motor cortex (iMC) (PLI_iPC-iMC↑). This enhanced local synchronization may represent an adaptive compensatory mechanism within the ipsilesional sensorimotor network, consistent with previous research [37, 38] suggesting that optimized local phase synchronization could support basic motor function recovery through improved sensory-motor integration.

However, this compensatory enhancement co-existed with damage to directed connectivity in higher-order cognitive control pathways. GC analysis revealed impairment in the ipsilesional higher-order control network, including reduced directed connectivity from the ipsilesional prefrontal cortex (iPFC) to iMC (GC_iPFC→iMC↓), suggesting disruption of cognitive control over motor execution. Additionally, directed regulation from impaired core regions (iMC/iPC) to the ipsilesional sensory cortex (iSC) was significantly weakened (GC_iMC→iSC↓, GC_iPC→iSC↓), indicating potential impairment in both motor-to-sensory and sensory integration pathways. These observations align with previous studies [39–43] highlighting the importance of these pathways for motor function.

A key observation of this exploratory study is the strong negative correlation between weakened directed connectivity from the ipsilesional motor cortex to the sensory cortex (GC_iMC→iSC) and the MBI (r = -0.61). This finding raises the hypothesis that the capacity for directed regulation of sensory input by the motor cortex may be an important neural mechanism supporting independence in daily activities post-stroke. This observation is supported by multi-level evidence [44–46] suggesting that the iMC→iSC pathway could play a central role in sensorimotor integration and functional recovery.

Prefrontal high-order integration deficits

This study observed that after stroke, the prefrontal cortex (PFC) exhibited altered directed information flow bilaterally. On the ipsilesional side, directed information flow from the ipsilesional prefrontal cortex (iPFC) to the ipsilesional motor cortex (iMC) and sensory cortex (iSC) (GC_iPFC→iMC↓, GC_iPFC→iSC↓) was significantly weakened, suggesting potential impairment of high-order cognitive control over motor execution and sensory processing. On the contralesional side, information flow from the contralesional motor cortex (cMC) to the contralesional prefrontal cortex (cPFC) (GC_cMC→cPFC↓) was also reduced, indicating possible disruption of feedback integration from motor execution areas. These observations [47] suggests that bilateral prefrontal involvement may be relevant to understanding complex task-execution deficits in stroke patients.

While ipsilesional local sensorimotor networks may achieve partial compensation for basic functions through enhanced phase synchronization (PLI↑), the widespread alterations in prefrontal directed connectivity (GC↓) observed in this study may limit adaptive network reorganization. Research evidence [43, 48] provides preliminary directions for investigating potential therapeutic approaches targeting these pathways.

In summary, the results of this exploratory study suggest that the bilateral motor cortices (cMC, iMC) and the ipsilesional sensory cortex (iSC) may serve as key hubs of network reorganization after stroke. The characteristic connectivity changes observed in these regions—including local compensatory enhancement, widespread disruption of directed information flow, and loss of interhemispheric synchronization—may collectively contribute to network dysfunction.

Methodologically, this study demonstrates the feasibility of integrating multiple EEG connectivity measures (COH, PLI, GC) to capture different aspects of network reorganization in resting-state stroke patients. The correlation between the ipsilesional motor-to-sensory pathway (GC_iMC→iSC) and functional independence (MBI, r = -0.61) generates a hypothesis for future investigations into the role of sensorimotor integration in post-stroke recovery.

Limitations

This exploratory study has several limitations that should be considered when interpreting the findings. First, the sample was not stratified by disease stage, precluding analysis of dynamic changes in cortical network reorganization throughout stroke recovery. Second, relying solely on resting-state EEG, without task-based paradigms, limits our ability to validate causal relationships between connectivity changes and real-time behaviors. Third, the EEG’s spatial resolution was insufficient to precisely distinguish the differential contributions of prefrontal subregions to the observed connectivity alterations. Fourth, the reported behavior-neural mechanism correlations are based on a relatively small sample (n = 22) and lack correction for multiple comparisons, increasing the risk of Type I error and underscoring the exploratory nature of these findings. The clinical implications of the identified pathways were inferred solely from statistical correlations, highlighting the need for cross-validation with multimodal assessments in larger, independent cohorts. Fifth, the broadband COH estimates (0.5–80 Hz) may be partially biased by the 1/f spectral slope of EEG, a well‑known confound for coherence measures. Although this does not affect our narrowband PLI results, future studies should consider frequency‑resolved methods (e.g., multitaper) or spectral whitening to disentangle true connectivity from 1/f‑driven artifacts.

We also computed wPLI and iCOH as more conservative connectivity measures; however, no significant group differences were observed (Supplementary Table S2). This may reflect the exploratory nature of our study and the relatively small sample size, and does not diminish the validity of our primary findings obtained with well-established PLI and COH metrics, though these findings should be considered preliminary.

Conclusion

This exploratory EEG study investigated resting-state network reorganization in stroke patients using integrated connectivity analyses (COH/PLI/GC). The findings suggest a complex pattern of post-stroke reorganization characterized by local compensatory enhancement alongside alterations in directed regulation. The ipsilesional motor-to-sensory pathway (iMC→iSC) emerged as a potential candidate for functional recovery, given its strong correlation with the MBI. These observations suggest that hierarchical intervention strategies—prioritizing sensorimotor integration followed by modulation of the prefrontal regulatory network—might warrant investigation in future studies. However, given the exploratory nature of this work and the methodological limitations discussed, these findings require confirmation in larger, well-powered studies that employ multiple-comparison correction and task-based validation. Future research should aim to elucidate the dynamic mechanisms of these pathways and investigate whether neuromodulation techniques can selectively target directed connectivity to promote recovery.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

Xie Qiurong: Writing – original draft, Software, Methodology, Formal analysis, Conceptualization, Project administration, Resources, Funding acquisition.Zheng Nan: Writing – review & editing, Software, Methodology, Visualization, Formal analysis. Zhang Xuan: Writing – review & editing, Software, Methodology, Visualization, Formal analysis. Shi Wuxiang: Writing – review & editing, Methodology.Li Yurong: Writing – review & editing, Methodology, Funding acquisition. Wang Xiaoling: Writing – review & editing, Methodology, Funding acquisition.

Funding

This work was supported by the National Natural Science Foundation of China (Grant Nos. 82305357 and 62373108), the Fujian Provincial Natural Science Foundation (Grant Nos. 2025J01948 and 2023J01153) and the Science and Technology Development Project of the Chinese Association of Rehabilitation Medicine (Grant No. KFKT-2022-014). Please see Supplementary File 3.

Data availability

The datasets generated and analyzed during the current study are not publicly available but can be obtained from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the Helsinki Declaration and its later amendments, or with comparable ethical standards. Ethical approval was granted by the Ethics Committee of the Rehabilitation Hospital Affiliated with Fujian University of Traditional Chinese Medicine (Ethics Approval No.: 2023KY-048-002). Informed consent was obtained from all individual participants included in the study. Please see Supplementary File 2.

Consent for publication

All participants provided consent for the publication of their anonymized data.

Competing interests

The authors declare no competing interests.