This study recruited 22 individuals after their initial unilateral hemispheric chronic stroke (age: 60.73 ± 7.00 years, 14 men and 8 women; time after stroke onset: 63.73 ± 45.74 months, 13 left hemispheric stroke and 9 right hemispheric stroke, 8 dominant-hand affected, no sign of cognitive impairment, that is, above 6/10 in the abbreviated mental test [12] via convenience sampling). Patients’ hemiparetic arm functions were assessed using the FMA-UE [13] and the action research arm test (ARAT) [14]. Patients’ clinical information related to stroke was collected by reviewing their discharge summaries and/or radiological reports from local hospitals. Table 1 presents patients’ demographic and clinical information. These participants were part of a randomized controlled trial published elsewhere [15]. The current study utilized their resting-state EEG data and analyzed their correlation with clinical scales at baseline (before any intervention was initiated). The human ethics subcommittee of the Hong Kong Polytechnic University approved the study protocol (reference number: HSEARS 20190718003).

In addition, EEG data from 19 age-matched healthy adults (Age: 61.32 ± 11.18, all right-hand dominant, 11 mean and 8 women) and 16 younger-age healthy adults served as the controls (Age: 26.00 ± 1.83 years, all right-hand dominant, eight men and eight women,). We included the data from healthy controls in two different age ranges, in order to rule out any possible age-related neurophysiological alterations apart from those induced by stroke. These healthy individuals participated in other two experimental studies published elsewhere [16] (Reference numbers: HSEARS20180120003; HSEARS20121012008). The participants in this study provided written informed consent before their participation. The resting-state EEG data from patients with stroke and healthy adults at baseline have not been previously published. The current study was a secondary analysis of data generated from previous studies and all these studies were conducted in accordance with the Declaration of Helsinki.

Our EEG data were recorded using a 64-channel cap (Quik-Cap, Compumedics Neuroscan, USA), an EEG Amplifier (SynAmps RT 64-channel Amplifier, Compumedics Neuroscan, USA), and Neuroscan Curry 7 software (Compumedics Neuroscan, USA). Electrode impedance was maintained below 10 kOhm, and the signal was sampled at 1024 Hz. Resting-state EEG was recorded for approximately 3-min with participants seated upright in an electromagnetically shielded EEG chamber with their eyes closed. The participants were instructed to minimize head and body movements during the recording.

The raw EEG signals were downsampled to 250 Hz. Signals with considerable movement artifacts were initially rejected via visual inspection. We filtered the data using band-pass filters within the frequency range of 1–40 Hz. Subsequently, an independent component analysis was utilized to remove any ocular components. The EEG data were referenced to a common average. We defined 26 channels on the left hemisphere (FP1, AF3, F7, F5, F3, F1, FT7, FC5, FC3, FC1, T7, C5, C3, C1, TP7, CP5, CP3, CP1, P7, P5, P3, P1, PO7, PO5, PO3, and O1) and other 26 channels on the right hemisphere (FP2, AF4, F2, F4, F6, F8, FC2, FC4, FC6, FT8, C2, C4, C6, T8, CP2, CP4, CP6, TP8, P2, P4, P6, P8, PO4, PO6, PO8, and O2) to explore the hemispheric effect associated with stroke-induced lesions on cortical EEG rhythms and network features [17]. The data from patients with right-hemispheric stroke were flipped for the convenience of visualization and data analysis.

Spectral power values within bins of 0.5 Hz were calculated using Fourier decomposition of data epochs with the mtmfft method. The power values within the delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), beta-1 (12–16 Hz), and beta-2 (16–30 Hz) bands [16] were normalized to the relative percentage of the average power over the five bands for each channel. The calculations were performed using a custom-made MATLAB script modified based on a tutorial of the Fieldtrip toolbox.

(https://www.fieldtriptoolbox.org/workshop/madrid2019/tutorial_stats/).

The hemispheric asymmetry was calculated based on the normalized powers over the ipsilesional and contralesional hemispheres using the following formula [18]:

$$Asymmetric \,\, index= \frac_-_}__}$$

Based on the Hilbert transform, the weighted phase lag index (wPLI) was computed for each frequency band. The wPLI is insensitive to volume conduction, possesses stronger statistical power to detect changes in phase synchronization, and is less affected by uncorrelated noise sources relative to the phase lag index [19]. The wPLI values range from 0 to 1, where a higher value indicates a stronger interregional coupling of neural oscillations and vice versa.

We selected the following three graph theory-based network metrics for this study: (1) node strength, (2) clustering coefficient, and (3) global efficiency. The network metrics were calculated using the brain connectivity toolbox (www.brain-connectivity-toolbox.net) [8]. Density-based thresholding was initially applied to remove spurious connections. The 60-by-60 connectivity matrices were thresholded to retain between 50 to 5% (in 5% increments) of the largest wPLI values [20]. Subsequently, we computed the following network measures for each threshold:

Node strength was estimated as the sum of the edge weights connected to the channels.

where i indicates a specific sensor and j represents the other sensors (j = 1,..., n–1, n = the number of channels). The strength of each channel was computed by summing the strength values of all the connecting channels.

The Clustering coefficient was the geometric mean of all triangles associated with each channel, evaluating the functional segregation. First, the values of the neighboring triangles of the channels were calculated using the following:

$$_= \frac\sum__\times _\times _)}^\frac$$

where N represents all the channels, and j and h are all possible pairs of neighboring sensors that create triangles with a specific channel. Subsequently, the clustering coefficient for each channel was computed as follows:

$$_= \frac\sum_\frac_}_(_-1)}$$

where N is the number of sensors and ki is the number of all connected channels for a specific channel.

Global efficiency was a measure of functional integration and was calculated as the average of the inverse shortest path length using the following:

where lij is the shortest path length from node j to node i, computed using Dijkstra’s algorithm.

Statistical analyses were performed with SPSS 22.0. Initially, we compared the between-group differences of the hemispheric asymmetry using one-way analysis of variance (ANOVA) with post hoc pairwise Bonferroni-corrected comparisons. Levene’s test for equality of variances was used to test the homogeneity of variances between both groups. When the homogeneity of variance was violated, the degrees of freedom were adjusted using the Welch–Satterthwaite method. Subsequently, we compared the within-hemisphere differences using paired t-tests for the EEG rhythmic powers over different frequency bands, separately for each group. Following previous practice, regional power was defined as the mean power over adjacent electrodes within the following areas: frontal (left side: FP1, AF3, AF7, F1, F3, F5, and F7; right side: FP2, AF4, AF8, F2, F4, F6, and F8), central (left side: FT7, FC5, FC3, FC1, C5, C3, C1, CP5, CP3, CP1, T7, and TP7; right side: FT8, FC2, FC4, FC6, C2, C4, C6, T8, CP2, CP4, CP6, and TP8), and posterior (left side: P7, P5, P3, P1, PO7, PO5, PO3, and O1; right side: P2, P4, P6, P8, PO4, PO6, PO8, and O2) [17]. Finally, the potential correlation between EEG metrics and upper limb clinical scores was explored using Pearson’s correlation in the group of stroke patients. Statistical significance was set at p < 0.05 (two-tailed).

Regarding the network outcomes, we initially compared the within-hemisphere differences between the two network measures (node strength and clustering coefficient) under every threshold using paired t-tests for the network measures, separately for each group. Subsequently, the areas under the curve (AUC) were calculated by integrating the measures across the entire threshold range. All 60 channels were included to measure global efficiency, while the channels in the ipsilesional and contralesional hemispheres were computed separately for the other two measurements (i.e., node strength and clustering coefficient). Multiple t tests with Bonferroni corrections were performed to explore the differences in the AUC between the ipsilesional and contralesional hemispheres in patients with stroke, as well as the two control groups. Finally, the potential correlation between the AUC of the network measures and the upper limb clinical outcomes was explored in the group of stroke patients using Pearson’s correlation. Statistical significance was set at p < 0.05 (two-tailed).