Slower EEG alpha generation, synchronization and “flow”—possible biomarkers of cognitive impairment and neuropathology of minor stroke

Subjects

This research was conducted at the Clinic of Neurology of the Clinical Centre in Nis, Serbia. We included 21 subjects: 10 ischemic stroke patients and 11 healthy controls. One of the stroke patients died during the follow-up period. The clinical diagnosis of stroke and its classification were determined by a detailed patient history of symptoms, a neurological examination, and computed tomography (CT) and/or magnetic resonance imaging (MRI). Patients were assessed 9.70 ± 2.50 days following the onset of their stroke (the subacute stage for the first assessment), as well as 13.67 ± 5.85 months after their stroke (the chronic stage for the follow-up assessment). Neurological impairment was measured using the National Institute of Health Stroke Scale (NIHSS) on admission and at the time of both the first assessment and the follow-up assessment. Cognitive function was assessed by the Montreal Cognitive Assessment (MoCA) (Nasreddine et al., 2005) and the MoCA memory index score (MoCA-MIS) (Julayanont et al., 2014) both in the subacute stage and during the follow up. The Edinburgh handedness inventory-short form (EHI-SF) with four claims and a five-point response scale was used for assessing hand dominance (Veale, 2014).

The inclusion criteria for the minor stroke patients were the following: the first ever stroke, a confirmed ischemic stroke in the distribution of right middle cerebral artery (MCA), an NIHSS of four or less, the absence of aphasia, right hand dominance (EHI-SF score >  60) and the absence of other neurological or psychiatric diseases. The control subjects were volunteers with no history of neurological or psychiatric disorders who were matched by age and years of formal education with the stroke group. We used the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE)—short form (Jorm, 1994) to identify and exclude patients with pre-stroke dementia (IQCODE score >  3). The study was carried out in accordance with the Declaration of Helsinki (The Code of Ethics of the World Medical Association), and approved by the Ethical Committee of the Medical Faculty of the University of Nis, Republic of Serbia (No 12-9808-2/1). All the participants gave informed consent.

EEG data acquisition

EEG was recorded on the day of assessment, both in the subacute and the chronic stages of stroke, using a 19 channel NK-9100K EEG system (Nihon Kohden, Japan). We used standard EEG electrode placement and positioning according to the 10–20 International System (Fp1, Fp2, F7, F3, F4, F8, T3, C3, C4, T4, T5, P3, P4, T6, O1, and O2). Recordings were referenced to (C3+C4)/2, the physical reference of the NK-9100K EEG system, with the ground electrode placed on the forehead. After conventional amplification and filtering (high pass 0.1 Hz, notch filter 50 Hz) the analog signals were digitized (250/s), and recorded for approximately 15 min in a resting, awake state with eyes closed.

EEG data analysis

All the EEG recordings were visually inspected and discriminated using EEGLAB to eliminate any ocular, muscular, and other types of artefacts before further analysis. The final EEG signals, consisting of 10 min artefact-free periods, were analyzed in MATLAB R2011a, using software originally developed in MATLAB 6.5 (Ciric et al., 2015; Ciric et al., 2016; Kalauzi, Vuckovic & Bojić, 2012; Lazic et al., 2015; Lazic et al., 2017; Petrovic et al., 2013a; Petrovic et al., 2013b; Petrovic et al., 2014; Saponjic et al., 2013). We particularly analyzed the EEGs derived from the four lateral frontal (F3, F7, F4, F8), and corresponding lateral posterior (P3, P4, T5, T6) electrodes (Schleiger et al., 2014). After additional filtering (1–50 Hz band pass filter), we applied FFT algorithm to each individual recording, using a 10 s non-overlapping moving window resulting in 0.1 Hz frequency resolution (60 10 s Fourier epochs per each electrode), to calculate the relative EEG amplitudes of all the conventional frequency bands (δ = 1.1–4 Hz; θ = 4.1–8 Hz; α = 8.1–13 Hz; β = 13.1–30 Hz; γ = 30.1–50 Hz).

Following the concept of “individual alpha frequency” (Klimesch, 1999), for each EEG recording and each electrode we calculated the weighted average of the alpha frequency (αAVG), using the alpha amplitude as a weighting factor: ${\displaystyle \alpha AVG=\frac{\sum _f\left(Amp\left(f\right)xf\right)}{\sum _{f}Amp\left(f\right)}}$ where f is a specific frequency ranging across the alpha frequency band and Amp(f) is the amplitude at f (Hooper, 2005).

In addition, to analyze the EEG amplitude changes (the EEG microstructure), we calculated the group probability density distributions of all the conventional EEG frequency band relative amplitudes using the Probability Density Estimate (PDE) routine supplied with MATLAB R2011a. As in our previous studies, in order to eliminate any influence from absolute signal amplitude variations on the recordings, we calculated the relative Fourier amplitudes (Ciric et al., 2015; Ciric et al., 2016; Lazic et al., 2015; Lazic et al., 2017; Petrovic et al., 2013a; Petrovic et al., 2013b; Petrovic et al., 2014; Saponjic et al., 2013): ${\displaystyle {\left(RA\right)}_b=\frac{\sum _{b}Amp}{\sum _{tot}Amp},b=\left\{\delta ,\theta ,\alpha ,\beta ,\gamma \right\}.}$ For each electrode and each frequency band, PDE analysis was performed on the ensembles of relative amplitudes by pooling the measured values from all subjects belonging to the corresponding group (the control, the subacute stage of the stroke or the chronic stage of the stroke). For the statistical analysis of PDEs the relative amplitude means were calculated for each 2 min.

We also calculated the inter-hemispheric and intra-hemispheric coherences by using the “mscohere” routine of the MATLAB R2011a Signal Processing Toolbox, which computes the magnitude squared coherence between the signals x and y as: ${\displaystyle C_{\left(xy\right)}\left(f\right)=\frac{|P_{xy}\left(f\right){|}^2}{P_{xx}\left(f\right)P_{yy}\left(f\right)}}$ where $P_{xy}\left(f\right)$ is the cross spectrum of x and y, while $P_{xx}\left(f\right)$ and $P_{yy}\left(f\right)$ denote the power spectra of the two signals. For this purpose, the individual EEG signals, derived from the corresponding electrodes, were concatenated and pooled within each group. For each electrode pair, a coherence value of between 0 and 1 was calculated for every 10 s and each frequency point (at a 0.1 Hz frequency resolution), within the overall 0.1–50 Hz range. Then the values within each conventional frequency band were averaged for each spectrum, and their means were calculated from the collection of all the available spectra. We calculated the group mean coherence values for each 2 min, for each frequency band, each pair of electrodes, and each group.

We also analyzed the topographic distribution of the alpha carrier frequency phase potentials (PPs) through changes in their carrier frequency phase shifts (PSs; Kalauzi, Kesic & Saponjic, 2009; Kalauzi, Vuckovic & Bojić, 2012). For this particular analysis, the individual EEG signals, derived from all 19 electrodes, were re-referenced offline to the common average reference (the average of all 19 electrodes). To be specific, for each pair of channels, their EEG recordings were treated as signals with variable amplitudes, but with the same alpha frequency (model details can be found in Kalauzi, Vuckovic & Bojić, 2012, Appendix B). Since alpha frequency is expected to be different between individuals (“individual alpha frequency“, Klimesch, 1999), the alpha frequency range limits used for PP analysis were determined individually, based on the previously determined αAVG, following the natural limits of the alpha peak position within each spectrum. For each subject and each pair of electrodes we calculated an alpha carrier frequency PS and alpha carrier frequency PP derived from the corresponding Fourier component PSs. Finally, we calculated the group alpha carrier frequency PPs (for the control, the subacute stage of the stroke and the chronic stage of the stroke), and the group alpha carrier frequency PP differences (for the subacute stage of the stroke vs. the control, the chronic stage of the stroke vs. the control, and the chronic stage of stroke vs. the subacute stage of stroke) for each electrode. All the calculations (the alpha carrier frequency PSs, the alpha carrier frequency PPs, the group mean alpha carrier frequency PPs and the group alpha carrier frequency PP differences) were done using angular arithmetic (Kalauzi, Vuckovic & Bojić, 2012, Appendix A, C, E). Since alpha carrier frequency PP is an integral phase measure of a particular EEG channel in the alpha range, incorporating all alpha Fourier components, it is possible to compare and subtract carrier frequency PP values even if the αAVG frequency differs between groups, as was the case in this study. The final results of PP analysis are presented as circles over corresponding electrodes, with diameters (small or large) referring to the PP absolute values, and line types (solid or dashed) referring to the PP signs (positive or negative). To be clear, since the PP calculation is based on angular variables, small circles correspond to abs (PP) ≈0°, while large circles correspond to abs (PP) ≈ ± 180° (with a solid line for positive and a dashed line for negative PP values). Therefore, due to their angular nature, the large diameter circles represent similar PP values, regardless of the sign (e.g., PP =178° and PP = − 179° should be treated as close to each other since their actual difference is not 357°, but rather 3°). In this way all channels can be organized in a series (order) based on their PP values: those channels leading in phase to all others will have the highest (large positive) PP values, while the channels with the lowest (large negative) PP values would be the one following all the others (Kalauzi, Vuckovic & Bojić, 2012). Based on these results, we have also drawn a topographic pattern of “alpha flow” in the control group, and the subacute and chronic stage of the stroke, by following the phase potential gradient (from the highest to the lowest PP values).

Statistical analysis

All the values are presented as means ± SDs. The statistical analysis was done using the Kruskal–Wallis ANOVA with a post-hoc Mann–Whitney U two-tailed test, and Spearman’s correlation. In all cases the difference was considered significant for p ≤ 0.05.

Clinical and demographic assessments

Table 1 summarizes the relevant demographic and clinical data of the control group and the stroke patients. There were six male and four female subjects among the stroke patients and five male and six female subjects in the control group. The age difference between stroke patients (63.80 ± 7.00 in the subacute stage and 63.89 ± 6.83 in the chronic stage) and the healthy controls (60.64 ± 9.91) was not statistically significant (z ≥  − 0.78; p ≥ 0.44). There was no statistically significant difference in the years of education between the two groups (z =  − 1.52; p = 0.13). The average NIHSS was 8.20 ± 3.05 on admission, 2.50 ± 0.85 at the time of the first assessment, and 2.11 ± 0.78 at a time of the follow-up assessment.

All the patients in the stroke group had their stroke in the vascular territory of the right MCA. The average lesion volume measured by the ABC/2 method was 9.34 ± 6.49 cm³. The ischemic stroke lesion volumes and the affected structures for each patient are shown in Table 2.

As indicated in Table 1, MoCA scores in stroke patients during the subacute stage were significantly lower than in the healthy controls (20.80 ± 5.32 vs. 25.64 ± 3.20; z =  − 2.12; p = 0.03), in contrast with the chronic stage of the stroke where these MoCA score differences were not statistically significant in stroke patients vs. controls (23.44 ± 4.98 vs. 25.64 ± 3.20; z =  − 0.88; p = 0.38).

In addition, there was a significant difference between the subacute stroke patients and the control group concerning MoCA-MIS (Table 1, 8.80 ± 3.52 vs. 11.91 ± 2.98; z =  − 1.98; p = 0.05), whereas in the chronic stroke patients, the MoCA-MIS score returned to the control values (Table 1, 10.22 ± 3.35 vs. 11.91 ± 2.98; z =  − 1.15; p = 0.25).

EEG alpha activity following a minor stroke

For all the analyzed frontal and posterior EEGs, the group mean relative amplitude EEG spectra showed prominent amplitude peaks in the alpha frequency range in all groups, but with the alpha peaks shifted toward lower frequencies in stroke patients, particularly in the subacute stage (Fig. 1). Further analysis of αAVG frequency revealed a generalized decrease in αAVG following a stroke. In particular, and in contrast to the αAVG in the control group that was in the range of 10.38–10.54 Hz, the subacute stroke group had an αAVG in the range of 10.11–10.23 Hz, whereas the chronic stroke group had an αAVG in the range of 10.22–10.38 Hz.

A statistically significant slower alpha frequency in stroke patients vs. controls was demonstrated only for the subacute stage at F3, F4, F7, F8, P3 and P4 (χ² ≥ 6.11; p ≤ 0.05; z ≥  − 3.32; p ≤ 0.03), but with no change in the αAVG at T5 and T6 (χ² ≥ 3.09; p ≥ 0.18; z ≥  − 1.66; p ≥ 0.10). However, all the stroke induced alpha frequency changes returned to the control values (z ≥  − 1.90; p ≥ 0.06) in the chronic stage.

In our study we did not demonstrate the amplitude change of the slower alpha in subacute and chronic stroke patients vs. controls (χ² ≥ 1.44; p ≥ 0.14; z ≥  − 1.72; p ≥ 0.09; data not shown), but we have shown the delta, theta, and beta amplitude alterations (Fig. 2; χ² ≥ 6.61; p ≤ 0.04). The frontal cortex exhibited a general beta amplitude decrease in both subacute and chronic stroke patients vs. controls (Figs. 2I–2L, beta; χ² ≥ 15.21; p = 10⁻⁴; z ≥  − 4.08; p ≤ 0.01). This beta amplitude decrease was followed by a long-lasting theta amplitude increase across all the frontal and posterior channels (Figs. 2E–2H and 2Q–2T, theta; χ² ≥ 7.86; p ≤ 0.02; z ≥  − 5.64; p ≤ 0.03), with the exception of F8 and T5 where the theta amplitude returned to the control values in the chronic stage (z ≥  − 1.50; p ≥ 0.13). Delta amplitude alterations were in evidence only in the posterior channels (Figs. 2M–2P, delta; χ² ≥ 6.61; p ≤ 0.04): the delta amplitude increased at P3 in the chronic stage of stroke patients vs. controls (z =  − 2.42; p = 0.02) in contrast to a transient increase at T5 and T6 in the subacute stage (z ≥  − 2.90; p ≤ 0.01).

Inter-hemispheric and intra-hemispheric coherences following a minor stroke

Inter-hemispheric coherence analysis between the frontal electrodes, in both subacute and chronic stroke patients, revealed significantly increased coherences in the alpha frequency range in contrast to the control group (Figs. 3A and 3B; χ² ≥ 6.56; p ≤ 0.04; z ≥  − 3.24; p ≤ 0.02). This sustainably increased synchronization of slower alpha, caused by the stroke, was particularly expressed at the F3–F4 electrodes, and co-localized with the underlying neuropathology of the ischemic stroke (see Table 2). Conversely, the posterior electrodes expressed a generalized long-lasting desynchronization in all the frequency ranges (Figs. 3C and 3D; χ² ≥ 6.87; p ≤ 0.03; z ≥  − 4.82; p ≤ 0.04), but with no change in alpha coherences (χ² ≥ 2.59; p ≥ 0.14; z ≥ 10.63; p ≥ 0.10) in either subacute or chronic stroke patients vs. the controls.

In addition, in the chronic stage we demonstrated increased theta and beta synchronizations at the F3–F4 electrodes (Fig. 3A; χ² ≥ 7.67; p ≤ 0.02; z ≥  − 2.84; p ≤ 0.02) alongside broad spectrum (delta, theta, beta and gamma) desynchronization at the frontal F7–F8 electrodes (Fig. 3B; χ² ≥ 6.59; p ≤ 0.04; z ≥  − 3.00; p ≤ 0.05).

The increased synchronization of slower alpha in stroke patients vs. the controls was also shown within the ipsi-lesional hemisphere (Figs. 4D–4F; χ ² ≥ 6.06; p ≤ 0.05). We demonstrated sustainably increased alpha coherences across the F4–F8, F4–T6 and F4–P4 electrodes during both stages of stroke vs. the controls (z ≥  − 5.11; p ≤ 0.04). In addition, there were also inconsistent alterations in the delta, theta, beta and gamma coherences within the ipsi-lesional hemisphere in the subacute and chronic stages of stroke patients vs. the controls (Figs. 4D–4F; χ² ≥ 6.84; p ≤ 0.03; z ≥  − 4.24; p ≤ 0.02). Conversely, the contra-lesional hemisphere demonstrated an increased synchronization in the theta frequency range, but only in subacute stroke patients (Figs. 4A–4C; χ² ≥ 6.49; p ≤ 0.04; z ≥  − 4.51; p ≤ 0.05). This transient theta coherence augmentation was followed by inconsistent delta, alpha, beta and gamma coherence alterations in both the subacute and chronic stroke stages vs. the controls (Figs. 4A–4C; χ² ≥ 9.08; p ≤ 0.01; z ≥  − 4.86; p ≤ 0.03).

Topography of αAVG phase potentials following a minor stroke

The group mean phase potentials (PPs) of the newly emerged slower alpha in subacute stroke patients, and the alpha in the chronic stroke patients and healthy controls (Fig. 1) with their corresponding PP differences are depicted in Fig. 5. We have shown that the frontal cortex of the stroke patients vs. the controls expressed the greatest and long-lasting PP differences (Figs. 5A and 5B). This long-term impact of the stroke on the alpha PP differences was additionally confirmed by the almost non-existing PP differences between the subacute and chronic stage of the stroke patients (Fig. 5C). In addition, whereas the control group of patients showed relatively similar (large) alpha PPs (Figs. 5A and 5B, Control), the subacute stroke group of patients (Fig. 5A, Subacute Stroke) and the chronic stroke group of patients (Fig. 5B, Chronic Stroke) demonstrated a clear antero-posterior difference in alpha PPs, with the smallest frontal PPs vs. the largest posterior PPs. Consequently, the great majority of channels over the frontal cortex exhibited negative PP differences (dashed circles, Fig. 5A, Subacute Stroke vs. Control; Fig. 5B, Chronic Stroke vs. Control), meaning that the frontal alpha waves were more delayed in the stroke patients than in the controls compared to other channels. We identified the greatest decrease of PPs over the frontal cortex at F4 (Fig. 5A, Subacute Stroke vs. Control; Fig. 5B, Chronic Stroke vs. Control) and during both stages of stroke, which, as a sign of the largest delay in the newly emerged slower alpha, might be related to the confirmed neuropathology of an ischemic stroke.

Another way to present the group mean alpha PPs is to draw them as vectors in a unit circle (Figs. 6A–6C). In cases where the PPs of all channels are closely distributed in a unit circle, forming “a bundle”, this formation could be considered as having a “source” (the channel with the highest PP, where the alpha wave is first recorded) or “drain” (the channel with the lowest PP, where the alpha wave is last recorded). Since all the demonstrated alpha PPs were distributed across the whole circle, rather than being concentrated within “a bundle”, the unit circle could be considered only as a closed loop or no “source” or “drain” unit circle. Therefore, we have drawn a topographic pattern of the “alpha flow” for the control group (Fig. 6D), the subacute stroke group (Fig. 6E), and the chronic stroke group (Fig. 6F) by connecting the neighboring PP EEG channels with arrows following the descending PP gradient. The greatest alterations in the “alpha flow” in stroke patients vs. the controls were demonstrated within the frontal cortex (Figs. 6G–6I, indicated by solid red arrows), with F4 undergoing the most intensive PP change, leading to the reorganization of the PP channel sequence. To be specific, in control group the F4 PP neighbors were positioned longitudinally in the same ipsi-lateral hemisphere (Fig. 6G), whereas in the subacute (Fig. 6H) and chronic stage of the stroke patients (Fig. 6I) F4 was included in the newly established antero-posterior inter-hemispheric spiral PP pattern, indicating that the stroke had induced a sustained augmentation of inter-hemispheric interactions.

Correlations of the average EEG alpha frequency and cognitive functions

The results of correlation analysis between the MoCA and MoCA-MIS scores and the αAVG are presented in Table 3 and Fig. 7. We did not find a significant correlation between the MoCA and MoCA-MIS scores in the 11 control subjects and their αAVGs at any EEG channel (p ≥ 0.37).

While the MoCA scores in our 10 subacute stroke patients were significantly correlated with the αAVGs at F4 (ρ = 0.74, p = 0.02), P3 (ρ = 0.65, p = 0.04), P4 (ρ = 0.77, p = 0.01) and T6 (ρ = 0.74, p = 0.02), the MoCA-MIS scores were not significantly correlated with the αAVGs at any EEG channel (p ≥ 0.09).

Although the correlation between the MoCA scores and the αAVGs in our nine chronic stroke patients was not significant (they returned to the control correlation values, p ≥ 0.38), the MoCA-MIS became significantly correlated with their αAVGs at P3 (ρ = 0.72, p = 0.03), P4 (ρ = 0.92, p < 0.01), T5 (ρ = 0.77, p = 0.02) and T6 (ρ = 0.92, p < 0.01).