EEG Alpha Rhythms and Brain Network Flexibility Predict Stroke Recovery

Stroke remains one of the world’s leading causes of long-term disability, and for the millions of survivors who live with persistent weakness or paralysis, one of the most frustrating uncertainties is prognosis: how much function will return, and how quickly? A new study published in Frontiers in Stroke suggests that the answer may be written into the brain’s own electrical activity. By analyzing resting-state electroencephalography (EEG) recordings from stroke patients at one week and again at three months after their strokes, researchers have identified a set of electrophysiological biomarkers that together explain more than 78 percent of the variability in motor recovery—a striking figure for a field where prognostic tools remain crude and imprecise.

The research team, led by Majid Khalili-Ardali and Vivek Sharma of the Donders Institute for Brain, Cognition and Behaviour at Radboud University in the Netherlands, alongside colleagues at the University of Amsterdam, set out to bridge a persistent gap in stroke neuroscience. While it has long been recognized that the brain at rest is far from idle—displaying intricate, ever-changing patterns of spontaneous activity that predict behavior and reflect the intrinsic dynamics of large-scale neural networks—few studies have integrated the multiple layers of EEG information into a single framework capable of tracking recovery. Motor impairments, the most common consequence of stroke, affect up to 85 percent of patients in the acute phase, and understanding how the injured brain reorganizes its network dynamics could transform how rehabilitation is planned and personalized.

The study analyzed data from 17 patients with first-ever ischemic strokes in the territory of the middle and/or anterior cerebral artery, drawn from an original cohort of 20 after three were excluded for missing follow-up scores. The cohort spanned a wide clinical spectrum: ages ranged from 37 to 79 years, stroke severity on the National Institutes of Health Stroke Scale ranged from 3 to 27, and lesion side was roughly balanced between the two hemispheres. EEG was recorded using a 128-channel Biosemi ActiveTwo system at a sampling rate of 512 Hz while participants rested awake with their eyes closed, at one week post-stroke and again at three months. Motor recovery was quantified using change in the Fugl-Meyer scale, a standardized clinical measure of motor function, normalized to the maximum achievable score and calculated as the difference between the two sessions.

The analytical pipeline was rigorous and technically sophisticated. Preprocessing in MATLAB using the FieldTrip toolbox involved zero-phase Butterworth filtering between 0.5 and 45 Hz, notch filtering at 50 Hz to remove line noise, visual and metric-based rejection of noisy segments, and common average referencing. Critically, the team applied a technique called Fitting Oscillations and One Over F (FOOOF) to the power spectral densities, separating the true oscillatory component of the signal from the aperiodic 1/f background activity that otherwise confounds band-power measures. Only the periodic, oscillatory component was carried forward into subsequent analyses of delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), and beta (13–30 Hz) bands.

The study’s headline finding concerns the alpha band. Change in relative alpha power—the proportion of total spectral power falling within the 8–13 Hz range, averaged across all channels—showed a robust positive correlation with change in contralesional Fugl-Meyer score, with a Pearson correlation coefficient of r = 0.72 (p = 0.00099). This single spectral measure accounted for 52.57 percent of the variance in motor recovery on its own. Notably, the maximum shift in alpha peak frequency, a measure traditionally proposed as a marker of cortical reorganization, showed no significant relationship with improvement (r = 0.09, p = 0.71), suggesting that it is the balance of power across frequency bands rather than the precise frequency of the alpha peak that tracks functional gains.

Interpreting this alpha finding requires care, because it appears at first glance to contradict prior reports that higher alpha at a single time point portends worse function. The authors resolve this apparent conflict by emphasizing that their metric captured longitudinal change rather than absolute level. The acute stroke literature emphasizes global slowing—elevated delta and theta power alongside reduced alpha and beta—rather than excess alpha, with indices such as the delta-to-alpha ratio well established as markers of stroke severity. The observed increase in relative alpha power over the subacute period is therefore best understood as a normalization of spectral content away from the slowed, low-frequency profile characteristic of the acute phase, consistent with prior longitudinal work showing that patients who shift from slow-wave toward alpha-beta-dominant activity during rehabilitation achieve greater motor recovery.

Beyond spectral measures, the study examined two measures of large-scale network dynamics derived from the Kuramoto Order Parameter (KOP), a well-established index of global phase synchrony borrowed from the physics of coupled oscillators. After band-pass filtering the EEG into the four standard bands, applying Gram-Schmidt orthogonalization to reduce spatial leakage between electrodes, and extracting instantaneous phase via the Hilbert transform, the team computed the mean KOP over time as a measure of global synchrony and its standard deviation as a measure of metastability—the brain’s capacity to transition between transiently stable states while maintaining dynamic flexibility. In the alpha band, metastability correlated with recovery at r = 0.438 and synchrony at r = 0.417; in the theta band, the corresponding values were r = 0.356 and r = 0.418. While none of these correlations individually reached statistical significance at the α = 0.05 threshold in this modest sample, the consistent positive trends in the lower frequency bands suggested that enhanced large-scale oscillatory coordination favors better outcomes. The beta band, whose broad relative bandwidth makes its instantaneous phase less reliably interpretable, was excluded from the composite network index on both statistical and methodological grounds.

The third analytical pillar was EEG microstate analysis, which identifies quasi-stable topographical patterns in scalp electrical activity lasting roughly 40 to 120 milliseconds, thought to represent the building blocks of spontaneous cognition and the transient states of large-scale neural networks. Using a modified k-means clustering algorithm based on topographical dissimilarity, the team determined six optimal microstate classes per subject and session via the silhouette score, then derived six metrics from the resulting state sequences: mean duration, Hurst exponent, Lempel-Ziv-inspired complexity, global field power peak frequency, global explained variance, and the state transition probability matrix, with statistical significance assessed against a null distribution generated from 1,000 permutations and Bonferroni-corrected across the 36 possible transitions. None of the individual microstate measures correlated significantly with motor recovery—mean duration showed r = 0.09, complexity r = 0.05, GFP peaks r = −0.18, and global explained variance r = −0.15—and the average graph density of significant transitions, which increased from 0.65 in Session 1 to 0.70 in Session 2, showed no meaningful relationship with improvement (r = 0.079, p = 0.76).

The study’s most compelling result emerged when these three domains were integrated into a hierarchical regression model. Relative alpha power change alone explained 52.57 percent of the variance in motor recovery. Adding the network dynamics composite—standardized and averaged changes in synchrony and metastability across the alpha and delta bands—significantly improved the model, raising explained variance to 78.61 percent (ΔR² = 0.2604, F(1,14) = 17.04, p = 0.001). By contrast, adding the microstate composite produced only a negligible and non-significant increment (ΔR² = 0.0025, p = 0.700), indicating that microstate dynamics contributed no unique explanatory information once spectral and network-level features were accounted for. Multicollinearity diagnostics confirmed that these results were not artifacts of predictor overlap: variance inflation factors remained low for all three predictors, and the pairwise correlations among them were modest and non-significant.

The authors are careful to frame these findings as exploratory rather than definitive. The sample of 17 patients is small, limiting statistical power and generalizability, and the two-time-point design cannot disentangle baseline impairment, spontaneous recovery trajectory, and within-participant EEG-behavior coupling. The sequential Gram-Schmidt orthogonalization used to reduce spatial leakage before phase extraction is order-dependent and was originally validated for amplitude-based rather than phase-based connectivity measures, raising questions about the robustness of the KOP estimates. Moreover, the correlational nature of the study cannot establish whether these EEG signatures play a causal role in enabling recovery, whether they merely reflect the extent of underlying injury, or whether they are downstream consequences of the recovery process itself. The authors also note that microstate analysis was developed largely in healthy cohorts and in conditions with distributed rather than focal pathology; a focal stroke lesion alters local tissue conductivity and can distort scalp topography near the lesion boundary, potentially biasing microstate maps, and microstate classes are more strongly associated with attentional, salience, and self-referential networks than with motor circuits, which may partly explain their weak association with a purely motor outcome measure.

Nevertheless, the implications are substantial. If the findings replicate in larger cohorts with more frequent longitudinal assessments, a brief resting-state EEG recording could become a practical bedside tool for stratifying patients early after stroke, identifying those likely to benefit from intensive rehabilitation and guiding personalized treatment strategies. The unified framework—combining spectral power, network-level synchrony and metastability, and spatiotemporal microstate dynamics—offers a template for how multimodal EEG biomarkers might move from the research laboratory into clinical decision-making. For a condition that affects millions worldwide each year and leaves so many facing an uncertain road to recovery, the prospect of reading the brain’s own forecast of its healing capacity represents a genuinely transformative step forward in stroke medicine.

Subject of Research: EEG biomarkers of motor recovery after ischemic stroke

Article Title: Electrophysiological signature of stroke recovery: investigating EEG biomarkers for prognostic insights

Article References: Khalili-Ardali, M., Sharma, V., Mandahar, T. S., Páscoa dos Santos, F., Tiesinga, P., & Ramsey, N. (2026). Electrophysiological signature of stroke recovery: investigating EEG biomarkers for prognostic insights. Frontiers in Stroke, 5, Article 1894659. https://doi.org/10.3389/fstro.2026.1894659

Image Credits: AI Generated

DOI: 10.3389/fstro.2026.1894659

Keywords: stroke, EEG, alpha oscillations, motor recovery, metastability, synchrony, microstates, biomarkers, neurorehabilitation, Kuramoto Order Parameter, Fugl-Meyer scale, brain networks