Background Lesion location is a major source of post-stroke neurophysiological heterogeneity, yet most electroencephalography (EEG) studies analyze patients as a single group, limiting lesion-specific biomarkers and translation. We proposed a lesion-centric, multi-scale EEG framework integrating local oscillations, inter-regional connectivity, and hemispheric asymmetry with machine learning to characterize and decode basal ganglia (P1), fronto-temporal/centrum semiovale (P2), and brainstem (P3) lesions. Methods Five-minute eyes-open, 128-channel resting EEG (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(1\,\text {kHz}\)</EquationSource> </InlineEquation>) was recorded in 57 subacute stroke patients (P1 = 22, P2 = 18, P3 = 17) and 22 matched controls. From artifact-minimized <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(90\,\text {s}\)</EquationSource> </InlineEquation> segments, ROI-averaged power spectral density (PSD) (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\theta \)</EquationSource> </InlineEquation>: 4–<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(7\,\text {Hz}\)</EquationSource> </InlineEquation>; <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\alpha \)</EquationSource> </InlineEquation>: 7–<InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(12\,\text {Hz}\)</EquationSource> </InlineEquation>; <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(\beta _{1}\)</EquationSource> </InlineEquation>: 12–<InlineEquation ID="IEq8"> <EquationSource Format="TEX">\(16\,\text {Hz}\)</EquationSource> </InlineEquation>; peak <InlineEquation ID="IEq9"> <EquationSource Format="TEX">\(\alpha \)</EquationSource> </InlineEquation> frequency), current source density (CSD)-based magnitude-squared coherence, and directional BSI (dirBSI) were computed. Between-group and subgroup differences were assessed using t-tests/Wilcoxon and ANOVA/Kruskal–Wallis with Benjamini–Hochberg FDR (<InlineEquation ID="IEq10"> <EquationSource Format="TEX">\(q=0.05\)</EquationSource> </InlineEquation>). EEG–behavior associations were examined with Spearman correlations. For machine learning, common spatial patterns (CSP) features were classified using linear discriminant analysis (LDA) with leave-one-subject-out cross-validation. To align with clinical workflow, we report HC vs P as “stroke detection/screening” and patient-only P1/P2/P3 classification as “lesion subtype decoding for stratification” (along with pairwise P1 vs P2, P1 vs P3, and P2 vs P3 models). An EEGNet baseline was evaluated for comparison. Results Increased <InlineEquation ID="IEq11"> <EquationSource Format="TEX">\(\alpha \)</EquationSource> </InlineEquation> power and a leftward peak <InlineEquation ID="IEq12"> <EquationSource Format="TEX">\(\alpha \)</EquationSource> </InlineEquation> shift were observed in patients (HC: <InlineEquation ID="IEq13"> <EquationSource Format="TEX">\(9.93 \pm 1.09\,\text {Hz}\)</EquationSource> </InlineEquation>; P: <InlineEquation ID="IEq14"> <EquationSource Format="TEX">\(8.75 \pm 1.02\,\text {Hz}\)</EquationSource> </InlineEquation>; <InlineEquation ID="IEq15"> <EquationSource Format="TEX">\(p = 6.82 \times 10^{-5}\)</EquationSource> </InlineEquation>). Pre-FDR, <InlineEquation ID="IEq16"> <EquationSource Format="TEX">\(\theta \)</EquationSource> </InlineEquation>-band frontal–motor connectivity was strengthened, while posterior P–O connectivity in <InlineEquation ID="IEq17"> <EquationSource Format="TEX">\(\alpha \)</EquationSource> </InlineEquation>/<InlineEquation ID="IEq18"> <EquationSource Format="TEX">\(\beta _{1}\)</EquationSource> </InlineEquation> was weakened. Ipsilesional dominance in <InlineEquation ID="IEq19"> <EquationSource Format="TEX">\(\theta \)</EquationSource> </InlineEquation> was indicated by dirBSI (HC: <InlineEquation ID="IEq20"> <EquationSource Format="TEX">\(-0.026 \pm 0.101\)</EquationSource> </InlineEquation>; P: <InlineEquation ID="IEq21"> <EquationSource Format="TEX">\(0.061 \pm 0.122\)</EquationSource> </InlineEquation>; <InlineEquation ID="IEq22"> <EquationSource Format="TEX">\(q=0.012\)</EquationSource> </InlineEquation>). Across lesions, <InlineEquation ID="IEq23"> <EquationSource Format="TEX">\(\beta _{1}\)</EquationSource> </InlineEquation> power differences in central/parietal/occipital ROIs were detected pre-FDR, with higher parietal <InlineEquation ID="IEq24"> <EquationSource Format="TEX">\(\beta _{1}\)</EquationSource> </InlineEquation> in P3; <InlineEquation ID="IEq25"> <EquationSource Format="TEX">\(\alpha \)</EquationSource> </InlineEquation>-band asymmetry was stronger in P1/P2 and more symmetric in P3 (<InlineEquation ID="IEq26"> <EquationSource Format="TEX">\(q=0.028\)</EquationSource> </InlineEquation>). EEG–behavior correlations did not survive FDR. Using CSP+LDA, accuracies of 92.41% (HC vs P), 94.87% (P1 vs P3), 85.71% (P2 vs P3), and 82.50% (P1 vs P2) were achieved; all binary AUCs exceeded 0.85; three-class accuracy reached 85.96%. Conclusion This multi-scale EEG framework identifies lesion-associated neurophysiological signatures and demonstrates feasible lesion subtype decoding, supporting the potential of EEG biomarkers for objective stratification and precision neurorehabilitation.

Lesion-specific EEG signatures in stroke: a multi-scale framework integrating oscillations, connectivity, and asymmetry for machine learning decoding

期刊论文
出版模式：

开放获取
发表日期：

2026年3月31日

查看全文

Wanting Liu,
Conghui Wei,
Yajing Lan,
Xumiao Peng,
Chenyuan Zhai,
Zhen Liang,
Li Zhang,
Yan Gong,
Gan Huang

摘要

Background

Lesion location is a major source of post-stroke neurophysiological heterogeneity, yet most electroencephalography (EEG) studies analyze patients as a single group, limiting lesion-specific biomarkers and translation. We proposed a lesion-centric, multi-scale EEG framework integrating local oscillations, inter-regional connectivity, and hemispheric asymmetry with machine learning to characterize and decode basal ganglia (P1), fronto-temporal/centrum semiovale (P2), and brainstem (P3) lesions.

Methods

Five-minute eyes-open, 128-channel resting EEG ( \(1\,\text {kHz}\) ) was recorded in 57 subacute stroke patients (P1 = 22, P2 = 18, P3 = 17) and 22 matched controls. From artifact-minimized \(90\,\text {s}\) segments, ROI-averaged power spectral density (PSD) ( \(\theta \) : 4– \(7\,\text {Hz}\) ; \(\alpha \) : 7– \(12\,\text {Hz}\) ; \(\beta _{1}\) : 12– \(16\,\text {Hz}\) ; peak \(\alpha \) frequency), current source density (CSD)-based magnitude-squared coherence, and directional BSI (dirBSI) were computed. Between-group and subgroup differences were assessed using t-tests/Wilcoxon and ANOVA/Kruskal–Wallis with Benjamini–Hochberg FDR ( \(q=0.05\) ). EEG–behavior associations were examined with Spearman correlations. For machine learning, common spatial patterns (CSP) features were classified using linear discriminant analysis (LDA) with leave-one-subject-out cross-validation. To align with clinical workflow, we report HC vs P as “stroke detection/screening” and patient-only P1/P2/P3 classification as “lesion subtype decoding for stratification” (along with pairwise P1 vs P2, P1 vs P3, and P2 vs P3 models). An EEGNet baseline was evaluated for comparison.

Results

Increased \(\alpha \) power and a leftward peak \(\alpha \) shift were observed in patients (HC: \(9.93 \pm 1.09\,\text {Hz}\) ; P: \(8.75 \pm 1.02\,\text {Hz}\) ; \(p = 6.82 \times 10^{-5}\) ). Pre-FDR, \(\theta \) -band frontal–motor connectivity was strengthened, while posterior P–O connectivity in \(\alpha \) / \(\beta _{1}\) was weakened. Ipsilesional dominance in \(\theta \) was indicated by dirBSI (HC: \(-0.026 \pm 0.101\) ; P: \(0.061 \pm 0.122\) ; \(q=0.012\) ). Across lesions, \(\beta _{1}\) power differences in central/parietal/occipital ROIs were detected pre-FDR, with higher parietal \(\beta _{1}\) in P3; \(\alpha \) -band asymmetry was stronger in P1/P2 and more symmetric in P3 ( \(q=0.028\) ). EEG–behavior correlations did not survive FDR. Using CSP+LDA, accuracies of 92.41% (HC vs P), 94.87% (P1 vs P3), 85.71% (P2 vs P3), and 82.50% (P1 vs P2) were achieved; all binary AUCs exceeded 0.85; three-class accuracy reached 85.96%.

Conclusion

This multi-scale EEG framework identifies lesion-associated neurophysiological signatures and demonstrates feasible lesion subtype decoding, supporting the potential of EEG biomarkers for objective stratification and precision neurorehabilitation.