# Peak alpha frequency as an objective biomarker for cognitive assessment in post-stroke cognitive impairment

## TL;DR

This study shows that peak alpha frequency in EEG readings can help identify cognitive impairment after stroke, offering a potential objective biomarker for diagnosis.

## Contribution

The study introduces peak alpha frequency (PAF) as a novel, objective biomarker for assessing post-stroke cognitive impairment (PSCI).

## Key findings

- PSCI patients had significantly lower peak alpha frequency (PAF) compared to healthy controls across all major brain regions.
- T3PAF and T4PAF were strongly correlated with cognitive scores and effectively distinguished PSCI from healthy controls.
- EEG-based biomarkers achieved high diagnostic accuracy with AUCs of 0.761 and 0.773 using logistic regression and Random Forest models.

## Abstract

To investigate regional associations between peak alpha frequency (PAF) and poststroke cognitive impairment (PSCI) and evaluate PAF as an objective biomarker for cognitive assessment in PSCI.

A cross-sectional study compared 103 participants [PSCI, poststroke non-impaired (PSN), and healthy controls]. Cognitive function was assessed using MoCA scores. PAF characteristics were analyzed across brain regions via EEG, with logistic regression and Random Forest identifying key predictors. We aimed to evaluate whether PAF can be an effective indicator of cognitive status in PSCI.

The Kruskal-Wallis test with post hoc Bonferroni correction revealed that PSCI exhibited significantly lower PAF compared to HC across all major brain regions (frontal, temporal, central, and parieto-occipital; all P < 0.05). Compared to PSN, the PSCI group showed significantly reduced PAF at specific electrodes (F3, F4, F7, T3, T6, Fz; P < 0.05). Spearman correlation analysis demonstrated that PAF at multiple leads was positively correlated with MoCA scores across all subjects. Notably, after FDR correction, only T3PAF and T4PAF remained significantly negatively correlated with MoCA in all subjects (q < 0.05). Binary logistic regression identified T4PAF as the most discriminative predictor for distinguishing PSCI from HC (OR = 2.525). Random Forest analysis corroborated these findings, identifying F7PAF, O2PAF, T3PAF, and T4PAF as the most important predictors. Both models demonstrated excellent discriminatory power, with AUCs of 0.761 (logistic regression) and 0.773 (Random Forest), indicating robust performance of EEG-based biomarkers for PSCI detection.

Peak alpha frequency serves as a robust electrophysiological biomarker for PSCI. Multi-region PAF analysis enhances diagnostic precision for poststroke cognitive decline.