# Source-Level Resting-State EEG Connectivity Reveals Frequency-Specific Neural Reorganization and Predicts Motor Recovery in Individuals Post Stroke Following Gait Rehabilitation

## TL;DR

This study shows that resting-state EEG can track brain changes after stroke rehabilitation and predict motor recovery, offering a new tool for neurorehabilitation.

## Contribution

The study introduces source-level EEG connectivity as a novel biomarker for tracking neural reorganization and predicting motor recovery post-stroke.

## Key findings

- Post-stroke individuals showed baseline reduced connectivity in SMA and vPM, with compensatory increases in the insula and dorsalA6.
- After rehabilitation, connectivity increased in M1 and PoG, with dorsalA6 showing enhanced node strength and correlating with motor recovery.
- Machine learning achieved high accuracy in differentiating groups using alpha and gamma frequency EEG features.

## Abstract

Stroke-induced motor impairments are linked to disrupted cortical connectivity and remain a major challenge in neurorehabilitation. The purpose of this investigation was to evaluate brain network reorganization and its relationship to motor recovery following a 10-week gait rehabilitation program (3 days/week for a total of 30 sessions).

Forty-four participants (22 healthy adults HC, 22 individuals diagnosed with stroke) were enrolled, individuals in the stroke group were randomized into Exoskeleton-Assisted Rehabilitation (ER=9) or Standard of Care (SOC=13) groups. Resting-state EEG was recorded pre- and post-intervention. Directed functional connectivity was estimated from localized sources using Partial Directed Coherence (PDC), followed by graph theory analysis, laterality index (LI) computation, and machine learning classification using frequency-specific EEG features.

At baseline, individuals in the stroke group demonstrated reduced connectivity relative to HC group in the affected supplementary motor area (SMA, p<0.005) and unaffected ventral premotor cortex (vPM, p<0.01), with compensatory increases in the affected insula (INS) and dorsalA6. Post-intervention, connectivity increased from the unaffected primary motor cortex (M1) and affected postcentral gyrus (PoG, p<0.01), with dorsalA6 showing enhanced node strength (p<0.01). LI analysis indicated reduced contralesional compensation in cingulate gyrus (CG) and middle temporal gyrus (MTG), and re-engagement of the affected hemisphere (PoG, dorsalA6). DorsalA6 connectivity correlated with Fugl-Meyer score improvements (r=0.64, p=0.001). Machine learning achieved 93.18% accuracy and 94.83% Area Under ROC Curve to differentiate between the groups with features from the alpha and gamma being the most discriminative.

Source-level EEG connectivity and frequency-specific features are sensitive biomarkers of neuroplasticity. The 10-week program enhanced motor network integration and hemispheric balance, reflected in connectivity and clinical outcomes. EEG-based machine learning tools offer scalable solutions for precision assessment of neurorehabilitation.