Predicting functional recovery after stroke through eeg connectivity analysis and machine learning approaches

 

Do you not understand, prediction is completely useless for stroke survivors? It does nothing to get them recovered. There are a lot of mentors and senior researchers that need to be re-educated on the purpose of stroke research. The only goal in stroke is 100% recovery: not biomarkers, prediction, prognosis or other useless shit! I'd fire all of you for incompetence!

Predicting functional recovery after stroke through eeg connectivity analysis and machine learning approaches

MDPI

Journal of Clinical Medicine (JCM)

October 202514(20):7358

DOI:10.3390/jcm14207358

LicenseCC BY 4.0

Authors:

SeungHeon An

DongGeon Lee

DongMin Park

Kyeongbong Lee

Request Full-text Paper PDF

Request full-text PDF

 

Introduction

Stroke is a leading cause of long-term motor disability and dependency worldwide. The early post-stroke phase is a critical window for initiating targeted neurorehabilitation. However, outcomes vary widely even among patients with similar initial severity. Electroencephalography (EEG) offers a non-invasive, affordable solution with excellent temporal resolution. Functional connectivity metrics such as total coherence (TotCoh) and Small World (SW), when combined with artificial intelligence (AI), can provide valuable insights into the functional integrity of motor-related brain networks [1,2]. Here, we developed and validated an AI-based framework to classify stroke status, lesion side, and forecast recovery potential using EEG-derived features and baseline clinical evaluation.

 

MethodsAA

We enrolled 127 patients with first-ever subacute ischemic stroke (<15 days post-onset) and 90 healthy controls matched for age and sex. All participants underwent resting-state EEG. EEG connectivity metrics (TotCoh and SW) were computed for each hemisphere across Delta (2–4 Hz), Theta (4–8 Hz), Alpha1 (8–10 Hz), Alpha2 (10–13 Hz), Beta1 (13–20 Hz), Beta2 (20–30 Hz), and Gamma (30–40 Hz) frequency bands. NIHSS was administered at baseline (T0) and after 40 days of standard rehabilitation (T1). 

Results

After feature selection, each classifier retained a minimal yet highly informative set of EEG connectivity features. Stroke detection relied mainly on TotCoh values in the right hemisphere across Delta, Theta, and Gamma bands. Lesion lateralization was best captured by right hemisphere Theta TotCoh and SW indices from both hemispheres, highlighting interhemispheric asymmetries. For recovery prediction, the most relevant features included baseline NIHSS, TotCoh in Beta2 and Gamma bands, and SW

Discussion

Our findings confirm the utility of EEG-based functional connectivity as a reliable, early biomarker of motor recovery potential in subacute stroke. These findings highlight the robustness of the approach and the clinical relevance of the selected connectivity features, particularly those in the beta and alpha frequency ranges, which are tightly linked to motor function and plasticity mechanisms. The AI-driven approach allowed the development of accurate and interpretable models, easily