Enhancing upper limb motor recovery prediction after acute stroke using EEG and subacute data

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Science

Background:

• Stroke recovery of upper limb function is challenging to predict.
• Existing prediction methods, like the proportional recovery rule, have limitations.
• Electroencephalography (EEG) offers potential biomarkers for functional recovery assessment.

Purpose of the Study:

• To develop and validate a novel EEG-based machine learning model (StrokeRecovNet) for predicting upper limb motor recovery in stroke survivors.
• To compare the predictive performance of StrokeRecovNet against the proportional recovery rule.
• To identify key EEG biomarkers contributing to accurate recovery prediction.

Main Methods:

• Developed StrokeRecovNet, a feed-forward neural network, using 221 EEG biomarkers (spectral and functional connectivity) and clinical data.
• Trained and validated the model on two independent datasets of acute and subacute stroke patients.
• Evaluated model performance using median absolute error (MAE) for predicting Fugl-Meyer Assessment of the Upper Extremity (FMAUE) scores.

Main Results:

• StrokeRecovNet significantly outperformed the proportional recovery rule in predicting FMAUE scores in the subacute stage (MAE: 5.85 vs. 19.00).
• Incorporating subacute EEG data improved predictions for acute stroke patients (MAE: 5.87 vs. 8.80 for PRR).
• Brain symmetry indices and functional connectivity measures were key predictive features, varying with recovery stage.

Conclusions:

• EEG-based biomarkers can effectively predict individual upper limb motor recovery trajectories in stroke survivors.
• StrokeRecovNet provides a data-driven approach for forecasting recovery, outperforming traditional benchmarks.
• Subacute EEG data enhances early prediction accuracy, supporting personalized post-stroke rehabilitation planning.