Identification of Neuromotor Patterns Associated with Hand Rehabilitation Movements from EMG and EEG Signals

Stroke rehabilitation is essential for restoring upper limb function, impaired in nearly 85% of stroke survivors. Accurate classification of movement intention and execution can enhance rehabilitation through improved patient monitoring and robotic assistance. This study explores integrating electroencephalography (EEG) and electromyography (EMG) signals to classify five hand movements commonly used in rehabilitation therapy: four finger pinches and full hand closure. Biological signals were acquired from 22 healthy participants using a synchronized setup of EEG and EMG systems. Signal processing methods were applied to extract frequency and connectivity features from EEG and muscle activity metrics from EMG. These features were used to train machine learning models, including Ensemble Bagged Trees, neural networks, and SVMs, for classifying movement intention and execution. Results demonstrated that the Ensemble Bagged Trees model achieved the highest accuracy classified 5 classes, particularly when employing EEG power spectral density features, with 84.7% and 67.4% accuracy for movement execution and intention, respectively. Correlation analysis identified distinct motor-related brain connectivity patterns, offering insights into neuromotor processes. These findings support the integration of EEG-EMG analysis with machine learning to enhance stroke rehabilitation, improving patient monitoring and robotic-assisted therapy.