A novel methodology based on static visual stimuli and kinesthetic motor imagery for upper limb neurorehabilitation

This work proposes a protocol to be used in a Brain-Computer Interface (BCI) for hand and finger movement rehabilitation aimed at post-stroke patients. This strategy is based on Electroencephalography (EEG) and uses concepts of both Static Visual Stimuli (SVS) of the own subject’s hand (OwnSVS) and Kinesthetic Motor Imagery (KMI) to improve the motor task recognition of the same limb.

The protocol proposed here takes into account several aspects, such as physical rehabilitation strategies, Active Hand Orthosis (AHO) — represented by a robotic glove —, and the acquisition of the EEG signal over the brain motor area. Power Spectral Density (PSD) and Riemannian Geometry (RG) are used for feature extraction in this work, considering Mu ( \(\mu \) $\mu$ , 8–12 Hz), Low-Beta (Low- \(\beta \) $\beta$ , 13–17 Hz) and High-Beta (High- \(\beta \) $\beta$ , 18–24 Hz) frequency bands. Moreover, a feature selection stage using Pair-Wise Feature Proximity (PWFP) is also used before input to a Machine Learning (ML) classifier. Here, Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), k-Nearest Neighbors (kNN), and Decision Tree (DT) are compared.

As results, evaluated in 10 healthy subjects, features using PSD-PWFP and kNN achieved a maximum ACC of 89.81% (with AHO movements), improving by 3% the standard protocol (without AHO movements).

The findings of this work indicate that AHO movement assistance can improve the recognition of hand opening and closing MI tasks, which can be implemented in the design of robotic BCI controllers for hand and finger movement neurorehabilitation in stroke patients.