Prosthetic control system based on motor imagery.

A brain-computer interface (BCI) can be used for function replacement through the control of devices, such as prostheses, by identifying the subject's intent from brain activity. We process electroencephalography (EEG) signals related to motor imagery to improve the accuracy of intent classification. The original signals are decomposed into three layers based on db4 wavelet basis. The wavelet soft threshold denoising method is used to improve the signal-to-noise ratio. The sample entropy algorithm is used to extract the features of the signal after noise reduction and reconstruction. Combined with event-related synchronisation/desynchronisation (ERS/ERD) phenomenon, the sample entropy in the motor imagery time periods of C3, C4 and Cz is selected as the feature value. Feature vectors are then used as the input of three classifiers. From the evaluated classifiers, the backpropagation (BP) neural network provides the best EEG signal classification (93% accuracy). BP neural network is thus selected as the final classifier and used to design a prosthetic control system based on motor imagery. The classification results are wirelessly transmitted to control a prosthesis successfully via commands of hand opening, fist clenching, and external wrist rotation. Such functionality may allow amputees to complete simple activities of daily living. Thus, this study is valuable for subsequent developments in rehabilitation.