Translation of single channel electro encephalic signals into limb motion

Abstract

Neural prostheses (NPs) are devices that can translate brainwaves into motion. The non-invasive multi-channel headset used in the study of Brain–Computer Interface (BCI) systems for the development of NPs, presents high resolution in data collection, but also presents high computing expenses and hardware costs. To overcome the barrier of the costs and present an accessible technology for these studies, this manuscript presents the implementation of a method that uses a single-channel headset to sample the Electro Encephalo Graph (EEG) wave. The headset provides 8 individual brain waves (delta, theta, low alpha, high alpha, low beta, high beta, low gamma, mid gamma), operating in their characteristic frequency intervals. A Multi-layer Perceptron (MLP) was trained with the Alpha and Beta waves (4 signals), reaching a $73,9\text{\%}$ accuracy rate for detecting the movement (open/close) of the subject’s right hand. The conclusion on the subject hand status is fed into a kinematic (Denavit Hartenberg) model of the hand, to simulate the opening/ closing of a robotic hand. The results confirm the usability of the single-channel headset to extract information from the motor cortex for the development of cheaper and more accessible NPs. The advantages of this method are: (a) lower hardware expense and (b) lower computing load. The disadvantages of our approach lie in the time needed for the 15 s to react to the real-time patient brain signal and to produce the Open/Close command to the Neural Prosthesis. Future endeavors include the online usage of the trained NN by the subject. An additional interest domain is the usage of intention-of-movement brain waves for forecasting.