EEG feature extraction and classification in multiclass multiuser motor imagery brain computer interface u sing Bayesian Network and ANN

Abstract

Brain Computer Interface (BCI) is a direct communication channel between a trained human brain and an external device. For people who are paralysed, BCI acts as an interface to control and regulate external devices and replaces their lost motor functionality, using brain activity alone. A motor imagery BCI translates a person's imagination about a movement into control signals which in turn controls the intended device. For this, the EEG signals that are produced according to the motor imaging need to be processed and analysed using various signal processing algorithms. During a motor activity, only the sensorimotor rhythms, which are the μ and β rhythms, get activated. Wavelet decomposition, which could provide both frequency and time localization, is employed to extract the sensorimotor rhythms. Since different brain regions are dependent during an activity, the correlation between the EEG channels must also be considered during the extraction of features. For this, Bayesian Network approach is used to obtain the maximum probable channels during each motor activity. Most of the BCI systems are subject specific systems. The proposed system uses Artificial Neural Network for feature classification and is a multi-user BCI with increased accuracy, which can be an adaptive system for multiple users.