Cerebral Palsy EEG Signals Classification: Facial Expressions and Thoughts for Driving an Intelligent Wheelchair

Abstract

Brain Computer Interfaces (BCI) enables interaction between users and hardware systems, through the recognition of brainwave activity. However, the current BCI systems still have a very low accuracy on the recognition of facial expressions and thoughts. This makes it very difficult to use these devices to enable safe and robust commands of complex devices such as an Intelligent Wheelchair. This paper presents an approach to expand the use of a brain computer interface for driving an intelligent wheelchair by patients suffering from cerebral palsy. The approach was based on appropriate signal preprocessing based on Hjorth parameters, a forward approach for variable selection and several data mining algorithms for classification such as naive Bayes, neural networks and support vector machines. Experiments were performed using 30 individuals suffering from IV and V degrees of cerebral palsy on the Gross Motor Function (GMF) measure. The results achieved showed that the preprocessing and variable selection methods were effective enabling to improve the results of a commercial BCI product by 57%. With the developed system it was also possible for users to perform a circuit in a simulated environment using just facial expressions and thoughts.