Towards improved EEG interpretation in a sensorimotor BCI for the control of a prosthetic or orthotic hand.
Date
2011-10-12
Authors
Mohamed, Abdul-Khaaliq.
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A brain computer interface (BCI), which reroutes neural signals from the brain to
actuators in a prosthetic or orthotic hand, promises to aid those who suffer from hand
motor impairments, such as amputees and victims of strokes and spinal cord injuries.
Such individuals can greatly benefit from the return of some of the essential
functionality of the hand through the renewed performance of the basic hand
movements involved in daily activities. These hand movements include wrist
extension, wrist flexion, finger extension, finger flexion and the tripod pinch. The
core of this sensorimotor BCI solution lies in the interpretation of the neural
information for the five essential hand movements extracted from EEG
(electroencephalogram). It is necessary to improve on the interpretation of these EEG
signals; hence this research explores the possibility of single-trial EEG discrimination
for the five essential hand movements in an offline, synchronous manner.
The EEG was recorded from five healthy test subjects as they performed the actual
and imagined movements for both hands. The research is then divided into three
investigations which respectively attempt to differentiate the EEG for: 1) right and
left combinations of the different hand movements, 2) wrist and finger movements on
the same hand and 3) the individual five movements on the same hand. A general
method is applied to all three investigations. It utilizes independent component
analysis (ICA) and time-frequency techniques to extract features based on eventrelated
(de)synchronisation (ERD/ERS) and movement-related cortical potentials
(MRCP). The Bhattacharyya distance is used for feature reduction and Mahalanobis
distance clustering and artificial neural networks are used as classifiers. The best average accuracies of 89 %, 71 % and 57 % for the three respective
investigations are obtained using ANNs and features related to ERD/ERS. Along with
accuracies around 70 % for a few subjects in the five-movement differentiation
investigation, these results indicated the possibility of offline, synchronous
differentiation of single-trial EEG for the five essential hand movements. These hand
movements can be used in part or in combination as imagined and performed motor
tasks for BCIs aimed at controlling prosthetic or orthotic hands.