Single-trial EEG Discrimination between Five Hand Movement Imagery and Execution towards the control of a prosthetic/orthotic hand using a brain-computer interface Abdul-Khaaliq Mohamed School of Electrical and Information Engineering University of Witwatersrand Johannesburg, South Africa abdul-khaaliq.mohamed@wits.ac.za Lester Ryan John Department of Human Biology University of Cape Town Cape Town, South Africa lester.john@uct.ac.za Abstract— A brain computer interface (BCI) holds the potential to allow motor-impaired individuals to perform simple daily tasks by translating their neural intentions into movements of a robotic prosthetic/orthotic hand. These daily tasks can be facilitated through the performance of key kinematic hand movements; however, few BCI studies to date focus on functional kinematic unilateral hand movements. This paper presents the first attempt to discriminate the neural control signals for unilateral wrist extension and flexion, finger extension and flexion and the tripod pinch from single-trial electroencephalogram (EEG). High-resolution EEG was recorded from test subjects as they executed and imagined five essential hand movements using both hands. Independent component analysis (ICA) and time-frequency techniques were used to extract spectral features based on event-related (de)synchronisation (ERD/ERS). The Bhattacharyya distance (BD) was used for feature reduction, while Mahalanobis distance (MD) clustering and artificial neural networks (ANN) were used as classifiers. Average accuracies achieved ranged between 51 % and 63 %, however some subjects showed accuracies around 70 % for either real or imagined movements. This indicates the possibility of individual kinematic movement discrimination between this new combination of BCI hand movements. Keywords— Artificial neural networks (ANN), brain-computer interface (BCI), electroencephalogram (EEG), event-related (de)synchronisation (ERD/ERS), group classifier, independent component analysis (ICA), Mahalanobis distance (MD) I. INTRODUCTION Based on the assumption that neural activity can be translated into intended movements, which can then be executed by a device, such as a prosthetic or orthotic hand, scientists and engineers have been challenged to replicate the sensory-motor functions of the human hand [1] – [3]. People who suffer from motor impairments can benefit from these efforts, which aim to return some of the essential functionality of the human hand [1]. In the case of an amputee, a robotic prosthetic hand can replace the lost hand, while the non- functional hand of a stroke or spinal cord victim can be mobilised by a robotic exoskeletal orthotic hand [1]. Through thought, the user can then control one of these external devices using a brain-computer interface (BCI) to reroute the signals directly from the brain to actuators in the device [1], [2]. This solution can allow motor-impaired individuals to perform essential hand movements that will aid daily tasks, such as grasping or writing [1], [4]. Considering the movements that patients learn during motor rehabilitation [5], [6], [7], five basic hand movements are considered i.e. wrist extension (WE), wrist flexion (WF), finger extension (FE), finger flexion (FF) and the tripod pinch (TR). Offline single- trial investigations using neural information taken from electroencephalogram (EEG) for the combination of the WE, WF, FE, FF and the TR have been recently undertaken [8], [9]. The possibility of discriminating EEG between right and left groupings of the essential hand movements has been shown by Mohamed et al (a) [8], while Mohamed et al (b) [9] demonstrated that EEG for key parts of each hand i.e. the wrist and fingers, can be distinguished. However, a study to discriminate the EEG associated with these five hand movements on the same hand has not been published. There are two possible reasons for this: 1) multiclass BCI studies to date have yielded low accuracies [10] – [13], and 2) extracting relevant information originating from adjacent areas of the motor cortex (such as those controlling movements on the same limb) is challenging [14]. Hence, this paper outlines a first attempt to discriminate between the EEG associated with unilateral WE, WF, FE, FF and TR. II. BACKGROUND A. Sensorimotor BCI A BCI allows the actuation of an external device, such as a prosthetic hand, by interpreting the user’s intent from his brain, thus circumventing the natural neuromuscular pathway of the human body [1], [15]. BCIs that deal with the motor functions from and sensory inputs to the sensorimotor cortex of the brain are termed sensorimotor BCIs. BCIs can capture neural information invasively using (ECoG – electrocorticography for example) or non-invasively (using EEG for example).