Brain Machine Interface (BMI) or Brain Computer Interface (BCI) technologies provide the prospect of regenerating or replacing functions lost due to motor disabilities. BCIs connect the brain to a computer which translates the electrical activity of the brain into commands used to control external devices, hence allowing people with motor disability to control their external environment through a non-muscular communication channel. A BCI operates by transforming electrophysiological signals, known as Electroencephalogram (EEG) signals, from the user into device commands under an operating protocol. The protocol initialises and defines the nature of the communication (i.e., discrete or continuous). It also determines the strategy which underpins the generation of the signals used by the system (i.e., what triggers the changes within the EEG signals). While protocols involving brisk and very constrained movements have been widely explored in BCI studies, more natural movements have barely been considered. The choice of protocols that entail non-realistic movements emanates from the generation of well understood neuronal correlates modulated by the execution of such movements, leading to a lack of freedom in the design of BCI protocols.The development of algorithms translating EEG signals into commands that control external devices,a task termed as event detection in the present research, are a central part of any BCI system. However, most of the time, complex methods are used and most event detection in BCI development is devoted to the optimisation of such methods. Furthermore, the methods require extensive training of the user, putting a mental load on the user. The present study aims to investigate simple, but powerful, event detection methods requiring minimal training and the use of a protocol involving natural movements. Scalp EEG data was recorded from nine participants using natural hand movements.In particular, self-paced reaching hand movements were considered. The data was investigated in a pseudo time frequency domain using continuous wavelet coefficients. Methods using wavelet modulus maxima, the Mahalanobis distance, and bootstrapping of the Mahalanobis distance were developed for event detection. The data was analysed over a frequency range from 0.1 Hz to 25 Hz, covering the Slow Cortical Potentials (SCP), Mu and Beta frequency bands.The results showed that the method using wavelet modulus maxima was able to predict reaching hand movements onset in the SCP, Mu and Beta frequency bands about 1 s before movement onsetand yielded an maximum average prediction rate of approximately 80%. The Mahalanolobisdistance and the bootstrap methods were able to predict reaching hand movements initiation inthe SCP band about 1 s before movement onset with a maximum prediction rate of approximately 70%. The study has demonstrated that human voluntary movements can be predicted approximatively 1 s prior to movement onset with a good prediction rate. The study may contribute to the understanding of the planning and the control of human voluntary movements. Furthermore, the present research may contribute in designing advanced assistive devices in general and in particular may contribute in improving BCI systems design. Finally, the results may encourage the use of natural movements during BCI protocols design aiming to predict movement initiation and the monitoring of the mental state of BCI users.
|Date of Award||28 Jun 2016|
- University Of Strathclyde
|Supervisor||Heba Lakany (Supervisor) & Bernard A Conway (Supervisor)|