Brain-computer interface (BCI) has been known as a cutting-edge technology in the current research. It is able to measure the brain activity directly instead of using the natural peripheral nerves and muscles and translates the user’s intent brain activity into useful control signals. There is still a need for a technology for severely disabled individuals who suffer from locked-in syndromes, such as amyotrophic lateral sclerosis (ALS), cervical spinal cord injury (SCI) or tetraplegia and brain stem stroke. A brain-computer interface (BCI) could be used here as an alternative solution for control and communication. The main aim of this research is to develop a BCI system to assist mobility as hand-free technology for people with severe disability, with improved accuracy, which provides effective classification accuracy for wheelchair control.
Electroencephalography (EEG) is the chosen BCI technology because it is non-invasive, portable and inexpensive. Currently, BCI using EEG can be divided into two strategies; selective attention and spontaneous mental signal. For the selective attention strategy, BCI relies on external stimuli which might be uncomfortable for severely disabled individuals who need to focus on external stimuli and the environment simultaneously. This is not the case for BCIs which rely on spontaneous mental signals initiated by the users themselves. BCI that uses sensorimotor rhythm (SMR) is one of the examples of the spontaneous mental strategy. There have been many reports in research using SMR-based BCI; however, there are still some people who are unable to use this. As a result, in this thesis, mental task-based EEG is used as an alternative.
This thesis presents the embedded EEG system for mental task classification. A prototype wireless embedded EEG system for mental task BCI classification is developed. The prototype includes a wireless EEG as head gear and an embedded system with a wireless receiver. The developed wireless EEG provides a good common mode rejection ratio (CMRR) performance and a compact size with a low current consumption coin cell battery for power. Mental tasks data are collected using the prototype system from six healthy participants which include arithmetic, figure rotation, letter composing and counting task with additional eyes closed task. The developed prototype BCI system is able to detect the dominant alpha wave between 8-13Hz during eyes closed. Using the FFT as the features extractor and artificial neural network (ANN) as the classifier, the developed prototype EEG system provides high accuracy for the eyes closed and eyes open tasks. The classification of the three mental task combinations achieve an overall accuracy of around 70%.
Also, an optimized BCI system for mental task classification using the Hilbert-Huang transform (HHT) feature extractor and the genetic algorithm optimization of the artificial neural network (GA-ANN) classifier is presented. Non motor imagery mental tasks are employed, including: arithmetic, letter composing, Rubik’s cube rolling, visual counting, ringtone, spatial navigation and eyes closed task. When more mental tasks are used, users are able to choose the most effective of tasks suitable for their circumstance. The result of classification for the three user chosen mental tasks achieves accuracy between 76% and 85% using eight EEG channels with GA-ANN (classifier) and FFT (feature extractor). In a two EEG channels classification using FFT as the features extractor, the accuracy is reduced between 65% and 79%. However, the HHT features extractor provides improved accuracy between 70% and 84%.
Further, an advanced BCI system using the ANN with fuzzy particle swarm optimization using cross-mutated operation (FPSOCM-ANN) for mental task classification is presented. This experiment involves five able-bodied subjects and also five patients with tetraplegia as the target group of the BCI system. The three relevant mental tasks used for the BCI concentrates on mental letter composing, mental arithmetic and mental Rubik’s cube rolling forward. Although the patients group has lower classification accuracy, this is improved by increasing the time-window of data with the best at 7s. The results classification for 7s time-window show the best classifier is using the FPSOCM-ANN (84.4% using FPSOCM-ANN, 77.4% using GA-ANN, 77.0% using SVM, 72.1% using LDA, and 71.0% using linear perceptron). For practical use of a BCI, the two channels EEG is also presented using this advanced BCI classification method (FPSOCM-ANN). For overall, O1 and C4 are the best two channels at 80.5% of accuracy, followed by the second best at P3 and O2 at 76.4% of accuracy, and the third best at C3 and O2 channels at 75.4% of accuracy.