A Brain-Computer Interface (BCI) provides a person with the ability to communicate with a computer through the use of the brain signals rather than the use of the peripheral nerves and muscles. This research area focuses on the development of signal processing techniques for the classification of different mental states to activate commands that permit control of an external device via the brain activity.
BCI technology relies on the acquisition of electrical signals generated by billions of neurons inside the brain. The electrical fluctuations that arise from these neurons reach the scalp where they can be detected and recorded by means of non-invasive metal electrodes through a process known as electroencephalography (EEG). In a BCI system, EEG data is recorded from the human subject, processed to extract reliable features, and then mapped into computer based commands such as moving a cursor or selecting icons on the screen.
There are various neurophysiological phenomena evoking distinctive characteristics in brain signals that are suitable as control signals for BCI systems such as, sensorimotor rhythms, visual evoked potentials and P300 evoked potentials. The Department has worked on both motor imagery data as well as steady state visual evoked potentials (SSVEPs) to control BCI systems. Two main BCIs that were developed, based on SSVEPs, include a music player and a motorised bed application. Both systems exploit the electrical potentials evoked in the brain in response to repetitive visual stimulation and provide the user with an interface including a set of flickering stimuli. The user can attend to any of these stimuli to activate the desired control functions. The goal is to improve on the practicality aspect of these systems, including reduced training requirements, and reduced annoyance of the flickering stimuli, such that these systems can be used more reliably and comfortably outside a lab environment.
This research on BCI systems can lead to improved assisted living devices which can have significant benefits for individuals with restricted mobility. Furthermore, these systems can also be used by healthy users as they can provide a novel way of interacting with entertainment and gaming applications.
Scoring of sleep EEG data, to identify transient events of interest, may provide useful information regarding the occurrence of conditions, such as dyslexia, epilepsy and schizophrenia. This research area concerns the development of segmentation algorithms for automatic labeling of Stage 2 sleep EEG, characterised by spindles and K-complexes, for the detection of multiple events in EEG data.
Human beings spend about one third of their lives asleep. During this time however the brain is not sleeping and a look at the different electrical brain signals recorded during a night’s sleep shows that the sleep patterns cycle through five different stages. One of the most informative stages for an electroencephalographer searching for clinical abnormalities is Stage 2, which is characterized by spindles and K-complexes. These two transient events, which are observed from surface electroencephalography (EEG), are of particular interest as they vary across different subject groups. Specifically they are known to correlate with age and with various conditions such as dyslexia, epilepsy and schizophrenia.
Scoring of sleep EEG data to identify transient events of interest is typically done manually but this is time consuming, tedious and risks being subjectively interpreted. To address this problem, research at the Department of Systems and Control Engineering is proposing the use of switching multiple models for the segmentation and automatic labeling of Stage 2 sleep EEG characterized by spindles and K- complexes. Through supervised learning, models are trained to learn the characteristics of these different wave patterns and a probabilistic approach is then used to find which of the candidate models best represents the recorded EEG data. This approach offers a unified framework of detecting multiple events in sleep EEG and it requires very little training data, making it feasible to use in practice. A semi- supervised model allocation approach can also be adopted allowing new unknown modes to be learnt in real time.
The electrical activity of the brain during continuous activations and deactivations of specific brain areas may be recorded non-invasively by various electrodes placed on the scalp, which method is referred to as electroencephalography (EEG). This research area focuses on modelling the dynamics of recorded EEG signals, hence allowing for better understanding of the underlying processes, as well as localisation of the sources giving rise to the brain activity recorded by the electrodes.
