Engineering in brain research : Processing electroencephalograms and chaos in neural networks
Thesis DisciplineElectrical Engineering
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
The structure and function of the brain, as it is presently understood, is outlined. The importance of technology in acquiring this knowledge is illustrated by tracing the history of brain research, and the contributions that engineering is currently making to brain research are discussed. The manifestation of epilepsy in recordings of the electrical activity of the brain – electroencephalograms (EEGs) is outlined. A new PC-based system for the automated detection of this epileptiform activity is presented. The system consists of three stages: data collection, feature extraction and event detection. The feature extraction stage detects candidate epileptiform transients on individual channels, while an expert system is used to detect focal and non-focal epileptiform events. Considerable use of spatial and temporal contextual information present in the EEG aids both in the detection of epileptiform events and in the rejection of artifacts and background activity as events. Classification of events as definite or probable overcomes, to some extent, the problem of maintaining satisfactory detection rates while eliminating false detections. Test results are presented which indicate that this system should be capable of performing reliably in routine clinical EEG screening. Neural networks are introduced and their application to real-world problems examined. In particular, the application of back-propagation neural networks to the detection of epileptiform transients is discussed. A number of modifications to the back-propagation learning algorithm are proposed which should enable the desired network performance to be achieved. The phenomenon of deterministic chaos is reviewed and evidence for chaos in the brain presented. A new model of a neuron, termed the versatile neural unit and based on a chaotic system is proposed. This neural unit is relatively simple, yet produces a wide range of activity reminiscent of that observed in neurons. The versatile neural unit provides a means for introducing chaos into neural networks to enhance their performance and, at the same time, provide insights into the roles of chaos in the brain. Networks of versatile neural units are shown to produce activity similar to that observed in EEGs and the introduction of chaos into the self-organizing map is shown to improve its ability to cluster input patterns and model their probability density function.