Computational Modelling of Spatio-Temporal EEG Brain Data with Spiking Neural Networks

The research presented in this thesis is aimed at modelling, classification and understanding of functional changes in brain activity that forewarn of the onset and/or the progression of a neurodegenerative process that may result in a number of disorders, including cognitive impairments, opiate addiction, Epilepsy and Alzheimer’s Disease. The study of neural plasticity and disease onset have been the centre of attention for researchers; especially as the population is ageing there is a need to deal with the increase in cognitive decline and the early onset of neurological diseases. As a consequence, large amounts of brain data has been collected and even more is expected to be collected, by means of novel computational techniques and biochemistry measurements. However, brain data is difficult to analyse and understand, especially since many of the traditional statistical and AI techniques are not able to deal with it appropriately. Driven by these issues and aiming to achieve the proposed goals, this study undertook to explore the potential of an evolving spatio-temporal data processing machine called the NeuCube architecture of spiking neurons, to analyse, classify and extract knowledge from electroencephalography spatio-temporal brain data. Firstly, the research undertaken in this thesis proposes a biologically plausible spiking neural network methodology for electroencephalography data classification and analysis. Secondly, it proposes a methodology for understanding functional changes in brain activity generated by the spatio-temporal data in the spiking neural network model. Thirdly, a new unsupervised learning rule is proposed for the investigation of the biological processes responsible for brain synaptic activity with the aim of targeting pharmacological treatments. The research undertaken achieved the following: high accuracy classification of electroencephalography data, even when fewer EEG channels and/or unprocessed data was used; personalised prognosis and early prediction of neurological events; the development of a tool for visualization and analysis of connectivity and spiking activity generated in the computational model; a better understanding of the impact of different drug doses on brain activity; a better understanding of specific neurological events by revealing the area of the brain where they occurred; and the analysis of the impact of biochemical processes on the neuronal synaptic plasticity of the model. Further improvement of the understanding and use of the proposed methodologies would contribute to the advancement of research in the area of prediction of neurological events and understanding of brain data related to neurological disorders, such as Alzheimer’s Disease