Developments in signal processing for computerised diagnosis in clinical neurophysiology

The aim of this study was to apply signal processing techniques to a potential known as the contingent negative variation (CNV) in order to aid detection of schizophrenia, Parkinson's disease (PO) and Huntington's Disease (lID). A data recording system was constructed and used to obtain data from 20 schizophrenic patients, 16 PO patients, 21 -at-risk- of HD patients, 11 HD patients and 43 normal control subjects. The data included the CNV, electro-oculograms (required for the preprocessing of the CNV) and the subjects reaction times to an acoustic stimulus. The CNV waveforms were initially preprocessed. This reduced the effects of background electroencephalogram and ocular artefact potentials. The CNV waveforms were then processed using a method which involved the discrete Fourier transform (OFf) and discriminant analysis. This method developed from the work of Martin Nichols and Michael Coelho. It was possible to successfully identify the majority of the patients using this method. In order to reduce the complexity of patients' Identification a different method of CNV signal processing was considered. This involved obtaining the CNV features in the time domain and using them in neural networks. This method was as effective as the method which used OFf and discriminant analysis in identifying the patients. To establish whether HO could presymptomatically be detected in the at-risk of HD group, the CNV was analysed using principal component analysis (PCA) and Ward's clustering method. This resulted in identification of 7 patients who were suggested would develop HO. The subjects' reaction times were also analysed. This indicated that the reaction times of schizophrenic, PO, HO and some at-risk of HD patients were significantly different from the reaction times of their normal control subjects.