Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.617386
Title: Early detection of neurodegenerative diseases from bio-signals : a machine learning approach
Author: Iram, Shamaila
ISNI:       0000 0004 5350 4798
Awarding Body: Liverpool John Moores University
Current Institution: Liverpool John Moores University
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Given the fact that people, especially in advanced countries, are living longer due to the advancements in medical sciences which resulted in the prevalence of age-related diseases like Alzheimer’s and dementia. The occurrence of such diseases continues to increase and ultimately the cost of caring for these groups will become unsustainable. Addressing this issue has reached a critical point and failing to provide a strategic way forward will negatively affect patients, national health services and society as a whole. Three distinctive development stages of neurodegenerative diseases (Retrogenesis, Cognitive Impairment and Gait Impairment) motivated me to divide this research work into two main parts. To fully achieve the purpose of early detection/diagnosis, I aimed at analysing the gait signals as well as EEG signals, separately, as both of these signals severely get affected by any neurological disease. The first part of this research work focuses on the discrimination analysis of gait signals of different neurodegenerative diseases (Parkinson’s, Huntington, and Amyotrophic Lateral Sclerosis) and also of control subjects. This involves relevant feature extraction, solving the issues of imbalanced datasets and missing entries and lastly classification of multiclass datasets. For the classification and discrimination of gait signals, eleven (11) classifiers are selected representing linear, non-linear and Bayes normal classification techniques. Results revealed that three classifiers have provided us with higher accuracy rate which are UDC, LDC and PARZEN with 65%, 62.5% and 60% accuracy, respectively. Further, I proposed and developed a new classifier fusion strategy that combined classification algorithms with combining rules (voting, product, mean, median, maximum and minimum). It generates better results and classifies subjects more accurately than base-level classifiers. The last part of this research work is based on the rectification and computation of EEG signals of mild Alzheimer’s disease patients and control subjects. To detect the perturbation in EEG signals of Alzheimer’s patients, three neural synchrony measurement techniques; phase synchrony, magnitude squared coherence and cross correlation are applied on three different databases of mild Alzheimer’s disease (MiAD) patients and healthy subjects. I have compared right and left temporal parts of brain with rest of the brain area (frontal, central and occipital), as temporal regions are relatively the first ones to be affected by Alzheimer’s. Two novel methods are proposed to compute the neural synchronization of the brain; Average synchrony measure and PCA based synchrony measure. These techniques are evaluated for three different datasets of MiAD patients and control subjects using the Wilcoxon ranksum test (Mann-Whitney U test). Results demonstrated that PCA based method helped us to find more significant features that can be used as biomarkers for the early diagnosis of Alzheimer’s.
Supervisor: Aljumeily, Dhiya; Fergus, Paul; Randles, Martin Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.617386  DOI: Not available
Share: