Use this URL to cite or link to this record in EThOS:
Title: MCI progression classification for early diagnosis of Alzheimer's disease using machine learning and deep learning methods
Author: Yan, Yu
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2020
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Restricted access.
Access from Institution:
Alzheimer’s disease (AD), the most common cause of dementia, affects more than 520,000 people in the UK. It is a progressive disease and its causes still remain unclear. Besides the traditional neuropsychological test, imaging biomarkers are playing an increasingly important role for AD detection and especially for early stage diagnosis including prediction of the transition from mild cognitive impairment (MCI) to AD. The use of structural imaging such as T1 MRI for AD diagnosis has been widely explored, however, the application of PET imaging to AD is relatively novel, with new tracers appearing. In addition, the development of machine learning and deep learning algorithms can help researchers and clinicians build computer-based image analysis tools, which can be used to further enhance the AD diagnosis results by utilising image features that cannot be easily interpreted directly by human eyes. This thesis focuses on the development of such image analysis methods for the prediction of MCI to AD conversion. We first propose the use of novel features, based on complexity measurement, from volumetric amyloid PET images, which we use as input to a nonlinear Support Vector Machine (SVM) classifier. We tested this method using the novel PET features alone or in combination with standard PET metrics and MRI-derived features, obtaining state of the art accuracy. Our second main contribution involves the development of a deep learning model using two parallel 3D Convolutional Neural Networks (CNNs) to combine the information from amyloid PET and T1 MRI. Deep learning methods in this area are typically limited by the small size of available PET datasets. We show that prediction accuracy is improved by combining “diagnostically similar” datasets together as a form of augmentation (progressed MCI + pAD against stable MCI, as well as progressed MCI + pAD against stable MCI + healthy volunteer). Our third major contribution explores a new method for augmentation using data synthesis through a conditional Generative Adversarial Network (cGAN), which can generate new amyloid PET images from volumetric T1 MRI scans, for which there is much larger availability. The generated amyloid images are used to augment the original progressed MCI dataset to address the problem of data imbalance. The results showed that the accuracy for MCI to AD prediction using generated data augmentation method outperformed traditional augmentation methods substantially.
Supervisor: Grau, Vicente Sponsor: China Scholarship Council
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Biomedical Engineering