Machine learning for image-based classification of Alzheimer's disease
Author(s)
Gray, Katherine Rachel
Type
Thesis
Abstract
Imaging biomarkers for Alzheimer's disease are important for improved diagnosis and monitoring,
as well as drug discovery. Automated image-based classification of individual patients
could provide valuable support for clinicians. This work investigates machine learning methods
aimed at the early identification of Alzheimer's disease, and prediction of progression in mild
cognitive impairment. Data are obtained from the Alzheimer's Disease Neuroimaging Initiative
(ADNI) and the Australian Imaging, Biomarker and Lifestyle Flagship Study of Ageing
(AIBL).
Multi-region analyses of cross-sectional and longitudinal FDG-PET images from ADNI are
performed. Information extracted from FDG-PET images acquired at a single timepoint is
used to achieve classification results comparable with those obtained using data from research-quality
MRI, or cerebrospinal fluid biomarkers. The incorporation of longitudinal information
results in improved classification performance.
Changes in multiple biomarkers may provide complementary information for the diagnosis and
prognosis of Alzheimer's disease. A multi-modality classification framework based on random
forest-derived similarities is applied to imaging and biological data from ADNI. Random forests
provide consistent similarities for multiple modalities, facilitating the combination of different
types of features. Classification based on the combination of MRI volumes, FDG-PET intensities,
cerebrospinal fluid biomarkers, and genetics out-performs classification based on any
individual modality.
Multi-region analysis of MRI acquired at a single timepoint is used to show volumetric differences
in cognitively normal individuals differing in amyloid-based risk status for the development
of Alzheimer's disease. Reduced volumes in temporo-parietal and orbito-frontal regions in
high-risk individuals from both ADNI and AIBL could be indicative of early signs of neurodegeneration.
This suggests that volumetric MRI can reveal structural brain changes preceding
the onset of clinical symptoms.
Taken together, these results suggest that image-based classification can support diagnosis
in Alzheimer's disease and preceding stages. Future work may lead to more finely meshed
prognostic data that may be useful clinically and for research.
as well as drug discovery. Automated image-based classification of individual patients
could provide valuable support for clinicians. This work investigates machine learning methods
aimed at the early identification of Alzheimer's disease, and prediction of progression in mild
cognitive impairment. Data are obtained from the Alzheimer's Disease Neuroimaging Initiative
(ADNI) and the Australian Imaging, Biomarker and Lifestyle Flagship Study of Ageing
(AIBL).
Multi-region analyses of cross-sectional and longitudinal FDG-PET images from ADNI are
performed. Information extracted from FDG-PET images acquired at a single timepoint is
used to achieve classification results comparable with those obtained using data from research-quality
MRI, or cerebrospinal fluid biomarkers. The incorporation of longitudinal information
results in improved classification performance.
Changes in multiple biomarkers may provide complementary information for the diagnosis and
prognosis of Alzheimer's disease. A multi-modality classification framework based on random
forest-derived similarities is applied to imaging and biological data from ADNI. Random forests
provide consistent similarities for multiple modalities, facilitating the combination of different
types of features. Classification based on the combination of MRI volumes, FDG-PET intensities,
cerebrospinal fluid biomarkers, and genetics out-performs classification based on any
individual modality.
Multi-region analysis of MRI acquired at a single timepoint is used to show volumetric differences
in cognitively normal individuals differing in amyloid-based risk status for the development
of Alzheimer's disease. Reduced volumes in temporo-parietal and orbito-frontal regions in
high-risk individuals from both ADNI and AIBL could be indicative of early signs of neurodegeneration.
This suggests that volumetric MRI can reveal structural brain changes preceding
the onset of clinical symptoms.
Taken together, these results suggest that image-based classification can support diagnosis
in Alzheimer's disease and preceding stages. Future work may lead to more finely meshed
prognostic data that may be useful clinically and for research.
Date Issued
2012-10
Date Awarded
2012-11
Copyright Statement
Attribution NoDerivatives 4.0 International Licence (CC BY-ND)
Advisor
Rueckert, Daniel
Hammers, Alexander
Sponsor
Engineering and Physical Sciences Research Council
Publisher Department
Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)