Alzheimer's disease (AD) is a form of progressive neurodegenerative dementia and one of the most common diseases in the aging population. Early diagnosis of AD is strongly recommended for several reasons. First, it can helps to significantly reduce the social and economic impacts caused by AD and allow people to better manage and plan ahead. Second, it may provide more information for researchers seeking new scientific approaches for early treatment and intervention. However, in current clinical practice early diagnosis is often a challenge. While neuroimaging is routinely collected in hospitals, it is very hard for radiologists to manually read the high-dimensional image data for analysis and interpretation. This project proposes untested but potentially transformative research approaches to identify high-dimensional image features for AD early diagnosis based on Magnetic Resonance Imaging (MRI). This project will advance the research in machine learning, optimization, statistics, image science and bioinformatics, and potentially be used to address other high-dimensional images besides brain images. The project also has broader impacts through cross-disciplinary research, training and education.
This project has the following two aims: 1) Develop sparse coding based algorithms to identify features of structural MRI images for classifying AD patients and other diagnostic groups. This will allow the key structural features of images that separate AD patients, individuals with mild cognitive impairment (MCI) or healthy individuals to be identified. 2) Develop optimization and machine learning algorithms based on tensor Tucker core decomposition for high-dimensional image-marker detection from longitudinal functional MRI images. This approach should reduce the high computational complexity of marker detection from the longitudinal MRI images of AD patients. It is anticipated that the developed algorithms will enhance high-dimensional neuroimaging marker detection and diagnostic classification. This research project, if successful, will greatly impact the current practice of AD diagnosis by providing clinical doctors with the information from a larger population and also significantly easing the burden of radiologists.
