Lin-Ching ChangIBRAHIM AL MUBARK2022-06-052021-02-102022-06-0585261https://drepo.sdl.edu.sa/handle/20.500.14154/67274Alzheimer’s disease (AD) is a neurodegenerative disease typically affecting the elderly population. Recent advances in neuropsychological tests, as well as the utilization of data collected via cognitive tasks undertaken with computational methods such as machine learning (ML) provide promising opportunities to assist in the early detection of AD. This dissertation develops several computational methods including both conventional ML algorithms and artificial neural networks (ANNs) for early AD detection and differentiation between different stages of clinical presentation of AD. The proposed solutions built on behavioral data from several standard neuropsychological tests, a double cue spatial inhibition of return (IOR) task, or both combined. Data from the Georgetown University Medical Center and a public database from the Alzheimer's Disease Neuroimaging Initiative (ADNI) were used to train and test the models developed in this dissertation. The methodological contributions include investigate and compare different ML algorithms and deep learning (DL) approaches to classify different cognitive groups using standard neuropsychological tests and/or a simple 5-minute cognitive task. Cognitive groups are formed based on the stage of illness, ranging from cognitively normal (CN) to mild cognitive impairment (MCI) to AD. In addition, identifying the MCI individuals at risk of progression is important for clinical management. Another main contribution of this dissertation is to develop a lightweight DL architecture, multilayer perceptron (MLP), to predict the probability of converting from MCI-to-AD within a 3-year follow-up period. The comprehensive results presented in this dissertation demonstrate that the 5-minute cognitive task could produce critical data that is useful in diagnosing MCI/AD either alone or together with neuropsychological data. In particular, MLP with a combination of summarized scores from neuropsychological tests and the spatial (IOR) task achieved ~90% sensitivity, ~92% specificity, and ~92 accuracy%. This indicates its status as an ideal task in clinical practice to be administered to individuals with dementia. Combining features from different standard neuropsychological tests provides elevated performance results compared to a single test using MLP networks. The classification results across different cognitive group pair ranges between ~90% - ~99% for sensitivity, ~89% - ~100% for specificity, and ~90% - ~99% for accuracy, indicating good potential to assist in early AD diagnosis. ANNs demonstrate strong predictive performance in diagnostic classification compared to conventional ML. This indicates the potential for DL to assist early diagnosis across different cognitive groups using the standard neuropsychological tests and/or the simple 5-minute cognitive task.217enThesis