Date of Award
Doctor of Philosophy
Dr. Alvis Fong
Dr. Ikhlas Abdel Qader
Dr. Shameek Bhattacharjee
Dr. Mohammad Niaz
Prediction, early signs of Alzheimer’s, machine learning framework
Dementia is a collective term used to indicate a loss of memory functions with the presence of at least one additional loss of a major cognitive ability that hinders a person’s previous level of functioning. Studies show that dementia is highly age- associated and that the most common cause of dementia is Alzheimer’s disease. Early recognition of Alzheimer’s disease, before irreversible damage to the brain has already occurred, is paramount to slowing or preventing the disease. Therefore, algorithms for the prediction of early signs of dementia are essential. Machine learning approach has been reported to use several data sources such as neuroimaging, biological, neuropsychological, or a combination of different biomarkers to generate a system to predict the onset of Alzheimer’s.
This thesis introduced a Risk Monitoring System that integrates heterogeneous multisource Alzheimer’s disease data. It identified the most significant predictors in terms of the clinical and cognitive features associated with early signs of AD. Appropriate statistical survival analysis methods (semi- parametric and parametric methods) integrated with machine learning approaches such as Support Vector Machine and Random Survival Forest have been investigated. Scores from the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA) used in an integrated framework along with individuals’ demographic data such as age, sex, education level and race to develop a system for prediction with higher accuracy rates than those of current individual methods or systems.
Working with big data set, the biggest challenges is that the size of the dataset does not guarantee the validity of the results. This thesis thus attempts to develop predicting system focusing on the new facilities that big data offers; optimally extract information from massive data and having inference on the factors that affect the conversion between AD stages; mild, moderate and severe. Utilizing Big data tools helped improve the efficiency and speed of applied algorithms, especially those that need large computation (i.e., Survival Decision Tree). The tested algorithms (COX, SDT, and RNNs) showed a good prediction accuracy, however, RNN (LSTM and GRU) outperforms the other two algorithms and state-of-art algorithms.
Restricted to Campus until
Alsaedi, Abdalrahman, "On Prediction of Early Signs of Alzheimer’s— A Machine Learning Framework" (2021). Dissertations. 3717.