Artificial Intelligence Modeling of Alzheimer's Disease and Environmental Science

Graduation Year

2024

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Mathematics and Statistics

Major Professor

Chris P. Tsokos, Ph.D.

Committee Member

Kandethody M. Ramachandran, Ph.D.

Committee Member

Lu Lu, Ph.D.

Committee Member

Yicheng Tu, Ph.D.

Keywords

Alzheimer’s Diseas, Deep Learning, Desirability Function, Global Warmin, Mild Cognitive Impairmen, Supervised Learning

Abstract

A data-driven statistical model operates as a mathematical illustration of a tangible issuefaced in reality, enabling the creation of predictions or decisions based on data. The process of utilizing probability and statistical principles in statistical models is essential for deriving meaningful conclusions from the data. The research conducted in this dissertation utilizes artificial intelligence (AI) models to integrate findings across health and environmental sci- ence.

The first research study in this dissertation, Alzheimer’s disease is a mental health issue and a brain aging dilemma that makes it difficult for older people to complete daily tasks without assistance. The physician uses several risk factors to evaluate the patient’s impaired memory and for the development of Alzheimer’s disease, such as neuropsychological test measurement, cerebrospinal fluid (CSF), volumetric measurements from resonance imaging (MRI), positron emission tomography (PET), and genetic biomarkers. In this study, we developed an accurate data-driven analytical model that identifies significant risk factors and interactions that significantly contribute to predicting the time of conversion to AD from MCI patients within an 8-year follow-up, using the baseline dataset from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). We also ranked the risk factors based on the contribution to the conversion time to the AD of patients diagnosed with MCI. The proposed model will help examine the effectiveness and safety of prospective treatments for individuals with mild cognitive impairment.

Further, based on baseline measurements, we developed a statistical classification model using deep learning-based convolutional neural networks (CNN) to correctly distinguish pa- tients with Alzheimer’s disease (AD) from those with mild cognitive impairment (MCI). To achieve better classification performance, we utilize the Focal Loss function to address the imbalance between majority and minority classes in the training process. Thus, the proposed analytical mode recall of 89% for converted to Alzheimer’s Disease level, which means 89% of progression cases from Mild Cognitive Impairment (MCI) to Alzheimer’s Disease (AD) can be detected by this model.

Another research area in this dissertation is in the area of environmental science; Global warming is an environmental issue defined by the interaction between atmospheric temper- ature and carbon dioxide. It is caused by human activities that generate these emissions, including the combustion of fossil fuels for transportation and industrial processes, deforesta- tion, and cement production. To the best of our knowledge, no statistical model developed within the suggested logical structure in the African context. Furthermore, research indicates that Africa is anticipated to be the continent most impacted by climate change. Thus, it was essential to construct a statistical model aimed at identifying the significant risk factors and their interactions while also ranking them based on their contribution to carbon dioxide emissions across Africa.

Finally, we proposed a real data-driven analytical model for fossil-fuel (CO2) emissions in Africa, utilizing data obtained from the Carbon Dioxide Information Analysis Center (CDIAC), and this actual annual data has been collected from 1963 to 2014. The proposed model is of high quality, satisfies all necessary assumptions, well-validated, and predicts the (CO2) emissions with a high degree of accuracy. It identifies the significant risk factors and interactions among them and ranks them according to their contribution to carbon dioxide emissions in Africa. The developed statistical model is evaluated in comparison to alternative penalization methods due to the presence of multicollinearity among the risk factors, which yielded promising outcomes as indicated by the root mean square errors (RMSE) statistic. The results derived from the proposed model are also compared with previous findings from various countries worldwide. We further performed an optimization analysis of the carbon dioxide (CO2) emissions using the desirability function approach, obtaining the optimum value of carbon dioxide emission and the optimal values of the risk actors needed to minimize the carbon dioxide emission in Africa. The minimum value of the (CO2) was obtained along with a 95% confidence region, as well as surface response plots to assess the bivariate interaction effect of the risk factors on the (CO2) emission.

This document is currently not available here.

Share

COinS