Doctor of Philosophy (PhD)
First Advisor's Name
First Advisor's Committee Title
Second Advisor's Name
Second Advisor's Committee Title
Third Advisor's Name
Third Advisor's Committee Title
Fourth Advisor's Name
Fourth Advisor's Committee Title
Fifth Advisor's Name
Fifth Advisor's Committee Title
Sixth Advisor's Name
Sixth Advisor's Committee Title
Deep learning, Alzheimer's disease, Multimodal data, Multi-class Classification, Early diagnosis, Missing data challenge, Machine learning, Imputation methods, Neural Networks, Hippocampus segmentation, Volume estimation
Date of Defense
One of the challenges facing accurate diagnosis and prognosis of Alzheimer’s Disease (AD) is identifying the subtle changes that define the early onset of the disease. This dissertation investigates three of the main challenges confronted when such subtle changes are to be identified in the most meaningful way. These are (1) the missing data challenge, (2) longitudinal modeling of disease progression, and (3) the segmentation and volumetric calculation of disease-prone brain areas in medical images. The scarcity of sufficient data compounded by the missing data challenge in many longitudinal samples exacerbates the problem as we seek statistical meaningfulness in multiclass classification and regression analysis. Although there are many participants in the AD Neuroimaging Initiative (ADNI) study, many of the observations have a lot of missing features which often lead to the exclusion of potentially valuable data points that could add significant meaning in many ongoing experiments. Motivated by the necessity of examining all participants, even those with missing tests or imaging modalities, multiple techniques of handling missing data in this domain have been explored. Specific attention was drawn to the Gradient Boosting (GB) algorithm which has an inherent capability of addressing missing values. Prior to applying state-of-the-art classifiers such as Support Vector Machine (SVM) and Random Forest (RF), the impact of imputing data in common datasets with numerical techniques has been also investigated and compared with the GB algorithm. Furthermore, to discriminate AD subjects from healthy control individuals, and Mild Cognitive Impairment (MCI), longitudinal multimodal heterogeneous data was modeled using recurring neural networks (RNNs). In the segmentation and volumetric calculation challenge, this dissertation places its focus on one of the most relevant disease-prone areas in many neurological and neurodegenerative diseases, the hippocampus region. Changes in hippocampus shape and volume are considered significant biomarkers for AD diagnosis and prognosis. Thus, a two-stage model based on integrating the Vision Transformer and Convolutional Neural Network (CNN) is developed to automatically locate, segment, and estimate the hippocampus volume from the brain 3D MRI. The proposed architecture was trained and tested on a dataset containing 195 brain MRIs from the 2019 Medical Segmentation Decathlon Challenge against the manually segmented regions provided therein and was deployed on 326 MRI from our own data collected through Mount Sinai Medical Center as part of the 1Florida Alzheimer Disease Research Center (ADRC).
Aghili, Maryamossadat, "Deep Learning for Multiclass Classification, Predictive Modeling and Segmentation of Disease Prone Regions in Alzheimer’s Disease" (2021). FIU Electronic Theses and Dissertations. 4843.
In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/
This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).