Title
Brain Metabolic Responses To Alzheimer Pathologies With Molecular Imaging And Machine Learning
Date of Award
2022
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Graduate Group
Bioengineering
First Advisor
Ilya M. Nasrallah
Second Advisor
David A. Wolk
Abstract
Alzheimer Disease (AD) is defined by amyloid (A) and tau (T) pathologies, with T better correlated to neurodegeneration (N) than is A. However, T and N have complex regional relationships in part related to non-AD factors that may influence N. Using machine learning, we assessed heterogeneity in 18F-Flortaucipir vs. 18F-Fluorodeoxyglucose positron emission tomography as markers of T and neuronal hypometabolism (NM) in 289 symptomatic patients from the Alzheimer Disease Neuroimaging Initiative (ADNI) and 115 cognitively normal older adults from the Harvard Aging Brain Study (HABS). Fromboth cohorts, we identified six T/NM clusters with differing limbic and cortical patterns. The canonical group was defined as the T/NM pattern with the lowest regression model residuals, while non-canonical groups reflected either resilience or susceptibility, with either less or greater hypometabolism than expected relative to T. Resilient groups displayed better cognition and less copathology-related factors than the canonical group. Susceptible groups exhibited worse cognitive decline and had imaging and clinical measures consistent with the presence of copathologies, including factors associated with vascular, α-synuclein and TDP-43 pathologies. Mismatch analyses were applied with a loglinear model and compared to a generative adversarial network with dual contrastive learning objectives. We performed theoretical and empirical investigations to optimize this contrastive learning model. Our proof-of-concept experiments demonstrate the advantage of multi-domain contrastive losses, the utility of training set and contrastive sampling diversity and the ability of image-to-image translation models to accurately map between T and NM domains. These findings provide the basis for further biological and statistical inquiries into the translation accuracy and reconstruction error of models that map between types of images in AD. Together, T/NM mismatch in AD reveals distinct imaging signatures with pathobiological and prognostic consequences that may improve clinical trial design, diagnosis and management of patients living with neurodegenerative diseases.
Recommended Citation
Duong, Michael Tran, "Brain Metabolic Responses To Alzheimer Pathologies With Molecular Imaging And Machine Learning" (2022). Publicly Accessible Penn Dissertations. 5184.
https://repository.upenn.edu/edissertations/5184