Design, implementation and interpretation of algorithms to predict the progression of Alzheimer's disease

Abstract

Alzheimer's Disease (AD) is a serious public health issue that results in significant social and financial burdens on the individuals and communities impacted. In order to tackle this public health crisis it is critical that the clinical and computational research communities collaborate to identify possible causes of this progressive memory disease. Close collaboration between these two communities has the potential to result in promising therapeutic treatments for AD and other health conditions. This dissertation presents a collection of algorithms and associated derivations designed to predict the progression of AD using multi-task and structured regularization techniques, clustering membership by way of nonnegative matrix factorization, and COVID-19 clinical outcome prediction using multi-instance learning methods. This work presents novel algorithms for handling multimodal and longitudinal data and details approaches for multitask and multi-instance learning techniques that can be applied in other fields. Extensive discussions on algorithm predictive performance, interpretability, and implementation are provided for each method and are designed to serve as a framework for future research.