ADInsight: A Multimodal and Explainable Framework for Alzheimer's Disease Progression and Conversion Prediction
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Subject
Abstract
ADInsight represents the crux of this dissertation, introducing an integrated and explainable framework centered on predicting Alzheimer's disease (AD) conversion, particularly for those at the early stage of mild cognitive impairment (EMCI). Beginning with an examination of models grounded in individual research modalities, such as clinical data and advanced imaging, the research underscores the potential and limitations of singular approaches. As a response to these findings, this dissertation introduces a multimodal ensemble conversion prediction model that combines Diffusion Tensor Imaging (DTI) scans with clinical data. This ensemble not only increases the accuracy of predictions but is also notable for its dedication to explainability, bridging the gap between intricate neural network predictions and understandable medical interpretations. Upon further exploration a unique framework is revealed, combining the advantages of Random Forest Regression alongside the latest over-sampling methods. This framework unravels the intricacies of AD's nonlinear progression, leading to the formulation of patient progression groupings. The dissertation is then concluded with the Cognitive Visual Recognition Tracker (CVRT) application. This application marks an exploration into cognitive focus and visual identification, which are essential elements in the development of Alzheimer's disease. Benefiting both clinicians and patients, CVRT paves the way for innovative treatment strategies. In summary, our ADInsight framework provides a novel approach to understanding and predicting the progression of AD, providing a beacon of hope and knowledge in the ongoing struggle against this debilitating condition.
Table of Contents
Introduction -- Feature-based random forest model -- Diffusion tensor imaging (DTI) deep neural network (DNN) model -- Multimodality ensemble model -- Charting ad progression: time-to-event predictive models and novel categorization -- Clinical decision support application -- Conclusions
DOI
PubMed ID
Degree
Ph.D. (Doctor of Philosophy)