Research Project
9559512
SUMMARY This Phase I grant will use innovative machine learning approaches and brain image data to address the critical need to identify the presence and extent of Alzheimer?s disease pathology for clinical diagnosis and treatment evaluation. Misdiagnosis rates currently exceed 20% and diagnosis is not available in early stages of disease, impeding patient care and the development of effective treatments. Combining innovations in machine learning with recently available information from tau PET imaging, classifier and regression models will be developed that can predict amyloid plaque and tau distribution, using structural and/or functional magnetic resonance imaging (MRI) sequences. These advances can greatly improve the early, accurate diagnosis of Alzheimer?s disease, enable the selection of patients for clinical trials, and aid in the development of effective new therapeutics. In the first Specific Aim, multivariate machine learning classifiers will be developed using structural MRI, functional MRI (ASL), and FDG PET as a comparator to characterize amyloid and tau pathology and disease stage in patients with Alzheimer?s disease ranging from presymptomatic through dementia stages. Second, within-classifier and across-classifier performance will be evaluated with respect to the objectives to: discriminate subjects with amyloid and tau pathology; provide a metric of tau burden and spatial distribution using MRI and FDG PET modalities; and identify neurodegenerative patterns that may reflect differences in clinical severity among patients with the same tau burden and distribution. In addition, the relationship between classifier scores and cognitive endpoints will be evaluated. Third, similar classifiers will be developed using structural MRI and FDG PET to characterize amyloid and CSF tau burden in a genetically predisposed, early onset AD population, and findings compared to those in the late onset AD population. This work makes use of data acquired in the Alzheimer?s Disease Neuroimaging Initiative (ADNI), the DIAN study of early onset autosomal dominant Alzheimer?s Disease (ADAD), and additional data sets. Innovations of this work include: prediction of tau burden and distribution using imaging measures of neurodegeneration; application of our machine learning methods and optimization to recently available modalities and measures; unique approaches in machine learning optimization and classifier design; and the inclusion of Late Onset Alzheimer?s Disease (LOAD) and ADAD data sets and initial comparison between these forms of AD. Achievement of these aims will result in diagnostic and prognostic image analysis tools to aid in accurate diagnosis and prognosis supporting patient care and the clinical evaluation of therapeutic interventions.
