Topological Machine Learning to Characterize Higher-Order Brain Connectivity Alterations in Alzheimer's Disease - Current and upcoming clinical treatments for Alzheimer’s disease (AD) rely on early intervention during the prodromal stage of mild cognitive impairment (MCI) to alter prognosis. In order to achieve this, noninvasive imaging biomarkers are being developed to discern the progression of disease. One such class of biomarkers are networkbased biomarkers, which detect topological alterations in structural and functional brain networks that are related to disease stages and may additionally provide insight into the disease mechanism at a systems level. However, literature on network-based biomarkers in AD has offered diverging and conflicting interpretations of brain network topological alterations throughout the disease’s progression, leading to a significant barrier of progress in the field. The discrepancies are attributed to differences in brain network construction methods across studies, such as a priori network edge thresholding, which are not standardized and have a confounding impact on downstream observations, and thus demonstrates a clear and present need for more robust methodology in network-based analysis of neurodegenerative disease. Our project addresses the issue by leveraging threshold-free topological data analysis (TDA), which promises to be more robust than traditional graph theory for measuring topology by circumventing a priori thresholding. Firstly, we will investigate the applicability of TDA-based statistical models, comparing properties of topological profiles to observe robust brain network topological differences among cognitively normal (CN), MCI, and AD subjects, and associating these topological properties with other diagnostic biomarkers. Secondly, we will use novel machine learning methodology incorporating features derived from TDA in a classification paradigm, using properties of brain network topology to predict diagnostic classes of CN, MCI, and AD. The project will further establish the robust methodology of TDA in studying AD, as well as constructing novel network-based biomarkers for early identification of the disease. In addition, this NRSA F31 proposal offers novel training opportunities for the applicant in areas of applying advanced machine learning, artificial intelligence, statistical analysis, and computational approaches for analyzing big biobank and neuroimaging data to advance the study of AD and AD Related Dementia (ADRD), along with the training in professional development that will further prepare the trainee for the next phase of his career as an independent researcher. A world-class mentoring team of sponsor and co-sponsor will provide complementary advising and research resources as the NRSA predoctoral trainee gains independence over the course of this project.
