ADNI Psychometrics: Machine Learning to discern Natural History Aim 2 Supplement

点击翻译按钮获取中文摘要

Project Summary/Abstract This supplement is to a funded R01 called ADNI Psychometrics. This Supplement builds on the Second Specific Aim of the funded parent grant. That Aim focused on characterizing brain structure and functioning for people with different cognitively- defined subgroups of Alzheimer's disease. The Supplement adds one technique for analyzing the longitudinal structural data we are already analyzing. The new technique for structural data is machine learning. We have the opportunity to collaborate with a talented faculty member in Biomedical Informatics who has specific expertise in machine learning approaches to anatomical data (J Gennari). Dr. Genarri will supervise machine learning approaches to complement the various analytical approaches we already have underway for the longitudinal structural imaging data of Aim 2. Longitudinal imaging data are particularly significant, as any differences we find in the evolution of brain structure over time across subgroups supports the notion that the subgroups have distinct natural histories, which in turn goes a long way towards the provocative conclusion that these subgroups of “Alzheimer's disease” represent distinct conditions. Machine learning approaches to these data were not envisioned in the initial proposal, but represent a particularly valuable complementary approach that may identify similarities and differences in trajectories of the evolution of brain structure that would not be apparent using the more traditional analytic pipelines we outlined in the proposal. This then is the perfect fit for an Administrative Supplement – this is an opportunity to enhance the value of the parent study by adding new expertise to investigate in a complementary and valuable fashion a question that was already addressed by the parent grant. This Supplement builds on the same infrastructure and questions asked in Aim 2, but augments our analytical armamentarium with novel machine learning approaches.