Using connectomics to characterize risk for Alzheimer's Disease

点击翻译按钮获取中文摘要

PROJECT DESCRIPTION The prevalence of Alzheimer’s Disease (AD) is expected to increase significantly in the next 30 years. While research efforts continue to focus on the causes of AD and to develop effective medical treatments, there is also a pressing need to characterize risk for AD before the disease is diagnosed. There may be subtle changes in brain and cognitive function that are detectable before major symptoms emerge. One approach for characterizing these vulnerabilities is the use of functional and structural neuroimaging to identify risk profiles for AD. The present study proposes and tests a model of AD pathology using neuroimaging network analysis and machine learning approaches to provide insight on widespread changes in information processing in the AD brain. The proposed model hypothesizes that some aspects of network information processing reflect neurodegeneration and cognitive decline associated with AD pathology (e.g., hub connectivity and global connectivity) whereas other network properties reflect attempts to compensate for compromised information processing (e.g., diffusion of information). In addition, this proposal compares the efficacy of models with respect to discriminating diagnostic categories (e.g., machine learning classification of AD and clinically normal subjects) versus isolating underlying dimensions of AD cognitive decline (e.g., machine learning prediction of memory and language scores from network features). Finally, this study will determine whether features of AD pathology are present in an at-risk sample of subjects; namely, individuals diagnosed with amnestic mild cognitive impairment (aMCI). This will be examined by transferring the AD network models to aMCI subjects and testing whether the model can discriminate aMCI from clinically normal matched controls and whether the model can predict scores on cognitive tests. The analytic approach will using resting state fMRI data as a primary assay of network integrity, but diffusion imaging and task fMRI data will also be examined in an exploratory aim. The general approach will recruit individuals with AD, aMCI and clinically normal matched controls for each diagnostic group. The groups that are compared directly will be matched for amyloid status, as indicated by florbetapir PET imaging. The novel contributions of this project include (a) testing a network model of AD pathology that unifies various measures of network functioning, (b) comparing efficacy of modeling with respect to delineating diagnostic categories versus capturing underlying cognitive dimensions of AD, and (c) transferring the AD model to an at-risk group to assess disease vulnerability. This latter innovation can be applied in future studies to any group of subjects that is defined at risk, such as those with genetic vulnerability or positive family history. The present study also sets the stage for a subsequent longitudinal follow-up study to validate whether individuals identified at risk using network modeling actually convert to AD. Ultimately, network modeling of this sort may be used as a relatively less expensive and less invasive biomarker than PET imaging, which could lead to earlier treatments and interventions.