Manifold-valued statistical models for longitudinal morphometic analysis in preclinical Alzheimer's disease (AD)

项目摘要 在临床前阶段定量表征早期阿尔茨海默病（AD）病理学的能力是 这是早期干预的关键一步，包括疾病改善治疗和设计有效的临床试验 来测试治疗效果该项目的重点是推导统计图像分析方法，用于识别 早期阶段的形态学变化与AD病理学的直接指标（如 淀粉样蛋白沉积）和各种风险因素，例如认知能力低下的中年人的家族史 健康.拟定分析将在迄今为止组装的最大临床前AD队列中进行， 有助于阐明临床认知成像AD表型是如何出现在无症状个体中的， AD.我们分析的核心是一组算法，允许直接对强大的“流形值”进行操作。 并因此在拾取真实的时产生高灵敏度， 疾病过程的统计学弱多模态模式。 假设：1）检测临床前AD受试者中的形态测量变化与 淀粉样蛋白负担和各种风险因素是可能的，使用新的算法，直接与代表性 从纵向成像数据导出的变形场的雅可比矩阵。2)进行这种分析 在具有确定的AD病理学的无症状的有风险的临床前AD个体的大型多中心队列中， a）揭示了痴呆症发生15年以上时早期疾病过程的重要见解，B）提供了 临床前AD生物标志物和c）提供用于在个体水平预测临床终点的框架 科目 具体目标：1）开发新的算法，用于执行流形表示的统计分析， 与代表风险因素，AD病理学标志物， 和临床/认知测量。2)对两个独立的临床前AD进行端到端分析 队列，以确定形态学变化与各种预测因子以及测试/重测的关系 跨站点验证。3)分析迄今为止最大的临床前AD队列，以表征 对中年晚期个体的临床终点的整个预测谱。4)提供工业- 加强开源软件，实现全套模型，与现有软件集成 工具箱，用于工作站、高吞吐量集群或云上的部署。 意义：该项目将对脑成像研究的三个不同领域产生重大影响： 1)临床认知成像AD表型如何在无症状个体中出现的表征 AD的最早阶段，2）用于神经成像中的形态变化分析的严格算法， 神经科学，3）开放源代码的端到端的算法实现，供社区内使用。

Project Summary The ability to quantitatively characterize incipient Alzheimer's disease (AD) pathology in its preclinical stage is a critical step for early interventions involving disease modifying therapy and for designing efficient clinical trials to test therapy efficacy. This project focuses on deriving statistical image analysis methods for identifying the relationship of morphometric changes in this early stage with direct indicators of AD pathology (such as amyloid deposition) and various risk factors such as family history in late midlife adults who are cognitively healthy. The proposed analysis will be conducted on the largest preclinical AD cohort assembled to date and help elucidate how clinical-cognitive-imaging AD phenotypes emerge in asymptomatic individuals at risk for AD. The core of our analyses is a set of algorithms that allow operating directly on powerful “manifold-valued” representations of morphometric change and consequently yield high sensitivity in picking up real but statistically weak multi-modal patterns of the disease process. Hypothesis: 1) Detecting precise associations between morphometric changes in preclinical AD subjects with amyloid burden and various risk factors is possible using new algorithms that work directly with representations of the Jacobians of the deformation fields derived from longitudinal imaging data. 2) Conducting such analyses on a large multi-site cohort of asymptomatic at-risk preclinical AD individuals with identified AD pathology will a) reveal important insights into early disease processes when dementia is 15+ years away, b) provide preclinical AD biomarkers and c) provide frameworks for predicting clinical endpoints at the level of individual subjects. Specific Aims: 1) To develop new algorithms for performing statistical analysis of manifold representations of morphometric changes concurrently with multiple covariates representing risk factors, AD pathology markers, and clinical/cognitive measures. 2) Conducting an end-to-end analysis of two independent preclinical AD cohorts to identify the relationship of morphometric changes with various predictors as well as test/retest validation across sites. 3) Analyzing the largest preclinical AD cohort to date for characterizing the relationship of the entire spectrum of predictors to clinical endpoints for late midlife individuals. 4) Providing industry- strength open-source software implementing the full suite of models, integrated with existing software toolboxes, for deployments on a workstation, a high throughput cluster or the cloud. Significance: This project will have a significant impact across three distinct areas of brain imaging research: 1) characterization of how clinical-cognitive-imaging AD phenotypes emerge in asymptomatic individuals in the earliest stages of AD, 2) rigorous algorithms for morphometric change analysis in neuroimaging and neuroscience, 3) open-source end-to-end implementations of the algorithms for use within the community.