Statistical methods for large and complex databases of ultra-high-dimensional

摘要 医学影像学是基础科学和临床实践的基石。发现新 疾病的机制和标志物及其对临床实践的重要意义， 大型多中心成像研究正在获取数TB的复杂多模态 几十年来的横截面和纵向成像数据。 由于这些研究的复杂性，对这些研究的数据进行统计分析是具有挑战性的。 所获得的成像数据的结构和超高维度。此外，委员会认为， 解剖学、病理学和成像协议的异质性导致不稳定性， 许多当前最先进的图像分析方法的失败。该基金建议 通过生物医学成像研究人口的统计框架， 用于病理学识别和准确定量的稳健方法，以及 分析工具的横向和纵向检查的病因和 疾病进展。 这些技术将被应用于解决激励大型和多个关键目标， 在约翰斯进行的多发性硬化症和阿尔茨海默病的中心研究 霍普金斯医院，国家神经疾病和中风研究所，以及 地球仪。该项目将创建揭示和量化脑损伤的方法 病理学发病率和轨迹根据这项赠款开发的方法将有针对性地 这些神经成像的目标，但将形成统计图像分析的基础 方法广泛适用于生物医学科学。

Abstract Medical imaging is a cornerstone of basic science and clinical practice. To discover new mechanisms and markers of disease and their crucial implications for clinical practice, large multi-center imaging studies are acquiring terabytes of complex multi-modality imaging data cross-sectionally and longitudinally over decades. The statistical analysis of data from such studies is challenging due to the complex structure of the imaging data acquired and the ultra-high dimensionality. Furthermore, the heterogeneity of anatomy, pathology, and imaging protocols causes instability and failure of many current state-of-the-art image analysis methods. This grant proposes statistical frameworks for studying populations through biomedical imaging, scalable and robust methods for the identification and accurate quantification of pathology, and analytic tools for the cross-sectional and longitudinal examination of etiology and disease progression. These techniques will be applied to address key goals of the motivating large and multi- center studies of multiple sclerosis and Alzheimer's disease conducted at Johns Hopkins Hospital, the National Institute of Neurological Disorders and Stroke, and across the globe. The project will create methods for uncovering and quantifying brain lesion pathology, incidence, and trajectory. Methods developed under this grant will be targeted towards these neuroimaging goals, but will form the basis for statistical image analysis methods applicable broadly in the biomedical sciences.