Statistical Methods for Multilevel Multivariate Functional Studies

摘要 虽然成像研究在临床实践和研究中被广泛应用，但神经成像的数量- 以生物标志物为基础的很小。例如，在多发性硬化症免疫调节疗法的临床试验中， 只有常用的影像生物标志物是病变的总体积以及新发和新发肿瘤的数量。 有明显的损伤。这些生物标志物是必不可少的，但不能捕捉到病变的恢复过程， 它被认为在更严重的进展性疾病中会下降。部分或完全恢复 损伤的程度可能取决于大脑的愈合能力和外部因素，如治疗- 或环境和行为暴露。在这项提议中，我们采取了自然的下一步 根据观察到的病变的形成和变化提出MS的影像生物标志物 多序列结构磁共振成像。为了解决这个问题，我们提出了几种通用的方法- 科学问题：1)发展模型和方法，对几幅图像进行纵向分析 相同的大脑；2)识别和估计估计恢复所需的病史长度；3) 研究与已知的疾病生物标志物的相关性(在这种情况下，总体积和数量 新的和增强的病变)；4)开发对成像方案的变化具有健壮性的方法， 在纵向神经成像研究中不可避免地出现；以及5)开发计算工具，以允许 以便在实践中无缝地实施复杂的方法。而我们的科学fic问题是 本文提出的统计方法具有一定的通用性，可适用于各种不同的经度分布。 神经影像研究。例如，有许多正在进行的纵向神经成像研究， 包括ADNI、AIBL、HBC和Mistie，我们的方法可以用来研究微妙的或 病变或白质和灰质信号有较大变化。

Abstract While imaging studies are widely used in clinical practice and research, the number of neuroimaging- based biomarkers is small. For example, in clinical trials of immunomodulatory therapies for MS, the only commonly used imaging biomarkers are the total lesion volume and the number of new and en- hancing lesions. These biomarkers are essential, but do not capture the recovery process of lesions, which is thought to decline in more severe, progressive disease. The partial or complete recovery of lesions may depend both on the ability of the brain to heal and on external factors, such as treat- ment or environmental and behavioral exposures. In this proposal we take the natural next step of proposing imaging biomarkers for MS based on the formation and change of lesions as observed on multi-sequence structural MRIs. To solve this problem we propose to address several general method- ological problems: 1) develop models and methods for the longitudinal analysis of several images of the same brain; 2) identify and estimate the length of history that is necessary to estimate recovery; 3) study the association with known biomarkers of the disease (in this case total volume and number of new and enhancing lesions); 4) develop methods that are robust to changes in imaging protocols that inevitably arise in longitudinal neuroimaging studies; and 5) develop the computational tools that allow for sophisticated methods to be implemented seamlessly in practice. While our scientific problem is focused, the proposed statistical methods are general and can be applied to a wide variety of longitu- dinal neuroimaging studies. For example, there are many ongoing longitudinal neuroimaging studies, including the ADNI, AIBL, HBC, and MISTIE, where our methods could be used to study subtle or large changes in lesions or in white and gray matter intensities.