Statistical Methods for Brain Image Registration and Tensor-Based Morphometry

描述（由申请人提供）：基于张量的形态测量（TBM）是一种越来越流行的脑MRI和DTI数据组分析方法。分析中的主要步骤包括非线性配准，以将每个单独的扫描与公共空间对齐，以及随后的统计分析，以确定组之间的形态差异或纤维结构差异。在这里，我们提出了一种方法，以提高非线性配准和统计分析的TBM。 传统的TBM非线性配准是在T1加权MR图像上执行的，要么单独在分割的2D皮质上，要么在整个3D脑图像上，然后对这些域进行相应的统计分析。迄今为止，这两种选择都没有为整个大脑提供令人满意的解决方案，因为2D皮质TBM忽略了大脑的其余部分，而3D体积TBM难以匹配皮质，并且可能无法很好地匹配白色物质中的神经元纤维结构。在这里，我们描述了一种新的统计非线性配准算法的3D体积的TBM，结合了皮质匹配的优势，在整个大脑体积的3D统计流体配准。此外，我们的目标是通过在从扩散张量成像数据导出的成本函数中添加扩散张量之间的距离来准确地匹配底层纤维结构。 此外，我们建议通过使用变形场的雅可比矩阵中的所有信息以多变量的方式，并通过建立推断，使其可以在体积变化和变形方向方面进行解释，来提高TBM中统计分析的检测能力。 公共卫生相关性：我们从两个方面改进了基于张量的形态测量法用于组分析，首先通过使用皮质，结构MR和DTI信息到组合的皮质和统计流体配准算法中，其次通过使用多元统计方法来分析完整的雅可比矩阵以及其中的体积和方向信息。

DESCRIPTION (provided by applicant): Tensor-Based Morphometry (TBM) is an increasingly popular method for group analysis of brain MRI and DTI data. The main steps in the analysis consist of a nonlinear registration to align each individual scan to a common space, and a subsequent statistical analysis to determine morphometric differences, or difference in fiber structure between groups. Here, we propose a method to improve both the nonlinear registration and statistical analyses for TBM. The traditional nonlinear registration for TBM is performed on T1-weighted MR images, either on the seg- mented 2D cortices alone, or on the whole 3D brain images, followed by corresponding statistical analyses on those domains. To date, neither option provides a satisfactory solution for the entire brain, since 2D cortical TBM ignores the rest of the brain, while 3D volumetric TBM has difficulty matching the cortex and may not match well neuronal fiber structures in the white matter. Here we describe a new statistical nonlinear registration algorithm for 3D volumetric TBM that combines the advantages of cortical matching to those of a 3D statistical fluid registration on the whole brain volume. In addition, we aim to match the underlying fiber structure accurately by adding a distance between diffusion tensors in the cost function derived from diffusion tensor imaging data. Furthermore, we propose to improve the detection power in the statistical analysis in TBM by using all the information available in the Jacobian of the deformation field in a multivariate fashion, and by setting up the inference so that it can be interpreted in terms of both volumetric changes and directions of deformation. PUBLIC HEALTH RELEVANCE: We improve on Tensor-Based Morphometry for group analysis in two ways, first by using cortical, structural MR and DTI information into a combined cortical and statistical fluid registration algorithm, and secondly by using multivariate statistical methods to analyze the full Jacobian matrix and the volumetric and directional information in it.