Discussion of LESA: Longitudinal Elastic Shape Analysis of Brain Subcortical Structures
Discussion of LESA: Longitudinal Elastic Shape Analysis of Brain Subcortical Structures
复制标题
LESA的讨论：大脑皮层下结构的纵向弹性形状分析
DOI：
10.1080/01621459.2022.2120399
复制
发表时间：
2023
影响因子：
3.7
通讯作者：
Lila, Eardi
中科院分区：
文献类型：
--
作者：
Aston, John A.;Lila, Eardi
我们祝贺作者的一个有趣的文件，结合了形状分析的想法与纵向数据分析。随着涉及成像数据的实验越来越复杂，包括一些重要的纵向成像研究，尤其是那些提出的，重要的是开发方法，允许考虑分析这些数据，这篇文章走了很长的路，以帮助解决这些问题。形状本身的注册是通过一种方法完成的，该方法首先在Srivastava等人的统计功能数据上下文中获得了突出地位。（2011），其中平方根速度函数扮演着与平方根正常场（SRNF）类似的角色（杰明等人，2012）。这种方法具有非常有益的特性，允许定义本文中针对SNRF方法所描述的重新参数化不变度量。非常有趣的是，使用混合效应模型的方法再次与功能数据和配准文献相关（Tucker、Wu和Srivastava 2013; Hadjipantelis等人2015; Marron等人2015）。然而，与许多纵向数据分析的情况一样，特别是在稀疏观测的情况下，将数据的时间分量视为函数数据的一种形式是非常自然的，因此PACE方法（Yao，Müller和Wang 2005）是一个更合适的框架来模拟这些数据。对最初的计算机设备行动伙伴关系设想有许多可能的扩展，如果与应用有关，所有这些都可以实施。当然，这也会产生一些具有挑战性的理论问题，特别是考虑到形状空间的非欧几里得性质。
We congratulate the authors on an interesting paper which combines the ideas of shape analysis with longitudinal data analysis. With the increasing complexity of experiments involving imaging data, including a number of important longitudinal imaging studies, not least those presented, it is important to develop methodology that allows for a considered analysis of such data, and this article goes a long way to helping address these issues.There are a number of important choices made in the paper. The registration of the shapes themselves is done through an approach which first received prominence in a statistical functional data context in Srivastava et al.(2011), where the squareroot velocity function plays an analogous role to the squareroot normal field (SRNF)(Jermyn et al. 2012). This approach has highly beneficial properties that allow for the definition of reparameterization invariant metrics as described for the SNRF approach in this article. It is very interesting to see the approach using mixed-effect models which again has connections to the literature in functional data and registration (Tucker, Wu, and Srivastava 2013; Hadjipantelis et al. 2015; Marron et al. 2015). However, as with many instances of longitudinal data analysis, particularly with sparse observations, it is very natural to consider the temporal components of the data to be a form of functional data, and hence the PACE approach (Yao, Müller, and Wang 2005) is a more suitable framework to model these in. There are many possible extensions to the original PACE idea, and all these could be implemented if relevant to the application. Of course, this would also set up a number of challenging theoretical questions, particularly given the non-Euclidean nature of shape spaces.