Ferreira Fábio S, Pereira João M S, Duarte João V, Castelo-Branco Miguel
Institute for Biomedical Imaging and Life Sciences (IBILI), Faculty of Medicine, University of Coimbra, Portugal.
Laboratory of Biostatistics and Medical Informatics, Faculty of Medicine, University of Coimbra, Portugal.
Open Neuroimag J. 2017 May 29;11:32-45. doi: 10.2174/1874440001711010032. eCollection 2017.
Although voxel based morphometry studies are still the standard for analyzing brain structure, their dependence on massive univariate inferential methods is a limiting factor. A better understanding of brain pathologies can be achieved by applying inferential multivariate methods, which allow the study of multiple dependent variables, different imaging modalities of the same subject.
Given the widespread use of SPM software in the brain imaging community, the main aim of this work is the implementation of massive multivariate inferential analysis as a toolbox in this software package. applied to the use of T1 and T2 structural data from diabetic patients and controls. This implementation was compared with the traditional ANCOVA in SPM and a similar multivariate GLM toolbox (MRM).
We implemented the new toolbox and tested it by investigating brain alterations on a cohort of twenty-eight type 2 diabetes patients and twenty-six matched healthy controls, using information from both T1 and T2 weighted structural MRI scans, both separately - using standard univariate VBM - and simultaneously, with multivariate analyses.
Univariate VBM replicated predominantly bilateral changes in basal ganglia and insular regions in type 2 diabetes patients. On the other hand, multivariate analyses replicated key findings of univariate results, while also revealing the thalami as additional foci of pathology.
While the presented algorithm must be further optimized, the proposed toolbox is the first implementation of multivariate statistics in SPM8 as a user-friendly toolbox, which shows great potential and is ready to be validated in other clinical cohorts and modalities.
尽管基于体素的形态测量学研究仍是分析脑结构的标准方法，但其对大量单变量推断方法的依赖是一个限制因素。通过应用推断性多变量方法可以更好地理解脑部病变，这种方法允许研究同一受试者的多个相关变量及不同成像模态。
鉴于SPM软件在脑成像领域的广泛应用，本研究的主要目的是在该软件包中实现大规模多变量推断分析工具箱。将其应用于糖尿病患者和对照组的T1和T2结构数据。将此实现方法与SPM中传统的协方差分析和类似的多变量广义线性模型工具箱（MRM）进行比较。
我们实现了新的工具箱，并通过对28名2型糖尿病患者和26名匹配的健康对照者进行研究来测试它，使用来自T1加权和T2加权结构MRI扫描的信息，分别使用标准单变量体素形态测量法以及同时使用多变量分析法。
单变量体素形态测量法主要重现了2型糖尿病患者基底神经节和岛叶区域的双侧变化。另一方面，多变量分析重现了单变量结果的关键发现，同时还揭示丘脑是另外的病变部位。
虽然所提出的算法必须进一步优化，但所提议的工具箱是SPM8中首个作为用户友好型工具箱的多变量统计实现，它显示出巨大潜力，并准备好在其他临床队列和模态中进行验证。
