A large sample machine learning network analysis of vertex cortical thickness measures for high resolution definition of PTSD related cortical structure abnormalities

项目摘要 创伤后应激障碍（PTSD）影响着全球数百万人。现有研究将创伤后应激障碍 症状到皮质结构变化，包括例如前额叶和其它皮质区域的变薄。 PTSD在不同感兴趣区域（ROI）变化的相关性表明涉及多个网络。 ROI结果使用大皮质区域的平均厚度测量，从而定位病灶 厚度的变化非常困难。最近的基于顶点的工作，使用个体下方的厚度测量 顶点表面积约为1 mm2，已开始识别具有局灶性变薄的皮质顶点的限定簇 在创伤后应激障碍患者中，但是顶点之间的变化的关联仍然没有研究。现有研究约30万 样本量小的每个受试者的顶点可能缺乏统计功效、再现性和 解决交叉顶点关系。因此，大样本，基于顶点的工作可以说是需要推进 了解PTSD相关的皮质厚度变化。这将需要在分析和 方法的变革，因为目前的方法不能联合评估皮质厚度， 在研究中，每个受试者包含数十万个顶点， 数千个主题。 基于上述理由，计划中的工作采用了新的方法，首先，汇编现有的最大的 基于顶点的皮质厚度以及相关人口统计学和共病数据库（约19，000例受试者） 用于比较PTSD和非PTSD受试者的数据，第二，应用数据驱动的多顶点模式， 网络分析（MVPNA），以联合识别本地化顶点和顶点簇的分布式网络， 有相关的结构异常可以预测创伤后应激障碍这项工作将通过以下方式促进当前的理解： (1)确定PTSD患者中目前未解决的皮质厚度变化的局灶性部位，（2）提供 对PTSD厚度变化网络的基于顶点的模式的开创性见解，以及（3）开发新的 大样本皮质顶点数据系统评估的MVPNA方法。拟议工作 将对皮质结构变化的基础研究和临床研究产生进一步的影响， 与其他精神和神经疾病一起发生。

Project Summary Posttraumatic stress disorder (PTSD) affects millions of people globally. Existing studies link PTSD symptoms to cortical structural changes including, for example, thinning in prefrontal and other cortical regions. Correlations of PTSD changes in different regions of interest (ROIs) suggest involvement of multiple networks. ROI findings use average measures of thickness across large cortical regions, thus making localization of foci of thickness changes very difficult. Recent vertex-based work, using measures of thickness beneath individual vertex surface areas of ~1 mm2, has begun to identify delimited clusters of cortical vertices with focal thinning in PTSD patients, but associations of changes across vertices remain unstudied. Existing studies of ~300,000 vertices per subject with small sample sizes may lack statistical power, reproducibility and the capacity to resolve cross vertex relationships. Thus large sample, vertex-based work is arguably needed to advance understanding of PTSD related cortical thickness changes. This will require innovations in analyses and transformative changes in approach, because current approaches cannot jointly assess cortical thickness and network associations from data incorporating hundreds of thousands of vertices per subject in studies involving thousands of subjects. With the above rationale, the planned work uses novel approaches, first, to compile the largest existing database (~19,000 subjects) of vertex-based cortical thickness and associated demographic and comorbid data for comparing PTSD and non-PTSD subjects and, second, to apply data driven, multi-vertex pattern and network analysis (MVPNA) to jointly identify localized vertices and distributed networks of vertex clusters that have associated structural abnormalities which predict PTSD. This work will advance current understanding by: (1) identifying currently unresolved focal cortical sites of thickness change in PTSD patients, (2) providing seminal insight into vertex-based patterns of PTSD thickness change networks, and (3) developing new MVPNA approaches for systematic assessment of cortical vertex data from large samples. The proposed work will have further implications for basic research and clinical investigation of cortical structural changes that occur with other psychiatric and neurological disorders.