The Neuroimaging Brain Chart Software Suite

本研究建议对神经影像脑图(NIBCH)软件进行提炼、整合和推广 工具箱和机器学习(ML)模型库，一个软件组件生态系统，支持 跨研究的建设性整合、统计协调和以ML为中心的数据分析。NIBCH 通过将多模式脑MRI数据映射到紧凑的坐标，实现对这些数据的大规模分析 由我们的ML模型库实现的信息性神经成像签名系统。这件事的轴心 坐标系代表两种类型的信息：1)各种结构(sMRI和dMRI)和 功能连接(RsfMRI)成像衍生表型(IDP)，如多尺度脑包 和脑网络；2)基于ML的复杂成像特征(ML-IDP)，捕捉多变量 反映大脑老化、神经退行性变以及 神经精神障碍，以前是从仔细处理和精选的数据中得出的 超过65,000个个体。使用我们的软件工具箱(TBX)，研究人员将能够将新数据映射到 NIBCH，并因此使用在NIBCH中训练的ML-IDP模型，以及对NIBCH执行统计测试 并将它们的结果与使用相同TBX的其他研究的结果进行比较。该软件 套件将包括一套集装箱化的预处理和分析管道，以及统计 协调和ML推理工具箱，可通过独立的python前端访问 可视化，作为基于云的容器，并通过我们的高性能支持的Web界面 计算集群。讨论了几个传播计划，包括GitHub用户社区、教程 在重要的技术和临床会议上，并支持本地或云上的独立管道， 以及基于网络的协调和ML推理模块的访问。 我们计划的首要目标是提供软件工具，使用户能够 促进积极发展的基于社区的立体神经成像系统，该系统将利用 机器学习模型，提供丰富、精确、紧凑、简明和信息丰富的 大脑结构、功能和连通性。

This study proposes to refine, integrate and disseminate the NeuroImaging Brain Chart (NIBCh) software toolbox and machine learning (ML) model library, an ecosystem of software components enabling constructive integration, statistical harmonization, and ML-centric data analyses across studies. NIBCh enables large-scale analyses of multi-modal brain MRI data by mapping such data into a compact coordinate system of informative neuroimaging signatures implemented by our library of ML models. The axes of this coordinate system represent two types of information: 1) a variety of structural (sMRI and dMRI) and functional connectomic (rsfMRI) imaging derived phenotypes (IDPs), such as multi-scale brain parcelations and brain networks; 2) complex ML-based imaging signatures (ML-IDPs), which capture multi-variate imaging patterns that reflect the heterogeneity of brain aging, neurodegeneration, as well as of neuropsychiatic disorders and have been previously derived from carefully processed and curated data of over 65,000 individuals. Using our software toolboxes (Tbx), researchers will be able to map new data into NIBCh, and hence to use ML-IDP models trained in NIBCh, as well as perform statistical tests against NIBCh normative ranges and compare their results with those of other studies using the same Tbx. The software suite will include a set of containerized pre-processing and analysis pipelines, as well as statistical harmonization and ML inference toolboxes, which will be accessible via a standalone python front-end visualization, as cloud-based containers, and via a web-interface supported by our high-performance computing cluster. Several dissemination plans are discussed, including a github user community, tutorials at major technical and clinical meetings, and support of both standalone pipelines locally or on the cloud, and web-based access of harmonization and ML inference modules. The over-arching primary goal of our program is to provide the software tools that will allow users to contribute to an actively growing community-based dimensional neuroimaging system that will utilize machine learning models to provide rich, yet precise, compact, concise, and informative representations of brain structure, function and connectivity.