A High Performance Research Image Repository (RIR) for the Washington University Center of High Performance Computing (CHPC)

项目摘要/摘要： 我们建议建立一个研究图像库(RIR)来存储大量的生物医学成像数据。 RIR将包括华盛顿大学当地生产的数据集：Connectome协调 设施(CCF)(本身包括人类连接组项目(HCP)青年研究，寿命 相关项目、疾病相关项目和各类与HCP有关的项目)、阿尔茨海默病骑士 疾病研究中心(ADRC)，青少年大脑认知发展(ABCD)研究， 全面的神经肿瘤学数据存储库(CONDR)和基于临床的PACS图像存储库。 此外，外部数据收集的副本，如英国生物库，阿尔茨海默氏症 将维持神经成像倡议(ADNI)和癌症图像档案馆(TCIA)。RIR包括一个 数据管理软件解决方案，将引入许多新功能(如丰富数据标记 数据集和高级搜索功能)，并将使我们能够利用现有存储，包括 高性能计算(CHPC)1.4PB的BeeGFS暂存、固态NVMe驱动器 集成到计算节点中，以及10PB的基于ZFS的存储。所有存储都将呈现给用户 作为单个文件系统，而数据将根据存储迁移到不同的性能层 对数据集或处理算法的要求。RIR将被整合到CHPC中用于数据 正在处理。该提案还包括两台NVIDIA DGX A100 GPU服务器，提供最先进的GPU- 基于处理能力。将高质量、多样化的生物医学成像数据与Next- 发电计算能力将对生物医学成像处理管道和 在图像处理的深度学习这一新兴领域，其影响最为深远。

Project Summary/Abstract: We propose to build a Research Image Repository (RIR) to house large collections of biomedical imaging data. The RIR will include datasets produced locally at Washington University: The Connectome Coordination Facility (CCF) (which itself includes the Human Connectome Project (HCP) Young Adult study, The Lifespan related projects, the Disease related projects, and assorted HCP-related projects), The Knight Alzheimer Disease Research Center (ADRC), the Adolescent Brain Cognitive Development (ABCD) Study, The Comprehensive Neuro-Oncology Data Repository (CONDR), and the clinically-based PACS image repository. In addition, copies of external data collections such as the UK Biobank, The Alzheimer's Disease Neuroimaging Initiative (ADNI), and The Cancer Image Archive (TCIA) will be maintained. The RIR includes a data management software solution that will introduce many novel features (such as `data tagging' to enrich datasets, and advanced search features) and will allow us to leverage existing storage including the Center for High Performance Computing's (CHPC) 1.4PB of BeeGFS `scratch' storage, solid-state NVMe drives integrated into the compute nodes, and 10PB of ZFS-based storage. All storage will be presented to the user as a single file-system, while data will be migrated to different performance tiers based on the storage requirements of the datasets or processing algorithms. The RIR will be integrated into the CHPC for data processing. The proposal also includes two NVIDIA DGX A100 GPU servers providing state-of-the-art GPU- based processing power. The combination of high-quality, diverse sets of biomedical imaging data with next- generation computing power will have a transformative effect on biomedical imaging processing pipelines and nowhere will the effects be more profound than in the emerging field of Deep Learning for image processing.