Protected Radiomics Analysis Commons for Deep Learning in Biomedical Discovery

摘要 用于生物和医学成像的新兴成像系统的开发通常涉及 多维采集协议，因此相应地需要复杂的重建 算法和高级机器学习算法，以生成所需的结果图像 和定量肿瘤特征。该提案要求为高绩效 受保护的放射组学分析共享用于生物医学中的深度学习 发现，以进一步开发和应用深度学习技术，以定量 图像分析和图像重建。有12个具体的NIH项目将受益于 计算基础设施系统。我们通过内部的例子介绍这12个项目 四个具体的研究主题领域：（a）断层图像重建，（B）定量图像 分析，（c）功能定量，和（d）放射组学（图像数据）与表型 和基因组数据。所提出的系统是一个计算集群，它使用ScaleMP的多功能 SMP软件将群集节点聚合到单个对称多处理计算机中。的 主要硬件组件包括1台HP Enterprise ProLiant DL 380服务器和8台Apollo 6500 计算节点，共2.1 TB主内存，18个Intel Xeon E5- 2640 v4 10核CPU， 32个nVidia Tesla P100 GPU。服务器将通过100 Gbps EDR Infiniband网络连接。在 此外，三个重要的软件组件，旨在降低计算的复杂性， 环境和提高研究人员的生产力，将被集成到硬件组件： 使用上述ScaleMP vSMP从群集节点Cendio创建单个虚拟计算机 ThinLinc提供远程桌面图形登录服务，Bitfusion Flex AI Platform 提供GPU虚拟化、调度和优化，以及 深度学习框架。深度学习的受保护放射组学分析共享资源 在生物医学发现将允许研究人员加快，扩展和翻译他们目前的NIH 多维图像重建和图像分析算法领域的资金-所有 这将大大受益于深度学习方法-在一个安全的，受保护的，和HIPAA- 兼容的共享环境。

Abstract Development of emerging imaging systems for biological and medical imaging often involves multi-dimensional acquisition protocols and thus correspondingly the need for complex reconstruction algorithms and advanced machine learning algorithms in order to produce the desired resulting image and quantitative tumor characteristics. This proposal requests funding for a high performance computing system, entitled Protected Radiomics Analysis Commons for Deep Learning in Biomedical Discovery, to further the development and application of deep learning techniques to quantitative image analysis and image reconstruction. There are 12 specific NIH projects that will benefit from the proposed computing infrastructure system. We present the 12 projects through examples from within four Specific Research Topics areas: (a) tomographic image reconstruction, (b) quantitative image analysis, (c) functional quantification, and (d) association of radiomics (image data) with phenotypic and genomic data. The proposed system is a computing cluster, which uses ScaleMP's Versatile SMP software to aggregate the cluster nodes into a single symmetric multiprocessing computer. The major hardware components consist of 1 HP Enterpris\e ProLiant DL380 server and 8 Apollo 6500 compute nodes, with a total of 2.1 TB of main memory, 18 Intel Xeon E5-2640v4 10-core CPUs, and 32 nVidia Tesla P100 GPUs. The servers will be connected via a 100Gbps EDR Infiniband network. In addition, three important software components, which aim to reduce the complexity of the computing environment and increase researcher productivity, will be integrated into the hardware components: the aforementioned ScaleMP vSMP to create a single virtual computer from the cluster nodes, Cendio ThinLinc to provide remote desktop graphical login services, and Bitfusion Flex AI Platform which provides GPU virtualization, scheduling, and optimization, as well as curated container deployment of common deep learning frameworks. The Protected Radiomics Analysis Commons for Deep Learning in Biomedical Discovery will allow researchers to expedite, extend, and translate their current NIH funding in areas of many-dimensional image reconstruction and image analysis algorithms – all of which will benefit greatly from deep learning approaches – within a secure, protected, and HIPAA- compliant sharable environment.