Global Shared Memory (GSM) Supercomputer for Quantitative PET Imaging

描述（由申请人提供）：马萨诸塞州总医院的ET和SPECT成像得到了NIH资助研究的积极支持，并解决了神经病学、肿瘤学、心脏病学、神经科学、肿瘤生物学和分子成像领域的重要问题。在核医学成像中有几个重要的进展，例如用于新型PET放射性药物的动力学建模的动态成像、PET-MR、飞行时间PET和定量SPECT-CT，这些进展已经引发了对这些新应用的建模、模拟和图像重建的积极研究。使用来自能量、飞行时间和动态时间帧的额外维度的图像重建需要前所未有的计算机存储器和速度来建模极大的系统矩阵。具有非常大的共享内存的超级计算机对于这样的应用是至关重要的，因为在由网络连接的多个节点组成的标准计算机集群中使用分布式内存受到节点到节点网络连接缓慢的严重限制。我们建议成立一家硅图形公司。Altix UV 1000全球共享内存超级计算机，具有2 TB内存和256个英特尔至强内核。该系统将为大量受资助的NIH研究人员提供服务，并将通过提高定量PET和生物医学成像的图像质量和准确性来大幅提高其研究质量。

DESCRIPTION (provided by applicant): ET and SPECT imaging at Massachusetts General Hospital is actively supported by NIH- funded research and addresses questions of fundamental importance in the fields of neurology, oncology, cardiology, neuroscience, tumor biology and molecular imaging. There are several significant advances in nuclear medicine imaging, such as dynamic imaging for kinetic modeling of novel PET radiopharmaceuticals, PET-MR, time-of-flight PET, and quantitative SPECT-CT, that have triggered active research in modeling, simulation and image reconstruction for these new applications. Image reconstruction using additional dimensions from energy, time-of-flight, and dynamic time frame requires unprecedented computer memory and speed to model extremely large system matrices. A supercomputer with a very large shared memory is critical for such applications because using distributed memory across a standard computer cluster, which consists of multiple nodes connected by network, is severely limited by the slow node-to-node network connections. We propose a Silicon Graphics Inc. Altix UV 1000 global shared memory supercomputer with 2 TB memory and 256 Intel Xeon cores. This system will serve a large group of funded NIH investigators and will yield substantial improvement in the quality of their research by enhancing image quality and accuracy in quantitative PET and biomedical imaging.