GPU-based Computational Advancements for Neuroscience MATLAB Programs

描述(由申请人提供)：该项目的目的是推进开发的夹克：图形处理器引擎的MatLab，以包括旨在增强计算神经科学的功能。我们将开发工具，使MatLab(R)程序员能够享受图形处理器(GPU)的性能和速度优势。今天，据估计，在医疗保健行业有150万MATLAB用户，其中很大一部分人使用MatLab来解决与神经科学相关的问题。MatLab的用户，特别是那些处理大型神经科学数据集的用户，如脑MRI、fMRI、DW-MRI、PET和CT体积以及显微图像，目前在使用MatLab进行神经科学研究时存在两个主要问题：1)与其他编程语言(如C/C++)相比，MatLab速度慢；2)MatLab可视化无法处理大量数据或轻松地渲染解剖结构的3D模型。因此，神经学家经常付出昂贵和耗时的努力，将神经科学的MatLab代码移植到C/C++，代价是减缓研究工作和合作，并最终分散研究人员解决生物学问题的主要重点。然而，由于最近计算机处理器的进步，特别是由于NVIDIA的Tesla、AMD的Firestream和英特尔即将推出的Larrabee GPU，新一波桌面处理技术使个人研究人员有可能直接在MATLAB中获得更快的速度和增强的可视化效果。在过去的两年里，我们已经开发并发布了我们的第一个产品，Jacket：用于MatLab的GPU引擎，它使科学家能够在GPU上执行低级别的MatLab计算。在第一阶段，我们建议通过GPU来扩展Jacket--支持神经学家使用的最常见的MATLAB函数，例如那些在MATLAB的信号处理、图像处理和统计工具箱中找到的函数。在第二阶段，我们计划启用GPU-针对神经科学的更高级别的MatLab任务，例如开源SPM(统计参数映射)工具包和MatLab的生物信息学工具箱中提供的任务。此外，在第二阶段，我们计划通过启用Handle Graphics API的GPU和使用最新的光线跟踪技术(如NVIDIA的NVIRT中出现的那些技术)来提供最先进的体积渲染功能，从而极大地增强MATLAB的可视化。为了实现这些目标，需要进一步研究和开发这些工具，并对其进行优化，以实现最佳性能和最高标准的稳定性和用户友好性。 与公共健康相关：该项目的目的是推进Jacket的开发：用于MatLab的GPU引擎包括旨在增强计算神经科学的功能。目前，使用MatLab进行神经科学研究的主要问题有两个：1)计算速度；2)缺乏高性能的可视化技术。由于最近计算机处理器的进步，特别是由于NVIDIA的Tesla、AMD的Firestream和英特尔即将推出的Larrabee图形处理器，新一波桌面处理技术使个人研究人员有可能直接在MATLAB中获得更快的速度和增强的可视化效果。在这项工作中，我们将通过GPU扩展Jacket--支持神经学家使用的最常见的MATLAB函数，例如那些在MATLAB的信号处理、图像处理和统计工具箱中找到的函数。这些努力将使神经科学家能够专注于科学，而不是计算实现，从而加速全球神经科学的努力。

DESCRIPTION (provided by applicant): The purpose of this project is to advance the development of Jacket: The GPU Engine for MATLAB to include functionality aimed at enhancing computational neuroscience. We will develop tools which will allow MATLAB(R) programmers to access the performance and speed benefits of graphics processing units (GPUs). Today, there are an estimated 1.5 million MATLAB users in the healthcare industry, with a substantial portion of those using MATLAB to solve neuroscience-related problems. MATLAB users, especially those dealing with large neuroscience datasets, such as brain MRI, fMRI, DW-MRI, PET, and CT volumes as well as microscopy imagery, currently have two major problems in using MATLAB to conduct neuroscience research: 1) MATLAB is slow when compared to other programming languages such as C/C++, and 2) MATLAB visualizations are unable to handle large amounts of data or to render 3D models of anatomical structures with ease. Therefore, neuroscientists often undertake costly and time-consuming efforts to port neuroscience MATLAB code to C/C++, at the expense of slowing down research efforts, collaborations, and ultimately detracting from the researcher's primary focus of solving biological problems. However, due to recent advances in computer processors, specifically due to NVIDIA's Tesla, AMD's Firestream, and Intel's upcoming Larrabee GPUs, a new wave of desk-side processing technology makes it possible for individual researchers to get increased speed and enhanced visualizations directly in MATLAB. Over the last two years, we have developed and released our first product, Jacket: The GPU Engine for MATLAB, which enables scientists to perform low-level MATLAB computations on the GPU. In Phase I, we propose to extend Jacket by GPU-enabling the most common MATLAB functions used by neuroscientists, such as those found in MATLAB's Signal Processing, Image Processing, and Statistics Toolboxes. In Phase II, we plan to GPU-enable higher-level neuroscience- targeted MATLAB tasks, such as those available in the open source SPM (Statistical Parametric Mapping) toolkit and MATLAB's Bioinformatics Toolbox. Also, in Phase II, we plan to greatly enhance MATLAB's visualizations by GPU-enabling the Handle Graphics API and by using recent ray tracing technologies, such as those emerging in NVIDIA's NVIRT, to provide state-of-the-art volume rendering functions. In order to achieve these goals, further research and development is needed to build these tools and optimize them to achieve the best performance and highest standards of stability and user-friendliness. PUBLIC HEALTH RELEVANCE: The purpose of this project is to advance the development of Jacket: The GPU Engine for MATLAB to include functionality aimed at enhancing computational neuroscience. MATLAB users, especially those dealing with large neuroscience datasets, such as brain MRI, fMRI, DW-MRI, PET, and CT volumes as well as microscopy imagery, currently have two major problems in using MATLAB to conduct neuroscience research: 1) computational speed, and 2) lack of high-performance state-of-the-art visualizations. Due to recent advances in computer processors, specifically due to NVIDIA's Tesla, AMD's Firestream, and Intel's upcoming Larrabee GPUs, a new wave of desk-side processing technology makes it possible for individual researchers to get increased speed and enhanced visualizations directly in MATLAB. In this work, we will extend Jacket by GPU- enabling the most common MATLAB functions used by neuroscientists, such as those found in MATLAB's Signal Processing, Image Processing, and Statistics Toolboxes. These efforts will accelerate neuroscience efforts worldwide by empowering neuroscientists to focus on science, rather than computational implementations.