High-Performance Computer Cluster for Image Analysis

描述（由申请人提供）：我们建议购买一个分布式内存计算集群，以加强我们在图像分析，图像重建和技术评估方面的研究，我们研究的长期目标是提高我们在计算机辅助诊断（CAD），图像重建，乳腺癌和前列腺癌的MRI/MRS以及ROC分析方面的知识。具体而言，拟议的系统将在四个领域取得进展。首先，我们计划对需要在线计算的CAD进行临床评价。由于计算的复杂性，这些需要在高性能计算机上进行，以实现CAD系统的1 s响应。其次，我们将开发用于分析MRI和MRS数据集的图像分析技术，以改善乳腺癌和前列腺癌的早期检测和分期。MRI和MRS的数据集很大-大小超过千兆字节。为了最大限度地从这些数据中提取信息，预处理将开发先进的图像重建技术，结合计算机模拟来模拟图像采集系统，并使用模拟优化重建方法。模拟和重建技术都是计算密集型的。第四，我们将开发公开可用的软件，用于ROC分析，并结合该领域的最新进展。这些包括基于由于病例样本变异和图像读取器变异而引起的变异的ROC曲线之间的差异的统计分析。研究的相关性，这将采用计算集群，是四倍。首先，我们将在前瞻性临床前研究中证明计算机可以帮助放射科医生确定乳腺病变或肺结节是良性还是恶性。这可以减少不必要的活检次数，减少癌症被遗漏的机会。其次，通过提取更多的MRI检查信息，可以提高早期发现、诊断的准确性。我们的研究使用的影响提出的系统是大的。例如，旧版本的ROC软件被全球10，000多名研究人员使用。更新后的软件无疑将使许多研究人员受益。 ROC分析可以更快、更便宜地实现最有前途的新医学进展，因为可以设计更有效、更经济的评估研究，

DESCRIPTION (provided by applicant): We are proposing to purchase a distributed memory computation cluster to enhance our research in image analysis, image reconstruction, and technology assessment, The long-term objectives of our research is to advance our knowledge in computer-aided diagnosis (CAD), image reconstruction, MRI/MRS for the breast and prostate cancers, and ROC analysis. Specifically, the proposed system will allow for advances in four areas. First, we plan to conduct clinical evaluation of CAD where online calculations are needed. Because of the complexity of the calculations, these need to conducted on high performance computers to achieve 1 s response of the CAD system. Second, we will develop image analysis techniques for analyzing MRI and MRS datasets for improving the early detection and staging of breast and prostate cancers. The datasets for MRI and MRS are large - over a gigabyte in size. To maximize the extraction of information from these data, preprocesswill develop advanced image reconstruction techniques using a combination of computer simulation to model image acquisition systems and optimizing reconstruction methods using the simulations. Both the simulation and reconstruction techniques are compute intensive. Fourth, we will develop publicly available software for ROC analysis that incorporates recent advances in the field. These include statistical analyses of differences between ROC curves based on variations that due to case-sample variation and image-reader variation. The relevance of the research, which will employ the computation cluster, is four fold. First, we will demonstrate in prospective pre-clinical studies that computers can help radiologists determine whether a breast lesion or a lung nodule is benign or malignant. This can reduce the number of unnecessary biopsies and reduce the chances that a cancer is missed. Second, by extracting more information from MRI studies, we can improve the early detection, diagnosisctive. The impact of our research using the poposed system is large. For example, the old version of ROC software is used by more than 10,000 researchers worldwide. The updated software will undoubtedly benefit many researchers. ology (ROC analysis) can lead to faster and cheaper implementation of the most promising of new medical advances, because evaluation studies can be designed that are more efficient and cost effe