SPECT RECONSTRUCTION ALGORITHMS AND PARALLEL COMPUTING

这个应用程序是调查的图像重建算法 以及用于核医学和其他领域的并行计算硬件 医学影像学。 还包括研究新的方法， 评估和预测图像质量。 的长期目标 算法研究是在定量准确性方面的改进， 重建图像和改善诊断性能时，人类 观察者使用图像。 的硬件组件的目标 研究是生产一个经济的计算机系统，非常适合图像 重建，性能接近超级计算机。 的 图像质量研究的目的是发展模型观测器， 对于成像系统，哪些临床上有意义的品质因数可以 推导 算法研究将研究贝叶斯重建， 数据模型的不同形式（精确泊松或最小二乘 近似）和关于对象的不同形式的先验信息 正在重建。 补偿分散的新方法 将研究SPECT中的辐射和衰减， 将开发感兴趣区域的定量。 将为现有的并行计算机（TRIMM）开发软件， 允许详细探索在该资助下开发的算法， 以及更传统的算法。 一台升级的并行计算机， 具有1-2 GFLOPS的能力，也将被设计和建造。 图像质量的品质因数基于人的表现或 模拟观察者完成现实的、临床相关的任务。 特别 重点将放在最佳的线性观测器，但非线性 模型也将进行研究。 模型观察员的使用将 通过广泛的心理物理学研究证实了这一点。 当有效模型 发现，他们将被用来解决许多长期存在的问题，在形象 重建，包括最佳停止点和正则化 迭代算法中的函数以及各种形式的先验 信息.

This application is for investigation of image-reconstruction algorithms and parallel-computing hardware for use in nuclear medicine and other areas of medical imaging. Also included is investigation of new ways of assessing and predicting image quality. The long-term goals of the algorithmic research are improvements in quantitative accuracy in the reconstructed images and improved diagnostic performance when human observers use the images. The objective of the hardware component of the research is to produce an economical computer system, well suited to image reconstruction, with a performance approaching that a supercomputers. The image-quality studies are aimed at development of model observers from which clinically meaningful figures of merit for imaging systems may be derived. The algorithmic studies will investigate Bayesian reconstruction, with different forms for the data model (exact Poisson or least-squares approximations) and different forms of prior information about the object being reconstructed. New approaches to compensation for scattered radiation and attenuation in SPECT will be studied, and new methods for region-of-interest quantitation will be developed. Software for the existing parallel computer (TRIMM) will be developed, allowing a detailed exploration of the algorithms developed on this grant, as well as more conventional algorithms. An upgraded parallel computer, with 1-2 GFLOPS capability, will also be designed and constructed. The figures of merit for image quality are based on performance of human or model observers on realistic, clinically relevant tasks. Particular emphasis will be placed on the optimum linear observer, but nonlinear models will also be investigated. The use of model observers will be validated by extensive psychophysical studies. When valid models are found, they will be used to address many long-standing problems in image reconstruction, including the optimum stopping point and regularizing function in iterative algorithm and the effects of various forms of prior information.