Development of a Statistical Medical Image and Data Processing System

本研究项目的目的是在我们以前对统计图像处理的研究成果的基础上，开发一个统计医学图像和数据处理系统。为了实现这一目标，我们按照以下三个步骤进行了研究。首先，从计算开销的角度改进了PET和SPECT重建算法以及我们之前开发的MRI图像分割算法。通常，统计图像和数据处理算法需要大量的计算代价，因为它们是基于预定义的代价函数的最小化。然而，改进的算法具有与常用算法相当的合理计算成本。为了达到这一目的，我们引入了块迭代的概念和数学规划领域中的一些优化方法。其次，针对医学成像中出现的新问题，提出了基于统计估计的图像处理算法。例如，这些方法包括基于少量血管造影X射线投影数据的三维血管重建，用于PET和SPECT的分段衰减图重建，以及来自医学图像序列的光流计算。为了达到这一目的，我们使用了数学规划领域中的优化方法、拓扑约束图像标记和小波变换等新概念。最后，所有开发的软件统一在一个工作站上，使它们具有相同的数据格式和相同的用途，并将图像显示等用户界面结合在一起。我们利用现有的几个数据对该图像和数据处理系统的性能进行了评估。

The aim of this research project is to develop a statistical medical image and data processing system based on our previous research results on statistical image processing. To achieve this aim, we performed the research according to the following three steps. First, we improved the PET and SPECT reconstruction algorithm and the MRI image segmentation algorithm which we developed previously in terms of computational costs. In general, statistical image and data processing algorithms require a lot of computational costs because they are based on the minimization of pre-defined cost functions. However, the improved algorithms worked with reasonable computational costs comparable to those of the commonly used algorithms. To achieve this aim, we introduced the concept of block iterations and some optimization methods in the field of mathematical programming. Second, we developed image processing algorithms based on statistical estimation for new problems appearing in medical imaging. For example, these include 3-D blood-vessel reconstruction from a few angiographc x-ray projection data, segmented attenuation map reconstruction for PET and SPECT, and optical flow computation from medical image sequences. To achieve this aim we used new concepts such as optimization methods in the field of mathematical programming, topology-constrained image labeling, and wavelet transform. Finally, all the developed softwares are unified in a workstation such that they work with the same data format and the same usage, and some user-interfaces such as image display are combined too. We evaluated the performances of this image and data processing system with several existing data.