医学影像处理与分析软件平台设计与实现

发布时间：2018-05-03 17:21

本文选题：医学影像处理与分析 + 分割　；参考：《电子科技大学》2016年硕士论文

【摘要】：随着医疗成像设备突飞猛进的发展,针对不同成像设备的图像处理算法也层出不穷。为了充分利用已有算法,避免重复开发,一些国内外研究机构研发了医学影像处理与分析软件平台。现有的医学影像处理与分析软件平台在特定领域已经取得了巨大的成功,但是还存在如下不足:一些功能强大的医学影像处理与分析综合平台需要配套医疗设备或特殊硬件才能运行,价格昂贵,不利于学习和在医学领域的普遍使用;开源框架的软件平台二次开发学习成本过高,这主要是由于其依赖的开源软件库过于庞大和复杂造成;由于医学影像数据本身的复杂性,使得现有软件平台的专业性太强,通用性不足,使用不便。本文的设计目标是在Windows操作系统下集成常用的医学影像处理与分析算法,以减轻算法研究者在算法工具使用上的学习成本,使得他们能够专注于新算法的设计和对比分析。论文的主要内容如下:1、医学影像处理与分析算法平台的设计。该算法平台划分为三层:第一层,底层,基于VTK、ITK、FSL等成熟开源算法库构建;第二层,中间层,将底层算法库中关于医学影像格式转换、滤波、配准、分割和三维重建的算法按照统一的图像读写接口封装,以便于应用层的调用;第三层,应用层,把中间层提供的算法按照一定的规则组合、连接,设计了图像格式转换、配准与分割、显示三个交互模块。2、医学影像处理与分析软件的实现。依据医学影像处理与分析算法平台的设计,对底层ITK和VTK算法库中相关算法进行了封装,将FSL算法库中关于脑部影像配准与分割的功能模块移植到了Windows操作系统下运行,把MATLAB环境下运行的脑部海马体三维分割算法集成到了算法平台中,所有算法均按照中间层图像读写接口的设计规则进行封装。软件中实现了应用层的三个交互模块:第一个,图像格式转换模块,实现了DICOM切片序列转换为三维NIf TI文件、三维NIfTI文件重切片为DICOM序列等格式转换功能;第二个,配准与分割模块,实现了FLIRT(脑部影像线性配准)、FIRST(脑部影像兴趣区域分割)、BET(脑组织提取)、HST(脑部海马体分割)四个算法;第三个,显示模块,实现了医学影像二维显示和三维重建,二维显示支持兴趣区域勾画,三维重建支持面绘制、网格绘制和体绘制,体绘制中支持标签图像的叠加显示。3、软件功能测试。使用CT、MRI影像测试了格式转换模块、配准与分割模块、显示模块,验证了这三个模块的正确性。特别地,使用ADNI数据库中的脑部MRI影像对FIRST算法在Windows和Linux操作系统下分别进行了18组实验,根据Precision相似测度对FIRST算法分割结果进行比较,两个操作系统下的分割结果没有明显差别,证明了移植后的FIRST算法能够正确运行。
[Abstract]:With the rapid development of medical imaging equipment, image processing algorithms for different imaging equipment are emerging in endlessly. In order to make full use of existing algorithms and avoid repeated development, some domestic and foreign research institutions have developed medical image processing and analysis software platform. The existing software platform for medical image processing and analysis has achieved great success in specific fields. But there are the following shortcomings: some powerful medical image processing and analysis platform needs medical equipment or special hardware to run, the price is expensive, which is not conducive to learning and widely used in the field of medicine; The cost of secondary development of open source framework software platform is too high, which is mainly due to the huge and complex open source software library it relies on, and the complexity of medical image data makes the existing software platform too professional. The generality is not enough, the use is inconvenient. The design goal of this paper is to integrate common medical image processing and analysis algorithms under the Windows operating system, so as to reduce the learning cost of algorithm researchers in the use of algorithm tools, so that they can focus on the design and comparative analysis of new algorithms. The main contents of this paper are as follows: 1, the design of medical image processing and analysis algorithm platform. The algorithm platform is divided into three layers: the first layer, the bottom layer, based on VTKKITK / FSL and other mature open source algorithm library, the second layer, the middle layer, the bottom algorithm library about the medical image format conversion, filtering, registration, The algorithms of segmentation and 3D reconstruction are encapsulated according to the unified image reading and writing interface, so as to facilitate the call of the application layer, the third layer, the application layer, combines the algorithms provided by the middle layer according to certain rules, and designs the image format conversion. Registration and segmentation, display three interactive modules. 2, medical image processing and analysis software implementation. According to the design of medical image processing and analysis algorithm platform, this paper encapsulates the related algorithms in the underlying ITK and VTK algorithm library, and transplants the function module of brain image registration and segmentation in the FSL algorithm library to the Windows operating system to run. The 3D segmentation algorithm of brain hippocampus running in MATLAB environment is integrated into the algorithm platform, and all the algorithms are encapsulated according to the design rules of the middle layer image read-write interface. Three interactive modules of the application layer are implemented in the software: the first is the image format conversion module, which realizes the conversion function of DICOM slice sequence to 3D NIf TI file, 3D NIfTI file reslice to DICOM sequence and so on. In the module of registration and segmentation, four algorithms of FIRST (brain image linear registration) are implemented, and the third, display module, realizes two-dimensional display and 3D reconstruction of medical image. 2D display supports drawing of region of interest, 3D reconstruction support surface rendering, mesh rendering and volume rendering, and volume rendering supports superposition display of label image. The format conversion module, registration and segmentation module and display module are tested with CTT MRI image, and the correctness of the three modules is verified. In particular, using brain MRI images in ADNI database, 18 groups of experiments were carried out on FIRST algorithm under Windows and Linux operating system respectively, and the results of FIRST segmentation were compared according to Precision similarity measure. There is no obvious difference in the segmentation results between the two operating systems, which proves that the transplanted FIRST algorithm can run correctly.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】