基于云计算的脑部MR图像可视化的研究与实现

发布时间：2018-03-08 09:05

本文选题：云计算　切入点：体绘制　出处：《电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：脑部疾病相对人体其他器官的疾病更难治疗,因为人的脑部结构更为复杂,治疗手段也相对欠缺,如何能在有限的医疗条件下给医生提供直观的脑部三维图像,更为方便并且精确的找出脑部病灶所在,是目前的研究热门,具有非常重要的研究价值和临床意义。本文首先应用了目前热门的云计算技术搭建云服务平台,介绍了HDFS分布式文件系统,并使用了MapReduce分布式的执行框架实现了批量转换脑部MR图像格式,即DICOM到JPEG的格式转换,实现过程中对单机和集群以及不同节点的集群之间的转换效率进行了对比,得出多个节点的集群环境下转换效率相对较高的结论,此为后续三维重建及可视化提供了基础。其次介绍了医学图像三维绘制的流程,并就其中的分割提取提出了算法实现,对分割算法中的医学图像分水岭算法提出了改进,利用其各向异性扩散设计了新的分割算法,并进行了实验,新的算法在保持物体原有轮廓的显示效果比传统方法更令人满意,过分割情况也并不严重。最后对现有的可视化技术进行研究,尤其是对体绘制中的光线投射算法及其加速技术进行研究,并提出了光线投射法的一种改进算法,即基于接近云算法的光线投射法。这个算法对于空体素较多的图像绘制的加速效果更佳,主要利用的原理是在投射光线穿越三维数据场时直接跳过空体素,不对空体素进行颜色值与不透明度的累积,可以大大减少绘制所需时间。另外为了弥补了非空体素较多的图像加速效果不明显的局限性,在改进算法中还应用了快速三线性插值算法,对于非空体素采用小步长重采样的方法,使得新的算法比传统的基于硬件加速技术的光线投射法效率更高,绘制效果也令人满意。综上所述,本文主要对脑部MR图像的三维重建及可视化技术进行了研究,并对传统的体绘制算法上进行了改进,加速了绘制效率,对于用户对脑部MR图像实时交互的技术发展起到了推动的作用。
[Abstract]:Brain diseases are more difficult to treat than diseases of other organs of the human body, because people's brain structures are more complex and treatment methods are relatively lacking. How can doctors be provided with intuitive three-dimensional images of the brain under limited medical conditions? More convenient and accurate location of brain lesions, is the current hot research, has very important research value and clinical significance. Firstly, this paper uses the current popular cloud computing technology to build cloud service platform. This paper introduces the HDFS distributed file system, and uses the MapReduce distributed execution framework to realize the batch conversion of brain Mr image format, that is, the format conversion from DICOM to JPEG. In the process of implementation, the conversion efficiency between single machine and cluster and among different nodes is compared, and the conclusion that the conversion efficiency is relatively high in the cluster environment of multiple nodes is obtained. This provides the foundation for the subsequent 3D reconstruction and visualization. Secondly, the flow of 3D rendering of medical image is introduced, and the algorithm of segmentation and extraction is proposed, and the improvement of watershed algorithm of medical image is put forward. A new segmentation algorithm based on anisotropic diffusion is designed and experimented. The new algorithm is more satisfactory than the traditional method in maintaining the original contour of the object. Finally, the existing visualization techniques, especially the ray-casting algorithm and its acceleration in volume rendering, are studied, and an improved ray-casting algorithm is proposed. That is, ray-casting method based on nearing cloud algorithm, which can accelerate the rendering of images with more voxels. The main principle is to skip voxels directly while projecting light through 3D data fields. By not accumulating the color value and opacity of voxels, the time required for rendering can be greatly reduced. In addition, in order to make up for the limitation of the non-empty voxel image acceleration effect is not obvious, In the improved algorithm, the fast trilinear interpolation algorithm is also applied. The new algorithm is more efficient than the traditional ray-casting method based on hardware acceleration, and the method of small step size resampling is used for the non-empty voxel. The rendering effect is also satisfactory. In conclusion, this paper mainly studies the 3D reconstruction and visualization technology of brain Mr image, and improves the traditional volume rendering algorithm, which accelerates the rendering efficiency. It plays an important role in the development of real-time interaction of brain Mr images.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R445.2;TP391.41

【共引文献】