本文選題：云計(jì)算　切入點(diǎn)：體繪制　出處：《電子科技大學(xué)》2014年碩士論文　論文類型：學(xué)位論文

【摘要】：腦部疾病相對人體其他器官的疾病更難治療,因?yàn)槿说哪X部結(jié)構(gòu)更為復(fù)雜,治療手段也相對欠缺,如何能在有限的醫(yī)療條件下給醫(yī)生提供直觀的腦部三維圖像,更為方便并且精確的找出腦部病灶所在,是目前的研究熱門,具有非常重要的研究價(jià)值和臨床意義。本文首先應(yīng)用了目前熱門的云計(jì)算技術(shù)搭建云服務(wù)平臺,介紹了HDFS分布式文件系統(tǒng),并使用了MapReduce分布式的執(zhí)行框架實(shí)現(xiàn)了批量轉(zhuǎn)換腦部MR圖像格式,即DICOM到JPEG的格式轉(zhuǎn)換,實(shí)現(xiàn)過程中對單機(jī)和集群以及不同節(jié)點(diǎn)的集群之間的轉(zhuǎn)換效率進(jìn)行了對比,得出多個(gè)節(jié)點(diǎn)的集群環(huán)境下轉(zhuǎn)換效率相對較高的結(jié)論,此為后續(xù)三維重建及可視化提供了基礎(chǔ)。其次介紹了醫(yī)學(xué)圖像三維繪制的流程,并就其中的分割提取提出了算法實(shí)現(xiàn),對分割算法中的醫(yī)學(xué)圖像分水嶺算法提出了改進(jìn),利用其各向異性擴(kuò)散設(shè)計(jì)了新的分割算法,并進(jìn)行了實(shí)驗(yàn),新的算法在保持物體原有輪廓的顯示效果比傳統(tǒng)方法更令人滿意,過分割情況也并不嚴(yán)重。最后對現(xiàn)有的可視化技術(shù)進(jìn)行研究,尤其是對體繪制中的光線投射算法及其加速技術(shù)進(jìn)行研究,并提出了光線投射法的一種改進(jìn)算法,即基于接近云算法的光線投射法。這個(gè)算法對于空體素較多的圖像繪制的加速效果更佳,主要利用的原理是在投射光線穿越三維數(shù)據(jù)場時(shí)直接跳過空體素,不對空體素進(jìn)行顏色值與不透明度的累積,可以大大減少繪制所需時(shí)間。另外為了彌補(bǔ)了非空體素較多的圖像加速效果不明顯的局限性,在改進(jìn)算法中還應(yīng)用了快速三線性插值算法,對于非空體素采用小步長重采樣的方法,使得新的算法比傳統(tǒng)的基于硬件加速技術(shù)的光線投射法效率更高,繪制效果也令人滿意。綜上所述,本文主要對腦部MR圖像的三維重建及可視化技術(shù)進(jìn)行了研究,并對傳統(tǒng)的體繪制算法上進(jìn)行了改進(jìn),加速了繪制效率,對于用戶對腦部MR圖像實(shí)時(shí)交互的技術(shù)發(fā)展起到了推動的作用。
[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.
【學(xué)位授予單位】：電子科技大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：R445.2;TP391.41

【共引文獻(xiàn)】

相關(guān)博士學(xué)位論文 前1條

1 張霞;基于工業(yè)CT切片數(shù)據(jù)的三角網(wǎng)格模型簡化及優(yōu)化方法研究[D];重慶大學(xué);2013年

相關(guān)碩士學(xué)位論文 前1條

1 梁俊;基于多核學(xué)習(xí)支持向量機(jī)的貨幣識別[D];中南大學(xué);2014年

，

本文編號：1583247