自适应放疗中变形图像配准算法相关关键技术的研究

发布时间：2018-03-24 16:24

本文选题：自适应放疗　切入点：锥形束CT　出处：《南方医科大学》2017年博士论文

【摘要】：放射治疗是治疗肿瘤的三大有效手段之一,其治疗效果取决于两个方面:一是保证充分的靶区受照剂量,二是减少正常组织受照剂量。放疗技术也正是以此为目标向前发展,从三维适形放疗(three dimensional conformal radiotherapy,3D-CRT)到调强放疗(intensity-modulated radiotherapy,IMRT),再到图像引导放疗(image-guided radiotherapy,IGRT)以及自适应放疗(adaptive radiotherapy,ART)。具体而言,3D-CRT实现射野与靶区在形状上的一致,IMRT在此基础上实现剂量适形,而IGRT利用治疗中(in-room)影像校正了每治疗分次的摆位误差。但是,由于治疗效应和正常生理过程的影响,某些感兴趣区域(region of interest,ROI)的位置、形状和体积在治疗期间会发生变化,进而可能导致肿瘤欠照射或危及器官(organ at risk,OAR)过照射。ART通过重新设计或调整计划来适应当前的解剖结构情况,是解决上述问题的一种有效手段。作为ART中的关键技术之一,图像变形配准(deformable image registration,DIR)算法关联着计划图像和in-room图像,关联的准确程度决定了 ART的有效性和可靠性。因此验证DIR的精度是非常必要的。本文以此为出发点开展了一系列相关研究。本文首先回顾并实现四类(十种)代表性的基于图像灰度的DIR算法。这四类算法分别是光流场类(HS,HSLK和FFD)、demons类(OD,MD,SFD,DFD和DISC)、水平集类(LS)和样条类(BSpline),其中DISC是在图像处理单元(graphic processing unit,GPU)上实现,其他 DIR 算法均用 Matlab 实现。针对关于多模态图像验证DIR算法精度的研究相对缺乏的问题,本文回溯性地采集21例鼻咽癌患者的CT和CBCT图像,由经验丰富的医生勾画双侧腮腺、双侧颌下腺、颈椎椎体和椎孔四种ROI。然后利用所实现的DIR算法将CT图像上的ROI轮廓线推衍到CBCT图像上,并与医生勾画的真实轮廓线进行比较,以此评估其在ROI轮廓线推衍方面的精度和性能差异。结果表明,所有DIR算法的表现并不一定好于刚性配准。一般来说,DIR算法在刚性结构上的表现要比其在软组织上的表现出色,且DIR精度还随ROI的不同而变化。另外,算法的表现还与ROI的形变程度(时间推移)相关。针对目前体模验证研究中体模复杂度缺乏的问题,本文以真实患者的腹部CT影像为参考,设计并制备一种高仿真的物理体模。该体模包含肝、肾、脾、胃和脊椎等多种仿真结构,这些结构具有与真实解剖结构相似的体积、形状、CT值以及空间位置。另外,体模还包含195个标记点(直径为1～2 mm的金属球),均匀分布于各仿真器官的内部、表面以及器官间区域,标记点在变形前后的位置变化被作为精度验证的标准。体模验证结果表明,绝大多数DIR算法能显著改善刚性配准的精度,其表现与图像对比度及ROI性质(体积、形状等)等因素有关。尽管器官表面的DIR精度在总体上要显著高于器官内部的DIR精度,但对于同一实质器官而言,表面与内部之间一般不存在显著的精度差异。本文在DIR算法验证方面取得了一些初步成果,但在某些细节上仍有待进一步的研究,比如各DIR算法实现过程中参数优化问题、临床数据验证中勾画误差量化问题以及物理体模验证中基准标记点分布问题等等。此外,根据验证结果来改进现有DIR算法将会是我们未来的研究方向。例如,本研究表明算法精度与ROI的对比度有关,据此我们可人为提高原图像中某些ROI的对比度,将局部对比度增强后的图像的配准结果作为先验信息或约束条件加入到原图像的配准中,从而提高原图像的配准精度。
[Abstract]:Radiotherapy is one of the three effective methods to treat cancer, its treatment effect depends on two aspects: one is to ensure the target full dose, two is to reduce the dose of normal tissue. Radiotherapy technology is for this target forward, from three-dimensional conformal radiotherapy (three dimensional conformal radiotherapy, 3D-CRT in intensity-modulated radiotherapy (intensity-modulated) radiotherapy, IMRT), and then to image guided radiotherapy (image-guided radiotherapy, IGRT (adaptive radiotherapy) and adaptive radiotherapy, ART). Specifically, 3D-CRT fields and target zone is consistent in shape, the dose of conformal IMRT on the basis of this, and use in the treatment of IGRT (in-room) the image correction setup error in each treatment time. However, due to the influence of treatment effect and the normal physiological process, some regions of interest (region of, interest, ROI) the location, shape and body Product will change during treatment, which may lead to tumor under irradiation or organs (organ at risk, OAR) after irradiation.ART through redesign or adjust the plan to meet the current situation of the anatomical structure, is an effective means to solve the above problems. As one of the key technologies in ART, image registration (deformation deformable image registration, DIR) algorithm associated with the program image and in-room image, the accurate degree of correlation determines the validity and reliability of ART. It validates the accuracy of the DIR. It is very necessary to carry out a series of related research in this article as a starting point. This paper firstly reviews and realize the four (ten) representative DIR algorithm based on image gray value. These four algorithms are opticalflow class (HS, HSLK and FFD), Demons (OD, MD, SFD, DFD and DISC), the level set class (LS) and spline type (BSpline), which is DISC in image processing Unit (graphic processing unit, GPU) on the implementation of other DIR algorithms are implemented in Matlab. According to the relative lack of research on multimodal image authentication algorithm DIR problem, in this paper, a retrospective collection of CT and CBCT images of 21 cases of nasopharyngeal carcinoma patients, by experienced doctors draw the outline of bilateral parotid gland, submandibular glands, cervical vertebra four kinds of ROI. vertebral body and intervertebral foramen and then use the DIR algorithm to achieve the CT image on the ROI contours extrapolated to CBCT images, and the doctor and outline the true contour comparison, in order to evaluate the accuracy and performance of the difference in ROI derived contour aspect. The results show that all the performance of DIR algorithm is not necessarily better than the rigid registration. In general, performance of DIR algorithm in the rigid structure than its performance in the soft tissue on the job, and the accuracy of DIR with ROI also changed. In addition, the performance of the algorithm with ROI The degree of deformation (time). According to the present study phantom verification phantom complexity problem, based on the real patient abdominal CT images as reference, design and preparation of a high physical phantom simulation. The phantom consists of liver, kidney, spleen, stomach and other spinal structure simulation of these, and the real anatomical structure has similar structure shape, size, CT value and spatial position. In addition, the phantom also contains 195 markers (1~2 mm diameter metal ball), internal uniform distribution in each simulation organs, organs and the surface area was used as markers to verify the accuracy of the standard in the position changes before and after deformation. The validation results show that the vast majority of phantom, the DIR algorithm can significantly improve the accuracy of rigid registration, its performance and image contrast and ROI properties (size, shape) and other factors. Although the organ surface in general to show the accuracy of DIR Compared with the internal organs of DIR precision, but for the same substance between the surface and the internal organ, there is generally no significant difference in accuracy. This paper has made some preliminary results in the DIR algorithm, but some details still need to be further studied, such as the DIR algorithm to realize the optimization of parameters in the process of clinical data a validation standard outlined in the quantization error and physical phantom validation marker distribution problems and so on. In addition, our future research direction to improve the existing DIR algorithm will be based on the test results. For example, this study shows that the algorithm accuracy and contrast of ROI, so we can be provided some ROI image contrast in plateau that will be the local contrast enhanced image registration results as a priori information or constraints into the original image in the original registration, so as to improve the precision of image registration.

【学位授予单位】：南方医科大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41;R730.55

【相似文献】