三维放射治疗计划系统仿真建模研究

发布时间：2018-08-08 20:01

【摘要】：放射治疗是肿瘤治疗的重要手段,三维放射治疗计划系统(Three dimensional radiation therapy planning system,3D-RTPS)是精确放射治疗核心子系统。3D-RTPS利用计算机程序模拟整个治疗过程,计算出患者体内剂量分布数据,通过分析与评估,制定出合理的治疗方案,从而有效的减少放射治疗的副作用,增加肿瘤控制率。本文基于3D-RTPS产品需求,对3D-RTPS涉及的核心技术进行了比较全面和详细的研究,并进行了相关功能的验证与分析,力求提供一个满足临床需求、具有良好扩展性的软件产品平台。本文研究工作包括三维可视化、组织分割、剂量计算、逆向调强、GPU加速及软件开发实现。在三维可视化领域,主要做了两方面创新工作。一方面基于Phong光照模型,提出预先计算体元法向量并基于球坐标索引进行存储。该方法在光线追踪过程直接获得当前体元的法向量,避免了法向量重复计算,有效减少了可视化时间。基于球坐标索引进行存储避免了存储法向量三个浮点分量,减少了内存开销。另一方面,将光线投射方法的体绘制技术应用于剂量三维分布显示。在光线投射采样过程中,系统使得医生根据临床需要对阻光度及颜色进行分类,使用该方法,医生能直观的判断器官的剂量分布情况。在组织分割方面,实现了体轮廓、肺及脊髓的自动提取功能。基于CT影像人体结构的特征知识,提出了三个主要步骤实现脊髓的自动提取功能。在检测脊髓概率区关键步骤中,基于脊髓及其周围结构的特征知识,建立了一个全新的特征模型用于脊髓内一点的检测,基于该点进行区域增长得到脊髓概率区后,在该区域内实现脊髓的检测。引入特征模型自适应修正,实现了60例患者CT图像序列脊髓自动提取100%的成功率。软件运行于笔记本电脑,患者CT图像序列脊髓检测时间可以达到3秒左右,满足临床要求。在剂量计算方面,基于点核卷积叠加剂量计算模型,本文将治疗床CT影像象素加入到患者CT影像数据中,使系统在剂量计算过程引入治疗床对X射线束的衰减作用,降低了治疗床引起的剂量偏差,提高了系统剂量计算精度。在模型匹配方面,基于模型参数自身特征,提出了基于模拟退火优化算法进行模型自动匹配,降低了软件对操作人员业务能力的依赖,降低了产品维护成本,增加了产品市场竞争力。在逆向调强方面,提出用点核叠加构建笔形束核进行剂量计算,该方法提高了优化迭代过程中的剂量计算速度,使得基于点核叠加技术的计划系统得以集成直接孔隙的逆向调强技术。模体及临床实际病例试验表明,该方法与使用精确剂量计算模型得到的优化结果一致,可用于调强优化过程中的剂量计算。在GPU加速技术方面,对原有剂量计算模型进行了修改,基于CUDA编程技术将NVIDIA的GPGPU模型应用于点核卷积/迭加模型的3D-RTPS产品。在程序架构设计中使用MFC导出类及动态库技术,避免了大量代码移植工作。对结果数据进行比较与分析,确定了基于特定显卡效率最高的thread数目。基于以上工作,作者开发了我国首套基于点核卷积叠加剂量计算模型的3D-RTPS商业化产品软件,并已在实际临床中应用。
[Abstract]:Radiation therapy is an important means of cancer treatment. Three dimensional radiation therapy planning system (3D-RTPS) is the core subsystem of accurate radiation therapy,.3D-RTPS uses computer program to simulate the whole process of treatment, and calculates the dose distribution data in the patient's body. Through analysis and evaluation, it is made out. Based on the 3D-RTPS product demand, this paper makes a comprehensive and detailed study of the core technologies involved in 3D-RTPS, and carries out the verification and analysis of the related functions, and strives to provide a soft and scalable soft. The research work includes three-dimensional visualization, organization segmentation, dose calculation, reverse intensity adjustment, GPU acceleration and software development. In the field of three-dimensional visualization, the main work is two aspects of innovation. On the one hand, based on the Phong illumination model, it is proposed to calculate the voxel vector and store it based on the spherical coordinate index. The normal vector of the current body element is obtained directly in the light tracing process, which avoids the repeated calculation of the normal vector and reduces the visualization time effectively. The storage vector based on spherical coordinates avoids the three floating-point components of the storage vector and reduces the memory overhead. On the other hand, the volume rendering technique of the ray projection method is applied to the dose three-dimensional distribution. In the light projection sampling, the system enables the doctor to classify the resistance and color according to the clinical needs. By using this method, the doctor can intuitively judge the dose distribution of the organs. In the aspect of tissue segmentation, the automatic extraction function of the body contour, the lung and the spinal cord is realized. Based on the characteristic knowledge of the body structure of the CT image, the system is proposed. Three main steps to realize the automatic extraction of spinal cord. In the key step of detecting the spinal cord probability area, based on the characteristic knowledge of the spinal cord and its surrounding structure, a new feature model is established for the detection of one point in the spinal cord, and the spinal cord detection is realized in the region after regional growth of spinal cord probability area based on this point. The successful rate of automatic extraction of 100% in the spinal cord of 60 patients with CT image sequence is realized by introducing the adaptive correction of the feature model. The software runs on the notebook computer and the time of detection of the spinal cord of the patient's CT image sequence can reach to about 3 seconds. In the dose calculation, the dose calculation model based on the point kernel convolution superposition is used in the treatment bed C. The T image pixel is added to the patient's CT image data, so that the system is introduced into the treatment bed for the attenuation of the X ray beam in the dose calculation process, reduces the dose deviation caused by the treatment bed, and improves the accuracy of the system dose calculation. In the model matching, based on the model parameter's own characteristics, the model based annealing optimization algorithm is proposed to model the model. Automatic matching reduces the dependence of the software on operator's business ability, reduces the cost of product maintenance and increases the competitiveness of the product market. In the reverse intensity aspect, the point kernel superposition is proposed to build the pen shaped beam core for dose calculation. This method improves the dose calculation speed in the optimization iteration process and makes the technology based on the point kernel Superposition Technology. The plan system is able to integrate the reverse intensity modulation technique of the direct pore. The model body and the clinical case test show that the method is consistent with the optimization results obtained by using the accurate dose calculation model, and can be used for the dose calculation in the process of the optimization process. In the GPU acceleration technology, the original dose calculation model is modified and based on the CUDA programming. The technology of NVIDIA's GPGPU model is applied to the 3D-RTPS product of the point kernel convolution / superposition model. The MFC export class and the dynamic library technology are used in the program architecture design to avoid a large number of code transplanting. The results are compared and analyzed, and the number of thread based on the highest efficiency of the specific graphics card is determined. Based on the above work, the author develops The first commercial product software of 3D-RTPS based on point-core convolution superposition dose calculation model in China has been developed and applied in clinical practice.
【学位授予单位】：东北大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：R730.55;TP391.41

【参考文献】