Radiation Field Modeling and Computerized Treatment Planning

我们使用的是一家商业视频处理器板(GPU)供应商，其价格比上一代定制体积渲染板低得多，而且在性能方面具有更大的增长潜力。该商业供应商还提供了利用电路板进行蒙特卡罗计算的软件(CUDA)，这进一步支持了我们的研究目标。我们已经从三维图像融合过渡到探索四维放射成像。我们的研究已经建立了三维躯干外部体积的时间运动与内部器官的时间运动之间的基本关系，尤其是肺。我们的研究开发了一种利用躯干外部体积变化进行图像跟踪的体积方法，该方法已申请专利。基于电子伽马簇射(EGS-4)的蒙特卡罗剂量计算引擎(DCE)已经在Linux和Windows环境下完全实现。在Windows环境中，DCE已集成到功能齐全的治疗计划系统中。现在的工作集中在相空间源模型的开发上。目前，该系统主要用于小视野立体定向放射外科的研究。由于减小了场大小，边缘效应变得重要，用于访问辐射输出的探测器的大小影响测量结果。对这些输出测量的蒙特卡罗模拟极大地有助于选择最适合这些测量的探测器系统。蒙特卡罗算法与实验测量结果吻合得很好，小到4 mm。在一个相关的项目中，我们采用了代数DCES和蒙特卡罗DCES来预测从诊断CT扫描中接收到的器官剂量。来自GE CT扫描仪的X射线束在国家儿童医学中心用于临床扫描的特征已经完成。我们还完成了绝对剂量测量，将计算机断层扫描剂量指数(CTDI)的标准诊断测量与实际吸收剂量联系起来。这包括对传统的CT剂量学方法与新的CTDS进行直接比较，以突出个性化剂量学的优势并演示特定器官的剂量描述。我们的下一个目标是使用小的点剂量计进行螺旋扇束CT扫描仪的调试，而不是使用诊断放射学中使用的传统CTDI测量。希望这一系统将为流行病学剂量-反应研究的患者群体提供更完整的剂量评估框架。Rob已经开始与RRP、DCTD、NCI合作建设一台功能强大的计算机，该计算机将利用现场可编程网格阵列(FPGA)板来补充GPU卡，以将基础计算机的计算能力提高几个数量级。目前，该系统在5个扩展机箱中有35块FPGA卡和4个GPU。当完全运行时，这将大大减少完成复杂蒙特卡罗计算的时间，从几周减少到几分钟。Rob在使用最初选择用作FPGA阵列主机的Hewlett-Packard(HP)ProLiant Gen-10服务器时遇到了技术困难。当前的系统BIOS与多个PCIe扩展机箱的使用不兼容，多个PCIe扩展机箱是容纳所有FPGA板所必需的。重写或升级基本输入输出系统大约需要两年时间。作为替代方案，Rob收购了OSS(One Stop Systems)2U EOS服务器，该服务器保证支持多个PCIe扩展机箱，因为它们都来自同一供应商。这一新系统已经安装并正在进行测试，目前可配置32块FPGA卡和2块GPU卡。原来的HP服务器配置了单个PCIe扩展机箱(Trenton Systems)，该机箱承载4个FGA卡和2个GPU卡。服务器BIOS将支持单个PCIe机箱。在将最终产品输出到生产服务器之前，将利用该系统进行软件开发和测试。

We are using a commercial vendor of video processor boards (GPUs) which are both substantially less expensive than the previous generation of custom volume rendering boards as well as having a greater potential for growth in performance. The commercial vendor also provides software (CUDA) for utilizing the boards for Monte Carlo calculations, which further supports our research objectives. We have transitioned from 3-dimensional image fusion to exploring 4-dimensional radiotherapy imaging. Our research has established a fundamental relationship between the temporal motion of the 3-dimensional external torso volume and those of internal organs, especially the lungs. Our research has developed a volumetric methodology for image tracking using external torso volume change for which a patent has been applied for. The Electron-Gamma Shower (EGS-4)-based Monte Carlo Dose Calculation Engine (DCE) has been fully implemented in both a LINUX and Windows environment. In the Windows environment, the DCE has been integrated into a full featured treatment planning system. Work is now centered on the development of phase-space source models. Currently, this system is used to investigate small field stereotactic radiosurgery. Due to the reduced field size, edge effects become important and the size of detectors used to access the radiation output affect the measurement results. Monte Carlo simulation of these output measurements greatly assisted in the selection of the detector system that is most suitable for these measurements. The Monte Carlo algorithm provided good agreement with experimental measurements down to applicators as small as 4mm. In a related project, we have adapted both algebraic and Monte Carlo DCEs to predict organ doses received from diagnostic CT scans. The characterization of the x-ray beam from a GE CT Scanner used for clinical scanning at Children's National Medical Center is complete. We have also completed absolute dosimetry measurements linking the standard diagnostic measurement of Computer Tomography Dose Index (CTDI) to actual absorbed dose. This includes a direct comparison of traditional CT dosimetry methodology versus the new CTDS to highlight the advantages of individualized dosimetry and demonstrate organ-specific dose descriptions. Our next goal is to undertake the commissioning of a helical fan-beam CT scanner using a small point-dosimeter, as opposed to the traditional measurement of CTDI employed in Diagnostic Radiology. It is hoped that this system will provide the framework for a more complete dose assessment of patient populations for epidemiological dose-response studies. ROB has embarked on a joint project with RRP, DCTD, NCI on the construction of a powerful computer that will make use of Field Programmable Grid Array (FPGA) boards to supplement the GPU cards to boost the computational power of the base computer by several orders of magnitude. Currently, the system has 35 FPGA cards and 4 GPUs in 5 expansion chassis. When fully functional, this should substantially reduce the time to complete a complex Monte Carlo calculation from weeks to minutes. ROB has experienced technical difficulties with the Hewlett-Packard (HP) Proliant Gen-10 server originally selected to perform as the host for the FPGA array. The current system BIOS is not compatible with the use of multiple PCIe expansion chassis that are necessary to accommodate all of the FPGA boards. A BIOS rewrite or upgrade will take around 2 years. As an alternative, ROB has acquired an OSS (One Stop Systems) 2U EOS server, which is guaranteed to support multiple PCIe expansion chassis, since they are all from the same vendor. This new system has been installed and is being tested and is currently configured for 32 FPGA cards and 2 GPU cards. The original HP server has been configured with a single PCIe expansion chassis (Trenton Systems), which hosts 4 FPGA cards and 2 GPU cards. The server BIOS will support a single PCIe chassis. This system will be utilized for software development and testing prior to exporting the finalized product to the production server.