Generation of parametric images for FDG PET using dual-time-point scans

项目总结/摘要 正电子发射断层扫描联合计算机断层扫描（PET/CT），使用放射性标记示踪剂2- 脱氧-2-（18 F）氟-D-葡萄糖（FDG）已经成为癌症患者管理的标准成像工具。 半定量参数标准化摄取值（SUV）在临床上常规用于肿瘤摄取 定量，其在特定时间（通常为60分钟）后采集的静态PET图像上计算 示踪剂注入的时间间隔较短（通常为5-15分钟）。然而，SUV的量化准确性从一个 单次PET扫描受到示踪剂血浆清除和采集开始时间的变化的影响。双- 时间点FDGPET成像已经被深入研究并用于临床和研究， 通常一次扫描在60分钟，另一次在120分钟，显示出增强诊断的潜力。 FDG PET通过区分恶性肿瘤与炎症和正常组织的准确性。但目前 临床双时间点FDG PET研究使用两次扫描之间的相对SUV变化作为定量 指数，这不能消除示踪剂血浆清除率的变化。同时，双时间点 协议目前尚未优化和标准化，导致相互矛盾的结果。全定量 参数，示踪剂净摄取速率常数Ki，是量化FDG PET的最准确参数， 使用动态成像和房室模型计算。Ki与血浆清除率无关，或 收购开始时间。然而，长而复杂的采集协议（通常至少60分钟）， 需要动态扫描和连续动脉血液采样（或图像衍生的血液活性）， 从注射时间输入功能，限制了其在临床实践中的应用。与此同时， 参数Ki图像可以为FDG PET提供额外的异质性信息，具有挑战性 由于计算成本和对以下因素敏感，临床上使用逐体素房室建模 使用非线性最小二乘的噪声。Patlak图可用于简化Ki计算，Ki 通过逐体素拟合来生成图像。但仍需从15-30 min开始进行动态扫描 注射后和注射时的输入功能。该提案的目的是：（1）优化 使用Patlak图进行精确Ki定量的双时间点方案，无需个体患者的 输入函数，以及2）使用新技术生成高质量低噪声双时间点Ki图像 基于深度学习。在这个项目的成功，我们提出的方法可以获得可靠的肿瘤Ki 量化和参数Ki图像“免费”，而不增加任何额外的复杂性，现有的双重， 时间点方案目前在临床实践中使用，具有很大的潜力，以改善诊断和治疗 肿瘤学评估。我们希望将这种方法快速转化为临床研究，因为这是 一种后处理方法，基于使用临床使用的方案已经获得的数据， 对技术人员造成额外负担。

Project Summary/Abstract Positron emission tomography combined with computed tomography (PET/CT) using the radiolabeled tracer 2- deoxy-2-(18F)fluoro-D-glucose (FDG) has become a standard imaging tool for cancer patient management. The semi-quantitative parameter standardized uptake value (SUV) is routinely used in clinical for tumor uptake quantification, which is computed on the static PET image acquired at a certain time (typically 60 min) post tracer injection for a short interval (typically 5-15 min). However, the quantification accuracy of SUV from a single PET scan suffers from the variabilities of tracer plasma clearance and acquisition start time. The dual- time-point FDG PET imaging has been intensively investigated and used in both clinical and research studies, typically one scan at 60 min and the other at 120 min, showing the potential to enhance the diagnostic accuracy of FDG PET by differentiating malignancy from inflammation and normal tissue. However, the current clinical dual-time-point FDG PET studies use the relative SUV change between two scans as the quantification index, which cannot eliminate the variations in tracer plasma clearance. Meanwhile, the dual-time-point protocol has not been optimized and standardized currently, leading to conflicting results. The fully-quantitative parameter, tracer net uptake rate constant Ki, is the most accurate parameter to quantify FDG PET, which is calculated using dynamic imaging with compartmental modeling. Ki is independent on the plasma clearance or acquisition start time. However, the long and complex acquisition protocol (typically at least 60 min), which requires dynamic scanning and sequential arterial blood sampling (or image-derived blood activity) used as input function from the time of injection, limits its application in clinical practice. Meanwhile, generation of the parametric Ki image, which can provide additional heterogeneity information for FDG PET, is challenging clinically using voxel-by-voxel compartmental modeling due to the computational cost and being sensitive to noise using non-linear least squares. The graphical Patlak plot, can be used for simplified Ki calculation and Ki image generation by voxel-by-voxel fitting. However, it still needs dynamic scanning starting from 15-30 min after injection and input function from the time of injection. The aims of this proposal are 1) to optimize the dual-time-point protocol for accurate Ki quantification using Patlak plot without the need for individual patient's input function, and 2) to generate high-quality low-noise dual-time-point Ki images using novel techniques based on deep learning. Upon the success of this project, our proposed approach can obtain reliable tumor Ki quantification and parametric Ki image "for free" without adding any additional complexity on the existing dual- time-point protocol currently used in clinical practice, with great potential of improving diagnosis and therapy assessment in oncology. We expect the translation of this approach to clinical investigation to be fast, as this is a post-processing approach and is based on data already acquired using clinically used protocol without imposing additional burden to technologists.