Characterizing, optimizing, and harmonizing cancer detection with PET imaging

项目摘要 对较小和早期癌症的检测和诊断显著提高了患者的生存机会。 生存使用氟18-氟脱氧葡萄糖（FDG-PET）的正电子发射断层扫描（PET）成像 提供正常组织与癌组织的功能或代谢评估，自2000年以来， 在临床上广泛用于许多癌症的检测和诊断。十多年前，我们的研究小组 其他研究表明，测量和提高FDG-PET的探测能力是可行的 通过调整采集和图像重建参数来进行癌症成像。为此可 系统地通过评估对复制人类表现的观察者模型（即放射科医生）的影响 或核医学医师）。然而，在当时，要理解这一点在不同国家之间的变化是具有挑战性的。 具有不同分辨率、灵敏度和重建算法的系统，或者如果它们被操作 在不同的成像部位。 在过去的十年里，扫描仪的分辨率、灵敏度和 重建算法，以及飞行时间PET成像的常规采用。与此同时， 改进对示范观察员的理解和采用，以及采用或 协调多个PET制造商和成像站点的方法。最近， 机器智能算法的发展，如深度神经网络，用于图像重建 和图像分析，这有可能提高性能。 我们建议利用这些发展来描述、优化和协调癌症 PET成像检测。三个具体目标是：（1）开发表征方法（即， 测量）用于FDG PET成像的检测性能。(2)使用校准到 现代物理系统，我们将确定如何优化癌症检测作为采集的函数 和图像重建参数。(3)最后，我们将开发一个平台无关（供应商不可知） 可以应用于系统和成像站点的标准。这将导致多站点路线图， 多供应商实施方法，优化癌症检测能力，从而改善患者 结果。

Project summary Detection and diagnosis of smaller and earlier-stage cancers significantly improves a patient's chances of survival. Positron emission tomography (PET) imaging using fluorine 18–fluorodeoxyglucose (FDG-PET) provides a functional or metabolic assessment of normal versus cancerous tissues, and since 2000 has been widely used clinically for the detection and diagnosis of many cancers. Studies over a decade ago by our group and others had shown that it was feasible to both measure and improve the detection ability of FDG-PET imaging for cancer by adjusting acquisition and image reconstruction parameters. This could be done systematically by evaluating the effect on observer models that replicated human performance (i.e. radiologists or nuclear medicine physicians). At the time, however, it is challenging to understand how this varied across systems with different resolutions, sensitivities, and reconstruction algorithms, or if they were operated differently across imaging sites. Over the last decade there have been dramatic improvements in scanner resolution, sensitivity, and reconstruction algorithms, as well as the routine adoption of time-of-flight PET imaging. In parallel there has been an improved understanding and adoption of model observers, as well as pathways for adoption or harmonization of methods across multiple PET manufacturers and imaging sites. Most recently there has been the development of machine intelligence algorithms, such deep neural networks, for both image reconstruction and image analysis, which have the potential to improve performance. We are proposing to take advantage of these developments to characterize, optimize, and harmonize cancer detection with PET imaging. The three specific aims are: (1) Develop methods for characterization (i.e. measurement) of detection performance for FDG PET imaging. (2) Using a model system calibrated to a modern physical system we will then determine how to optimize cancer detection as a function of acquisition and image reconstruction parameters. (3) Finally we will develop a platform-independent (vendor agnostic) standard that can be applied across systems and imaging sites. This will lead to a roadmap for multi-site and multi-vendor implementation approaches that optimizing cancer detectability and thus improved patient outcomes.