Quantitative Oncological PET Image Generation and Analysis

目的：我的研究计划的长期目标是显着推进定量断层扫描核医学（PET和SPECT成像），并帮助使定量分析成为常规实践的一个组成部分，显着改善癌症的评估，并帮助实现精确（个性化）治疗。我的提案的短期目标是开发和优化适用于PET成像的先进（i）数据采集，（ii）图像重建和（iii）图像分析方法。 方法：我于2018年从约翰霍普金斯大学招募，在那里我领导了13年的高分辨率PET成像物理学项目。我在UBC/BC癌症的计划旨在开发强大的定量PET成像范例，可以转化为癌症的常规成像。我们的研究工作将在3个前沿： 1)数据采集：我的实验室率先提出了动态全身（DWB）PET成像的概念，最终推出了PET供应商临床产品（2017年由西门子推出的多参数PET套件）。与此同时，这一框架也面临着挑战，因为常规PET扫描的持续时间越来越短：我们的目标是研究和发现DWB PET的新协议，以便在常规成像中广泛采用。 2)图像重建：我们将开发和验证先进的3D和时空4D重建算法，包括先进的模型，动态以及运动信息。我们将利用机器学习、深度学习、字典学习以及内核方法来生成具有上级质量和定量准确性的图像。或者，重要的是，我们将研究是否可以通过显着降低剂量获得质量相当的图像。 3)图像量化：我们的实验室一直处于放射组学分析的最前沿，它生成形状和纹理特征，以更好地量化放射图像。在拟议的研究计划中，我们将采用两个框架：（i）使用明确定义的放射组学特征，然后应用机器学习来构建评估癌症的高级模型;（ii）直接使用深度学习方法来隐式推导PET中的重要摄取模式。 影响：我们的努力旨在改变常规成像中使用的PET协议，目前面向定性而不是定量评估。我们提出的PET数据采集和重建方法能够生成更恰当地反映潜在生理过程的定量图像。我们的工作还实现了一个范例，其中在常规评估中量化肿瘤形状和摄取的异质模式（例如肿瘤缺氧）。总的来说，我们的放射组学和机器学习工作旨在改变PET图像处理和定量领域，并实现对PET图像的更全面评估，最终可以显著改善癌症的管理和治疗。

Objectives: The long-term objective of my research program is to significantly advance quantitative tomographic nuclear medicine (both PET and SPECT imaging), and to help make quantitative analysis an integral part of routine practice, significantly improving assessment of cancer and to help make precision (personalized) therapy a reality. The short-term objective of my proposal is to develop and optimize advanced (i) data acquisition, (ii) image reconstruction, and (iii) image analysis methods as applied to PET imaging. Methods: I was recruited in 2018 from the Johns Hopkins University, where I led a program of high-resolution PET imaging physics for 13 years. My program at UBC / BC Cancer aims to develop powerful paradigms of quantitative PET imaging that can translate to routine imaging of cancer. Our research efforts will be on 3 frontiers: 1) Data acquisition: My lab pioneered the concept of dynamic whole-body (DWB) PET imaging, culminating in a PET vendor clinical product release (Multiparametric PET Suite by Siemens in 2017). At the same time, there is a challenge to this framework, in that routine PET scans are performed in increasingly short durations: we aim to investigate and discover novel protocols for DWB PET that can enable wide adoption in routine imaging. 2) Image reconstruction: We will develop and validate advanced 3D and spatiotemporal 4D reconstruction algorithms, including advanced models, dynamic as well as motion information. We will utilize machine learning, deep learning, dictionary learning, as well as kernel methods, to produce images with superior quality and quantitative accuracy. Alternatively, and importantly, we will study whether images of comparable quality can be obtained by significantly lower doses. 3) Image quantification: Our lab has been at the forefront of radiomics analysis, which generates shape and texture features to better quantify radiological images. In the proposed research program, we will pursue two frameworks: (i) use of explicitly-defined radiomic features followed by application of machine learning to construct advanced models for assessment of cancer; and (ii) direct use of deep learning methods to implicitly derive important patterns of uptake in PET. Impact: Our efforts aim to alter PET protocols as used in routine imaging, presently geared towards qualitative as opposed to quantitative assessment. Our proposed PET data acquisition and reconstruction methods enable generation of quantitative images that more appropriately reflect the underlying physiological processes. Our work also enables a paradigm where heterogeneous patterns of tumour shape and uptake are quantified in routine assessment (e.g. tumour hypoxia). Overall, our radiomics and machine learning efforts aim to alter the PET image processing and quantitation landscape, and to enable more comprehensive assessment of PET images, that can ultimately result in significant improvements in management and treatment of cancer.