数字化的核医学影像图谱在核医学影像分析、药物研发、精准治疗等领域发挥着重要作用。目前脑部核医学影像图谱已经促进了脑科学研究的进步，但是全身范围图谱仍然相对缺乏。本课题基于上千例正常中国人全身18F-FDG PET/CT影像构建PET/CT影像图谱，开发自动算法实现图谱与个体PET/CT影像的配准。本研究提出构建一种参数化的可变图谱，采用统计建模方法从大量中国人样本数据中学习解剖形态和功能代谢参数，利用这些参数调控图谱的器官形态和像素灰度发生变化，使图谱通过自身变化模拟不同人的影像，给出针对受试个体的形态和功能参数描述。本研究将设计自动参数优化算法实现图谱与个体影像的配准，研发层级式深度神经网络完成全身器官快速定位，借助器官间的形态与灰度相关性来配准不清晰或病变器官。本课题将为核医学影像分析研究提供基础性的、具有我国自主知识产权的、包含着中国人影像特征的图谱及相应的配准算法支持。

Digital nuclear medicine image atlases are playing important roles in nuclear medicine image analysis, drug development and individualized precision medicine. So far, the usage of brain atlases has dramatically promoted the progress in neuroscience, but the whole-body atlases are still lacking. This project aims to construct a set of whole-body PET/CT atlases of Chinese cohort based on over 1000 18F-FDG PET/CT images of normal Chinese subjects. Automated atlas registration method will be developed to match the atlas to individual PET/CT images. For atlas construction, we propose to develop a parametric atlas which can adapt its organ shapes and pixel intensities to match individual PET/CT images. Anatomical and functional parameters of the atlas will be learnt from a large training dataset using the statistical modelling approach. An automated parameter optimization method will be developed to register the atlas to individual images. A hierarchical deep neural network architecture will be constructed to rapidly detect the organs in whole-body range, and the inter-organ relationships of shape and intensity will be used to robustly detect the low-contrast and diseased organs. This project will yield intellectual properties of China’s own PET/CT atlas, providing fundamental data and methodology supports to Chinese-specific nuclear medicine image analysis.