模型与数据耦合驱动的快速四维EPRI肿瘤氧成像

基于氧图像导航的精准放疗是指对肿瘤的不同氧气浓度区域施加不同的剂量。电子顺磁共振成像（EPRI）是一种高精度高分辨氧气浓度成像方法。当前EPRI氧成像的瓶颈问题是扫描时间过长。为通过减少投影数据的采集而提高扫描速度，本项目研究四维EPRI稀疏重建方法，以实现快速成像。在前期模型驱动三维重建方法基础上，项目拟将基于压缩感知的模型驱动方法和基于深度学习的数据驱动方法耦合，设计全新的四维图像重建方法。根据EPRI投影信号的特点，确定最优数据保真项；用核TV范数构建先验型正则器，以压制稀疏伪影；通过深度学习构建经验型正则器，以压制噪声。采用Chambolle-Pock算法框架设计模型求解算法。用数学模体、物理模体及小鼠肿瘤数据验证、评估并定型整套稀疏重建方案。模型驱动与数据驱动的耦合、噪声的多层次建模和压制、直接四维重建等新技术的综合运用，将进一步提高EPRI肿瘤氧成像的扫描速度和重建精度。

The precise radiation based on oxygen image navigation means tumor region of different oxygen concentration should be applied different dose. Electron paramagnetic resonance imaging (EPRI) is an advanced oxygen imaging method of high precise and high resolution. Currently, the bottleneck of EPRI oxygen imaging is too long scan time. To speed up the scan process by reducing projection data, the project focuses on four dimensional EPRI sparse reconstruction method to implement fast imaging. Based on our three dimensional model-driven reconstruction methods, the project will combine the model-driven method based on compressed sensing and the data-driven method based on deep learning to develop new four dimensional image reconstruction method. We will explore the optimal data fidelity term according to the characteristics of the EPRI projection signals, construct a prior regularization term via nuclear TV norm to suppress the sparse artifacts and construct an experimental regularization term by deep learning to suppress noise. We will design model solving optimization algorithm based on Chambolle-Pock algorithm framework. The mathematical phantoms, physical phantoms and mouse-tumor data will be used for validating, evaluating and finalizing the whole sparse reconstruction program. The combined use of model- and data-driven methods, the multi-layered noise modeling and suppression, the fully four dimensional reconstruction and other new techniques will further improve the scan speed and reconstruction accuracy of EPRI tumor-oxygen imaging and lay the foundation for finalizing the fast imaging technology in EPR oxygen imager for small animals.