放射治疗是癌症治疗的主要手段之一。治疗过程中肿瘤退缩和结构改变会使放疗剂量偏离计划而影响疗效。以影像引导评估这种剂量偏差，及时调整放疗计划实施自适应放疗对保证疗效和减少并发症有重要意义。有良好软组织分辨率且无额外辐射的磁共振影像引导自适应放疗(MRIgART)是未来发展方向，但需解决两个关键问题：(1)如何从MRI获得组织密度支持放疗剂量计算；(2)实时快速勾画靶区与危及器官支持临床剂量体积评估。通过深度学习方法从MRI生成合成CT(SCT)图像和建立智能勾画算法是解决途径之一。目前的SCT模型研究多基于配对图像学习，而临床难以同时采集配对MR/CT图像；智能勾画研究则主要集中在CT图像。本项目拟基于申请人前期研究，采用不需配对图像的循环对抗生成网络并增加MRI序列组合训练获得高质量SCT图像，同时结合CT/MR图像勾画数据，训练SCT智能勾画模型，为MRIgART提供剂量体积验证手段。

Radiation treatment is the major therapy for malignant disease. However, the tumor shrinking and the anatomic changes during the treatment course will make the radiation dose distribution deviate from the original plan and affect the curative effect. Image-guided evaluation of these dose deviation and timely adjustment of treatment plan, implementing the so-called adaptive radiation therapy(ART), are of great significance to ensure the therapy outcome and reduce complications. With an excellent soft tissue resolution and no extra radiation, the magnetic resonance images (MRI) guided adaptive radiotherapy (MRIgART) is the emerging development but two key problems need to be solved: (1) how to obtain tissue density from MRI to support radiotherapy dose calculation; (2) if it is achievable to rapidly contour the volume of treatment target and organs at risk from the online images to support the real time dose-volume evaluation. One of the solutions is to generate synthetic CT (SCT) images from MRI by deep learning and establish an intelligent auto-segmentation algorithm model. At present, the research of SCT model is mostly based on paired image learning, yet it is difficult to collect paired MR/CT images at the same time in clinic, while the research of intelligent segmentation is mainly focused on CT images. Based on the applicant's previous research, this project proposes to use the cycle generative adversarial network of deep leaning, which do not need paired images, and add MRI sequence combination for training to obtain high quality SCT images. At the same time, training another unique intelligent segmentation model with combined manual-contoured CT and MR data set to provide applicable tools of dose-volume verification for MRIgART.