肿瘤CT/CBCT/MRI/PET影像深度学习配准、分割与放疗剂量预测

癌症已成为威胁人类健康的第二大死因疾病。60%-70%的癌症患者需要进行放射治疗。恶性肿瘤临床放疗计划过程中的CT、CBCT、MRI、PET影像高精度配准、放疗靶区高精度分割、放疗剂量分布高精度预测和放疗计划多目标优化是当前实施精确放疗（ART, IGRT, IMRT, VMAT, TOMO 和 SBRT）的技术瓶颈。本项目拟联合肿瘤放疗计划过程中的所有可用影像，通过多任务、多级的生成对抗网络、变分自编码-解码网络和深度卷积神经网络无监督、半监督学习，集成多模态影像三维空间和多层次特征通道注意力机制，建立肿瘤多模态影像高精度配准、放疗靶区高精度分割、放疗剂量分布高精度预测模型，使用预测的剂量作为目标剂量进行放疗计划多目标优化，解决肿瘤智能放疗关键技术瓶颈科学问题，为使高精度、自适应、个性化的肿瘤智能放疗走向临床放疗，提供科学方法、理论依据和临床实验验证结果，具有重要科学意义和临床应用价值。

Cancer is the second leading cause of death, which is harmful in terms of human health. Radiotherapy is taken in 60%-70% of cancer patients. Now, there are also some technical bottlenecks of precision radiotherapy, e.g. ART,IGRT,IMRT,VMAT,TOMO and SBRT, for tumors, such as accurate registration of CT,CBCT,MRI and PET images of tumors, high precision segmentation of radiotherapy target volumes, high precision prediction and multi-objective optimization of radiotherapy dose distribution of radiotherapy planning for tumors. In order to break these bottlenecks in intelligent radiotherapy for tumors, we are going to integrate attention mechanisms for 3D spatial and multi-level feature channels of deep convolution neural networks for the accurate registration, high precision segmentation and prediction radiotherapy dose distribution. We intend to fuse CT,CBCT,MRI and PET images via multi-task and multi-stage generative adversarial networks, variational encoder-decoder networks or deep convolution neural networks by unsupervised learning or semi-supervised learning. We will conduct the multi-objective optimization of radiotherapy dose distribution of radiotherapy planning by the predicted radiotherapy dose distribution as the desired radiotherapy dose. The project will provide scientific approaches, theoretical basis and clinical experimental verification results, which is helpful to the clinical high precision, adaptive, personalized intelligent tumor radiotherapy. It is of great significance to scientific development and clinical application.