前列腺肿瘤图像的分割关系到放疗中靶区勾画与术中定位的精度，是前列腺癌放射治疗的关键问题。当前临床采用的医生手工勾画靶区的方法，存在工作强度大、精度低等问题，对前列腺癌的放疗精度有显著影响。研究前列腺肿瘤图像的自动分割方法具有重要的科学与临床意义。本项目将针对多模态前列腺肿瘤图像分割展开研究，主要包括：①优化图像特征提取；②形状先验构建与引入；③多模信息融合与维数约简；④高维空间分类方法；⑤多模图像联合分割框架等。本项目旨在为多模态图像分割中的若干科学问题提供知识积累，同时为前列腺癌放疗中的靶区自动勾画与肿瘤术中定位提供实用方法与工具。本项目前期研究工作已发表SCI论文10余篇，为项目实施奠定了坚实基础。

Image segmentation is a key issues in prostate cancer radiotherapy, and have a direct relation to the target area outlining and the tumor location accuracy. In current clinic, the segmentation is usually completed by doctors through manual interactions. However, this method is time-consumed with high labor intensity and sometimes not accurate enough, which can reduce the radiation precision dramatically. Reseach of automatic prostate tumor segmentation technique is with great scientific and clinical significance. Our research will focus on the key technologies about the prostate tumor segmentation based on multi-modal images. The main content includes: (1) Propose an optimal image feature extraction scheme. (2) Construct an anatomic shape prior.(3) Introduce a multi-modal information fusion model and a dimensionality reduction method. (4) Design a new classification method in high-dimensional feature space. (5)Design a co-segmentation framework combined with multi-modal image information. This project aims to accumulate scientific knowledge for multi-modal image segmentation, and also provide practical methods and tools in the target area outlining and automatic tumor location during the prostate cancer radiotherapy. Previous relevant work has published more than 10 papers cited by SCI, which laid a solid foundation for the implementation of this project.