癌症是严重威胁人类健康的疾病，“癌症和肿瘤基因图谱”计划整合了30多种癌症的病理图像与基因数据，为探索与癌症预后相关的生物标志物提供了强有力的支持。由于癌症的发病机制极其复杂，而所采集的影像与基因数据维度高，异构性强，且对其精确标记代价昂贵，给探索病理机制并构建有效的预后预测模型带来相当大挑战。为此，本项目 1)通过结合已有诊断结果提升预后分析的性能，拟提出一种诊断信息辅助下联合基因影像数据的特征选择框架，实现有效的预后预测。2)考虑到癌症的预后与不同层次基因调控的相关性，拟提出一种多层次融合影像与多模态基因数据的学习方法实现预后预测。3）考虑对高维基因影像数据精确标记代价昂贵，拟提出一种粗粒度标记下联合基因影像数据的预后预测方法。研究成果将为癌症的病理理解与预后预测提供可靠的技术支持。

Cancer is the disease of serious threat to mankind health. The Cancer Genome Atlas project has incorporated the pathological image data and genomic data for more than 30 cancer types, which provides a strong support for exploring the survival-associated biomarkers for prognosis prediction. Due to the complex pathogenesis of cancer disease, along with the problem that the collected high-dimensional imaging and genomic data are usually with heterogenous patterns and the high cost of precise labeling, it has become a huge challenge to explore the pathological mechanism and build effective prognosis prediction models for cancers.Therefore, 1) Concerning the auxiliary diagnosis information can help improve the prognosis performance, we construct a feature selection framework that combine both image and genomic data for prognosis prediction by the aid of diagnosis information. 2）Concerning the gene regulation is a complex process controlled at multiple levels, we propose a novel multi-level association approach to fuse image and multi-modal genomic data for prognosis prediction. 3) Concerning the high-cost of precisely labelling the high-dimensional image and genomic data, we propose a novel algorithm to fuse image and genomic data for survival analysis that only relies on coarse-grained label information. This project will provide data-driven support for the pathological understanding and prognosis prediction of cancers.