面向肺癌临床辅助诊疗决策的多模态数据融合分析关键技术研究

项目名称： 面向肺癌临床辅助诊疗决策的多模态数据融合分析关键技术研究

项目编号： No.61502091

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 其他

项目作者： 曹鹏

作者单位： 东北大学

项目金额： 21万元

中文摘要： 肺癌的死亡率位居最常见恶性肿瘤之首，其发病率在我国亦逐年升高。研究表明早期诊断和个性化治疗对提高肺癌患者的治愈率和预后效果起着至关重要的作用。由于肺癌疾病种类的多样性和病理变化的复杂性，有效提高肺癌定性诊断和预后分析的准确率是对现有知识结构和技术水平的巨大挑战。因此本课题将以多模态的临床数据为研究对象，开展肺结节/肿块良恶性诊断、预后判断分析、治疗方案推荐和治疗疗效评估临床任务的关键技术研究。针对高维度、错分代价不同、类别未标记、分布复杂及时序性等肺癌临床数据分析的需求，基于深度学习网络的多模态融合分析方法，结合半监督学习、自适应聚类学习、相关性分析及多任务学习技术，实现跨模态数据统一表示，并根据临床任务设计有效的智能数据分析方法，构建预测模型。在此基础上为了验证理论研究的正确性，构建肺癌临床辅助诊疗决策系统。项目研究为肺癌的诊断和治疗提供科学的决策，具有一定的学术价值与研究意义。

中文关键词： 多模态融合；临床决策支持；深度学习

英文摘要： The mortality of Lung cancer is primary among the most common malignant tumors, and its incidence increased year by year in China. Some research studies show that early diagnosis and personalized treatment plays a vital role in improving the cure rate and the prognosis of patients with lung cancer. Due to the diversity and the complexity of pathological, improving the accuracy of diagnosis and prognosis of lung cancer is a great challenge. This research will focus on the study of fusion analysis of lung cancer clinical data with multi-modality for diagnosis, prognosis decision, recommendation personalized therapy and assessment of the therapy. The aim of the study is to design intelligent data analysis for high dimension, different class misclassification cost, limited labeled, complex data distribution and time series based on the fusion of multi-modality analysis, combined with the methods of semi-supervised learning, adaptive clustering, correlation analysis and multi-task learning, in order to achieve homogeneous representation of multi-modality, develop effective data analysis methods and construct prediction modal for clinical tasks. This project can provide the new ideas for the Clinical Decision Support for early diagnosis and treatment of lung cancer, and bring academic value and research significance.

英文关键词： Fusion of multi-modality ;Clinical decision support;Deep learning