结合影像和文本信息的医学病例检索  

Medical Case Retrieval Based on Combination of Images and Textual Information

作　　者：吴梦麟[1,2] 陈强[1] 孙权森[1] 

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094 [2]南京工业大学电子与信息工程学院,南京211816

出　　处：《计算机辅助设计与图形学学报》2014年第9期1430-1437,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61273251);江苏省"青蓝工程"(D040201)

摘　　要：针对计算机辅助诊断中的相似病例查找,提出一种基于多图半监督学习模型的多模态信息检索算法.首先将医学影像和文本视为互为补偿的不同模态的信息,根据2种信息分别构建图模型,并定义多图融合的半监督学习框架;然后在融合图上进行流形排序,得到最终的病例检索结果.在乳腺X线影像数据库和肺部CT影像数据库上的实验结果表明,该算法能够有效地利用影像和文本信息进行病例检索,具有更好的检索性能.For similar medical case searching in computer-aided diagnosis, a multi-modality information retrieval algorithm based on multi-graph semi-supervised learning model is proposed. This paper first regards medical images and texts as compensatory multi-modality information. The graph models for both kinds of information are constructed respectively and a multi-graph fusion framework is defined. Then, a manifold ranking procedure is generated based on this fusion graph to obtain the final medical case retrieval results. Experimental results on mammograph dataset and lung CT dataset demonstrate that the proposed algorithm can effectively utilize the image and textual information for medical case retrieval, and achieves a more effective and efficient retrieval performance.

关 键 词：病例检索 多模态信息 多图半监督学习 流形排序 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]