项目名称: 面向在线检索的医学影像多特征降维方法研究
项目编号: No.61502319
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 计算机科学学科
项目作者: 申华磊
作者单位: 河南师范大学
项目金额: 21万元
中文摘要: 基于内容的医学影像检索一直是近年来学术界的研究热点。使用数据降维方法对从医学影像提取的多种高维特征向量进行降维、消除其中的冗余信息,是实现高效医学影像检索的关键之一。在大数据环境下,医学影像检索呈现在线检索的新应用形态,医学影像多特征降维因而面临三个亟待解决的关键问题:特征互补问题、增量更新问题和跨库检索问题。项目针对这三个关键问题、围绕面向在线检索的医学影像多特征降维一个主题展开研究。提出多特征空间组合模型,在此基础上:1)提出基于子空间互补的医学影像多特征组合降维方法,以解决特征互补问题;2)提出基于局部近邻结构更新对齐的医学影像低维特征表达更新方法,以解决增量更新问题;3)提出基于子空间数据分布差异最小化的医学影像低维特征表达迁移方法,以解决跨库检索问题。项目将在多个医学影像数据集上,设计基于内容的医学影像检索实验,验证上述研究内容的有效性。
中文关键词: 基于内容的医学影像检索;数据降维;多特征互补;增量更新;跨库检索
英文摘要: Content-based medical image retrieval (CBMIR) continues to be a hot topic in research community in recent years. To effectively enhance performance of CBMIR, employing dimensionality reduction methods to remove redundancy contained in high dimensional visual feature vectors extracted from medical images is one of the critical issues. With the forthcoming of big data era, CBMIR enters the new stage termed online retrieval. Dimensionality reduction for medical images thus faces three critical problems: multiple features complementation, incremental update, and cross-dataset retrieval. To address these challenges, this proposal proposes an online-retrieval-oriented medical image dimensionality reduction strategy. A multiple feature spaces combination model is designed, based on which: 1) a new method for medical image multiple features dimensionality reduction based on subspaces complementation is proposed, to solve the “multiple features complementation” problem; 2) a novel approach for incremental update of low-dimensional representation of medical images based on global alignment of local neighborhood structure update is designed, to address the “incremental update” problem; and 3) a novel method for low-dimensional representation transfer of medical images based on minimization of data distribution difference among multiple subspaces is proposed, to solve the “cross-dataset retrieval” problem. This proposal will design CBMIR experiments on multiple medical image datasets to verify effectiveness of these methods.
英文关键词: Content-based medical image retrieval;Dimensionality reduction;Multiple features complementation;Incremental update;Cross-dataset retrieval