基于SOM聚类的多模态医学图像大数据挖掘算法  

Large data mining algorithms of multi-modal medical image based on SOM clustering

作　　者：李晓峰[1] 李东[2] LI Xiaofeng;LI Dong(Department of Information Engineering,Heilongjiang International University,Harbin 150025,China;School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]黑龙江外国语学院信息工程系,黑龙江哈尔滨150025 [2]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《西安工程大学学报》2019年第4期462-467,共6页Journal of Xi’an Polytechnic University

基　　金：黑龙江省自然科学基金面上项目(F2016038);中国管理科学研究院项目(CMAS180305)

摘　　要：针对现有医学图像大数据挖掘算法不能较好解决帧率重叠行为,影响医学图像成像质量,从而引发图像数据冗余度攀升现象的问题,提出一种基于SOM聚类的多模态医学图像大数据挖掘算法。在SOM拓扑结构中,设定聚类变量的初始权值条件,完成基于SOM聚类的多变量分析。再通过预处理原始医学图像的方式,判断多模态图像的集成正负性相关属性,建立唯一的图像聚类融合标准,完成多模态医学图像的集成聚类融合处理。在此基础上,构建完整的大数据决策树结构,通过判定医学图像多模态种类的方式,达到并行化改进图像挖掘节点的目的,实现基于SOM聚类的多模态医学图像大数据挖掘算法。在标准化的大数据挖掘平台中,连接图像属性监测设备,分析多次运行数据结果可知,在SOM聚类策略影响下,医学图像成像质量提升明显,图像数据自身冗余度不再无限攀升,帧率重叠行为得到有效控制。In view of the fact that the existing big data mining algorithms for medical images fail to solve the problem of frame rate overlapping and affect the imaging quality,the problem of rising redundancy of image data arises a multi-modal medical image big data mining algorithm based on SOM clustering is thus proposed.In the SOM topology,the initial weights of clustering variables are set to complete the multivariate analysis based on SOM clustering.Then,by preprocessing the original medical images,the positive and negative related attributes of the integration of multi-modal images are judged,and the only image clustering fusion standard is established to complete the integrated clustering fusion processing of multi-modal medical images.On this basis,a complete decision tree structure for large data is constructed.By determining the multi-modal types of medical images,the aim of improving image mining nodes in parallel is achieved,and the research into the algorithm is realized.In the standardized large data mining platform,the image attribute monitoring equipment is connected and the results of running data are analyzed.Under the influence of SOM clustering strategy,the image quality is improved obviously,and the redundancy of image data itself is no longer unlimited,and thus the effective control of frame rate overlap is achieved.

关 键 词：SOM聚类 多模态图像 大数据挖掘 初始权值 集成正负性 节点并行化 帧率重叠 

分 类 号：TP[自动化与计算机技术]