基于数据挖掘的多模医学图像连续形变识别  

Continuous Deformation Recognition of Multimode Medical Images Based on Data Mining

作　　者：刘亮[1] 王静宇[2] 唐思源[1] LIU Liang;WANG Jing-yu;TANG Si-yuan(Baotou Medical College,Inner Mongolia University of Science and Technology,Baotou Inner Mongolia 014040,China;Department of Information Engineering,Inner Mongolia University of Science and Technology,Inner Mongolia Baotou 014040,China)

机构地区：[1]内蒙古科技大学包头医学院,内蒙古包头014040 [2]内蒙古科技大学信息工程学院,内蒙古包头014040

出　　处：《计算机仿真》2023年第12期247-251,共5页Computer Simulation

基　　金：内蒙古自然科学基金(2020MS06001);教育部高等教育司2021年第二批产学合作协同育人项目(202102490017);内蒙古自治区卫生健康科技计划项目(202201395)。

摘　　要：为了提升病变区域识别精度,得出有效的医学图像信息,提出基于数据挖掘的多模医学图像连续形变识别方法。通过直方图修正检测图像灰度值,利用均衡化将像素均匀分布在各个区间,突出影像中重点灰度范围,通过灰度共生矩阵明确目标像素灰度和位置参数;根据数据库事务集合,将图像分成不同标识区域;在信息增益率中加入分裂信息,分割挖掘出对应形变区域;考虑到相邻图像帧会产生时间差,在不发生粘连和分离的前提下迭代更新,完成医学图像的连续形变区域识别。仿真中以胃镜影像作为实验样本,证明所提算法能够精准识别存在形变的区域,清晰划分出形变区域。In order to improve the accuracy of identifying lesion regions and obtain effective information of medical images,this article presented a method of identifying continuous deformation of multi-modal medical images based on data mining.Firstly,the gray value of the image was detected by histogram correction,and then pixels were evenly distributed in every interval by equalization.In the meanwhile,the key gray range in images was highlighted.Moreo⁃ver,the gray and position parameters of the target pixel were determined by the gray-level co-occurrence matrix.Ac⁃cording to the database transaction set,the image was divided into different identification areas.After that,split in⁃formation was added into the information gain ratio to segment and mine the deformation region correspondingly.With consideration of the time difference between adjacent image frames,iterative updating was performed without adhesion and separation.Finally,continuous deformation regions in medical images were identified.In the simulation,gastro⁃scope images are taken as experimental samples.The proposed algorithm can accurately identify the region with de⁃formation,and clearly divide deformation regions.

关 键 词：数据挖掘 多模医学图像 图像预处理 灰度均衡 连续形变识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]