基于模糊C均值和随机漫步的CT肝脏图像分割算法研究  

A Hybrid Method Based on Fuzzy C-Means Algorithm and Random Walker Algorithm for CT Liver Image Segmentation

作　　者：王琴琴 WANG Qinqin(Department of Radiology,Women's Hospital of Nanjing Medical University,Nanjing Maternity and Child Health Care Hospital,Nanjing Jiangsu 210004,China)

机构地区：[1]南京医科大学附属妇产医院(南京市妇幼保健院)放射科,江苏南京210004

出　　处：《中国医疗设备》2020年第9期107-110,共4页China Medical Devices

摘　　要：目的提出一种基于模糊C均值(Fuzzy C-Means,FCM)和随机漫步的CT肝脏图像分割算法。方法仿真实验图像选自MIDAS和3Dircadb数据库,先采用中值滤波器进行图像预处理,平滑且保持图像边缘信息。接着,采用FCM算法将图像初步划分为灰度特征相似的目标与背景区域。最后,采用随机漫步算法进一步勾画肿瘤边界。采用重叠错误率、相对误差和Dice相似性系数作为图像分割结果评价指标。结果定性分析显示,基于本研究算法能准确勾画CT肝脏肿瘤边界。定量评价显示,采用本研究分割算法获得的图像分割评价指标均优于其他文献报道的算法,且与5名经验丰富技师分割结果相吻合,其中平均重叠错误率、相对误差和Dice相似性系数分别达到15.61%±5.32%、4.02%±3.01%、0.81±0.06。结论联合使用FCM算法和随机漫步算法能准确有效的分割CT肝脏肿瘤图像,具有较高的临床应用价值。Objective To propose a hybrid segmentation method for liver tumors from CT imaging based on Fuzzy C-means(FCM)clustering algorithm with random walker algorithm.Methods The simulation images were selected from Midas and 3 Dircadb databases.Firstly,the median filter was used for image preprocessing to smooth and keep the edge information of the image.Then,the CT liver image was initially divided into objects and background which has similar features by FCM method.Finally,random walker algorithm was adopted to draw the boundary of liver tumor for further segmentation.The accuracy ofimage segmentation resultsweremeasured by mean overlap error(OE),relative difference(RD),and Dice similarity coefficient(DSC).Results Qualitative analysisresults showed that the boundary of liver tumors was delineated accurately with the proposed method.Quantitative evaluation results showed that the image segmentation evaluation indexes obtained by using the segmentation algorithm of this study were superior to the algorithms reported in other literatures,and were consistent with the segmentation results of 5 experienced technicians.Moreover,the average overlap error rate,relative error and Dice similarity coefficient were 15.61%±5.32%,4.02%±3.01%,0.81±0.06,respectively.Conclusion The combination of FCM algorithm and random walk algorithm can segment CT liver tumor image accurately and effectively,which has high clinical application value.

关 键 词：图像分割 模糊C均值 随机漫步 肝脏肿瘤 CT显像 

分 类 号：R197.39[医药卫生—卫生事业管理] TP391.41[医药卫生—公共卫生与预防医学]