基于CT图像3D特征的肺结节检测  

Pulmonary Nodules Detection Based on 3D Features from CT Images

作　　者：王彬 赵海 朱宏博 朴春赫 WANG Bin;ZHAO Hai;ZHU Hong-bo;PAK Chun-hyok(School of Computer Science ＆ Engineering,Northeastern University,Shenyang 110169,China.)

机构地区：[1]东北大学计算机科学与工程学院,辽宁沈阳110169

出　　处：《东北大学学报（自然科学版）》2018年第2期181-185,共5页Journal of Northeastern University(Natural Science)

基　　金：中央高校基本科研业务费专项资金资助项目(N140405004;N161608001);辽宁省科技厅重点实验室建设基金资助项目(LZ201015)

摘　　要：为了提高肺结节检测的性能,提出一种基于中心点连续性的肺结节检测算法.该算法使用基于简单线性迭代聚类超像素方法分割CT图像,并根据相似度合并超像素,进而得到肺部区域及疑似肺结节区域,降低了疑似肺结节的漏检率.根据各帧CT图像中疑似肺结节区域的中心点偏移程度评价其中心点连续性,最终判断出阳性肺结节.文中的实验数据来自于上海市胸科医院和LIDC数据库.实验结果表明,改进后算法的敏感度达到86.36%,假阳性率为1.76.A detection method based on the center continuity is proposed to improve the performance of pulmonary nodules detection. CT images are segmented by using SLIC（ simple linear iterative clustering） superpixel algorithm in the method. Superpixels are merged according to the similarity to extract pulmonary areas and suspected pulmonary nodule areas,which reduces the missing rate of suspected pulmonary nodules. Suspected pulmonary nodules are diagnosed as positive which keeping center continuous in 3 D space. All of CT images in experiments are obtained from Shanghai chest hospital and LIDC database. The experimental results of the improved algorithm show that sensitivity is 86. 36% and false positive is 1. 76.

关 键 词：肺结节 肺结节检测 肺结节分割 超像素 SILC 中心点连续性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]