摘要
针对肺部CT切片图像中血管横截面与肺结节成像特征类似导致无法有效剔除假阳性结节的问题,提出根据孤立性肺结节的特征对感兴趣的区域进行进一步提取候选结节来剔除干扰肺结节检测的假阳性结节。为提高肺结节特征提取的有效性和肺结节的分类准确性,通过Relief特征加权极限学习机(ELM-extreme learning machine)对候选结节特征数据进行预处理,减小训练样本中关联度较小的特征对分类结果的影响。对LIDC数据集中肺部CT影像进行实验,实验结果表明,所提Relief-ELM可有效降低误诊率和漏诊率,提高孤立性肺结节识别的准确率。
In the CT image slices,the appearance of the vascular cross section is similar to the pulmonary nodules so that the false positives can not be recognized effectively.It is suggested that the candidate nodules should be extracted from the region of inte-rest according to the characteristics of the pulmonary nodules to remove the false positive nodules that interfered with the detection of pulmonary nodules.To improve the effectiveness of pulmonary nodule extraction and the accuracy of classification of pulmonary nodules,the feature data of candidate nodules were pretreated by Relief feature weighted extreme learning machine(ELM)to reduce the impact of the characteristics of small correlation in training samples on classification results.Experimental results show that the proposed Relief-ELM can effectively reduce the misdiagnosis rate and missed diagnosis rate,and improve the accuracy of solitary pulmonary nodule recognition by experimenting with CT images of LIDC data.
作者
陈树越
黄萍
朱军
刘佳镔
CHEN Shu-yue;HUANG Ping;ZHU Jun;LIU Jia-bin(School of Information Science and Engineering,Changzhou University,Changzhou 213164,China)
出处
《计算机工程与设计》
北大核心
2018年第10期3252-3258,共7页
Computer Engineering and Design
基金
国家自然科学基金项目(61502058)
江苏省高等学校大学生创新创业训练计划基金项目(201610292023Z)
关键词
孤立性肺结节
假阳性
候选结节
Relief特征加权
极限学习机
solitary pulmonary nodules
false positive
candidate nodules
Relief feature weighting
extreme learning machine
