支持向量机在肺结节CT图像中的应用  

Application of support vector machine for pulmonary nodules in CT image

作　　者：王晶晶[1] 孙涛[1] 赵枫朝[2] 李霞[1,3] 蔡博文[2] 朱晓萌[2] 郭秀花[1,3] 

机构地区：[1]首都医科大学公共卫生学院,北京100069 [2]北京工业大学计算机学院,北京100022 [3]北京市临床流行病学重点实验室,北京100069

出　　处：《北京生物医学工程》2013年第5期528-530,535,共4页Beijing Biomedical Engineering

基　　金：国家自然科学基金(81172772;30972550);北京市自然科学基金(4112015);北京市属高等学校人才强教计划(PHR201007112)资助

摘　　要：目的探讨基于孤立性肺结节建立支持向量机预测模型效果,提高肺癌的早期诊断率。方法对收集的55例患者的三正交位的4135张肺结节CT图像,应用Curvelet变换进行纹理提取,对提取的476个特征值应用支持向量机进行良恶性分类并预测,应用符合率、敏感度和特异度对预测结果进行评估。结果 CT图像符合率为78.0%,敏感度为88.6%,特异度为24.0%。结论 Curvelet转换提取三正交位肺结节纹理,用支持向量机建立预测模型,一定程度上有助于早期发现和诊断肺癌。Objective To evaluate the prediction results of solitary pulmonary nodules using support vector machine model in order to improve the detection and diagnosis of early-stage lung cancer. Methods We collected 4135 CT images of benign or malignant solitary pulmonary nodules in three dimensions from 55 patients. Four hundred and seventy-six Curvelet transform textural features were used as parameters to establish support vector machine model, and the classification consistency, sensitivity and specificity were used to evaluate the forecast results. Results The classification consistency, sensitivity and specificity for the model were 78.0% ,88.6% and 24.0% , respectively. Conclusions Based on Curvelet transform to extract textural features, support vector machine can improve the diagnosis of early-stage lung cancer to some extent.

关 键 词：CT图像 肺结节 Curvelet转换 纹理提取 支持向量机 

分 类 号：R318.04[医药卫生—生物医学工程]