摘要
针对传统肺结节识别中对感兴趣区域(ROI)进行特征计算时造成的一些隐含结构信息丢失的问题,提出了矩阵输入模式的多核学习矩阵化最小二乘支持向量机识别算法(MKLMatLSSVM)。该算法将多核方法与矩阵化最小二乘支持向量机(MatLSSVM)相结合,继承了二者优点,涵盖了多种类型的核。为验证算法的有效性,将其应用于肺结节识别。实验采用20个患者的CT图像,提取的ROI中含80个结节及190个假阳。结果表明,MKL-MatLSSVM算法在使用混合核及RBF核时,能兼顾敏感度、准确度和特异度指标,且其接收者操作特征(ROC)曲线下面积均可达到0.96以上,优于先前两种包括MatLSSVM在内的支持向量机(SVM)算法。
Traditional methods for lung nodule recognition need to extract the features of the Region of Interests （ROIs）, which usually leads to loss of some implicit structure information. To avoid this problem, a novel Multiple Kernel Learning method based on Matrix Least Square Support Vector Machine （MKL-MatLSSVM） is proposed. This method combines the advantages of both MKL method and MatLSSVM, and supports direct matrix input, suitable for image identification. To verify the effectiveness of the proposed method, it was applied to identify lung nodules in CT images of 20 patients, where the extracted ROIs contain 80 nodules and 190 false positives. The results show that when using hybrid or Radial Basis Function （RBF） kernels in MKL-MatLSSVM, the resulting sensitivity, accuracy and specificity can be balanced, and the area under the Receiver Operating Characteristic （ROC） curve can reach 96~, better than other two previous Support Vector Machine （SVM） methods that include MatLSSVM.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2014年第2期508-515,共8页
Journal of Jilin University:Engineering and Technology Edition
基金
吉林省科技发展计划项目(201201129)
长春工业大学理工科基金项目(2011LG04)
2012年国家级'大学生创新创业训练计划'项目(201210190017)
吉林省教育厅科研专项项目(2014142)
