摘要
目的:探讨C4.5决策树算法结合主成分分析法(PCA)在新疆高发病食管癌X钡剂造影图像分类中的应用。方法:选取新疆高发病食管癌图像200张,其中蕈伞型和溃疡型图像各100张。对图像进行归一化、去噪和空间转换等预处理;对图像进行二尺度小波变换提取图像的低频信息,然后对其进行灰度共生矩阵法提取图像的特征;采用主成分分析(PCA)法对所提取的特征进行筛选;通过构造决策树C4.5算法分类器来验证特征的分类能力。结果:使用决策树C4.5算法分类器,对主成分分析获取的特征及综合特征进行分类。PCA选择的特征分类准确率为95%;使用综合特征分类准确率为80%。结论:综合特征的分类准确率与PCA选择的特征相比较低,表明在进行分类时,冗余特征可能会降低分类准确率;而本研究采用PCA选择后的特征分类准确率较高,表明该算法能有效减少冗余特征,弥补了过高维数的特征向量易引起维数灾难的问题,从而使得分类准确率得到了提高。一定程度上为后续的其它组织器官的特征提取提供了依据。
bjective：In this paper, to discussing C4.5 decision tree combine principal component analysis the application of the classification in Xinjiang of Kazak esophagus cancer of X- ray barium images.Methods：It select 200 picture in Xinjiang of Kazak esophagus cancer of X-ray barium angiogram images,among mushroom umbrella,esophageal ulcer types images,100 copies. For one thing,for Xinjiang of Kazak esophagus cancer of X-ray barium angiogram images are processed normalized, the noise removed andenhanced;then those images are decomposed in two level by using wavelet transformation,which extract their low frequency information.Finally use the texture feature extraction method basing on GLCM to extract Xinjiang of Kazak esophagus cancer of X-ray barium. using of PCA method for the selection of the characteristic value. Used of decision tree C4.5 algorithm in Xinjiang Uygur Herbs image on classification characteristics of the comprehensive evaluation. Results：Using C4.5 decision tree to construct classifier,for PCA to obtain the characteristic and the characteristics of comprehensive two kinds of methods to classify.with the result that characteristics of the PCA to choose classification accuracy rate reached to 95%;characteristics of comprehensive classification accuracy reached 80%.Conclusion：Comprehensive characteristics of classification accuracy is lower than the features of PCA to select,show that redundant features can reduce classification accuracy; after using PCA selected the characteristics, the classification accuracy is higher,its the algorithm can effectively reduce the redundant features. It make up a high dimension feature vector easy cause the problem of dimension disaster, and thus improving the overall classification efficiency.It can to a certain extent, for the subsequent provide a basis for the feature extraction of other tissues and organs.Xinjiang of kazak esophagus cancerPCAC4.5 decision treetexture featureimage
作者
孔喜梅
木拉提·哈米提
严传波
孙静
阿布都艾尼·库吐鲁克
艾赛提·买提木沙
姚娟
King Ximei Murat Hamit Yan Chuanbo Abdugheni Kutluk Asat Matmusa Yao Juan(College of Medical Engineering Technology,Xinjiang Medical University, Urumqi 830011, China College of Public Health,Xinjiang Medical University, Urumqi 830011, China Department of Radiology,First Affiliated Hospital,Xinjiang Medical University, Urumqi 830011, China)
出处
《科技通报》
北大核心
2016年第9期52-57,共6页
Bulletin of Science and Technology
基金
国家自然科学基金(81460281,81160182,61201125)
江西民族传统药协同创新项目(JXXT201401001-2)
