基于集成分类器的新疆哈萨克族食管癌分型的研究

Classification on Xinjiang Kazak Esophageal Disease based on Integrated Classifier

探讨Bagging、Adaboost、Random Forest(RF)三种集成分类器在新疆哈萨克族食管癌分型中的应用。使用Matlab软件编程并提取图像的灰度-梯度共生矩阵和Tamura纹理特征;利用SPSS软件对提取到的混合纹理特征进行主成分分析(PCA)降维并得到新的主成分矩阵;将三种集成分类器并应用于主成分矩阵对食管癌进行分型;采用受试者工作特征(ROC)分析技术和参数评估对各分类模型进行评估。三种食管癌两两分类时:溃疡型和缩窄型、蕈伞型食管癌X线图像的纹理特征存在着一定差异性,三种分类器的分类准确率、ROC分析曲线及各参数评估值都很理想。三种食管癌综合分类时:RF分类器的分类效果明显优于其他两种分类器。将集成分类器应用于哈萨克族食管癌分型中,为哈萨克族食管癌影像学诊断提供了一定的依据,也为新疆哈萨克族食管癌的计算机诊断系统的研发奠定了基础。

To acquire a strong classification capability dealing with X- ray image of Kazakh esophageal cancer,by means of Bagging,Adaboost,Random Forest integrated classifier. The features were preprocessed and extracted based on gray gradient co- occurrence matrix and Tamura texture features using matlab. The dimensionality of features data set was reduced and the new PCA principal components matrices were gotten using spss; the classification was applied on new features matrices using integrated classifiers; the classification model was evaluated using ROC curve and parameter estimation. In terms of image classification of three different patterns of X- ray image of esophagus,the textures feature X- ray images of esophagus of ulcerated and narrowed,fungating type had some differences,the classifications of three types of classifiers were accuracy,analysis of the ROC curve and the parameter estimation were very satisfactory. Three comprehensive classification of esophageal cancer: the accuracy of RF classifier classification was superior to the other two classifiers. The integrated classifiers applied to classify Kazakh esophageal cancer can improve the classification ability and provide a certain basis judgment of imaging diagnosis. It lay the foundation for research and development of computer diagnostic system of esophageal cancer in Xinjiang Kazakh.