摘要
目的结合灰度共生矩阵和小波变换的纹理分析方法提取新疆哈萨克族高发病食管癌X射线钡剂造影图像的特征,旨在为放射科医生的诊断决策提供具有实际参考价值的辅助信息,提高食管癌诊断的准确率和效率。方法选取2种中晚期食管癌——蕈伞型和缩窄型,以及正常食管图像各100张,利用基于灰度共生矩阵的纹理特征提取方法分别提取食管癌X射线图像的角二阶矩、熵、惯性矩、逆差矩及相关性的方差作为纹理特征,同时使用小波变换对食管癌X射线图像进行二层小波分解,获取其高频子图,并提取高频子图的能量特征作为纹理特征。然后,使用C4.5决策树算法构造一个分类器,对正常食管和中晚期食管癌图像进行分类研究。结果共计提取11维特征,利用单一特征算法进行分类,灰度共生矩阵法分类准确率为64.66%,小波变换法分类准确率为77%。而综合的灰度共生矩阵和小波变换法的分类准确率为81.67%,更适用于正常食管和中晚期食管癌的分类。结论本研究将灰度共生矩阵、小波变换算法与决策树C4.5相结合,对正常食管与蕈伞型和缩窄型食管癌进行特征提取及分析,结果表明本算法分类准确率较高,为开发食管癌的计算机辅助诊断系统奠定了基础。
Objective In this paper,combining gray level co-occurrence matrix （GLCM） with wavelet transform of texture arithmetic method extracted X-ray image texture of esophageal cancer with high morbidity in Xinjiang.This method provided auxiliary information with reference value for radiologist and enhanced accuracy rate and efficiency of esophageal cancer.Methods Two types of middle and terminal esophageal cancer：fungating type esophageal cancer and constrictive esophageal cancer were selected in the experiment,and 100 images of the two types of esophageal cancers and normal esophagus were selected,respectively.We used GLCM texture feature method to extract angular second moment,entropy,contrast,correlation and inverse difference moment of variance respectively for X-ray images of esophageal cancer.Meanwhile,we employed wavelet transform to process two-dimensional discrete wavelet decomposition at second level,obtained its highfrequency content images,and extracted the energy features of high-frequency content images as texture features.The C4.5 decision tree was employed as a classifier.Results Eleven features were extracted by GLCM and wavelet transform methods.The experimental results showed that using single feature classification,the accuracy rate of GLCM classification and wavelet transform classification reached to 64.66% and 77%,respectively.The accuracy rate of the comprehensive of GLCM and wavelet transform method was 81.67%,more suitable for the classification of normal esophagus and advanced esophageal cancer.Conclusions This method achieved high classification performance and lay the foundation of the computer-aided diagnosis system of Kazakh esophageal cancer in Xinjiang.
出处
《北京生物医学工程》
2017年第1期37-44,共8页
Beijing Biomedical Engineering
基金
国家自然科学基金(81460281
81560294)
江西民族传统药协同创新项目(JXXT201401001-2)资助
关键词
食管癌
灰度共生矩阵
小波变换
特征提取
图像分类
esophageal cancer
gray level co-occurrence matrix
wavelet transformation
feature extraction
image classification
