摘要
目的:讨论Bagging、Adaboost、Random Forest(RF) 3个集成分类器对新疆哈萨克族食管图像分型中的分类能力。方法:使用Matlab图像处理软件,对食管X线图像进行预处理,对预处理后的图像使用灰度共生矩阵和Hu不变矩特征进行图像特征的提取;然后,使用主成分分析法对特征值进行筛选优化,得到分类能力较强的特征值;最后,使用Weka软件,将3个不同的集成分类器对正常食管和早期食管癌图像进行分类,并进行分类模型的评估。结果:使用Bagging、Adaboost、Random Forest(RF) 3个集成分类器结合降维后的灰度共生矩阵特征值对食管图像进行分类时,正常食管的分类准确率是82%、94%、88%,早期食管癌的分类准确率是94%、88%和94%;使用降维后的Hu不变矩特征值和3种集成分类器对正常食管和早期食管癌进行分类时,正常食管的分类准确率是60%和64%、61%,早期食管癌的分类准确率是57%、68%和65%;结论:3种集成分类器结合灰度共生矩阵对正常食管和早期食管癌X线图像进行分类,其分类准确率与Hu不变矩相比分类效果更显著。说明灰度共生矩阵结合3种集成分类器更适合用于区分正常食管和早期食管癌X线图像。
Objective: The application of three kinds of integrated classifiers for Bagging,Adaboost and Random Forest( RF) in xinjiang kazakh esophageal image. Methods: Using Matlab image processing software,the X-ray esophageal image preprocessing,feature extraction of images after preprocessing using gray level co-occurrence matrix and Hu invariant moments feature; then,analysis of the two kinds offeature extraction algorithm to extract the characteristic values were optimized using principal component,classification ability of the feature value; finally,the use of Weka software,the three different integrated classifier combination feature selection respectively in normal esophagus and early esophageal cancer image classification,and use the parameter values to evaluate the classification model. Results: in normal esophageal and early esophageal X-ray image classification,Bagging,Adaboost,the Random Forest( RF) three integrated classifier using three integrated classifier using gray level co-occurrence matrix classification accuracy rate were 82% and 94%,respectively,82% and 94%,88% and 94%; Using the Hu invariant characteristics,the classification accuracy rate was 60% and 64%,61% and 57%,68%and 65%. Conclusion: three integrated classifier using gray level co-occurrence matrix of normal esophageal and early esophageal X-ray image classification accuracy is higher,than the Hu invariant moment features classification effect is more prominent. The gray-scale symbiosis matrix is applied to the characteristics of normal esophageal and early esophageal cancer X-ray images.
引文
[1]左婷婷,郑荣寿,曾红梅,等.中国食管癌发病状况与趋势分析[J].中华肿瘤杂志,2016,38(9):703-708.
[2]赵洪斌,刘丹.CAD技术在乳腺癌诊断中的应用研究[J].中国民康医学,2011,23(6):769-770.
[3]王伟胜,骆嘉伟,林红利.医学图像计算机辅助诊断数据平台研究[J]. Chinese Journal of Biomedical Engineering,2013,32(1):105-108.
[4]杨韬,夏建国,缪锦林,等.基于灰度共生矩阵的乳腺钼靶图像分析[J].中国医学影像学杂志,2012,20(4):306-309.
[5]褚晶辉,李英敏,宋垣,等.一种医学图像去噪程序的并行优化[J].中国医学物理学杂志,2012,29(6):3781-3783.
[6] Wang X,Zhang W,Ji Q. Image object extraction with shape and edge-driven Markov random field model[J].Institution of Engineering and Technology,2014,8(7):383-396.
[7]闫晶莹,王晓军.不变矩在图像特征提取及目标识别中的应用[J].机械工程与自动化,2011,1(2):52-54.
[8] Hu M K.Visual pattern recognition by moment invariants[J].JRE Tram on Info Theory,1962,(8):179-187.
[9] Dragisa Z,Jovisa Z. Shape ellipticity based on the first Hu moment invariant[J]. Information processing letters,2013,113(19):807-810
[10]梁曼,黄富荣,何学佳,等.荧光光谱成像技术结合聚类分析及主成分分析的藻类鉴别研究[J].光谱学与光谱分析,2014,34(8):2132-2136.
[11]黎远鹏,黄富荣,董佳,等.荧光光谱成像技术结合主成分分析与Fisher判别快速鉴别肉苁蓉[J].光谱学与光谱分析,2015,35(3):689-694.
[12]欧阳震诤.不平稳数据流的分类技术研究[D].湖南省:国防科学技术大学,2009.