介绍和评估了描述肝癌超声图像纹理特征的孔隙度方法.用14幅正常肝和14幅原发性肝癌图像为样本,将用5种方法得到的正常肝和原发性肝癌图像的孔隙度值进行正态分布检验,5种方法孔隙度值基本都呈正态分布.只有用立方盒质量和盒柱平均值法得到的正常肝和原发性肝癌的孔隙度平均值通过了差异显著性Student-t检验,用这两种方法得到的正常肝孔隙度值有较小的平均值和标准差,原发性肝癌图像孔隙度值有较大的平均值和标准差.对用5种方法得到的孔隙度进行ROC分析结果表明盒柱平均值法得到了最大的ROC曲线下的面积为0.959 2.用10折交叉验证和不同核函数的SVM进行分类,立方盒质量和盒柱平均值法分别得到了96.428 6％和92.857 1％的分类正确率.实验结果表明所提出的盒柱平均值法具有较强的描述肝癌超声图像纹理特征的能力.%This paper intriduced and evaluated lacunarity method for texture characterization of normal ultrasonic liver parenchyma and primary liver cancer image. Five methods of lacunarity estimation for the texture characterization of normal liver and primary liver cancer image were tested and evaluated based on 14 normal liver and 14 primary liver cancer images, and only cube-box-mass and box-column-mean passed student-t significance deviation test. Experimental result Analysis showed that lacunarity mean and standard deuiation of images of normal liver parenchyma were statistically significantly lower than that of primary liver cancer for cube-box-mass and box-column mean methods. ROC analysis showed that box-column-mean method obtained the highest area (0.959 2) under ROC curve. Through 10-fold cross validation and different kernel function SVM analysis, cube-box-mass and box-column-mean method achieved classification accuracy of 96. 428 6% and 92.857 1% , respectively. The experimental results indicated that box-column mean method is the most appropriate method for characterizing the texture feature of ultrasonic liver image among the five methods.
展开▼
