Abstract: The performance of describing ultrasonic liver cancer image texture feature with fractal and lacunarity and their combined factors is studied. The fractal dimension and lacunarity values are estimated with 4 fractal dimension and a lacunarity methods based on the samples of 14 ultrasonic images each for normal liver and liver cancer. ROC（Receiver Operating Characteristic） analysis shows that the single factors for FPS and LBCM hold higher AUCs（area under ROC curve）. The train and test results for the single factors and the combined factors with SVM（Support Vector Machine）exhibit that FPS（Fourier Power Spectrum）+LBCM（Lacunarity of Box Column Mean）（4 kernals） and DBC（Differential Box Counting）+LBCM（except SIGMOID） get higher classification accuracy rate than the single factors.

Key words: ultrasonic liver cancer image, texture analysis, fractal dimension, lacunarity, synergy, Support Vector Machine（SVM）

摘要： 对分维和孔隙度及其组合因子表征超声肝癌图像纹理特征的性能进行了对比研究。以正常肝和肝癌各14幅超声图像为样本，用4种分维和一种孔隙度方法计算分维和孔隙度值。用ROC进行评估，单因子的傅里叶功率谱分维和盒柱平均值孔隙度值有较大的ROC曲线下面积。用SVM对单因子和组合因子进行训练和检验表明，傅里叶功率谱分维与盒柱平均值孔隙度（FPS+LBCM）（4个核）和差分盒计数分维与盒柱平均值孔隙度（DBC+LBCM）（除SIGMOID外）构成的组合因子有比单因子较高的分类准确率。

关键词: 肝癌超声图像, 纹理分析, 分维, 孔隙度, 协同作用, 支持向量机