肝细胞癌MR图像的纹理分类研究

Texture Classification Study of MR Images for Hepatocellular Carcinoma

结合小波多分辨率分析方法与统计分析方法提出了一种复合纹理分类模型,评估其基于核磁共振(MR)图像辅助诊断肝细胞癌(HCC)与正常肝脏组织的价值。首先,训练样本按类别分成两组,在每组中执行小波系数统计;其次,对新样本的小波系数基于两组统计结果执行两次离散化,以直方图、共生矩阵、游程长度矩阵等方法提取到两组特征;最后,基于两组特征执行两次分类以计算新样本的类别属性概率并决策。实验结果显示,该模型获得了比传统方法更好的分类性能,表明采用该模型对基于MR图像进行HCC与正常肝脏组织的计算机辅助诊断是有益的。

Combining wavelet multi-resolution analysis method and statistical analysis method, a composite texture classification model is proposed to evaluate its value in computer-aided diagnosis of hepatocellular carcinoma (HCC) and normal liver tissue based on magnetic resonance (MR) images. First, training samples are divided into two groups by two categories, statistics of wavelet coefficients are calculated in each group. Second, two discretizations are performed on wavelet coefficients of a new sample based on the two sets of statistical results, and two groups of features can be extracted by histogram, co-occurrence matrix, and run-length matrix, etc. Finally, classification is performed twice based on the two groups of features to calculate the category attribute probabilities, then a decision is conducted. The experimental results demonstrate that the proposed model can obtain better classification performance than routine methods, it is rewarding for the computer-aided diagnosis of HCC and normal liver tissue based on MR images.