基于医学图像纹理特征量优化的肝纤维化分级

Staging of Hepatic Fibrosis Based on Optimization of Selected Texture Features

摘要: 目前基于肝纤维化计算机辅助诊断的纹理特征量的选取和图像特征量的优化方案尚缺指导性结论。本文针对大量采集的MR和CT图像数据进行肝纤维化纹理特征量的提取，采用SVM分类器进行留一算法检测，通过对各项准确率的分类统计，得出了计算机辅助肝纤维化诊断的优化方案。结果显示肝脏MRI和CT图像的感兴趣区域的大小为20 × 20像素时获得的效果最优；在特征量个数为3至7个的组合时获得的分类效果最好；统计出MRI图像和CT图像的各个特征量在计算机辅助诊断肝纤维化程度实验中的权重值；发现MRI比CT能够更好地反映肝脏纤维化程度，而且MRI图像的有效特征量分布更为集中。

Abstract: By now, it is still lack of guidance conclusion on texture features selection and image modalities optimization for the computer-aided diagnosis of hepatic fibrosis. Based on the extraction of texture features from large amount of MRI and CT images, lack of research exists on computer-aided diagnosis of hepatic fibrosis on medical images, this paper is to study the optimization of computer-aided diagnosis program of hepatic fibrosis, we can not only conclude options on Regions of Interest of hepatic MRI and CT images, but also get optimal selection of medical image features using SVM algorithm classifier; we find that the optimal experiment performance is equilibrium phase and 20 × 20 pixels size of ROI; the optimal number of features is confirmed from 3 to 7. The weighted values of 15 features for computer aided diagnosis of hepatic fibrosis in hepatic MRI and CT images are calculated and ranked; the hepatic MRI images reflect the degree of liver fibrosis is better than CT images; the distribution of effective features in liver MRI images is more concentrated than that in liver CT images.