基于纹理特征的肝硬化MRI分期

Application of MRI in Cirrhosis Staging Based on Texture Features

借助MRI影像的计算机辅助诊断已逐渐被应用于肝硬化分期研究中。为了提高分期的准确率,本文基于MR8滤波器组的纹理特征提取对肝硬化MRI分期,构建了一个适用于肝硬化MRI的T1、T2、动脉期、门静脉期和平衡期五种序列图像的分期系统。MR8滤波器组具有响应维度低和旋转不变性等优点,因此,本文利用MR8纹理特征构建纹理基元统计直方图将肝硬化分为三个阶段,分别为正常、早期和中晚期。与经典的灰度共生矩阵(GLCM)的纹理特征提取方法相比,MR8滤波器组纹理特征对每一序列肝硬化的分期准确率均高于GLCM方法,其中T2和平衡期,三个阶段的ROI分期准确率均达到100%;在此基础上,将五个序列的分类结果进行融合,最终病例的分期准确率也达到100%。

Computer aided diagnosis by means of MRI（Magnetic Resonance Imaging） has been gradually used in staging of cirrhosis. In this paper, MRI is applied in staging of cirrhosis based on MR8 filter banks. A staging system that applies to five sequences of liver MRI, including T1-weighted, T2-weighted, arterial phase, portal venous phase and equilibrium phase, is structured to improve staging accuracy. A significant advantage of MR8 filter banks is the rotation invariance. Therefore, MR8 filter banks are used to extract texture features and to construct statistical histogram of texton. The statistical histograms are then used to separate the MRI sequences into three stages： normal, early stage and the middle and advanced stage. The experimental results show that the method employed in this paper has an excellent performance of accuracy compared with the classic GLCM（Gray Level Co-occurrence Matrix） method. Especially in T2-weighted sequence and equilibrium phase sequence, the ROI in three stages of both sequences appears a 100% accuracy. On this basis, the ROI results of the five sequences are integrated into the final case accuracy results, which is also 100%.