使用MRI磁敏感加权相位图的影像纹理分析技术对预测肝脏纤维化程度的研究

Liver fibrosis is regarded as the intermediate stage from chronic liver disease developing to liver cirrhosis and even hepatocellular carcinoma. Accurate fibrosis staging can be beneficial in delaying the onset of liver cirrhosis. MR, although being a non-invasive imaging modality that provides objective information, still has limitation in recognition of hepatic fibrosis and mild liver cirrhosis. Image texture is a set of metrics calculated in image processing designed to quantify the perceived texture of an image. Image Texture gives us information about the spatial arrangement of color or intensities in an image or selected region of an image. Previous study has shown that susceptibility weighted (SWI) phase image, compared to other conventional MR images, is more sensitive to the reticulated and nodular change of the liver in cirrhosis patients. Texture analysis helps to detect subtle texture difference between images by using a variety of quantifiable texture features. Therefore, this study aims to use image texture analysis technique based on SWI phase image, the sensitive imaging method, to extract texture features of liver, to compare the texture features between difference fibrosis stage groups, and to find the optimal subset of texture features which could accurately predict the fibrosis stage, and finally, to evaluates the accuracy of this texture analysis method. The success of this study may provide a novel alternative method in diagnosing liver fibrosis, which is of significant clinical importance.

肝脏纤维化是慢性肝病演变为肝硬化并向肝癌发展的中间阶段，准确识别不同程度纤维化有助于延缓肝硬化的发生。MR作为无创客观的成像方法对肝纤维化甚至肝硬化的准确识别仍存在一定局限性。纹理是图像的某种局部性质，是对局部区域中像素之间关系的一种度量。前期研究发现，肝脏磁敏感加权成像（Susceptibility-weighted imaging, SWI）相位图能较常规的MR成像序列更敏感地显示肝硬化者网格结节状的肝脏纹理改变，而图像纹理分析技术则可以将细微的纹理差异通过不同纹理特征量化。因此本研究拟利用SWI相位图这一敏感的成像手段为基础，结合影像纹理分析技术，提取肝脏的纹理特征，比较不同程度肝纤维化者肝脏纹理特征的差异，找到准确预测纤维化程度的最佳纹理特征集，并评估该最佳特征集预测纤维化程度的准确性。本项研究的成功有助于为肝纤维化的诊断提供一种全新且无创的方式，对肝纤维化临床干预具有深远的意义。

本研究拟利用SWI相位图，结合影像纹理分析技术，提取纹理特征，比较不同程度纤维化纹理特征的差异，找到准确预测的最佳纹理特征集，并评估该方法的准确性。研究最终纳入了 45例肝硬化患者和22例志愿者，在3T磁共振下使用2D屏气磁敏感加权成像。利用MaZda纹理分析软件，由两名医师独立提取了相位图上的纹理特征，包括直方图（Histogram）、梯度（Gradient）、共生矩阵（Co-occurrencematrix）、游程矩阵（Run-length matrix）、自回归模型（autoregressive model）。使用大数据分析软件Weka及其CfsSubsetEval评估和BestFirst搜索方法其对获得的纹理特征进行特征筛选，分析优化子集在两组之间分布的差异。进一步地将筛选所得优化子集进一步使用决策树模型（J48）、贝叶斯网络、支持向量机SVM、Logistic回归等方法进行训练，并评估各种分类方法的准确性。最后分析纹理特征在不同观测者下的一致性。经过分析，获得了5个最优化纹理特征MinNorm, Variance, AngScMom, Correlat, RLNonUni，这些特征在两组之间具有显著差异，单因素分析发现其区分准确性的ROC曲线下面积分别为0.89、0.88、0.67、0.77和0.84。观测者一致性评分在0.85至0.99。J48决策树、贝叶斯网络、SVM、Logistic回归分类的准确性分别为80.70%、85.57%、85.57%、76.12%。从而证明基于SWI相位图的纹理分析可以有效区分肝硬化与正常肝组织。

内容获取失败，请点击重试