3.0 T MRI T2 mapping纹理特征在膝关节骨性关节炎软骨损伤分级中的价值

The value of T2 mapping texture features of 3.0 T MRI in grading cartilage injury of knee osteoarthritis

目的探究基于3.0 T MRI T2 mapping的纹理特征在膝关节骨性关节炎(keen osteoarthritis,KOA)患者不同程度软骨损伤分级中的诊断性能。材料与方法回顾性分析实验组骨性关节炎患者72个膝关节及对照组健康志愿者22个膝关节。通过矢状位T2 mapping生成T2伪彩图,在T2伪彩图中画取ROI并标记国际软骨修复学会(International Cartilage Repair Society,ICRS)分级,选取ICRS MRI分级与关节镜分级一致的201个关节面图像,采用OK软件提取、分析纹理参数。按7∶3的比例随机选取143个关节面图像作为训练集,剩余58个关节面图像作为验证集。对训练集的参数用Spearman及sbf(select by filter)进行特征过滤,用随机森林函数进行特征选择,用ctree建立模型,给出特征在鉴别正常软骨及不同软骨损伤分级中的权重。用曲线下面积(area under the curve,AUC),敏感度、特异度,准确度来评价模型预测正常软骨及不同软骨损伤分级的性能。结果MinLocation、MaxSize及Maximun3DDiameter权重均一致较大,其中MinLocation在各损伤分级中权重均最大,超过0.75。集训集中正常软骨的AUC值为0.91,Ⅰ级损伤的AUC值为0.82,Ⅱ级损伤的AUC值为0.84,Ⅲ级损伤的AUC值为0.88;验证集中正常软骨的AUC值为0.87,Ⅰ级损伤的AUC值为0.74,Ⅱ级损伤的AUC值为0.84,Ⅲ级损伤的AUC值为0.96。AUC最高的是验证集中Ⅲ级损伤软骨,为0.96;其次是训练集中正常软骨,为0.91。无论在训练集还是验证集中都表现出了良好的预测价值。敏感度最高的是训练集中Ⅰ级损伤软骨,为0.83;特异度最高的是训练集中Ⅲ级损伤软骨,为0.98。结论通过T2 mapping提取的纹理参数在不同软骨损伤程度中有较好的鉴别能力。

Objective:To explore the diagnostic performance of texture analysis based on 3.0 T MRI T2 mapping in different levels of cartilage injury in knee osteoarthritis patients.Materials and Methods:Retrospective analysis of experimental group keen osteoarthritis patients(KOA)72 knee joints and control group healthy volunteers(H)22 knee joints.Through the sagittal T2 mapping T2 pseudocolor,to draw ROI and mark ICRS grading in a T2 artifact,a total of 201 articular surface images were selected consistent with MRI ICRS grading and arthroscopic grading,OK software was used to extract and analyze texture parameters.According to the ratio of 7∶3,143 articular surface images were randomly selected as the training set,and the remaining 58 articular surface images were used as the verification set.The parameters of the training set were filtered by Spearman and sbf(select by filter),the characteristics were selected by random forest function,and the model was established by ctree to give the weight of the characteristics in the identification of normal cartilage and different cartilage injury grades.AUC,sensitivity,specificity and accuracy were used to evaluate the performance of the model in predicting normal cartilage and different cartilage injury grades.Results:MinLocation,MaxSize and Maximun3 DDiameter weights are consistent,among them MinLocation was the largest weight of each damage grade,over 0.75.AUC value of normal cartilage in training set was 0.91,gradeⅠdamage AUC 0.82,gradeⅡdamage AUC 0.84,gradeⅢdamage AUC 0.88;verified that the AUC value of concentrated normal cartilage was 0.87,gradeⅠdamage AUC 0.74,gradeⅡdamage AUC 0.84,gradeⅢdamage AUC 0.96.AUC highest was the validation of the gradeⅢdamage to cartilage,0.96.The second was the training of normal cartilage,0.91.Both in the training set and the verification set show good predictive value.The most sensitive is the injury of cartilage in training setⅠ,0.83 percent.The highest specificity was the injury of cartilage in training setⅢ,0.98.Conclusions:Texture parameters extracted by T2 mapping have better ability to distinguish different cartilage damage.