摘要
目的探讨基于MRI纹理分析对眼眶神经纤维瘤(NF)与神经鞘瘤(SN)的鉴别诊断价值。方法回顾性分析经术后病理证实的11例NF患者和16例SN患者的资料。应用MaZda软件在MRI常规序列图像上勾画ROI并提取纹理特征,计算观察者间及观察者组内相关系数,应用LASSO算法进行特征降维,筛选出最佳特征参数,绘制ROC曲线下面积(AUC)评价其对鉴别NF与SN的诊断效能。结果在平扫T_(1)WI、T_(2)WI、DWI及T_(1)WI轴位增强4个序列分别提取的15个纹理分析特征中,对ICC均>0.75的特征进行LASSO选择,获得包括峰度、偏度、第99百分位数、游程不均匀性、灰度不均匀度、短游程增强在内的6个特征参数。其中T_(1)WI序列中,游程不均匀性和短游程增强的AUC分别为0.716、0.705,但游程不均匀性敏感度和特异度较高(87.50%、63.64%),具有较好的诊断价值;在T_(2)WI序列灰度不均匀度可以区分二者,AUC为0.676,最佳临界值为3670.80,敏感度为87.50%,特异度为45.45%;在DWI序列第99百分位数具有较高的AUC(0.764)和诊断特异度(100.00%);在增强T_(1)WI序列游程不均匀性鉴别诊断价值较大,AUC为0.742,最佳临界值为828.55,敏感度为86.67%,特异度为62.50%。结论基于常规MRI序列图像纹理分析可以帮助鉴别眼眶NF与SN。
Objective To evaluate the diagnostic performance of MRI texture analysis(TA)for differentiation of orbital Neurofibroma(NF)from Schwannoma(SN).Methods 11 patients with NF and 16 patients with SN confirmed by pathology were retrospectively analyzed.Mazda software was used in the conventional MR sequences outline extracting ROI and TA,Inter-and intra-class correlation coefficients(ICCs)were calculated from two different observers.The least absolute shrinkage and selection operator(LASSO)algorithm for characteristic dimensionality reduction were used to analyze the value of TA between the two groups.The ROC curve of the characteristic parameters with statistical significance was drawn,and the AUC was used as the main indicator to evaluate its effect of the diagnosis in distinguishing between NF and SN.Results Among the 15 characteristic parameters extracted from four sequences of plain T_(1)WI,T_(2)WI,DWI and T_(1)WI axial enhancement,all features with ICCs>0.75 were chosen to further select the LASSO algorithm,and 6 optimal features including Skewness,Kurtosis,Perc.99%,RLNonUnion,GLevNonLI and ShrtREmp demonstrated statistically significant differences.In T_(1)WI,the AUC of RLNonUnion and GLevNonLI were 0.716,0.705.However,RLNonUnion showed the highest sensitivity and the specificity(87.50%and 63.64%)with statistically significant difference(P<0.05).The diagnostic efficiency of GLevNonLI in T_(2)WI was the best,AUC,the optimal critical value,sensitivity and specificity scores were 0.676,3670.80,87.50%,45.45%,respectivety.The AUC of Perc.99%of the DWI group had the largest(AUC=0.764,specificity=100.00%,sensitivity=50.00%,the best critical value=33).On contrast enhanced T_(1)WI,the AUC of RLNonUnion was the largest of 0.742,the best critical value was 828.55,the sensitivity and the specificity scores were 86.67%,62.50%.showing better prediction efficacy.Conclusion TA based on conventional MR sequence images could contribute to discriminating NF from SN.
作者
张宝月
苏云杉
向述天
杨正林
ZHANG Baoyue;SU Yunshan;XIANG Shutian(Department of Radiology,the Affiliated Hospital of Yunnan University,Kunming,Yunnan Province 650021,P.R.China)
出处
《临床放射学杂志》
北大核心
2022年第7期1246-1251,共6页
Journal of Clinical Radiology
关键词
眼眶神经纤维瘤
眼眶神经鞘瘤
磁共振成像
纹理分析
Orbital neurofibroma
Orbital schwannoma
Magnetic resonance imaging
Texture analysis
