常规MRI纹理分析鉴别诊断眼眶淋巴瘤与炎性假瘤

目的 探讨常规MRI纹理分析在眼眶淋巴瘤和炎性假瘤鉴别诊断中的应用价值。方法 回顾性分析经病理或治疗随访证实的15例眼眶淋巴瘤及17例炎性假瘤患者的MRI资料。应用MaZda软件手工勾画ROI，并提取T1WI、脂肪抑制T2WI及脂肪抑制T1WI增强扫描图像中病变的纹理特征。通过Fisher系数、分类错误概率联合平均相关系数（POE+ACC）、交互信息（MI）及三者联合（FPM）的方法选择最佳纹理参数集合。使用线性判别分析（LDA）和非线性判别分析（NDA）进行纹理分类。比较最佳分类序列上两种病变的纹理特征差异。结果 T1WI及T2WI最佳纹理参数主要源于共生矩阵及游程矩阵，增强T1WI最佳纹理参数主要源于共生矩阵及直方图。T2WI纹理特征鉴别眼眶淋巴瘤及炎性假瘤能力最佳，其中FPM选择纹理特征联合NDA分类的误判率最低，为1.56%。眼眶淋巴瘤T2WI纹理特征参数中的能量及长游程补偿均高于炎性假瘤（P均<0.005），而熵及短游程补偿均低于炎性假瘤（P均<0.005）。结论 常规MR图像纹理分析可用于鉴别眼眶淋巴瘤和炎性假瘤。

MRI texture analysis in differential diagnosis of orbital lymphoma and inflammatory pseudotumor

Objective To discuss the application value of texture analysis of conventional MRI in differential diagnosis of orbital lymphoma from inflammatory pseudotumor. Methods The conventional MRI data of 15 patients with lymphoma and 17 patients with inflammatory pseudotumor proven by pathology or clinical follow-up were retrospectively reviewed. The texture features of lesions based on axial T1WI, fat-saturated T2WI and contrast enhanced fat-saturated T1WI were extracted by manually drawn ROIs with software MaZda. The subsets of optimized texture parameters were chosen by four different methods: Fisher coefficient, probability of classification error and average correlation coefficient (POE+ACC), mutual information measure (MI) and the combination of the above three methods (FPM), respectively. Linear discriminant analysis (LDA) and nonlinear discriminant analysis (NDA) were performed for texture classification. The texture features from the sequence with the best classification result of orbital lymphoma and inflammatory pseudotumor were compared. Results The optimal texture parameters were mainly derived from co-occurrence matrix and run-length matrix on T1WI and T2WI. The optimal texture parameters were mainly derived from co-occurrence matrix and histogram on contrast enhanced T1WI. The best classification of MRI texture was obtained within T2WI with lowest classification error of 1.56% achieved by FPM in combination with NDA. Comparing the texture parameters of orbital lymphoma and inflammatory pseudotumor on T2WI, the angular second moment and long length emphasis were significantly higher in orbital lymphoma (both P<0.005), while the entropy and short length emphasis were significantly lower in orbital lymphoma (both P<0.005). Conclusion It is feasible to use texture analysis on conventional MRI for the differentiation of orbital lymphoma and inflammatory pseudotumor.