新疆医科大学学报

目的探讨MRI纹理分析在眼眶良恶性肿瘤中鉴别诊断的价值。方法回顾性分析2016年5月-2018年10月经桓台县人民医院术后病理或随访证实的76例眼眶肿瘤患者,其中良性组45例,恶性组31例。所有患者均行常规序列和MR增强检查并提取扫描图像中病变的纹理特征,选择最具有鉴别瘤价值的参数进行眼眶良恶性肿瘤特征分类,统计误判率并与医师读片结果进行对比。结果鉴别眼眶良性与恶性肿瘤时,来自压脂T2WI图像纹理特征鉴别能力最佳。在纹理特征参数选择方法中,FPM选择的纹理特征参数鉴别眼眶良恶性肿瘤误判率最低,为2.6%～18.4%;其次是交互信息(MI),为3.9～28.9%;Fisher系数和分类错误概率联合平均相关系数(POE+ACC)的误判率接近,均为5.3%～34.2%。纹理特征分类分析方法中,PCA区分眼眶良恶性肿瘤的误判率(3.9%～15.8%)明显低于RDA(6.6%～34.2%)、NDA(2.6%～34.2%)和LDA(5.3%～26.3%)。影像医师的误判率为23.7%(18/76),高于采用纹理分析,差异有统计学意义(P均<0.05)。结论 MRI图像纹理分析可用于鉴别眼眶良恶性肿瘤;不同序列图像、纹理特征提取方法和分类方法在鉴别眼眶良恶性肿瘤中的诊断效能是不同的,推荐使用压脂T2WI图像的FPM联合NDA方法。

Objective To evaluate the value of MRI texture analysis in the differential diagnosis of orbital benign and malignant tumors. Methods A total of 76 cases of orbital tumors confirmed by postoperative pathology or follow-up in the hospital were analyzed retrospectively, including 45 cases of benign tumors and 31 cases of malignant tumors. All patients were undergone routine sequence and MR enhancement examination and extracted the texture features of the lesions in the scan images. The most valuable parameters were selected for the classification of the orbital benign and malignant tumors. The misjudgment rate was calculated and compared with the results of film reading by doctors. Results When differentiating the orbital benign and malignant tumors, the texture features of T2 WI images from pressed lipids have the best ability to distinguish the benign and malignant orbital tumors. In the texture feature parameter selection methods, the misjudgment rate of FPM selected the texture feature parameters for the differentiation of the orbital benign and malignant tumors was the lowest(2.6%～18.4%), followed by the interactive information(MI)(3.9%～28.9%). The misjudgment rate of Fisher coefficient and joint average correlation coefficient(POE+ACC) of classification error probability were close, both were 5.3%～34.2%. In the method of texture feature classification and analysis, the misjudgment rate of PCA in distinguishing between the benign and malignant orbital tumors(3.9%～15.8%) was significantly lower than that of RDA(6.6%～34.2%), NDA(2.6%～34.2%) and LDA(5.3%～26.3%). The misjudgment rate of the imaging doctors was 23.7%(18/76), which was higher than that of the texture analysis. The differences were statistically significant(P<0.05). Conclusion MRI image texture analysis can be used to distinguish the benign and malignant orbital tumors; different sequence images, texture feature extraction methods, and classification methods are different for the diagnosis of orbital benign and malignant tumors. It is recommended to use the FPM combined with NDA method of fat-suppressed T2 WI images.