基于磁共振成像图像的影像组学特征鉴别诊断眼眶淋巴瘤与炎性假瘤  

MRI-based radiomic features in differentiation of orbital lymphoma and inflammatory pseudotumor

作　　者：杨立娟[1] 谢小洋 王俊明[1] 蔚道清 王栋[1] 齐乃新[3] YANG Lijuan;XIE Xiaoyang;WANG Junming;WEI Daoqing;WANG Dong;QI Naixin(Department of Radiology,Xi'an Fourth Hospital,Xi'an 710004,China;School of Information Science and Techology,Northwest University,Xi'an 710069,China;Department of Radiology,the Second Affiliated Hospital of Xi'an Jiaotong University,Xi'an 710004,China)

机构地区：[1]西安市第四医院放射影像科,陕西西安710004 [2]西北大学信息科学与技术学院,陕西西安710069 [3]西安交通大学第二附属医院影像科,陕西西安710004

出　　处：《实用放射学杂志》2020年第12期1914-1917,1932,共5页Journal of Practical Radiology

基　　金：国家自然科学基金面上项目(11571012)。

摘　　要：目的探讨基于MRI图像的影像组学特征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证实的27例眼眶淋巴瘤及27例炎性假瘤患者的MRI资料.随机分为训练集和验证集,应用MITK软件手工勾画ROI进行图像分割,提取7个直方图特征和20个尺度不变特征变换(SIFT)特征,结合这2种特征,使用逻辑回归实现眼眶淋巴瘤和炎性假瘤的鉴别分类,以ROC曲线的AUC来评价鉴别能力.预测结果与放射科诊断医生分类结果进行比较.结果结合直方图特征和SIFT特征,最终得到训练集中AUC为0.9838,准确度为94.38%;验证集中AUC为0.7092,准确度为73.22%;放射诊断医师评估的准确性分别为72.22%(高年资放射诊断医生)和55.56%(低年资放射诊断医生).结论基于MRI图像的影像组学特征可用于鉴别诊断眼眶淋巴瘤和炎性假瘤,诊断效能与高年资放射诊断医师相当,明显高于低年资放射诊断医师.Objective To discuss the application value of the MRI based radiomic features in differentiating orbital lymphoma and inflammatory pseudotumor.Methods 27 patients with orbital lymphoma and 27 patients with inflammatory pseudotumor confirmed by pathology were enrolled retrospectively.MRI data of these subjects were acquired and divided into training dataset and validation dataset randomly.The ROI was manually delineated via MITK software,and then 7 histogram features and 20 scale invariant feature transform(SIFT)features were extracted from the ROI.After combining these features,Logistic regression was used to differentiate orbital lymphoma from inflammatory pseudotumor.Discriminating ability was assessed by the AUC of ROC curve.The predicted results were compared with the results from radiologists.Results The com bination of histogram features and SIFT features finally achieved an AUC of 0.9838 and accuracy of 94.38%in the training dataset,and achieved an AUC of 0.7092 and accuracy of 73.22%in the validation dataset.The accuracy of an experienced radiologist and a radiology resident were 72.22%and 55.56%respectively.Conclusion MRI based radiomic features may be useful for differentiating orbital lymphoma from inflammatory pseudotumor,and the diagnostic performance is comparable to the experienced radiologist,but is significantly higher than the radiology resident.

关 键 词：眼眶淋巴瘤 炎性假瘤 磁共振成像 影像组学 

分 类 号：R739.7[医药卫生—肿瘤] R445.2[医药卫生—临床医学]