基于STIR-MRI的影像组学模型评估原发性下肢淋巴水肿的临床分期  

A preliminary study on the value of STIR-MRI-based radiomics to identify the clinical staging of primary lower extremity lymphedema

作　　者：刘梦珂 张怡梦 张妍 李兴鹏 郝琪 张晓杰 徐井旭 黄陈翠 王仁贵[1] LIU-Mengke;ZHANG-Yimeng;ZHANG-Yan(Department of Radiology,Beijing Shijitan Hospital,Capital Medical University,Beijing 100038,China)

机构地区：[1]首都医科大学附属北京世纪坛医院放射科,北京100038 [2]北京大学第九临床医学院放射科,北京100038 [3]北京深睿博联科技有限责任公司,北京100089

出　　处：《放射学实践》2024年第4期517-522,共6页Radiologic Practice

基　　金：国家自然科学基金资助项目(61876216)。

摘　　要：目的:探讨MRI影像组学模型对原发性下肢淋巴水肿(PLEL)临床分期的评估价值。方法:回顾性搜集2018年1月-2019年12月在本院就诊的132例单侧PLEL患者的临床和影像资料。依据2020年国际淋巴协会(ISL)淋巴水肿临床分期标准,Ⅰ期45例,Ⅱ期51例,Ⅲ期36例。将STIR序列原始图像导入深睿多模态科研平台,连续选择10层肢体肿胀最严重的横轴面图像,逐层在皮下软组织区域内的水肿区手动勾画ROI,提取影像组学特征1743个。采用相关性分析及F-Test算法进行特征的筛选,然后采用logistic回归分类器分别建立Ⅰ期与Ⅱ期、Ⅱ期与Ⅲ期的分类模型,并采用ROC曲线评价模型的鉴别效能。采用决策曲线分析(DCA)评估模型的临床实用性。结果:鉴别Ⅰ期与Ⅱ期,共筛选出10个组学特征构建预测模型,其在训练集中的AUC为0.935(95%CI:0.886~0.983),在验证集中的AUC为0.917(95%CI:0.858~0.975)。鉴别Ⅱ期与Ⅲ期,共筛选出10个组学特征用于建模,其在训练集中的AUC为0.838(95%CI:0.749~0.927),在验证集中为0.760(95%CI:0.654~0.866)。DCA验证了影像组学模型鉴别PLEL临床分期的临床实用性。结论:基于下肢MRI建立的影像组学模型可较准确地评估PLEL的严重程度,与PLEL的临床分期一致性高。Objective:To explore the evaluation value of MRI radiomics models for clinical sta-ging of primary lower extremity lymphedema(PLEL).Methods:A total of 132 patients with unilateral PLEL were retrospectively collected in our hospital from January 2018 to December 2019,and were classified into stageⅠ,ⅡorⅢaccording to the clinical staging criteria formulated by International Society of Lymphology(ISL)for lymphoedema in 2020.The original MR images of STIR sequence were imported into the Deepray multimodal research platform,10 continuous slices of cross-sectional images on which the most severe limb swelling were found were selected,and on each slice of image the region of interest(ROI)were manually drawn at edema area in subcutaneous soft tissue,and then 1743 radiomics features were extracted.The features were filtered by correlation analysis and F-Test algorithm,and logistic regression classifier was used to build the classification models for stageⅠandⅡ,and stageⅡandⅢ,respectively.Decision curve analysis(DCA)was used to assess the clinical utility of the models.Results:For differentiating stageⅠandⅡ,10 radiomics features were selected out and were included for model construction,and the AUC of the model in training set was 0.935(95%CI:0.886~0.983)and the valid cohort AUC was 0.917(95%CI:0.858~0.975);for stageⅡvsⅢ,a total of 10 features were screened for model construction:the training cohort AUC was 0.838(95%CI:0.749~0.927)and the valid cohort AUC was 0.760(95%CI:0.654~0.866).The DCA confirmed the clinical utility of the imaging-radiomics model to identify the clinical stage of PLEL.Conclusion:The radiomics model based on lower limb MRI can accurately evaluate the severity of PLEL and has high consistency with the clinical staging of PLEL.

关 键 词：影像组学 磁共振成像 原发性淋巴水肿 下肢 临床分期 

分 类 号：R445.42[医药卫生—影像医学与核医学] R658.3[医药卫生—诊断学] R551.2[医药卫生—临床医学]