多参数MRI影像组学对甲状腺乳头状癌颈部淋巴结的术前评估价值  

Diagnostic value of machine learning based on multi-parameters of MRI radiomics to predict cervical lymph node status of papillary thyroid carcinoma

作　　者：马伟琼[1] 陈康胤 杨宁 江桂华[2] 蓝博文[1] 曾裕镜[1] MA Weiqiong;CHEN Kangyin;YANG Ning;JIANG Guihua;LAN Bowen;ZENG Yujing(Department of Radiology,Huizhou Central People's Hospital,Huizhou 516000,China;Department of imaging,Guangdong Second Provincial General Hospital,Guangzhou 510000,China)

机构地区：[1]惠州市中心人民医院放射科,惠州516000 [2]广东省第二人民医院影像科,广州510000

出　　处：《磁共振成像》2022年第10期108-113,共6页Chinese Journal of Magnetic Resonance Imaging

基　　金：广东省医学科研基金立项项目(编号:B2019201);惠州市科技局基金立项项目(编号:2018Y025);惠州市中心人民医院2018年重点扶持项目(编号:〔2018〕118)。

摘　　要：目的评估基于多参数MRI影像组学特征的机器学习方法对甲状腺乳头状癌(papillary thyroid carcinoma,PTC)患者术前预测颈部淋巴结状态的价值。材料与方法回顾性分析182例经手术病理确诊淋巴结状态的PTC患者的影像和临床资料,分为91例颈部淋巴结转移组和91例非转移组。在多种MRI影像序列[轴位T1WI、T2WI、T2WI压脂、T1WI增强、T1WI压脂增强以及弥散加权成像(diffusion weighted imaging,DWI)]中勾画感兴趣区(region of interest,ROI)获得纹理特征和直方图特征。最后,使用多参数MRI纹理特征和直方图特征作为输入,构建了支持向量机(support vector machine,SVM)模型对颈部转移和非转移淋巴结进行分类。结果结合多种磁共振序列的多参数模型具有良好的分类性能,分类准确度高达79.61%,敏感度为75.00%,特异度为83.00%,曲线下面积(area under the curve,AUC)值为0.911。结论基于多参数MRI影像组学的机器学习方法可以有效地预测PTC患者术前的颈部淋巴结状态。Objective:To explore the diagnostic value of machine learning based on multi-parameters of MRI radiomics to predict cervical lymph node status in patients with papillary thyroid carcinoma(PTC).Materials and Methods:The imaging and clinical data of 182 patients with PTC diagnosed by operation and pathology were retrospectively analyzed,according to the result of surgical pathology,the patients were divided into lymph node metastasis group(91 cases)and no lymph node metastasis group(91 cases).Radiomics analysis was performed on multi-parameters of MRI[T1WI,T2WI,fat-saturated T2WI,enhanced T1WI,fat-saturated and enhanced T1WI,diffusion weighted imaging(DWI)]to acquire texture features and histogram features.Based on the above features,a support vector machine(SVM)model was constructed to classify cervical metastatic and non-metastatic lymph nodes.Results:The diagnostic performance of machine learning models which were based multi-parameters of MRI radiomics was superior,with a classification accuracy of 79.61%,a sensitivity of 75.00%,a specificity of 83.00%,and an area under curve(AUC)value of 0.911.Conclusions:The machine learning method based on multi-parameters of MRI radiomics can effectively predict cervical lymph node status in patients with PTC.

关 键 词：甲状腺癌 甲状腺乳头状癌 颈部淋巴结转移 多参数 磁共振成像 影像组学 直方图特征 支持向量机 

分 类 号：R445.2[医药卫生—影像医学与核医学] R736.1[医药卫生—诊断学]