CT纹理分析鉴别甲状腺良恶性结节和预测淋巴结转移  

CT texture analysis in differentiating benign and malignant thyroid nodules and predicting lymph node metastasis

作　　者：陈镜键 方正[1] 钟维佳[1] 邹欣芯 游涛 CHEN Jingjian;FANG Zheng;ZHONG Weijia;ZOU Xinxin;YOU Tao(Department of Radiology,the Second Affiliated Hospital of Chongqing Medical University,Chongqing 400010,China)

机构地区：[1]重庆医科大学附属第二医院放射科,重庆400010

出　　处：《中国医学影像技术》2021年第1期35-39,共5页Chinese Journal of Medical Imaging Technology

摘　　要：目的探讨CT纹理分析鉴别甲状腺结节良恶性和预测恶性结节淋巴结转移的价值。方法回顾性分析174例经手术病理证实的甲状腺结节患者,包括122例良性病变(良性组)及52例恶性病变(恶性组);根据是否淋巴结转移将恶性组患者分为转移亚组(n=22)与无转移亚组(n=30)。采用Mazda软件于动脉期CT图像提取纹理特征,并以Fisher方法对纹理特征进行降维,分别获得良性组与恶性组、转移亚组与无转移亚组的最优纹理特征。比较良性组与恶性组、转移亚组与无转移亚组间最优纹理特征差异;针对差异有统计学意义的纹理参数采用二元Logistic回归分析影响甲状腺结节良恶性的独立预测因子,以ROC曲线法分析独立预测因子鉴别甲状腺结节良恶性的效能。结果共提取279个纹理特征,经降维后获得良性组与恶性组各10个最优纹理特征参数,除参数S(5,-5)InvDfMom外,其余参数组间差异均有统计学意义(P均<0.05)。参数Vertl_GlevNonU、45dgr_GlevNonU是甲状腺良恶性结节的独立预测因子(P均<0.05)。Vertl_GlevNonU、45dgr_GlevNonU鉴别甲状腺良恶性结节的最佳阈值分别为21.11和33.61,前者的AUC、敏感度和特异度分别为0.76、73.80%及73.10%,后者分别为0.77、68.90%及75.00%。恶性组内转移亚组与无转移亚组间最优纹理特征差异均无统计学意义(P均>0.05)。结论基于CT甲状腺结节纹理特征分析对鉴别甲状腺良恶性结节具有一定价值,而对预测甲状腺恶性结节淋巴结转移价值有限。Objective To investigate the value of CT texture analysis for differentiating benign and malignant thyroid nodules and predicting lymph node metastasis of malignant nodules.Methods Data of 174 patients with thyroid nodules confirmed by surgical pathology were retrospectively analyzed,including 122 benign lesions(benign group)and 52 malignant lesions(malignant group).Patients in malignant group were divided into metastatic subgroup(n=22)and non-metastatic subgroup(n=30)according to lymph nodes metastasis.Mazda software was used to extract texture features from arterial phase CT images,and Fisher method was used to reduce the dimensionality of texture features to obtain the optimal texture features between benign group and malignant group,also metastatic subgroup and non-metastatic subgroup in malignant group,respectively.The optimal texture features were compared between benign group and malignant group,also metastatic subgroup and non-metastatic subgroup.For texture features being statistical different between groups,binary Logistic regression analysis was used to analyze the independent predictors of benign and malignant thyroid nodules,and ROC curve method was used to analyze the diagnostic efficacy of independent predictors.Results A total of 279 texture features were extracted,and 10 optimal texture features were obtained between benign group and malignant group after dimension-reduction.Except for parameter S(5,-5)InvDfMom,statistical significant differences of 9 parameters were found between groups(all P<0.05).The parameter Vertl_GlevNonU and 45dgr_GlevNonU were independent predictors of benign and malignant thyroid nodules(both P<0.05).The optimal cutoff value of parameter Vertl_GlevNonU and 45dgr_GlevNonU to identify benign and malignant thyroid nodules was 21.11 and 33.61,and the AUC,sensitivity and specificity of the former was 0.76,73.80%and 73.10%,of the latter was 0.77,68.90%and 75.00%,respectively.No significant difference of texture features was detected between metastatic subgroup and non-metastat

关 键 词：甲状腺结节 诊断 体层摄影术 X线计算机 纹理分析 

分 类 号：R736.1[医药卫生—肿瘤] R814.42[医药卫生—临床医学]