计算机辅助诊断软件联合多学科建立甲状腺结节恶性风险预测模型  

A predictive model of thyroid nodules:based on the CAD software and multi-disciplinary indicators

作　　者：叶冯颖 杨文敏 李尚青 苏淇琛[1] 游剑虹[2] 王康健 吕国荣[1,4] YE Fengying;YANG Wenmin;LI Shangqing;SU Qichen;YOU Jianhong;WANG Kangjian;LYU Guorong(Department of Ultrasound,the Second Affiliated Hospital of Fujian Medical University,Quanzhou 362000,China;Department of Ultrasound,Zhongshan Hospital Affiliated to Xiamen University,Xiamen 361004,China;Department of Ultrasound,Zhangzhou Affiliated Hospital of Fujian Medical University,Zhangzhou 363000,China;Collaborative Innovation Center for Maternal and Infant Health Service Application Technology,Quanzhou 362011,China)

机构地区：[1]福建医科大学附属第二医院超声科,福建泉州362000 [2]厦门大学附属中山医院超声科,福建厦门361004 [3]福建医科大学附属漳州市医院超声科,福建漳州363000 [4]泉州医学高等专科学校母婴健康服务应用技术协同创新中心,福建泉州362011

出　　处：《中国医学物理学杂志》2021年第1期54-60,共7页Chinese Journal of Medical Physics

基　　金：泉州市科技计划项目(2019C076R);教育部泉州医学高等专科学校母婴健康服务应用技术协同创新中心经费资助;福建省临床重点专科建设项目。

摘　　要：目的:应用计算机辅助诊断(CAD)软件量化分析甲状腺结节的超声特点,结合临床及实验室指标,建立甲状腺结节恶性风险预测模型,检测预测模型诊断效能并与不同年资医师诊断对比。方法:模型建立组多中心、前瞻性地纳入2019年1月~9月在福建医科大学附属第二医院、厦门大学附属中山医院、漳州市医院接受甲状腺手术及术前超声检查的364例患者(共388个结节),采用CAD软件分析超声图像。收集CAD软件图像分析信息、临床信息及实验室信息作为相关因素。以病理为金标准,对比21种相关因素的良恶性组间差别,筛选出组间差异具有统计学意义的11种相关因素进行Logistic回归分析,筛选出对结节良恶性预测有统计学意义的6种相关因素进行模型建立。模型验证组纳入同期于3所医院行甲状腺细针穿刺(FNA)及穿刺前检查的105例患者(共105个结节)。由预测模型及3位不同年资的医师分别判断结节良恶性,对照病理结果,绘制受试者工作曲线(ROC),计算曲线下面积(AUC)、敏感度、特异度、阳性预测值(PPV)、阴性预测值(NPV),对比预测模型与不同年资医师的诊断效能。结果:建立预测模型为Logi(tp)=-5.218+2.601×(低回声指数)+1.981×(强回声指数)+3.079×(边缘模糊指数)+1.267×(纵横比>1)+0.614×(TSH)-0.071×(结节最大径)。计算可得模型的AUC为0.884,敏感度为85.50%,特异度为81.97%,PPV为91.1%,NPV为72.5%。预测模型的AUC、敏感度介于中、高年资医师间,特异度介于低、中年资医师间。结论:该模型具有较好的甲状腺结节恶性风险预测能力,可认为总体诊断效能介于中、高年资医师之间。Objective To quantitatively analyze the ultrasonic features of thyroid nodules using computer-aided diagnosis(CAD)software,to establish a predictive model of thyroid nodules based on CAD result,clinical information and laboratory indicators,and to compare the diagnostic efficacy of the predictive model with that of sonographers of different seniority.Methods The modelbuilding data set prospectively included 364 patients(a total of 388 nodules)who underwent thyroid surgery and preoperative ultrasound examination in three hospitals(the Second Affiliated Hospital of Fujian Medical University,Zhongshan Hospital Affiliated to Xiamen University,Zhangzhou Affiliated Hospital)from January 2019 to September 2019.The CAD analytic results,clinical information and laboratory indicators were collected.Using pathology as the gold standard,the differences of 21 related factors between benign and malignant thyroid groups was compared.Finally,11 factors were selected for logistic regression analysis,and 6 related factors with statistical significance for the prediction of benign and malignant nodules were selected for model establishment.The model validation data set included 105 patients(a total of 105 nodules)who underwent fine needle aspiration(FNA)in the three hospitals during the same period.The predictive model and three sonographers of different seniority made their judgments of the thyroid nodules separately.Using the pathology as the gold standard,the receiver operating characteristics curve(ROC)was drawn,the area under the curve(AUC),sensitivity,specificity,positive predictive value(PPV),negative predictive value(NPV)were calculated.The diagnostic efficacy of predictive model and sonographers of different seniority was compared with each other.Result A predictive model was built as follow:Logit(p)=-5.218+2.601×(hypoechogenicity index)+1.981×(echogenic foci index)+3.079×(edge blurring index)+1.267×(aspect ratio>1)+0.614×(TSH)-0.071×(tumor maximum diameter).The AUC of the model was 0.884,the sensitivity was 85.50%,t

关 键 词：计算机辅助诊断 甲状腺结节 超声 Logistics回归预测模型 

分 类 号：R318[医药卫生—生物医学工程] R581[医药卫生—基础医学]