基于超声特征及基因检测技术构建甲状腺结节恶性风险预测模型  

Construction of malignant risk prediction model of thyroid nodules based on ultrasound features and gene detection techniques

作　　者：何巧灵 何美男 湛婉华 龙娟 魏雪云 邹梦晨[2] HE Qiao-ling;HE Mei-nan;ZHAN Wan-hua;LON Juan;WEI Xue-yun;ZOU Meng-chen(Zengcheng Central Hospital of Guangzhou(Zengcheng Branch of Nanfang Hospital),Guangzhou 511340,China;Nanfang Hospital,Southern Medical University,Guangzhou 510515,China)

机构地区：[1]广州市增城区中心医院(南方医院增城分院),广东广州511340 [2]南方医科大学南方医院,广东广州510515

出　　处：《广东医科大学学报》2023年第2期159-164,共6页Journal of Guangdong Medical University

基　　金：广东省医学科研基金(B2022025)。

摘　　要：目的探讨基于超声特征及基因检测技术构建甲状腺结节恶性风险预测模型的可行性。方法回顾性分析202例行甲状腺手术患者的临床资料(模型建立组),所有患者均完善超声检查,而后采用甲状腺超声计算机辅助诊断(CAD)软件对其超声图像进行分析。以病理结果为金标准,对所有患者的临床信息、实验室信息及CAD软件图像分析信息进行logstic回归分析,并构建风险模型。另纳入同期完善穿刺前检查并行甲状腺细针穿刺(FNA)的200例患者作为模型验证组。采用受试者工作曲线(ROC)分析预测模型和不同年资医师对甲状腺结节良恶性的判断结果,比较诊断效能差异。结果低回声强度、强回声点、边缘模糊程度、纵横比>1及血清促甲状腺激素(TSH)、甲状腺过氧化物酶抗体(TPOAb)、癌胚抗原(CEA)水平升高、基因突变阳性是甲状腺结节恶性的高危因素,结节最大径是保护因素(P<0.05或0.01)。甲状腺结节恶性风险预测模型的ROC线下面积AUC为0.891(95%CI 0.773~0.981),最佳预测临界值为70.67%,敏感度为85.63%,特异度为82.13%,阳性预测值为91.36%,阴性预测值为76.81%。研究所构建的甲状腺结节恶性风险模型诊断效能高于中、低年资医师,低于高年资医师,特异度介于中、低年资医师之间(P<0.05)。结论根据超声特征、细针穿刺病理、TSH水平、TPOAb、基因检测联合logistic回归分析所构建的甲状腺结节恶性风险模型可有效辅助临床预测恶性甲状腺结节的发生。Objective To explore the feasibility of constructing a malignant risk prediction model of thyroid nodules based on ultrasound features and gene detection techniques.Methods The clinical data of 202 patients who underwent thyroid surgery(Model Establishment Group)were retrospectively analyzed.All the patients have received ultrasonic testing,and then their ultrasonic images were analyzed by CAD software.Taking the pathological results as the gold standard,the logistic regression analysis was performed on the clinical information,laboratory information and image analysis information of CAD software of all patients,and a risk model was constructed.In addition,another 200 patients who received prepuncture examination and underwent fine thyroid needle aspiration(FNA)during the same period were included as the Model Verification Group.The receiver operating curve(ROC)was used to analyze the prediction model and the judgment results of different experienced physicians on the benign and malignant thyroid nodules,and the diagnostic efficiency was compared.Results Hypoechoic intensity,strong echo points,indistinct edge,aspect ratio>1,elevated serum TSH,TPOAb,TGAb and CEA levels,positive gene mutation were the high risk factors for malignant thyroid nodules,while the maximum diameter of nodules was the protective factor(P<0.05 or 0.01).The ROC AUC of the malignant risk prediction model of thyroid nodules was 0.891(95%CI 0.773-0.981),with the optimal predictive threshold of 70.67%,the sensitivity of 85.63%,the specificity of 82.13%,the PPV of 91.36% and the NPV of 76.81%.The malignant risk prediction model of thyroid nodules constructed in this study had the diagnostic efficiency higher than the physicians with medium and low seniority,but lower than the physicians with high seniority,and had the specificity between the physicians with low and those with middle seniority(P<0.05).Conclusion The malignant risk prediction model of thyroid nodules constructed according to ultrasonic features,fine needle aspiration pathology,TSH

关 键 词：超声特征 细针穿刺病理 促甲状腺激素 甲状腺过氧化物酶抗体 甲状腺结节 风险预测模型 

分 类 号：R445.1[医药卫生—影像医学与核医学]