摘要
目的 应用超声影像组学评分(Rad-score)、甲状腺结节超声恶性危险分层中国指南(C-TIRADS)判定甲状腺结节良恶性质,探讨二者联合诊断恶性甲状腺结节的价值。方法 2020年1月—2022年1月湖北医药学院附属人民医院诊治甲状腺结节患者689例(847个结节)为训练组,同期诊治甲状腺结节患者300例(403个结节)为验证组。训练组二维超声声像应用labelme软件分割感兴趣区,提取超声影像特征,应用LASSO回归筛选有能力预测结节良恶性质的影像组学优势特征,进行10倍交叉验证后构建超声影像组学模型,计算出评判恶性甲状腺结节的Rad-score阈值为-0.23。绘制ROC曲线评估Rad-score诊断验证组恶性甲状腺结节的效能;应用C-TIRADS对验证组进行恶性风险分层,判定结节性质。采用多因素logistic回归分析Rad-score及C-TIRADS分类中的超声声像特征对训练组甲状腺结节性质的影响,并依据多因素logistic回归分析结果构建Rad-score联合C-TIRADS预测甲状腺结节性质的的列线图模型,采用ROC曲线验证列线图模型对验证组恶性甲状腺结节的预测价值。以组织病理结果为金标准,计算并比较Rad-score、C-TIRADS单独及联合诊断验证组恶性甲状腺结节的准确率、特异度、灵敏度。结果 (1)Rad-score阈值为-0.23时,诊断验证组恶性甲状腺结节的AUC为0.918(95%CI:0.887~0.950,P=0.030),灵敏度为93.9%,特异度为89.9%。验证组403个结节中,Rad-score诊断良性134个,恶性269个。Rad-score诊断验证组恶性甲状腺结节的准确率为92.1%(371/403),灵敏度为92.8%(257/277),特异度为90.5%(114/126)。(2)验证组403个结节中,C-TIRADS分类2类34个、3类42个、4A类27个、4B类33个、4C类104个、5类163个。C-TIRADS诊断恶性甲状腺结节的准确率为86.6%(349/403),灵敏度为88.4%(245/277),特异度为82.5%(104/126)。(3)Rad-score(OR=3.247,95%CI:3.821~107.623,P=0.001)、内部成分(OR=2.427,95%CI:1.906~8.271,P=0.017)、内部回声(
Objective To determine the benign and malignant thyroid nodules(TNs) by using ultrasound radiomics score(Rad-score) and Chinese guidelines for thyroid nodules malignant risk stratification by ultrasound(C-TIRADS),and to explore the value of the combination of them two to the diagnosis of malignant TNs.Methods A total of 689 patients with TNs(847TNs)diagnosed and treated in People's Hospital Affiliated to Hubei University of Medicine from January2020to January 2022 were selected as training group,and another 300patients with TNs(403TNs)diagnosed and treated during the same period were selected as validation group.Labelme software was used to segment the region of interest and extract the ultrasound image features.LASSO regression was used to select the radiomics dominant features that could predict the benign and malignant TNs,and the radiomics model was constructed after 10-fold cross-validation.The Rad-score threshold for malignant TNs was-0.23.ROC curves were drawn to evaluate the efficiency of Rad-score on the diagnosis of malignant TNs in validation group.C-TIRADS was used to stratify the malignant risk of validation group and determine the nature of TNs.Multivariate logistic regression analysis was used to analyze the influence of Rad-score combined with C-TIRADS classification on the nature of TNs in training group,by which a combined nomogram model was constructed to predict the nature of TNs.ROC curves were used to verify the value of the nomogram model to the prediction of malignant TNs in validation group.Taking the histopathological results as the gold standard,the accuracy,specificity and sensitivity of Rad-score and C-TIRADS alone and in combination in the diagnosis of malignant TNs were calculated and compared in validation group.Results(1)When the Rad-score threshold was-0.23,the AUCfor diagnosing malignant TNs was 0.918(95%CI:0.887-0.950,P=0.030),the sensitivity was93.9%,and the specificity was 89.9%in validation group.Among the 403TNs in validation group,the Rad-score model diagnosed 134benign TN
作者
曹婧芳
胡培
郑霜
朱圆圆
刘文婷
刘玺
肖彬
CAO Jingfang;HU Pei;ZHENG Shuang;ZHU Yuanyuan;LIU Wenting;LIU Xi;XIAO Bin(Department of Ultrasound,People's Hospital Affiliated to Hubei University of Medicine,Shiyan,Hubei 442000,China;Healthcare Big Data Center,School of Public Health of Hubei University of Medicine,Shiyan,Hubei 442000,China)
出处
《中华实用诊断与治疗杂志》
2023年第7期678-684,共7页
Journal of Chinese Practical Diagnosis and Therapy
基金
湖北省卫生健康委员会资助项目(WJ2021F041)。
关键词
甲状腺结节
超声影像组学模型
甲状腺结节超声恶性危险分层中国指南
thyroid nodules
ultrasound radiomics
Chinese guidelines for thyroid nodules malignant risk stratification byultrasound
