基于多模态超声构建预测甲状腺结节性质的定量诊断模型

Construction of a quantitative diagnosis model for predicting the nature of thyroid nodules based on multi-modality ultrasound images

目的通过分析患者临床信息及甲状腺病灶多模态超声图像特征构建定量诊断列线图模型,以期术前准确预测可疑结节恶性概率,为临床决策提供有效参考。方法收集在2020年9月1日至2021年6月10日于哈尔滨医科大学附属第二医院行手术治疗的甲状腺结节超声恶性危险分层中国指南(C-TIRADS)3~5类的933例患者共计1121个结节,按照8∶2随机分为训练集897个,测试集224个。基于逻辑回归分析筛选甲状腺结节良恶性相关因素构建列线图,并以曲线下面积(AUC)评估其诊断效能。结果①经单因素分析初筛后,多因素分析显示年龄、回声、方位、局灶强回声、边缘、后方回声特征、弹性评分与结节良恶性显著相关(均P<0.001),良恶性结节的声晕差异有统计学意义(P=0.012);②列线图在测试集AUC为0.903(95%CI=0.862~0.944),并在最大径≤10 mm或>10 mm的结节中分别达到0.889(95%CI=0.832~0.946)、0.960(95%CI=0.925~0.994),显示出较高的诊断性能。结论该列线图能够术前有效鉴别甲状腺结节良恶性,对最大径>10 mm的结节具有较高的诊断性能,可有效减少不必要的穿刺活检和手术。

Objective To construct a quantitatively diagnostic nomogram model by analyzing the clinical information of patients and the features of multi-modality ultrasound images of thyroid lesions,so as to preoperatively predict the malignant probability of suspicious thyroid nodules and provide effective references for clinical decision-making.Methods A total of 933 patients,1121 thyroid nodules of C-TIRADS 3-5 categories,who underwent surgery in the Second Affiliated Hospital of Harbin Medical University from September 1,2020 to June 10,2021 were collected.The nodules were randomly divided into training(n=897)and test groups(n=224)in 8∶2 ratio.Finally,the diagnostic performance was evaluated by area under the curve(AUC).Results①After preliminary screening by univariate analysis,multivariate analysis showed that age,echogenicity,orientation,echogenic foci,margin,posterior features,and elastic score were significantly correlated with benign and malignant nodules(all P<0.001),and the difference of halo between benign and malignant nodules was also statistically significant(P=0.012).②The AUC of nomogram was up to 0.903(95%CI=0.862-0.944)in the test set,and 0.889(95%CI=0.832-0.946)and 0.960(95%CI=0.925-0.994)in nodules with maximum diameter of≤10 mm and of>10 mm respectively,which showed high diagnostic performance.Conclusions The nomogram model could accurately differentiate malignant from benign thyroid nodules preoperatively,with the highest diagnostic performance for the nodules with maximum diameter of>10 mm,and effectively avoid the unnecessary fine-needle biopsy and surgical operation.