基于多模态超声的甲状腺乳头状癌决策树模型的构建及其诊断效能评估

A Multimodal Ultrasound-based Decision-making Tree Model for the Diagnosis of Papillary Thyroid Carcinoma:Development and Efficacy Evaluation

背景随着多种超声检查技术不断发展及城乡居民保健意识增强,近年来甲状腺乳头状癌(PTC)检出率逐年升高,而构建相应的决策树模型有助于及早发现和诊断PTC。目的基于多模态超声构建PTC决策树模型并评估其诊断效能。方法选取2018年1月至2020年10月在华北理工大学附属医院住院并行甲状腺结节切除术的PTC患者180例,共发现186个结节,其中良性结节99个(非PTC组),恶性结节87个(PTC组)。比较PTC组与非PTC组常规超声、实时剪切波弹性成像(SWE)、超声造影(CEUS)检查结果,分别构建基于常规超声、SWE、CEUS及多模态超声的PTC决策树模型并进行诊断效能评估。结果两组结节回声、纵横比、边缘、局灶强回声、弹性最大值(Emax)、弹性最小值(Emin)、弹性均值(Emean)、弹性标准差(Esd)、与周围正常组织弹性比值(Eratio)、增强程度、增强特点、造影剂分布、造影剂进入时间、造影剂消退时间、达峰浓度(Peak)、时间-强度曲线下面积(AUCt)、平均渡越时间(MTT)比较,差异有统计学意义(P<0.05)。分别构建基于常规超声、SWE、CEUS及多模态超声的PTC决策树模型,结果显示其根节点分别为局灶强回声、Emax、AUCt、Emax;十折交叉验证法测试结果显示,基于常规超声、SWE、CEUS、多模态超声的PTC决策树模型误判率分别为33.9%、19.4%、37.6%、7.0%。基于多模态超声的PTC决策树模型的灵敏度、特异度、准确率、阳性似然比、阴性似然比、Kappa值分别为88.5%、99.0%、94.1%、88.5、0.12、0.880,诊断效能明显高于基于常规超声、SWE、CEUS的PTC决策树模型。结论本研究基于多模态超声成功构建了PTC决策树模型,并具有较高的诊断效能,有利于提高PTC的诊断准确率并为临床提供新的诊断思路。

Background The detection rate of papillary thyroid carcinoma(PTC)is increasing in recent years,which may be due to advances in various ultrasonic imaging technologies and residents’increased awareness of participating in the screening for PTC.The development of a decision-making tree model is helpful for identifying and diagnosing PTC timely.Objective To develop a multimodal ultrasound-based decision-making tree model for the diagnosis of PTC,and to assess its efficacy.Methods One hundred and eighty inpatients with excision of thyroid nodules were recruited from North China University of Technology Affiliated Hospital from January 2018 to October 2020.One hundred and eighty-six thyroid nodules were found in them,and 87 of which were malignant(PTC group),and other 99 were benign(non-PTC group).The results of routine ultrasound,shear wave elastography(SWE),and contrast-enhanced ultrasound(CEUS)of PTC and non-PTC groups were compared.And the imaging parameters of PTC detected by routine ultrasound,SWE,CEUS,and the combination of routine ultrasound,SWE and CEUS(hereinafter referred to as“multimodal ultrasound”)were used to develop a decision-making tree model,respectively,then the efficacies of these four models were evaluated.Results There were significant differences in nodule echo,aspect ratio,edge,focal hyperecho,maximum elasticity(Emax),minimum elasticity(Emin),mean elasticity(Emean),standard deviation of elasticity(Esd),elasticity ratio to normal surrounding tissue(Eratio),enhancement degree,enhancement characteristics,contrast medium distribution,contrast medium arrival time,contrast medium subsidence time,peak concentration(Peak),area under the time-intensity curve(AUCt)and mean transit time(MTT)between PTC and non-PTC groups(P<0.05).Root nodes of decision-making tree models based on imaging parameters of PTC measured by routine ultrasound,SWE,CEUS and multimodal ultrasound were focal hyperecho,Emax,AUCt and Emax,respectively.Ten-fold cross-validation test showed that,misdiagnosis rates of decision-making tree models for PTC based on routine ultrasound,SWE,CEUS and multimodal ultrasound were 33.9%,19.4%,37.6%and 7.0%,respectively.The sensitivity,specificity,accuracy rate,positive likelihood ratio,negative likelihood ratio and Kappa value of multimodal ultrasound-based decision-making tree model for PTC were 88.5%,99.0%,94.1%,88.5,0.12 and 0.880,respectively.And it had higher diagnostic efficiency than other three models.Conclusion We successfully constructed the multimodal ultrasound-based decision-making tree model for PTC with relatively high diagnostic efficiency.Moreover,it contributes to the improvement of diagnostic accuracy of PTC,which may be considered as a new approach to diagnosing PTC.