关键词: 甲状腺癌, 乳头状；甲状腺结节；决策树；多模态成像；多模态超声；诊断

Abstract: 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.

Key words: Thyroid cancer, papillary；Thyroid nodule；Decision trees；Multimodal imaging；Multimodal ultrasound；Diagnosis