目的 探讨基于超声影像特征与定量参数构建甲状腺结节癌变诊断模型的价值，并进行验证。方法 回顾性选取我院2019年1月至2021年2月253例甲状腺结节患者，根据病理结果将甲状腺良性结节患者作为良性组、甲状腺恶性结节作为癌变组。按照国际诊断模型建模共识，根据时间顺序将2019年1月至2020年1月（良性组112例、癌变组74例）组成训练集，2020年2月至2021年2月（良性组36例、癌变组31例）组成验证集，对比两组超声影像特征与定量参数[收缩期峰值流速（PSV）、血流血管指数（VFI）、舒张末期血流速度（EDV）、阻力指数（RI）]，分析甲状腺结节癌变的影响因素，构建诊断模型，评价该模型的诊断价值并进行验证。结果 训练集和验证集中良性组与癌变组内部回声、包膜完整性、形态、边界、质地、纵横比、是否微钙化、血流情况、PSV、VFI、EDV、RI差异有统计学意义（P＜0.05）；训练集中良性组与癌变组结节大小差异有统计学意义（P＜0.05）；Logistic回归分析，内部低回声、形态不规则、质地以实性为主、纵横比≥1、微钙化、PSV、VFI、RI是甲状腺结节癌变的独立危险因素（P＜0.05）；利用上述高危因素构建诊断模型为-4.178+内部回声×2.593+形态×2.585+质地×2.113+纵横比×2.053+微钙化×2.518+PSV×1.681+VFI×1.583+RI×1.844，似然比卡方检验提示模型建立具有统计学意义，Wald卡方检验提示模型构建有效，Hosmer-Lemeshow拟合优度检验显示模型拟合效果较好；根据诊断模型绘制ROC曲线，训练集的AUC为0.945，95%CI为0.902～0.973，敏感度为89.19%，特异度为86.61%；验证集的AUC为0.925，95%CI为0.834～0.975，敏感度为90.32%，特异度为83.33%。结论 基于超声影像特征与定量参数构建甲状腺结节癌变诊断模型具有可靠诊断价值，可作为临床优选的影像辅助诊断方式。

Objective To investigate the value of constructing a diagnostic model of thyroid nodule canceration based on ultrasound image features and quantitative parameters, and to verify it. Methods A total of 253 patients with thyroid nodules in our hospital from January 2019 to February 2021 were retrospectively selected. According to the pathological results, patients with benign thyroid nodules were selected as the benign group and malignant thyroid nodules as the cancerous group. According to the international diagnostic model modeling consensus, the training set was composed from January 2019 to January 2020 (112 cases in the benign group and 74 cases in the cancerous group) according to the time sequence, and from February 2020 to February 2021 (36 cases in the benign group). , 31 cases of cancer group) to form a validation set, and compared the two groups of ultrasound image features and quantitative parameters [peak systolic velocity (PSV), blood flow vascular index (VFI), end-diastolic blood velocity (EDV), resistance index (RI) ], analyze the influencing factors of thyroid nodule carcinogenesis, build a diagnostic model, evaluate the diagnostic value of the model and verify it. Results There were significant differences in internal echo, capsule integrity, morphology, boundary, texture, aspect ratio, microcalcification, blood flow, PSV, VFI, EDV and RI between the benign group and the cancerous group in the training set and validation set (P <0.05); there was a statistically significant difference in the nodule size between the benign group and the cancerous group in the training set (P<0.05); Logistic regression analysis showed that the internal hypoechoic, irregular shape, mainly solid texture, aspect ratio ≥1, microscopic Calcification, PSV, VFI and RI were independent risk factors for thyroid nodule carcinogenesis (P＜0.05); Using the above high-risk factors to construct a diagnostic model as -4.178+internal echo×2.593+morphology×2.585+texture×2.113+aspect ratio×2.053+microcalcification×2.518+PSV×1.681+VFI×1.583+RI×1.844, likelihood ratio The square test indicates that the model is statistically significant, the Wald chi-square test indicates that the model is effective, and the Hosmer-Lemeshow goodness-of-fit test shows that the model is well fitted; according to the ROC curve drawn from the diagnostic model, the AUC of the training set is 0.945, 95 The %CI was 0.902-0.973, the sensitivity was 89.19%, and the specificity was 86.61%; the AUC of the validation set was 0.925, and the 95% CI was 0.834-0.975, the sensitivity was 90.32%, and the specificity was 83.33%. Conclusion The construction of a thyroid nodule cancer diagnosis model based on ultrasound image features and quantitative parameters has reliable diagnostic value, and can be used as a clinically preferred imaging-assisted diagnosis method.