不同TI-RADS分类系统对甲状腺结节的诊断效能对比研究

目的比较美国放射学会甲状腺影像报告与数据系统（ACR TI-RADS）、人工智能算法简化版甲状腺影像报告与数据系统（AI TI-RADS）和Kwak甲状腺影像报告与数据系统（Kwak TI-RADS）对甲状腺结节良恶性的诊断价值。 方法收集2013年5月至2017年7月于郑州大学第一附属医院超声检查诊断为甲状腺结节并经术后病理证实的2114例患者共3134个结节的超声图像资料。应用ACR TI-RADS、AI TI-RADS和Kwak TI-RADS对所有结节进行分类，构建ROC曲线，比较三者对甲状腺结节良恶性的诊断效能。 结果3134个结节中，病理证实良性1566个，恶性1568个。将结节按最大径分为＜10 mm组1450个和≥10 mm组1684个，＜10 mm组的结节恶性率高于≥10 mm组（59.7% vs 41.7%，χ²=101.399，P＜0.001）。AI TI-RADS和Kwak TI-RADS诊断甲状腺结节良恶性的ROC曲线下面积分别为0.897、0.899，均高于ACR TI-RADS（0.879，P均＜0.05）。AI TI-RADS与Kwak TI-RADS的诊断敏感度分别为86.4%和88.2%，准确性分别为86.1%和86.1%，均高于ACR TI-RADS（敏感度和准确性分别为80.9%、83.0%），差异均有统计学意义（P均＜0.05）。三者特异度两两比较差异均无统计学意义（P均＞0.05）。3种分类方法诊断≥10 mm组甲状腺结节良恶性的ROC曲线下面积均大于＜10 mm组（Kwak TI-RADS：0.922 vs 0.853，AI TI-RADS：0.924 vs 0.845，ACR TI-RADS：0.907 vs 0.830）；无论≥10 mm组还是＜10 mm组，AI TI-RADS和Kwak TI-RADS诊断甲状腺结节良恶性的ROC曲线下面积均大于ACR TI-RADS（P均＜0.05）。 结论Kwak TI-RADS和AI TI-RADS对甲状腺结节的综合诊断效能优于ACR TI-RADS。Kwak TI-RADS操作简单，临床实用性强，而AI TI-RADS对结节的分类更全面，更有利于甲状腺结节的风险管理。

Efficacy of different TI-RADS classification systems in diagnosis of thyroid nodules

ObjectiveTo compare the diagnostic value of American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS), Artificial Intelligence Thyroid Imaging Reporting and Data System (AI TI-RADS), and Kwak Thyroid Imaging Reporting and Data System (Kwak TI-RADS) for benign and malignant thyroid nodules. MethodsThe ultrasound images of 3134 thyroid nodules were collected from 2114 patients diagnosed as having thyroid nodules by ultrasound examination and confirmed by postoperative pathology at the First Affiliated Hospital of Zhengzhou University from May 2013 to July 2017. All nodules were categorized by ACR TI-RADS, AI TI-RADS, and Kwak TI-RADS, and receiver operating characteristic (ROC) curves were plotted to compare the diagnostic efficacy of the three. ResultsAmong the 3134 nodules, 1566 were benign and 1568 were malignant. According to the maximum diameter, the nodules were divided into <10 mm group (1450 nodules) and ≥10 mm group (1684 nodules). The malignant rate of nodules in the <10 mm group was higher than that in the ≥10 mm group (59.7% vs 41.7%, χ²=101.399, P＜0.001). The areas under the ROC curves of AI TI-RADS and Kwak TI-RADS for diagnosing benign and malignant thyroid nodules were 0.897 and 0.899, respectively, which were higher than that of ACR TI-RADS (0.879; P＜0.05).The sensitivities of AI TI-RADS and Kwak TI-RADS were 86.4% and 88.2%, and the accuracies were 86.1% and 86.1%, respectively, which were significantly higher than those of ACR TI-RADS (80.9% and 83.0%; P＜0.05). Pairwise comparisons of the specificities of the three showed no statistically significant difference (P＞0.05). The areas under the ROC curves of the three classification systems for diagnosing benign and malignant thyroid nodules were higher in the ≥10 mm group than in the <10 mm group (Kwak TI-RADS: 0.922 vs 0.853; AI TI-RADS: 0.924 vs 0.845; ACR TI-RADS: 0.907 vs 0.830). Whether ≥10 mm group or <10 mm group, the areas under the ROC curves of AI TI-RADS and Kwak TI-RADS performed better than ACR TI-RADS (P＜0.05). ConclusionKwak TI-RADS and AI TI-RADS are more effective than ACR TI-RADS in the comprehensive diagnosis of thyroid nodules. Kwak TI-RADS is simple to operate and has strong clinical practicability, while AI TI-RADS is more comprehensive in the classification of nodules, which is more conducive to the risk management of thyroid nodules.