计算机辅助诊断系统鉴别甲状腺结节良恶性的诊断效能及其影响因素

目的评估计算机辅助诊断系统（CAD）鉴别甲状腺结节良恶性的诊断效能及其影响因素。 方法回顾性分析2013年1月至2017年12月就诊于中山大学附属肿瘤医院的1035例患者资料（共1065个甲状腺结节）,所有甲状腺结节均经超声引导下细针穿刺活检或外科手术证实。利用CAD系统基于Kwak提出的甲状腺影像报告与数据系统（K-TI-RADS）、美国甲状腺协会（ATA）指南、美国放射协会（ACR）发表的ACR-TI-RADS指南分为3组并进行分析,以病理结果为"金标准",建立受试者工作特征（receiver operator characteristic,ROC）曲线,计算曲线下面积（area under the ROC curve,AUC）,采用Z检验比较组间AUC的差异。根据约登指数最大的切点值确定最佳诊断临界点,计算以最佳诊断临界点区分良恶性时,不同组间的敏感度、特异度、阳性预测值、阴性预测值和准确性,采用McNemar检验比较各组间差异。 结果ACR-TI-RADS指南诊断甲状腺结节的曲线下面积（AUC）高于K-TI-RADS及ATA指南（0.743 vs 0.703,0.743 vs 0.693）,差异均有统计学意义（Z=3.026、3.669,P均<0.01）。ACR-TI-RADS指南敏感度高于K-TI-RADS指南（91.95% vs 63.98%）,差异有统计学意义（P<0.001）,与ATA指南比较（91.95% vs 90.04%）,差异无统计学意义（P=0.7123）。以4c类为诊断界值时,K-TI-RADS指南特异度最高（66.23% vs 49.74%,66.23% vs 47.38%）,差异均有统计学意义（P均<0.001）。结节大小影响CAD的诊断效能,其中当结节最大径线为5~<10 mm或≥20 mm时,CAD的AUC最大,差异有统计学意义（P<0.001）。而CAD在正常甲状腺背景与桥本甲状腺炎背景下,AUC比较,差异无统计学意义（P=0.82）。 结论甲状腺CAD基于ACR-TI-RADS指南进行诊断时,显示出更好的诊断效能,而结节大小影响CAD的诊断效能。

Diagnostic efficacy of computer aided diagnosis system in differentiating benign and malignant thyroid nodules and its influencing factors

ObjectiveTo determine the diagnostic efficacy of the computer aided diagnosis (CAD) system in discriminating malignant and benign thyroid nodules and the possible influencing factors. MethodsThis retrospective study analyzed 1035 patients with 1065 thyroid nodules at Sun Yat-sen University Cancer Center between May 2013 and October 2017. All the nodules was proven by ultrasound guided fine needle aspiration cytology or thyroidectomy. The CAD system was used to analyze the thyroid nodules, according to the categories defined by the K-TI-RADS, ATA, and ACR-TI-RADS guidelines. Using pathology as the gold standard, receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic efficacy of the CAD system. In addition, the differences of CAD diagnostic efficacy were compared among different thyroid nodule sizes and thyroid background. ResultsThe area under the ROC curve (AUC) of the ACR-TI-RADS was higher than those of the K-TI-RADS (Z=3.026, P=0.0025) and ATA (Z=3.669, P<0.001). The sensitivity of the ACR-TI-RADS was better than that of the K-TI-RADS (P<0.001), and was comparable to that of the ATA (P=0.7123). Using 4C as the cutoff value, the K-TI-RADS demonstrated a significantly higher specificity than the ATA and ACR-TI-RADS (P<0.001). The size of thyroid nodules influenced the diagnostic efficacy of CAD. The AUCs of CAD in the 5~<10 mm and ≥20 mm group were higher than those in the 10~<15 mm and 15~<20 mm group (P<0.001). There was no significant difference in AUC of the CAD between the normal background and Hashimoto thyroiditis background groups (P=0.82). ConclusionThe CAD system exhibits a better diagnostic efficacy based on the ACR-TI-RADS, and the size of thyroid nodules influences the diagnostic efficacy of CAD