摘要目的 探讨计算机辅助检测和诊断(CAD)软件中 K‐TIRADS 、 ACR‐TIRADS 、 ATA风险分层的诊断效能以及辅助超声医师诊断甲状腺结节的应用价值.方法 回顾性分析192个有术后病理的甲状腺结节资料,使用CAD软件对甲状腺结节分别进行K‐TIRADS 、 ACR‐TIRADS 、 ATA 分类,统计ROC曲线下面积、敏感性、特异性等得出最佳指南,以最佳指南为分类标准,采用双盲法对比同一超声医师在使用CAD前、结合CAD后对甲状腺结节的诊断能力.结果 CAD软件中 K‐TIRADS 、ACR‐T IRADS 、 A T A的AUC分别为 0 .88 、 0 .77 、 0 .62 ,两两比较差异有统计学意义( P ＜ 0 .05 ) ;K‐T IRADS 与AT A的特异性比较差异无统计学意义( P ＝ 0 .176 ) ,但均高于 ACR‐T IRADS ( P ＜0 .05) . CAD 、超声医师、超声医师结合CAD使用K‐TIRADS的AUC分别为0 .88 、 0 .80 、 0 .93 ,两两比较差异有统计学意义( P ＜0 .05) ;CAD和超声医师结合CAD敏感性比较差异无统计学意义( P ＝0 .163) ,但均高于超声医师人工判读( P ＜0 .05) ;CAD 、超声医师、超声医师结合CAD特异性两两比较差异有统计学意义( P ＜0 .05) .结论 CAD软件中3种甲状腺超声风险分层系统均具有良好的诊断价值,其中K‐TI RADS的AUC最大,CAD软件辅助超声医师能提高对诊断甲状腺结节的诊断效能,具有较好的临床应用前景.

abstractsObjective To explore the diagnostic efficiency of K‐T IRADS ,ACR‐T IRADS and AT A risk stratification in computer‐aided detection and diagnosis ( CAD ) software and the application value of CAD‐assisted ultrasound physicians in diagnosing thyroid nodules . Methods One hundred and ninety‐two thyroid nodules with postoperative pathological results were retrospectively analyzed . All of them were graded by K‐T IRADS ,ACR‐T IRADS and A T A with CAD software ,and the best guide was recognized by calculating the area under the ROC curve ,sensitivity and specificity . T hen ,based on the best guidelines for the classification criteria , the double‐blind method was used to compare the ability of the same ultrasonologist to diagnose thyroid nodules before and after CAD . Results T he AUC value of K‐T IRADS , ACR‐T IRADS ,A T A was 0 .88 ,0 .77 ,0 .62 respectively in the CAD software . T he difference between the two groups was statistically significant ( P ＜0 .05 ) . T here was no significant difference in the specificity between K‐T IRADS and A T A ( P ＝0 .176 ) ,w hich were both higher than ACR‐T IRADS with statistically significant differences ( P ＜ 0 .05 ) . T he AUC value of the diagnosis among CAD itself , ultrasound physicians and physicians combined CAD was 0 .88 ,0 .80 ,0 .93 ,respectively . T he difference between the two groups was statistically significant ( P ＜0 .05) . T here was no significant difference in the sensitivity between CAD itself and physicians combined CAD ( P ＝0 .163 ) ,w hich were both higher than ultrasound physicians with statistical significant differences( P ＜0 .05) . Among ultrasound physicians ,CAD itself and physicians combined CAD ,the difference in specificity between the two groups was statistically significant ( P ＜0 .05) . Conclusions All the three risk stratification systems of thyroid ultrasound in CAD software have good diagnostic values ,among w hich K‐T IRADS has the largest AUC . T he CAD software can assist ultrasound physicians to improve the thyroid nodule diagnostic performance , and has a good clinical application prospect .