探讨人工智能（AI）超声结合品管圈在低年资超声医师对甲状腺结节风险评估医疗质量中的效果。

以广东省第二人民医院2019年1月至10月共119个有手术病理结果的甲状腺结节的二维声像图作为图像资料。利用AI超声结合品管圈活动对2位低年资住院医师（医师1、2）的甲状腺结节风险评估能力进行持续质量改进。甲状腺结节的良恶性以手术病理作为金标准。甲状腺结节的征象以高年资医师组识别结果为金标准。活动前、后，2位低年资超声医师均采用2017年美国放射学会发布的甲状腺影像报告与数据系统甲状腺结节超声指南评估甲状腺结节，并统计活动前、后2名低年资医师对甲状腺超声检查操作的规范性、图像存储合格率及患者对低年资医师的信任度。绘制活动前、后2位医师对甲状腺结节良恶性诊断的受试者操作特征（ROC）曲线，并采用DeLong检验比较2位低年资超声医师诊断效能的差异。采用McNemar检验比较2位低年资超声医师在活动前、后对甲状腺结节超声征象的识别准确率的差异。

AI超声结合品管圈活动前、后，2名低年资医师对甲状腺结节声像图回声的识别准确率均有提高[医师1：47.90% vs 53.78%，医师2：45.38% vs 53.78%），差异具有统计学意义（P=0.031、0.004），其中医师2在活动前、后对甲状腺结节成分、形态、点状强回声方面的识别准确率均有所提高（69.75% vs 80.67%；58.82% vs 63.87%；52.10% vs 56.30%），差异具有统计学意义（P=0.004、0.021、0.031）。活动前、后2名低年资医师诊断甲状腺结节良恶性的ROC的曲线下面积明显提高（医师1：0.878 vs 0.921，P=0.036；医师2：0.824 vs 0.883，P=0.001）。此外，低年资医师的甲状腺超声检查操作规范合格率由60%提高至95%，图像存储合格率由50%提高至90%，患者对低年资医师的信任度由70%提高至90%。

AI超声结合品管圈活动可全方位、多维度提高低年资超声医师的甲状腺结节恶性风险评估能力，改善医疗服务质量。

To explore the impact of artificial intelligence (AI) ultrasound and quality control circle (QCC) activity on junior sonographers' ability to accurately estimate the risk of thyroid nodules.

From January to October 2019, 119 two-dimensional ultrasonographic images of thyroid nodules with surgical pathology outcomes were collected in our hospital. AI ultrasound in conjunction with QCC activity was used to continually enhance two junior physicians' (doctors 1 and 2) ability to assess thyroid nodule risk. The gold standard for diagnosing benign and malignant thyroid nodules was surgical pathology. The signs of thyroid nodules recognized by two senior physicians were used as the gold standard. Two junior sonographers used the 2017 American College of Radiology Thyroid Imaging Reporting and Data System Thyroid Nodules (ACR TI-RADS) to assess thyroid nodules both before and after the activity. Then, both before and after the activity, whether thyroid ultrasound examination procedures were standard, the qualified rate of image storage, and patients' trust in younger physicians were examined. Receiver operating characteristic (ROC) curves for benign and malignant thyroid nodules were plotted before and after the activity, and the DeLong test was used to compare the difference in diagnostic efficiency between the two junior sonographers. The McNemar test was used to compare the recognition accuracy of the two junior sonographers (doctor 1 and doctor 2) for thyroid nodule ultrasound signs before and after the activity.

The two young doctors' ability to identify the echo of thyroid nodules was increased by the use of AI ultrasound (doctor 1: 47.90% vs 53.78%, P=0.031; doctor 2: 45.38% vs 53.78%, P=0.004). Doctor 2's ability to recognize thyroid nodules' component (69.75% vs 80.67%, P=0.004), shape (58.82% vs 63.87%, P=0.021), and punctate hyperechogenicity (52.10% vs 56.30%, P=0.031) was similarly enhanced at the same time. The AUC values of the two junior doctors for diagnosing thyroid nodules were significantly higher after QCC activity than before (doctor 1: 0.878 vs 0.921, P=0.036; doctor 2: 0.824 vs 0.883, P=0.001). The qualified rate of image storage grew from 50% to 90%, the qualified rate of thyroid ultrasound examination by junior physicians improved from 60% to 95%, and the patients' trust in junior physicians increased from 70% to 90% as a result of QCC activity.

AI ultrasound combined with QCC activity can improve the ability of junior sonographers to assess the malignant risk of thyroid nodules in an all-round and multi-dimensional way, and improve the quality of medical services.