S-Detect技术在甲状腺可疑结节诊断中的应用价值

The application value of S-Detect technology in the diagnosis of suspected thyroid nodules

目的探讨S-Detect智能辅助诊断技术在甲状腺可疑结节诊断中临床应用价值。方法选取2019年9-12月在河南省肿瘤医院行超声检查并按美国放射学会(ACR)甲状腺影像报告与数据系统(TI-RADS)标准诊断为TI-RADS 3~4类的患者153例,分别由高年资医师、低年资医师及S-Detect对其进行良恶性鉴别诊断,然后再由高年资医师组和低年资医师组分别联合S-Detect诊断,以病理结果为金标准,绘制受试者工作特征(ROC)曲线,比较不同组间诊断结果的ROC曲线下面积、敏感性、特异性和准确性。结果153例甲状腺结节经手术病例证实良性108例(108个病灶),恶性45例(45个病灶)。S-Detect技术研究的敏感性、特异性、准确性分别为77.78%、87.04%、84.31%,高于低年资医师组的68.89%、79.63%、76.47%,但低于高年资医师组的84.44%、94.44%、91.50%;低年资组联合S-Detect后诊断的敏感性、特异性、准确性明显提高(77.78%、85.18%、83.01%)。低年资医师组联合S-Detect诊断的ROC曲线下面积(0.815)高于低年资医师组(0.743),差异有统计学意义(χ2=8.332,P=0.004);高年资医师组联合S-Detect诊断的ROC曲线下面积(0.901)与高年资医师组(0.894)比较,差异无统计学意义(χ2=0.095,P=0.756)。以病理结果为"金标准",高年资医师组联合S-Detect诊断的一致性最高(Kappa=0.797)。结论S-Detect技术诊断甲状腺的准确性较高,该技术有助于提高低年资医师诊断的特异性和准确性。

Objective To explore the clinical application value of S-Detect intelligent assistant diagnosis technology in the diagnosis of suspicious thyroid nodules.Methods A total of 153 patients who were diagnosed as TI-RADS 3-4 according to the ACR TI-RADS standard by ultrasound examination in Henan Provincial Cancer Hospital from September 2019 to December 2019 were included in the study. Differential diagnosis of benign and malignant thyroid nodules were made by senior doctors, junior doctors and S-Dectect technology, respectively. Then using pathological results as the standard, receiver operating charateric curve(ROC) curves were plotted to compare the area under curve (AUC), sensitivity, specificity and accuracy of S-Detect, senior and junior doctor groups, as well as their combinations.Results Of the 153 patients with thyroid nodules, 108 cases(108 nodules) were comfirmed benign and 45 cases(45 nodules) were malignant afeer operation. The diagnostic sensitivity, specificity and accuracy of S-Detect were 77.78%, 87.04%, 84.31%. They were higher than junior doctors(68.89%, 79.63%, 76.47%), but lower than senior doctors(84.44%, 94.44%, 91.50%). The diagnostic sensitivity, specificity and accuracy in junior doctors combined with S-Detect were significantly improved(77.78%, 85.18%, 83.01%). The area under curve of S-Detect diagnosis was higher in the group of junior doctors combined with S-Detect (0.815) than that in the group of junior doctors (0.743), and the difference was statistically significant (χ2=8.332, P=0.004). There was no statistically significant difference (χ2=0.095, P=0.756) in the AUC of diagnosis between the group of senior doctors combined with S-Detect(0.901) and senior doctors(0.894). Using pathological results as the "gold standard" , the highest consistency of diagnosis was found in the senior doctors combined with S-Detect(Kappa=0.797).Conclusions S-Detect technology has a high accuracy in the differential diagnosis of thyroid nodules. With the aid of this technology, it can improve the specificity and accuracy of diagnosis for the junior doctors.