人工智能辅助胸部低剂量 CT 在肺部结节良、恶性诊断中的准确率研究

目的? 探讨人工智能辅助胸部低剂量 CT 在肺部结节良、恶性诊断中的应用。方法? 选取 2020 年 3 月至 2021 年 7 月在新乡同盟医院行胸部低剂量 CT 检查患者 83 例,根据阅片方式不同分为人工阅片组和人工智能辅助阅片组,观察两组肺部结节的诊断结果。结果? 人工智能辅助阅片诊断肺部结节的阳性率为 86.75%,高于人工阅片的阳性率 68.67%,差异具有统计学意义（x 2 =6.549,P=0.013）;Kappa 检验两种阅片方式的一致性较弱（Kappa 值 = 0.196）,P ＞ 0.05;人工智能辅助阅片对3～7mm直径肺部结节的检出率为93.44%,明显高于人工阅片的检出率85.25%,差异具有统计学意义（P＜0.05）,而两种阅片方式对 0～3mm、7～20mm 直径肺部结节的检出率差异无统计学意义（P ＞ 0.05）;以病理结果为“金标准”绘制 ROC 曲线,结果显示,人工阅片、人工智能辅助阅片诊断恶性肺部结节的 AUC 分别为 0.742（95%CI：0.514,0.921）、0.830（95%CI：0.701,1.00）,且两种阅片方式的特异度、灵敏度、阳性预测值、阴性预测值差异均无统计学意义（P ＞ 0.05）。结论? 人工智能辅助胸部低剂量 CT 能提高肺部结节良、恶性诊断的准确率,并能提高 3～7mm 直径肺部结节的检出率,但与人工阅片对肺部结节诊断的特异度、灵敏度、阳性预测值、阴性预测值基本一致。

Research on the Accuracy of Artificial Intelligence Assisted Low-Dose Chest CT in the Differential Diagnosis of Benign and Malignant Pulmonary Nodules

Objective To explore the application of artificial intelligence assisted low-dose chest CT in the differential diagnosis of benign and malignant pulmonary nodules. Methods 83 patients who underwent low-dose chest CT examination at Xinxiang Tongmeng Hospital from March 2020 to July 2021 were selected. They were divided into two groups based on different methods of examination: simple manual examination group and artificial intelligence assisted examination group. The diagnostic results of lung nodules in both groups were observed. Results The positive rate of artificial intelligence assisted film reading in diagnosing pulmonary nodules was 86.75%, which was higher than that of manual film reading of 68.67%, and the difference was statistically significant (x2=6.549, P =0.013); The Kappa test showed weak consistency between the two reading methods (Kappa value=0.196), with P >0.05; The detection rate of pulmonary nodules with a diameter of 3~7mm using artificial intelligence assisted film reading was 93.44%, significantly higher than that of 85.25% using manual film reading, and the difference was statistically significant (P<0.05). However, there was no statistically significant difference in the detection rate of pulmonary nodules with a diameter of 0~3mm and 7~20mm between the two film reading methods (P >0.05); Using pathological results as the gold standard, the ROC curve was plotted, and the results showed that the AUC for diagnosing malignant pulmonary nodules using manual film reading and artificial intelligence assisted film reading were 0.742 (95%CI: 0.514, 0.921) and 0.830 (95%CI: 0.701, 1.00) respectively. The specificity, sensitivity, positive predictive value, and negative predictive value of the two film reading methods were not statistically significant (P >0.05). Conclusion Artificial intelligence assisted low-dose chest CT can improve the accuracy of diagnosing benign and malignant pulmonary nodules, and increase the detection rate of pulmonary nodules with a diameter of 3~7mm. However, it is consistent with the specificity, sensitivity, positive predictive value, and negative predictive value of manual film reading for the diagnosis of pulmonary nodules.