人工智能在肺小结节诊断中的临床研究与应用

Clinical research and application of AI in the diagnosis of pulmonary nodules

目的:探讨人工智能(AI)在肺小结节诊断中的临床研究与应用价值。方法:收集在医院行肺部CT检查的2000例患者,分别采用人工阅片与杏脉悦影肺结节AI初步阅片两种方式进行阅片。比较两种阅片方式在肺小结节分析中的临床应用价值。结果:杏脉悦影肺结节AI初步阅片与人工阅片两种阅片方式发现的结节数量、结节类型及准确率均比较接近,肺成像报告和数据系统(LungRADS)1~5类准确率分别为93%、90%、79%、73%和100%。两种阅片方式诊断肺恶性结节的受试者工作特征(ROC)曲线下面积(AUC)分别为0.739和0.828,二者AUC均>0.7,提示杏脉悦影肺结节AI初步阅片的诊断准确率较高。结论:杏脉悦影肺结节AI初步阅片与人工阅片在肺小结节诊断中的准确率接近,且均处于较高水平。杏脉悦影肺结节AI初步阅片可以协助并部分取代人工阅片,以提高工作效率,缓解医务人员工作压力。

Objective:To explore the value of clinical research and application of Artificial Intelligence(AI)in the diagnosis of pulmonary nodules.Methods:A total of 2000 patients who underwent pulmonary computed tomography(CT)examination in hospital were selected as the study subjects.Two kinds of methods of reviewing image included manually reviewing image and Xingmai Yueying pulmonary nodule AI reviewing image were adopted.And the clinical application values of two kinds of reviewing image methods in analysis of pulmonary nodules were compared.Results:The number of nodules,nodules type and accuracy were relatively close between Xingmai Yueying pulmonary nodule AI preliminary reviewing and manually reviewing image.And the accuracies of pulmonary imaging report and data system(LungRADS)1-5 classification were 93%,90%,79%,73%and 100%.The area under curve(AUC)values of receiver operating characteristics(ROC)curve of two kinds of reviewing image methods were 0.739 and 0.828 in diagnosing pulmonary malignant nodules,and the AUC both two methods were larger than 0.7,which indicated that the accuracy of Xingmai Yueying pulmonary nodule AI preliminary reviewing image was higher.Conclusion:The accuracy of Xingmai Yueying pulmonary nodule AI preliminary reviewing image is close to that of manual reviewing image,and both of them are in higher level.Xingmai Yueying pulmonary nodule AI preliminary reviewing image can assist and partially replace manual reviewing image so as to improve work efficiency and relieve working pressure of medical personnel.