“互网+”人工智能影像辅助诊断系统在肺结核领域的使用效果分析

Analysis of the application effect of"Internet+"artificial intelligence imaging assisted diagnosis system in the field of tuberculosis

目的通过分析"互联网+"人工智能(AI)影像辅助诊断系统对肺结核的诊断效果,探讨人工智能在基层肺结核诊断领域的实用性。方法收集2020年6-9月山东省4个县区结核病定点医疗机构372例可疑肺结核患者的胸片,利用AI进行诊断,与省级专家组诊断结果进行比对,对基本情况进行描述性分析,利用SPSS 16.0制作受试者工作特征(ROC)曲线、计算曲线下面积(AUC)。结果经专家诊断,确诊为肺结核患者160例(43.01%)。随着AI诊断百分值从>0%升至≥90%,灵敏度从100.00%降至73.75%,阴性预测值从100.00%降至81.25%,差异具有统计学意义(P<0.05),特异度从5.19%升至85.85%,阳性预测值从44.32%升高到79.73%,差异具有统计学意义(P<0.05);ROC曲线下面积(AUC)为0.907(P<0.05);AI诊断百分值≥70%时,诊断正确率最高82.26%。结论AI具有较高的诊断正确率,可辅助临床医生进行筛查、诊断肺结核,适合在基层推广。

Objective By analyzing the diagnostic effect of"Internet+"artificial intelligence imaging assisted diagnosis system on tuberculosis,we discuss the practicability of artificial intelligence in the field of tuberculosis diagnosis at the grassroots level.Methods Chest radiographs of 372 suspected tuberculosis patients from designated TB medical institutions in four counties and districts of Shandong province from June to September 2020 were collected,and diagnosis results by AI were compared with the diagnosis results from provincial expert group.Descriptive analysis was carried out;the receiver operating characteristic(ROC)curve was plotted,and the area under curve(AUC)was calculated with SPSS16.0.Results A total of 160 cases of pulmonary tuberculosis confirmed by experts,accounting for 43.01%.With AI diagnostic score from>0%to≥90%,sensitivity decreased from 100.00%to 73.75%,negative predictive value decreased from 100.00%to 81.25%,the difference was statistically significant(P<0.05),the specificity increased from 5.19%to 85.85%,and the positive predictive value increased from 44.32%to79.73%,with statistical significance(P<0.05).Area under ROC curve(AUC)was 0.907(P<0.05).As AI diagnostic score≥70%,the diagnostic accuracy was 82.26%.Conclusion AI has high diagnostic accuracy,and can assist clinicians in screening and diagnosis of pulmonary tuberculosis,which is suitable for promotion at the grassroots level.