摘要
目的 观察人工智能(AI)知识图谱和图像分类对胸部后前位X线片(简称胸片)质量控制(QC)的价值.方法 回顾性分析安徽省影像云平台中595家医疗机构共9 236幅胸片,构建包含21个分类标签的QC知识图谱.先由10名技师据此对胸片进行2轮单人QC和1轮多人QC,分别将结果记为A、B、C;再以AI算法进行分类评估,将结果记为D.最后由1名QC专家对C、D进行审核并确定最终QC结果,以之为参考评估上述4种QC效果.结果 AI算法用于胸片QC的曲线下面积(AUC)均≥0.780,平均AUC为0.939.A、B、C、D胸片QC的平均精确率分别为81.15％、85.47％、91.65％、92.21％.结论 AI知识图谱和图像分类技术可有效用于胸部后前位X线片QC.
Abstract
Objective To observe the value of artificial intelligence(AI)knowledge graph and image classification for quality control(QC)of chest posterior-anterior position X-ray radiograph(abbreviated as chest film).Methods Totally 9 236 chest films from 595 medical institutions in Anhui province imaging cloud platform were retrospectively enrolled.QC knowledge graph containing 21 classification labels were constructed.Firstly,QC of chest films based on the above knowledge graph were performed by 10 technicians for 2 rounds of single person and 1 round of multi person,and the results were recorded as A,B and C,respectively.Then AI algorithms were used to classify and evaluate based on knowledge graph,and the result was recorded as D.Finally,a QC expert reviewed results C and D to determine the final QC results and taken those as references to analyze the efficiency of the above 4 QC.Results The area under the curve(AUC)of AI algorithm for QC of chest films were all ≥0.780,with an average value of 0.939.The average precision of QC for chest films of A,B,C and D was 81.15％,85.47％,91.65％and 92.21％,respectively.Conclusion AI knowledge graph and image classification technology could be effectively used for QC of chest posterior-anterior position X-ray radiograph.
基金项目
安徽高校协同创新项目(GXXT-2022-031)
NETL
NSTL
万方数据