胸部正位片人工智能自动质控系统模型的研究与应用  

Research and Application of Artificial Intelligence Automatic Auality Control System Model for Chest Orthography

作　　者：胡君花 辛小燕 唐堂 于芷轩 胡安宁 HU Junhua;XIN Xiaoyan;TANG Tang;YU Zhixuan;HU Anning(Department of Radiology,Nanjing Drum Tower Hospital,The Affiliated Hospital of Nanjing University Medical School,Nanjing Jiangsu 210008,China)

机构地区：[1]南京大学医学院附属南京鼓楼医院医学影像科,江苏南京210008

出　　处：《中国医疗设备》2023年第12期63-68,共6页China Medical Devices

基　　金：南京大学医院管理研究所课题项目;南京鼓楼医院医学发展医疗救助基金会资助项目(NDYG2020025)。

摘　　要：目的 基于人工智能图像识别技术在医学影像领域的应用,构建一套人工智能自动质控系统。方法 通过U-Net框架,VGG16架构完成对胸部数字X线摄影术(Digital Radiography,DR)正位胸片的语义分析,结合制定的人工智能评片质控标准,构建人工智能自动质控系统,并进行应用检验。结果 U-Net的预测结果与手动标注的区域相似度极高(肺野96.73%,肩胛骨98.02%,锁骨98.71%);VGG16架构对异物的分辨能力最终测试精度为87.58%。在集中质控评价300份数据中,质量管理小组一级片232份、二级片62份、三级片6份、废片0份,人工智能质控系统一级片228份、二级片67份、三级片5份、废片0份,机器质控与人工质控的一致性较强(Kappa=0.901,P<0.001)。结论 胸部正位片人工智能自动质控系统模型可以高效、准确、客观地实现影像的自动智能质控,前端质控提高了拍片质量;集中质控可以批量完成图像质控工作,节约人员成本。Objective To construct an artificial intelligence automatic quality control system based on the application of artificial intelligence image recognition technology in medical image field.Methods Using U-Net framework and VGG16 architecture to complete the semantic analysis of digital radiography(DR),combined with the established artificial intelligence film evaluation quality control standards,artificial intelligence automatic quality control system was constructed,and the application test was carried out.Results The prediction results of U-Net were extremely similar to those of manually labeled areas(96.73%of lung field,98.02%of scapula,98.71%of clavicle).The final test accuracy of VGG16 architecture for foreign object resolution was 87.58%.In the 300 data of centralized quality control evaluation,quality management team(232 of first-level tablets,62 of second-level tablets,6 of third-level tablets and 0 of rejected tablets),artificial intelligence quality control system(228 of first-level tablets,67 of second-level tablets,5 of third-level tablets and 0 of rejected tablets),and according to the score results,the coefficient of consistency between machine quality control and manual quality control showed strong consistency(Kappa=0.901,P<0.001).Conclusion The artificial intelligence automatic quality control system model of chest position-image can realize automatic intelligent quality control of image efficiently,accurately and objectively,and the front-end quality control improves the quality of the shot.And centralized quality control can complete image quality control work in batches and save personnel cost.

关 键 词：人工智能 胸部正位片 数字化摄影 质量控制 

分 类 号：R197.39[医药卫生—卫生事业管理] R144[医药卫生—公共卫生与预防医学]