图像识别在凝血试验自动审核方案中的应用  

Application of image recognition in automatic review scheme of coagulation test

作　　者：董争华 佐玉琴 赵晓明[1] 纪凌云 杨冀[1] DONG Zhenghua;ZUO Yuqin;ZHAO Xiaoming;JI Lingyun;YANG Ji(Department of Clinical Laboratory,South Branch of East Hospital Affiliated to Tongji University,Shanghai 200123,China;Department of Clinical Laboratory,Shanghai Pudong New Area Traditional Chinese Medicine Hospital,Shanghai 201205,China)

机构地区：[1]同济大学附属东方医院南院检验科,上海200123 [2]上海市浦东新区中医医院检验科,上海201205

出　　处：《国际检验医学杂志》2024年第11期1368-1374,共7页International Journal of Laboratory Medicine

摘　　要：目的建立带有图像识别功能的凝血试验自动审核方案,评价方案的正确性和有效性。方法使用人工智能软件与硬件联合建立对标本性状、标本血量、血细胞比容可以自动判定结果的图像识别系统,与目测法比较标本性状判定结果的正确性,与手工测量法比较标本血量判定结果的正确性,与血细胞分析仪比较血细胞比容判定结果的正确性。根据流程图、参考区间、医学决定水平、危急值范围、相关文献、工作经验及历史数据制订出凝血试验自动审核规则。对制订的规则进行人工验证,分别计算自动审核通过率、真阳性率、真阴性率、假阳性率、假阴性率。评估自动审核方案实施后实验室内周转时间的变化情况。结果图像识别系统标本性状判定正确率为96.72%,将溶血、黄疸、脂血标本判定为正常标本的假阴性率为0.04%。图像识别系统与手工测量的两组标本血量数据进行比较,P=0.4881,图像识别方法不劣于手工测量方法。图像识别系统与血细胞分析仪的两组血细胞比容数据进行比较,P=0.1130,图像识别系统不劣于血细胞分析仪。该研究建立凝血试验自动审核规则61条,包括数值异常、逻辑异常、Delta Check、标本质量异常、反应曲线异常等,自动审核通过率为76.19%,真阳性率为23.77%,真阴性率为76.19%,假阳性率为0.04%,假阴性率为0.00%。实施自动审核方案后各分位数标本周转时间均缩短,平均缩短13.66 min。结论图像识别技术应用到凝血试验自动审核中,使自动审核功能更加自动化、更具科学性,将标本质量判定标准化,提升检验结果的准确性,有效提高工作效率和节省人力。Objective To establish an automatic review plan for coagulation tests with image recognition function,and evaluate the correctness and effectiveness of the plan.Methods Artificial intelligence software and hardware were combined to establish an image recognition system that could automatically determine the characteristics of specimens,blood volume and hematocrit.The correctness of the determination results of specimen character was compared with the visual method,the correctness of the determination results of blood volume was compared with the manual measurement method,and the correctness of hematocrit was compared with the hematology analyzer.According to the flow chart,reference interval,medical decision level,critical value range,relevant literature,work experience and historical data,the autoverification rules of coagulation tests were formulated.The autoverification rules were manually verified,and the autoverification pass rate,true positive rate,true negative rate,false positive rate,and false negative rate were calculated.The change of turnaround time in the laboratory after the implementation of the autoverification scheme was evaluated.Results The accuracy rate of sample trait determination in the image recognition system was 96.72%,and the false negative rate of judging hemolytic,jaundice,and lipoid blood samples as normal samples was 0.04%.The image recognition system was compared with the blood volume data of two groups of specimens measured manually,P=0.4881.The image recognition method was not inferior to the manual measurement method.Comparing the two sets of hematocrit data from the image recognition system and the blood cell analyzer,P=0.1130,the image recognition system was not inferior to the blood cell analyzer.A total of 61 automatic review rules for coagulation tests had been established,including numerical abnormalities,logical abnormalities,Delta Check,sample quality abnormalities,reaction curve abnormalities,etc.The automatic review pass rate was 76.19%,true positive rate was 23.77%,tr

关 键 词：人工智能 图像识别 凝血试验 自动审核 实验室信息系统 

分 类 号：R446.9[医药卫生—诊断学]