基于B/S架构的新生儿疼痛面容图像标注系统研发

Development of Neonatal Pain Face Image Labeling System based on B/S architecture

目的构建基于浏览器/服务器(browser/server,B/S)架构的新生儿疼痛面容图像标注系统(neonatal pain face image labeling system,NPFILS),评价其质量及效果。方法采集2019年9月至2020年9月在湖南医药学院附属医院产科住院接受足跟采血操作的新生儿疼痛面容视频,基于B/S架构构建NPFILS;20名新生儿科护士先后使用labelImg和NPFILS对新生儿疼痛面容图像进行标注,比较两种标注系统精确率、召回率及系统使用用户体验评分(system usability scale,SUS)。结果NPFILS与LabelImg标注系统的模型混淆矩阵精确率分别为(0.882±0.112 vs.0.853±0.128)、召回率分别为(0.735±0.098 vs.0.741±0.075),差异无统计学意义(P>0.05)。新生儿疼痛面容图像标注系统的用户体验评分明显高于labelImg系统,差异有统计学意义(72.500±3.535 vs.26.667±6.831,P<0.05)。结论NPFILS标注质量高,用户体验评分高于labelImg系统,为新生儿疼痛图像标注工作提供了一种快捷、标准化的新系统。

Objective To construct a neonatal pain face image labeling system(NPFILS)based on B/S architecture and evaluate its quality and effect.Methods Newborns’painful facial videos were collected when they were hospitalized in the obstetrics department of the Affiliated Hospital of Hunan Medical University from September 2019 to September 2020 and underwent heel puncture blood collection.The NPFILS constructed is based on B/S architecture.20 neonatal nurses Labeled 1681 neonatal pain face photoes which were randomly selected both by labelImg system and NPFILS.The annotation accuracy,recall rate and system usability scale(SUS)were compared between the two systems.Results The accuracy of the model confusion matrix of the NPFILS and the labelImg annotation system were(0.882±0.112 vs.0.853±0.128,P>0.05)and the recall rates were(0.735±0.098 vs.0.741±0.075,P>0.05)respectively,and there were no statistical differences between two systems.The user score of SUS of NPFILS was significantly higher than that of labelImg system(72.500±3.535 vs.26.667±6.831,P<0.05).Conclusion NPFILS has high annotation quality and higher user SUS score than of labelImg system,which provides a fast and standardized new system for neonatal pain image annotation.