智能医生助理关键技术及应用研究  

Research on Key Technology and Application of Intelligent Doctor Assistant

作　　者：赵侠 贺轲[2] ZHAO Xia;HE Ke(Department of Surgery,Guangdong Workers'Hospital,Guangzhou Guangdong 510000,China;Department of General Surgery,Second People's Hospital of Guangdong Province,Guangzhou Guangdong 510000,China)

机构地区：[1]广东省工人医院外科,广东广州510000 [2]广东省第二人民医院普通外科,广东广州510000

出　　处：《中国卫生标准管理》2022年第4期49-52,共4页China Health Standard Management

基　　金：广州市科技计划项目(202102081210060002)。

摘　　要：目的探讨智能医生助理关键技术及应用研究。方法本院影像科数据采用人工智能深度学习的方法,建立并采用预先发表的数据集进行训练,随后采用测试集进行系统测试,再采用(consensus-based image description evaluation,CIDEr)评分方法对测试集和训练集数据进行评估,于原始数据集中随机抽取2018年1月—2020年1月由本院影像科医生给出报告(专家组)100份由训练完成的系统给出报告(模型组)100份。比较两组生成报告评分;比较两组诊断价值,并绘制受试者工作曲线(ROC)。结果本次评分结果中,5分报告模型组有79份(79.00%),专家组有83份(83.00%),组间比较差异无统计学意义(P>0.05);4分报告模型组有10份(10.00%),专家组有6份(6.00%),组间比较差异无统计学意义(P>0.05);3分报告模型组有6份(6.00%),专家组有5份(5.00%),组间比较差异无统计学意义(P>0.05);2分报告模型组有0份(0),专家组有3份(3.00%),组间比较差异无统计学意义(P>0.05);1分报告模型组有5份(5.00%),专家组有3份(3.00%),组间比较差异无统计学意义(P>0.05);两组生成报告评分平均分值比较差异无统计学意义(P>0.05);专家组报告的灵敏度和特异度分别为90.57%和89.36%,AUC值为0.889,95%CI(0.810~0.943),模型组报告的灵敏度和特异度分别为88.68%和87.23%,AUC值为0.852,95%CI(0.767~0.915)。结论智能医生助理可有效帮助影像科医生自动生成图像报告,进而为临床医生诊疗方案的制定提供更佳的数据支持。Objective To explore the key technology and application of intelligent doctor assistant.Methods Selected the Imaging Department data of our hospital,adopted the method of artificial intelligence deep learning,established and used published data set for training in advance,then test set was adopted to improve the system test,then the(consensus-based image description evaluation,CIDEr)score method was used to evaluate and training sets of test data,from January 2018 to January 2020,100 reports from the radiologists in the imaging department of our hospital(expert group)and 100 reports from the trained system(model group)were randomly selected from the original data set.The two groups were compared to generate report scores;the diagnostic values of the two groups were compared,and receiver operating curves(ROC)were drawn.Results Among the scoring results,there were 79 reports(79.00%)in the model group and 83 reports(83.00%)in the expert group of 5 points,the difference was not significant difference(P>0.05).There were 10 reports in the model group(10.00%)and 6 reports(6.00%)in the expert group of 4 points,the difference was not significant difference(P>0.05).There were 6 reports(6.00%)in the model group and 5 reports(5.00%)in the expert group of 3 points,the difference was not significant difference(P>0.05).There were 0 reports(0)in the model group and 3 reports(3.00%)in the expert group of 2 points,the difference was not significant difference(P>0.05).There were 5 reports(5.00%)in the model group and 3 reports(3.00%)in the expert group of 1 point,the the difference was not significant difference(P>0.05).There was no significant difference between the two groups in the mean score of the generated report score(P>0.05).The sensitivity and specificity reported by the expert group were 90.57%and 89.36%,respectively,with an AUC value of 0.889,95%CI(0.810-0.943),and the sensitivity and specificity reported by the model group were 88.68%and 87.23%,respectively,with an AUC value of 0.852,95%CI(0.767-0.915).Conclusion Intell

关 键 词：智能医生助理 关键技术 应用 CIDEr评分 人工智能 灵敏度 

分 类 号：R445[医药卫生—影像医学与核医学]