摘要
目的利用患者历史对比数据,建立基于机器学习中支持向量机(support vector machine,SVM)算法识别临床混淆样本的方法,并验证该方法的临床有效性。方法收集约45万例血常规检测结果,经过数据清洗过滤,同一患者只保留2次检验结果并计算差值校验的绝对值。用同一患者与不同患者的差值(delta)分别制作正配样本与错配样本。采用SVM分类算法实现用多项目识别2种样本,与基于单项目识别的参考变化值法(reference change value,RCV)作比较。结果SVM算法识别正配和错配2种样本,精确率达到92.21%。基于统计学的RCV方法在不同项目下效果不同,其中红细胞平均血红蛋白量(MCH)在绝对值delta下的准确率最高,为81.51%。结论基于多项目的SVM算法可识别混淆样本,可在结果审核中应用。
Objective To establish a method for identifying sample mix-up based on support vector machine(SVM)algorithm in machine learning using patients′history data,and verify its clinical effectiveness.Methods About 450000 blood routine test results were collected.After cleaning and filtering data,only two test results were kept for the same patient and the absolute value of delta check was calculated.The delta data from same patients and different patients were used to generate matched and mismatched samples,respectively.The SVM classification algorithm was used to detect mix-up samples with the multi-items,and was compared with the reference change value(RCV)method only using the single-item.Results The accuracy rate of SVM algorithm identified matched and mismatched samples was 92.21%.However,the RCV method based on statistics had different effects in various items,and the highest accuracy rate was 81.51%,which was found in the item of mean corpuscular hemoglobin(MCH)with the absolute delta value.Conclusion The SVM algorithm based on the multi-items can effectively identify sample mix-up and be applied in the result auto-verification.
作者
张金杰
梁玉芳
王清涛
王哲
冯祥
韩泽文
宋彪
高志琪
周睿
ZHANG Jinjie;LIANG Yufang;WANG Qingtao;WANG Zhe;FENG Xiang;HAN Zewen;SONG Biao;GAO Zhiqi;ZHOU Rui(Department of Laboratory Medicine,Beijing Chaoyang Hospital,Capital Medical University,Beijing 100020;Beijing Center for Clinical Laboratories,Beijing 100020;Inner Mongolia Wesure Data Technology Co.,Ltd,Hohhot 010000,Inner Mongolia;Inner Mongolia University of Finance and Economics,Hohhot 010051,Inner Mongolia,China)
出处
《临床检验杂志》
CAS
2021年第12期945-949,共5页
Chinese Journal of Clinical Laboratory Science
基金
北京市临床重点专科卓越项目(检验科)。
关键词
混淆样本
差值校验
支持向量机
sample mix-up
delta check
support vector machine
