基于半监督学习的医嘱辅助决策研究

Research on Assistant Decision Making of Medical OrdersBased on Semi-supervised Learning

针对常规数据挖掘方式难以从海量电子病历中有效挖掘出病历潜在价值的问题,根据电子病例数据特点,结合半监督学习方法,采用半监督k-means模型优化患者相似组划分,并综合考虑患者相似性及医嘱类别采用基于半监督LP算法确定最佳医嘱选择。通过对某三甲医院白血病电子病历进行医嘱辅助决策实验,对提出的基于半监督学习的医嘱辅助决策方法进行验证。结果表明,所提出的医嘱辅助决策方法可利用已有的医疗数据资源辅助医护人员诊疗,且其准确率优于传统基于患者相似性度量的医嘱选择方法,有助于医护人员开具准确的医嘱。

Due to the time-consuming of data preprocessing and annotation caused by various forms of electronic cases,it is difficult to effectively mine the potential problems of electronic cases.According to the characteristics of electronic case data,combined with semi-supervised learning method,this paper studies the assistant decision-making of medical orders from two aspects of patients’similar groups and medical order scheme selection.By constructing a semi-supervised k-means model with pairwise constraints,the division of similar groups is optimized.By comprehensively considering the influence of patient similarity and order category on the decision-making of doctor's order,and based on semi-supervised LP algorithm,the best choice of doctor's order is determined.Taking the leukemia electronic medical record of a 3A hospital as an example,this paper verifies the proposed method.The results show that the proposed method can make use of the existing medical data resources to assist medical staff in diagnosis and treatment,and its accuracy is better than the traditional method based on patient similarity measurement,which is helpful for medical staff to issue accurate orders.