Research On Intelligent Diagnosis Based On Imbalanced Electronic Medical Record Dataset

Electronic medical records are the most detailed and direct record form of clinical medical activities,and it is the data basis of intelligent diagnosis.The diagnosis results of electronic medical records are multi-disease,including normal diagnosis,pathological diagnosis and complications,so intelligent diagnosis can be treated as multi-label classification problem.The development of medical informatization has accumulated a large amount of medical data.Intelligent diagnosis based on electronic medical records can significantly improve the efficiency and accuracy of diagnosis,and alleviate the contradiction between supply and demand of medical resources.The diagnostic results in real electronic medical records data are imbalanced distributed,with rare diseases occurring far less frequently than common diseases.Taking the obstetric electronic medical records dataset as the entry point,this paper discusses the improvement of intelligent diagnosis performance based on the imbalanced electronic medical records dataset,and mainly completes the following works:(1)For high coupling diagnosis results in dataset,Double Decoupled Network(DDN)based intelligent diagnosis algorithm is proposed.DDN firstly decouples representation learning and classifier learning.In the representation learning stage,CNN is used to learn the original features of the data and then the learning parameters are fixed.In the classifier learning stage,a decoupled rebalancing algorithm is proposed to decouple the highly coupled diagnostic results and rebalance the dataset,and then the balanced dataset is used to train the classifier.Experimental results on the obstetric electronic medical records data set show that for the highly coupled diagnosis results,compared with the model without dual decoupled,the accuracy of DDN is increased by 4.36% to 84.17%.By analyzing the accuracy increment of highly coupled diagnosis results,DDN can effectively increase the number of low-frequency samples,learn good feature representation,and improve the diagnostic performance of low-frequency diseases.(2)For all diagnosis results in dataset,Incorporating Label Semantics(ILS)based intelligent diagnosis algorithm is proposed.Firstly,BERT model is used to encode the input electronic medical records to obtain word representation.An adjustive attention mechanism is proposed to model the semantic relationship between labels and documents,and to obtain a label-specific word representation.The combination of the BERT encoded word representation and the label-specific word representation is input into the BI-LSTM document encoder for disease diagnosis.The experimental results in the obstetric electronic medical records dataset show that,compared with the BERT model,the diagnostic precision is increased by2.15% to 86.16% for all the labels in the dataset.By analyzing the n DCG@K value of the diagnosis results of low-frequency diseases,ILS can make full use of label semantics to supplement the information of low-frequency diseases and improve the diagnostic performance of low-frequency diseases.An intelligent diagnosis system based on ILS called Xiao Yi is developed to realize good human-computer interaction and assist doctors in making decisions...