GANs For Electronic Health Record Synthesis

At present,due to the lack of sharing of information between different medical institutions,the confidentiality of medical data,and the small number of cases of a certain disease,it is difficult to obtain a large amount of effective data.The algorithm of machine learning needs a lot of data to support.Only enough data can effectively extract features and improve the generalization ability of the model.In the case of less data,the model of machine learning or deep learning is difficult to give full play to its advantages.In order to obtain enough effective training data,we can use the method of data generation to generate data that conforms to the characteristics of the actual data.Based on this,the disease diagnosis or prediction model is trained,which is of great significance for the application of machine learning or deep learning algorithm in medical field.This article takes the generation of diabetes and heart failure disease data as an example.First,using the generation model to generate data for patients with diabetes and heart failure,this is a new attempt in the field of medical data generation.Second,the data mapped by ICD-9 is extracted and processed into image form.Therefore,some models that are effective in image generation can be used for generation.Thirdly,based on the GAN that performs well in terms of generation,the generation model of this paper is designed by changing the network structure parameters.Fourth,the recently proposed evaluation in GAN is adopted.Generate a good two-sample test of the classifier to evaluate the quality of the model generated in this paper.Fifth,through the specific disease risk prediction experiment to further verify whether the data generated in this paper can improve the accuracy of disease prediction,the experiment proves that the method proposed in this paper Synthetic data can improve the accuracy of disease prediction.