Research On Atrial Fibrillation Prediction Model Based On Convolutional And Recurrent Neural Network

Atrial fibrillation is a burdensome disease.It is of great social and clinical significance to predict patients with atrial fibrillation based on patients’ electronic health record data.Most of the existing prediction models of atrial fibrillation based on patients’ electronic health record data are realized by using traditional feature engineering combined with statistical methods,which have limited processing capacity for highdimensional data such as electronic health record.Some of the latest disease prediction models based on deep learning algorithm have such problems: models based on convolutional neural network(CNN)cannot extract the timing characteristics of patients’ electronic health record data;However,models based on recurrent neural network(RNN)ignored the difference characteristics of each medical variable.In view of this situation,this thesis aims to study a method that can predict the future occurrence of atrial fibrillation using patients’ electronic health record data.The main contributions of the thesis can be summarized as follows:(1)This thesis proposes a prediction model of atrial fibrillation without professional assistance.Since the CNN models have adavantages of capturing the complex relationship,and the RNN models can capture the temporal characteristics,the model combines the advantages of the both models,i.e.a independent CNN structure is adopted to extract the difference characteristics among the medical variables,and meanwhile a RNN structure integrated whith attention mechanism is utilized to capture the temporal characteristics and correlation among the medical varaibles.(2)In order to verify the effectiveness of the model proposed in this study,a series of comparative experiments were designed to compare the model proposed in this thesis with various existing models.First,machine learning models and deep learning models are compared.Then,CNN models is compared with RNN models.Finally,the model proposed in this thesis is compared with some of existing models.Experimental results on real hospital data show that the proposed model outperforms the existing prediction models in accuracy,with an increase of 2.14% in F1 and 1.32% in AUC.