基于CNN和双向LSTM的房颤预测模型

PREDICTION MODEL OF ATRIAL FIBRILLATION BASED ON CNN AND BIDIRECTIONAL LSTM

现有基于CNN的模型无法提取患者数据中的时序特征,而基于RNN的模型忽略了各医学变量的差异性特征。针对这种情况,提出一种结合CNN和RNN的房颤预测模型,利用一个独立CNN模块捕获电子病历数据中各医学变量间的差异性特征,同时使用一个独立的RNN模块捕获电子病历数据中时序性特征以及各医学变量间的相关性特征。在真实医院数据集上的实验结果表明,与最新的一些基于电子病历数据的疾病预测方法相比,该模型在房颤的预测方面表现得更加突出,F1值提高了2.14%,AUC值提高了1.32%。

The existing model based on CNN cannot extract the temporal characteristics from patient data,while the models based on recurrent neural network ignore the different characteristics of various medical variables.To solve these problems,a predictive model of atrial fibrillation(AF)combined with CNN and RNN is proposed.This model used an independent CNN module to capture the different characteristics among medical variables in the electronic health records(EHR)data.At the same time,an independent RNN module was used to capture the temporal characteristics and correlation characteristics among medical variables in the EHR data.Experimental results on real hospital data sets show that compared with some of the latest disease prediction methods based on EHR data,the model performs better in predicting AF,with an increase of 2.14%in F1 and 1.32%in AUC.