基于一维卷积神经网络的心肌梗死诊断研究

Research on the Diagnosis of Myocardial Infarction Based on One-dimensional Convolutional Neural Network

目的:心电图判读主要依靠医师经验性分析,容易出现差错,本研究提出一维卷积神经网络预测模型实现对心肌梗死的智能诊断。方法:本研究利用Keras及Tensorflow搭建一维卷积神经网络。利用PTB-XL公共数据平台上的21 837条ECG数据,经过删除异常值、中值滤波、裁剪等数据预处理后,将数据划分为训练集、验证集及测试集,用训练集数据训练CNN模型,用验证集验证模型,并采取正则化、Dropout及早停止法策略抑制过拟合,得到最终的预测模型。结果:模型对测试集预测的AUC、准确率、敏感度、特异度及阳性预测值分别达到0.981、0.952、0.932、0.957、0.840。结论:该预测模型对心肌梗死的预测性能优异,为心肌梗死的诊断提供了新方法。

Objective:ECG interpretation mainly depends on the empirical analysis of doctors,which is prone to errors.In this Study,a one-dimensional convolutional neural network was proposed to realize the intelligent diagnosis of myocardial infarction.Methods:One-dimensional convolutional neural network was constructed by using Keras and Tensorflow.Using 21837 ECG data on the PTB-XL public data platform,after data preprocessing such as removing outliers,median filtering,clipping,the data was divided into training set,validation set and test set.The model was trained with training set,and verified by validation data.And regularization,Dropout and early stopping strategy were adopted to suppress over fitting.Then the final prediction model was obtained.Results:The AUC,accuracy,sensitivity,specificity and positive predictive value predicted by the model on the test set reached to 0.981,0.952,0.932,0.957,0.840.Conclusion:The prediction model has excellent performance in predicting myocardial infarction and provides a new method for the diagnosis of myocardial infarction.