基于双通道输入深度神经网络的心律失常检测方法研究

Method of Arrhythmia Detection Based on Dual-Channel Deep Input Neural Network

本文对心律失常的自动分类问题进行研究,提出一种基于双通道输入深度神经网络的心律失常检测方法。采用改进的基于小波变换的滤波算法对心电信号进行预处理后,将一个心跳片段和扩展心跳分段,输入卷积神经网络(CNN)与长短时记忆网络(LSTM)串行融合的神经网络,同时提取心跳的局部特征和前后依赖关系,对心跳进行分类;针对数据集不平衡问题,在训练集划分和损失函数中引入加权改进。应用MIT-BIH心律失常数据库,验证模型的有效性,最终准确率99.3%,在心血管疾病的临床辅助诊断应用中有很大的潜力。

This paper studies the automatic classification of arrhythmia,and proposes a method of arrhythmia detection based on dual-channel deep input neural network.After the ECG signal is preprocessed by an improved filtering algorithm based on wavelet transform,a heartbeat segment and an expanded one are input into a neural network that serially combines a convolutional neural network(CNN)and a long short term memory network(LSTM);simultaneously the local features and dependencies of the heartbeats are extracted to classify the heartbeats.Considering the imbalance of data set,the training set division and the loss function are both improved with weights.The MIT-BIH arrhythmia database is used to verify the effectiveness of the model,getting the accuracy rate of 99.3%,which shows that it has great potential in clinical auxiliary diagnosis of cardiovascular diseases.