基于卷积神经网络的心律失常分类研究

Classification of Arrhythmia Based on Convolutional Neural Networks

心律失常是引起心肌梗塞、突发性心脏死亡等严重疾病的重要原因,常借助心电图进行早期诊断。然而,传统的心电信号分类方法有着复杂的特征提取任务,计算量大、费时费力。为此通过在经典卷积神经网络的基础上加入Dropout层,设计了一种改进的卷积神经网络模型,该模型可以进一步提升模型的泛化能力、提高准确率。其利用CNN自动提取特征,将经过预处理的心电信号直接作为模型的输入,自动识别5种不同类型的心拍。在MIT-BIH心律不齐数据库上进行实验,准确率达到99.68%,特异性为98.94%,灵敏度为99.76%。实验结果表明,相较于经典卷积神经网络,本文提出的方法能够精确、高效地识别不同类型的心律失常疾病。

Arrhythmia is a significant cause of serious conditions such as myocardial infarction and sudden cardiac death,and is often diagnosed early using electrocardiogram.However,the traditional ECG signal classification method has complicated task of feature extraction,which is time-consuming and laborious.To address this,we design an improved Convolutional Neural Network(CNN)model by incorporating Dropout layers into the classic CNN framework.This enhancement aims to improve the model's generalization ability and increase accuracy.By using CNN to automatically extract features,the preprocessed ECG signal is directly used as the input of the model to automatically identify 5 different types of heart beats.Experiments conducted on the MIT-BIH Arrhythmia Database achieve the accuracy of 99.68%,the specificity of 98.94%,and the sensitivity of 99.76%.The experimental results show that compared with the classic CNN,the proposed method can accurately and efficiently identify different types of arrhythmia diseases.