基于多尺度融合特征卷积神经网络的房颤自动检测算法
投稿时间:2024-02-28  修订日期:2024-05-10  点此下载全文
引用本文:
摘要点击次数: 23
全文下载次数: 0
作者单位邮编
郭禧斌* 郑州科技学院 450064
王瞧 河南工业大学 
程晓琦 佛山科学技术学院 
基金项目:河南省科技攻关项目(222102310222)资助、河南省教育厅高校重点研究项目(21B460018)资助
中文摘要:心房颤动简称房颤,是目前临床上较为常见的心律失常之一,其发病率和死亡率较高。传统的房颤诊断主要依赖医生目视检查心电图来完成,费时耗力、效率较低。本文提出了一种基于多尺度融合特征的卷积神经网络用于房颤的自动检测,网络中多尺度卷积核可以提升心电信号的特征信息捕获能力,从而克服传统单尺度卷积核的局限性。所提模型在MIT-BIH两个公共数据库中取得了97.4%的准确率、97.3%的敏感度和97.4%的特异度。同时与现有多个算法相比,本文所提的房颤检测算法策略具有更佳的模型精度和鲁棒性。
中文关键词:心房颤动  心电信号  多尺度卷积核  卷积神经网络
 
Automatic Detection Algorithm of Atrial Fibrillation Based on Multi-scale Fusion Feature Convolution Neural Network
Abstract:Atrial fibrillation(AF)is the most common arrhythmia in clinic at present, with high incidence rate and mortality.The traditional diagnosis of atrial fibrillation mainly depends on visual examination of ECG,which is time-consuming,labor-consuming and inefficient.In this work,we present a convolution neural network (CNN) based on multi-scale fusion features for automatic detection of AF.Multi-scale convolution kernel (MCK) in this network can improve the feature information capture ability of ECG signals,and overcome the limitation of traditional single-scale convolution kernel.The proposed model achieves an accuracy of 97.4%,sensitivity of 97.3% and specificity of 97.4% on the two public databases of MIT-BIH.Compared with several existing algorithms,the AF detection algorithm proposed in this paper shows better model accuracy and robustness.
keywords:atrial fibrillation  ECG  multi-scale convolution kernel  convolution neural network
查看全文   查看/发表评论   下载pdf阅读器