摘要
有数据显示,每年约有1500万人因房颤住院治疗,及时诊断可以有效地降低患病风险。本文提出一种改进CNN和LSTM的深度学习模型,在CNN模块使用了Swish函数作为激活函数,并在LSTM中引入了点积注意力机制,用于心电信号的特征提取和心律失常的诊断。利用卷积神经网络提取心电信号中的特征,LSTM可以对心电信号中的特征进行深度的挖掘,引入注意力机制,完成房颤的识别。所提模型的准确率为0.9771,F1为0.9609,精确率为0.9765,召回率为0.9799。
Data show that about 15 million people are hospitalized due to atrial fibrillation every year.Timely diagnosis can effectively reduce the risk of disease.In this paper,we propose a deep learning model with improved CNN and LSTM.The Swish function is used as the activation function in the CNN module,and the dot product attention mechanism is introduced in the LSTM for feature extraction of ECG signals and diagnosis of arrhythmia.CNN is used to extract features in ECG signals,and LSTM can deeply mine the features in ECG signals.Attention mechanism is introduced to complete the recognition of atrial fibrillation.The accuracy rate of the proposed model is 0.9771,the F1 is 0.9609,the precision rate is 0.9765,and the recall rate is 0.9799.
作者
王锐
周作建
李灿
李红岩
郎许锋
宋懿花
Wang Rui;Zhou Zuojian;Li Can;Li Hongyan;Lang Xufeng;Song Yihua(School of Artificial Intelligence and Information Technology,Nanjing University of Traditional Chinese Medicine,Nanjing,Jiangsu 210023,China)
出处
《计算机时代》
2023年第8期69-73,共5页
Computer Era
关键词
房颤识别
卷积神经网络
长短期记忆网络
注意力机制
atrial fibrillation recognition
convolutional neural network(CNN)
long short-term memory network(LSTM)
attention mechanism