摘要
目的:构建一种有效的心电图分类算法,以实现对正常心电图与异常心电图的准确、可靠识别。方法:对来源于PhysioNet开源数据库的基于.dat文件的心电图信号进行预处理及扩充,构建卷积神经网络模型并优化批尺寸(batchsize)、正则化参数(l2_regularizer)、学习率(learningrate)、丢弃值(dropout)、训练步数(epoch)5个超参数,并进行算法性能评价实验。选用的性能评价指标包括准确率、灵敏度和整体指标F1值。结果:经过研究表明,提出的基于.dat文件的心电图正异常识别算法准确率达90%,灵敏度为89.7%,F1值为90.4%。结论:构建的基于心电图信号的分类识别算法能够高效、可靠地识别心电图,具有潜在的临床应用价值。
Objective To construct an effective ECG classification algorithm to achieve accurate and reliable identification of normal ECG and abnormal ECG images. Methods The.dat-files-based ECG signals based on PhysioNet open source database were pre-processed and expanded. A convolutional neural network model was constructed, and its five hyperparameters were optimized, such as batchsize, regularizer, learning rate learningrate, dropout and training steps epoch. Algorithm performance evaluation experiments were carried out. The selected performance evaluation metrics included accuracy,sensitivity and the overall metric F1 value. Results The proposed.dat-file-based ECG identification algorithm had an accuracy of 90%, a sensitivity of 89.7% and an F1 value of 90.4%. Conclusion The constructed algorithm can identify ECG signals efficiently and reliably.
作者
马晶
李林献
邱筱岷
王正杰
王小花
MA Jing;LI Lin-xian;QIU Xiao-min;WANG Zheng-jie;WANG Xiao-hua(Wuxi Maternity and Child Health Care Hospital,Wuxi 214002,Jiangsu Province,China)
出处
《医疗卫生装备》
CAS
2020年第12期31-34,43,共5页
Chinese Medical Equipment Journal
关键词
卷积神经网络
心电图分类
超参数优化
数据预处理
数据扩充
convolutional neural network
ECG classification
hyperparameter optimization
data preprocessing
data expansion
作者简介
马晶(1994-),女,硕士,主要从事医疗器械与人工智能方面的研究工作,E-mail:1183794950@qq.com;通信作者:李林献,E-mail:llxtlp@sina.com。
