| 摘 要: 心房颤动(AF)是一种最常见的心律失常类型,为了提高房颤预测的准确率和可靠性,提出了一种基于连续小波变换和残差神经网络的房颤预测方法。首先,采用软阈值小波去噪方法去除心电图信号的噪声干扰;其次,通过连续小波变换生成二维时频图;最后,使用带下采样的残差神经网络进行房颤预测。为了全面评估所提方法的性能,新建立了一个包含2 160条心电图(ECG)记录的综合数据集,并在此数据集上进行了实验。实验结果表明,该方法在新数据集和公开数据集(AFPDB)上分别得到92.4%和96.1%的精确度,相较于当前的深度学习方法,实现了显著提升。 |
| 关键词: 房颤;心电图;连续小波变换;残差网络 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 浙江省自然科学基金项目(LQ22F010006; LTGY23H170004) |
|
| Research on Atrial Fibrillation Prediction Based on Continuous Wavelet Transform and Residual Neural Networks |
|
ZHU Yuxiang1, TONG Jijun1, XIA Shudong2, ZHU Haihang1
|
(1.School of Inf ormation Science and Engineering, Zhejiang Sci-Tech University, Hangzhou 310018, China;
2.The Fourth Af f iliated Hospital Zhejiang University School of Medicine, Yiwu 322000, China)
zhyxang7@163.com; jijuntong@zstu.edu.cn; shystone@126.com; haihangzhu_zstu@163.com
|
| Abstract: Atrial Fibrillation (AF) is the most common type of arrhythmia. To improve the accuracy and reliability of AF prediction, a method based on continuous wavelet transform and residual neural network is proposed. Firstly, a soft-threshold wavelet denoising method is used to remove noise interference from the Electrocardiogram (ECG) signal. Secondly, a two-dimensional time-frequency map is generated through continuous wavelet transform. Finally, a downsampled residual neural network is used for AF prediction. To comprehensively evaluate the performance of the proposed method, a new comprehensive dataset containing 2 160 ECG records has been established, and experiments have been conducted on this dataset. Experimental results show that the method achieves accuracy of 92.4% on the new dataset and 96.1% on the publicly available dataset (AFPDB), respectively, realizing significant improvements compared to current deep learning methods. |
| Keywords: AF; Electrocardiogram; continuous wavelet transform; residual network |
