基于卷积注意力的单导联心电图房颤检测方法

Single Lead Electrocardiogram Atrial Fibrillation Detection Method Based on Convolutional Attention

随着可穿戴心电设备的普及,从单导联心电图中自动检测房颤的方法越来越重要。针对可穿戴心电设备采集的单导联心电图中存在噪声干扰的问题,提出一种基于卷积注意力的残差神经网络模型Resnet34-CAB。通过融合卷积注意力块(CAB),在模型复杂度少量增加的情况下,选择性地关注心电图的关键特征,自适应地抑制噪声,提高了模型的检测性能。在公开数据集上的实验结果表明,Resnet34-CAB模型优于Resnet34、Resnet34-Transformer模型,验证了融合CAB的有效性。

With the popularity of wearable electrocardiographic devices,the method of automatically detecting atrial fibrillation from single lead electrocardiograms is becoming increasingly important.A residual neural network model Resnet34-CAB based on convolutional attention is proposed to address the issue of noise interference in single lead electrocardiograms collected by wearable electrocardiographic devices.By integrating Convolutional Attention Blocks(CAB),the detection performance of the model is improved by selectively focusing on key features of the electrocardiogram and adaptively suppressing noise,with a small increase in model complexity.The experimental results on public datasets show that the Resnet34 CAB model outperforms the Resnet34 and Resnet34 Transformer models,verifying the effectiveness of the fusion CAB.