基于注意力机制与卷积神经网络的不均衡多标签心电图分类方法研究

Attention-Based Convolutional Neural Network for Imbalanced Multi-Label Electrocardiogram

心血管疾病是全球主要死因之一,及早发现心脏病的风险因素对于预防猝死至关重要,基于心电图的心律失常自动检测是心脏病筛查的重要手段,对降低心脏病死亡率至关重要。临床上同一患者可能同时出现多个心电异常类型,因此,解决心电图数据中多标签问题且考虑标签关系显得尤为重要。本文提出了一个多标签心电图分类模型,包括卷积神经网络、Transformer和注意力机制等模块,其中卷积神经网络用于提取局部特征,Transformer用于提取全局特征,采用非对称损失函数平衡正负标签。对两个心电数据集进行实验,实验结果表明,本文提出的模型相较于现有心电图分类方法,表现更优异。

Cardiovascular disease is one of the leading causes of death worldwide.Early detection of risk factors for heart disease is essential to prevent sudden death.The automatic detection of arrhythmia based on electrocardiogram is an important means of heart disease screening,which is crucial to reduce the mortality of heart disease.In clinical settings,patients may exhibit multiple abnormal categories rather than just one.Therefore,it becomes necessary to address the issue of multiple labels in electrocardiogram data and consider the relationships between these labels.We propose a multi-label classification model composed of convolutional neural networks,Transformers and attention mechanisms.The convolutional neural network is used to extract local features,Transfor-mer is used to extract global features,and an asymmetric loss function is used to balance positive and negative labels.Experimental results on two electrocardiogram dataset show that the proposed model has better performance than the existing electrocardiogram classification methods.