Atrioventricular nodal reentrant tachycardia (AVNRT) and atrioventricular reentrant tachycardia (AVRT) are two common arrhythmias with high similarity. Automatic electrocardiogram (ECG) detection using machine learning and neural networks has replaced manual detection, but few studies distinguishing AVNRT from AVRT have been reported. This study proposed a classification algorithm using bottleneck attention module (BAM)-based deep residual network (ResNet) through two-lead ECG records. Specifically, ResNet possessed sufficient network depth to extract abundant features, and BAM was introduced to optimize weight assignment of feature maps by fusing together channel and spatial information. Seven types of ECG signals from four public databases were used to pretrain the proposed classification model, which was then fine-tuned using the experimental dataset. The AVNRT and AVRT detection precisions were 98.95% and 87.47%, sensitivities were 87.52% and 98.58%, and the F1-scores were 92.82% and 92.68%, respectively. These findings showed that our proposed classification model achieved excellent inter-patient classification performance and can assist doctors in the diagnosis of AVNRT and AVRT.

中文翻译：

---

使用基于注意力的残差网络对室上性心动过速进行分类

房室结折返性心动过速（AVNRT）和房室折返性心动过速（AVRT）是两种常见的具有高度相似性的心律失常。使用机器学习和神经网络的自动心电图 (ECG) 检测已取代手动检测，但很少有研究将 AVNRT 与 AVRT 区分开来。本研究通过双导联心电图记录提出了一种使用基于瓶颈注意力模块 (BAM) 的深度残差网络 (ResNet) 的分类算法。具体来说，ResNet 拥有足够的网络深度来提取丰富的特征，并引入 BAM 通过融合通道和空间信息来优化特征图的权重分配。来自四个公共数据库的七种心电图信号用于预训练提出的分类模型，然后使用实验数据集对其进行微调。AVNRT 和 AVRT 检测精度分别为 98.95% 和 87.47%，灵敏度分别为 87.52% 和 98.58%，F1 分分别为 92.82% 和 92.68%。这些发现表明，我们提出的分类模型实现了出色的患者间分类性能，可以帮助医生诊断 AVNRT 和 AVRT。