Abstract：Objective To detect paroxysmal atrial fibrillation (PAF) accurately, efficiently and rapidly from dynamic electrocardiograms
(ECG) and reduce the morbidity and mortality of patients, a hybrid model for paroxysmal atrial fibrillation detection by convolutional
temporal neural network bases on attention mechanism is designed. Methods Firstly, the algorithm extracted ECG signal features
by the convolutional neural network. Then, the attention mechanism was used to help the network focus on the key information, and
finally inputted bidirectional gated recurrent unit for contextual information linkage to detect PAF more accurately. Results The
model was pre-trained using the Physionet 2021 database, and then migration learning was performed on the CPSC2021 database.
The sensitivity, specificity and accuracy of this model were 96.86%, 98.56%, 98.54%, respectively. Conclusion This algorithm is
effective in detecting PAF compared with other algorithms and has potential clinical application value.
刘风雅，余睿，宾光宇，周著黄，吴水才. 基于注意力机制的卷积时序神经网络检测阵发性房颤模型[J]. 中国医疗设备, 2023, 38(3): 1-7.
LIU Fengya, YU Rui, BIN Guangyu, ZHOU Zhuhuang, WU Shuicai. Model of Attention-Based Convolutional Temporal Neural Network for Detection of
Paroxysmal Atrial Fibrillation. China Medical Devices, 2023, 38(3): 1-7.
