【摘要】 为解决当前深度学习模型进行心房颤动检测泛化能力差的问题，提出了一种基于长时RR间期的心房颤动检测算法。基于心电信号的一维时序特性以及心房颤动的特殊RR间期，设计了CNN与LSTM结合的深度学习模型，深度挖掘长时RR序列的时间与空间特征，使得它能够在未知数据集上取得良好的结果。使用MIT-BIH心房颤动数据集的全部可用样本划分训练、验证与盲法测试集（3名个体）。通过10倍交叉验证后在盲法测试集上的准确率为99.11%、敏感性为98.86%、特异性为99.47%、阳性预测率为99.62%、F1分数为99.24%。模型与现有方法进行了对比，证实所提模型用于心房颤动检测的可行性，能够有效识别出未知数据集的心房颤动病例，泛化能力强。更多还原

【Abstract】 An atrial fibrillation detection algorithm based on long-time RR interval is proposed to solve the problem of the poor generalization ability of current deep learning models for atrial fibrillation detection. Based on the one-dimensional timing characteristics of electrocardiogram signals and the specific RR interval of atrial fibrillation, a deep learning model combining convolutional neural network and long short-term memory network is designed to deeply mine the temporal and spatial characteristics of long-term RR sequences, which enables the model to perform well on unknown data sets. All available samples of the MIT-BIH atrial fibrillation data set are divided into training, validation, and blindfold detection test sets(3 subjects). After 10-fold cross-validation, the accuracy, sensitivity, specificity, positive prediction rate, and F1score on the blind detection test set are 99.11%, 98.86%, 99.47%, 99.62%, and 99.24%, respectively. The comparison with the existing methods confirms the feasibility of the proposed model for atrial fibrillation detection. The proposed method can effectively identify atrial fibrillation cases from unknown data sets and has a high generalization ability.更多还原