基于谱带融合的EEG自动睡眠分期算法

EEG Automatic Sleep Stage Staging Algorithm Based on Spectral Band Fusion

由于当前数据集类别不均衡造成基于深度学习的睡眠阶段分期模型性能较差。论文提出了一种基于谱带融合数据增强的自动睡眠分期算法(EEGFusionNet,EFNet),以解决上述问题提高模型性能。该算法首先对样本数据进行谱带融合的数据增强处理,通过由卷积神经网络与双向门控循环单元提取EEG中的时频特征与时序特征,将特征融合进行睡眠阶段分期。将Sleep EDF数据集中153名健康受试者的EEG信号作为本模型的样本数据进行训练,得到模型的准确率约为84.4%,kappa值为83.8%。与传统的基线模型相比,论文提出的基于谱带融合的睡眠分期模型在准确性和一致性方面具有显著提升。

Due to the unbalanced data set,the performance of sleep stage staging models based on deep learning is poor.In this paper,an automatic sleep staging algorithm based on spectral band fusion data enhancement is proposed to solve the above prob-lems and improve the performance of the model.In this algorithm,the time-frequency characteristics of EEG signals are extracted by convolutional neural network and bidirectional gated cyclic unit,and then the sleep stage is segmented.The EEG signals of 153 healthy subjects in the sleep EDF dataset are trained as the sample data of this model.The accuracy of this model is about 85.3%,and the Kappa value is 0.78.Compared with the traditional baseline model,the proposed sleep staging model based on band fusion has significantly improved accuracy and consistency.