基于原始单通道脑电图的高效睡眠自动分期方法

Efficient Automatic Sleep Staging Method Based on Original Single-channel Electroencephalogram

提出一种基于单通道脑电数据的睡眠自动分期方法。利用多个并行的卷积操作学习脑电的多尺度空间特征,使用长短期记忆网络挖掘局部时不变特征中的时间信息。针对类别不平衡问题,采用时移滚动方法和加权交叉熵损失函数。在公开数据集Sleep-EDF上的实验结果表明,所提方法仅使用单通道数据实现了端到端的高效睡眠自动分期,缓解了不平衡数据集的分类问题。

An automatic sleep staging method based on the single-channel electroencephalogram data was proposed.Multiple parallel convolutional operations were employed to learn electroencephalogram multi-scale spatial features,and long-short term memory networks were used to mine temporal information in local time-invariant features.To treat the class imbalance problem,a time-shifted rolling method and a weighted cross-entropy loss function were applied.The results on the public dataset Sleep-EDF showed that the proposed method achieved efficient end-to-end automatic sleep staging using only single-channel data and alleviated the classification problem of unbalanced datasets.