A single-channel EEG based automatic sleep stage classification method leveraging deep one-dimensional convolutional neural network and hidden Markov model

A Multilevel Temporal Context Network for Sleep Stage Classification

“…The accuracy of this model is 83,8%, which is 0.4% lower than that of the MLTCN model. Yang et al [ 10 ] utilized HMM to learn the temporal features of long epochs and obtained an accuracy of 83.98%, but did not consider the intra-epoch temporal features, and the performance was lower than that of MLTCN. MLTCN learns the three-level temporal features from intra-epoch, adjacent epochs, and long epochs at the same time.…”

“…ese signals are divided into 30-second epochs, and sleep specialists manually label each epoch according to some standard criteria, such as the American Academy of Sleep Medicine (AASM) rules [2] or Rechtscha en and Kales rules [3]. According to the AASM rules, each epoch is classi ed into one of the ve stages: Wake, REM, N 1 , N 2 , and N [7][8][9][10][11][12][13] and time-frequency images [14,15]. Tsinalis et al [7] used the raw EEG signals to learn features and the relationship between features by two-layer convolutions and pooling.…”

Sleep Staging Based on Multi Scale Dual Attention Network

“…Meanwhile, because of its ability to learn high-dimensional and hierarchical features directly from large amounts of data, it is becoming significant in sleep staging. For example, in [16], Jadhav et al used the wave transform (WT) combined with migration learning to automatically exploit the time-frequency spectrum of EEGs without manual feature extraction, which achieves an accuracy of 83.17%; Afterwards, it combined with multiple classifiers to classify the extracted features, reaching an accuracy of 91.31%; In [17], Tsinalis et al constructed a CNN-based neural network to implement end-to-end learning of raw EEG and finally achieve an average accuracy of 82% and an overall accuracy of 74%; In [18], VGG-16 is used for spectrogramming of EEG and achieves an accuracy of 86%; Yang et al [19] used one-dimensional CNN to automatically extract features from the raw EEG with HMM for classification which receives an accuracy of 83.98%; In [20], Seo et al used a deep CNN-based on a modified ResNet-50 to extract sleep-related features and a two-layer BIL-STM to learn the transition rules between sleep stages to achieve end-to-end learning. Zhang et al proposed a new unsupervised CNN in [21] to automatically extract sleep-related features from physiological signals and achieved an accuracy of 83.4%.…”

“…In the meantime, the calculation methods of the overall accuracy are inconsistent. For example, in [8,19], the average accuracy of each sleep stage is taken as the final accuracy; in [33], the weighted average accuracy of each sleep stage is taken as the final accuracy. In order to make a fair comparison, in Table 7, according to the confusion matrix given by the authors and the calculation methods used in this paper, the overall accuracy and F1 score of the methods listed are recalculated.…”

Competitive Multi-Verse Optimization with Deep Learning Based Sleep Stage Classification

“…In recent times, deep learning (DL) methodology namely recurrent neural network (RNN), convolution neural network (CNN), and other forms of deep neural networks (DNN) have become a common tool in pattern identification in biomedical signal processing. Long short-term memory (LSTM) method that has taken advantageous factor of sequential data learning to advance categorization performance was highly recommended for automatic nap phase [9,10].…”