A Deep Transfer Learning Framework for Sleep Stage Classification with Single-Channel EEG Signals

Abstract

The polysomnogram (PSG) is the gold standard for evaluating sleep quality and disorders. Attempts to automate this process have been hampered by the complexity of the PSG signals and heterogeneity among subjects and recording hardwares. Most of the existing methods for automatic sleep stage scoring rely on hand-engineered features that require prior knowledge of sleep analysis. This paper presents an end-to-end deep transfer learning framework for automatic feature extraction and sleep stage scoring based on a single-channel EEG. The proposed framework was evaluated over the three primary signals recommended by the American Academy of Sleep Medicine (C4-M1, F4-M1, O2-M1) from two data sets that have different properties and are recorded with different hardware. Different Time–Frequency (TF) imaging approaches were evaluated to generate TF representations for the 30 s EEG sleep epochs, eliminating the need for complex EEG signal pre-processing or manual feature extraction. Several training and detection scenarios were investigated using transfer learning of convolutional neural networks (CNN) and combined with recurrent neural networks. Generating TF images from continuous wavelet transform along with a deep transfer architecture composed of a pre-trained GoogLeNet CNN followed by a bidirectional long short-term memory (BiLSTM) network showed the best scoring performance among all tested scenarios. Using 20-fold cross-validation applied on the C4-M1 channel, the proposed framework achieved an average per-class accuracy of 91.2%, sensitivity of 77%, specificity of 94.1%, and precision of 75.9%. Our results demonstrate that without changing the model architecture and the training algorithm, our model could be applied to different single-channel EEGs from different data sets. Most importantly, the proposed system receives a single EEG epoch as an input at a time and produces a single corresponding output label, making it suitable for real time monitoring outside sleep labs as well as to help sleep lab specialists arrive at a more accurate diagnoses.

Sleep Stage Classification in Children Using Self-Attention and Gaussian Noise Data Augmentation

“…In many studies [ 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 ], feature engineering algorithms such as Fourier Transform (FT), Wavelet Transform (WT), Spectral Features Analysis (SFA), and Time-frequency Analysis (TA), etc., were used to generate and extract hand-crafted features from PSG recordings. Then various machine learning methods (e.g., Support Vector Machine (SVM), Decision Tree (DT), Adaptive Boosting (Adaboost) and RF, etc.)…”

Multi-task learning for arousal and sleep stage detection using fully convolutional networks

“…This tabulation offers a nuanced understanding of key aspects within these studies, including the chosen architecture, the layer topologies employed (1D vs. 2D), the utilized input format (sequence vs. time-frequency representation (TFR)), the adopted classification scheme (oneto-one, one-to-many, many-to-one, many-to-many), and the incorporation of transfer learning. In terms of architectures, the convolutional neural network (CNN) emerges as the prevailing choice, primarily due to its adeptness in local feature extractionan essential aspect for precise sleep stage detection [12,37,[41][42][43][44][45][46][47][48][49][50][51]. Complementing local feature extraction, the inclusion of global feature extraction, often referred to as sequence modeling, becomes imperative.…”

“…This stems from the realization that sleep stages are influenced not solely by transitory events but are also intricately linked to preceding contexts. In pursuit of sequence modeling, researchers have predominantly turned to Recurrent Neural Networks (RNNs), attention mechanisms, or transformer structures [37,41,45,47,48,[51][52][53].…”

“…Furthermore, the integration of TFRs into the preprocessing pipeline has been explored to advance the comprehension of sleep patterns [41,42,47,49,52,53]. By simultaneously capturing both temporal and frequency information, TFRs hold the potential to improve the detection accuracy of systems.…”

“…However, the integration of TFRs introduces intricacies due to their demanding computational requisites, which can prolong processing times. Additionally, transfer learning has been investigated as an avenue to enhance model training efficacy in scenarios where data availability is constrained [41,49]. Transfer learning leverages the knowledge distilled from preexisting models trained on analogous tasks or datasets.…”