Sleep Staging Method Research Articles

Automatic sleep staging method of EEG signal based on transfer learning and fusion network

Automatic sleep staging technology is a current research hotspot in brain-computer interfaces, freeing sleep specialists from the time-consuming task of manually diagnosing sleep. It performs complex sleep staging tasks by characterising sleep waves. Current research cannot fully interpret the meaning of EEG signals due to drawbacks such as high signal-to-noise ratio and insufficiency of EEG signals, and weak interpretability of depth models. In this research, we proposed an automatic sleep staging network of EEG signal based on transfer learning and integration of single-channel and multi-channel features. In Epoch processing block (EPB) stage and sequence processing block (SPB) stage, the frequency information and long-term features were extracted from the raw EEG and time-frequency data. Beside that, the transfer learning strategy was still adopted to overcome the data-variability and data-inefficiency issues and enable transferring knowledge from a large dataset to a small cohort. And then, the learning results of the two neural networks were fused adaptively and classified by using LightGBM. Lastly, we used the public dataset sleep-edf expanded to evaluate the performance of the system. The overall highest accuracy rate obtained on the sleep-edf expanded sleep-cassette (sc) subset was 87.84%, on the basis of not relying on massive training data, the accuracy rate is 1.57%-6.73% higher than other methods in the same experimental environment.

Multi-Modal Physiological Signals Based Squeeze-and-Excitation Network With Domain Adversarial Learning for Sleep Staging

Sleep staging is the basis of sleep medicine for diagnosing psychiatric and neurodegenerative diseases. However, the existing sleep staging methods ignore the fact that multi-modal physiological signals are heterogeneous, and different modalities contribute to sleep staging with distinct impacts on specific stages. Therefore, how to model the heterogeneity of multi-modal signals and adaptively utilize the multi-modal signals for sleep staging remains challenging. Moreover, existing methods suffer from the individual variance of physiological signals. How to generalize the sleep staging model for the variance among subjects is also challenging. To address the above challenges, we propose the multi-modal physiological signals based Squeeze-and-Excitation Network with Domain Adversarial Learning (SEN-DAL) to capture the features of electroencephalogram (EEG) and electrooculogram (EOG) for sleep staging. The SEN-DAL is made up of two independent feature extraction networks for modeling the heterogeneity, a Multi-modal Squeeze-and-Excitation feature fusion module for adaptively utilizing the multi-modal signals, and a Domain Adversarial Learning module to extract subject-invariant sleep feature. Experiments demonstrate that the SEN-DAL is superior to the baseline models on a public sleep staging dataset, reaching an F1-Score of 82.1%, which is similar to human experts. Through the ablation experiments, we found that the proposed mechanisms, including modal-independent feature extraction, adaptive utilization of multi-modal signals, and domain adversarial learning, are effective for sleep staging tasks. The code of SEN-DAL is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/xiyangcai/SEN-DAL</uri> .

DSSNet: A Deep Sequential Sleep Network for Self-Supervised Representation Learning Based on Single-Channel EEG

Sleep staging by highly trained specialists is laborious. Although automatic sleep staging with supervised learning methods has been implemented for almost a decade, it requires lots of manually annotated data. Self-supervised learning methods have recently been gaining attention. They can learn representations with unlabeled data, which alleviates the cost of labeling work. However, the problem is that these self-supervised sleep staging methods either require prior knowledge, as with frequency information, or they produce unsatisfactory results. Thus, we propose a deep sequential sleep network (DSSNet), a self-supervised framework that aims to perform multi-view representations based on contrastive learning. It utilizes a single-channel electroencephalogram but achieves competitive performance. We also explore the impact of different contrastive mechanisms on DSSNet performance. The results of the Sleep-EDF dataset prove that the consistency of negative samples is crucial for improving performance. We evaluate DSSNet on Sleep-EDF and ISRUC-Sleep and achieve accuracies of 80.0% and 71.4%.

A lightweight automatic sleep staging method for children using single-channel EEG based on edge artificial intelligence.

With the development of telemedicine and edge computing, edge artificial intelligence (AI) will become a new development trend for smart medicine. On the other hand, nearly one-third of children suffer from sleep disorders. However, all existing sleep staging methods are for adults. Therefore, we adapted edge AI to develop a lightweight automatic sleep staging method for children using single-channel EEG. The trained sleep staging model will be deployed to edge smart devices so that the sleep staging can be implemented on edge devices which will greatly save network resources and improving the performance and privacy of sleep staging application. Then the results and hypnogram will be uploaded to the cloud server for further analysis by the physicians to get sleep disease diagnosis reports and treatment opinions. We utilized 1D convolutional neural networks (1D-CNN) and long short term memory (LSTM) to build our sleep staging model, named CSleepNet. We tested the model on our childrens sleep (CS) dataset and sleep-EDFX dataset. For the CS dataset, we experimented with F4-M1 channel EEG using four different loss functions, and the logcosh performed best with overall accuracy of 83.06% and F1-score of 76.50%. We used Fpz-Cz and Pz-Oz channel EEG to train our model in Sleep-EDFX dataset, and achieved an accuracy of 86.41% without manual feature extraction. The experimental results show that our method has great potential. It not only plays an important role in sleep-related research, but also can be widely used in the classification of other time sequences physiological signals.

An automatic sleep staging method based on CRNN-HMM model

Sleep staging is an important process for detecting sleep quality and diagnosing sleep disorders. However, traditional sleep staging is a labor-intensive task, and it is prone to subjective errors. Therefore, this paper innovatively proposes an automatic sleep staging model based on single-channel EOG—CRNN-HMM. The CRNN-HMM in this paper combines Convolutional recursive neural networks(CRNN) and hidden Markov model(HMM). The main idea of this model is to use CRNN to automatically extract features from EOG, and send the feature signals to a variant of RNN, Bi-directional Long Short-Term Memory(BiLSTM), to mine the dependencies between sleep stages and realize automatic staging of sleep data. Finally, a Hidden Markov Model is used to convert the prior information of the sleep phase of the adjacent EOG cycle in order to improve the classification performance of S1, thereby improving the classification performance of CRNN. The simulation results show that the overall accuracy of the model on the CAP-Sleep data set reaches 95.0%, which proves that the model can provide a way for the evaluation of sleep quality.

Investigation of automated sleep staging from cardiorespiratory signals regarding clinical applicability and robustness

ObjectiveIn this contribution, we present an automated sleep stage classifier using modern machine learning methods with cardiorespiratory time series as input signals. We investigate (1) which classification quality we can yield, (2) how variations in the input data affect the results and (3) to which extend the proposed classifier model can serve as basis for medical diagnoses. MethodsWe extracted R-to-R intervals from electrocardiograms and breath-to-breath intervals from thoracic respiratory effort. Our analysis used 5036 patients from the Sleep Heart Health Study, 998 random patients served as independent hold-out test data. Our model architecture consists of convolutional and recurrent layers. From the predicted hypnograms, we calculated sleep metrics to estimate if the proposed model can serve as basis for medical diagnoses. We further modified the RR interval data and trained new models on that data to investigate its robustness to different data sources and resulting classification quality. ResultsThe proposed method yields a Cohen’s kappa score of 0.80 for distinguishing Wakefulness, NREM and REM, which is considered an almost perfect classification. Further, the proposed method is very robust concerning differing data sources. We found that some summarizing sleep metrics calculated from our predictions are reliable (e.g. sleep efficiency), others, however, are rather unreliable (e.g. percentage of deep sleep). ConclusionsOur results demonstrate the applicability and robustness of the proposed method for sleep staging. Additionally, they underline the requirement to consider not only scoring accuracy but also sleep metrics.

Unsupervised multi-subepoch feature learning and hierarchical classification for EEG-based sleep staging

As the medium of developing brain–computer interface system, the recognition of EEG signals is complicated and difficult due to the complex nonstationary characteristics and the individual difference between subjects. In this paper, we investigate the EEG signal classification problem and propose a novel unsupervised multi-subepoch feature learning and hierarchical classification method for automatic sleep staging. First, we divide the EEG epoch into multiple signal subepochs, and each subepoch is mapped to amplitude axis and time axis respectively to obtain two kinds of feature information with amplitude–time dynamic characteristics. Then, the statistical classification features are extracted from the mapped feature information. Furthermore, we conduct unsupervised feature learning for consistent and specific classification features from the perspective of time series. Finally, according to the differences and similarities of EEG signals in different sleep stages, a hierarchical weighted support vector machine-based classification model (H-WSVM) is established, which can use different feature subsets at each classification level and different weighting parameters for unbalanced data samples. To select the optimal feature subset for detecting each sleep stage, we propose a novel evaluation criterion for feature classification ability based on rough set theory. Experimental results on the most commonly used dataset show that the proposed method has better sleep staging performance and can effectively promote the development and application of EEG sleep staging system.

Inter-database validation of a deep learning approach for automatic sleep scoring.

Study objectivesDevelopment of inter-database generalizable sleep staging algorithms represents a challenge due to increased data variability across different datasets. Sharing data between different centers is also a problem due to potential restrictions due to patient privacy protection. In this work, we describe a new deep learning approach for automatic sleep staging, and address its generalization capabilities on a wide range of public sleep staging databases. We also examine the suitability of a novel approach that uses an ensemble of individual local models and evaluate its impact on the resulting inter-database generalization performance.MethodsA general deep learning network architecture for automatic sleep staging is presented. Different preprocessing and architectural variant options are tested. The resulting prediction capabilities are evaluated and compared on a heterogeneous collection of six public sleep staging datasets. Validation is carried out in the context of independent local and external dataset generalization scenarios.ResultsBest results were achieved using the CNN_LSTM_5 neural network variant. Average prediction capabilities on independent local testing sets achieved 0.80 kappa score. When individual local models predict data from external datasets, average kappa score decreases to 0.54. Using the proposed ensemble-based approach, average kappa performance on the external dataset prediction scenario increases to 0.62. To our knowledge this is the largest study by the number of datasets so far on validating the generalization capabilities of an automatic sleep staging algorithm using external databases.ConclusionsValidation results show good general performance of our method, as compared with the expected levels of human agreement, as well as to state-of-the-art automatic sleep staging methods. The proposed ensemble-based approach enables flexible and scalable design, allowing dynamic integration of local models into the final ensemble, preserving data locality, and increasing generalization capabilities of the resulting system at the same time.

EOGNET: A Novel Deep Learning Model for Sleep Stage Classification Based on Single-Channel EOG Signal.

In recent years, automatic sleep staging methods have achieved competitive performance using electroencephalography (EEG) signals. However, the acquisition of EEG signals is cumbersome and inconvenient. Therefore, we propose a novel sleep staging approach using electrooculogram (EOG) signals, which are more convenient to acquire than the EEG. A two-scale convolutional neural network first extracts epoch-wise temporary-equivalent features from raw EOG signals. A recurrent neural network then captures the long-term sequential information. The proposed method was validated on 101 full-night sleep data from two open-access databases, the montreal archive of sleep studies and Sleep-EDF, achieving an overall accuracy of 81.2 and 76.3%, respectively. The results are comparable to those models trained with EEG signals. In addition, comparisons with six state-of-the-art methods further demonstrate the effectiveness of the proposed approach. Overall, this study provides a new avenue for sleep monitoring.

Machine learning with ensemble stacking model for automated sleep staging using dual-channel EEG signal

Sleep staging is an important part of diagnosing the different types of sleep-related disorders because any discrepancies in the sleep scoring process may cause serious health problems such as misinterpretations of sleep patterns, medication errors, and improper diagnosis. The best way of analyzing sleep staging is visual interpretations of the polysomnography (PSG) signals recordings from the patients, which is a quite tedious task, requires more domain experts, and time-consuming process. This proposed study aims to develop a new automated sleep staging system using the brain EEG signals. Based on a new automated sleep staging system based on an ensemble learning stacking model that integrates Random Forest (RF) and eXtreme Gradient Boosting (XGBoosting). Additionally, this proposed methodology considers the subjects’ age, which helps analyze the S1 sleep stage properly. In this study, both linear (time and frequency) and non-linear features are extracted from the pre-processed signals. The most relevant features are selected using the ReliefF weight algorithm. Finally, the selected features are classified through the proposed two-layer stacking model. The proposed methodology performance is evaluated using the two most popular datasets, such as the Sleep-EDF dataset (S-EDF) and Sleep Expanded-EDF database (SE-EDF) under the Rechtschaffen & Kales (R&K) sleep scoring rules. The performance of the proposed method is also compared with the existing published sleep staging methods. The comparison results signify that the proposed sleep staging system has an excellent improvement in classification accuracy for the six-two sleep states classification. In the S-EDF dataset, the overall accuracy and Cohen’s kappa coefficient score obtained by the proposed model is (91.10%, 0.87) and (90.68%, 0.86) with inclusion and exclusion of age feature using the Fpz-Cz channel, respectively. Similarly, the Pz-Oz channel’s performance is (90.56%, 0.86) with age feature and (90.11%, 0.86) without age feature. The performed results with the SE-EDF dataset using Fpz-Cz channel is (81.32%, 0.77) and (81.06%, 0.76), using Pz-Oz channel with the inclusion and exclusion of the age feature, respectively. Similarly the model achieved an overall accuracy of 96.67% (CT-6), 96.60% (CT-5), 96.28% (CT-4),96.30% (CT-3) and 97.30% (CT-2) for with 16 selected features using S-EDF database. Similarly the model reported an overall accuracy of 85.85%, 84.98%, 85.51%, 85.37% and 87.40% for CT-6 to CT-2 with 18 selected features using SE-EDF database.

250 AI-Supported Sleep Staging from Activity and Heart Rate

Abstract Introduction Polysomnography (PSG) is considered the gold standard for sleep staging but is labor-intensive and expensive. Wrist wearables are an alternative to PSG because of their small form factor and continuous monitoring capability. In this work, we present a scheme to perform such automated sleep staging via deep learning in the MESA cohort validated against PSG. This scheme makes use of actigraphic activity counts and two coarse heart rate measures (only mean and standard deviation for 30-s sleep epochs) to perform multi-class sleep staging. Our method outperforms existing techniques in three-stage classification (i.e., wake, NREM, and REM) and is feasible for four-stage classification (i.e., wake, light, deep, and REM). Methods Our technique uses a combined convolutional neural network coupled and sequence-to-sequence network architecture to appropriate the temporal correlations in sleep toward classification. Supervised training with PSG stage labels for each sleep epoch as the target was performed. We used data from MESA participants randomly assigned to non-overlapping training (N=608) and validation (N=200) cohorts. The under-representation of deep sleep in the data leads to class imbalance which diminishes deep sleep prediction accuracy. To specifically address the class imbalance, we use a novel loss function that is minimized in the network training phase. Results Our network leads to accuracies of 78.66% and 72.46% for three-class and four-class sleep staging respectively. Our three-stage classifier is especially accurate at measuring NREM sleep time (predicted: 4.98 ± 1.26 hrs. vs. actual: 5.08 ± 0.98 hrs. from PSG). Similarly, our four-stage classifier leads to highly accurate estimates of light sleep time (predicted: 4.33 ± 1.20 hrs. vs. actual: 4.46 ± 1.04 hrs. from PSG) and deep sleep time (predicted: 0.62 ± 0.65 hrs. vs. actual: 0.63 ± 0.59 hrs. from PSG). Lastly, we demonstrate the feasibility of our method for sleep staging from Apple Watch-derived measurements. Conclusion This work demonstrates the viability of high-accuracy, automated multi-class sleep staging from actigraphy and coarse heart rate measures that are device-agnostic and therefore well suited for extraction from smartwatches and other consumer wrist wearables. Support (if any) This work was supported in part by the NIH grant 1R21AG068890-01 and the American Association for University Women.

Attention-Based LSTM for Non-Contact Sleep Stage Classification Using IR-UWB Radar.

Manual scoring of sleep stages from polysomnography (PSG) records is essential to understand the sleep quality and architecture. Since the PSG requires specialized personnel, a lab environment, and uncomfortable sensors, non-contact sleep staging methods based on machine learning techniques have been investigated over the past years. In this study, we propose an attention-based bidirectional long short-term memory (Attention Bi-LSTM) model for automatic sleep stage scoring using an impulse-radio ultra-wideband (IR-UWB) radar which can remotely detect vital signs. Sixty-five young (30.0 ± 8.6 yrs.) and healthy volunteers underwent nocturnal PSG and IR-UWB radar measurement simultaneously; From 51 recordings, 26 were used for training, 8 for validation, and 17 for testing. Sixteen features including movement-, respiration-, and heart rate variability-related indices were extracted from the raw IR-UWB signals in each 30-s epoch. Sleep stage classification performances of Attention Bi-LSTM model with optimized hyperparameters were evaluated and compared with those of conventional LSTM networks for same test dataset. In the results, we achieved an accuracy of 82.6 ± 6.7% and a Cohen's kappa coefficient of 0.73 ± 0.11 in the classification of wake stage, REM sleep, light (N1+N2) sleep, and deep (N3) sleep which is significantly higher than the conventional LSTM networks (p < 0.01). Moreover, the classification performances were higher than those reported in comparative studies, demonstrating the effectiveness of the attention mechanism coupled with bi-LSTM networks for the sleep staging using cardiorespiratory signals.

Transfer learning from ECG to PPG for improved sleep staging from wrist-worn wearables

Objective. To develop a sleep staging method from wrist-worn accelerometry and the photoplethysmogram (PPG) by leveraging transfer learning from a large electrocardiogram (ECG) database. Approach. In previous work, we developed a deep convolutional neural network for sleep staging from ECG using the cross-spectrogram of ECG-derived respiration and instantaneous beat intervals, heart rate variability metrics, spectral characteristics, and signal quality measures derived from 5793 subjects in Sleep Heart Health Study (SHHS). We updated the weights of this model by transfer learning using PPG data derived from the Empatica E4 wristwatch worn by 105 subjects in the ‘Emory Twin Study Follow-up’ (ETSF) database, for whom overnight polysomnographic (PSG) scoring was available. The relative performance of PPG, and actigraphy (Act), plus combinations of these two signals, with and without transfer learning was assessed. Main results. The performance of our model with transfer learning showed higher accuracy (1–9 percentage points) and Cohen’s Kappa (0.01–0.13) than those without transfer learning for every classification category. Statistically significant, though relatively small, incremental differences in accuracy occurred for every classification category as tested with the McNemar test. The out-of-sample classification performance using features from PPG and actigraphy for four-class classification was Accuracy (Acc) = 68.62% and Kappa = 0.44. For two-class classification, the performance was Acc = 81.49% and Kappa = 0.58. Significance. We proposed a combined PPG and actigraphy-based sleep stage classification approach using transfer learning from a large ECG sleep database. Results demonstrate that the transfer learning approach improves estimates of sleep state. The use of automated beat detectors and quality metrics means human over-reading is not required, and the approach can be scaled for large cross-sectional or longitudinal studies using wrist-worn devices for sleep staging.

A Study of Human Sleep Stage Classification Based on Dual Channels of EEG Signal Using Machine Learning Techniques

Sleep staging is one of the important methods for the diagnosis of the different types of sleep-related diseases. Manual inspection of sleep scoring is a very time-consuming process, labor-intensive, and requires more human interpretations, which may produce biased results. Therefore, in this paper, we propose an efficient automated sleep staging system to improve sleep staging accuracy. In this work, we extracted both linear and non-linear properties from the input signal. Next to that, a set of optimal features was selected from the extracted feature vector by using a feature reduction technique based on the ReliefF weight algorithm. Finally, the selected features were classified through four machine learning techniques like support vector machine, K-nearest neighbor, decision tree, and random forest. The proposed methodology performed using dual-channel EEG signals from the ISRUC-Sleep dataset under the AASM sleep scoring rules. The performance of the proposed methodology compared with the existing similar methods. In this work, we considered the 10-Fold cross validation strategy; our proposed methods reported the highest classification accuracy of 91.67% with the C4-A1 channel, and 93.8% with the O2-A1 channel using the Random forest classification model. The result of the proposed methodology outperformed the earlier contribution for two-class sleep states classification. The proposed dual-channel sleep staging method can be helpful for the clinicians during the sleep scoring and treatment for the different sleep-related diseases.

A Multi-Class Automatic Sleep Staging Method Based on Photoplethysmography Signals.

Automatic sleep staging with only one channel is a challenging problem in sleep-related research. In this paper, a simple and efficient method named PPG-based multi-class automatic sleep staging (PMSS) is proposed using only a photoplethysmography (PPG) signal. Single-channel PPG data were obtained from four categories of subjects in the CAP sleep database. After the preprocessing of PPG data, feature extraction was performed from the time domain, frequency domain, and nonlinear domain, and a total of 21 features were extracted. Finally, the Light Gradient Boosting Machine (LightGBM) classifier was used for multi-class sleep staging. The accuracy of the multi-class automatic sleep staging was over 70%, and the Cohen’s kappa statistic k was over 0.6. This also showed that the PMSS method can also be applied to stage the sleep state for patients with sleep disorders.

XSleepNet: Multi-View Sequential Model for Automatic Sleep Staging.

Automating sleep staging is vital to scale up sleep assessment and diagnosis to serve millions experiencing sleep deprivation and disorders and enable longitudinal sleep monitoring in home environments. Learning from raw polysomnography signals and their derived time-frequency image representations has been prevalent. However, learning from multi-view inputs (e.g., both the raw signals and the time-frequency images) for sleep staging is difficult and not well understood. This work proposes a sequence-to-sequence sleep staging model, XSleepNet,1 that is capable of learning a joint representation from both raw signals and time-frequency images. Since different views may generalize or overfit at different rates, the proposed network is trained such that the learning pace on each view is adapted based on their generalization/overfitting behavior. In simple terms, the learning on a particular view is speeded up when it is generalizing well and slowed down when it is overfitting. View-specific generalization/overfitting measures are computed on-the-fly during the training course and used to derive weights to blend the gradients from different views. As a result, the network is able to retain the representation power of different views in the joint features which represent the underlying distribution better than those learned by each individual view alone. Furthermore, the XSleepNet architecture is principally designed to gain robustness to the amount of training data and to increase the complementarity between the input views. Experimental results on five databases of different sizes show that XSleepNet consistently outperforms the single-view baselines and the multi-view baseline with a simple fusion strategy. Finally, XSleepNet also outperforms prior sleep staging methods and improves previous state-of-the-art results on the experimental databases.

PSG Kayıt Sinyalleri Kullanılarak Uyku Evrelerinin Sınıflandırılması

Automatic sleep staging is aimed within the scope of this paper. Sleep staging is a study by a sleep specialist. Since this process takes quite a long time and sleep is a method based on the knowledge and experience, it is inevitable for each person to show different results. For this, an automatic sleep staging method has been introduced. In the study, EEG (Electroencephalogram), EOG (Electrooculogram), EMG (Electromyogram) data recorded by PSG (Polysomnography) device for seven patients in Necmettin Erbakan University sleep laboratory were used. 81 different features were taken from the data in time and frequency environment. Also, PCA (Principal component analysis) and SFS (Sequential forward selection) feature selection methods were used. The classification success of the sleep phases in different machine learning methods was measured by using the received features. Linear D. (Linear Discriminant Analysis), Cubic SVM (Support vector machine), Weighted kNN (k nearest neighbor), Bagged Trees, ANN (Artificial neural network) were used as classifiers. System success was achieved with a 5 fold cross-validation method. Accuracy rates obtained were respectively 55.6%, 65.8%, 67%, 72.1%, and 69.1%.

End-to-end sleep staging using convolutional neural network in raw single-channel EEG

ObjectiveManual sleep staging on overnight polysomnography (PSG) is time-consuming and laborious. This study aims to develop an end-to-end automatic sleep staging method in single-channel electroencephalogram (EEG) signals from PSG recordings. MethodsA convolutional neural network called CCN-SE is proposed to address sleep staging tasks. The proposed method was efficiently constructed by stacking a collection of consecutive convolutional micro-networks (CCNs) and squeeze-excitation (SE) block. The designed model took multi-epoch (3 epochs) raw EEG signals as its input and relabeled the input. We trained and tested this model on different single-channel EEG (C4-A1 and Fpz-Cz) signals from two open datasets and then explored the model’s generalization ability and the channel mismatch problem using clinical PSG files. ResultsResults of the five-fold cross-validation show that our model achieved the good overall accuracies in SHHS1 (88.1%) and Sleep-EDFx (85.3%) datasets. Furthermore, the observed scores on 10 healthy clinical sleep recordings using the single EEG channel (C4-M1) based on two trained weights were 72.3% and 81.9%. ConclusionThe obtained performance on two sleep datasets reveals the efficiency and generalization capability of the proposed method in sleep staging in EEG. Furthermore, the results on the clinical PSG recordings suggest that the proposed model can alleviate the problem of channel mismatch to some extent. SignificanceThis study proposes a novel method for automatic sleep staging that can be easily utilized in portable sleep monitoring devices and draws attention to the channel mismatch in sleep staging.

Towards More Accurate Automatic Sleep Staging via Deep Transfer Learning

Despite recent significant progress in the development of automatic sleep staging methods, building a good model still remains a big challenge for sleep studies with a small cohort due to the data-variability and data-inefficiency issues. This work presents a deep transfer learning approach to overcome these issues and enable transferring knowledge from a large dataset to a small cohort for automatic sleep staging. We start from a generic end-to-end deep learning framework for sequence-to-sequence sleep staging and derive two networks as the means for transfer learning. The networks are first trained in the source domain (i.e. the large database). The pretrained networks are then finetuned in the target domain (i.e. the small cohort) to complete knowledge transfer. We employ the Montreal Archive of Sleep Studies (MASS) database consisting of 200 subjects as the source domain and study deep transfer learning on three different target domains: the Sleep Cassette subset and the Sleep Telemetry subset of the Sleep-EDF Expanded database, and the Surrey-cEEGrid database. The target domains are purposely adopted to cover different degrees of data mismatch to the source domains. Our experimental results show significant performance improvement on automatic sleep staging on the target domains achieved with the proposed deep transfer learning approach. These results suggest the efficacy of the proposed approach in addressing the above-mentioned data-variability and data-inefficiency issues. As a consequence, it would enable one to improve the quality of automatic sleep staging models when the amount of data is relatively small.11The source code and the pretrained models are published at https://github.com/pquochuy/sleep_transfer_learning.

Automatic Sleep Staging Based on a Hybrid Stacked LSTM Neural Network: Verification Using Large-Scale Dataset

Previously reported automatic sleep staging methods have usually been developed using healthy groups of fewer than 100 subjects. In this study, an automatic sleep staging method based on hybrid stacked long short-term memory (LSTM) was proposed and evaluated using a large-scale dataset of subjects with sleep disorders. Twenty-four features, including temporal and spectrum factors, were extracted from physiological signals and normalized after extracting the features. A variety of hybrid stacked LSTM structures and hidden units were used to determine the most suitable structure and parameters for the automatic sleep staging method. Finally, the proposed method was validated using a large-scale sleep disorder dataset from the PhysioNet Challenge 2018. To validate the robustness of the proposed system, half of the 994 subjects were randomly assigned to the training set, and the other half were assigned to the testing set. The best accuracy and kappa coefficient of the proposed method are 83.07% and 0.78, respectively. The best hybrid stacked structure was LSTM combined with bidirectional LSTM, which has 125 hidden units. In addition, four common sleep indices, including sleep efficiency, sleep onset time, wake after sleep onset, and total sleep time, were evaluated. The results, according to the intraclass correlation coefficient, indicated a moderate agreement with the results of the expert. The performance of the proposed method was compared with that of conventional machine learning, and it was noted that the hybrid stacked LSTM is a promising solution for automatic sleep staging. In future work, this method may assist clinical staff in reducing the time required for sleep staging.