Automatic Sleep Scoring Research Articles

Intra- and inter-epoch temporal context network (IITNet) using sub-epoch features for automatic sleep scoring on raw single-channel EEG

A deep learning model, named IITNet, is proposed to learn intra- and inter-epoch temporal contexts from raw single-channel EEG for automatic sleep scoring. To classify the sleep stage from half-minute EEG, called an epoch, sleep experts investigate sleep-related events and consider the transition rules between the found events. Similarly, IITNet extracts representative features at a sub-epoch level by a residual neural network and captures intra- and inter-epoch temporal contexts from the sequence of the features via bidirectional LSTM. The performance was investigated for three datasets as the sequence length (L) increased from one to ten. IITNet achieved the comparable performance with other state-of-the-art results. The best accuracy, MF1, and Cohen's kappa (κ) were 83.9%, 77.6%, 0.78 for SleepEDF (L = 10), 86.5%, 80.7%, 0.80 for MASS (L = 9), and 86.7%, 79.8%, 0.81 for SHHS (L = 10), respectively. Even though using four epochs, the performance was still comparable. Compared to using a single epoch, on average, accuracy and MF1 increased by 2.48%p and 4.90%p and F1 of N1, N2, and REM increased by 16.1%p, 1.50%p, and 6.42%p, respectively. Above four epochs, the performance improvement was not significant. The results support that considering the latest two-minute raw single-channel EEG can be a reasonable choice for sleep scoring via deep neural networks with efficiency and reliability. Furthermore, the experiments with the baselines showed that introducing intra-epoch temporal context learning with a deep residual network contributes to the improvement in the overall performance and has the positive synergy effect with the inter-epoch temporal context learning.

0441 An Automatic Sleep Scoring System Based on Ensemble Convolutional Neural Network and Spectrogram of Sleep Physiological Signal

Abstract Introduction Manual sleep stage scoring is time consuming and subjective. Therefore, several studies focused on developing automated sleep scoring algorithms. The previously reported the automatic sleep scoring have been develop usually using small dataset, which less than 100 subjects. In this study, an automatic sleep scoring system based on ensemble convolutional neural network (ensemble-CNN) and spectrogram of sleep physiological signal was proposed and evaluated using a large dataset with sleep disorder. Methods The spectrograms were computed from each 30-s EEG and EOG of 994 subjects from PhysioNet 2018 challenge dataset, using the continuous wavelet transform, which were fed into an ensemble-CNN classification for training. The ensemble-CNN contained five pretrained models, ResNet-101, Inception-v4, DenseNet-201, Xception, and NASNet models, because these models’ architectures are different which can learn different features from the spectrograms to obtain high accuracy. The probabilities of five models were averaged to decide the sleep stage for each spectrogram. After classifying sleep stage, a smoothing process was used for sleep continuity. Moreover, the total 80% data from PhysioNet dataset were randomly assigned to the training set, and the remaining data were assigned to the testing set. Results To validate the robustness of the proposed system, the validation procedure was repeated five times. The performance measures were averaged over the five runs. The overall agreement and kappa coefficient of the proposed method are 82% and 0.73, respectively. The sensitivity of the sleep stages of Wake, N1, N2, N3, and REM are 90.0%, 48.6%, 84.9%, 84.2%, and 81.9%, respectively. Conclusion The performance of the proposed method was achieved expert level, and it was noted that the ensemble-CNN is a promising solution for automatic sleep stage scoring. This method can assist clinical staff in reducing the time required for sleep stage scoring in the future. Support This work was supported by the Ministry of Science and Technology, Taiwan. (MOST 106-2218-E-035-013-MY2, 108-2221-E-035-064, and 108-2634-F-006-012).

1211 Assessing the Accuracy of a Dry-EEG Headband for Measuring Brain Activity, Heart Rate, Breathing and Automatic Sleep Staging

Abstract Introduction The development of new sleep study devices, adapted for daily use, is necessary for diagnosis of sleep disorders. However, this requires to be both suitable for daily use and capable of recording accurate electrophysiological data. This study assesses the signal acquisition of a comfortable sleep headband, using dry electrodes, and the performance of its automatic sleep staging algorithms compared to the gold-standard clinical PSG scored by 4 sleep experts. Methods 42 participants slept at a sleep center wearing both the Dreem headband (DH) and a PSG simultaneously. We measured 1) the EEG signal similarity between both devices, 2) heart rate, breathing frequency and respiration rate variability (RRV) agreement, and 3) the performance of the headband automatic sleep scoring compared to PSG sleep experts manual scoring. Results Results demonstrate a strong correlation between the EEG signals acquired by the headband and those from the PSG, and the signals acquired by the headband enable monitoring of alpha (r= 0.75 ± 0.11), beta (r= 0.74 ± 0.14), delta (r = 0.78 ± 0.16), and theta (r = 0.63 ± 0.15) frequencies during sleep. The mean absolute error for heart rate, breathing frequency, and RRV was 2.2 ± 0.8 bpm, 0.3 ± 0.2 cpm and 3.1 ± 0.4 %, respectively. Automatic Sleep Staging reached an overall accuracy of 84.1 ± 7.5% (F1 score: 83.0 ± 8.4) for the headband to be compared with an average of 86.4 ± 5.5% (F1 score: 86.5 ± 5.5) for the 4 sleep experts. Conclusion These results demonstrate the capacity of the headband to both precisely monitor sleep-related physiological signals and process them accurately into sleep stages. This device paves the way for high-quality, large-scale, longitudinal sleep studies. Support This Study has been supported by Dreem sas.

Long Short-Term Memory Networks for Unconstrained Sleep Stage Classification Using Polyvinylidene Fluoride Film Sensor.

Sleep stage scoring is the first step towards quantitative analysis of sleep using polysomnography (PSG) recordings. However, although PSG is a gold standard method for assessing sleep, it is obtrusive and difficult to apply for long-term sleep monitoring. Further, because human experts manually classify sleep stages, it is time-consuming and exhibits inter-rater variability. Therefore, this article proposes a long short-term memory (LSTM) model for automatic sleep stage scoring using a polyvinylidene fluoride (PVDF) film sensor that can provide unconstrained long-term physiological monitoring. Signals were recorded using a PVDF sensor during PSG. From 60 recordings, 30 were used for training, 10 for validation, and 20 for testing. Sixteen parameters, including movement, respiration-related, and heart rate variability, were extracted from the recorded signals and then normalized. From the selected LSTM architecture, four sleep stage classification performances were evaluated for a test dataset and the results were compared with those of conventional machine learning methods. According to epoch-by-epoch (30 s) analysis, the classification performance for the four sleep stages had an average accuracy of 73.9% and a Cohen's kappa coefficient of 0.55. When compared with other machine learning methods, the proposed method achieved the highest classification performance. The use of LSTM networks with the PVDF film sensor has potential for facilitating automatic sleep scoring, and it can be applied for long-term sleep monitoring at home.

Addressing database variability in learning from medical data: An ensemble-based approach using convolutional neural networks and a case of study applied to automatic sleep scoring

In this work we examine some of the problems associated with the development of machine learning models with the objective to achieve robust generalization capabilities on common-task multiple-database scenarios. Referred to as the “database variability problem”, we focus on a specific medical domain (sleep staging in sleep medicine) to show the non-triviality of translating the estimated model's local generalization capabilities into independent external databases. We analyze some of the scalability problems when multiple-database data are used as inputs to train a single learning model. Then, we introduce a novel approach based on an ensemble of local models, and we show its advantages in terms of inter-database generalization performance and data scalability. In addition, we analyze different model configurations and data pre-processing techniques to determine their effects on the overall generalization performance. For this purpose, we carry out experimentation that involves several sleep databases and evaluates different machine learning models based on convolutional neural networks.

Exploring scoring methods for research studies: Accuracy and variability of visual and automated sleep scoring.

Sleep studies face new challenges in terms of data, objectives and metrics. This requires reappraising the adequacy of existing analysis methods, including scoring methods. Visual and automatic sleep scoring of healthy individuals were compared in terms of reliability (i.e., accuracy and stability) to find a scoring method capable of giving access to the actual data variability without adding exogenous variability. A first dataset (DS1, four recordings) scored by six experts plus an autoscoring algorithm was used to characterize inter-scoring variability. A second dataset (DS2, 88 recordings) scored a few weeks later was used to explore intra-expert variability. Percentage agreements and Conger's kappa were derived from epoch-by-epoch comparisons on pairwise and consensus scorings. On DS1 the number of epochs of agreement decreased when the number of experts increased, ranging from 86% (pairwise) to 69% (all experts). Adding autoscoring to visual scorings changed the kappa value from 0.81 to 0.79. Agreement between expert consensus and autoscoring was 93%. On DS2 the hypothesis of intra-expert variability was supported by a systematic decrease in kappa scores between autoscoring used as reference and each single expert between datasets (.75-.70). Although visual scoring induces inter- and intra-expert variability, autoscoring methods can cope with intra-scorer variability, making them a sensible option to reduce exogenous variability and give access to the endogenous variability in the data.

Towards a user-friendly sleep staging system for polysomnography part I: Automatic classification based on medical knowledge

Abstract Manual sleep scoring is a time-consuming task that requires a high level of medical expertise. For this reason, a number of automatic sleep scoring algorithms have recently been implemented. However, their use by physicians remains limited for various reasons: a lack of transparency of the approach used, insufficient heterogeneity among the patients used for testing, or a lack of practicality. This paper presents a system for facilitated sleep scoring that will overcome these limitations. The proposed system, a user-friendly tool based on electrophysiological channels, was trained and tested on large datasets of 300 and 100 distinct recordings from patients with various sleep disorders. The method replicates the manual sleep scoring process, in accordance with the American Academy of Sleep Medicine (AASM) guidelines and generates patient-dependent sleep scoring (using the SATUD system). For an improved level of precision and confidence with regard to scoring, our approach also provides a table that gives indications about the confidence level of the algorithm when scoring sleep. In contrast to recent deep learning approaches, the algorithms used were chosen for their resilience and as they are easy to understand. Medical knowledge was included in the process as much as possible. Results showed that the system is consistent with manual scoring (mean Cohen's Kappa of 0.69 and accuracy rate of 77.8%). It proves that a facilitated interpretation of the model, crucial in such fields as sleep diagnosis, can be provided when using automatic tools. This new system thereby generates sleep scoring decision support tools, which should easily contribute to significant time-saving and help sleep specialists to perform sleep diagnosis.

Automatic sleep scoring system using a patch-type EEG sensor based on deep learning

Automatic sleep scoring system using a patch-type EEG sensor based on deep learning

Accurate whole-night sleep monitoring with dry-contact ear-EEG

Sleep is a key phenomenon to both understanding, diagnosing and treatment of many illnesses, as well as for studying health and well being in general. Today, the only widely accepted method for clinically monitoring sleep is the polysomnography (PSG), which is, however, both expensive to perform and influences the sleep. This has led to investigations into light weight electroencephalography (EEG) alternatives. However, there has been a substantial performance gap between proposed alternatives and PSG. Here we show results from an extensive study of 80 full night recordings of healthy participants wearing both PSG equipment and ear-EEG. We obtain automatic sleep scoring with an accuracy close to that achieved by manual scoring of scalp EEG (the current gold standard), using only ear-EEG as input, attaining an average Cohen’s kappa of 0.73. In addition, this high performance is present for all 20 subjects. Finally, 19/20 subjects found that the ear-EEG had little to no negative effect on their sleep, and subjects were generally able to apply the equipment without supervision. This finding marks a turning point on the road to clinical long term sleep monitoring: the question should no longer be whether ear-EEG could ever be used for clinical home sleep monitoring, but rather when it will be.

Development of a human-computer collaborative sleep scoring system for polysomnography recordings.

The overnight polysomnographic (PSG) recordings of patients were scored by an expert to diagnose sleep disorders. Visual sleep scoring is a time-consuming and subjective process. Automatic sleep staging methods can help; however, the mechanism and reliability of these methods are not fully understood. Therefore, experts often need to rescore the recordings to obtain reliable results. Here, we propose a human-computer collaborative sleep scoring system. It is a rule-based automatic sleep scoring method that follows the American Academy of Sleep Medicine (AASM) guidelines to perform an initial scoring. Then, the reliability level of each epoch is analyzed based on physiological patterns during sleep and the characteristics of various stage changes.Finally, experts would only need to rescore epochs with a low-reliability level. The experimental results show that the average agreement rate between our system and fully manual scorings can reach 90.42% with a kappa coefficient of 0.85. Over 50% of the manual scoring time can be reduced. Due to the demonstrated robustness and applicability, the proposed approach can be integrated with various PSG systems or automatic sleep scoring methods for sleep monitoring in clinical or homecare applications in the future.

Automatic Sleep Scoring Based on Refined Composite Multi-Scale Entropy and Support Vector Machine

Automatic Sleep Scoring Based on Refined Composite Multi-Scale Entropy and Support Vector Machine

Multi-modality of polysomnography signals’ fusion for automatic sleep scoring

Abstract Objective The study aims to develop an automatic sleep scoring method by fusing different polysomnography (PSG) signals and further to investigate PSG signals’ contribution to the scoring result. Methods Eight combinations of four modalities of PSG signals, namely electroencephalogram (EEG), electrooculogram (EOG), electromyogram (EMG), and electrocardiogram (ECG) were considered to find the optimal fusion of PSG signals. A total of 232 features, covering statistical characters, frequency characters, time-frequency characters, fractal characters, entropy characters and nonlinear characters, were derived from these PSG signals. To select the optimal features for each signal fusion, four widely used feature selection methods were compared. At the classification stage, five different classifiers were employed to evaluate the validity of the features and to classify sleep stages. Results For the database in the present study, the best classifier, random forest, realized the optimal consistency of 86.24% with the sleep macrostructures scored by the technologists trained at the Sleep Center. The optimal accuracy was achieved by fusing four modalities of PSG signals. Specifically, the top twelve features in the optimal feature set were respectively EEG features named zero-crossings, spectral edge, relative power spectral of theta, Petrosian fractal dimension, approximate entropy, permutation entropy and spectral entropy, and EOG features named spectral edge, approximate entropy, permutation entropy and spectral entropy, and the mutual information between EEG and submental EMG. In addition, ECG features (e.g. Petrosian fractal dimension, zero-crossings, mean value of R amplitude and permutation entropy) were useful for the discrimination among W, S1 and R. Conclusions Through exploring the different fusions of multi-modality signals, the present study concluded that the multi-modality of PSG signals’ fusion contributed to higher accuracy, and the optimal feature set was a fusion of multiple types of features. Besides, compared with manual scoring, the proposed automatic scoring methods were cost-effective, which would alleviate the burden of the physicians, speed up sleep scoring, and expedite sleep research.

Symbolic dynamics of electroencephalography is associated with the sleep depth and overall sleep quality in healthy adults

Sleep electroencephalographic (EEG) provides the opportunity to study sleep scientifically. Slow wave activity (SWA), presenting EEG spectral power in the low-frequency range, has proven to be a useful parameter in sleep medicine. Drawing inspiration from the adaptive and noise-assist features of symbolic dynamics, we introduced a symbolic analogue of SWA as EEG signal was generally considered as non-linear and non-stationary. Moreover, we investigated whether the proposed metrics can capture patterns that characterize and differentiate different sleep stages, and whether EEG dynamical features during the wake to sleep transition after light-off share a correlation with the overall sleep quality during the whole night. Single-channel EEGs derived from the polysomnography (PSG) of 111 healthy adults in the Sleep Heart Health Study were analyzed retrospectively. Every 30-second epoch of EEG data was transformed into a symbolic sequence using equiprobable symbolization and then the percentage of constant word (PCW) was calculated. The results revealed that the proposed metric, PCW, exhibits a correlation with wake/sleep stages over the night. More importantly, average PCW in short sections (15–60 min) at the beginning of the night shows a correlation with various indices of sleep quality for the entire night, suggesting PCW as a potential indicator for the requirement for an early sleep intervention. In conclusion, the results validate the use of symbolic dynamics in automatic sleep scoring and evaluation, and might further expand the application of SWA measurement to the early intervention of sleep disorders.

A convolutional neural network for sleep stage scoring from raw single-channel EEG

Abstract We present a novel method for automatic sleep scoring based on single-channel EEG. We introduce the use of a deep convolutional neural network (CNN) on raw EEG samples for supervised learning of 5-class sleep stage prediction. The network has 14 layers, takes as input the 30-s epoch to be classified as well as two preceding epochs and one following epoch for temporal context, and requires no signal preprocessing or feature extraction phase. We train and evaluate our system using data from the Sleep Heart Health Study (SHHS), a large multi-center cohort study including expert-rated polysomnographic records. Performance metrics reach the state of the art, with accuracy of 0.87 and Cohen kappa of 0.81. The use of a large cohort with multiple expert raters guarantees good generalization. Finally, we present a method for visualizing class-wise patterns learned by the network.

The genetic liability for insomnia is associated with lower amount of slow wave sleep in young and healthy individuals

The genetic liability for insomnia is associated with lower amount of slow wave sleep in young and healthy individuals

A hierarchical classification method for automatic sleep scoring using multiscale entropy features and proportion information of sleep architecture

BackgroundSleep scoring is a critical step in medical researches and clinical applications. Traditional visual scoring method is based on the processing of physiological signals, such as electroencephalogram (EEG), electrooculogram (EOG) and electromyogram (EMG), which is a time consuming and subjective procedure. It is an urgent task to develop an effective method for automatic sleep scoring. MethodThis paper presents a hierarchical classification method for automatic sleep scoring by combining multiscale entropy features with the proportion information of the sleep architecture. Based on a three-layer classification scheme, sleep is categorized into five stages (Awake, S1, S2, SWS and REM). Specifically, the first layer is a standard SVM which performs classification between Awake and Sleep, while the second and third layers are implemented by combining probabilistic output SVM with proportion-based clustering. Multiscale entropy (MSE) from electroencephalogram (EEG) is extracted to represent signal characteristics in multiple temporal scales. ResultsThe proposed method is evaluated with 20 sleep recordings, including 10 subjects with mild difficulty falling asleep and 10 healthy subjects. The overall accuracy of the proposed method is 91.4%. Compared with traditional methods, the classification accuracy of the proposed method is more balanced and the global performance is much better. The dataset includes both healthy subjects and subjects with sleep disorders, which means the presented method has good generalization capacity. Experimental results demonstrate the feasibility of the attempt to introduce proportion information into automatic sleep scoring.

A Novel Computer Assisted Automatic Sleep Scoring System by Employing Artificial Neural Network–A review

Sleep is an essential element for an individual’s well-being and is considered vital for the overall mental and physical heath of a person. Sleep can be considered as a virtual detachment of an individual from his environment. In normal humans, about 30% of their life-time is spent for sleep. Artificial neural networks (ANNs) or connectionist systems are computing systems inspired by the biological neural networks that constitute animal brains. Sleep scoring is under taken by the examination and visual inspection of polysomnograms (PSG) done by sleep specialist. PSG is specialty test, the conduction of which includes the recording of various physiological signals. The signals obtained are processed using digital processing tools so as to extract information. Soft computing techniques are used to analyze the signals. ANNs are parallel adaptive systems suitable for solving of non-linear problems. Using ANN for automatic sleep scoring is especially promising because of new ANN learning algorithms allowing faster classification without decreasing the performance. Both appropriate preparations of training data as well as selection of the ANN model make it possible to perform effective and correct recognizing of relevant sleep stages. Such an approach is highly topical, taking into consideration the fact that there is no automatic scorer utilizing ANN technology available at present. The high performances observed with systems based onneural networks highlight that these tools may be act new tools in the field of sleep research. In this scenario we are surmised the review regarding the computer assisted automatic scoring of sleep and soft computing technique Artificial Neural Network.

An automated method for sleep staging from EEG signals using normal inverse Gaussian parameters and adaptive boosting

Sleep stage screening based on visual inspection is burdensome, time-consuming, subjective, and error-prone owing to the large bulk of data which have to be screened. Consequently, automatic sleep scoring is essential for both sleep research and sleep disorder diagnosis. In this work, we present the application of newly proposed tunable-Q factor wavelet transform (TQWT) to devise a single channel EEG based computerized sleep staging algorithm. First, we decompose the sleep-EEG signal segments into TQWT sub-bands. Then we perform normal inverse Gaussian (NIG) pdf modeling of TQWT sub-bands wherein NIG parameters are used as features. The effects of various TQWT parameters are also studied. The suitability of NIG parameters in the TQWT domain is inspected. In this study, we employ adaptive boosting (AdaBoost) for sleep stage classification. To assess the performance of the classification model and to determine the optimal choices of AdaBoost parameters, 10 fold cross-validation is performed. The performance of the proposed scheme is promising in terms of sensitivity, specificity, accuracy, and Cohen's Kappa co-efficient. Comparative analysis of performance suggests that the algorithmic performance of the proposed scheme, as opposed to that of the state-of-the-art ones is better. Further, the proposed algorithm also gives superior S1 and REM stage detection accuracy. The computerized sleep scoring scheme propounded herein can expedite sleep disorder diagnosis, contribute to the device implementation of a sleep monitoring system, and benefit sleep research.

An efficient sleep scoring system based on EEG signal using complex-valued machine learning algorithms

Sleep staging is a significant step in the diagnosis and treatment of sleep disorders. Sleep scoring is a time-consuming and difficult process. Given that sleep scoring requires expert knowledge, it is generally undertaken by sleep experts. In this study, a new hybrid machine learning method consisting of complex-valued nonlinear features (CVNF) and a complex-valued neural network (CVANN) has been presented for automatic sleep scoring using single channel electroencephalography (EEG) signals. First of all, we should note that in this context, nine nonlinear features have been obtained as those are often preferred for the classification of EEG signals. These obtained features were then converted into a complex-valued number format using a phase encoding method. In this way, a new complex-valued feature set was obtained for sleep scoring. The obtained attributes have been presented as input to the CVANN algorithm. We have used a number of different statistical parameters during the evaluation process. The results that have been obtained are based on two sleep standards: Rechtschaffen & Kales (R&K) and American Academy of Sleep Medicine (AASM). Finally, a 91.57% accuracy rate was obtained according to R&K standard; a 93.84% accuracy rate was obtained according to the AASM standard using the proposed method. We therefore observed that the proposed method is promising in terms of the sleep scoring.

Multiple classifier systems for automatic sleep scoring in mice

BackgroundElectroencephalogram (EEG) and electromyogram (EMG) recordings are often used in rodents to study sleep architecture and sleep-associated neural activity. These recordings must be scored to designate what sleep/wake state the animal is in at each time point. Manual sleep-scoring is very time-consuming, so machine-learning classifier algorithms have been used to automate scoring. New methodInstead of using single classifiers, we implement a multiple classifier system. The multiple classifier is built from six base classifiers: decision tree, k-nearest neighbors, naïve Bayes, support vector machine, neural net, and linear discriminant analysis. Decision tree and k-nearest neighbors were improved into ensemble classifiers by using bagging and random subspace. Confidence scores from each classifier were combined to determine the final classification. Ambiguous epochs can be rejected and left for a human to classify. ResultsSupport vector machine was the most accurate base classifier, and had error rate of 0.054. The multiple classifier system reduced the error rate to 0.049, which was not significantly different from a second human scorer. When 10% of epochs were rejected, the remaining epochs’ error rate dropped to 0.018. Comparison with existing method(s)Compared with the most accurate single classifier (support vector machine), the multiple classifier reduced errors by 9.4%. The multiple classifier surpassed the accuracy of a second human scorer after rejecting only 2% of epochs. ConclusionsMultiple classifier systems are an effective way to increase automated sleep scoring accuracy. Improvements in autoscoring will allow sleep researchers to increase sample sizes and recording lengths, opening new experimental possibilities.