Personalization of Automatic Sleep Scoring: How Best to Adapt Models to Personal Domains in Wearable EEG.

Wearable EEG enables us to capture large amounts of high-quality sleep data for diagnostic purposes. To make full use of this capacity we need high-performance automatic sleep scoring models. To this end, it has been noted that domain mismatch between recording equipment can be considerable, e.g. PSG to wearable EEG, but a previously observed benefit from personalizing models to individual subjects further indicates a personal domain in sleep EEG. In this work, we have investigated the extent of such a personal domain in wearable EEG, and review supervised and unsupervised approaches to personalization as found in the literature. We investigated the personalization effect of the unsupervised Adversarial Domain Adaptation and implemented an unsupervised method based on statistics alignment. No beneficial personalization effect was observed using these unsupervised methods. We find that supervised personalization leads to a substantial performance improvement on the target subject ranging from 15% Cohen's Kappa for subjects with poor performance ( ) to roughly 2% on subjects with high performance ( ). This improvement was present for models trained on both small and large data sets, indicating that even high-performance models benefit from supervised personalization. We found that this personalization can be beneficially regularized using Kullback-Leibler regularization, leading to lower variance with negligible cost to improvement. Based on the experiments, we recommend model personalization using Kullback-Leibler regularization.

L-SeqSleepNet: Whole-cycle Long Sequence Modeling for Automatic Sleep Staging.

Human sleep is cyclical with a period of approximately 90 minutes, implying long temporal dependency in the sleep data. Yet, exploring this long-term dependency when developing sleep staging models has remained untouched. In this work, we show that while encoding the logic of a whole sleep cycle is crucial to improve sleep staging performance, the sequential modelling approach in existing state-of-the-art deep learning models are inefficient for that purpose. We thus introduce a method for efficient long sequence modelling and propose a new deep learning model, L-SeqSleepNet, which takes into account whole-cycle sleep information for sleep staging. Evaluating L-SeqSleepNet on four distinct databases of various sizes, we demonstrate state-of-the-art performance obtained by the model over three different EEG setups, including scalp EEG in conventional Polysomnography (PSG), in-ear EEG, and around-the-ear EEG (cEEGrid), even with a single EEG channel input. Our analyses also show that L-SeqSleepNet is able to alleviate the predominance of N2 sleep (the major class in terms of classification) to bring down errors in other sleep stages. Moreover the network becomes much more robust, meaning that for all subjects where the baseline method had exceptionally poor performance, their performance are improved significantly. Finally, the computation time only grows at a sub-linear rate when the sequence length increases.

At-home sleep monitoring using generic ear-EEG.

Introduction: A device comprising two generic earpieces with embedded dry electrodes for ear-centered electroencephalography (ear-EEG) was developed. The objective was to provide ear-EEG based sleep monitoring to a wide range of the population without tailoring the device to the individual. Methods: To validate the device ten healthy subjects were recruited for a 12-night sleep study. The study was divided into two parts; part A comprised two nights with both ear-EEG and polysomnography (PSG), and part B comprised 10 nights using only ear-EEG. In addition to the electrophysiological measurements, subjects filled out a questionnaire after each night of sleep. Results: The subjects reported that the ear-EEG system was easy to use, and that the comfort was better in part B. The performance of the system was validated by comparing automatic sleep scoring based on ear-EEG with PSG-based sleep scoring performed by a professional trained sleep scorer. Cohen's kappa was used to assess the agreement between the manual and automatic sleep scorings, and the study showed an average kappa value of 0.71. The majority of the 20 recordings from part A yielded a kappa value above 0.7. The study was compared to a companioned study conducted with individualized earpieces. To compare the sleep across the two studies and two parts, 7 different sleeps metrics were calculated based on the automatic sleep scorings. The ear-EEG nights were validated through linear mixed model analysis in which the effects of equipment (individualized vs. generic earpieces), part (PSG and ear-EEG vs. only ear-EEG) and subject were investigated. We found that the subject effect was significant for all computed sleep metrics. Furthermore, the equipment did not show any statistical significant effect on any of the sleep metrics. Discussion: These results corroborate that generic ear-EEG is a promising alternative to the gold standard PSG for sleep stage monitoring. This will allow sleep stage monitoring to be performed in a less obtrusive way and over longer periods of time, thereby enabling diagnosis and treatment of diseases with associated sleep disorders.

Self-applied ear-EEG for sleep monitoring at home.

High quality sleep monitoring is done using EEG electrodes placed on the skin. This has traditionally required assistance by an expert when the equipment needed to mounted. However, this creates a limitation in how cheap and easy it can be to record sleep in the subject's own home. Here we present a data set of 120 home recordings of sleep, in which subjects use self-applied ear-EEG monitoring equipment. We compare this data set to a previously recorded data set with both ear-EEG and polysomnography, which was applied by an expert. Clinical relevance - On all tested metrics, self applied sleep recordings behaved the same as expert applied. This indicates that ear-EEG can reliably be used as a home sleep monitor, even when subjects apply the equipment themselves.

Sleep Monitoring Using Ear-Centered Setups: Investigating the Influence From Electrode Configurations.

Modern sleep monitoring development is shifting towards the use of unobtrusive sensors combined with algorithms for automatic sleep scoring. Many different combinations of wet and dry electrodes, ear-centered, forehead-mounted or headband-inspired designs have been proposed, alongside an ever growing variety of machine learning algorithms for automatic sleep scoring. OBJECTIVE: Among candidate positions, those in the facial area and around the ears have the benefit of being relatively hairless, and in our view deserve extra attention. In this paper, we seek to determine the limits to sleep monitoring quality within this spatial constraint. METHODS: We compare 13 different, realistic sensor setups derived from the same data set and analysed with the same pipeline. RESULTS: All setups which include both a lateral and an EOG derivation show similar, state-of-the-art performance, with average Cohen's kappa values of at least 0.80. CONCLUSION: If large electrode distances are used, positioning is not critical for achieving large sleep-related signal-to-noise-ratio, and hence accurate sleep scoring. SIGNIFICANCE: We argue that with the current competitive performance of automated staging approaches, there is a need for establishing an improved benchmark beyond current single human rater scoring.

Investigation of low dimensional feature spaces for automatic sleep staging.

BACKGROUND: Automatic sleep stage classification depends crucially on the selection of a good set of descriptive features. However, the selection of a feature set with an appropriate low computational cost without compromising classification performance is still a challenge. This study attempts to represent sleep EEG patterns using a minimum number of features, without significant performance loss. METHODS: Three feature selection algorithms were applied to a high dimensional feature space comprising 84 features. These methods were based on a bootstrapping approach guided by Gini ranking and mutual information between the features. The algorithms were tested on three scalp electroencephalography (EEG) and one ear-EEG datasets. The relationship between the information carried by different features was investigated using mutual information and illustrated by a graphical clustering tool. RESULTS: The minimum number of features that can represent the whole feature set without performance loss was found to range between 5 and 11 for different datasets. In ear-EEG, 7 features based on Continuous Wavelet Transform (CWT) resulted in similar performance as the whole set whereas in two scalp EEG datasets, the difference between minimal CWT set and the whole set was statistically significant (0.008 and 0.017 difference in average kappa). Features were divided into groups according to the type of information they carry. The group containing relative power features was identified as the most informative feature group in sleep stage classification, whereas the group containing non-linear features was found to be the least informative. CONCLUSIONS: This study showed that EEG sleep staging can be performed based on a low dimensional feature space without significant decrease in sleep staging performance. This is especially important in the case of wearable devices like ear-EEG where low computational complexity is needed. The division of the feature space into groups of features, and the analysis of the distribution of feature groups for different feature set sizes, is helpful in the selection of an appropriate feature set.

EEGs Vary Less Between Lab and Home Locations Than They Do Between People.

Given the rapid development of light weight EEG devices which we have witnessed the past decade, it is reasonable to ask to which extent neuroscience could now be taken outside the lab. In this study, we have designed an EEG paradigm well suited for deployment "in the wild." The paradigm is tested in repeated recordings on 20 subjects, on eight different occasions (4 in the laboratory, 4 in the subject's own home). By calculating the inter subject, intra subject and inter location variance, we find that the inter location variation for this paradigm is considerably less than the inter subject variation. We believe the paradigm is representative of a large group of other relevant paradigms. This means that given the positive results in this study, we find that if a research paradigm would benefit from being performed in less controlled environments, we expect limited problems in doing so.

Ear-EEG for sleep assessment: a comparison with actigraphy and PSG.

PURPOSE: To assess automatic sleep staging of three ear-EEG setups with different electrode configurations and compare performance with concurrent polysomnography and wrist-worn actigraphy recordings. METHODS: Automatic sleep staging was performed for single-ear, single-ear with ipsilateral mastoid, and cross-ear electrode configurations, and for actigraphy data. The polysomnography data were manually scored and used as the gold standard. The automatic sleep staging was tested on 80 full-night recordings from 20 healthy subjects. The scoring performance and sleep metrics were determined for all ear-EEG setups and the actigraphy device. RESULTS: The single-ear, the single-ear with ipsilateral mastoid setup, and the cross-ear setup performed five class sleep staging with kappa values 0.36, 0.63, and 0.72, respectively. For the single-ear with mastoid electrode and the cross-ear setup, the performance of the sleep metrics, in terms of mean absolute error, was better than the sleep metrics estimated from the actigraphy device in the current study, and also better than current state-of-the-art actigraphy studies. CONCLUSION: A statistically significant improvement in both accuracy and kappa was observed from single-ear to single-ear with ipsilateral mastoid, and from single-ear with ipsilateral mastoid to cross-ear configurations for both two and five-sleep stage classification. In terms of sleep metrics, the results were more heterogeneous, but in general, actigraphy and single-ear with ipsilateral mastoid configuration were better than the single-ear configuration; and the cross-ear configuration was consistently better than both the actigraphy device and the single-ear configuration.

Predicting Sleep Classification Performance without Labels.

When generating automatic sleep reports with mobile sleep monitoring devices, it is crucial to have a good grasp of the reliability of the result. In this paper, we feed features derived from the output of a sleep scoring algorithm to a 'regression ensemble' to estimate the quality of the automatic sleep scoring. We compare this estimate to the actual quality, calculated using a manual scoring of a concurrent polysomnography recording. We find that it is generally possible to estimate the quality of a sleep scoring, but with some uncertainty ('root mean squared error' between estimated and true Cohen's kappa is 0.078). We expect that this method could be useful in situations with many scored nights from the same subject, where an overall picture of scoring quality is needed, but where uncertainty on single nights is less of an issue.

Muscle Activity Detection during Sleep by Ear-EEG.

Muscle activation during sleep is an important biomarker in the diagnosis of several sleep disorders and neurodegenerative diseases. Muscle activity is typically assessed manually based on the EMG channels from polysomnography recordings. Ear-EEG provides a mobile and comfortable alternative for sleep assessment. In this study, ear-EEG was used to automatically detect muscle activities during sleep. The study was based on a dataset comprising four full night recordings from 20 healthy subjects with concurrent polysomnography and ear-EEG. A binary label, active or relax, extracted from the chin EMG was assigned to selected 30 s epoch of the sleep recordings in order to train a classifier to predict muscle activation. We found that the ear-EEG based classifier detected muscle activity with an accuracy of 88% and a Cohen's kappa value of 0.71 relative to the labels derived from the chin EMG channels. The analysis also showed a significant difference in the distribution of muscle activity between REM and non-REM sleep.

Automatic sleep stage classification based on subcutaneous EEG in patients with epilepsy.

BACKGROUND: The interplay between sleep structure and seizure probability has previously been studied using electroencephalography (EEG). Combining sleep assessment and detection of epileptic activity in ultralong-term EEG could potentially optimize seizure treatment and sleep quality of patients with epilepsy. However, the current gold standard polysomnography (PSG) limits sleep recording to a few nights. A novel subcutaneous device was developed to record ultralong-term EEG, and has been shown to measure events of clinical relevance for patients with epilepsy. We investigated whether subcutaneous EEG recordings can also be used to automatically assess the sleep architecture of epilepsy patients. METHOD: Four adult inpatients with probable or definite temporal lobe epilepsy were monitored simultaneously with long-term video scalp EEG (LTV EEG) and subcutaneous EEG. In total, 11 nights with concurrent recordings were obtained. The sleep EEG in the two modalities was scored independently by a trained expert according to the American Academy of Sleep Medicine (AASM) rules. By using the sleep stage labels from the LTV EEG as ground truth, an automatic sleep stage classifier based on 30 descriptive features computed from the subcutaneous EEG was trained and tested. RESULTS: An average Cohen's kappa of [Formula: see text] was achieved using patient specific leave-one-night-out cross validation. When merging all sleep stages into a single class and thereby evaluating an awake-sleep classifier, we achieved a sensitivity of 94.8% and a specificity of 96.6%. Compared to manually labeled video-EEG, the model underestimated total sleep time and sleep efficiency by 8.6 and 1.8 min, respectively, and overestimated wakefulness after sleep onset by 13.6 min. CONCLUSION: This proof-of-concept study shows that it is possible to automatically sleep score patients with epilepsy based on two-channel subcutaneous EEG. The results are comparable with the methods currently used in clinical practice. In contrast to comparable studies with wearable EEG devices, several nights were recorded per patient, allowing for the training of patient specific algorithms that can account for the individual brain dynamics of each patient. Clinical trial registered at ClinicalTrial.gov on 19 October 2016 (ID:NCT02946151).

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.

Discrimination of Sleep Spindles in Ear-EEG.

Sleep spindles are brief oscillatory events observed in EEG measurements during sleep, related to both sleep staging and basic neuroscience. The objective of this study was to investigate to which extent sleep spindles are observable from ear-EEG. The analysis was based on single-night recordings from 12 subjects, wearing both a polysomnography setup and two light-weight mobile EEG devices (ear-EEG). By introducing a sleep spindle index capable of discriminating between epochs with distinct spindles and distinctly spindle-free epochs, we describe to which extent the most clear cut sleep spindles (as labeled using scalp EEG) can be detected using ear-EEG. We find that ear-EEG can be used to detect sleep spindles, at a performance level similar to scalp derivations. We speculate that part of the observed discrepancy between ear-EEG and the gold standard (scalp EEG) could be caused by the visibility of different spindles in the ear-EEG.

Machine-learning-derived sleep-wake staging from around-the-ear electroencephalogram outperforms manual scoring and actigraphy.

Quantification of sleep is important for the diagnosis of sleep disorders and sleep research. However, the only widely accepted method to obtain sleep staging is by visual analysis of polysomnography (PSG), which is expensive and time consuming. Here, we investigate automated sleep scoring based on a low-cost, mobile electroencephalogram (EEG) platform consisting of a lightweight EEG amplifier combined with flex-printed cEEGrid electrodes placed around the ear, which can be implemented as a fully self-applicable sleep system. However, cEEGrid signals have different amplitude characteristics to normal scalp PSG signals, which might be challenging for visual scoring. Therefore, this study evaluates the potential of automatic scoring of cEEGrid signals using a machine learning classifier ("random forests") and compares its performance with manual scoring of standard PSG. In addition, the automatic scoring of cEEGrid signals is compared with manual annotation of the cEEGrid recording and with simultaneous actigraphy. Acceptable recordings were obtained in 15 healthy volunteers (aged 35 ± 14.3 years) during an extended nocturnal sleep opportunity, which induced disrupted sleep with a large inter-individual variation in sleep parameters. The results demonstrate that machine-learning-based scoring of around-the-ear EEG outperforms actigraphy with respect to sleep onset and total sleep time assessments. The automated scoring outperforms human scoring of cEEGrid by standard criteria. The accuracy of machine-learning-based automated scoring of cEEGrid sleep recordings compared with manual scoring of standard PSG was satisfactory. The findings show that cEEGrid recordings combined with machine-learning-based scoring holds promise for large-scale sleep studies.

Sleep EEG Derived From Behind-the-Ear Electrodes (cEEGrid) Compared to Standard Polysomnography: A Proof of Concept Study.

Electroencephalography (EEG) recordings represent a vital component of the assessment of sleep physiology, but the methodology presently used is costly, intrusive to participants, and laborious in application. There is a recognized need to develop more easily applicable yet reliable EEG systems that allow unobtrusive long-term recording of sleep-wake EEG ideally away from the laboratory setting. cEEGrid is a recently developed flex-printed around-the-ear electrode array, which holds great potential for sleep-wake monitoring research. It is comfortable to wear, simple to apply, and minimally intrusive during sleep. Moreover, it can be combined with a smartphone-controlled miniaturized amplifier and is fully portable. Evaluation of cEEGrid as a motion-tolerant device is ongoing, but initial findings clearly indicate that it is very well suited for cognitive research. The present study aimed to explore the suitability of cEEGrid for sleep research, by testing whether cEEGrid data affords the signal quality and characteristics necessary for sleep stage scoring. In an accredited sleep laboratory, sleep data from cEEGrid and a standard PSG system were acquired simultaneously. Twenty participants were recorded for one extended nocturnal sleep opportunity. Fifteen data sets were scored manually. Sleep parameters relating to sleep maintenance and sleep architecture were then extracted and statistically assessed for signal quality and concordance. The findings suggest that the cEEGrid system is a viable and robust recording tool to capture sleep and wake EEG. Further research is needed to fully determine the suitability of cEEGrid for basic and applied research as well as sleep medicine.

Automatic sleep staging using ear-EEG.

BACKGROUND: Sleep and sleep quality assessment by means of sleep stage analysis is important for both scientific and clinical applications. Unfortunately, the presently preferred method, polysomnography (PSG), requires considerable expert assistance and significantly affects the sleep of the person under observation. A reliable, accurate and mobile alternative to the PSG would make sleep information much more readily available in a wide range of medical circumstances. NEW METHOD: Using an already proven method, ear-EEG, in which electrodes are placed inside the concha and ear canal, we measure cerebral activity and automatically score the sleep into up to five stages. These results are compared to manual scoring by trained clinicians, based on a simultaneously recorded PSG. RESULTS: The correspondence between manually scored sleep, based on the PSG, and the automatic labelling, based on ear-EEG data, was evaluated using Cohen's kappa coefficient. Kappa values are in the range 0.5-0.8, making ear-EEG relevant for both scientific and clinical applications. Furthermore, a sleep-wake classifier with leave-one-out cross validation yielded specificity of 0.94 and sensitivity of 0.52 for the sleep stage. COMPARISON WITH EXISTING METHOD(S): Ear-EEG based scoring has clear advantages when compared to both the PSG and other mobile solutions, such as actigraphs. It is far more mobile, and potentially cheaper than the PSG, and the information on sleep stages is far superior to a wrist-based actigraph, or other devices based solely on body movement. CONCLUSIONS: This study shows that ear-EEG recordings carry information about sleep stages, and indicates that automatic sleep staging based on ear-EEG can classify sleep stages with a level of accuracy that makes it relevant for both scientific and clinical sleep assessment.

On the Keyhole Hypothesis: High Mutual Information between Ear and Scalp EEG.

We propose and test the keyhole hypothesis-that measurements from low dimensional EEG, such as ear-EEG reflect a broadly distributed set of neural processes. We formulate the keyhole hypothesis in information theoretical terms. The experimental investigation is based on legacy data consisting of 10 subjects exposed to a battery of stimuli, including alpha-attenuation, auditory onset, and mismatch-negativity responses and a new medium-long EEG experiment involving data acquisition during 13 h. Linear models were estimated to lower bound the scalp-to-ear capacity, i.e., predicting ear-EEG data from simultaneously recorded scalp EEG. A cross-validation procedure was employed to ensure unbiased estimates. We present several pieces of evidence in support of the keyhole hypothesis: There is a high mutual information between data acquired at scalp electrodes and through the ear-EEG "keyhole," furthermore we show that the view-represented as a linear mapping-is stable across both time and mental states. Specifically, we find that ear-EEG data can be predicted reliably from scalp EEG. We also address the reverse view, and demonstrate that large portions of the scalp EEG can be predicted from ear-EEG, with the highest predictability achieved in the temporal regions and when using ear-EEG electrodes with a common reference electrode.

Threshold Games and Cooperation on Multiplayer Graphs.

OBJECTIVE: The study investigates the effect on cooperation in multiplayer games, when the population from which all individuals are drawn is structured-i.e. when a given individual is only competing with a small subset of the entire population. METHOD: To optimize the focus on multiplayer effects, a class of games were chosen for which the payoff depends nonlinearly on the number of cooperators-this ensures that the game cannot be represented as a sum of pair-wise interactions, and increases the likelihood of observing behaviour different from that seen in two-player games. The chosen class of games are named "threshold games", and are defined by a threshold, M > 0, which describes the minimal number of cooperators in a given match required for all the participants to receive a benefit. The model was studied primarily through numerical simulations of large populations of individuals, each with interaction neighbourhoods described by various classes of networks. RESULTS: When comparing the level of cooperation in a structured population to the mean-field model, we find that most types of structure lead to a decrease in cooperation. This is both interesting and novel, simply due to the generality and breadth of relevance of the model-it is likely that any model with similar payoff structure exhibits related behaviour. More importantly, we find that the details of the behaviour depends to a large extent on the size of the immediate neighbourhoods of the individuals, as dictated by the network structure. In effect, the players behave as if they are part of a much smaller, fully mixed, population, which we suggest an expression for.

Reference configurations for ear-EEG steady-state responses.

Ear-EEG is a non-invasive EEG recording method, where EEG is recorded from electrodes placed in the ear. Ear-EEG could be implemented into hearing aids, and provide neurofeedback for e.g. objective hearing assessment through measurements of the auditory steady-state response. In cases where the objective is to measure a specific feature of an event-related potential, there will be a subject specific optimal reference configuration. This work presents a method for optimizing the reference configuration for steady-state type potentials. For given electrode positions, the method maximizes the signal-to-noise (SNR) ratio of the first harmonic of the steady-state response. This is obtained by estimating a set of weights applied to the electrode signals. The method was validated on a dataset recorded from 12 subjects. The weights were estimated from one part of the dataset, and the validation was performed on another part of the dataset. For all subjects the proposed method demonstrated a robust SNR estimate, yielding on par or better SNR compared to other well-known methods.