Long Term Event
Report on
“SLEEP APNEA DETECTION BASED ON AN EEG SIGNAL AND
VERIFICATION IS DONE USING HARDWARE”
Subject Name: Signals and Systems Subject Code: 22EC430
CO4: Implement and Demonstrate skill sets related to software tools in the analysis of signals and
systems in group
Submitted by
Name USN Marks (20)
Dharti 02JST23UEC017
Poorvi M Bellavara 01JST23UEC034
Under the guidance of
Dr. Shashidhar R.
Associate Professor
Department of ECE5
SJCE, JSSSTU, Mysore
DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING
JSS SCIENCE AND TECHNOLOGY UNIVERSITY
SRI JAYACHAMARAJENDRA COLLEGE OF ENGINEERING
JSS TECHNICAL INSTITUTIONS CAMPUS
MYSURU-570006
2024-2025
INTRODUCTION
Sleep apnea is a prevalent sleep disorder characterized by frequent interruptions in breathing during
sleep, leading to decreased levels of blood oxygen. This approach involves the classification of subject-
specific sleep apnea and non-apnea events. Sleep apnea is a detrimental respiratory disorder
characterized by recurrent breathing interruptions lasting around 10-20 seconds during sleep,
accompanied by a marked decrease in blood oxygen saturation levels. It affects approximately 15% of
men and 5% of women. This disorder poses an elevated risk of numerous adverse health outcomes. In its
early stages, sleep apnea has immediate effects such as fatigue, intermittent hypoxia, irregular heart
rates, and mood swings. Prolonged exposure can escalate into severe complications, including
cardiovascular issues, cognitive decline strokes, and even mortality. The American Association of Sleep
Medicine (AASM) ranks sleep apnea among the top 10% of high-risk disorders leading to death. A
recent 18-year study at the University of Wisconsin-Madison revealed that individuals with sleep apnea
are three times more likely to die from various diseases compared to those without sleep apnea
symptoms. Although treatments are available for Sleep apnea, it still remains undiagnosed and untreated
in approximately 90% of affected individuals. Sleep apnea, diagnosed and monitored through cardio-
respiratory polysomnography (PSG), is classified into obstructive sleep apnea (OSA), central, and mixed
types. OSA occurs when muscle relaxation during sleep blocks the upper airway, leading to reduced or
ceased airflow. Central sleep apnea is caused by neurological disorders affecting breathing signals, while
Mixed Sleep Apnea (MSA) combines features of both. The severity of sleep apnea is measured using the
apnea-hypopnea index (AHI), which indicates the number of apnea and hypopnea episodes per hour of
sleep. Episodes typically last 30 to 60 seconds, causing blood oxygen desaturation and increasing health
risks like heart issues, stroke, diabetes, depression, and headaches. Therefore, developing reliable sleep
apnea detection devices is vital for accurate diagnosis, ongoing monitoring, and effective management of
these health risks [2]. Electroencephalography (EEG) is a well-established technique (Pravdich-
Neminsky, 1913), providing valuable insights into brain activity, with applications both in clinical
practice and in basic and applied neuroscience (Nunez and Srinivasan, 2007). Despite the widespread
adoption, a number of important applications of EEG are prohibited by the requirements for mobility
and discreetness of the EEG equipment used. Examples include brain-computer interfaces (BCI), long-
term monitoring of neurological patients and sleep monitoring. To this end, several EEG recording
methods have recently been proposed (Casson et al., 2010; Debener et al., 2012), including some which
benefit from a relatively unrestricted access to the area around and inside the ears (Looney et al., 2012;
Bleichner et al., 2015; Norton et al., 2015). A very promising solution, belonging to the latter class, is the
ear-EEG platform (Looney et al., 2012; Kidmose et al., 2013) which promises a robust, unobtrusive and
non-invasive means for monitoring the brain activity outside the laboratory, both within and outside a
clinical context [3]. Sleep monitoring, meaning the recording and subsequent analysis of physiological
signals during sleep, is important in the study and diagnosis of a long list of diseases. However, the gold
standard method, the polysomnography (PSG) is both an encumbrance for the patient as well as
expensive to perform. Therefore, the need for monitoring must always be balanced against practical
constraints, and it is rarely feasible to perform more than one or two full-night recordings at a time.
However, in both treatment and study of several diseases, it is very valuable to monitor quality and
extent of patient sleep over extended periods (weeks or longer). Important examples are idiopathic
hypersomnia and circadian rhythm sleep-wake disorders, but studies such as Ciano et al. 2017 and Ye et
al. 2018 also demonstrate the value of longitudinal sleep monitoring in more general care as well as
research. These circumstances (the limitations of the PSG and the need for long term sleep monitoring)
fuel an ongoing effort to develop a cheap, precise alternative to the PSG, suitable for long-term
monitoring. Of special interest in this endeavour is the method of ear-EEG4, in which recording
electrodes are placed in or around the ears. This placement is useful because the area is largely hairless
and relatively hidden (meaning the possibility of good signals and a low-profile device). Previously, a
good correspondence has been shown between ear-EEG and scalp-EEG [4]. In the past decade, multiple
studies have explored the use of digital wearable (e.g. actigraphy) and bed-side (e.g. radar based) sensors
to quantify various aspects of sleep, but failed to capture the neurophysiological signatures that underpin
the quantification of sleep based on the AASM convention. With the introduction of various wearable
EEG sensors (ear-EEG, cEEGrid, sleep zeo, Dreem) which capture brain activity from unconventional
places (e.g. on the forehead, in or around the ears or using a headband), personalized long-term sleep
monitoring on the general population is within reach. Visually reviewing the large amount of time series
sleep data that could be recorded with this new generation of wearable EEG would be time-consuming
and costly, in addition to requiring re-training of the human scorers for each new wearable (which would
be highly inefficient. Initial approaches to automatic sleep scoring were based on traditional, hand-
crafted features, designed using domain knowledge of sleep experts. These features were fed into a
suitable machine learning algorithm [5]. The characteristics of EEG signals depends on the location of
the electrodes, the state of the brain, and the stimulation of the brain e.g. through the peripheral nervous
system or the optic nerve/retina. An auditory stimulus, for instance, will induce modulated patterns into
the EEG, and the changes in the EEG from the otherwise spontaneous EEG are termed auditory evoked
potentials (EPs). Despite the clinical value of EEG, several issues limit its use beyond the clinical
environment. These allow for recording outside the clinical environment, but are largely inadequate for
everyday use. This is both because of their limited ability to integrate into daily activities, requirement
for a skilled assistant, and the fact that the ambulatory systems, like the conventional systems, still have
clearly visible electrodes and leads from the electrodes to the recording device. Despite some advances
in addressing these issues, on-scalp systems inevitably require a means for stable attachment (cap,
headset, and/or adhesive), making the recording process uncomfortable, obtrusive, and stigmatizing [6].
LITEARATURE SURVEY
Mikkelsen et al. presented two open-access datasets—EESM19 and EESM23—designed to explore the
potential of ear-EEG for sleep monitoring in home settings. Across 320 recorded nights involving 30
healthy subjects, the datasets provide comprehensive EEG recordings, some paired with scalp-EEG and
actigraphy for validation. Their study demonstrated that ear-EEG is a viable tool for mobile EEG
research, offering high-quality repeated sleep recordings that could contribute to advancements in
biomedical research and brain-computer interfaces [1].
Alam et al. described an innovative, energy-efficient digital system for real-time detection of sleep apnea
using EEG signals. The system was implemented on an Artix-7 FPGA and aimed at reducing power
consumption while maintaining high detection accuracy [2].
Mikkelsen et al. introduced and thoroughly characterized a novel method for recording
electroencephalographic (EEG) signals using electrodes embedded within the human ear, referred to as
the ear-EEG platform. The primary objective of their work was to investigate whether ear-EEG could
serve as a reliable and non-invasive alternative to traditional scalp EEG, especially in mobile and real-
world environments [3].
Mikkelsen et al. introduced a novel approach for long-term, non-intrusive sleep monitoring using dry-
contact ear-EEG, aiming to bridge the performance gap between lightweight EEG alternatives and the
gold-standard polysomnography (PSG). Their study addressed the need for clinically viable home-based
sleep tracking solutions that are both user-friendly and accurate enough for diagnostic purposes [4].
Mikkelsen et al. investigated the influence of electrode configurations on ear-centered sleep monitoring.
They compared 13 different sensor setups derived from the same dataset, analysing their performance
using identical machine learning pipelines for automatic sleep scoring. The study found that setups
incorporating both lateral and electrooculography (EOG) derivations achieved state-of-the-art sleep
classification performance, with Cohen's kappa values exceeding 0.80. These findings suggest that
electrode distance significantly impacts signal quality and sleep stage classification accuracy [5].
Kidmose et al. introduced the concept of ear-EEG as a discreet, unobtrusive brain monitoring method
using electrodes embedded in a personalized earpiece. They compared ear-EEG with conventional on-
scalp EEG for auditory and visual evoked potentials, utilizing steady-state and transient models across
multiple subjects. Their study demonstrated that ear-EEG signals, despite having lower amplitude, offer
comparable signal-to-noise ratios, making them viable for wearable brain monitoring applications [6].
Mikkelsen et al. explored the use of ear-EEG for automatic sleep staging as a portable alternative to
polysomnography (PSG). Their method involved placing electrodes in the concha and ear canal to record
brain activity, which was then classified into sleep stages using machine learning techniques. The results
showed a strong correlation between ear-EEG-based sleep staging and conventional PSG, with Cohen’s
kappa values ranging from 0.5 to 0.8, indicating its potential for long-term sleep monitoring in scientific
and clinical applications [7].
Delorme and Makeig developed EEGLAB, an open-source MATLAB toolbox for analysing single-trial
EEG dynamics, including Independent Component Analysis (ICA). Their work introduced a
comprehensive graphical user interface that enables EEG data preprocessing, visualization, and
advanced statistical methods such as time-frequency decomposition and event-related spectral
perturbation. The study demonstrated how ICA can effectively separate EEG components related to
neural and artifact signals, paving the way for more precise EEG analysis in neuroscience research [8].
HuyPhan et al. introduced SeqSleepNet, a hierarchical recurrent neural network designed for sequence-
to-sequence automatic sleep staging. Their approach redefines sleep staging as a sequential classification
problem, utilizing a filter bank layer for frequency preprocessing and attention-based recurrent layers for
short-term and long-term temporal modelling. Their study demonstrated that SeqSleepNet outperformed
previous methods, achieving 87.1% accuracy and a Cohen’s kappa of 0.815 on a dataset of 200 subjects,
showing its effectiveness for automating sleep stage classification [9].
Kappel et al. investigated optimized reference configurations for ear-EEG steady-state responses,
proposing a method to enhance signal-to-noise ratio (SNR) through weighted electrode combinations.
Their study validated the approach using auditory steady-state response (ASSR) recordings from 12
subjects, demonstrating that the optimized reference configuration (ORC) consistently yielded
comparable or better SNR compared to conventional methods. The findings highlight the potential of
ear-EEG for neurofeedback applications, including integration into hearing aids for objective hearing
assessments [10].
Kjaer et al. investigated the feasibility of ear-EEG as an alternative to manually scored
polysomnography (PSG) for sleep monitoring. Their study compared automatic sleep scoring based on
ear-EEG with conventional PSG recordings across 20 healthy subjects, each monitored over four nights.
Results showed that ear-EEG provided accurate estimations of sleep metrics such as total sleep time,
sleep efficiency, and wake after sleep onset, although slight discrepancies were observed in REM latency
and N2/N3 fraction estimations. The findings support ear-EEG as a practical and unobtrusive solution
for long-term sleep tracking [11].
Babaei et al. introduced a novel multimodal deep learning approach for detecting sleep apnea events
using physiological signals. Their model combines a convolutional neural network (CNN) with gated
recurrent units (GRU) and an attention mechanism to capture both spatial and temporal features.
Additionally, they introduced a multi-domain feature extractor (MDFE) block with a Squeeze and
Excitation (SE) block to enhance feature representation. Their approach, validated on the Sleep Heart
Health Study (SHHS) dataset, achieved high accuracy (83.87), demonstrating its potential for real-time
monitoring and personalized treatment strategies for sleep disorders [12].
Talal Sarheed Alshammari explored the application of machine learning algorithms for classifying sleep
disorders, emphasizing the importance of automated approaches in enhancing diagnostic accuracy. The
study compared deep learning models, such as artificial neural networks (ANN), with conventional
techniques like support vector machines (SVM) and random forests (RF), optimizing parameters using a
genetic algorithm for improved classification. Results demonstrated that the ANN model achieved the
highest classification accuracy of 92.92%, showcasing the effectiveness of machine learning in sleep
disorders detection and potentially aiding personalized healthcare solutions [13].
Jaehoon et al. conducted a validation study comparing an in-home sleep electroencephalography (EEG)
device with simultaneous polysomnography (PSG) in patients with obstructive sleep apnea (OSA). Their
research assessed sleep staging accuracy, finding that combining EEG with pulse rate and oxygen
saturation (SpO2) improved performance. Statistical analyses demonstrated significant agreement
between the two methods, with over 95% accuracy in Bland-Altman plots and high interclass correlation
coefficients. The study suggests that portable EEG devices provide reliable at-home sleep assessments
for OSA patients, reducing the need for laboratory PSG [14].
Mikkelsen et al. conducted an alternative to expert-applied polysomnography (PSG). Their methodology
involved 120 sleep recordings using ear-EEG, comparing them to an earlier dataset where electrodes
were applied by experts. The results showed similar data quality, sleep metrics, and algorithm
confidence, demonstrating that self-applied ear-EEG is a reliable method for home sleep monitoring[15].
Kappel et al. conducted the study that explores the feasibility of dry-contact electrode ear-EEG as an
improvement over wet-electrode methods for non-invasive and discreet long-term EEG monitoring. A
new ear-EEG platform was developed with actively shielded, nanostructured electrodes embedded in an
individualized soft-earpiece. The technology was tested in a study with 12 subjects across four EEG
paradigms. The dry-contact electrode ear-EEG platform demonstrated comparable performance to
conventional scalp EEG setups, with significant responses across all paradigms, highlighting its potential
for practical, user-friendly EEG monitoring without gel-based electrode interfaces [16].
Gorgolewski et al. developed the Brain Imaging Data Structure (BIDS) to standardize neuroimaging data
organization. BIDS integrates existing formats, metadata, and common practices for easier sharing and
analysis resulting in improved accessibility, reproducibility, and efficiency in neuroimaging research
through structured datasets [17].
Phan et al. explored about sleep transformer, a deep-learning model for automatic sleep staging with
interpretability and uncertainty quantification. It uses a sequence-to-sequence transformer model with
self-attention to enhance explainability and confidence estimation in sleep scoring. This achieved
performance comparable to state-of-the-art models, offering interpretability via attention heat maps and
confidence quantification to assist clinical integration[18].
Phan et al. (2023) explored L-SeqSleepNet, a deep learning model designed for efficient long sequence
modeling in automatic sleep staging. They hypothesized that capturing whole-cycle sleep dependencies
improves staging accuracy and developed a method that folds, processes, and unfolds long EEG
sequences. The model was tested on four EEG datasets, including scalp EEG, in-ear EEG, and around-
the-ear EEG, demonstrating state-of-the-art performance while reducing classification errors in N3 and
REM stages. Results showed that L-SeqSleepNet significantly outperformed existing models, including
SeqSleepNet and SleepTransformer, with improved robustness and computational efficiency[19].
Tabar et al. introduced assessing automatic sleep staging using three different ear-EEG setups and
comparing their performance with polysomnography (PSG) and wrist-worn actigraphy. They conducted
sleep staging on 80 full-night recordings from 20 healthy subjects using single-ear, single-ear with
ipsilateral mastoid, and cross-ear electrode configurations. Their results showed that the cross-ear setup
achieved the highest accuracy (kappa = 0.72), outperforming actigraphy and other ear-EEG
configurations in sleep metrics. The study concluded that cross-ear EEG significantly improves sleep
staging accuracy, making it a promising alternative to traditional PSG[20].
Hemmsen et al. proposed monitoring trait-like characteristics of sleep EEG using ear-EEG. They
conducted an in-home study with 20 subjects over multiple nights, comparing ear-EEG with scalp-EEG.
Their results showed high intra-individual similarity, confirming ear-EEG as a stable and reliable method
for tracking neural signatures[21].
Borup et al. proposed automatic sleep scoring using patient-specific ensemble models and knowledge
distillation for ear-EEG data. They trained ensembles of neural networks and distilled them into a single
model, improving predictive performance by 2.4%. Their approach leverages semi-supervised learning,
allowing the model to adapt to individual sleep patterns using unlabeled personal data[22].
MOTIVATION
The function of sleep in overall physical health and well-being is significant with regards to memory
consolidation, emotional control and physical recovery. Unoptimized sleep, whether qualitative or
quantitative, leads to dementia, depression, cardiometabolic health issues, and increased mortality.
Approximately one billion middle-aged adults worldwide are estimated to be affected by obstructive
sleep apnea (OSA) with or without symptoms. Furthermore, while the number of patients with OSA in
Japan is estimated to be over 9 million, the number of PSG tests carried out annually is approximately
800,007 indicating a low test-to-number of patients. As of recent estimates, approximately 104 million
Indians may be suffering from obstructive sleep apnea (OSA), according to a study conducted by the All
India Institute of Medical Sciences (AIIMS) in New Delhi. This figure represents about 11% of the adult
population. To address these limitations, several comparative validations were conducted in both
laboratory and home environments using equipment such as unattended PSG. The latest developments
on wearable technologies and signal processing have enabled the emergence of a variety of portable,
user friendly sleep monitoring solutions. Notably, ear-centric EEG technologies offer a unique approach
that symbolizes the integration of comfort, portability, and effectiveness for neural recording solutions.
Notably, ear-centric EEG technologies offer a unique approach that symbolizes the integration of
comfort, portability, and effectiveness for neural recording solutions.
PROBLEM STATEMENT
To develop a machine learning-based system for the detection of sleep apnea events using ear-EEG
signals that delivers PSG-comparable accuracy in an unobtrusive and user-friendly format suitable for
home-based and continuous monitoring.
OBJECTIVES
To collect the datasets based on the sleep recordings, all performed using the ear-EEG method.
https://doi.org/10.1038/s41597-025-04579-8
To preprocess the signals of each dataset and removing the background noises present, if any.
To feature extract the data which further reduces the number of resources required for processing,
retaining the relevant and accurate information.
To input the extracted features into machine learning models trained to classify sleep stages and
detect apnea events. Various models can be employed depending on the complexity of the task.
To design the system in order to operate in real-time, with a lightweight processing unit
analysing incoming EEG data streams and making real-time predictions.
To validate it through hardware interface which could be understandable by the humans.
METHODOLOGY
DATASET EAR- FEATURE
DATA PRE- DATA
EEG EXTRACTION
PROCESSING AUGMENTATION
ACQUISITION
ML/DL MODEL
OUTPUT AND APNEA EVENT
EVALUATION DETECTION (CNN, CNN-LSTM)
Fig.1 Methodology for detecting sleep apnea
In the fig.1, the methodology for detecting sleep apnea using ear-EEG and machine learning is shown.
1. Ear-EEG Acquisition: EEG signals are first collected using an ear-EEG device, which consists of
strategically positioned electrodes within or around the ear to capture electrical activity of the
brain. The primary advantage of using ear-EEG lies in its ability to collect high-fidelity data in a
comfortable and unobtrusive manner, making it suitable for long-term home-based monitoring.
These raw EEG signals, once acquired, are then transmitted to a processing unit for further
analysis.
2. Data augmentation: In a sleep apnea event data augmentation involves generating synthetic or
altered physiological data (like ECG, airflow, or oxygen saturation) to enhance model training,
especially when real data is limited. Techniques include adding noise, time warping, signal
flipping, or using generative models like GANs. This improves model generalization and
robustness to variability in patient signals.
3. Preprocessing: This step is critical for enhancing signal quality and involves filtering the data to
remove noise and artifacts that can result from muscle movements, environmental interference,
or hardware limitations. Signal processing techniques such as band-pass filtering, notch filtering,
and artifact rejection methods are applied to retain meaningful components of the EEG signal.
Once the signal is cleaned, it is segmented into periods, often in windows of 30 seconds, which
aligns with standard practices for sleep analysis.
4. Feature Extraction/Raw Signal Input: In this phase, significant features are derived from the EEG
signal, capturing essential information from both the time and frequency domains. Time-domain
features include mean, variance, and skewness, while frequency-domain features often include
power spectral density across relevant EEG bands such as delta, theta, alpha, and beta.
Additionally, nonlinear features like entropy or fractal dimension may be extracted to capture
complex signal patterns indicative of apnoeic events.
5. ML/DL Model: The extracted features are input into machine learning models trained to classify
sleep stages and detect apnea events. Various models can be employed depending on the
complexity of the task, such as Support Vector Machines (SVM), Random Forests, or deep
learning models including Convolutional Neural Networks (CNNs) and Long Short-Term
Memory (LSTM) networks. These models are trained on labelled datasets where apnea events
and sleep stages have been annotated by experts. The training process involves dividing the data
into training, validation, and test sets, ensuring robust evaluation and preventing overfitting.
6. Apnea Event Detection: Once trained, the model is capable of identifying patterns in new, unseen
data. In the deployment phase, the system is designed to operate in real-time, with a lightweight
processing unit analysing incoming EEG data streams and making real-time predictions. The
predicted outcomes, such as detection of apnoeic events or classification of sleep stages, are then
used to generate actionable insights.
7. Output and Evaluation: Finally, the performance of the system is evaluated using metrics such as
accuracy, sensitivity, specificity, and area under the curve (AUC). The output can be visualized
via a user-friendly interface or integrated into a health monitoring app, providing real-time alerts
and reports for users or clinicians. The entire system is optimized for low power consumption
and portability, making it suitable for practical use in everyday environments. This methodology
ensures the development of a reliable, efficient, and user-centric solution for sleep apnea
detection using ear-EEG and machine learning.
PERT CHART/TIMELINE
OBJECTIVES WEEK WEEK WEEK WEEK WEEK WEEK WEEK WEEK
1 2 3 4 5 6 7 8
Synopsis
Dataset
Preparation
Data
Preprocessing
Data
Augmentation
Feature
extraction
Model
training
Explainability
visualization
Fig2.Pert Chart
REFERENCES
[1]. Bjarke Mikkelsen, K., Rezai Tabar, Y., Rævsbæk Birch, L. et al. (2025) “Ear-EEG sleep
monitoring data sets”. Sci Data 12, 301 https://doi.org/10.1038/s41597-025-04579-8
[2]. M. S. Alam, Y. Siddiqui, M. Hasan, O. Farooq and Y. Himeur (2024),"Energy-Efficient FPGA
Based Sleep Apnea Detection Using EEG Signals," in IEEE Access, vol. 12, pp. 40182-40195, ,
doi: 10.1109/ACCESS.2024.3374223.
[3]. Mikkelsen KB, Kappel SL, Mandic DP and Kidmose P (2015) “EEG Recorded from the Ear:
Characterizing the Ear-EEG Method”. Front. Neurosci. 9:438. doi: 10.3389/fnins.2015.00438
[4]. Mikkelsen, K.B., Tabar, Y.R., Kappel, S.L. et al. (2019), “Accurate whole-night sleep monitoring
with dry-contact ear-EEG”. Sci Rep 9, 16824. https://doi.org/10.1038/s41598-019-53115-3
[5]. K. B. Mikkelsen, H. Phan, M. L. Rank, M. C. Hemmsen, M. de Vos and P. Kidmose (2022),
"Sleep Monitoring Using Ear-Centered Setups: Investigating the Influence from Electrode
Configurations," in IEEE Transactions on Biomedical Engineering, vol. 69, no. 5, pp. 1564-
1572, doi: 10.1109/TBME.2021.3116274.
[6]. P. Kidmose, D. Looney, M. Ungstrup, M. L. Rank and D. P. Mandic(2013) "A Study of Evoked
Potentials From Ear-EEG," in IEEE Transactions on Biomedical Engineering, vol. 60, no. 10, pp.
2824-2830, doi: 10.1109/TBME.2013.2264956.
[7]. Mikkelsen, K.B., Villadsen, D.B., Otto, M. et al. “Automatic sleep staging using ear-
EEG”. BioMed Eng OnLine 16, 111 (2017). https://doi.org/10.1186/s12938-017-0400-5
[8]. Arnaud Delorme, Scott Makeig(2004) “EEGLAB: an open source toolbox for analysis of single-
trial EEG dynamics including independent component analysis”,Journal of Neuroscience
Methods, Volume 134, Issue 1,Pages 9-21,ISSN 0165-0270,
https://doi.org/10.1016/j.jneumeth.2003.10.009.
[9]. H. Phan, F. Andreotti, N. Cooray, O. Y. Chén and M. De Vos,( 2019) "SeqSleepNet: End-to-End
Hierarchical Recurrent Neural Network for Sequence-to-Sequence Automatic Sleep Staging,"
in IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 27, no. 3, pp. 400-
410, doi: 10.1109/TNSRE.2019.2896659.
[10]. S. L. Kappel, C. B. Christensen, K. B. Mikkelsen and P. Kidmose(2016), "Reference
configurations for ear-EEG steady-state responses," 38th Annual International Conference of the
IEEE Engineering in Medicine and Biology Society (EMBC), Orlando, FL, USA, 2016, pp. 5689-
5692, doi: 10.1109/EMBC.2016.7592018.
[11]. Kjaer TW, Rank ML, Hemmsen MC, Kidmose P, Mikkelsen K.,(2022) “Repeated automatic
sleep scoring based on ear-EEG is a valuable alternative to manually scored polysomnography”.
PLOS Digit Health.;1(10):e0000134. doi: 10.1371/journal.pdig.0000134. PMID: 36812563;
PMCID: PMC9931275.
[12]. A. F. Babaei, J. Tanha, M. A. Balafar and S. Roshan, (2025),"A Novel Multimodal Deep
Learning Approach With Loss Function for Detection of Sleep Apnea Events," in IEEE Access,
vol. 13, pp. 52085-52099, doi: 10.1109/ACCESS.2025.3552254.
[13]. T. S. Alshammari,(2024), "Applying Machine Learning Algorithms for the Classification of
Sleep Disorders" in IEEE Access, vol. 12, pp. 36110-36121,
doi:10.1109/ACCESS.2024.3374408.
[14]. Seol, J., Chiba, S., Kawana, F. et al. (2024) “Validation of sleep-staging accuracy for an in-home
sleep electroencephalography device compared with simultaneous polysomnography in patients
with obstructive sleep apnea”, Sci Rep 14, 3533 https://doi.org/10.1038/s41598-024-53827-1
[15]. Mikkelsen, K. B. et al. (2022), “ Self-applied ear-EEG for sleep monitoring at home”. In 2022
44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society
(EMBC), 3135-3138, https://doi.org/10.1109/EMBC48229.2022.9871076, ISSN:2694-0604.
[16]. S. L. Kappel, M. L. Rank, H. O. Toft, M. Andersen and P. Kidmose,(2019), "Dry-Contact
Electrode Ear-EEG," in IEEE Transactions on Biomedical Engineering, vol. 66, no. 1, pp. 150-
158,doi: 10.1109/TBME.2018.2835778.
[17]. Gorgolewski, K. J. et al.(2016), “The brain imaging data structure, a format for organizing and
describing outputs of neuroimaging experiments”. Scientific Data 3, 160044,
https://doi.org/10.1038/sdata.2016.44.
[18]. H. Phan, K. Mikkelsen, O. Y. Chén, P. Koch, A. Mertins and M. De Vos,(2022),
"SleepTransformer: Automatic Sleep Staging With Interpretability and Uncertainty
Quantification," in IEEE Transactions on Biomedical Engineering, vol. 69, no. 8, pp. 2456-2467,
doi: 10.1109/TBME.2022.3147187
[19]. Phan, H. et al. (2023), “L-SeqSleepNet: Whole-cycle Long Sequence Modelling for Automatic
Sleep Staging”. IEEE Journal of Biomedical and Health Informatics1. 10,
https://doi.org/10.1109/JBHI.2023.3303197.
[20]. Tabar, Y. R. et al. (2020), “ Ear-EEG for sleep assessment: a comparison with actigraphy and
PSG”. Sleep and Breathing https://doi.org/10.1007/s11325-020-02248-1.
[21]. Hemmsen, M., Mikkelsen, K., Rank, M. & Kidmose, P. (2021), “ Long-term monitoring of trait-
like characteristics of the sleep electroencephalogram using ear-EEG”. Sleep 44, A109–A109,
https://doi.org/10.1093/sleep/zsab072.271.
[22]. Borup, K., Kidmose, P., Phan, H. & Mikkelsen, K. (2023), “ Automatic sleep scoring using
patient-specific ensemble models and knowledge distillation for ear-EEG data”. Biomedical
Signal Processing and Control 81, 104496, https://doi.org/10.1016/j.bspc.2022.104496.
