Introduction

According to medical guidelines, measurements of sleep shoud be conducted either in a hospital by polysomnography (PSG) or in the home environment by sleep logs and actigraphy. PSG is considered to be gold standard however, it requires a vast amount of ressources and ample capacities to perform, which can lead to long waiting lists. In comparison to PSG, the other mentioned ambulatory techniques are not as accurate in recording sleep profiles. These issues drive the motivation for developing, clinically testing, and validating new ambulatory methods of measuring sleep profiles objectively in an outpatient setting.

Methods

Recent developments such as self-applicable EEG based medical proved devices can objectively measure sleep profiles similar to those obtained by traditional PSG. Current studies in sleep labs use such devices to investigate different aspects of sleep. For example, differences in accuracy between home and lab measurement, variability of sleep pattern across several days, and capability of phenotyping of sleep disorders. .

Results

Additionally, there are other surrogate parameters that are designed to estimate sleep profiles which are of great interest such as wearables. These include rings, smartwatches, mattress sensors, bedside radar devices, and even smartphone applications using built-in sensors. Such devices are initially sold as consumer products. Later, through collaborations with scientists and clinicians, many endeavor to transition them into clinically validated diagnostic devices via validation research, ultimately resulting in a medically certified product.Typically, these validation studies have a specific target population namely insomnia, sleep-wake-rhythm disorders, and sleep apnea dependet on the device capability. This will lead to faster and more tailored specific treatment for patients with sleep disorders, as well as long-term monitoring of clinical outcomes. Moreover, it will help identify limitations and pitfalls in clinical practice.

Conclusion

New developments of measurement in the home environment have the capability to retrive sleep profiles in a less-obstrusive way, to produce more realistic results, and increase the identification of previously undiagnosed indivuals.

Consequently, these developments could represent a significant advancement in enhancing the clinical management of individuals with sleep disorders.

Arousals are a crucial aspect of human sleep and are linked to sleep-related disorders, such as obstructive sleep apnea or periodic limb movement disorder. Although it is known that arousal is associated with an interaction between the cardiovascular and the cerebral system, the exact underlying functions are not fully understood. Current research indicates a functional coupling between the autonomic and the central nervous system. We investigated this behavior by means of graph analysis to enhance the understanding of the underlying regulatory mechanisms of cardio-cerebral effects influenced by arousals.

We analyzed polysomnography recordings from 2,651 participants in the Sleep Heart Health Study by extracting electroencephalography band power as measure of activity in the central nervous system, and heart rate variability and QT interval variability as measures of activity in the autonomic nervous system. Transfer entropy (TE) was used to characterize the coupling behavior between central and autonomic nervous system measures. We introduced the influence factor ρ to quantify the bidirectional change in TE. To visualize the effects of arousal and sleep stages, we utilized graph analysis.

Our results show a significant increase in TE from the central to the autonomic nervous system by 1.23% (p < 0.05) due to arousals. At the same time, TE from the autonomic to the central nervous system decreased by 1.16% (p < 0.05). Observable changes in graphs were found across all sleep stages, with deep sleep showing the largest difference of ρ between both directions by 15.6%. Consequently, we conclude that arousal from sleep provokes a cardiovascular response primarily through the central, instead of vice versa.

In conclusion, arousals can influence cardiocerebral regulation as a central nervous activation. Further investigations will examine the effects of sex, age, and disease on cardiocerebral effects using graph theory methods for improved quantitative interpretation.

Classification of apnoea events with artificial intelligence based on a segmentation technique

Ralf Seepold, Computer Science, HTWG Konstanz, Konstanz, Germany,

ralf.seepold@htwg-konstanz.de

Ángel Serrano Alarcón, School of Informatics, Reutlingen University, Reutlingen, Germany,

angel.serrano_alarcon@reutlingen-university.de

Natividad Martínez Madrid , School of Informatics, Reutlingen University, Reutlingen, Germany,

natividad.martinez@reutlingen-university.de

Introduction

There are many scientific articles in the literature on the recognition of automatic sleep apnea. However, most such works do not have a method for calculating the exact duration of apnea events or the Apnoea-Hypopnoea Index (AHI). This work presents the development and validation of an artificial intelligence (AI) model using a segmentation technique through an engineering approach to calculate sleep apnea events and AHI.

Methods

We used data from two well-known studies that collected polysomnographic data for calculating apnoea events: the Sleep Heart Health Study (SHHS) and the Multiethnic Atherosclerosis Study (MESA). During the develop-ment of the AI model, we applied as little preprocessing as possible to the data (removing large artifacts and standardization) and fed the model with such data. We worked with three signals (oxygen saturation, heart rate, respiratory, abdominal force) per patient at a 1 Hz for 8 hours. For the AHI calculation, the duration of the ap-noea events is first calculated, verified to be longer than 10 seconds, and counted. The final sum is divided by the total sleep time or, in this case, the total recording time.

Results

After training and evaluating the AI model, results for metrics such as accuracy, specificity, and sensitivity of about 80% were obtained. It is important to remember that this is a binary classification and that we have more examples of non-apnea than apnea. However, this fact is somewhat challenging to counter as we need the whole signal to study the different apnea events and the subsequent AHI calculation. The classification is done on a second-by-second basis.

Conclusion

From the point of view of accuracy, sensitivity, or specificity, a result of approximately 80% was achieved, mak-ing the use of this method promising. As a future step, an artificial intelligence model will be developed for multiple classification tasks to distinguish different types of apnoea.

Sleep Arousal Detection with a Single Channel EEG

Tugce Canbaz Gümüssu, Walter Karlen

Institute of Biomedical Engineering, Ulm University, Ulm, Germany

Introduction

Sleep arousals (SA) are transient increases in vigilance level during sleep, often identified by brief frequency shifts in the electroencephalogram (EEG). It is crucial to investigate the occurrence and characteristics of SA in a natural environment by monitoring them at home. While wearable EEG devices allow for home-based sleep EEG recordings over multiple nights, they typically support only a limited number of channels that do not offer the standard configuration for characterizing SA. Here, we propose a machine learning approach using a single-channel EEG to detect SA.

Methods

Seven participants (2 females, mean age 42 ± 15 years) recorded 29 nights of single-channel EEG using the MHSL-Sleepband v3. Each recording was scored for reference sleep stages and SA labels according to the American Academy of Sleep Medicine (AASM) guidelines. We segmented the EEG into 1 s long segments and extracted from these segments 24 frequency-based features from 5 frequency bands, including delta (2-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), sigma (12-16 Hz) and beta (16-40 Hz), and three time-domain features. We divided the recordings into subject-exclusive training and test sets with approximately 70% of the data assigned to training. We then trained a decision tree (DT) classifier with adjusted class weights and a max depth of 7. We evaluated the classification performance using the accuracy, sensitivity, specificity, and area under receiver operating characteristics curve (AUROC) metrics. We then compared the performance with two approaches from our previous work that were based on head movement and alternative EEG signal processing.

Results

The training set included 29 recordings (4 subjects) with 1720 arousals and the test set included 7 recordings (3 subjects) with 465 arousals. The sensitivity was 83%, the specificity 76%, the accuracy 76%, and the AUROC 0.88. The sensitivity was in the same range as previous approaches (81% head movement and 85% alternative EEG).

Conclusion

Detecting SA using DT on single-channel EEG is feasible. Further work on algorithms to perform more detailed characterization of SA from reduced channel, home-based recordings is needed.

Recent advancements in the field of acoustic stimulation have shown promising effects on the autonomic nervous system, influencing parameters such as heart rate [1] and cognitive performance [2]. In this study, we analyzed polygraphy data from patients with heart failure related central sleep apnea. We compared the Apnea-Hypopnea Index (AHI) during a baseline night with that during a night where acoustic stimulation was applied for 40 minutes before bedtime. Seven out of eight patients showed a significant reduction in AHI during the intervention night (p < 0.01). These preliminary findings suggest that further investigation in a more controlled environment is warranted.