Towards Integration of Sleep Quality in Sports Monitoring – Improving Wearable Sleep Detection Algorithms through Respiratory Information

Krauß D, Richer R, Küderle A, Rohleder N, Eskofier B (2022)

Publication Language: English

Publication Type: Conference contribution, Abstract of lecture

Publication year: 2022

Series: 13th World Congress of Performance Analysis of Sport 2022 & 13th International Symposium on Computer Science in Sport 2022

Book Volume: 13

Event location: Wien

Abstract

Sleep affects a variety of cognitive and physiological functions like reaction time, muscle strength, and motivation (Halson and Juliff 2017). Thus, sufficient sleep and regeneration are crucial for performance in sports. Nevertheless, current data suggests that many elite athletes do not experience optimal sleep quality or quantity (Halson and Juliff 2017). To offer athletes the possibility to adjust their sleep and gain better performance, it is important to reliably detect sleep. The gold standard for sleep detection is Polysomnography (PSG). However, long-term sleep tracking via PSG is not feasible, due to the requirement of performing it in a laboratory setting (Iber et al. 2004). Wearable sensors are a promising alternative since they are low-cost and widely used throughout the sports community. During sleep, body movements decrease compared to wakefulness (Wilde-Frenz and Schulz 1983). In addition, cardiac and respiratory activity changes during sleep (Douglas et al. 1982; Silvani and Dampney 2013). However, current commercial systems are mostly based on actigraphy for unobtrusive sleep detection (Imtiaz 2021). But, movement-based systems tend to overestimate sleep due to a lack of movements shortly before falling asleep or in short periods of wakefulness (De Zambotti et al. 2019). Previous research shows promising improvements in sleep detection by combining actigraphy with cardiac and respiratory information (Devot, Dratwa, and Naujokat 2010). Recent work has only assessed the combination of actigraphy and heart rate variability (HRV) but has not investigated respiration in large-scale studies (Zhai et al. 2020). For that reason, this work aims to compare actigraphy-based sleep detection with multimodal approaches combining actigraphy and HRV, and to investigate the influence of adding respiratory information in a large-scale dataset by comparing different machine learning algorithms. The data used in this work were collected in a sleep study of 2,237 participants, which contains actigraphy, ECG, and unattended overnight PSG recordings including respiratory information (Zhang et al. 2018). After excluding participants with incomplete data, 1,740 remained for further analysis. In total, 370 actigraphy features, as well as 30 HRV features were extracted according to Zhai et al. (2020). In addition, 62 respiration rate variability (RRV) features were extracted using the Neurokit2 library (Makowski et al. 2021). Results show that the best-performing algorithm to discriminate between sleep and wake phases was a Multi-Layer Perceptron (MLP). To find the best set of hyperparameters, a grid search with embedded 5-fold cross validation was performed over a defined search space. In this work, adding respiratory information significantly improved the key metrics of assessing sleep/wake detection performance (Figure 1). Adding RRV features resulted in a larger boost in detection accuracy than adding HRV to actigraphy (Table 1). In particular, specificity, which is a good marker for assessing the overprediction of sleep, showed a strong increase in performance. The extraction of respiratory information from ECG might be advantageous, as no additional sensors are required. Better sleep detection can help athletes to adjust their sleep habits and, thus, gain better performance. Furthermore, this offers the unique possibility to further investigate the link between physical performance and sleep quality.

Authors with CRIS profile

Additional Organisation(s)

Related research project(s)

How to cite

APA:

Krauß, D., Richer, R., Küderle, A., Rohleder, N., & Eskofier, B. (2022, September). Towards Integration of Sleep Quality in Sports Monitoring – Improving Wearable Sleep Detection Algorithms through Respiratory Information. Paper presentation at 13th World Congress of Performance Analysis of Sport 2022 & 13th International Symposium on Computer Science in Sport 2022, Wien, AT.

MLA:

Krauß, Daniel, et al. "Towards Integration of Sleep Quality in Sports Monitoring – Improving Wearable Sleep Detection Algorithms through Respiratory Information." Presented at 13th World Congress of Performance Analysis of Sport 2022 & 13th International Symposium on Computer Science in Sport 2022, Wien Ed. IACSS & ISPAS, 2022.

BibTeX: Download