Work package 5: Algorithm development

Lead: University of Eastern Finland


Automatic algorithm development 


The current diagnostic practice for obstructive sleep apnea is based on metrics developed decades ago. These were limited by the analog recordings printed on paper and simplistic analyses. However, these metrics remain diagnostic cornerstones despite the advancements in digital recordings and computational methods.  




To overcome the limitations in the current clinical practice, we aim to develop artificial intelligence-based methods for the estimation of the severity of obstructive sleep apnea and related severe health consequences. We will produce fully automated artificial intelligence-based algorithms to analyze polysomnographic recordings and home sleep apnea tests. To achieve this, we will introduce novel analytical pathways to track novel biomarkers from signals recorded during sleep studies. These pathways will be exploited to represent not only the respiratory event severity but also the short-term physiological changes. Further, this information will be linked with symptoms of obstructive sleep apnea to determine the true severity of the disease. Finally, the aim is to use the information on novel biomarkers to determine the minimum sensor and sampling setups for an accurate diagnosis and severity estimation of obstructive sleep apnea. 




This work package will provide information on which signals should be assessed, sampled, and analyzed in sleep studies. More precisely, it is essential to understand the effect of respiratory event severity on short- and long-term health consequences before novel diagnostic concepts can be implemented in clinical practice. This is of paramount importance to reliably define the severity of the disease reflecting the severe health consequences ultimately enabling the individualization of treatment and targeting the limited treatment resources to those individuals in most need. 


Link to other work packages  


This work package is strongly linked with work packages 3 and 4 which together form a basis for the European Sleep Scoring Manual. Within these three work packages, the SleepRevolution project will determine the most essential signals required for sleep studies to make the recording setup as simple as possible without compromising the diagnostic accuracy. Together these work packages will also provide methods for optimized scoring and analysis of recordings to reduce the amount of manual work and induced costs – again without compromising the diagnostic accuracy. Furthermore, work package 5 has an important role in the SleepRevolution research chain. Its results will be directly utilized in work package 6 from which the results are utilized in work package 7 and further in retrospective parts of the project.


University Of Eastern Finland

Reykjavik University

University Of Queensland

Nox Medical Ehf