Work package 3: Retrospective data mining
Lead: The University of Queensland
Retrospective data mining
The current severity estimation of obstructive sleep apnea is highly simplified mostly accounting only for the number of respiratory events during the night and symptoms of daytime sleepiness. The role of acute systematic effect, sleep disturbance, the risk of comorbidities, and their interlinkage to respiratory events remains overlooked. The state-of-the-art analysis methods alongside large international databases may be a solution to overcome these issues.
The goal of this work package is to develop methods that go through sleep recordings and score them automatically using machine learning methods. This means for example to identify the sleep stages throughout the night and finding events such as breathing difficulties and apnoeas. Currently, it takes a trained sleep technologist two hours to do this manually, but with automated methods, we hope to achieve this in a matter of seconds, and the sleep technologists will only have to give their feedback on the grey areas where the methods are not able to give a clear answer. To arrive at a more physiologically accurate way to assess the severity of obstructive sleep apnea, we aim to utilize the substantial database, a total of 30,000 sleep studies, already gathered by our strong ESADA network from all over Europe. We will apply modern artificial intelligence and signal processing techniques to create far more accurate and detailed diagnostic indexes, tailored to patient demographics and medical conditions.
This work package will quantify the respiratory events, the acute systemic effect, sleep disturbance, cardiovascular effects, and the interaction between the characteristics of these domains to better assess the severity of obstructive sleep apnea. This work package will provide algorithms based on sophisticated signal analysis methods and artificial intelligence for quantifying the disease severity and estimating the risk of developing various comorbidities. Moreover, we will provide methods for assessing the complex interactions between different physiological signals to reveal patterns linked to disease phenotypes.
Link to other work packages
This work package is strongly linked with work packages 4 and 5. These together will provide methods for optimized diagnosis and severity assessment of obstructive sleep apnea while reducing the amount of manual work and induced costs. Together, these work packages will also determine the most essential signals required for sleep studies to simplify the recording setup without compromising the diagnostic accuracy. Finally, this work package forms the basis for the novel technologies to be applied and validated in work packages 8-10.
University Of Queensland
University Of Eastern Finland
Unversity Of Oslo
European Sleep Research Society e.V.
Nox Medical Ehf
European Sleep Apnoea Database