Sleep disorders are an important and growing public health problem affecting a considerable proportion of the world's population.The study of such disorders is made possible by polysomnography (PSG): an ad hoc examination that allows a series of physiological parameters to be recorded during the night, such as brain activity, eye movements, muscle movements, oxygen levels, heartbeat and respiration.
For decades, numerous AI algorithms have been developed to automate the scoring of PSG, i.e. the extraction of clinically relevant information from physiological signals according to official standards.
Recently, thanks to increased computing power, deep learning has also been employed with promising results, highlighting the high capacity to learn from a highly heterogeneous data set.
For several years now, there has been a fruitful collaboration between the Biomedical Signal Processing (BPS) research group led by Francesca Faraci, of the Institute of Digital Technologies for Personalised Healthcare (MeDiTech), the , the and the ). This resulted in several publications and the European project SPAS, Sleep Physician Assistant System, for the creation of a platform to support and optimise the work of healthcare professionals active in the analysis of sleep disorders.
As part of this research, Luigi Fiorillo and Giuliana Monachino, Researcher and PhD student at MeDiTech, carried out numerous tests using access to InselSpitalBern's large database and the clinical support of the renowned sleep centre, achieving excellent performance.
The study, co-funded by the and the European Union, demonstrates the resilience of a deep learning algorithm in identifying different sleep phases. Specifically, it shows how a deep learning algorithm succeeds in solving a clinical procedure using information (derivations of physiological signals) that differs from that normally used by a sleep physician. This study is the result of analyses performed on the highest number of PSGs ever used to date in the literature.
More information:
Important findings have been published in the NPJ scientific journal Nature: .