Open Access

Ángel Serrano AlarcónORCiD

• The importance of quality sleep and optimal sleep duration for the maintenance of good health is well-established. However, it should be noted that this hypothesis is not invariably manifested by the occurrence of sleep disorders. Obstructive sleep apnea (OSA) is one of the most prevalent sleep disorders, with polysomnography (PSG) being the most effective procedure for identifying this disorder. Despite its accuracy, polysomnography is associated with numerous disadvantages, including protracted waiting periods for patients and considerable financial expense. Consequently, a number of alternatives have been developed to replace or complement polysomnography. Among these options, solutions incorporating artificial intelligence models have been identified. The purpose of this PhD thesis is to broaden the spectrum of analysis beyond the traditional boundaries by addressing the concept of the digital twin and its application in the field of sleep medicine. The concept of the digital twin has seen application across a variety of industrial domains. However, its implementation within the medical field, and more specifically within the domain of sleep medicine, has been less prevalent. In the course of this study, a comprehensive review of the methodologies and computational technologies that are essential for the implementation of a digital twin for the detection of sleep disorders, specifically the recognition of obstructive sleep apnea, has been conducted. The digital twin is comprised of two principal components: firstly, data of good quality, and secondly, artificial intelligence models that generate predictions from the data and facilitate data analysis. This study has therefore investigated the minimum set of signals required for the detection of obstructive sleep apnea. In addition, deep learning models have been developed for use with these signals, which have obtained 84.3% accuracy, 82.5% sensitivity, 86% specificity and an AUC (area under the curve) of 92.1%. The emphasis has also been placed on one of the main challenges associated with the use of artificial intelligence: so-called 'black boxes'. The utilisation of visualisation techniques served to enhance the explainability of the models. The present study has the potential to lay the groundwork for detecting sleep patterns and disorders more explicitly and accurately following the application and development of a digital twin. It is hypothesised that this development has the potential to facilitate a diagnosis that is more appropriate and less complex than the current polysomnography method.