« Conception and development of a multimodal and innovative medical device
for the screening of Sleep Apnea Syndrome (SAS) »
Place : (confidential defense)
- Pierre-Yves Gumery, Professeur des Universités, Université Grenoble Alpes, laboratoire TIMC, Director
- Julie Fontecave-Jallon, Maîtresse de conférences, Université Grenoble Alpes, laboratoire TIMC, Co-director
- Régine Le Bouquin Jeannes, Professeur des Universités, Université Rennes 1, Reporter
- Mounir Chennaou, Ingénieur de Recherche Hdr, Irba – Bretigny-Sur-Orge, Reporter
- Patrick Mallea, Docteur-Ingénieur, Société Nehs Digital – Malakof, Examiner
- Frédéric Gagnadoux, Professeur des Universités - Praticien Hospistralier, Université d’Angers, Examiner
- Jean-Louis Pepin, Professeur des Universités - Praticien Hospistralier, Université Grenoble Alpes, Examiner
Sleep Apnea, Accelerometry, Multimodality, Machine Learning
Sleep Apnea Syndrome (SAS) is a sleep-related condition characterized by the occurrence of frequent and abnormally distributed episodes of cessation (apneas) or decrease (hypopneas) of ventilation for a duration of at least 10 seconds. The prevalence of SAS is estimated around 10% of the worldwide population. However, this pathology remains largely under-diagnosed, in particular because of the cumbersomeness and complexity of the reference diagnostic method, the polysomnography (PSG). The congestion of sleep centers and the still manual expertise required to analyze recordings penalizes patient management. In this context, one of the challenges of industrial research is to lighten investigation methods. In the spirit of the international classification of the American Academy of Sleep Medicine (AASM), which prioritizes the different diagnostic approaches from polysomnography in a monitored condition to ventilatory polygraphy with one or two sensors, we are seeking to minimize the number of sensors. Performance evaluation is to be assessed with respect to uses that can range from hospital diagnosis to home screening.
This thesis proposes a double objective. The first is to design an innovative lightweight device based on the coupling of an accelerometric system and an electrocardiography (ECG) sensor. The accelerometric system using two levels of measurement (thoracic and abdominal) aims at an indirect estimation of airflow for the detection of respiratory events. The evaluation of this estimation is conducted under physiological and pathophysiological conditions. In this second case, the other objective of the thesis is to develop an automated performance evaluation environment based on a Machine Learning solution. This solution is also based on the implementation of a specific database with, to date, 28 patients. The decision-making strategy is based on a model that detects abnormal segments and estimates the Apnea-Hypopnea Index (AHI).
In terms of results, an analysis of the performance of the decision-making tool in polysomnographic conditions showed that the classification of apneic subjects according to the AHI reached an accuracy of 100%. The analysis of the performance of the innovative device using the validated decision tool led to a precision of 79%, a sensitivity of 80% and a specificity of 77% based on available data. These results provide a proof of concept for the use of the device in SAS screening. In order to go beyond the proof of concept, an increase in the number of patients and the statistical dimension of the results remain necessary.