• Citer

Development of a wearable technique for cardiovascular and sleep assessment: from space research to clinical applications

par Balali, Paniz
Président du jury Nonclercq, Antoine
Promoteur Debeir, Olivier
Co-Promoteur Van De Borne, Philippe
Publication Non publié, 2026-04-23

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

AbstractSleep apnea remains significantly underdiagnosed, despite its well-established impact on cardiovascular, metabolic, and cognitive health. Gold-standard diagnostics such as polysomnography are complex, costly, and often inaccessible in real-world settings, ranging from rural clinics to high-altitude expeditions and even space missions. Against this backdrop, this doctoral research investigated whether physiological signals captured by simple wearable sensors could serve as reliable indicators of sleep apnea, particularly in environments where conventional diagnostics are impractical.The work began with a seemingly simple question: whether heart rate variability (HRV), measured during wakefulness, predict sleep apnea severity, as demonstrated in a recent multicenter study involving over 1,000 patients. Our findings supported these results using our in-house patient dataset. However, during a one-year winter-over expedition in Antarctica, we found it challenging to attribute changes in wakefulness HRV exclusively to sleep disturbances, given the multifactorial influences on autonomic regulation in such extreme environments.Historically, wearable-based sleep apnea detection has often focused on single-lead electrocardiography (ECG) and HRV parameters, a trend fueled by extensive literature and benchmark datasets, such as those from PhysioNet challenges. Building on this insight, the focus of this work shifted to nighttime HRV, and classification models for apnea detection were developed using both open-access datasets and newly acquired recordings. Machine learning approaches, including support vector machines and random forests, were used to train models to distinguish apneic from non-apneic events based solely on HRV-derived autonomic patterns. Through these studies, both the potential and limitations of HRV-based diagnostics were revealed: while effectiveness was observed within the original datasets, generalizability was lacking when the models were applied to new recordings.To address this, the physiological scope was expanded by integrating additional wearable signals, such as seismocardiography (SCG) and ballistocardiography (BCG), which capture the mechanical activity of the heart and its interaction with respiration. Drawing from a comprehensive review of cardiorespiratory interactions, a multimodal sensing approach was pursued for the simultaneous acquisition of ECG, SCG, and impedance cardiography signals. This approach enabled the extraction of advanced features, such as cardiac time intervals (e.g., pre-ejection period, left ventricular ejection time), which offered deeper insights into autonomic and cardiac function than HRV alone. The utility of detecting these features was first tested on data collected on 17 cosmonauts before, during, and after a long-duration spaceflight onboard the International Space Station. Even though small changes were detected, they remained subclinical and dependent on the position (supine or sitting) used as a reference on Earth, overall demonstrating the efficacy of the countermeasure regime used in space.To further improve the performance of sleep apnea detection beyond ECG-based methods, the idea was to integrate parameters derived from SCG and ECG together. The groundwork for this combined approach has been initiated, and its full development remains a perspective for future work. Throughout this work, the themes of portability, generalizability, and physiological relevance remained central. Whether applied to high-altitude expeditions, long-duration space missions, or resource-limited clinics, the global goal was to transform wearable cardiopulmonary signals into actionable clinical insights, without the need for a hospital bed. In doing so, this thesis puts into perspective the reliance on HRV alone for assessing sleep quality and highlights the potential of emerging technologies, which can be integrated into wearable systems to provide complementary and clinically meaningful information for sleep monitoring.