| TAZ-TFM-2022-1290 |
Clasificación automatizada de las fases del sueño mediante el uso de bioseñales.
Sierra Torralba, María
Montesano del Campo, Luis (dir.)
Universidad de Zaragoza,
EINA,
2022
Departamento de Informática e Ingeniería de Sistemas, Área de Lenguajes y Sistemas Informáticos
Máster Universitario en Robótica, Gráficos y Visión por Computador
Resumen: Good quality sleep is vital for good health. It supports different physiological body functions, including immune, metabolic, and cardiovascular systems. Furthermore, adequate sleep facilitates optimal learning, memory, attention, mood, and decision-making processes. Nevertheless, sleep disorders are prevalent worldwide.
Sleep monitoring and scoring is crucial in the study and diagnosis of these diseases. Today,the only widely accepted method in clinical practice is the polysomnography (PSG), which is both intrusive for patients and expensive to perform for health systems. Accurate monitoring requires at least one night in a sleep laboratory and a time-consuming setup by technicians. The classification of sleep stages across the night provides information on the overall architecture of sleep, as well as the duration and proportion of the sleep stages, all of which inform the diagnosis of sleep disorders. Currently, this task is performed visually by human experts, requiring each 30-second epoch of a full night recording to be assigned a sleep stage. As a result, wating times for diagnostics are often larger than six months, depriving many patients of effective treatment and, thus, representing a pragmatic bottleneck. The main goals of this thesis are to automate such labor-intensive and routine process, leading to a great reduction in workload for clinicians, as well as addressing the increasing need for longitudinal monitoring in home environments. In order to accomplish them, an AI-powered technique is developed. This will constitute the main part of a wearable EEG monitoring device based on a new textile sensor technology, which can comfortably assess everyone’s sleep at home. For that purpose, the existing literature is explored. Machine learning algorithms and, in particular, emerging deep learning approaches have shown to be outstanding approaches. Accordingly, two different deep neural networks are proposed. After that, they are implemented and applied to sleep scoring. Their goodness is evaluated considering a wide range of datasets with very different characteristics, as well as applying diverse validation and testing methods. The results presented in this project demonstrate the validness of the models to perform real-time sleep staging with a limited number of channels in realistic settings. Moreover, one of the designed approaches in particular leads to a performance very similar to human sleep experts. Consequently, this work serves as a proof-of-concept for future sleep technology, and lays the foundation for a diverse scope of brain-computer interfaces for real-world applications.
Sleep monitoring and scoring is crucial in the study and diagnosis of these diseases. Today,the only widely accepted method in clinical practice is the polysomnography (PSG), which is both intrusive for patients and expensive to perform for health systems. Accurate monitoring requires at least one night in a sleep laboratory and a time-consuming setup by technicians. The classification of sleep stages across the night provides information on the overall architecture of sleep, as well as the duration and proportion of the sleep stages, all of which inform the diagnosis of sleep disorders. Currently, this task is performed visually by human experts, requiring each 30-second epoch of a full night recording to be assigned a sleep stage. As a result, wating times for diagnostics are often larger than six months, depriving many patients of effective treatment and, thus, representing a pragmatic bottleneck. The main goals of this thesis are to automate such labor-intensive and routine process, leading to a great reduction in workload for clinicians, as well as addressing the increasing need for longitudinal monitoring in home environments. In order to accomplish them, an AI-powered technique is developed. This will constitute the main part of a wearable EEG monitoring device based on a new textile sensor technology, which can comfortably assess everyone’s sleep at home. For that purpose, the existing literature is explored. Machine learning algorithms and, in particular, emerging deep learning approaches have shown to be outstanding approaches. Accordingly, two different deep neural networks are proposed. After that, they are implemented and applied to sleep scoring. Their goodness is evaluated considering a wide range of datasets with very different characteristics, as well as applying diverse validation and testing methods. The results presented in this project demonstrate the validness of the models to perform real-time sleep staging with a limited number of channels in realistic settings. Moreover, one of the designed approaches in particular leads to a performance very similar to human sleep experts. Consequently, this work serves as a proof-of-concept for future sleep technology, and lays the foundation for a diverse scope of brain-computer interfaces for real-world applications.
Tipo de Trabajo Académico: Trabajo Fin de Master
Enlace permanente:
El registro pertenece a las siguientes colecciones:
Trabajos académicos > Trabajos Académicos por Centro > Escuela de Ingeniería y Arquitectura
Trabajos académicos > Trabajos fin de máster