Sleep is a fundamental process for human beings. Sleep quality and quantity are essential features: their lack may indicate sleep disorders. One of the most common sleep disorder is Obstructive Sleep Apnea Syndrome (OSAS), a condition characterized by breathing interruptions during sleep, due to partial or complete obstructions of the upper respiratory tract, that may cause a blood oxidative imbalance able to raise the risk of cardiovascular and cerebrovascular pathologies. OSAS is typically diagnosed with a laboratory polysomnography and commonly treated with Continuous Positive Airway Pressure (CPAP) technique. Sleep position may trigger apneic events, therefore, its detection and monitoring during CPAP treatment may be useful to correlate the occurrence of an apneic event with the position occupied at that moment. The aim of this thesis is to develop a low-power wearable device able to identify sleep position through an embedded classification model, exploiting Inertial Measurement Unit (IMU) signals. The device communicates data via Bluetooth Low Energy (BLE) and its functionalities are managed through a Real Time Operative System (RTOS). This classifier has been trained on 12 signals per position, to identify thresholds character izing each class, based on the trunk angles of rotation in the transversal (β) and sagittal (δ) planes, calculated exploiting accelerometer signals. The obtained classifier discriminates between up, supine, prone, lateral left and right po sitions, acting as a decision tree algorithm, comparing β and δ to the identified thresholds to retrieve the final class. It has been tested on 5 signals and reached 96% accuracy, perfectly comparable to the 93% achieved by a Linear Discriminant Analysis classifier found in literature. The device also underwent a successfully passed technical testing, in which the RTOS processes have been tested in their functionalities (both separately and integrated), and a power consumption optimization process.
Il sonno è un fondamentale processo per l’essere umano. Durata e qualità sono aspetti cruciali, con cui valutare la presenza di disturbi del sonno. Tra i più comuni vi è la Sindrome da Apnea Ostruttiva del Sonno (OSAS), un disturbo caratterizzato da pause respiratorie durante il sonno, provocate da ostruzioni parziali o complete delle vie aeree superiori, che possono portare a squilibri ossidativi del sangue ed aumentare il rischio di patologie cardiovascolari e cerebrovascolari. Solitamente viene diagnosticata con una polisonnografia e trattata con CPAP. La posizione nel sonno può influire sul verificarsi degli eventi apneici: identificarla e monitorarla durante il trattamento CPAP può essere utile per correlare l’insorgenza di un evento apneico con la posizione occupata in quel momento. Lo scopo di questa tesi è sviluppare un dispositivo indossabile a basso consumo, dotato di un algoritmo integrato per classificare la posizione del soggetto sfruttando i segnali dell’unità inerziale (IMU). Il dispositivo comunica i dati tramite Bluetooth Low Energy (BLE) e le sue funzionalità sono gestite da un sistema operativo real-time (RTOS). Il classificatore è stato addestrato con 12 segnali per posizione, per identificare le soglie caratterizzanti ciascuna classe usando gli angoli di rotazione nel piano trasversale (β) e sagittale (δ), ricavati dai segnali dell’accelerometro. Il classificatore ottenuto lavora come un algoritmo ad albero decisionale: identifica le posizioni sollevata, supina, prona, laterali sinistra e destra, confrontando β e δ con le soglie trovate. È stato testato su 5 segnali e ha raggiunto un’accuratezza del 96%, in linea col 93% ottenuto dal classificatore di analisi discriminante lineare trovato in letteratura. Il dispositivo ha anche superato con successo una serie di verifiche tecniche volte a testare il funzionamento dei processi RTOS (sia singolarmente che integrati), e un processo di ottimizzazione dei consumi.
Development of a low-power wearable device to identify sleep positions for sleep apnea applications
CHIESA, ROBERTA
2022/2023
Abstract
Sleep is a fundamental process for human beings. Sleep quality and quantity are essential features: their lack may indicate sleep disorders. One of the most common sleep disorder is Obstructive Sleep Apnea Syndrome (OSAS), a condition characterized by breathing interruptions during sleep, due to partial or complete obstructions of the upper respiratory tract, that may cause a blood oxidative imbalance able to raise the risk of cardiovascular and cerebrovascular pathologies. OSAS is typically diagnosed with a laboratory polysomnography and commonly treated with Continuous Positive Airway Pressure (CPAP) technique. Sleep position may trigger apneic events, therefore, its detection and monitoring during CPAP treatment may be useful to correlate the occurrence of an apneic event with the position occupied at that moment. The aim of this thesis is to develop a low-power wearable device able to identify sleep position through an embedded classification model, exploiting Inertial Measurement Unit (IMU) signals. The device communicates data via Bluetooth Low Energy (BLE) and its functionalities are managed through a Real Time Operative System (RTOS). This classifier has been trained on 12 signals per position, to identify thresholds character izing each class, based on the trunk angles of rotation in the transversal (β) and sagittal (δ) planes, calculated exploiting accelerometer signals. The obtained classifier discriminates between up, supine, prone, lateral left and right po sitions, acting as a decision tree algorithm, comparing β and δ to the identified thresholds to retrieve the final class. It has been tested on 5 signals and reached 96% accuracy, perfectly comparable to the 93% achieved by a Linear Discriminant Analysis classifier found in literature. The device also underwent a successfully passed technical testing, in which the RTOS processes have been tested in their functionalities (both separately and integrated), and a power consumption optimization process.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Chiesa_Executive_Summary.pdf
solo utenti autorizzati a partire dal 19/03/2027
Descrizione: Executive summary
Dimensione
1.39 MB
Formato
Adobe PDF
|
1.39 MB | Adobe PDF | Visualizza/Apri |
2024_04_Chiesa_Thesis.pdf
solo utenti autorizzati a partire dal 19/03/2027
Descrizione: Testo della Tesi
Dimensione
19.03 MB
Formato
Adobe PDF
|
19.03 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/219254