Respiratory rate (RR) is a vital physiological parameter, sensitive to both pathological conditions and physiological stressors. Despite its clinical importance, RR is still rarely monitored by wearable devices. The COVID-19 pandemic has strengthened the interest in remote monitoring solutions, highlighting the gap between research and commercial devices. Conventional methods—such as direct observation, spirometry, or plethysmography—are not well suited for continuous, real-world monitoring. This thesis presents the design and testing of a wearable, non-invasive device for continuous monitoring of RR and other physiological parameters. The system is based on multi-wavelength photoplethysmography (PPG) ranging from visible to near-infrared (NIR), with a configurable LED setup. Data acquisition was structured into two protocols: one for evaluating the potential of multi-wavelength PPG in estimating RR under static conditions, and the other for testing the device's ability to measure heart rate and activity parameters in both static and dynamic conditions. Data was collected from nine participants per each protocol, using gold-standard reference devices for comparison. A range of previously-published PPG-based RR estimation algorithms was evaluated on the collected data using the RRest Matlab toolbox developed by Peter Charlton et al. In addition, the effectiveness of multi-wavelength ensembling for RR estimation was investigated. Four RR ensembling strategies were tested and compared. The best-performing approach, based on multi-wavelength PPG, achieved a median absolute error of 2.17 breaths per minute and an interquartile range (IQR) of 3.48 breaths per minute on aggregated data. Furthermore, the optimal single wavelengths for RR estimation at the finger and wrist level were identified. These results demonstrate that combining multiple PPG wavelengths could improve RR estimation accuracy, confirming the potential of this approach for more robust, continuous respiratory monitoring.
La frequenza respiratoria (RR) è un parametro vitale essenziale, sensibile a molteplici condizioni patologiche e stressori fisiologici. Nonostante la sua rilevanza clinica, è ancora poco monitorata in ambito domiciliare. La pandemia di COVID-19 ha rafforzato l’interesse verso soluzioni di monitoraggio remoto, evidenziando il divario tra la ricerca e l’offerta commerciale. La misurazione convenzionale avviene tramite osservazione diretta o tecniche come la spirometria o la pletismografia, poco adatte all’uso quotidiano. Questa tesi propone un dispositivo indossabile e non invasivo per il monitoraggio continuo della frequenza respiratoria e di altri parametri fisiologici. Il sistema si basa su una fotopletismografia a più lunghezze d'onda, dal visibile al vicino infrarosso (NIR), con configurazione modulabile dei LED. L’acquisizione dati è articolata in due protocolli: uno per frequenza cardiaca, SpO\textsubscript{2} e parametri di attività (in statica e dinamica), l’altro per la stima della RR in condizioni statiche. I test, condotti su gruppi di nove soggetti, hanno previsto l’uso simultaneo di dispositivi standard come riferimento. L’elaborazione tramite il toolbox RRest (MATLAB), sviluppato da Peter Charlton, ha permesso il confronto di diversi algoritmi di stima della RR, valutando anche l’efficacia della fusione di lunghezze d’onda. L’algoritmo più performante è quello ricavato dalla fusione finale delle frequenze respiratorie, ottenute da tre varianti, che condividono la stessa tecnica di stima (Count-oring) ma differiscono per la modalità di estrazione delle feature dal segnale PPG. Esso ha presentato un errore assoluto mediano di 2.17 respiri/minuto e un intervallo interquartile (IQR) di 3.48, considerando tutti i soggetti. Quattro metodi di fusione sono stati testati e confrontati, tramite questo algoritmo, con le migliori singole lunghezze d’onda per dito e polso. I risultati mostrano un miglioramento dell’accuratezza nella stima della RR per tutti e quattro i metodi, confermando il potenziale della fotopletismografia multi-lunghezza d’onda per il monitoraggio respiratorio continuo.
Proof of concept design and development of a multi-wavelength photoplethysmographic system for physiological monitoring and respiratory rate estimation
BODEI, NICOLA
2024/2025
Abstract
Respiratory rate (RR) is a vital physiological parameter, sensitive to both pathological conditions and physiological stressors. Despite its clinical importance, RR is still rarely monitored by wearable devices. The COVID-19 pandemic has strengthened the interest in remote monitoring solutions, highlighting the gap between research and commercial devices. Conventional methods—such as direct observation, spirometry, or plethysmography—are not well suited for continuous, real-world monitoring. This thesis presents the design and testing of a wearable, non-invasive device for continuous monitoring of RR and other physiological parameters. The system is based on multi-wavelength photoplethysmography (PPG) ranging from visible to near-infrared (NIR), with a configurable LED setup. Data acquisition was structured into two protocols: one for evaluating the potential of multi-wavelength PPG in estimating RR under static conditions, and the other for testing the device's ability to measure heart rate and activity parameters in both static and dynamic conditions. Data was collected from nine participants per each protocol, using gold-standard reference devices for comparison. A range of previously-published PPG-based RR estimation algorithms was evaluated on the collected data using the RRest Matlab toolbox developed by Peter Charlton et al. In addition, the effectiveness of multi-wavelength ensembling for RR estimation was investigated. Four RR ensembling strategies were tested and compared. The best-performing approach, based on multi-wavelength PPG, achieved a median absolute error of 2.17 breaths per minute and an interquartile range (IQR) of 3.48 breaths per minute on aggregated data. Furthermore, the optimal single wavelengths for RR estimation at the finger and wrist level were identified. These results demonstrate that combining multiple PPG wavelengths could improve RR estimation accuracy, confirming the potential of this approach for more robust, continuous respiratory monitoring.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_07_Bodei_Tesi_01.pdf
accessibile in internet per tutti
Descrizione: testo tesi
Dimensione
11.61 MB
Formato
Adobe PDF
|
11.61 MB | Adobe PDF | Visualizza/Apri |
|
2025_07_Bodei_Executive_Summary_02.pdf
accessibile in internet per tutti
Descrizione: testo executive summary
Dimensione
42.42 MB
Formato
Adobe PDF
|
42.42 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/10589/240608