The aim of this thesis was to study, develop and validate systems for telemedicine applications to monitor vital parameters. The focus has been on photoplethysmography (PPG) devices to obtain non-invasive, cuff-less, wireless, and repeated measurements of blood pressure (BP), in addition to heart rate (HR), oxygen saturation of blood (SpO2) and respiration rate (RR). It includes software design and implementation to estimate BP from the PPG signal, with the development of the processing algorithm, including filtering and noise elimination, and evaluating the critical issues on which it is necessary to intervene. The characteristic points of PPG signal have been identified and features have been extracted, including novel features extracted from the Maximal Overlap Discrete Wavelet Transform (MODWT) enhanced PPG signal. Then the most significant features have been identified by several selection algorithms. This has permitted to implement, train and compare the performance of several machine learning (ML) models with the aim of estimating systolic and diastolic BP using the processed PPG signal. Afterwards, several deep learning (DL) algorithms have been implemented using the whole PPG signal instead of the features extracted from it, and the impact of the loss function, model’s input and duration of the input has been investigated. Then a first prototype of a wearable embedded solution has been developed for a telemedicine application.
L’obiettivo di questa tesi è stato quello di studiare, sviluppare e validare sistemi per applicazioni di telemedicina per il monitoraggio dei parametri vitali. Il focus è stato sui dispositivi di fotopletismografia (PPG) per ottenere misure non invasive, senza bracciale, wireless, e ripetute della pressione sanguigna (BP), oltre alla frequenza cardiaca (HR), alla saturazione di ossigeno nel sangue (SpO2) e alla frequenza respiratoria (RR). La tesi include la progettazione e l’implementazione del software per stimare la BP dal segnale PPG, con lo sviluppo dell’algoritmo di elaborazione, compresi il filtraggio e l’eliminazione del rumore, e la valutazione delle criticità su cui è necessario intervenire. I punti caratteristici del segnale PPG sono stati identificati e sono state estratte le features, comprese le nuove features estratte dal segnale PPG migliorato a seguito dell’utilizzo della Maximal Overlap Discrete Wavelet Transform (MODWT). Successivamente le features più significative sono state identificate da diversi algoritmi di selezione. Ciò ha permesso di implementare, addestrare e confrontare le prestazioni di diversi modelli di machine learning (ML) con l’obiettivo di stimare la BP sistolica e diastolica utilizzando il segnale PPG elaborato. In seguito, sono stati implementati diversi algoritmi di deep learning (DL) utilizzando l’intero segnale PPG anziché le features da esso estratte, ed è stato analizzato l’impatto della funzione loss, dell’input del modello e della durata dell’input. Infine, è stato sviluppato un primo prototipo di soluzione embedded indossabile per un’applicazione di telemedicina.
Accurate blood pressure measurement from photoplethysmography signals using machine learning and deep learning techniques for innovative telemedicine / De Palma, Luisa. - ELETTRONICO. - (2024). [10.60576/poliba/iris/de-palma-luisa_phd2024]
Accurate blood pressure measurement from photoplethysmography signals using machine learning and deep learning techniques for innovative telemedicine
De Palma, Luisa
2024-01-01
Abstract
The aim of this thesis was to study, develop and validate systems for telemedicine applications to monitor vital parameters. The focus has been on photoplethysmography (PPG) devices to obtain non-invasive, cuff-less, wireless, and repeated measurements of blood pressure (BP), in addition to heart rate (HR), oxygen saturation of blood (SpO2) and respiration rate (RR). It includes software design and implementation to estimate BP from the PPG signal, with the development of the processing algorithm, including filtering and noise elimination, and evaluating the critical issues on which it is necessary to intervene. The characteristic points of PPG signal have been identified and features have been extracted, including novel features extracted from the Maximal Overlap Discrete Wavelet Transform (MODWT) enhanced PPG signal. Then the most significant features have been identified by several selection algorithms. This has permitted to implement, train and compare the performance of several machine learning (ML) models with the aim of estimating systolic and diastolic BP using the processed PPG signal. Afterwards, several deep learning (DL) algorithms have been implemented using the whole PPG signal instead of the features extracted from it, and the impact of the loss function, model’s input and duration of the input has been investigated. Then a first prototype of a wearable embedded solution has been developed for a telemedicine application.File | Dimensione | Formato | |
---|---|---|---|
37 ciclo-DE PALMA Luisa.pdf
accesso aperto
Descrizione: Tesi di dottorato
Tipologia:
Tesi di dottorato
Licenza:
Non specificato
Dimensione
7.89 MB
Formato
Adobe PDF
|
7.89 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.