Recent progress in wearable technologies and computational modeling has created new pathways for deriving meaningful physiological insights from sleep. This research follows two interconnected avenues: developing deep learning methods for analyzing multimodal sleep signals, and examining cardiovascular dynamics during NREM sleep influenced by auditory stimulation. The wearable-focused investigations aimed to extract sleep-related information using a chest-mounted device in real-world, home settings. Custom deep learning models were crafted for multitask classification, addressing signal quality evaluation, detection of respiratory and sleep-associated events, and early-stage, cuffless blood pressure estimation. These solutions showcased promising capabilities for delivering precise, explainable, and scalable sleep monitoring beyond traditional clinical environments. The second part of the work established an analytical pipeline to quantify how discrete slow wave events interact with cardiovascular dynamics during sleep. Using high-resolution EEG, ECG, and blood pressure signals collected under closed-loop auditory stimulation, slow waves were classified by synchronization type and analyzed in relation to wave-locked autonomic responses. This approach revealed that highly synchronized slow waves, when enhanced through auditory stimulation, elicited stronger heart rate and blood pressure modulations — physiological signatures that correlated with the next-morning cardiac performance. These results highlight how event-based analysis of multimodal sleep signals can disentangle the dynamics of neural-cardiovascular coupling, offering a quantitative framework to study how specific sleep patterns influence cardiovascular physiology. Taken together, these contributions advance the methodological and conceptual tools needed to investigate sleep as a dynamic interface between neural and cardiovascular systems. By bridging algorithmic development with physiological insight, this work contributes to a growing effort to characterize and contextualize sleep as a window into cardiovascular function — both through scalable monitoring tools and deeper understanding of how auditory stimulation and slow wave dynamics shape autonomic regulation.
Negli ultimi anni, il rapido progresso dei dispositivi indossabili e della tecnologia computazionale ha aperto nuove prospettive per la comprensione dei processi fisiologici durante il sonno. Questo lavoro si muove lungo due direzioni parallele e complementari: da un lato la progettazione di strumenti basati su deep learning per l’elaborazione multimodale dei segnali del sonno; dall’altro l’indagine delle dinamiche cardiovascolari nel sonno NREM, con particolare attenzione al ruolo della stimolazione acustica nella modulazione delle slow waves. La prima linea di ricerca ha portato allo sviluppo di modelli capaci di analizzare in modo automatico i dati raccolti, in ambiente domestico, da un dispositivo toracico indossabile. Le architetture di deep learning proposte assolvono simultaneamente più compiti: valutazione della qualità del segnale, rilevazione di eventi respiratori e correlati al sonno, oltre alla stima preliminare della pressione arteriosa. I risultati dimostrano la fattibilità di un monitoraggio del sonno accurato, interpretabile e scalabile anche al di fuori del contesto clinico. Il secondo filone di ricerca ha portato alla definizione di un modello analitico per quantificare il legame tra slow waves e risposte cardiovascolari durante il sonno. Grazie a registrazioni ad alta risoluzione di EEG, ECG e pressione arteriosa, ottenute in condizioni controllate con stimolazione acustica a ciclo chiuso, le slow waves sono state classificate in base al grado di sincronizzazione e analizzate rispetto alle risposte autonomiche evento-correlate. È emerso che le onde più sincronizzate, soprattutto se potenziate acusticamente, producono variazioni più marcate della frequenza cardiaca e della pressione sanguigna — segnali associati a un miglioramento della funzione cardiaca osservato al risveglio. L’approccio evento-centrico adottato ha permesso di scomporre le dinamiche di interazione cervello-cuore durante il sonno, offrendo un quadro quantitativo dell’impatto di specifici pattern elettroencefalografici sulla regolazione cardiovascular. Nel loro insieme, questi contributi delineano un percorso che va dall’innovazione tecnologica alla scoperta fisiologica, rafforzando gli strumenti per studiare il sonno come interfaccia dinamica tra sistema nervoso e apparato cardiovascolare. Integrando sviluppo algoritmico e comprensione biologica, la tesi si colloca nel più ampio contesto della ricerca traslazionale, contribuendo a chiarire come la stimolazione acustica e i diversi fenotipi delle slow waves influenzino l’equilibrio autonomico e la funzione cardiaca.
Multimodal sleep analysis: deep learning for wearables and cardiovascular dynamics of auditory-evoked NREM patterns
Alessandrelli, Giulia
2024/2025
Abstract
Recent progress in wearable technologies and computational modeling has created new pathways for deriving meaningful physiological insights from sleep. This research follows two interconnected avenues: developing deep learning methods for analyzing multimodal sleep signals, and examining cardiovascular dynamics during NREM sleep influenced by auditory stimulation. The wearable-focused investigations aimed to extract sleep-related information using a chest-mounted device in real-world, home settings. Custom deep learning models were crafted for multitask classification, addressing signal quality evaluation, detection of respiratory and sleep-associated events, and early-stage, cuffless blood pressure estimation. These solutions showcased promising capabilities for delivering precise, explainable, and scalable sleep monitoring beyond traditional clinical environments. The second part of the work established an analytical pipeline to quantify how discrete slow wave events interact with cardiovascular dynamics during sleep. Using high-resolution EEG, ECG, and blood pressure signals collected under closed-loop auditory stimulation, slow waves were classified by synchronization type and analyzed in relation to wave-locked autonomic responses. This approach revealed that highly synchronized slow waves, when enhanced through auditory stimulation, elicited stronger heart rate and blood pressure modulations — physiological signatures that correlated with the next-morning cardiac performance. These results highlight how event-based analysis of multimodal sleep signals can disentangle the dynamics of neural-cardiovascular coupling, offering a quantitative framework to study how specific sleep patterns influence cardiovascular physiology. Taken together, these contributions advance the methodological and conceptual tools needed to investigate sleep as a dynamic interface between neural and cardiovascular systems. By bridging algorithmic development with physiological insight, this work contributes to a growing effort to characterize and contextualize sleep as a window into cardiovascular function — both through scalable monitoring tools and deeper understanding of how auditory stimulation and slow wave dynamics shape autonomic regulation.File | Dimensione | Formato | |
---|---|---|---|
2025_06_Giulia_Alessandrelli_PhD.pdf
accessibile in internet per tutti a partire dal 23/06/2028
Descrizione: Giulia Alessandrelli PhD Thesis - Multimodal Sleep Analysis: Deep Learning for Wearables and Cardiovascular Dynamics of Auditory-Evoked NREM Patterns
Dimensione
10.46 MB
Formato
Adobe PDF
|
10.46 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/10589/239774