Nowadays the diagnosis of sleep-related breathing disorders, performed mainly through polysomnography, is complex and expensive in terms of equipment, due to the high variety of signals that are recorded during the whole night, and physicians who must continuously follow the correct advancement of the recording. These two aren’t the only problematics linked to the polysomnographic recording, the golden standard for the sleep analysis, as a matter of fact the analysis must be carried in specialized centres, the foreign environment can modify the quality of the sleep of the patient and in addition the visual inspection of all the recorded signals is done manually by physicians causing waste of time and resources. Other less expensive alternatives have been introduced, such as portable monitors that permit to bring the analysis directly at patient’s home, without the need of supervisors during all night but the results on their effective applicability are still controversial. In this context the principal aim of this work of thesis is collocated; the main idea is to develop a way to discriminate between healthy and unhealthy patients through the application of an apnea detection algorithm to the ECG derived respiratory signal (EDR signal). As a matter of fact, the ECG is a simple signal to be recorded, is possible to obtain it also from wearable devices and its cost is surely lower than a full polysomnographic recording. It’s also possible to exploit ECG recordings recorded for other purposes to obtain a pre-screening over some sleep disturbances such as apnea events. So, starting from the ECG signal recorded during the whole night, through Holter analysis, specific estimations of the respiratory signal are obtained, applying different methods already presented in different articles in literature, such as QRS area estimation, Q-R distance and R waves trends. All the methods citated have own characteristics and advantages often linked to the specific subject or to the conditions of acquisition of ECG signal, such as the posture; therefore, it’s complex to define a priori which is the better method for a specific subject or for a specific time interval of data. The important step forward has been made by the application of PCA algorithm to EDR signal obtainment, the main idea below this decision is the possibility to generalize all the EDR estimations in one. The dataset consists in Holter and polysomnographic signals of 13 patients recorded at Niguarda Ca’ Granda Hospital in Milan, all are obese patients with a proven sleep-related breathing disorders with different degrees of severity; even if the total number of patients present in the dataset is 13, only 7 patients have been considered, due to absence of Holter report for one patient and the really bad quality signals for the other 5 patients. Once the respiratory signals have been obtained, a statistical analysis has been performed through the evaluation of Pearson’s correlation coefficient with a t-test, between estimated respiratory signal and thoracic one obtained from the polysomnographic recording. Firstly, the test has been applied on the entire length of the signal including the zones in which the signal to nose ratio was very low, secondly only on the clearest segments of signal. To reach the objective of detection of apnea events a specific algorithm has been exploiting based on the evaluation of the breath amplitude compared to the reference breath amplitude obtained over 8 previous respiratory acts. The results are encouraging, the correlation coefficients calculated over segments of signals range between moderate and high values and is interesting to notice the relation between quality of the estimation through correlation coefficient and quality of apnea detection. In terms of apnea detection, all patients result unhealthy as expected, but the number of apnea detected compared to the Apnea Hypopnea Index (AHI) reported on the sleep report is strongly dependent on the method chosen; the best results are given by Q-R distance, R wave trend and 1st PCA estimation. This thesis is organized in five chapters, firstly will be analysed the context in which this work arises, macro & micro structure of sleep, the breathing-related sleep disorders, the tests available for the diagnose of these disorders and the need for a new approach; secondly will be presented the state of the art in terms of ECG derived respiratory signal estimations. The core of the thesis is presented in the last three chapters, in materials and methods the main applied methods are reported, starting from the analysis of the ECG signals, through the application of PCA algorithm, to the presentation of the apnea detection algorithm, in the chapter related to the results both sides of this work of thesis are analysed, the quality of the estimation of the ECG derived respiratory signal and the number of apnoeic events effectively detected exploiting the apnea detection algorithm, last but not least the last chapter reports possible future improvements of this work starting from improvement of the quality of the original ECG signals and further steps forward that can really led to the development of a system that could become the future of sleep apnea detection and an important instrument of screening in the optics to reduce costs and improve the performance.
Al giorno d’oggi la diagnosi dei disturbi del sonno legati alla respirazione è complessa e costosa sia in termini di strumentazione utilizzata, a causa della grande varietà di segnali registrati durante tutta la notte, sia in termini di infermieri specializzati e medici che devono continuamente monitorare l’avanzamento dell’esame diagnostico. Ma queste non sono le uniche problematiche legate alla polisonnografia, il golden standard nell’analisi dei disturbi del sonno, infatti l’analisi deve essere eseguita in centri specializzati e l’ambiente estraneo può modificare la qualità del sonno del paziente, inoltre l’ispezione visiva di tutti i segnali registrati viene eseguita manualmente da medici specializzati causando spreco di tempo e risorse. Alcune alternative meno costose sono state introdotte, come i monitor portatili che permettono di portare l’analisi direttamente a casa del paziente, senza la necessità di medici ed infermieri a supervisionare durante tutta la notte, ma i risultati sulla loro effettiva applicabilità sono tutt’ora controversi. In questo contesto si pone l’obiettivo principale di questa tesi, l’idea fondamentale è quella di sviluppare un modo per discriminare fra soggetti sani e malati attraverso l’applicazione di un algoritmo di detezione delle apnee notturne su segnale respiratorio derivato da segnale ECG. Infatti, l’ECG è un segnale semplice da registrare, che al giorno d’oggi può essere ottenuto anche attraverso strumenti indossabili con un costo sicuramente inferiore rispetto ad un’intera registrazione polisonnografica. Inoltre, è possibile sfruttare registrazioni di ECG fatte per altri scopi anche al fine di ottenere un pre-screening su alcuni disturbi del sonno quali le apnee notturne. Quindi, partendo dal segnale ECG registrato durante tutta la notte, attraverso l’analisi Holter, specifiche stime dell’attività respiratoria vengono ottenute attraverso l’applicazione di diversi metodi già presenti in letteratura, come la stima dell’area sottesa al complesso QRS, la distanza Q-R e l’ampiezza dei picchi R rispetto alla baseline. I metodi qui elencati hanno specifiche peculiarità e vantaggi spesso dipendenti dal singolo soggetto o dalle condizioni di acquisizione del segnale ECG, quali esempio la postura. Pertanto, diventa difficile definire a priori quale sia il metodo migliore per uno specifico paziente o per uno specifico intervallo temporale di dati. Il passo in avanti davvero importante che è stato fatto in questo lavoro è stata l’applicazione della Principal Component Analysis, PCA, per l’ottenimento della stima del segnale respiratorio, con l’intento di generalizzare tutte le diverse stime ottenute con i diversi metodi in una sola stima in grado di sintetizzare l’elemento comune delle singole. Il dataset è formato dai segnali polisonnografici e dall’ECG holter, registrati durante la stessa notte, presso l’Ospedale Niguarda Cà Granda di Milano, di 13 pazienti obesi che soffrono di comprovati disturbi del sonno legati alla respirazione; anche se il numero totale di pazienti è 13 per l’analisi sono stati considerati solo 7 pazienti, a causa della mancanza dell’Holter report per un paziente e la qualità pessima in termini di rapporto segnale rumore sull’ECG holter degli altri 5 soggetti. Una volta ottenute le stime del segnale respiratorio, viene eseguita un’analisi statistica attraverso la valutazione del coefficiente di correlazione di Pearson tra segnale respiratorio stimato e segnale respiratorio del torace, ottenuto dalla polisonnografia, con un t-test. Inizialmente l’analisi viene eseguita sull’intera lunghezza dei segnali incluse le zone in cui il rapporto segnale rumore è molto basso, per poi passare all’analisi nelle sole zone in cui il segnale risulta maggiormente pulito. Per raggiungere l’obiettivo di detezione delle apnee da segnale respiratorio stimato viene implementato uno specifico algoritmo basato sulla valutazione dell’ampiezza del segnale del respiro confrontata all’ampiezza del respiro di riferimento calcolata su 8 atti respiratori. I risultati sono incoraggianti, i coefficienti di correlazione calcolati sui segmenti più puliti di segnale variano tra valori moderati ed elevati di correlazione ed è interessante notare la relazione esistente tra qualità del segnale stimato e qualità nella detezione delle apnee. Per quanto riguarda la detezione delle apnee, tutti i pazienti risultano affetti da apnee come previsto, ma il numero di apnee effettivamente individuate comparato con l’Apnea Hypopnea Index, AHI, riportato sullo sleep report è fortemente dipendente dal metodo scelto per la stima del segnale del respiro, i risultati migliori sono dati dalla distanza Q-R, dall’andamento dei picchi R rispetto alla baseline e dalla prima componente ottenuta dalla PCA. Questa tesi è organizzata in 5 capitoli, innanzitutto verrà esaminato il contesto in cui questo lavoro si pone, macro e microstruttura del sonno, le principali patologie legate alla respirazione durante il sonno, gli strumenti per la diagnosi presenti e il perché della necessità di avere un nuovo approccio, dopodiché viene presentato lo stato dell’arte in termini di stime esistenti per il respiro a partire dall’ECG. Il punto focale della tesi è presentato negli ultimi tre capitoli, nei materiali e metodi sono riportati i principali metodi applicati, partendo dall’analisi del segnale ECG, passando per la stima dell’attività respiratoria mediante l’applicazione dell’algoritmo PCA, fino alla presentazione dell’algoritmo di detezione delle apnee; nel capitolo relativo ai risultati sono stati analizzati entrambi gli aspetti che caratterizzano questo lavoro, la qualità della stima del segnale respiratorio e il numero di eventi apnoici effettivamente individuati sfruttando l’algoritmo di detezione delle apnee, ultimo ma non per importanza, il capitolo delle conclusioni riporta futuri implementazioni, a partire dal miglioramento della qualità del segnale ECG originale e altri passi in avanti nella direzione di ottenere uno strumento che possa essere il futuro della detezione di apnee e un importante strumento di screening nell’ottica di ridurre costi e migliorare la performance.
ECG-derived respiratory signals for detection of sleep apnea events in overweight patients
De ROSE, IRENE
2017/2018
Abstract
Nowadays the diagnosis of sleep-related breathing disorders, performed mainly through polysomnography, is complex and expensive in terms of equipment, due to the high variety of signals that are recorded during the whole night, and physicians who must continuously follow the correct advancement of the recording. These two aren’t the only problematics linked to the polysomnographic recording, the golden standard for the sleep analysis, as a matter of fact the analysis must be carried in specialized centres, the foreign environment can modify the quality of the sleep of the patient and in addition the visual inspection of all the recorded signals is done manually by physicians causing waste of time and resources. Other less expensive alternatives have been introduced, such as portable monitors that permit to bring the analysis directly at patient’s home, without the need of supervisors during all night but the results on their effective applicability are still controversial. In this context the principal aim of this work of thesis is collocated; the main idea is to develop a way to discriminate between healthy and unhealthy patients through the application of an apnea detection algorithm to the ECG derived respiratory signal (EDR signal). As a matter of fact, the ECG is a simple signal to be recorded, is possible to obtain it also from wearable devices and its cost is surely lower than a full polysomnographic recording. It’s also possible to exploit ECG recordings recorded for other purposes to obtain a pre-screening over some sleep disturbances such as apnea events. So, starting from the ECG signal recorded during the whole night, through Holter analysis, specific estimations of the respiratory signal are obtained, applying different methods already presented in different articles in literature, such as QRS area estimation, Q-R distance and R waves trends. All the methods citated have own characteristics and advantages often linked to the specific subject or to the conditions of acquisition of ECG signal, such as the posture; therefore, it’s complex to define a priori which is the better method for a specific subject or for a specific time interval of data. The important step forward has been made by the application of PCA algorithm to EDR signal obtainment, the main idea below this decision is the possibility to generalize all the EDR estimations in one. The dataset consists in Holter and polysomnographic signals of 13 patients recorded at Niguarda Ca’ Granda Hospital in Milan, all are obese patients with a proven sleep-related breathing disorders with different degrees of severity; even if the total number of patients present in the dataset is 13, only 7 patients have been considered, due to absence of Holter report for one patient and the really bad quality signals for the other 5 patients. Once the respiratory signals have been obtained, a statistical analysis has been performed through the evaluation of Pearson’s correlation coefficient with a t-test, between estimated respiratory signal and thoracic one obtained from the polysomnographic recording. Firstly, the test has been applied on the entire length of the signal including the zones in which the signal to nose ratio was very low, secondly only on the clearest segments of signal. To reach the objective of detection of apnea events a specific algorithm has been exploiting based on the evaluation of the breath amplitude compared to the reference breath amplitude obtained over 8 previous respiratory acts. The results are encouraging, the correlation coefficients calculated over segments of signals range between moderate and high values and is interesting to notice the relation between quality of the estimation through correlation coefficient and quality of apnea detection. In terms of apnea detection, all patients result unhealthy as expected, but the number of apnea detected compared to the Apnea Hypopnea Index (AHI) reported on the sleep report is strongly dependent on the method chosen; the best results are given by Q-R distance, R wave trend and 1st PCA estimation. This thesis is organized in five chapters, firstly will be analysed the context in which this work arises, macro & micro structure of sleep, the breathing-related sleep disorders, the tests available for the diagnose of these disorders and the need for a new approach; secondly will be presented the state of the art in terms of ECG derived respiratory signal estimations. The core of the thesis is presented in the last three chapters, in materials and methods the main applied methods are reported, starting from the analysis of the ECG signals, through the application of PCA algorithm, to the presentation of the apnea detection algorithm, in the chapter related to the results both sides of this work of thesis are analysed, the quality of the estimation of the ECG derived respiratory signal and the number of apnoeic events effectively detected exploiting the apnea detection algorithm, last but not least the last chapter reports possible future improvements of this work starting from improvement of the quality of the original ECG signals and further steps forward that can really led to the development of a system that could become the future of sleep apnea detection and an important instrument of screening in the optics to reduce costs and improve the performance.File | Dimensione | Formato | |
---|---|---|---|
2019_04_De Rose.pdf
solo utenti autorizzati dal 02/04/2020
Descrizione: Testo della tesi
Dimensione
4.2 MB
Formato
Adobe PDF
|
4.2 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/10589/146160