In the 21st century, along with the aging of the population due to the improved life expectancy, there has been an increase in the chronic conditions that is not expected to decline and require ongoing medical care with several implications such as health costs and worsening of quality of life. Chronic diseases, also known as Noncommunicable diseases (NCDs), tend to be of long duration and are the result of a combination of genetic, physiological, environmental and behavioral factors, mostly life-style related. For this reason, more attention should be paid on health surveillance rather than afterwards treatment: due to their intrinsic long-term feature, NCDs are not acute enough to be treated in hospital and sustaining health monitoring is critical to cope with this challenging issue. In this perspective, medical wearable devices could represent a useful instrument for ambulatory care monitoring and preventive medicine, providing an improvement for pathologies diagnosing, early detection of health’s conditions worsening and for administering the best treatment to improve life quality. The possibility to monitor physiological and biochemical parameters continuously, under natural physiological conditions and in any environment is recognized as one of the most promising platforms for minimally obtrusive and individualized health services. Focusing on chronic respiratory diseases, a wide category of subjects interested in a continuous ventilatory parameters monitoring are subjects affected by chronic obstructive pulmonary disease (COPD). Another group of pathological subjects that can benefit from a sensorized garment are those experiencing obstructive sleep apnea (OSAS). Untreated OSAS could lead to several negative effects, from decrease in productivity to cardiovascular diseases, rising the risk of hypertension, cardiac failure or coronary artery disease. In cases such as these, a reliable wearable device would provide a new way to diagnose specific pathologies, even at early stages, and monitor the patient during daily living. The availability of different technologies and devices has fostered a great interest in continuous and long-term monitoring of breathing parameters to assess ventilatory function in patients and/or healthy subjects. Respiratory rate assessment, for example, is essential for detecting acute changes and for monitoring the progression of illness. Although in the past it has been defined as the “neglected vital sign”, abnormal respiratory rate has been shown to be an important predictor of serious clinical events such as cardiac arrest and to be better in discriminating between stable patients and patients at risk. Furthermore, monitoring ventilation during sleep helps to detect apneas and hypopneas, and allows to quantify their occurrences in the screening for sleep disorders. To this extent, wearable devices for monitoring babies’ breath have great importance for preventing sudden infant death syndrome (SIDS) that can cause suffocation. Besides pointing out the importance of pulmonary ventilation assessment through breathing parameters, the possibility to monitor continuously and ubiquitously their values allows to investigate long term variability and correlation properties. Respiration in awake, healthy adult humans is characterized by high variability in the frequency, duration and amplitude of breaths. Whereas breathing is generally treated as a rhythmic process, the variability in cycle-by-cycle measurements of respiratory period and breath amplitude suggests a fractal nature revealed by long-range correlations among the fluctuations in the number of breaths, breath amplitudes and peak-to-peak breath interval. Such long-range correlations have been investigated with different fractal analyses, involving either detrended fluctuation analysis or other methods. Several researches have demonstrated that the complexity analysis of breathing dynamics can be used as a method to understand nonlinear behavior, to provide a physiological insight to the respiratory system and to become a tool for clinical assessment of respiratory disorders. Furthermore, analyzing respiratory dynamics with these approaches may provide information about response to treatment, and may be used in clinical practice as well as for home-based monitoring of disease progression. The objective of this thesis is to develop algorithms to perform signal processing and extract breath-to-breath breathing parameters from the sum of the five respiratory signals. In order to characterize and investigate variability and correlation properties of spontaneous breathing both descriptive statistics and detrended fluctuation analysis, used to compute the scaling exponent , are performed for each breathing parameter. Thanks to the availability of a large amount of data for each subject, it is possible to investigate different conditions of level of activity and perform a preliminary segmentation of wakefulness and sleep. Within a clinical trial performed at the Centro Cardiologico Monzino (Milan), healthy subjects and patients with cardiorespiratory pathologies (coronary artery disease, congestive heart failure and chronic obstructive pulmonary disease) are considered for our analyses in order to investigate possible differences in the fractal dynamics and variability in control and disease states. During the first phase of the study, a simple classifier for human activity recognition is implemented to distinguish generic postures, generic activity, postural transitions and walking. Taking into account several issues that are still ongoing, mostly related to the unknown and random orientation of the axes of the IMU, the performance of the classifier allows to obtain a training and testing accuracy above 90%. Thanks to its simplicity it could be easily implementable directly on the device and the sliding window of 3s with 50% overlapping allows to obtain a real-time classification. Regarding the respiratory signal, it is obtained as the sum of the five respiratory recordings acquired through elastic strain sensors placed on the thorax, the abdomen and at the xiphoid process. A proper filtering stage is implemented to remove noise and an algorithm to detect peaks and troughs is designed in a self-written MATLAB software. In order to investigate breath-to-breath fluctuations of breathing parameters time series of tidal volume, inspiratory time, respiration period, respiratory rate, duty cycle, minute ventilation and mean inspiratory flow are extracted. Moreover, these time series (sample at each breath) are obtained both for the entire respiratory signal and separated in diurnal and nocturnal tracts identified with time segmentation on an hourly basis. The descriptive statistics allows to better investigate the distribution of the extracted values for each parameter and reveals, as expected, higher variability during day and increased values of tidal volume, minute ventilation, mean inspiratory flow, duty cycle and respiratory rate. Moreover, significant differences are highlighted when comparing control and disease states, characterizing the outcome distributions up to the fourth order statistics (i.e. mean, spread of range of values, skewness and kurtosis). At the same time, detrended fluctuation analysis allows to compute the scaling exponent for each parameter. A qualitative analysis of the obtained log-log plot established the presence of a cross-over point around 100 breaths, suggesting a change in the fractal behavior for short and long ranges, with an increased scaling exponent approaching or exceeding α = 1 for a number of breaths higher than a hundred. In addition, day and night segmentation reveals that during night, when sleep is most likely to occur, the scaling exponents exhibit higher values compared to those obtained for diurnal tracts. Even in terms of fractal fluctuations some significant differences are brought out, revealing that most of them could be found between healthy subjects and patients with congestive heart failure. The heterogeneous population enrolled, with healthy subjects younger and with lower BMI than cardiorespiratory patients calls for correlation analysis with age and BMI effects. As a result, tidal volume, minute ventilation and mean inspiratory flow, which are obtained from the formed, show associations with age and body mass index: the increase of those characteristics seems to decrease the value of the parameters. An important result consists in the absence of correlations between scaling exponents and age and BMI effects, suggesting that significant differences obtained are attributable only to presence/absence of cardiorespiratory disease. In conclusion, both descriptive statistics and detrended fluctuation analysis have been demonstrated as useful tools to assess variability and long-term correlations in breathing pattern parameters. Decreased values of scaling exponents in elder cardiorespiratory patients might suggest a degradation of correlations that, though not fully understood, poses new basis to further investigation. The thesis is divided into four main chapters: Chapter 1: Introduction. Within this chapter the theoretical basis relevant to understand the objectives of this thesis is recalled. The potential role of wearable devices in continuous and ubiquitous monitoring is investigated with particular attention to respiratory diseases. L.I.F.E. Corporation and its medical garment are briefly described. Main respiratory pathologies and ventilation physiology parameters are exposed, describing also traditional and common non-invasive measurement techniques. Variability and fractal characterization of physiological signals are introduced to describe the state of art in the study of fractal analysis of human respiration and correlated variability in breathing patterns. The paragraph about the aims of this thesis concludes the first chapter. Chapter 2: Materials and methods. The second chapter describes the steps developed for the analysis of the time series extracted for each breathing parameter. First, a classifier for human activity recognition is described. The outcome of the classification is then used to quantify the level of activity of each subject. To extract breathing parameters from respiratory signal, an algorithm for peaks and troughs detection is introduced. A brief panoramic of descriptive statistics involved in the analysis of time series is presented. In conclusion of Chapter 2, detrended fluctuation analysis is briefly explained. Chapter 3: Results. The third chapter presents the statistical analysis of the results obtained by performing descriptive statistics and detrended fluctuation analysis for each breathing parameter. First, the analysis of time series extracted from the entire respiratory signal is performed and an evaluation of age and BMI effects further investigate the outcomes. Then, the level of activity is considered to define a day and night variability in healthy subjects and patients. At the same time, a comparative analysis is performed to determine differences between groups. Chapter 4: Conclusions. The last chapter reports conclusion achieved in the analysis of variability and long-term correlations and summarized results described previously. The last part of the chapter is dedicated to prospective outlooks, suggesting further studies to better investigate changes in fluctuations of respiration through a sensorized garment.
Nel ventunesimo secolo, insieme ad un progressivo invecchiamento della popolazione dovuto all’aumento dell’aspettativa di vita, si sta assistendo ad un aumento delle condizioni di cronicità che richiede continue cure mediche con conseguenti implicazioni legate sia ai costi della sanità sia al peggioramento della qualità della vita. Le malattie croniche, conosciute anche con l’acronimo inglese NCDs (Noncommunicable diseases) tendono a protrarsi nel tempo e sono il risultato di una combinazione di fattori genetici, fisiologici, ambientali e comportamentali, soprattutto legati allo stile di vita. Per queste ragioni, si dovrebbe prestare molta più attenzione al monitoraggio costante della salute piuttosto che al trattamento curativo: a causa della loro natura cronica, tali malattie non sono caratterizzate solamente da episodi acuti curabili in ospedale e questa problematica tende a rendere difficoltosa la possibilità di tenere sotto controllo lo stato di salute. In questa prospettiva, i dispositivi medici indossabili potrebbero rappresentare un utile strumento per il monitoraggio combinato all’assistenza ambulatoriale e alla medicina preventiva, migliorando la diagnosi di alcune patologie, diagnosticando precocemente peggioramenti di salute e intervenendo nella definizione di trattamenti specifici per migliorare la qualità della vita. La possibilità di monitorare parametri fisiologici e biochimici in maniera continua, in condizioni di vita quotidiana e in ogni ambiente, è riconosciuta come una delle direzioni più promettenti per un servizio di cura personalizzato e minimamente invasivo. Considerando le malattie croniche respiratorie, i soggetti affetti da broncopneumopatia cronica ostruttiva (conosciuta con l’acronimo inglese COPD) trarrebbero beneficio da un monitoraggio continuo dei parametri ventilatori. Un indumento sensorizzato, inoltre, permetterebbe di controllare anche i pazienti che soffrono di apnea ostruttiva durante il sonno (OSAS). Episodi di apnea trascurati possono infatti portare ad effetti negativi che spaziano da insufficienze e scompensi a malattie cardiovascolari, aumentando il rischio di ipertensione, scompenso cardiaco o coronopatie. In questi casi, un dispositivo indossabile affidabile dal punto di visa medico potrebbe fornire un nuovo strumento per diagnosticare patologie specifiche, perfino allo stadio iniziale, e permetterebbe di controllare il paziente i durante la vita quotidiana. La disponibilità di diverse tecnologie e dispositivi ha favorito la crescita dell’interesse verso il monitoraggio continuo e a lungo termine di parametri respiratori per valutare la funzione respiratoria in soggetti sani e/o pazienti. L’analisi della frequenza respiratoria, ad esempio, è essenziale per identificare alterazioni acute e per monitorare il progresso della malattia. Anche se in passato è stata definita come “il parametro vitale dimenticato”, è stato dimostrato che valori anomali nella frequenza respiratoria sono un importante strumento di predizione di seri venti clinici come l’arresto cardiaco, che consente di discriminare tra pazienti stabili e pazienti a rischio. Inoltre, monitorare il respiro durante il sonno consente di identificare episodi di apnea e ipopnea e permette di quantificarne le occorrenze per valutare disturbi del sonno. In questo senso, dispositivi indossabili per il monitoraggio del respiro nei neonati hanno un ruolo rilevante per prevenire la sindrome della morte in culla (SIDS) che causa soffocamento. Oltre a dimostrare l’importanza della valutazione del respiro tramite parametri respiratori, la possibilità di monitorare continuamente e ovunque tali valori consente di valutare proprietà correlative e variabilità nel lungo termine. La respirazione in adulti sani durante il giorno è caratterizzata da grande variabilità nella frequenza, nella durata e nell’ampiezza dei respiri. Pur essendo generalmente considerato un processo ritmico, il respiro presenta una variabilità tra cicli consecutivi in termini di periodo respiratorio che sembra suggerire una natura frattale, rivelata da correlazioni nelle fluttuazioni nel numero dei respiri, nelle ampiezze e negli intervalli tra picchi di inspirazione. Queste correlazioni sono state studiate tramite differenti metodi di analisi frattale, tra cui la detrended fluctuation analysis. Molte ricerche hanno dimostrato che l’analisi della complessità della dinamica respiratoria può essere usata per comprendere un comportamento non lineare, per fornire una comprensione fisiologica del sistema respiratorio e come strumento clinico di valutazione di patologie. Inoltre, tale analisi consentirebbe di indagare le risposte ai trattamenti utilizzati e potrebbe essere usata nella pratica clinica, così come per il monitoraggio casalingo dei progressi della malattia. L’obiettivo del presente progetto di tesi è lo sviluppo di algoritmi per elaborazione del segnale ed estrazione di parametri respiratori dalla traccia respiratoria ottenuta sommando i cinque segnali acquisiti dai sensori. Per caratterizzare e valutare la variabilità e le correlazioni del respiro spontaneo sono state utilizzate per ogni parametro respiratorio sia la statistica descrittiva, sia l’algoritmo di detrended fluctuation analysis, usato per calcolare il parametro . Grazie alla disponibilità di una grande quantità di dati per ciascun soggetto, è possibile, inoltre, studiare diverse condizioni di termini di livello di attività ed eseguire una preliminare identificazione di sonno e veglia. In un trial clinico condotto presso il Centro Cardiologico Monzino (Milano), sono stati arruolati soggetti sani e pazienti affetti da una patologia cardiorespiratoria (con coronopatia, scompenso cardiaco congestizio e broncopneumopatia cronica ostruttiva). Questo ha permesso di valutare l’alterazione in termini di variabilità e di fluttuazioni frattali causata dalla presenza di una malattia. Durante la prima fase dello studio, un semplice classificatore per il riconoscimento dell’attività umana è stato implementato per distinguere generiche posture statiche, generici movimenti, transizioni posturali e cammino. Pur considerando che alcune problematiche, legate soprattutto all’orientamento ignoto e casuale degli assi dell’IMU in ogni logger, il classificatore consente di ottenere con un buon grado di accuratezza (sopra il 90% sia per il dataset di training sia per il dataset di test) una corretta classificazione. Inoltre, grazie alla sua semplicità è facilmente implementabile direttamente sul dispositivo e la finestra temporale di 3s con una sovrapposizione del 50% permette una classificazione in tempo reale del movimento. Per quanto riguarda il segnale respiratorio, esso è ottenuto dalla somma delle cinque tracce acquisite tramite i sensori posti sul torace, sull’addome e in corrispondenza del processo xifoideo. È stato implementato un adeguato filtraggio per rimuovere il rumore e un algoritmo per identificare massimi e minimi del segnale è stato scritto in un software progettato in ambiente MATLAB. Per indagare fluttuazioni respiro per respiro di ciascun parametro respiratorio, sono ricavate le serie temporali di volume corrente, tempo di inspirazione, periodo respiratorio, frequenza respiratoria (1/periodo respiratorio * 60), duty cycle (tempo di inspirazione/periodo respiratorio), ventilazione minuto (volume corrente * frequenza respiratoria) e flusso inspiratorio medio (volume corrente/tempo di inspirazione). Inoltre, le serie temporali espresse in funzione del numero di respiri sono ottenute sia a partire dall’intero segnale sia per tratti diurni e notturni identificati su base oraria. La statistica descrittiva permette di analizzare più approfonditamente come sono distribuiti i valori di estratti di ogni parametro e rivela, come previsto, maggiore variabilità durante il giorno e valori più elevati per alcuni parametri respiratori, quali volume corrente, ventilazione minuto, flusso inspiratorio medio, duty cycle e frequenza respiratoria. Inoltre, differenze significative sono evidenziabili tra soggetti sani e pazienti sia in termini di valore medio, sia in termini di diffusione nell’intervallo di valori e l’analisi asimmetria e curtosi consente anche di valutare la forma della distribuzione. Contemporaneamente, la detrended fluctuation analysis consente di calcolare il parametro α per ogni parametro respiratorio. Un’analisi qualitativa del grafico logaritmico in funzione del numero n di respiri ha consentito di terminare intorno a 100 respiri un punto che segna un cambio di inclinazione della retta di regressione, suggerendo un diverso comportamento frattale per lunghi e brevi intervalli, con un esponente α maggiore, che si avvicina o supera il valore unitario per un numero di respiri superior a 100. Inoltre, la suddivisione di giorno e notte rivela che durante quest’ultima, quando cioè è più verosimile identificare il sonno, l’esponente presenta valori maggiori in media rispetto a quelli calcolati in tratti diurni. Anche in termini di fluttuazioni frattali sono emerse differenze significative che rivelano maggiori differenze tra soggetti sani e pazienti con scompenso cardiaco congestizio. La popolazione di soggetti coinvolta è molto eterogenea, con soggetti sani tendenzialmente più giovani e con un BMI più basso rispetto ai pazienti. Pertanto, un’analisi delle possibili correlazioni con fattori come età ed indice di massa corporea è necessaria. I risultati provano che il volume corrente, insieme a ventilazione minuto e flusso inspiratorio medio che da esso sono derivati, tendono a diminuire con l’aumentare dell’età. Un risultato importante è invece dato dall’apparente assenza di correlazioni tra esponenti α, calcolati per il lungo e il breve termine e questi fattori. Questo potrebbe suggerire che le differenze significative ottenute sono attribuibili solamente alla presenza/assenza di patologie cardiorespiratorie. In conclusione, sia la statistica descrittiva che la detrended fluctuation analysis rappresentano strumenti utili per determinare e valutare correlazioni nei pattern di parametri respiratori. L’abbassamento del valore di α in pazienti cardiorespiratori anziani sembra suggerire, infine, una degradazione delle correlazioni che, anche se non ancora completamente chiara, getta le basi per studi futuri. La tesi si divide in quattro capitoli principali: Capitolo 1: Introduzione. In questo capitolo sono richiamate le conoscenze di base utili a comprendere l’obiettivo della tesi. Viene commentato il ruolo potenziale dei dispositivi indossabili per un monitoraggio continuo ed ubiquitario, con una e il suo dispositivo medico indossabile sono brevemente descritte. Le principali patologie respiratorie e i parametri della fisiologia ventilatoria sono spiegati descrivendo anche le tecniche non invasive tradizionali e più comuni per la loro misura. La variabilità e la caratterizzazione frattale dei segnali fisiologici sono introdotte per contestualizzare lo stato dell’arte sull’analisi frattale dela respirazione umana e sulle correlazioni presenti nei parametri respiratori. Il paragrafo riguardante gli obiettivi della tesi conclude il primo capitolo. Capitolo 2: Materiali e metodi. Il secondo capitolo descrive i passaggi sviluppati per l’analisi delle serie temporali estratte per ogni parametro respiratorio. In primo luogo, viene descritto un classificatore per il riconoscimento dell’attività. Il risultato della classificazione è usato poi per calcolare il livello di attività di ogni soggetto. Per estrarre i parametri respiratori dal segnale respiratorio, viene introdotto un algoritmo per l’identificazione di massimi e minimi. Viene presentata una breve panoramica della statistica descrittiva usata per analizzare le serie temporali. Al termine del secondo capitolo viene brevemente spiegato l’algoritmo di detrended fluctuation analysis. Capitolo 3: Risultati. Il terzo capitolo riporta l’analisi statistica dei risultati ottenuti eseguendo un’analisi di statistica descrittiva e applicando la DFA per ogni parametro respiratorio. Nella prima parte del capitolo viene considerata l’analisi di serie temporali estratte dall’intera traccia respiratoria di ogni soggetto e vengono valutati gli effetti di fattori come età e indice di massa corporea. Successivamente viene considerato anche il livello di attività per definire la variabilità tra giorno e notte nei soggetti sani e nei pazienti. Contemporaneamente un’analisi comparativa viene eseguita per determinare differenze tra gruppi. Capitolo 4: Conclusioni. L’ultimo capitolo contiene le conclusioni ottenute analizzando la variabilità e l’esito della DFA e vengono riassunti i risultati descritti precedentemente. La parte conclusiva del capitolo è dedicata a possibili sviluppi futuri e vengono suggeriti studi ulteriori per poter investigare cambi nella fluttuazione del segnale respiratorio attraverso un dispositivo indossabile.
Assessment of correlated variability in breathing parameters through a sensorized garment
CATTAPAN, ANNA MARIA
2017/2018
Abstract
In the 21st century, along with the aging of the population due to the improved life expectancy, there has been an increase in the chronic conditions that is not expected to decline and require ongoing medical care with several implications such as health costs and worsening of quality of life. Chronic diseases, also known as Noncommunicable diseases (NCDs), tend to be of long duration and are the result of a combination of genetic, physiological, environmental and behavioral factors, mostly life-style related. For this reason, more attention should be paid on health surveillance rather than afterwards treatment: due to their intrinsic long-term feature, NCDs are not acute enough to be treated in hospital and sustaining health monitoring is critical to cope with this challenging issue. In this perspective, medical wearable devices could represent a useful instrument for ambulatory care monitoring and preventive medicine, providing an improvement for pathologies diagnosing, early detection of health’s conditions worsening and for administering the best treatment to improve life quality. The possibility to monitor physiological and biochemical parameters continuously, under natural physiological conditions and in any environment is recognized as one of the most promising platforms for minimally obtrusive and individualized health services. Focusing on chronic respiratory diseases, a wide category of subjects interested in a continuous ventilatory parameters monitoring are subjects affected by chronic obstructive pulmonary disease (COPD). Another group of pathological subjects that can benefit from a sensorized garment are those experiencing obstructive sleep apnea (OSAS). Untreated OSAS could lead to several negative effects, from decrease in productivity to cardiovascular diseases, rising the risk of hypertension, cardiac failure or coronary artery disease. In cases such as these, a reliable wearable device would provide a new way to diagnose specific pathologies, even at early stages, and monitor the patient during daily living. The availability of different technologies and devices has fostered a great interest in continuous and long-term monitoring of breathing parameters to assess ventilatory function in patients and/or healthy subjects. Respiratory rate assessment, for example, is essential for detecting acute changes and for monitoring the progression of illness. Although in the past it has been defined as the “neglected vital sign”, abnormal respiratory rate has been shown to be an important predictor of serious clinical events such as cardiac arrest and to be better in discriminating between stable patients and patients at risk. Furthermore, monitoring ventilation during sleep helps to detect apneas and hypopneas, and allows to quantify their occurrences in the screening for sleep disorders. To this extent, wearable devices for monitoring babies’ breath have great importance for preventing sudden infant death syndrome (SIDS) that can cause suffocation. Besides pointing out the importance of pulmonary ventilation assessment through breathing parameters, the possibility to monitor continuously and ubiquitously their values allows to investigate long term variability and correlation properties. Respiration in awake, healthy adult humans is characterized by high variability in the frequency, duration and amplitude of breaths. Whereas breathing is generally treated as a rhythmic process, the variability in cycle-by-cycle measurements of respiratory period and breath amplitude suggests a fractal nature revealed by long-range correlations among the fluctuations in the number of breaths, breath amplitudes and peak-to-peak breath interval. Such long-range correlations have been investigated with different fractal analyses, involving either detrended fluctuation analysis or other methods. Several researches have demonstrated that the complexity analysis of breathing dynamics can be used as a method to understand nonlinear behavior, to provide a physiological insight to the respiratory system and to become a tool for clinical assessment of respiratory disorders. Furthermore, analyzing respiratory dynamics with these approaches may provide information about response to treatment, and may be used in clinical practice as well as for home-based monitoring of disease progression. The objective of this thesis is to develop algorithms to perform signal processing and extract breath-to-breath breathing parameters from the sum of the five respiratory signals. In order to characterize and investigate variability and correlation properties of spontaneous breathing both descriptive statistics and detrended fluctuation analysis, used to compute the scaling exponent , are performed for each breathing parameter. Thanks to the availability of a large amount of data for each subject, it is possible to investigate different conditions of level of activity and perform a preliminary segmentation of wakefulness and sleep. Within a clinical trial performed at the Centro Cardiologico Monzino (Milan), healthy subjects and patients with cardiorespiratory pathologies (coronary artery disease, congestive heart failure and chronic obstructive pulmonary disease) are considered for our analyses in order to investigate possible differences in the fractal dynamics and variability in control and disease states. During the first phase of the study, a simple classifier for human activity recognition is implemented to distinguish generic postures, generic activity, postural transitions and walking. Taking into account several issues that are still ongoing, mostly related to the unknown and random orientation of the axes of the IMU, the performance of the classifier allows to obtain a training and testing accuracy above 90%. Thanks to its simplicity it could be easily implementable directly on the device and the sliding window of 3s with 50% overlapping allows to obtain a real-time classification. Regarding the respiratory signal, it is obtained as the sum of the five respiratory recordings acquired through elastic strain sensors placed on the thorax, the abdomen and at the xiphoid process. A proper filtering stage is implemented to remove noise and an algorithm to detect peaks and troughs is designed in a self-written MATLAB software. In order to investigate breath-to-breath fluctuations of breathing parameters time series of tidal volume, inspiratory time, respiration period, respiratory rate, duty cycle, minute ventilation and mean inspiratory flow are extracted. Moreover, these time series (sample at each breath) are obtained both for the entire respiratory signal and separated in diurnal and nocturnal tracts identified with time segmentation on an hourly basis. The descriptive statistics allows to better investigate the distribution of the extracted values for each parameter and reveals, as expected, higher variability during day and increased values of tidal volume, minute ventilation, mean inspiratory flow, duty cycle and respiratory rate. Moreover, significant differences are highlighted when comparing control and disease states, characterizing the outcome distributions up to the fourth order statistics (i.e. mean, spread of range of values, skewness and kurtosis). At the same time, detrended fluctuation analysis allows to compute the scaling exponent for each parameter. A qualitative analysis of the obtained log-log plot established the presence of a cross-over point around 100 breaths, suggesting a change in the fractal behavior for short and long ranges, with an increased scaling exponent approaching or exceeding α = 1 for a number of breaths higher than a hundred. In addition, day and night segmentation reveals that during night, when sleep is most likely to occur, the scaling exponents exhibit higher values compared to those obtained for diurnal tracts. Even in terms of fractal fluctuations some significant differences are brought out, revealing that most of them could be found between healthy subjects and patients with congestive heart failure. The heterogeneous population enrolled, with healthy subjects younger and with lower BMI than cardiorespiratory patients calls for correlation analysis with age and BMI effects. As a result, tidal volume, minute ventilation and mean inspiratory flow, which are obtained from the formed, show associations with age and body mass index: the increase of those characteristics seems to decrease the value of the parameters. An important result consists in the absence of correlations between scaling exponents and age and BMI effects, suggesting that significant differences obtained are attributable only to presence/absence of cardiorespiratory disease. In conclusion, both descriptive statistics and detrended fluctuation analysis have been demonstrated as useful tools to assess variability and long-term correlations in breathing pattern parameters. Decreased values of scaling exponents in elder cardiorespiratory patients might suggest a degradation of correlations that, though not fully understood, poses new basis to further investigation. The thesis is divided into four main chapters: Chapter 1: Introduction. Within this chapter the theoretical basis relevant to understand the objectives of this thesis is recalled. The potential role of wearable devices in continuous and ubiquitous monitoring is investigated with particular attention to respiratory diseases. L.I.F.E. Corporation and its medical garment are briefly described. Main respiratory pathologies and ventilation physiology parameters are exposed, describing also traditional and common non-invasive measurement techniques. Variability and fractal characterization of physiological signals are introduced to describe the state of art in the study of fractal analysis of human respiration and correlated variability in breathing patterns. The paragraph about the aims of this thesis concludes the first chapter. Chapter 2: Materials and methods. The second chapter describes the steps developed for the analysis of the time series extracted for each breathing parameter. First, a classifier for human activity recognition is described. The outcome of the classification is then used to quantify the level of activity of each subject. To extract breathing parameters from respiratory signal, an algorithm for peaks and troughs detection is introduced. A brief panoramic of descriptive statistics involved in the analysis of time series is presented. In conclusion of Chapter 2, detrended fluctuation analysis is briefly explained. Chapter 3: Results. The third chapter presents the statistical analysis of the results obtained by performing descriptive statistics and detrended fluctuation analysis for each breathing parameter. First, the analysis of time series extracted from the entire respiratory signal is performed and an evaluation of age and BMI effects further investigate the outcomes. Then, the level of activity is considered to define a day and night variability in healthy subjects and patients. At the same time, a comparative analysis is performed to determine differences between groups. Chapter 4: Conclusions. The last chapter reports conclusion achieved in the analysis of variability and long-term correlations and summarized results described previously. The last part of the chapter is dedicated to prospective outlooks, suggesting further studies to better investigate changes in fluctuations of respiration through a sensorized garment.File | Dimensione | Formato | |
---|---|---|---|
TESI_ANNAMARIACATTAPAN.pdf
Open Access dal 04/07/2021
Descrizione: Testo della tesi
Dimensione
5.9 MB
Formato
Adobe PDF
|
5.9 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/10589/141620