• Aucun résultat trouvé

Conclusion and perspectives

N/A
N/A
Protected

Academic year: 2021

Partager "Conclusion and perspectives"

Copied!
6
0
0

Texte intégral

(1)

Conclusion and perspectives

The principal objective of this thesis was to study the influence of gravity in respiration, heart rate and cardio-respiratory interactions during sleep. Sleep data from 5 astronauts, before, during and after two space shuttle flights was recorded in the framework of the study

“Sleep and respiration in microgravity”, in 1998. Poor sleep quality has been a constant complaint of astronauts since the early days of space flight, and respiratory related factors were cited as a possible cause of sleep disruption. Results from another group participating in the same study tell us they are not, on the contrary: the number of respiratory related arousals is greatly reduced in space when compared to pre-flight, and snoring is almost absent in weightlessness1.

The data used throughout this work, obtained in 1998, still remains today a unique data set. Respiration and heart rate during sleep have not been studied in space since and, due to past and current constraints in space flight, it will most likely remain unique in the near future. Therefore, the need to fully exploit this data set, extracting all useful information, was one of the driving forces behind the work presented here. This led to the development and validation of two new analysis tools, namely an algorithm for detection of inspiration and expiration onset in respiratory signals, and an algorithm for heart rate variability analysis during sleep.

The first practical obstacle we were confronted with was linked to the amount of data to be analyzed. Analysis tools that are less than optimal can be used in short (few minutes) recordings, and errors manually corrected, requiring a human expert spending time visually inspecting the outcome of the algorithm. This is simply not feasible for a data set that spans more than 600 hours of respiratory and ECG signal. Reliable automated tools were required.

To our surprise, in what concerns breath detection during sleep, none existed with the required efficiency. We have therefore set out to create one such tool. The result is the algorithm presented in chapter 4 (published in the journal IEEE Transactions on Biomedical Engineering, in 20022). The neural network based algorithm clearly outperforms previously published algorithms for breath detection, presenting higher specificity and similar or higher positive predictive values. Moreover, the algorithm was found to be indistinguishable from a

1A.R. Elliott, S.A. Shea, D.-J. Dijk, J.K. Wyatt, E. Riel, D.F. Neri, C.A. Czeisler, J.B. West, and G. Kim Prisk. Microgravity reduces sleep-disordered breathing in humans. Am J Resp Crit Care Med, 164(3):478-485, 2001

2R.C. S´a and Y. Verbandt. Automated breath detection on long-duration signals using feedforward backpropagation artificial neural networks. IEEE Trans Biomed Eng 49(10): 1130-1141, 2002.

245

(2)

group of 5 human experts performing the same task. The algorithm was successfully applied to breath detection in infant respiratory data, proving its use was not restricted to adult sleep. A similar algorithmic approach was successfully implemented in the analysis of a completely different time series, the doppler acquired time series of blood velocity in the middle cerebral artery. The goal of the algorithm was to detect the systolic and diastolic velocity of blood flowing in the middle cerebral artery.

Recently, a simpler, striped down version of this algorithm has been implemented as the building block of an automated, almost real-time quantification algorithm for cardiopul- monary resuscitation performance3. The ultimate goal of the efforts by this specific group is to create an algorithm capable of giving online corrective feedback to a human operator and improve the quality of cardiopulmonary resuscitation. The neural networks based approach we introduced was used by Risdal et al due to its performance in event detection, but even more so due to its resilience to drifts, high levels of noise, and signal corruption, all of which are frequent during cardiopulmonary resuscitation.

The study of heart rate variability during sleep, and more specifically of its high fre- quency component - respiratory sinus arrhythmia - was at the origin of the development of the wavelet based algorithm presented in chapter 6. Our goal was to analyze heart rate vari- ability during sleep, and to follow its variations through different sleep stages. Once sleep stages are usually scored in windows of 30s, our aim was to obtain a similar time resolution for the analysis.

Though time-domain methods for the study of heart rate variability have been proposed, quantification of heart rate variability is mostly performed using time-frequency analysis, mostly Fourier analysis, following established guidelines. Fourier analysis, though a powerful tool, presents limitations, namely at the level of the time-frequency compromise that can be implemented: a long window improves frequency accuracy, but decreases time localization, and vice-versa. For each frequency and case of interest a particular choice of window length is optimal, but studying different frequency bands, as in the case of heart rate variability, requires a compromise. Fourier windows should be long to improve frequency resolution, yet longer window length results in decreased time localization. The necessary tradeoffs can be quite impractical when analyzing specific frequencies, and this effect is worsened in short time series or in time series that are only locally stationary, resulting in outputs that are averaged over different states.

Instantaneous heart period is essentially determined by two competing autonomic path- ways, the vagal and the sympathetic system. Due to the time constants involved in vagal and sympathetic modulation of heart period, heart period variability in the 0.15 to 0.4 Hz range is mainly associated with vagal activity, while variability in the 0.04 to 0.15 Hz range is driven both by sympathetic and vagal modulation.

In order to quantify heart rate variability in the different frequency bands during sleep, two difficulties arise: the first is associated with changes in sleep stages, for as sleep stages vary, heart rate variability also changes, creating local non-stationarities; the second in as- sociated with breathing frequency. Heart period variability in the 0.15–0.4 Hz is usually centered around the respiratory frequency. When breathing frequency is low, which hap- pens frequently during sleep, the variability present in this frequency range is close to the 0.15Hz frontier that separates low and high frequency oscillations. In this situation, Fourier algorithms using short windows (30 s) are not able to distinguish between the two different components. Using longer Fourier windows (120 s) allows the correct separation of these

3M. Risdal, S. O. Aase, M. Stavland and T. Eftestol; Impedance-based ventilation detection during cardiopulmonary resuscitation. IEEE Trans Biomed Eng, 54(12):2237-45, 2007.

(3)

frequency components, yet at the expense of time localization, resulting in outputs that are averaged over a longer time period, sometimes different sleep stages, blurring transients.

Wavelet analysis is a multi resolution method: its time and frequency resolutions are not fixed, but depend on the frequency being analyzed. Wavelets analyzing higher frequencies will be more localized in time, while those analyzing lower frequencies will present less time localization. It is thus possible to establish a fairer compromise for the time-frequency resolution. Moreover, by using a complex wavelet, the determination of the phase difference between the respiratory and heart period time series can be computed.

As demonstrated in chapter 6, a wavelet algorithm offers a better time-frequency resolu- tion compromise than the traditional windowed Fourier transform approaches. The wavelet algorithm was tailored to meet the specific needs of the present experiment: a time res- olution, in the high frequency band, smaller than 30 s, and a frequency resolution in the frequency range of 0.10 to 0.20 Hz capable of correctly separating the low and high fre- quency components of HRV. A paper describing the algorithm and presenting its validation against standard windowed Fourier analysis was accepted for publication in the journal IEEE Transactions on Biomedical Engineering, in 20074.

Chapter 7 presents the most relevant findings concerning respiration during sleep in microgravity. Its contents have recently been submitted for publication in the Journal of Applied Physiology 5.

In short, respiratory mechanics during sleep is altered by exposure to microgravity. An early flight increase in abdominal contribution to tidal breathing was present for all sleep stages, followed by a decrease towards pre-flight values in subsequent days, denoting a slow adaptation to the new environment. The reasons for this slow adaptation are yet unknown, though the strongest candidates are muscular adaptation and/or changes in central neural control. This last cause is the most likely, probably resulting in either a reduction of drive to the diaphragm or in a better coordination in the sequence of activation of rib cage and abdominal muscles, improving the relative partition of respiratory effort with time spent in weightlessness.

Despite the initial change and subsequent adaptation in abdominal contribution, sleep induced differences in respiratory mechanics present on earth between NREM and REM sleep persisted in weightlessness, this despite the derecruitment of rib cage respiratory muscles that is reported to occur for awake subjects in space.

During REM sleep in microgravity, thoraco-abdominal asynchrony - a measure of the abdominal tendency to lead the rib cage in the movements that create a tidal excursion - was increased compared to pre-flight, indicating a less efficient coupling between the di- aphragm and the lower rib cage in space. On the other hand, despite the fact that abdominal contribution to tidal volume during NREM sleep in microgravity changed in parallel and with similar magnitude as for REM sleep, thoraco-abdominal asynchrony decreased in space during NREM sleep, with the rib cage leading the diaphragm.

Previous studies have reported that NREM sleep causes a significant increase in the activity of the rib cage respiratory muscles, while during REM periods a significant decrease in their activity is reported. Taking this into account, two reasons can explain these pre- flight–microgravity differences in thoraco-abdominal asynchrony: the diaphragm is in a less

4L. Cnockaert, P.F. Migeotte, L. Daubigny G. Kim Prisk, F. Grenez, R.C. S´a. A methods for the analysis of respiratory sinus arrhythmia using continuous wavelet transforms. IEEE Trans Biomed Eng, 55(5), 1640-1642, 2008.

5R.C. S´a, G. Kim Prisk, M. Paiva. Microgravity alters respiratory mechanics during sleep. Submitted to J Appl Physiol.

(4)

efficient operating position in microgravity, or the coupling between the two respiratory compartments is less effective in space.

Respiratory drive was found to be increased in weightlessness for all sleep classes, while the duty cycle (TI/TT ot) was decreased. If the diaphragm is indeed in a less efficient opera- tional position in microgravity than pre-flight supine, this increase in drive can compensate, or even override it, maintaining ventilation.

The wavelet algorithm described in chapter 6 was applied to the time series of heart pe- riod and respiration during sleep, in order to probe eventual changes in heart rate variability and cardio-respiratory interactions induced by exposure to weightlessness. The results ob- tained are presented in chapter 8.

When compared to its pre-flight average, heart period increased for all sleep classes early in space flight, and returned to pre-flight values for sleep episodes recorded late in the flight.

Post-flight, heart period decreased below pre-flight values. Total heart period variability was significantly decreased in-flight. Yet contrary to prior reports for awake subjects, during sleep the decline seen in early microgravity recordings continued into the late portion of the flight. Post-flight, total variability remained much lower than pre-flight.

Despite the relevant decrease in overall variability, sympatho–vagal balance in micrograv- ity remained unchanged for awake and NREM sleep, and presented a tendency towards a less vagal, more sympathetic state during REM sleep, as time spent in microgravity increased.

This resulted in an increase of the NREM–REM differences present pre-flight.

The gain and the phase delay between respiration and instantaneous heart period, es- timated at the breathing frequency, remained essentially unchanged when comparing pre- flight, microgravity and post-flight.

Perspectives Algorithms inspired in the work presented here on breath detection can be used and integrated with “smart textiles” or other types of sensors for ambulatory, real- time monitoring of respiration. As in the case of cardiopulmonary resuscitation, ambulatory monitoring is an environment where high levels of noise and drifts are present and where the use of a sophisticated detection algorithm can bring significant improvements in breath detection efficiency.

This work was, to the best of our knowledge, the first study of respiration during sleep where respiration was analyzed throughout the entire night, and not just in selected por- tions. The lack of efficient automated tools for respiratory event detection in long duration respiratory signals led previous researchers to choose portions of less than 5 minutes as representative of a sleep stage or sleep class, ignoring the remaining signal. Applying the same continuous all night approach to the study of respiratory sleep disorders might help to further characterize and understand their origins and consequences.

The wavelet based heart rate variability algorithm opens new windows on the study of sudden changes in heart rate control, namely when induced by changes in posture and/or gravitational load (tilt tables, parabolic flight), or changes in respiratory frequency.

For both respiratory and heart rate, this work focused on quantifying the effect of micro- gravity in respiration and respiratory mechanics, heart rate and heart rate variability, during sleep. Therefore no special attention was devoted to what happens inside each individual sleep episode.

(5)

In the future, the same data can be exploited, from the respiratory point of view, to study the dynamics of respiration during sleep, to characterize how sleep stage transitions alter respiratory mechanics, drive and ventilation. In what concerns heart rate, it is also interesting to study changes occurring with time spent sleeping (early–late night differences), but also the evolution of sympathetic–vagal control of the heart rate in the different sleep stages.

Studying heart rate control during sleep, and the changes happening within each night, might help to shed light into questions such as why sudden cardiac death is more likely to happen in the morning, if this is at all connected with sleep.

Special attention should also be devoted to the pre-flight data on its own, for it constitutes a rare longitudinal sleep study on healthy subjects, with 6 to 9 sleep episodes recorded for each subject over 3 months. Despite a large and increasing number of publications concerning respiratory disorders during sleep, there is a lack of normative data for healthy subjects.

Moreover, most studies that do focus on healthy subjects have studied teenagers or young adults, while in this study the average age of subjects was 41.0 years, closer to the age range of the population moslty affected by respiratory sleep disorders.

General conclusions and perspectives The primary objective of this work was to in- crease the basic knowledge of the effect of microgravity on human respiration, heart rate and their interactions, during sleep. It covers many different topics, from physics, engineering, signal processing to physiology, and is intrinsically interdisciplinary. It was necessary, for me, to acquire knowledge in respiratory, heart rate and sleep physiology, as well as on the impact of weightlessness in all these systems. On the signal processing side, applying arti- ficial neural networks and wavelets to biomedical signal processing was also a challenging, creative learning experience. Crossing borders of our own disciplines is not only enriching for all involved, but essential and necessary for biomedical sciences, and science in general.

As a physicist and engineer, I am glad that in the pursuit of this goal, I have been driven to develop signal processing tools that I hope will also help others in their quest of basic knowledge on respiratory and cardiac physiology. I am extremely happy that my work in classifying respiratory events has already inspired others, in a way that might improve health care and probably one day help to save lives.

(6)

Références

Documents relatifs

The ARO Observer is produced by the Astrophysics Branch of the National Research Council to serve as a forum for the Canadian astronomical community, and in

In an age of rising environmental concerns, financial recession, and political change, companies are increasing corporate social responsibility (CSR) activities in

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

Unwed mothers who decide to place their child for adoption were compared, both pre- and post-parturition, with those who chose to keep their child.. Dependent variables included

While on the local scale observed soil data should be used to minimise uncertainties of model applications, the German texture classification seems to be appropriate for regional

Dans ce court paragraphe, non seulement les diverses reprises permettent d’éviter la répétition, mais aussi l’emploi de termes plus spécifiques contribue à ajouter des

Pour organiser un atelier philosophique en classe, à partir de ces grandes images, découvrez des outils pédagogiques sur www.pommedapi.com?. À ton avis, un ami ça sert

[r]