Embedded neonatal respiration monitoring
Citation for published version (APA):Hoeben, R. M., & Technische Universiteit Eindhoven (TUE). Stan Ackermans Instituut. Design and Technology of Instrumentation (DTI) (2011). Embedded neonatal respiration monitoring. Technische Universiteit Eindhoven.
Document status and date: Published: 01/01/2011
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
Eindhoven University of Technology
Stan Ackermans Institute: Design and Technology of Instrumentation
School of Medical Physics and Engineering Eindhoven: Qualified Medical Engineering In collaboration with:
Philips Research: Biomedical Sensor Systems Máxima Medical Centre Veldhoven
Embedded neonatal respiration monitoring
Ir. R.M. Hoeben
August, 2011
Company confidential
Committee:
Martijn Schellekens, Senior Scientist Philips Research
Anton Janssen, Research Scientist Philips Research
Sidarto Bambang Oetomo, Neonatologist MMC
Astrid Osagiator, Senior nurse neonatal care MMC
Ward Cottaar, Director DTI and SMPE/e
Ivonne Lammerts, Coordinator SMPE/e
Jan Botman, Co-director DTI
trodes used to measure the heart and breathing rate, can cause skin irritations and skin lesions when being pulled off. Furthermore, all the wires create a barrier for parents to touch and interact with their child.
The E-Nemo (Embedded Neonatal Monitoring) project intends to change the way in which (pre-mature) neonates are monitored in the neonatal intensive care unit (NICU).
The aim of E-Nemo is to create a patient support system that assures comfort for the neonate and provides a more friendly environment for parental bonding, whilst keeping the current quality of vital sign monitoring.
This report concerns the work related to the monitoring of only one vitals sign, namely: respira-tion. The aim of the E-Nemo respiration monitoring project is to design and develop a neonatal respiration monitoring system using sensors embedded in a patient support system (e.g., a mat-tress). A key challenge of this system is achieving the same robustness and reliability as existing monitoring equipment for neonates.
Before the respiration sensor can be moved from the chest of the neonate into the underlying support system some questions need to be answered. Such as: Where can we place this sensor? Is one sensor enough? Which type of sensor is most suitable?
To answer these (and more) questions regarding the design of the neonatal respiration monitoring system, a clinical trial was conducted at the NICU of the Máxima Medical Centre in Veldhoven. During this trial firsthand knowledge on the position and movement of neonates in an incubator, and general NICU workflow issues was gained.
The clinical trial has resulted in a list of design specifications for the neonatal respiration moni-toring system and a better understanding of the workflow and possible measurement disturbances in a NICU.
Furthermore, this project has successfully demonstrated the possibility of measuring the neonatal respiration signal without direct skin contact with the neonate. However, in order to achieve the quality and reliability needed for intensive care respiration monitoring more research is necessary. Measuring the deformation of the mattress is expected to be a better measure for the respiration movements, than the pressure changes underneath the mattress which were measured in this study. Furthermore, more research is needed to determine the accuracy that will be demanded of the sys-tem, as this research has demonstrated that the current gold standard (transthoracic impedance plethysmography) does not function continously either.
Samenvatting
De bewakingssystemen gebruikt bij pasgeborenen zorg zijn niet optimaal voor het comfort van deze kinderen. Om de hartslag en ademhaling te bewaken wordt gebruik gemaakt van plakelec-troden. Zoals de naam al doet vermoeden zitten deze op rechtstreeks op de huid geplakt en dat kan huid irritatie en zelfs wondjes veroorzaken wanneer ze worden verwijderd. Tevens is het voor ouders moeilijker om een band met hun kind op te bouwen en hun kind aan te raken, wanneer hun kind aan allerlei draadjes ligt.
Het E-Nemo project wil de manier waarop (te vroeg geboren) babies op de neonatale intensive care unit (NICU) worden bewaakt veranderen. In plaats van plakelectroden is het de bedoeling om de sensoren te verwerken in het matras van de kinderen, zonder daarbij in te leveren op de betrouwbaarheid van de bewaking. Op deze manier wordt het comfort van de kinderen verhoogt en kunnen ouders hun kind makkelijker benaderen.
Dit verslag gaat over een klein deel van het E-Nemo project gericht op de ademhalingsbewak-ing. Het doel van het E-Nemo ademhalings bewakings project is om een systeem te ontwerpen en ontwikkelen dat de ademhaling van (te vroeg) geboren babies kan bewaken, met behulp van sensoren die in het matras zitten verwerkt. Een van de grootste uitdagingen binnen dit project is het behalen van dezelfde kwaliteitsstandaard als de huidige ademhalingsmonitor.
Voordat dit nieuwe systeem kan worden ontwikkeld moeten er eerst een aantal vragen beantwoord worden, zoals: Waar moeten de sensoren komen? Hoeveel sensoren zijn er nodig? En wat voor sensoren moeten dan gebruikt worden? Voor het ontwerp van het nieuwe ademhalings bewakings systeem is het nodig deze (en andere) vragen te beantwoorden. Daarom is een klinisch onderzoek uitgevoerd in de NICU van het Máxima Medisch Centrum in Veldhoven.
De resultaten van dit klinisch onderzoek hebben geleidt tot de ontwerp specificaties van het neona-tale ademhalings bewakings systeem en meer inzicht omtrend neonaneona-tale zorg.
Verder heeft dit project bewezen dat het mogelijk is om de ademhaling van babies te meten, zonder dat er direct contact is tussen het kind en de sensor. Om te zorgen dat het bewakingssysteem ook de kwaliteit krijgt die wordt verwacht op een intensive care is het belangrijk om nog meer onderzoek te doen naar het huidige bewakingssysteem en om in plaats van de drukverschillen onder het matras te kijken naar vervorming van het matras ten gevolge van de ademhaling.
Executive Summary 1
Samenvatting 2
1 Introduction: neonatal care 5
1.1 Embedded neonatal monitoring . . . 7
2 The project: E-Nemo respiration monitoring 9 3 The clinical trial: preparation 11 3.1 Neonatal respiration monitoring: Current practice . . . 11
3.2 Measurement set-up . . . 12
3.2.1 The sensor mat . . . 12
3.3 Preliminary tests . . . 15
3.4 The clinical trial . . . 16
4 The clinical trial: data analysis 19 4.1 Position analysis . . . 19
4.2 Respiration analysis . . . 20
5 The clinical trial: results 24 5.1 Position detection . . . 24
5.2 Respiration analysis . . . 26
5.2.1 Location of the respiration signal . . . 30
6 Design specifications 32 6.1 Design of a neonatal respiration monitoring system . . . 32
6.2 Future work . . . 34 7 Conclusions 35 8 Acknowledgements 36 Appendices 37 A Project Plan 38 B Neonatal Care 46 B.1 Developmental care . . . 48 B.2 Neonatal monitoring . . . 49
C Home monitoring equipment 51 D Measurement set-up 53 D.1 Pressure distribution: Sensor Mat . . . 53
D.2 Posture reference: Webcam . . . 55
D.3 Respiration reference: Porti . . . 55
D.4 Data acquisition: Laptop . . . 56
E The tekscan sensor system 62 E.1 Measurement properties . . . 62
E.2 Characterizing the sensor mat: static tests . . . 64
E.3 Characterizing the sensor mat: dynamic tests . . . 75
E.4 Influence on measurements . . . 78
F Data classification 80 F.1 Manual data annotation . . . 80
F.2 Automatic motion detection . . . 81
F.3 Blob detection . . . 83
F.4 Pressure dataset generation . . . 85
G Data analysis methods 87 G.1 Position analysis . . . 87
G.2 Respiration analysis . . . 88
H Results 101 H.1 Performance of the reference . . . 101
H.2 Position analysis . . . 105
H.3 Respiration analysis . . . 108
H.4 Summary . . . 123
I Design of the embedded neonatal monitoring system 125 I.1 Design specifications . . . 125
I.2 Sensor positioning . . . 128
I.3 Use models . . . 129
J ICBE forms 132 J.1 Global strategy plan, REQ-ENEMO-201012-GSP . . . 133
J.2 Trial strategy plan, REQ-ENEMO-201012-TSP . . . 137
J.3 Initial risk analysis, REQ-ENEMO-201012-IRA . . . 144
J.4 Consent form . . . 149
J.5 Instruction for nurses . . . 153
Neonatal care
Neonate
An infant in the first 4 weeks of life, a newborn child
There is no greater joy than holding your child in your arms for the very first time. Getting to know the child that was growing in the womb all that time is a very exciting period. Unfortunately, not all parents are able to enjoy this time in the comfort of their own home. Some babies need more intensive and specialised care which requires them to remain in the hospital. The majority of the children that cannot go home with their parents were born too early.
A normal pregnancy lasts 37 to 42 weeks. Children born before 37 weeks are called premature neonates [Horbar and Lucey, 1995]. The term premature is used because the child did not have time to fully develop within the mothers womb. A premature neonate is therefore much smaller and lighter than a full term neonate. Their length ranges from 25 to 40 cm, and weight from 500 to 3000 g, depending on the extent of prematurity, whilst full-term neonates are 50 cm in length and weigh 3000+ g [Fenton, 2003; Organization, 2006].
Due to the immaturity of their organs, premature neonates often require specialised care. There-fore, most premature neonates spend their first few weeks of life in a neonatal intensive care unit (NICU) until they are healthy and mature enough to go home with their parents. At a NICU, specialised physicians (neonatologists) and nurses care for the neonates in an environment where there is quick access to dedicated equipment and resources.
Thanks to the improvement in quality of neonatal care, the chances of survival of premature neonates at lower birth weight have increased over the last decades. Although more and more premature infants survive, it is uncertain how their long-term development is affected by their premature birth and if they will be able to have normal lives. Concerns have risen on the relation between preterm birth and an impaired psychomotor development, which results in delayed cogni-tive development and general behavioral problems. During follow-up studies, 15-25% of premature born infants at 2-5 years of age had a disability (physical, cognitive (mental), sensory, emotional, developmental or a combination). In comparison 30-40% of low-birth-weight infants had a lower
CHAPTER 1. INTRODUCTION: NEONATAL CARE
school performance and 20-40% had more behavioral problems than full-term infants [Kleberg et al., 2000; Westrup et al., 2004; Hack et al., 1995; Horbar and Lucey, 1995].
The stabilization of the survival and morbidity rates give rise to several questions [Hack and Fa-naroff, 2000]. What is the lower limit of gestational age1? Below which weight and gestational
age is treatment futile, or even doing more harm than good for the individual infant, its family and society? Has the maximum survival rate and best quality of life for premature neonates been reached?
Or, are there methods to further increase survival rate and decrease long-term morbidity of pre-mature neonates?
An answer to that last question might be: changing the NICU environment. There is an obvious difference between the NICU environment and the mother’s womb. As aggressive brain develop-ment occurs prior to birth (continuing to age three), a premature neonate is subjected to all sorts of stimuli that may not correspond to its developmental needs. Stress is known to have a nega-tive effect on the neonates neurobehavioral development and, in turn, impaired neurobehavioral development can cause long-term problems such as mentioned previously [Kleberg et al., 2000; Symington and Pinelli, 2003].
Furthermore, when a neonate has to remain in the hospital during the first weeks of life, the parent-child bonding process is also disturbed. Depending on the severity of illness of the child the parents will have less interaction with their baby [Kleberg et al., 2000].
Developmental care has been advocated to improve the development and long-term outcome of the premature neonate. It was introduced in the 1980s and provides a strategy to adjust the environment to decrease stress and involve the parents in the care for the neonate, whilst keeping the high quality of technological care. More information on developmental care can be found in Appendix B.
Studies show that premature neonates benefit from developmental care. They tend to recover more quickly, are shorter on a ventilator, feed on their own earlier and have improved short and long-term outcome compared to other preterm neonates [Als et al., 1994; McAnulty et al., 2010; Kleberg et al., 2000; Symington and Pinelli, 2003].
Technology plays a large part in the neonatal care environment. Physiological parameters such as increased heart rate and decreased oxygen saturation offer valuable insight in the condition of the neonate. It is therefore essential that the neonates vital signs are monitored continuously, so that a nurse can be alerted when the neonate is in distress.
Patient monitors provide real-time continuous information on the neonates vital signs and can alert nurses to changes in one or more vital signs. They make use of combinations of non-invasive optical and electrical sensors, and are typically equipped for monitoring the functioning of the heart (beat and rate , electrocardiogram (ECG)), lungs (oxygen saturation SpO2and respiration),
brain (EEG electroencephalogram/CFM cerebral function monitoring), and body temperature [Murković et al., 2003]. These monitoring techniques are described in more detail in Appendix B.2.
1Not only technically possible, but also ethically. The Dutch law currently recognizes a child born after a
pregnancy of 24 weeks as viable. This threshold is coming closer and closer to the upper limit for abortion [Dutch Ministery of Justice].
1.1
Embedded neonatal monitoring
Unfortunately, current neonatal monitoring methods are somewhat conflicting with patient well being. For instance, the sticky electrodes used to measure the electro-cardiogram, heart rate and breathing rate, shown in Figure 1.1, can cause skin irritations and skin lesions when being pulled off. Because of this, the electrodes are minimally adhesive which in turn causes the electrodes to fall off and cause many false alarms. Especially in the very premature neonates there is not enough surface area to properly place the electrodes.
Furthermore, for parents all the wires create a barrier to touch and interact with their child while parental bonding is essential for the newborn. Therefore, there is a need to develop an alternative to the sticky electrodes.
Figure 1.1: Premature neonatal twins in an incubator
The E-Nemo project aims to change the way in which (premature) neonates are monitored in the neonatal intensive care unit. E-Nemo stands for Embedded Neonatal monitoring. The E-Nemo project aims to create a patient support system that assures comfort for the neonate and provides a more friendly environment for parental bonding, whilst keeping the current quality of vital signs monitoring.
This patient support system (for example a mattress) has vital sign sensors embedded into the design, thus eliminating the need for sticky electrodes. A scheme of this idea is shown in Figure 1.2. The monitoring techniques include heart rate, breathing rate, pulse oximetry (SpO2), the electro-cardiogram (ECG), the peripheral and core temperature and eventually the electro-encephalogram (EEG). Although the feasibility of unobtrusive sensors for all these modalities has been demon-strated to a certain extend for adults in different projects more research is needed to demonstrate their performance when used for (premature) neonates and to investigate how they should be embedded in the patient support system.
CHAPTER 1. INTRODUCTION: NEONATAL CARE
Figure 1.2: The scheme of the patient support system: technologies to monitor the heart, breathing rate, temperature and oxygen saturation are embedded into the mattress.
Philips Research is not the only party involved in the E-Nemo project. A consortium of five partners is working together on the E-Nemo project:
Máxima Medical Centre Veldhoven (MMC): Clinical validation and user insight Philips Research: Sensor technology and market research
Eindhoven University of Technology (TU/e): System design and signal processing Royal Health Foam (RHF): Mattress technology and production
Applied Micro Electronics (AME): Electronics development and production
As stated previously, the aim of the E-Nemo project is to monitor several vital signs using sensors that are embedded in the patient support system. A key challenge of this system is achieving the same robustness and reliability as existing monitoring equipment for neonates. A method to achieve this robustness is using different types of sensors to measure the vital signs. By using different sensors, disturbances can be mitigated because not all sensors are influenced by that par-ticular disturbance. Thus, achieve intensive care quality of vital signs monitoring through sensor fusion .
The sensor development of the E-Nemo project consists of four subprojects, and each sub-projects focusses on a different vital sign and different sensor modality. The four subprojects are: Cardiac (ECG), Oxygen saturation (SPO2), Temperature, and Respiration monitoring. This report con-cerns the E-Nemo subproject focussed on unobtrusive respiration monitoring.
E-Nemo respiration monitoring
The aim of the E-Nemo respiration monitoring project is to design and develop a neonatal respi-ration monitoring system using sensors embedded in a patient support system.
Before the respiration sensor can be moved from the chest of the neonate into the underlying support system (e.g., the mattress) some questions need to be answered. Such as: Where can we place this sensor? Is one sensor enough? Which type of sensor is most suitable?
Basically at this stage we have an idea, to get from this idea to an actual product we need to go through several development phases in order to get the design specifications needed to create the actual prototype of the neonatal respiration monitoring system (Figure 2.1).
• Unobtrusive neonatal respiration monitoring
Idea
• Position and movement of neonates • Understanding NICU workflow Pre-development • Sensitivity • Accuracy • Sensor location Specifications • New neonatal respiration monitoring system Product designFigure 2.1: Project workflow
The NICU is a very complex working environment, and at this stage we have not enough knowl-edge about how the new system should and could function optimally in this environment. As we aim to change the way neonatal vital signs are monitored we need to be absolutely sure that the quality of care is increased by using this new neonatal respiration monitoring system. There-fore, the pre-development stage of the E-Nemo respiration project is focussed on gaining firsthand knowledge on the position and movement of neonates in an incubator, and general NICU workflow issues that should be taken into account in the design of the prototype.
CHAPTER 2. THE PROJECT: E-NEMO RESPIRATION MONITORING
Approach
In order to get an overview of the design specifications of the neonatal respiration monitoring system the following steps were taken:
1. Clinical trial preparations
• Study the current respiration monitoring system • Assemble the measurement set-up
• Obtain permission to perform a clinical trial
2. Perform a clinical evaluation to acquire data that will help in designing the patient support system
• Measure the pressure distribution of a neonate over time using a pressure sensitive sensor mat (gives information on position and movement).
• Obtain already recorded vital signs from current monitoring systems (for development and validation purposes).
• Increase understanding of environment and issues that a neonate experiences. • Get an overall understanding of the workflow and relating issues in the NICU. 3. Use this clinical data to:
• Determine locations where the neonates reside on the mattress.
• Determine locations where the respiration of the neonate can best be measured. • Evaluate the performance of the current respiration sensor
4. Design the first respiration system prototype using the information gathered during the clinical tests (requirements)
A more detailed description of the project plan and schedule can be found in Appendix A. During the clinical trial a pressure sensitive sensor mat (array of pressure sensors) was used to measure the pressure distribution of neonates over time. The changes in the pressure distribution correspond to the neonates’ movement, including the respiration movement.
A great benefit of using an array of pressure sensors is that this will also provide detailed infor-mation on the location of the neonate and in which locations respiration is detected. Thus aiding in the design of the new respiration sensor system, and the positioning of other vital sign sensors and overall design of the structure of the patient support system. Furthermore, by performing these tests in a NICU possible measurement disturbances and workflow issues will also become apparent. Besides the pressure sensor mat, the current respiration signal was recorded to verify the detection of respiration and evaluate the performance of the pressure sensor mat (reference). One of the main challenges is that this system will have to match or even exceed the performance of the current respiration sensor. Additionally, visual information is required to verify the position of the neonate and to be able to check the cause of disturbances .
The conclusions of the clinical tests will be translated into specifications of the new unobtrusive neonatal respiration monitoring system.
The clinical trial
During this project the position and movement (including respiratory movement) of premature neonates were recorded during a clinical trial. Knowledge on the position and movement behaviour is needed for the design of an unobtrusive neonatal respiration monitoring system. One of the main challenges is for this system is to match or even exceed the performance of the current respiration sensor. Therefore the current situation: its advantages and disadvantages was investigated first.
3.1
Neonatal respiration monitoring: Current practice
There are several techniques to monitor the respiration based on mechanical and electrical tech-niques. The most common method for monitoring respiration (aside from visual monitoring) is transthoracic impedance plethysmography (TTI) [Lindberg et al., 1992]. TTI is considered the current gold standard in monitoring neonatal respiration [Johansson et al., 1999].
In TTI, an alternating current is passed between two electrodes that are placed on the chest (Fig-ure 3.1). The changes in the chest volume and its composition (percentage of air relative to tissue) caused by respiration affect this current. The pattern that can be seen in the changing current allows for derivation of the respiration frequency.
Besides the changes in gas volume, the changes in blood volume also influence the impedance change. For adults this is rarely a problem as chest area and the tidal volume are much larger (500ml compared to 10-30 ml for neonates [Khurana, 2008; Cook et al., 1957; Keszler and Abubakar, 2004]), and the heart and breathing rate differ significantly. The adult resting heart rate is 60-70 beats per minute and respiration rate 12-15 breaths per minute, while neonates can have res-piration rates of 120 breaths per minute, which is very close to the neonatal heart rate which is approximately 120-180 beats per minute [Khurana, 2008; Engoren et al., 2009; Girling, 1972; Eiselt et al., 1993]. Thus, in neonates the cardiac activity can contribute significantly to the detected transthoracic impedance, which hinders respiratory monitoring [Li et al., 1977].
Monitoring the respiration rate in neonates is challenging because premature neonates have an irregular respiratory activity [Äärimaa et al., 1988]. The regular respiration frequency is estimated at 50 - 60 times a minute [Fenner et al., 1973], though rates between 10-80 (or even 120) times per minute are not uncommon [Lindberg et al., 1992; Engoren et al., 2009].
Apart from irregular breathing, neonates sometimes experience breathing pauses. When these pauses last longer than 10-15s, they are called apnea periods and require intervention (e.g., wak-ing the neonate) [Fenner et al., 1973]. A total of 25% of neonates who weighed less than 2,5 kg and 84% of the neonates below 1kg experience apnea during their first weeks of life. It is believed that these pauses are caused by the immaturity of the central respiratory control mechanisms. There are two underlying causes for apnea: the brain is not sending a signal to breathe (central apnea) or the airway is blocked (obstructive apnea). A central apnea occurs in about 50% of the cases, and obstructive and mixed apnea (combination of central and obstructive apnea) make up the other half in respectively 10 and 40% of the apnea cases [Kattwinkela et al., 1975; Abu-Shaweesh and Martin, 2008; Finer et al., 1992].
Current monitoring systems are very well capable of identifying periods of central apnea. How-ever, during an obstructive apnea the breathing motion continues normally which makes it more
CHAPTER 3. THE CLINICAL TRIAL: PREPARATION
Figure 3.1: Scheme of neonatal respiration measurement
difficult to detect. Examining the air flow or combining the respiration measurement with the heart rate or saturation values can be used to detect an obstructive apnea.
The respiration signal detected using TTI is used as a reference respiration signal, and is therefore also recorded during the clinical trial.
3.2
Measurement set-up
To record the position and (respiratory) movement of neonates four devices were used: 1. A laptop: to control all measurement devices and store all acquired data.
During measurements a custom written program was started that controlled the entire recording (details can be found in Appendix D.4). While the measurement was running, visual feedback was provided in the form of the pressure distribution and webcam image. 2. A webcam: to capture the inside of the incubator (posture of the neonate and activity e.g.,
presence of a nurse).
The webcam (Logitech C510 HD Webcam) was mounted on a camera stand with suction cup to allow flexible and easy attachment to the incubator..
3. An analog digital converter: to record the currently measured respiration signal (impedance measurement as reference).
The analog digital converter (Porti, Twente Medical Systems International BV) was con-nected to the analog out of the GE monitor.
4. A sensor mat: to measure the pressure distribution of a neonate (position and movement). This sensormat is part of the Tekscan Body Pressure Measurement System (BPMS). The properties of the sensormat are described in more detail in the next Section (3.2.1). To protect both the neonate and the sensor mat itself it was covered with a PU (polyurethane) coated fabric that is water resistant. This type of fabric is often used for mattress coverings in health care (Seyntex, Belgium).
These devices and their positioning are shown in Figure 3.2. A more extensive description of all these devices can be found in Appendix D.
3.2.1
The sensor mat
The sensor mat consists of ultra-thin flexible printed circuits. Each circuit consists of silver elec-trodes (conductive) and a layer of pressure-sensitive ink. Its electrical resistance changes upon a
Figure 3.2: The placement of the measurement set-up at the NICU.
change in applied pressure. When a force is applied to the sensor, its resistance decreases. The locations where strips of ink (array of rows and columns) intersect are the sensitive areas. In stead of referring to these areas as sensors (which suggest them being separate), they are called sensels. The reason for making this distinction is that it is important to note that they are not individual sensors but somehow connected due to the usage of strips of ink.
During the trial two tekscan models of sensor mats were used: the 5315 and the 5350N (Fig-ure 3.3). The reason for using two sensor mats is that a more sensitive version was desired and the 5350N is more sensitive (and was not on the market at the start of the clinical trial). A sensor mat is connected to a tekscan handle (Figure 3.3), which performs the data acquisition.
The system can also be calibrated to obtain either absolute pressure (Pa), load (N) or weight (kg) values, however, this requires an additional (expensive) component. As the aim is to measure the relative changes in load, absolute values are not necessary. It was therefore decided not to perform calibration.
The output of the sensor mat is digitized values (0-255), which will be referred to as colorvalues. The name was chosen because the pressure distribution is often displayed as a colored surface plot where 0 represents dark blue and 255 red.
The total bodyweight of the neonates is in the range of 500-5000 g, and head to toe length 25-40 cm, which roughly corresponds to 3 - 15 g/sensel (g/cm2) if the neonates would be placed directly
onto the sensor mat. In reality the mattress is distributing the mass over a larger area meaning this value is expected to be lower. This means that the measurements are conducted in the lower end of the sensitivity of the sensor mat, as can be seen from the measurement ranges of both sensor mats listed in Table 3.1. Therefore, preliminary tests were performed to check whether the sensor mat would be suitable for the clinical trial (Section 3.3 and Appendix E).
Table 3.1: Measurement range of the sensor mats
Sensor mat Measurement range (specs) Output range At the highest sensitivity 5315 34 - 207 kPa 350 - 2100 g/sensel 0-255 1 bit = 0.35-0.50 g 5350N 21-303 kPa 215 - 3100 g/sensel 0-255 1 bit = 0.15-0.25 g
CHAPTER 3. THE CLINICAL TRIAL: PREPARATION
(a) The sensormat type 5315. (b) The sensormat type 5350N.
(c) Scheme of type 5315. (d) Scheme of type 5350N.
(e) The sensormat handle.
Figure 3.3: The BPMS system: two types of sensormats and the handle that performs the data acquisition. The sensor mat dimensions are 5315 / 5350N: total length and width 622.3x529.8mm / 579.9x533.9mm, thickness 0.33mm tab length 130.9mm / 325.1mm, matrix width and height 487.7x426.7mm / 439.9x480.1mm, column and row width 6.35mm / 6.0mm and spacing 10.16mm / 10mm. In total there are 42x48 / 44x48 rows and columns resulting in a total of 2016 / 2112 sensels [BPMS User
3.3
Preliminary tests
Before starting the clinical trial, the sensor system was evaluated to verify that the system was suited to detect the position and respiration of a neonate. In Appendix E these (and more) tests are described in detail.
To test whether the sensor mat was able to register the position of a neonate, two dummy neonate models were created1. The pressure distributions generated by these dummies matched the pres-sure distribution generated by a neonate.
The dummies are shown in Figure 3.4: the smallest one weighs 500g and the largest one 1300g.
Figure 3.4: Two phantom babies were created for this study. They are linen shapes filled with flower. Left 1300g, right 500g. The small bellows shown in the left can be inserted into the dummy next to it.
Tests were performed by placing the dummies on top of two different mattresses (a white incubator mattress and the mattress as used in the incubator of the MMC (Figure 3.5, both from GE)), on top of both sensor mats.
Figure 3.5: The matresses used for testing: MMC (green) and white mattress. Top and side view. The MMC mattress consists of three layers: memory foam, cold foam, memory foam. (In Dutch memory foam
= traagschuim.)
Both sensor mats picked up the dummies location, and as expected the more sensitive 5350N registered a (slightly) larger area and higher pressure values than the 5315. Since both sensor mats detected the 500g dummy they appear well suited for position detection.
The mattress as used in the MMC, is likely very comfortable, however, these test already indicated it distributes the pressure so well that it makes it more difficult to detect the neonate, compared to the white mattress.
In order to detect the respiration signal, the sensor mat should be able to at least pick up varia-tions with frequencies ranging from 0 to around 3 Hz, which corresponds to the expected domain of the respiration frequency.
1The dummies were created by cutting two body shaped sections out of tightly woven linen and sewing these
CHAPTER 3. THE CLINICAL TRIAL: PREPARATION
First the response of the sensormat (only 5315) to a step change in load was evaluated. For that purpose, a set-up was created, which uniformly loaded the entire sensor mat using a bag of water. The pressure in the waterbag could be varied by changing the water level in the in- and outlet. During these measurements, three different loads were applied by varying the water height in the in- and outlet.
The first thing that was noticed in these measurements is the large spread in individual sensel values despite the uniform load. The average value indeed increased when the load increased, however, the spread of the values was almost as large as the measurement range.
The sensor mat has a very fast initial response to the step changes in load. This initial response should be sufficient to detect respiration. However, as it takes very long (»1s) to reach its final value, some amplitude loss will occur when measuring respiration.
Though the average response to a change in load was regular and repeatable, the behaviour of individual sensels did not exhibit the same pattern. Thus not only the initial value, but also the response to a change in pressure differs from sensel to sensel. Due to the complexity it is too time consuming to characterise all sensels individually.
To demonstrate that the sensor mat was indeed capable of detecting the respiration signal some dynamic tests were performed. These are described in Appendix E.
Neonatal respiration occurs with rates of around 10-120 breaths per minute [Lindberg et al., 1992; Engoren et al., 2009]. The sensor mat was capable of detecting signals with frequencies within this range. The amplitude of the signal, which could be interesting for tidal volumes, is unreliable due to hysteresis and the averaging algorithm of the sensor mat.
If the amplitude of the respiration signal is sufficient to detect it is unknown at this stage. It is very difficult to estimate the variation in load that respiration will cause, therefore clinical tests need to be performed to verify this.
3.4
The clinical trial
During the clinical trial the position, movement and respiration signal of (premature) neonates were measured and recorded during 24h . All these measurements were conducted at the NICU of the MMC Veldhoven.
Before measurements could be started, permission from the Internal Committee Biomedical Ex-periments (ICBE) from Philips and the Medical and Ethical Committee of the MMC was needed. They review whether an experiment is safe, useful and ethical. Safety is not only concerning avoid-ance of physical harm, but also ensuring the privacy of a test subject is respected. All documents handed into the ICBE are included in Appendix J.
Permission was obtained to perform clinical tests on (max) 14 neonates. The neonates participat-ing in this trial were selected by the clinical investigator, Prof.dr. S. Bambang Oetomo, based on their weight and, more importantly, health status. Their parents were quite often in or residing close to the hospital which allowed for a personal conversation and explanation of the measurement before they signed the permission form (which can be seen in Appendix J).
The fact that permission was obtained for testing on 14 patients does not mean that all these tests will be performed. The trial will start with (relatively) heavy neonates and if measurements are successful patients of lower weight will be selected. In the NICU heavier patients tend to be healthier patients, which is why we start with that patient group.
When a suitable candidate for measurements was available the set-up was be moved to the hospital and everything was installed such that a measurement could be started either during routine care or when the neonate was out of the incubator for "kangaroo care" (Moment when the parents hold
their baby on their chest, this skin to skin contact helps with the bonding). This was done as to not disturb the neonate during the placement of the sensor mat.
Sometimes, the set-up needed to be removed, for example in case an Xray photo needed to be taken. Therefore, all nurses received a short verbal instruction, and a manual and quick reference card were left next to the set-up.
In total measurements were performed on 7 neonates. In between measurements, the set-up was adjusted a few times to improve the measurement. The workflow is shown in Figure 3.6.
Permission • ICBE • METC • Contract
Start clinical trial • Patient selection • First measurement
Patient 1
• 5315 underneath mattress • Data analysis: low
amplitude respiration signal
Patient 2
• 5315 on top of mattress
• Data analysis: higher amplitudes, but unstable surface area
Patient 3
• 5315 underneath and on top of mattress • Best results yet,
relatively heavy patient • Sensitivity sensor? Update measurement system • Include more sensitive sensor mat (Tekscan type 5350N)
• Continue trial
Patient 4 and 5
• 5350N underneath mattress • Data analysis: low
amplitude, patient 5 no respiration
Adapt sensor mat • Placed knobs on top of sensels 5350N Patient 6 and 7 • 5350N with knobs underneath mattress • Improved respiration detection, not the signal amplitude
CHAPTER 3. THE CLINICAL TRIAL: PREPARATION
Initial measurements were performed with sensormat type 5315. However, analysis demonstrated a very low amplitude of the respiration signal. It was initially attempted to improve this ampli-tude by placing the sensor mat on top of the mattress. Though this indeed resulted in a bigger amplitude, other problems were introduced such as creasing which resulted in an unstable mea-surement surface area underneath the neonate. Meamea-surements were continued using a new more sensitive sensor mat (type 5350N). This improved the measurement underneath the mattress, but it was still impossible to obtain the respiration signal for the lightest patient. It is suspected that the mattress is doing such a good job at pressure distribution that there is no respiration signal left at the bottom of the mattress for the light weighted neonates. A final attempt was made to improve the performance of the sensormat by sticking small knobs onto the individual sensels, thus creating an higher initial load on each sensel and a flat loading surface (uniform load on top of individual sensels), and ensure all pressure is now distributed over the sensels (nothing on the plastic area in between sensels).
To summarize, four cases were evaluated during the trial: sensormat underneath mattress (5315, and 5350N), sensormat on top of matress (5315), and sensormat with knobs underneath matress (5350N).
The set-up was placed such that it would not hinder access to the incubator, while still being accessible for both nurses and researcher. During the 24h measurements the webcam (placed on the back of the incubator, underneath the cover) documented the inside of the incubator (position neonate, recorded at 1 Hz), the reference respiration was recorded (porti, 512 Hz), and the pressure distribution was measured and recorded with the sensor mat (65 Hz). An example of the dataset obtained during a measurement is shown in Figure 3.7.
After a 24h measurement this data was encrypted and transported from the MMC to Philips Research where it was analysed.
1
Time [s]
Impedance signal [μV]
Pressure Mat Webcam Reference
Figure 3.7: Example of the data recorded during the 24h measurement: pressure distribution, photo, and reference respiration signal.
The data analysis served two purposes: finding the position of the neonate over time and recov-ering a respiration signal in the pressure data.
Before the data obtained during the clinical trial is analysed, it is checked for disturbances and divided into categories, based on the neonates position and movement.
The full 24h measurements were viewed at high speed, using the data player developed for this study (Appendix D.4). This was done to check all webcam footage for the presence of faces (those incidental frames were removed for privacy reasons), note the moments when a nurse or parent was caring for the neonate, and note the posture of the neonate (e.g., back or belly). The posture of the neonate with respect to the mattress influences the resulting pressure distribution and it is interesting to see if there are differences in the respiration pattern depending on the posture. Besides the webcam data, the reference data was also checked for suitability. Sometimes the impedance signal is distorted or even absent e.g., due to motion (on average 15% of the time) or loose electrodes (incidental). During these periods it cannot be used for verification, therefore these periods were excluded during the respiration analysis. The methods used to analyse the reference are described in Appendix F.
4.1
Position analysis
Blob detection was used to find and select the area underneath the neonate. Blob detection finds groups of pixels that are connected to each other in order to identify objects / regions of interest. Besides giving information on the position of the neonate, this will decrease the complexity and necessary computation time of further analysis. Because only sensels that were underneath the neonate are selected and the unloaded sensels were discarded. A more extensive description of the blob detection method can be found in Appendix F.3. Figure 4.1 shows the steps from a frame with the pressure distribution, to the surface of the blob, and its resulting coordinate list. All surface plots are with respect to the location of the sensel on the sensor mat (which is either 42x48 or 44x48).
Blob detection was performed on a frame at 10 second intervals. This resulted in a nice overview of the changes in pixel coordinates over time, which of course corresponds to the neonates movement. The area information was then used to create a surface plot of all sensels, which shows the percent-age of time that sensel was loaded. All segments in which the neonate was not in the incubator were excluded. An example of the percentage of time a sensel is loaded is shown in Figure 4.2. The actual data analysis is only performed on another sub selection of sensels, namely the ones that are loaded more than 60% of the time.
Furthermore, in the end this list of coordinates is combined with sensels that contained the res-piration signal to demonstrate at which locations a sensor should be placed in the design of the respiration sensor system.
CHAPTER 4. THE CLINICAL TRIAL: DATA ANALYSIS 0 10 20 30 40 50 0 5 10 15 20 25 30 35 40 45 50 0 50 100 150 200 250
(a) Surface plot of pressure distribution (patient 1, sensormat underneath mattress) (colorbar shows the raw colorvalues)
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45 50
(b) Blob detection patient 1. This is the sur-face that was selected through blob detection
x y 20 13 21 13 26 13 27 13 33 13 15 14 16 14 17 14 18 14 : : 35 33 36 33 14 34 (c) Example of the corresponding list of sensel coordi-nates.
Figure 4.1: Illustration of how the coordinates of the sensels that are compressed by the neonate are detected. 0 10 20 30 40 50 0 10 20 30 40 50 0 20 40 60 80 100
(a) Surface plot before correction
0 10 20 30 40 50 0 10 20 30 40 50 0 20 40 60 80 100
(b) Surface plot after correction
Figure 4.2: Surface plot of the % of time the sensels are loaded during a 24h experiment. This example is from the second patient measurement. It clearly demonstrates the need for correction that the patient
was out of the incubator and for the period the sensel partially malfunctioned.
4.2
Respiration analysis
Respiration detection is performed to investigate the performance of the pressure mat compared to the reference, to determine when respiration is not detected, and to determine the locations where respiration was most successfully detected.
The method vital signs monitors use to determine the respiration rate is to count the amount of breaths and determine their period. This was done using peak detection. An example of peak detection performed on the reference signal is shown in Figure 4.3. In peak detection a maximum is identified in between two minima. The breathing rate (bpm, breaths per minute) is derived from the period between those identified peaks. It is important this rate is determined for every peak, because then cessation of breathing is detected earlier. However, the breathing rate shown on the patient monitor is an average of the last 8 breaths. The red circle in Figure 4.3 shows a section where the peak detection algorithm misses peaks during movement. Therefore, the fragments with motion are therefore discarded as explained in more detail in Appendix F.2.
930 935 940 945 950 −3 −2 −1 0 1 2 3x 10 5 Time [s]
Impedance signal [microvolt] Reference signal
Detected peak
Figure 4.3: An example of a processed reference signal. The peaks have been identified and labeled. As can be seen in the section marked with the circle when there is motion the algorithm does not identify all
peaks.
Before this method can be applied to the signals in the pressure mat the amount of signals to be analysed needs to be reduced. On average, there are 200-600 sensels (in the blob) that can potentially contain the respiration signal. Ideally, only one communal (respiration) signal will be analysed using peak detection and then compared to the respiration rate obtained from the reference.
Several methods to detect a respiration signal in the pressure distribution have been evaluated, such as:
• Center of mass movement: Breathing is associated with the movement of the diaphragm, unfortunately this motion could not be detected when looking at a shift in the center of mass. • Linear summation: Summation of all signals of the sensels that were on the same line with respect to the neonates shoulders. It was assumed that those signals will be in the same breathing phase. The summation helps to further reduce the amount of signals that have to be analysed. The downside of this method is that the line over which is summated is essential. Due to the support objects in the incubator that merged with the pressure distribution resulting from the neonate it was not possible to determine the neonates head to toe and shoulder axes most of the time. Therefore this method was also abandoned. • Frequency analysis: By taking the fourier transform of the signals the frequency
com-ponents of the ’respiration’ signal were determined. Though this method showed promising results on test data sets, it was unsuccessful on the clinical data. The main problem was that it is difficult to pinpoint which frequency corresponded to respiration. For example: the reference is strongly affected by the pumping of the heart, resulting in detection of the heart rate is in stead of the respiration rate.
In Appendix G.2 these methods are described in more detail and examples are provided.
None of these methods proved very successful at detecting the respiration signal. Therefore, an-other alternative was attempted: namely principal component analysis (PCA). PCA is often used to reduce the amount of variables in data, which is in this case really helpful as there are often 600+ signals to be analysed.
CHAPTER 4. THE CLINICAL TRIAL: DATA ANALYSIS
In source separation the original signal is derived without knowing how it was mixed into the measured signal. (Blind) source separation methods rely on statistical properties of the data to distinguish signals from different sources. For the purpose of this study, we want to find the res-piration movement among all other movements registered by the sensor mat. The derivation of the principal components is described in more detail in Appendix G.2.
In principal component analysis the eigenvectors and -values of all the signals are derived to and used to derive new (artificial) signals (so called principal components) that account for most of the variance in the recorded signals. The variance in the signal is the deviation from its average value.
In this case, a principal component is a linear weighted combination of all sensel signals. The weight assigned to a signal results from the eigen analysis. The amount of principal components is equal to the number of original signals. However, only the first few principal components account for a meaningful amount of variation in the signal, therefore only those will be examined. The first principal component accounts for the largest variance in the recorded sensel signals, and the second principal component for the 2nd largest variance and so on. Furthermore, all principal components are uncorrelated to each other.
As the respiratory movement is continuous it is expected to account for most of the variation in the sensel signals (as motion is more incidental). Therefore PCA should be able to detect the respiration signal.
The exact source signal is not found through these methods. They are able to reconstruct the original waveform, but the original scale, component order and any delays are not taken into ac-count.
The scale is lost because the principal component is a combination of the original signals, the exact (original) amplitude of the signal is unknown. The first component may not be respiration as the motion of the neonate can have a larger amplitude (and thus variance) than the respiratory motion of the neonate. The same signal with a delay will be recognized as different components (a sine and cosine are in essence the same signal, but with a delay, PCA will recover them as to separate signals, as can be seen in Figure 4.4).
Therefore, additional processing will be needed to determine which of the resulting components contains the respiration signal [Choi et al., 2005; Vullings et al., 2010; Bishop, 2006].
0 10 20 30 −2 0 2 4 6 8 Time sin(t) 2*sin(t)+5 0.5*cos(t)
(a) Surface plot before correction
0 10 20 30 −1.5 −1 −0.5 0 0.5 1 1.5 Time
Pricipal component value
PC1 PC2 PC3
(b) Surface plot after correction
Figure 4.4: Example of the result of principal component analysis performed on three periodic signals. The first two components are the result of the combination of the sines and cosine and the third component
In summary the process of determining the respiration rate involves the following steps:
• The signals coming from pressure sensel located underneath the baby are selected through blob detection.
• These selected signals are then bandpass filtered to remove noise (bandpass: butterworth 0.1-10Hz, 2nd order).
• The amount of pressure signals is further reduced to one ’respiration’ signal through principal component analysis.
• The respiration rate present in the principal component is determined through peak detec-tion.
Then in order to compare the detected respiration signal to the reference signal the following steps are taken:
• The respiration rate of the reference signal is also determined through peak detection. • The detected respiration rate is successfully detected if it is within a 5% (good) to 10%
(average) deviation of the reference. Note: the periods of motion are excluded from analysis. • When the respiration was successfully detected it is then determined which sensels con-tributed most to the principal component by examining the corresponding value of each sensel in the eigenvector.
Chapter 5
Results
During the clinical trial the position and movement of seven (premature) neonates have been recording during a 24h measurement. These recordings were analysed to determine where in the incubator the neonates are mostly residing and underneath which part of the body the respiratory movement was strongest. A complete overview of the results can be found in Appendix H.
5.1
Position detection
This research has shown that the neonates predominantly stay in one place: the middle of the incubator. Though the neonates move a lot none of them have moved away from the center of the incubator, as is demonstrated in Figure 5.1. As you can see the edges of the ’ellipse’ in the middle of the sensor mat are changing, this is due to motion of the neonate (arms and legs) and because nurses regularly change the posture of the neonate (e.g., belly to side).
The fact that the neonates are always placed exactly in the middle of the incubator is related to the incubators access locations for the nurses.
These results show a very promising location for the placement of the sensors: in the middle of the incubator. Furthermore, as it is the desire to have a patient support system that monitors vital signs and in itself is a comfortable support structure (which aids in the positioning), this data could also give valuable insight into the locations of support structures. This would require a more detailed analysis based on the posture of the neonates (side, back, belly).
0 10 20 30 40 50 0 10 20 30 40 50 0 20 40 60 80 100
(a) 5315 underneath mattress: the mattress created a large area over which the pressure was distributed.
0 10 20 30 40 50 0 10 20 30 40 50 0 20 40 60 80 100
(b) 5315 on top of mattress: the area of pressure is smaller as the sensor mat was on top of the mattress, and the body of the neonates is more recognizable.
0 10 20 30 40 50 0 10 20 30 40 50 0 20 40 60 80 100
(c) 5350N underneath mattress (with and without knobs): again a large area over which the pressure was distributed. 0 10 20 30 40 50 0 10 20 30 40 50 0 20 40 60 80 100
(d) Combination of all measurements with the 5315 and 5350N (the bottom and top row of the 5350N were ex-cluded).
Figure 5.1: Surface plot of the percentage of measurement time a sensor was loaded (combination of all patient measurements). The head of the patients is at the bottom of the figures. The colorbar represents the percentage. The sizes differ depending on which sensor mat was used during the measurement (either
CHAPTER 5. THE CLINICAL TRIAL: RESULTS
5.2
Respiration analysis
Respiration analysis was performed on the sensels that were underneath a neonate more than 60% of the time. This was determined per patient.
The results from the respiration detection prove that it is possible to measure respiration using the pressure sensitive sensor mat. An example of the analysis of the respiration signal of patients 3 and 7 is shown in Figures 5.2 and 5.3. For patient 7 the respiration rate was successfully detected (within 5% deviation) most times, Figure 5.3c, however this is only corresponds to 20% of the time (within 10% deviation is 45% of the time). However, for patient 3 the respiration rate was only suc-cessfully detected (within 5% deviation) 5% of the time and 20% of the time within 10% deviation. The calculated respiration rates are often too low or too high when compared to the reference, meaning the the detection of respiration is far from the 100% coverage.
The reason for this deviation is twofold:
Firstly, the detected reference respiration rate can be incorrect. For example, when the heart rate distorts the impedance signal, as can be seen in Figure 5.2, or when the reference is distorted by motion, Figures 5.4a to d.
Secondly, the sensor mat occasionally fails to detect the respiration signal (not sensitive enough?), as shown in Figure 5.4e and f, or also fails to detect respiration due to motion, as shown in Fig-ure 5.4c and d.
A source of concern is the minimal signal amplitude that was encountered during this study (and resulting quantization errors). An example of a raw pressure signal is shown in Figure 5.3e. All these measurements were conducted on (relatively) heavy patients. Despite the benefit of their weight the sensor mat’s results are still disappointing; the amplitude of the respiration signal is very low. It is therefore expected that the issues with the signal amplitude will only increase when expanding the patient population to include light weight patients.
Another source of concern is the high occurence of the 50 bpm rate in the derivated respiration rate of the sensor mat (visible when examining all patients measurements, Appendix H). A possibility is that another periodic signal coming from a part of the incubator is causing harmonic variations that are registered when the neonates movement (incl respiratory movement) is minimal. To verify this a measurement in an incubator without neonate (though still a fixed weight on the sensor mat) should be conducted.
16 17 18 19 20 −1 −0.5 0 0.5 1 x 10 5 Time [s]
Reference signals [microvolt]
ECG Respiration
(a) Reference respiration signal and ECG signal. The red squares mark the individual breaths. On the surface of a respiration smaller peaks can be seen that result from a heart beat.
28 29 30 31 −1 −0.5 0 0.5 1 x 10 5 Time [s]
Reference signals [microvolt]
Respiration ECG
(b) Reference ’respiration signal and ECG signal. The respiratory component is completely absent, only the heart beat can be recognized.
0 20 40 60 80 100 120 0 2000 4000 6000 8000 10000 Respiration rate [bpm] Occurence
(c) Histogram of the calculated respiration rates in the principal component. 0 50 100 150 0 2000 4000 6000 8000 10000 Respiration rate [bpm] Occurence
(d) Histogram of the calculated respiration rates in the reference signal. The spread in respiration rate is large, this is most likely caused by detection of the heart rate in stead of respiration rate.
−50 −40 −30 −20 −100 0 10 20 30 40 50 2000 4000 6000 8000 10000
Deviation from reference [%]
Occurence
(e) Histogram of the percentage the pressure respiration rate differs from the reference respiration rate. The rate of the sensor mat is the majority of the time lower than the reference rate.
0 1000 2000 3000 0 50 100 150 200 Time [s] Respiration rate [bpm] Reference Sensor Mat
(f) Derived respiration rate of the sensor mat and ref-erence signal. The respiration rate detected from the reference is actually the heart rate, which is why it is higher the majority of the time.
Figure 5.2: Respiration results Patient 3 (2200g), sensormat 5315 underneath mattress for 20h. Overall the respiration rate derived from the sensor mat is too low compared to the reference, mainly due to the
CHAPTER 5. THE CLINICAL TRIAL: RESULTS 0 20 40 60 80 100 120 0 500 1000 1500 2000 2500 3000 Respiration rate [bpm] Occurence
(a) Histogram of the calculated respiration rates in the principal component. This distribution appears to cor-respond quite well to the reference (b).
0 20 40 60 80 100 120 0 500 1000 1500 2000 2500 3000 Respiration rate [bpm] Occurence
(b) Histogram of the calculated respiration rates in the reference signal. −50 −40 −30 −20 −100 0 10 20 30 40 50 500 1000 1500 2000 2500 3000
Deviation from reference [%]
Occurence
(c) Histogram of the percentage the pressure respiration rate differs from the reference respiration rate.
0 1000 2000 3000 0 50 100 150 200 Time [s] Respiration rate [bpm] Sensor Mat Reference
(d) Derived respiration rate of the sensor mat and ref-erence signal, they clearly exhibit the same trend.
70 71 72 73 74 75 −500 0 500 Time [s] Signal [−] Principal component Reference
(e) Fragment of the 1st principal component of one recording recording and the matching reference signal.
84 86 88 90 92 94 200 210 220 230 Time [s]
Output pressure sensel Sensel [17,14]
Reference respiration (scaled)
(f) Fragment of a raw pressure signal of one recording recording and the reference signal, showing the quanti-zation problems
Figure 5.3: Respiration results Patient 7 (1900g), sensormat 5350N with knobs underneath mattress for 4h (Measurement stopped due to decrease in patients health). The derived respiration rates seem to correspond quite well. Though the rate derived from the sensor mat tends to be too high. In this case the
0 50 100 150 200 250 0 20 40 60 80 100 Time [s] Respiration rate [bpm] Sensor Mat Reference
(a) Fragment of the derived respiration rates. There is a good correspondence in rates at time 50-100 and 150-200s. 45 50 55 60 −1500 −1000 −500 0 500 1000 Time [s] Respiration signal [−] Principal component Reference
(b) Respiration signals, at t=50s a motion is causing a disturbance to the reference, causing a deviation be-tween both signals.
16400 1660 1680 1700 1720 1740 50 100 150 200 Time [s] Respiration rate [bpm] Sensor Mat Reference
(c) Fragment of the derived respiration rates, where the rates do not correspond well.
1620 1640 1660 1680 1700 1720 −1 −0.5 0 0.5 1 x 104 Time [s] Respiration signal [−]
(d) Respiration signals. The deviation in respiration
rates is likely caused by a motion artefact disturbing both signals. 1750 1760 1770 1780 30 40 50 60 70 Time [s] Respiration rate [bpm] Sensor Mat Reference
(e) Another fragment of the derived respiration rates, where the rates do not correspond well.
1758 1760 1762 −2000 −1000 0 1000 2000 Time [s] Respiration signal [−] Principal component Reference
(f) Respiration signals. This time the reference is func-tioning properly, but no respiration signal is detected in the principal component.
Figure 5.4: Respiration rate detection for Patient 7 (1900g), sensormat 5350N with knobs underneath mattress for 4h (Measurement stopped due to decrease in patients health). Illustration of the difficulties encountered when comparing the reference respiration rate to the respiration rate derived from the pressure
CHAPTER 5. THE CLINICAL TRIAL: RESULTS
5.2.1
Location of the respiration signal
The respiration signal was strongest near the abdomen of the neonates (independent of the posture of the neonate: back, belly or side) as is shown in Figure 5.5. Furthermore, when examining the results for individual neonates in more detail, it becomes evident that the head of the neonate ap-pears to be moving in opposite phase to the neonates abdomen (Figure 5.6). This could be a very interesting feature to use in the detection of the respiration signal. As examining the difference between the signals gives a respiration signal with a larger amplitude, and when the movement of the head an abdomen is not correlated it could be motion.
(a) Patient 1 (5315 underneath mattress) (b) Patient 3 (5315 underneath and on top of
mattress)
(c) Patient 4 (5350N underneath mattress) (d) Patient 6 (5350N underneath mattress)
Figure 5.5: Location of the main contributors of the principal component (during the entire measure-ments). Colorbar represents total (absolute) contribution. The head of the neonate is at the top and the
area shown as zero is the total area that was used for the analysis.
These results only demonstrate the first principal component, while in many cases the next two to three components also contain the respiration signal (different phase and lower amplitude). In the future, it would be interesting to also analyse these components for respiration. Additionaly, when a component does not contain respiration it might provide information on the neonates movement (such as feet movement). Smart combinations of the principal components could result in a more reliable respiration signal (free of motion artifacts). Therefore, analysis techniques to optimally combine the principal components need to be investigated further.
The analysis of the reference has demonstrated that there are frequently periods of motion. After examining all 24h recordings the conclusion is that neonates move very often. Usually only small movements (arm or leg), but there were children who were able to turn themselves from their side onto their belly. Fortunately, most of these motion periods last less than 10-15s, which is the alarm limit for apnea detection. The patient monitor must be able to alert a nurse when the neonate suffers from apnea, thus a measurement gap of >15s is unacceptable. The respiration detection method must therefore also be functioning during movement.
this stage if the periods where only the heart rate is detected, are in fact periods of apnea. As we use the reference to determine the reliability and effectiveness of the new respiration moni-toring system it is necessary to further investigate the performance of the reference.
(a) Sensel contribution to the principal component dur-ing a period where the neonate remained in the same position.
(b) Reference photo: the neonate was ly-ing on his back.
(c) Sensel contribution to the principal component dur-ing a period where the neonate remained in the same position.
(d) Reference photo: the neonate was
ly-ing on his side. (left side, image
mir-rored)
(e) Sensel contribution to the principal component dur-ing a period where the neonate remained in the same position.
(f) Reference photo: the neonate was ly-ing on his abdomen.
Figure 5.6: More detailed location of respiration: Patient 6 (1900g), sensormat 5350N with knobs underneath the mattress.
Chapter 6
Design specifications
6.1
Design of a neonatal respiration monitoring system
The aim of the E-Nemo respiration project is to develop a system using sensors embedded in the the patient support system that detects the respiration signal.
To gain an understanding of the requirements of this system and the general workflow at a NICU a clinical trail was conducted.
List of requirements
The results of that clinical trial have led to the following list of design requirements for the neonatal respiration monitoring system (Table 6.1).Appendix I contains a more elaborate description of the requirements and possible used cases for the prototype of the neonatal respiration monitoring system.
Table 6.1: Design requirements of the neonatal respiration monitoring system
Functional requirements:
Measurement range: 0-180 breaths per minute (detect apnea)
Resolution: minimally 0.1 g/cm2 or around 1 mm deformation Operational range: 0-25 g/cm2 and 0 - 2 cm deformation
(Neonates of 500 g up to 5 kg, length varying from 25- 50 cm) Accuracy: ? minimal of 5% or 10 breaths per minute
Sensitivity: Detect individual breaths Operating time: Continuous
Response time: <1s
Size: Appropriate for incubators and cribs Sensor requirements:
Location: Embedded into the patient support system (e.g., mattress) Invasiveness: Non-invasive, no direct skin contact
Measurement type: Respiration movement
Environmental requirements: Operating temperature 5-40oC
Operating humidity 30 - 99.9%
Compliance and safety requirements:
The neonatal respiration monitoring system is a class IIb medical device and as such requires a declaration of conformity (the manufacturer’s declaration of device safety and effectiveness), before it can be used for testing. It therefore needs to adhere to several safety regulations, which are listed in Appendix I.
