Toward Non-invasive Neonatal Gas Monitoring with Plasma-based Spectroscopy

(1)

TVE-F 18002

Examensarbete 15 hp Juni 2018

Toward non-invasive neonatal

gas monitoring with plasma-based spectroscopy

Hampus Fröjdholm Hedwig Haas

Carmen Lee

(2)

Teknisk- naturvetenskaplig fakultet UTH-enheten

Besöksadress:

Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0 Postadress:

Box 536 751 21 Uppsala Telefon:

018 – 471 30 03 Telefax:

018 – 471 30 00 Hemsida:

http://www.teknat.uu.se/student

Abstract

Toward Non-invasive Neonatal Gas Monitoring with Plasma-based Spectroscopy

Hampus Fröjdholm, Hedwig Haas, Carmen Lee

Transcutaneous gas monitoring of oxygen and carbon dioxide is an important method for monitoring the oxygenation and ventilation in prematurely born infants. The amount of gas diffusing through the skin is very small and current technology rely on heating of the skin to increase the gas amount, where transcutaneous oxygen requires the highest temperature. Heating damages the delicate skin of the infants and is a major concern among the nurses administering the treatment.

In this thesis we have investigated a prototype and developed components for a novel transcutaneous gas monitor capable of performing transcutaneous carbon dioxide measurements on adults without any external heating with a fast start-up time compared to conventional monitors. The technology is based on a microplasma source developed at the Microsystems Technology department at Uppsala university.

The thesis has focused on producing three key components of the prototype. Firstly the controller board and radio frequency (RF) amplifier, previously two separate circuit boards, have been combined into one. The design also incorporates improvements regarding the power supply of the board, where a buck converter instead of a linear regulator is used to step down most of the voltage. This eliminates the need to use a heat sink to remove heat generated during voltage transformation. Secondly a sensor pad has been developed from a silicone-based material which is soft, flexible and self-adhesive, allowing it to be placed anywhere on the patient without tape or glue. And lastly techniques have been developed to efficiently turn the raw data collected from the plasma source into a usable carbon dioxide concentration signal.

Though the plasma-based transcutaneous gas monitoring is yet far from clinical trials, the technology has shown promising results and it is deemed to be a viable alternative with better performance in patient comfort, sensitivity and response time.

Ämnesgranskare: Natalia Ferraz

Handledare: Anders Persson, Greger Thornell

(3)

Populärvetenskaplig sammanfattning av projektet

Projektet baseras på en innovation inom transkutan gasmätning som utvecklats på institutionen för Mikrosystemteknik på Uppsala universitet. Mätningen utförs med hjälp av en plasmakälla. Utöver plas- makällan behövs ett silikonplåster för att samla upp gas, samt en tunn kapillär för att föra gas från plåstret till plasman. I projektet används denna princip för att vidareutveckla mätningen av partial- trycket av koldioxid i blodet på vuxna, för att i framtiden kunna applicera principen på för tidigt födda barn. Fördelen med denna prototyp är att man undviker att värma upp huden, som är nödvändigt med dagens teknologi, och därigenom kan förlänga användningstiden utan risk för att skada patienten.

Projektet har bestått av tre moment för att modifiera och utveckla prototypen: sammanfoga två kretskort till ett, förbättra designen på silikonplåstret och förfina analysen av data. Den tidigare elektroniken var uppdelad på två separata kretskort. De båda kretsarna har förts samman och blivit eﬀektivare genom användandet av en buck converter utöver en linjärregulator. Denna stegar ner spänningen innan linjärregulatorn för att minska den eﬀekt som går förlorad och den värme som genereras. Mätningarna av gaser har visualiserats i diagram via ett Matlab-skript, för att kunna tyda vilka intensiteter som motsvarar respektive gas. Detta delmoment i projektet fokuserade på den slutliga användaren och dennes behov, något som stärktes via ett studiebesök på Akademiska sjukhuset. Därefter har ett nytt skript tagits fram för att ge en visuell representation av hur koldioxidhalten förändras över tid. Silikonplåstret har tillverkats med hjälp av ett kiselgummi, Dimetylpolysiloxan (PDMS), och även modifierats enligt ett recept framställt på Uppsala universitet. Plåstret har under projektet formgivits, testats och utvärderats med hjälp av olika recept och tekniker för att uppnå en tillförlitlig modell som är användbar för vidare mätningar.

Kretskortet sammanfogades men hann ej testas på grund av leveransproblem. Detta innebar att kretskortet inte kunde testas tillsammans med övrig utrustningen eftersom komponenterna ej hann lödas fast.

Plåstret gav efter många iterationer ett resultat som kunde användas för att genomföra transkutana mätningar av koldioxidhalten hos en testperson med hjälp av den signalbehandling som utförts. För vidareutveckling av detta projekt behövs ett mer utvecklat plåster som kan produceras på ett eﬀektivt sätt och säkerställer riskfri, långvarig applicering på huden. Dataanalysen har utförts i Matlab och för fortsatt utveckling är ett öppet program önskvärt för en ökad lättillgänglighet. Plasmaprototypen som vidareutvecklats är ett framtida alternativ för att genomföra transkutan mätning av blodgaser och kan täcka ett efterfrågat behov hos dess användare.

(4)

Acknowledgement

First of all we would like to express our gratitude to our supervisors Anders Persson and Greger Thornell for their guidance and valuable input throughout the project. Secondly, we would like to thank Natalia Ferraz, our project mentor. Her eﬀort in managing the stage-gates was a great help in keeping us on track. The last but not the least we are grateful to all the people around us for their generosity in sharing their knowledge and lending their help. Here to name a few:

• Martin Berglund, for giving us guidance and feedback when designing the new circuit board.

Our design was largely based on the previous work done by Martin.

• Anette Johansson, Kerstin Segelström & Jan Rutkowski the nurses and engineer at Akademiska sjukhuset, for helping us identify the end-user needs and demonstrate the current technology used at the neonatal intensive care unit.

• Lena Klintberg, for showing us the clean room and relevant equipment useful for our project.

• MakerSpace, for being available for 3D printing and the belonging material in the library at Ångström.

• Peter Sturesson, for helping us to mill negatives in the milling machine to the pad mould in the cleanroom.

• Pierre Sandin, for helping us with the PDMS production in the lab by showing us the equipment and the production process.

• Ragnar Seton, for helping us with the script communicating Matlab and the spectrometer in the lab. The script used was largely based on Ragnars previous work.

• Rickard Viik, for helping us with the S3-PDMS production.

• Erika Åkerfeldt, for helping us to produce the master to the pad mould in the cleanroom.

Without their contribution, this thesis would not have been possible.

(5)

Glossary

hyperoxia Hyperoxia occurs when cells and organs are exposed to excess supply of oxygen. 1 hypoxia Hypoxia occurs when cells and organs suﬀer from deficiency of oxygen. 1

PCB Printed circuit board. 5, 6, 7, 8, 9, 12, 13, 14, 15 pCO₂ Partial pressure of carbon dioxide. 16, 18 PDMS Polydimethylsiloxane. 4, 7, 8, 9, 13, 14, 15, 18 PEIE Ethoxylated polyethylenimine. 4, 7, 9, 13, 14, 18 pO2 Partial pressure of oxygen. 1, 3, 18

RF Radio frequency. 3, 4, 5, 6, 12, 13 RFPA Radio frequency power amplifier. 6

S3-PDMS Soft, stretchable and sticky polydimethylsiloxane-based elastomer. 4, 7, 9, 13, 14, 18 SSRR Stripline split-ring resonator. 1, 3

TCM Transcutaneous gas monitoring, measuring the local oxygen and carbon dioxide released through the skin from the capillaries. 1, 2, 3, 16, 18

TcpCO2 Transcutaneous partial pressure of carbon dioxide. 1, 2, 3, 5, 11, 16, 26, 27 TcpO2 Transcutaneous partial pressure of oxygen. 2

(8)

1 Introduction

1.1 Background

It is estimated that in 2010, 11.1% of all live births were born prematurely according to a 2012 report by WHO.[1] Out of the 15 million neonates, 1 million died within one month due to a variety of complications and infants born prematurely need constant monitoring to ensure their proper development. One of the challenges is oxygenation and the adequacy of ventilation. Blood oxygen levels can change rapidly due to the underdeveloped organs of the infant. Excessive levels of oxygen in the tissues (hyperoxia) can cause visual impairment or even blindness while deficiency of oxygen in the tissues (hypoxia) can lead to brain damage (encephalopathy).[2][3] Oxygenation and ventilation are closely linked, where both are important parameters to monitor in relation to neonatal care.

There are three common monitoring techniques today: arterial blood sampling, pulse oximetry and transcutaneous gas monitoring (TCM), each facing its own diﬃculties. Though arterial blood sampling is the most direct and accurate method, the frequency of taking arterial blood sample is limited both by the low blood volume of the infants as well as the delicate nature of the infant skin. The underdeveloped skin of neonates is extremely thin, easily damaged and unable to provide a barrier to minimize fluid and electrolyte losses, protect against infection and prevent absorption of toxic substances and regulate the internal thermal environment.[4][5] Frequently repeated arterial punctures can result in critical loss of blood and dermal infection.

Pulse oximetry monitors a person’s peripheral oxygen saturation (SpO2). It has become popular in recent years as a monitoring method in clinical settings due to the non-invasiveness and the ease of use compared to its non-invasive counterparts.[6] However, a study suggests that the move towards pulse oximetry from transcutaneous pO2 monitoring has resulted in infants being cared for at a lower oxygen tension than before and the method has a higher variability than other non-invasive monitoring techniques.[7][8] It is also shown that its results are negatively aﬀected by the patient’s movement, improper probe placement and possibly unreliable when the patient suﬀers from severe hypoxemia (deficiency of blood oxygen).[9]

Transcutaneous means entering, passing or penetrating the skin. The transcutaneous gas monitoring measures the amount of oxygen or carbon dioxide diffusing through the skin, as an indirect indicator of arterial partial pressure of oxygen (PaO2) or carbon dioxide (PaCO2). The commercially successful methods have faced resistance from the practitioners due to its potential of causing harm to the patients and the difficult procedure of attaching and storing the sensors.[6] Due to the heating requirement, the sensors need to re-sited after a certain time interval. However, a neonate has a very small surface area where the monitor sensors can be attached.[10] As a result, nurses are often reluctant to administer these methods on a frequent basis. In short, though it is crucial to monitor oxygenation and ventilation in neonates, all of the existing methods come with major trade-offs.

1.2 Objectives

A group of researchers at the Division of Microsystems Technology, Uppsala University, have developed a microplasma source, called a stripline split-ring resonator (SSRR), which can be used to measure small quantities of gases.[11] The SSRR has a gas inlet feeding a small amount of gas to the plasma cavity placed at the centre of the resonator. Through the window on the sides of the plasma cavity, the light emitted from the plasma is picked up by the spectrometer, which can be used to determine the composition of the gas with a high degree of accuracy (for detailed information, please refer to the article cited here).[11] The device was initially designed to search for signs of extraterrestrial life on Mars but later found its potential in the medical device industry.

Prior to our project, a proof of concept setup was assembled to measure transcutaneous partial pressure of carbon dioxide (TcpCO2) on adults. This result points to the possibilities of using it on infant patients, since the skin of premature infants is thinner, which increases the diﬀusion and thus the detectable signal.

(9)

This project produced and refined the components for a TCM prototype utilizing a SSRR to measure TcpCO2. We also utilized a special type of self-adhesive polymer developed at the same department to make a sensor pad that will stick to the patient’s skin without the use of glue or tape. Because of the sensitivity of the SSRR, oxygen and carbon dioxide diﬀusing through the skin can be measured without heating, which, paired with the adhesive sensor pad, would allow the device to operate for longer periods of time without causing harm to the patient.

1.3 Transcutaneous Gas Monitoring

As mentioned above TCM focuses on the measuring of blood gases diffusing through the skin. Current technology is based on an amperometric measurement. The electrode on the sensor consists of a platinum cathode and a silver reference anode covered with a membrane permeable to oxygen and an adhesive fixation ring, which forms a sealed cell containing an electrolyte solution. To receive a reliable reading, the gases have to first saturate in the electrolyte solution in contact with the skin.[12] In order to increase the amount of transcutaneous gases, the skin is heated to 42 C–45 C, depending on the type of gas being measured, to get an adequate signal.[13] In some cases, for the monitoring of TcpCO2, the temperature can be lowered to 37 C.[14] Heating melts the crystalline structure of the stratum corneaum (the outer layer of the skin), which makes it less effective a barrier to oxygen diffusion. The oxygen saturates the solution and diffuses through the membrane before being reduced at the cathode. The process generates an electric signal which is converted into the partial pressure measurements of oxygen on the surface[9].

The sensors are calibrated using a known gas mixture[6]. TcpCO2 and TcpO2 are important parameters for monitoring infant ventilation and oxygenation respectively.[6]

1.3.1 The Existing Industrial Standards

The largest brands of the conventional TCM device are Radiometer, Perimed and SenTec, all of which use the sensor electrode design screwed on an adhesive fixation ring. See Table 1 for a brief comparison of a few of their models.

Table 1: TCM on the market.

Brand HQ Model Primary use

Radiometer Denmark TCM5 special care, neonatal care

Perimed Sweden PeriFlux 6000 wound healing, amputation, oxygen therapy SenTec Switzerland SDMS neonatal care, anesthesiology, sleep diagnostics

In general, the existing devices have a touch screen interface with several (up to 8) channels for TcpO2. Some models are portable and others can be purchased as modules integrated into the hospitals’ data handling systems. The monitors have step-by-step manuals with instructions for the user. Measurements can be printed, exported as PDF or shown directly on screen. Radiometer and SenTec have a monitor which controls the heating temperature of the sensor in order to avoid burns or damage to the skin.

The monitors can allow up to 12 hours of measurements without making a new calibration according to manufacturers’ specification[15]. This also includes a 30 minute disconnection if transport or examination of the neonatal patient is needed. However, in practice calibrations can be required as frequent as every 90 minutes and need to be done when moving the sensor based on the interview with hospital staﬀ (see below).

1.3.2 Problems with the Current Design

During the interview with two nurses at the neonatal intensive care unit (NICU) and an engineer at the Uppsala University Hospital, a need for development of new TCM techniques was clearly expressed. The

(10)

nurses pointed out the fact that the current technology is based on research from the 1980s with minor improvements. A major concern was the apparent resistance from manufacturers to listen to feedback about the technology, especially in relation to transcutaneous measurements in neonatal intensive care.

In the traditional design, the monitor relies on the electrode to provide electrical signal for analyzing the gas content. The electrode sensor is snapped into a rigid fixation ring glued to the skin. This design is to ensure a tight seal for the containment of the electrolyte solution. Gases must saturate in the solution before the sensor can provide reliable reading. The response time for the existing devices needs to cover 15 min of heating to the skin before being ready to sample, which results in a non-optimal way of measuring, since it cannot provide real time results. Because the neonates have such delicate skin, the nurses sometimes need to put a less adhesive ring under the fixation ring to provide a more gentle adhesion. From our interview, we learned that over the years, the manufacturers have increased the adhesion of the ring instead of decreasing it, despite of the end-user feedback/complaints.

All sensors also need to be regulated regarding temperature and time in order to avoid burns on the sensitive skin and require to be re-sited at regular intervals of less than 90 min and re-calibrated before each monitoring sequence.[6] The procedure is diﬃcult to perform without inflicting pain on the infant patients. Moreover, a layer of silver chloride builds up on the surface of the electrode after a few uses.[13]

This will require a "re-membraning" every week or after each use on a patient with infectious disease.

In practice, some hospitals change the membrane between patients regardless of the use condition.

The aforementioned drawbacks are confirmed by our contacts at the hospital. At the Uppsala University Hospital TCM is used in the NICU on all infants with respiratory assistance which are almost all neonates.

Measurements are done 6 to 8 times per day up to several months depending on the status of the infant.

An exception is made for the smallest infants who has arterial catheters for drawing blood. For TcpCO2

a temperature of 42 C was required and for pO2 44 C is needed. A measurement takes at least 15 minutes, but preferably 30 minutes, to heat up the area before measuring. Because of the burns caused by the probe and the small surface area of the infants, it is diﬃcult to find a suitable spot for the device.

The high sensitivity of the plasma-source-based design of TCM can be used to detect signals from nanogram samples[16]. Unlike the conventional TCM sensor design, which requires a flat circular area of 2 cm–4 cm in diameter, the cell needed for the plasma-source-based design is less than 1 cm by 1 cm and can be placed on a curved surface. The use of a plasma source compared to electrodes contributes to a faster response time and requires less gas. Finally, the existing technique requires at least 15 min of heating before a valid measurement while the plasma-source-based design delivers useful data almost instantly with no demand for heating at all.

1.4 Peripheral Devices for the SSRR-based TCM

The SSRR-based TCM requires a few essential components. The ones we worked on and produced were:

1) a controller board and a RF amplifier which ignites and drives the plasma, 2) a sensor pad with a capillary that collects gas sample and feeds to the plasma source.

1.4.1 Controller Board and RF Amplifier

The current circuit boards were created to allow an Arduino to control and gather data from the plasma source. Currently it is designed as an Arduino shield containing all measurement and control hardware together with an add-on chip with the main amplifier and monitoring circuits.

The Arduino shield is divided into three sections. One section controls the DC bias of a probe inserted into the plasma; another creates the RF waves driving the plasma; and the third collects data (data acquisition). A star grounding scheme is used to electrically isolate the individual parts and reduce noise.

(11)

The second circuit board contains the main RF amplifier and monitoring circuits. Because of the high frequency and current the traces for the amplifier needs to be impedance matched, meaning that the trace widths and lengths as well as their overall shape must be precisely controlled to obtain the same electrical characteristics throughout the entire circuit. By doing so, maximum eﬃciency of the circuit is achieved. The monitoring circuits measures the power going into the amplifier and out to the SSRR and also the reflected portion of the wave at the interface between the chip and the plasma source.

1.4.2 S3-PDMS for the Sensor Pad

Polydimethylsiloxane (PDMS) is the most widely used organic silicon-based polymer within areas of microfluidic chips, stretchable electronics and bio-medical patches. To form elastomers-polymer with viscoelasticity, PDMS has to be cross-linked with itself. Cross-linking is a bonding process in polymers in which their chains are connected thereby changing the physical properties of the material. [17, p.37]

To be able to connect a PDMS microsystem to the skin, it has to be stretchy and sticky. This can be achieved using PDMS-based elastomers, with a small amount of the amine-based polymer, ethoxylated polyethylenimine (PEIE). The resulting product is called S3-PDMS where S3 stands for the special properties of this modified version of PDMS: soft, stretchy and sticky.[18]

To make S3-PDMS, a curing agent is needed together with a silicone base. Then PEIE is added to achieve the adhesive effect. PEIE changes the cross-linking structure and its amines form new strong complexes together with the platinum curing catalyst which is used for PDMS. This gives a structure containing both high and low density cross-linking (heterogeneous cross-linking) compared to the homogeneous cross-linking in ordinary PDMS. The PDMS is often made softer by changing the ratio of silicon base and cross-linker, but adding PEIE has been shown to increase all three investigated properties mentioned above. The research made by Jeong et al. [18] showed that the more PEIE added, the more sticky the sample became. The S3-PDMS was up to 10 times more adhesive to the skin than the original PDMS when using 1:10 ratio for base and cross-linker. The authors also investigated the effects of S3-PDMS on the human skin, during which test subjects carried the patches made from the material for 12 days with no observable negatives effects.[18]

The components should therefore be added according to a certain recipe for the desired adhesive result.

The base, curing agent and PEIE are all poured in a plastic cup and stirred with a glass rod for 2-3 minutes, left to de-gas in the freezer before cured at 90 C for about 2-3 hours.[18]

1.5 Specific Aims

We aim to produce and test the three major elements for the microplasma-based TCM: the driving circuit which provides control and the radio frequency (RF) wave to ignite the plasma source; the dermal sensor pad for transcutaneous gas sampling; and the software for data analysis.

The driving circuit consisted of two separate boards: 1) The controller that collects data and provides the power; 2) the RF amplifier which ignites and maintains the plasma. Major concerns are the heat generation of the power supply and interference from high-frequency components. The existing driving circuit uses a linear voltage regulator to step down the 12 V coming from the power supply to the required 5 V. The extra power is dissipated as heat which is cooled with a heat sink. An improvement was made by adding a buck converter, that drops the voltage down to 7 V, before the linear regulator to achieve the same operating properties without a lot of heat generation. Much of the components, especially on the RF amplifier board, are to handle signals of high frequency (higher than 2.4 GHz). As a result, the placement of the components and traces can aﬀect the signals to a great extent. Special care was taken to avoid "cross-talk" and other types of interference while extending the controller board to include the RF amplifier.

The sensor pad has been remodeled and manufactured with various recipes and processes. A few tech-

(12)

niques that work or will work with minor modification are selected and presented in the Results section.

In the current setting, the gas content is measured and the emission spectrum is plotted using a Matlab script. Since the script is written for a diﬀerent application (finding extraterrestrial life), the data representation shows the intensity of the spectral lines instead of the gas partial pressure we were interested in. In the project, we have looked at the requirements of the end users, e.g., connectors used, sampling frequency, and critical levels. Based on the literature and an interview with hospital staﬀ at the Uppsala University Hospital, we have produced a prototype script that processes the raw spectrum and generates a stable reading and a visual representation showing the CO2 partial pressure variation over time.

2 Methods

2.1 Experimental Setup

Figure 1 shows the schematic of the lab setup. The plasma source is controlled by the controller board and the Arduino. During the actual experiment, the RF/controller board was substituted by a large RF wave generator and an electronic box since we did not receive the PCB in time. This setup produces the same plasma properties as the PCB, but is more expensive and takes up more space. The gas collected from the sensor pad enters the plasma source through the capillary and the gas inlet; the vacuum pump connected to the other valve ensures a constant gas flow. Through the window, the spectrometer analyze the emitted light and sends a spectrum to the computer to be further analyzed.

spectrometer

plasma source

controller/RF-ampli er & Arduino

sensor pad gas in

signal out

s

signal out RF in

gas out to vacuum pump

valve

Figure 1: Schematic of the experimental setup.

For calibration, the sensor pad was removed and the capillary was inserted into a 100 mL syringe with CO2 mixed with air at diﬀerent concentrations. The integration time of the spectrometer was set to 0.1 s(to avoid clipping of the peaks at higher concentration) and concentrations of 2 %, 5 %, 10 %, 15 % and 20 % CO2 were used to find the relation between intensity of the spectrum and the concentration.

A total of 60 s of data was gathered at each fixed concentration, which was then averaged and used as a calibration point.

For the transcutaneous measurement the integration time was set to 0.5 s, 5 times higher than in the calibration. Peak height is linearly increasing with integration time and the resulting concentrations were therefore scaled down by a factor of 5. In the beginning of every measurement the pad was left in air for 30 s before attaching it to the skin. This initial data was used to remove the background noise by subtracting the median of the first 70 samples from the entire data set. The pad was removed after 500 sand left in air for the remainder of the experiment. Each experiment was run for a total of 600 s.

(13)

In a second experiment the sensor pad was attached to the forearm of a healthy adult. In order to further increase the TcpCO2 signal, the adult tried to hold breath for approximately one minute twice during the experiment.

2.2 Circuit Board Design

The circuit board was designed in the open source software package KiCAD and largely based on the existing work done at the department. The blueprint of the existing RF amplifier was used as the specification for extending the controller board to incorporate the RF amplification.

KiCAD contains programs for creating electronic schematics and printed circuit board (PCB) layouts.

It also has programs for creating custom schematic symbols and PCB footprints. All files generated by KiCAD are plain text in a human readable format. There are a number of limitations in KiCAD, when creating high power, high frequency designs such as the RF amplifier of this work. To overcome this, a combination of external tools, custom scripts and workarounds were used to get a PCB layout as similar as possible to the existing board. By keeping the design close to the functional board, the risks of making mistakes and the testing needed is minimized.

The central part of the amplifier board is the RFPA (radio frequency power amplifier). It uses traces of varying width as part of the design. These need to match the shape specified on the blueprint. The traces were drawn in Inkscape, a vector graphics program, and imported into KiCAD as a custom silk screen graphic through the bitmap-to-component utility. This layer was then manually changed in the component text file to the front copper layer. The trace was divided into a left and right part, where two versions were created, one normal and one mirrored, for increased flexibility when creating the layout.

A corresponding component in the schematic was created to connect other components to the trace.

Care had to be taken when drawing the traces as far away as possible from the "hot" side of the amplifier (after the transistor). Signal traces close to the amplifier were drawn on the back copper plane of the board when possible.

To create curved traces, as required by the specification, a custom Python script was written to generate arcs of straight trace segments. The generated traces were then copied into the PCB layout text file and the positions changed to match the start and end points of connecting traces.

The RF signal traces require a larger clearance between the trace and the surrounding copper fill than the rest of the circuit in order to isolate the high-frequencies from the rest of the board. KiCAD can increase the clearance on a net-by-net basis, but this has no eﬀect on the custom-made large traces. The workaround employed was to have a separate copper fill around the custom traces with a larger global clearance.

Calculations of trace impedance were done without taking into account the solder mask. Because of this the specification, a removal of the solder mask over the RF signal traces was required. This can be achieved by having a fill on the front solder mask layer. Since the solder mask layer is a "negative" layer, any mask in a filled zone will be removed. The area around the traces was marked by tracing the edges of the copper fill. The ground plane was also removed under the capacitors used in the amplifier to remove any stray capacitance between the capacitor and ground plane that could skew the total capacitance value.

Finally, to minimize disturbances, the copper layers had to be connected through vias. This procedure is called via stitching. A custom-made via footprint was created and its net was manually set to GND.

This component was then copied and placed with 50 mil spacing as evenly as possible across the board.

Around the RF signal traces, a denser stitching was used to shield high-frequency components.

In high frequency design traces have to be impedance matched, in this case to 50 ⌦ which is a common value for RF applications. Trace impedance is dependant on placement, e.g inner or outer copper layers,

(14)

and PCB material. Using the PCB Calculator track widths for so called microstrip lines were calculated using the PCB material specification from OshPark, the PCB manufacturer. Electronic components were ordered from a major online retailer.

2.3 Sensor Pad Design

The sensor pad is a critical part of the transcutaneous gas monitoring since it is in direct contact with the neonates and has been said to cause discomfort and potential harm when using existing designs on the market. In this project, the sensor pads were made by molding S3-PDMS and PDMS in custom-made molds. The molds consist of three basic components: a flat base, a rim to contain and shape the pad, and negative cavities to create a sampling cell. The diﬀerent techniques to manufacture such pads are documented in the following sections.

2.3.1 Technique I

OpenSCAD was used to create the 3D models for the outer rims of the pads, which were then 3D printed on a Creality CR-10 3D printer. The material used was polylactic acid (PLA), a type of plastic. The PLA was coated with epoxy to increase smoothness and to seal holes. The rim was then glued onto a piece of acrylic glass to get a smooth bottom surface. One concern with using PLA was its low glass transition temperature at 60 C where the plastic becomes soft and pliable. Tests with epoxy-coated PLA showed that structural integrity could be maintained at temperatures as high as 90 C.

Inkscape was used to create the patterns of the negative cavities for the sampling cells. The svg-file was exported dxf-format to be loaded on a PCB plotter for milling out the patterns in the chosen material.

The negative cavities were planned to be made from dicing tape with three diﬀerent free area densities (80 %, 40 % and 30 %) and three diﬀerent heights (1 layers–3 layers of tape). Milling and drilling in the dicing tape proved to be challenging and instead one pattern was drilled into polyimide and hand cut into shape. The diameter of area for the negative cavities was 1 cm and the thickness in the range of a 100 µm. To attach the pattern of the negative to the base of the mold, spray glue was used as an adhesive.

A number of diﬀerent designs were made for the pad. All designs had a central sampling cell with straps going out for mechanical support, much like a wristwatch. The size was exaggerated to fit an adult and to accommodate several testing areas for easier evaluation of the diﬀerent negative designs. The wristwatch was also made in smaller sizes. A schematic of the mold is seen in Figure 2, where the pattern of the negative cavities, 3D printed rim and the acrylic glass have been attached to each other, before pouring in PDMS.

The S3-PDMS was mixed according to Table 2 and left to de-gas in a freezer at 20 C over night. It was then poured into the mold carefully to avoid air bubbles and then cured in the oven at 90 C for 2-3 hours depending on the thickness of the mixture in the mold.

Table 2: Recipe for S3-PDMS using Sylgard ratio 1:10.

Curing agent Base PEIE

1 g 10 g 40µL

(15)

acrylic glass

epoxy coated PLA rim polyimide negative

Figure 2: Schematic of the epoxy coated PLA rim glued to acrylic glass. The negative in the centre was made from polyimide, glued to the acrylic glass with spray glue.

2.3.2 Technique II

For the second attempt we used 30 µL of PEIE substituting the 40 µL in the original recipe as described in Table 2 to decrease the degree of stickiness. The thickness of the PDMS was reduced significant from 2 mmto approximately 0.4 mm to decrease the curing time. One of the designed molds, the diamond- shaped, was used as well as two petri dishes, of which one of them had a negative design attached with spray glue at the bottom. The S3-PDMS was poured from the plastic cup used for mixing and the molds were left to de-gas in the freezer for 1-2 hours.

2.3.3 Technique III

A new kind of negative pattern was investigated. Instead of having a separate base and negative cavities, we chose to use a piece of PCB board and milled the negative cavities into it, about 100 µm. This technique avoided epoxy and spray glue, since they both were suspected to aﬀect the quality of the resulting PDMS.

microscope slide PDMS

sampling cell negative PCB

Figure 3: Schematic of the sensor pad mold. A piece of blank PCB was used as the base for drilling the holes. Microscope slides were glued using Sylgard PDMS to the sides for framing.

After the holes were milled, the cavities were cleaned and the edges were polished with sandpaper. For constructing the rim, microscope slides of glass were attached to the PCB with normal PDMS to act as glue and cured in the oven. Figure 3 shows the schematic of the sensor pad mold. Tape was used

(16)

on top of the microscope slides to strengthen the mold. The diameter of the cavities were 400 µm and the distance between the neighbouring holes was 800 µm or 1 mm from centre to centre depending on pattern. The microscope slides are commonly available in a chemistry lab and the benefit of using them to construct the mold was the easy control of the pad thickness in comparison to the visual inspection deployed previously.

The process of making the pad included: 1) The area of the cavities was covered with a small drop of normal PDMS as seen at Step 1 in Figure 4 to better register the patterns of the cavities. (An optional Step 2 can be deployed to increase the thickness of the centre part). 2) After the drop of PDMS was nearly cured, the edges were cut with a sharp knife. 3) The rest of the mold was filled with the sticky PDMS as seen in Figure 4. To investigate the most appropriate adhesive property for our application, S3-PDMS with 20, 30 and 40 µL of PEIE were tested.

(1) (2)

(3) (4)

PCB board cavities glass

rst layer PDMS second layer PDMS S3-PDMS Figure 4: Process of sensor pad production using Technique III. PDMS is used for the negative surrounded with S3-PDMS to get a self adhesive pad.

An optional second coat of PDMS can be employed to further increase the thickness of the sampling area before adding the S3-PDMS.

2.3.4 Technique IV

Given the diﬃculties of molding the pattern of the cavities onto the PDMS, (the pillars were too fine and nearly impossible to peel oﬀ without ripping apart) a new negative pattern was designed with a channelled structure that would provide better support when removing the pad from the mold. A preliminary experiment with the pattern made with a cutter plotter turned fruitful.

The pattern marked as e in Figure 5 was made on a sheet of tape using a Graphtec Craft Robo Pro cutter. It was scaled down to approximately 75 % from the drawing given the resolution allowance of the cutter. The patterned tapes was then transferred onto a large piece of glass used as the base for the mold. Microscope slides were used to frame the mold, similar to the one in Figure 3. The same method as shown in Figure 4 was employed.

2.3.5 Technique V

To provide a robust and rigid support for the sampling cell, another pattern, d in Figure 5, was milled onto a piece of PCB and cut out. This sampling cell was used as a backup to the flexible PDMS sensor pad.

(17)

1 cm

(a) (b) (c)

(d) (e)

Figure 5: The patterns of the cavities used to make the pad molds. Depending on the size and accuracy of the tools available, the resulting patterns diﬀer slightly from this schematic.

2.3.6 Attaching the Capillary

The glass capillary, which was needed for extracting gas from the sensor pad to the gas inlet of the plasma source, was attached to the pad with epoxy glue. The steps are shown in Figure 6: 1) Press the capillary through the center of the pad and apply a droplet of epoxy resin near the end of the capillary to secure it and to seal the opening into the sampling cell. Allow the epoxy to cure for a few minutes.

2) Pull back the capillary to sink it into the cavity of the sampling cell so it does not stand out from the pad. 3) Apply a second and larger droplet of epoxy on the top side to lock the capillary on the pad. 4) Clip the tip that protrudes from the surface on the pillar side.

The capillary used had an inner/outer diameter 75/150 µm. While testing the rigid sampling structure (Technique V), a capillary of the size 40/115 µm was used instead.

(1) (2)

(3) (4)

capillary pad

epoxy

Figure 6: Procedure of attaching the capillary (size not to scale). Epoxy is not applied to the tip of the capillary to avoid capillary forces drawing glue into it. When the epoxy has cured the capillary can be cut. A large amount of epoxy is applied to the top of the pad to provide mechanical support.

(18)

2.4 Signal Processing

2.4.1 Spectral Lines to Concentration Levels

The data we received from the emission spectroscopy records the normalized intensity against the wave length in the range of 350 nm to 700 nm. To identify relevant peaks, samples were taken at CO2 concentrations of 100 %, 50 %, 25 %, 12.5 % and 6.25 % in air. As the concentration of CO2 decreased, so did the height of the peaks at the wave lengths associated with CO2. A similar series was done with N2

and CO2 peaks at locations with low N2noise were selected for further analysis. In Figure 7 the typical spectra of CO2 and N2 are shown. Three peaks of wavelengths 483 nm, 519 nm and 560 nm (Figure 8) were used to determine the CO2concentration.

A preliminary analysis showed a linear relation between intensity and concentration at low levels. A more thorough analysis was later made at lower concentrations as mentioned in Experimental Setup (Section 2.1) and a linear fit of the calibration points was constructed. This was used as a model for the conversion between peak intensity and concentration.

350 400 450 500 550 600 650 700

0 0.2 0.4 0.6

Wavelength [nm]

In te ns it y

CO2

N2

Figure 7: CO2and N2spectrum. Used to identify CO2peaks that can be used to determine concentration.

478 480 482 484 0

2 4 6

·10 ²

Wavelength [nm]

Intensity

516 518 520 0

2 4 6

·10 ²

Wavelength [nm]

556 558 560 562 0

2 4 6

·10 ²

Wavelength [nm]

100 % 25 % 12.5 % 6.25 %

Figure 8: Relevant CO2 peaks. Peak height decreases with decreasing CO2

concentration.

(19)

482 483 484 0

0.2 0.4 0.6

Wavelength [nm]

Intensity

516 518 520 522 0

0.2 0.4 0.6

Wavelength [nm]

555 560 565 0

0.2 0.4 0.6

Wavelength [nm]

Original data Isolated peak Gaussian fit

Figure 9: Removing of background noise and fitting of peaks. The background noise was removed by using using fitting routines created with Matlab Curve Fitting Toolbox and Gaussian curves were then fitted to the peak to get the peak height.

To remove background noise from other peaks fitting routines were created using Matlab Curve Fitting Toolbox. The relevant peaks were excluded from the data and smoothing splines or Gaussian curves were used to approximate surrounding peaks. The background was then subtracted from the spectrum to yield the isolated peaks. To compensate for drift in peak location the isolated peaks were fitted with Gaussian curves and the height of that was used as the peak intensity (Figure 9). For more information on the exact fitting routines used and how the data was transformed into a TcpCO2 signal, refer to Appendix A.

2.4.2 Further Processing of Concentration Signal

The initial data processing with background-noise cancellation and linear fitting to determine the concentration levels had no eﬀects on the random noise generated by stochastic processes. To further remove noise additional processing was performed. First the concentrations from the 3 peaks were averaged to receive a combined CO2 concentration. Then a combination of a median filter of order 10 and a moving average filter of order 19 was used to smooth the signal This produced a smooth signal that still follows the overall trend of the original data. The chosen filtering methods cause a delay of the measurement by about 5 s since the moving average filter requires 9 data points ahead of the current one (at an integration time of 0.5 s). Details of the filter selection can be found in Appendix B.

3 Results and Discussion

3.1 Circuit Board

The circuit board combined the two separate boards that were used before: the controller board and the RF amplifier (see Figure 10 for the combined board before soldering). By combining the boards the performance and reliability of the setup should increase. The circuits were kept, to a large extent, similar to their initial designs.

The Arduino have a limited number of analog pins (the pins by the dashed line in Figure 10). Some of these pins were previously used as digital pins due to their proximity to the surrounding chips. These had to be moved to the digital pins on the other side of the board to accommodate the power sensors.

(20)

Figure 10: Render of the front side of the PCB by the manufacturer. The exposed RF amplifier traces are clearly visible on the lower part of the board.

The planes of the board were stitched together with vias. The amplifier part of the board is much sparser to avoid noise from the "hot" side of the amplifier.

The star ground pattern, previously used in the control circuit to remove noise, was removed in favour of one continuous ground plane due to technical diﬃculties. The RF amplifier (lower part of the PCB with the exposed copper traces), as an extension to the control circuit, has gone through most changes. The majority of components were re-positioned and placed around the amplifier traces with an emphasis on keeping components and traces away from the "hot" side of the amplifier (the left side of the amplifier trace), drawing traces on the back plane if possible. This led to the RF amplifier side being much sparser than the controller board side.

A buck converter was added to step down the voltage before the voltage regulator to 7 V. This addition is expected to decrease the heat generation substantially, removing the need for a heat sink on the linear regulator. A larger power supply was bought that will connect barrel jack on the new circuit board. The power supply will provide the 20 V needed by the RF amplifier.

The final design was converted to the Gerber format and ordered from a major PCB manufacture.

Unfortunately the delivery had a remarkable delay which forced us to abort the plan of soldering and testing the board. Therefore, all data for our analysis was gathered from the existing setup. Still, the work done is useful for further work with the prototype because of the improvement in heat reduction and the convenience of handling one integrated circuit board.

3.2 Sensor Pad

The best molds tested for the pad were the ones using the PCB or glass as base with microscope slides (Techniques III and IV). By using this technique we were able to omit the gluing step and the PDMS was easy to control in terms of thickness and curing. Together with the glass slides the best results were achieved when the cavities were milled into the PCB base as well as when the negative pattern was cut out on sheet tape and adhered to the bottom, seen in Figure 5. The PCB milled negatives were shaped as a hexagons with holes and lines, a-d, while the tape negative was shaped as a square with lines, e.

The pad was at first designed according to the recipe for S3-PDMS described in Table 2. For the first attempt, three batches of the recipe in Table 2 were made and divided into the two diﬀerent molds seen in Figure 11. The aim was to get around 2 mm of thickness. The curing time had to be adjusted since the S3-PDMS was still viscous after of 2-3 hours so it was left overnight to cure. The layer became thicker than the what was mentioned in the literature[18] and therefore took longer time to cure than expected.

(21)

Air bubbles in the PDMS were also found though the visual inspection showed no bubbles in the fluid prior to curing. Probably, there were tiny bubbles residing at the interface between the fluid and the substrate, which expanded upon heating and were trapped in the solidifying substance. The resulting pads were very sticky to the point where they were impossible to retrieve from the mold. Since the 40 µL of PEIE is on the verge of prohibiting in cross-linking entirely, a minuscule amount of additional PEIE can ruin the curing of S3-PDMS. The method was therefore modified (Technique II) and repeated with diﬀerent amounts of PEIE, diﬀerences in stickiness as well as curing times.

Figure 11: Molds for making the pad in S3-PDMS with the pattern of the negative cavities placed in the center. The length is about 20 cm and height 1 cm.

The level of stickiness using Technique I and II as seen in Figure 12 indicated that the PDMS was not properly cured. We later found out that the epoxy which we used to seal the PLA molds was likely the culprit as it potentially interfered with the cross-linking. A dark brown discoloration is another sign of substance interaction. The pad was hard to handle without damaging it. This lead to Technique III, where we used glass slides to get an even and thin layer of PDMS as well as being able to avoid the gluing process by milling or using tape as the negatives. According to the tested pad designs, it seemed that PCB base does not interfere with the PDMS and to use PCB for the base of the mold could be a good way of proceeding whereas the 3D printed model is more diﬃcult in matter of cutting out the pad and interference of components. We also realized that using the glass slides simplifies the laboratory process as it allowed for easy construction of several molds on a single piece of PCB or glass.

Figure 12: Failed attempt of the pad molding (Technique I, II). The substance was very sticky and diﬃcult to separate from the substrate.

Technique III was used for the investigation of diﬀerent amounts of PEIE. The 20 µL of PEIE was found to be the most suitable, 30 and 40 µL were to sticky and did not stay as one piece when removing it from the mold. Since the first PDMS attempts were made with S3-PDMS poured in the mold, the

(22)

modification with PDMS for the negative turned out to be more successful since it gave a more defined shape. The sticky PDMS was almost impossible to keep in the desired shape.

The milled PCB circular cavities a and b, seen in Figure 5 turned out to be hard to keep clean since PDMS got stuck in the holes, which also resulted in the expected pillars breaking at the roots when removing the pad from the mold. A modified pattern c as seen in Figure 5 with lines shaped as a hexagon was therefore made. Figure 13 compares the results from the two diﬀerent designs. In Figure 14 the resulting sensor pad with pattern c can be seen. This attempt turned out to be the best sensor pad regarding the negative, since it had the most smooth and clean surface on the sampling cell.

Figure 13: Comparison of the cavity designs. Left: the design with lines (Technique IV, negative c) has little residual. Right: the design with circular cavities (Technique III, negative a) is filled with residual PDMS.

Figure 14: Pad with PDMS according to Technique IV with a close-up on the negative c, the surface of which is not entirely smooth.

The negative design with lines named d in Figure 5 belongs to Technique V, in which the PCB cutout alone acts as the sampling cell with a capillary attached to it. It is then taped to the skin. This technique faced problem with air pressure, which in turn aﬀected the plasma source. A potential improvement could be to use a large capillary to improve the air flow.

3.3 Signal Processing

The linear fit of the diﬀerent CO2 peaks investigated showed that concentration is linear with respect to peak intensity with R²-values of 0.998, 0.996 and 0.996 for the 483 nm, 519 nm and 560 nm peak respectively (Figure 15). Only the slope of the linear fits was used to determine the CO2concentration during transcutaneous measurements. This is due to varying bias between diﬀerent measurements.

To compensate for this the data was shifted using the first 70 samples as described in the method.

The slopes for the peaks were (with 95 % confidence intervals): 483 nm: 2.79(36), 519 nm: 2.60(36) and 560 nm: 3.67(56). As can be seen in Figure 15 all the investigated peaks had a slightly diﬀerent relationship between intensity and concentration. However when used as a model during transcutaneous measurements concentrations estimated from all peaks were very similar. This allowed the concentrations to be easily combined. By averaging the 3 peaks’ contribution at each time step, the impact of the

(23)

0 1 2 3 4 5 6 7 8 9 10 11

·10 ² 0

10 20

R²= 0.996 R²= 0.998

R²= 0.996

Intensity

Concentration[%]

CO2483 nmpeak CO2519 nmpeak CO2560 nmpeak Linear fit (483 nm) Linear fit (519 nm) Linear fit (560 nm)

Figure 15: Linear fits of the CO2 peaks at constant concentrations. Due to varying bias between transcutaneous measurements only the slope was used to determine CO2concentration and bias was instead removed on a per measurement basis. The slopes for the peaks were (with 95 % confidence intervals): 483 nm: 2.79(36), 519 nm: 2.60(36) and 560 nm: 3.67(56). The linear relationship holds only for low concentrations as saturation was observed for concentrations above 50 %.

random noise is reduced. The linear model is only valid for lower concentrations as we observed that concentrations above 50 % caused saturation and a deviation from the linear relationship.

Using the linear model CO2 concentration was determined during transcutaneous measurements (Fig- ure 16). When the pad was attached to the forearm CO2 concentrations started increasing until it reached a maximum of about 1.4 % before the pad was removed. With the pad in air again the CO2

concentrations dropped down to atmospheric levels (0 %).

Experiment 1 was the control, in which the test subject was breathing normally. In experiment 2 the test subject was holding breath for about a minute twice during the experiment. Two elevations in the CO2 concentration can be seen at 250 s and 430 s. At the end of experiment 2 the capillary got clogged by sweat (at 450 s), causing the pressure to drop and leading to a spike in the measured concentration.

This is due to the varying properties of the plasma with changing pressure, thus does not represent a true change in CO2concentration. When the pressure started to decrease the pad was removed, but the eﬀects of the clogging persisted and as can be seen in Figure 16 it took a longer time for the concentration to return to atmospheric levels.

The TcpCO2 levels we measured are lower than the normal arterial pCO2 levels which is estimated to be about 5 %–5.6 %.[19] The reason could be attributed to the fact that the size of the sampling cell is too large compared with the scale of gas molecules and the sensor pad is not entirely airtight. The constant air exchange between the sampling cell and the atmosphere potentially brings down the TcpCO2

reading. It is possible that the measured value needs to be mapped (potentially scaling or biasing) into the correct range. However, it is diﬃcult to know, without simultaneous measurement of arterial pCO2

or with other equipment, whether the measurement is accurate and what transformations are needed.

What can be noted is that the signal was well above the noise levels and can be processed to provide a smooth stable reading, both on-line and oﬀ-line. After 10 minutes the CO2 concentrations had reached a fairly stable value. Compare this to the minimum 15 minutes and preferred 30 minutes start-up time of current TCM technology. This time can probably be improved with a more eﬃcient pad. The pad

(24)

0 50 100 150 200 250 300 350 400 450 500 550 600 0

1 2

Pad attached

Pad removed

Time [s]

Concentration[%]

Experiment 1 Experiment 2

Figure 16: Transcutaneous CO2 measurement on the forearm. Experiment 1 is a control measurement with normal breathing while in experiment 2 the test subject was holding breath twice during the measurement. The pad was attached to the skin after 30 s and removed for the last 100 s. In experiment 2, two elevations in the CO2concentration can be seen 250 s and 430 s. The large increase in CO2concentration at 450 s was due to the capillary getting clogged which also led to the slower descent back to atmospheric concentrations.

used for the experiment had a quite large sampling volume. With a smaller sampling volume and better adhesion to the skin start-up time and signal quality is expected to improve.

Another source of error was the changing pressure during a series of measurements. When starting the equipment and vacuum pump the pressure would quickly drop and stabilize. However during continuous operation the pressure would slowly continue to go down. This likely caused successive measurements over longer time periods to slightly drift towards higher CO2 concentrations. This was especially a problem when doing measurements where the test subject held breath. Sweat and heat being produced caused the pressure to drop more quickly. To mitigate this, measurements should be made at similar pressures. In our case experiment 1 was performed at mean pressure 78.7(28) Pa and experiment 2 at mean pressure 77.7(27) Pa.

There was also a debate whether holding the pad to the skin would improve the signal quality. On the one hand holding the pad might improve the seal and reduce the sampling volume, on the other hand it increased heat and sweat production. In the end we opted to carefully hold the pad to the skin, since the used pad quickly lost a lot of its initial stickyness.

Further improvement can also be made to the linear model. With the 100 mL syringe it was impractical to create CO2 mixtures of concentration lower than 5 % and the minimum resolution was 1 percent unit. Meanwhile the transcutaneous measurements yielded concentrations of maximally 1.4 %. With a better, more accurate way of creating low concentration CO2 mixtures, more closely spaced calibration points closer to the transcutaneous levels could be taken. The integration time could also be increased to eliminate the need for scaling the data before using the model.

(25)

4 Conclusions

In this project, we designed and manufactured some of the auxiliary components for a TCM prototype based on the microplasma technology. Unfortunately due to the logistic delay, it was not possible to assemble all the parts we had made for the actual experiment.

We observed clear signals of transcutaneous CO2 change in our data and managed to quantify the concentration levels. The response time and measurement sensitivity showed excellent results. However, before clinical trials several issues need to be addressed.

First of all, the handling of the S3-PDMS material is more difficult than we initially expected. The properties of the material are prone to change of the concentration of the additive PEIE and the curing process tends to be affected by other substances present. Though the recipe with 20 µl and PDMS cured at room temperature have provided the best results, the adhesion is still sub-optimal. Studies on its long-term effect are also necessary since it has only been briefly investigated by Jeong et al.[18]

Secondly, we saw evidence of transcutaneous O2 change but a series of oxygen and nitrogen data are required in order to infer and quantify the concentration levels. However, unlike the current technology with electrodes which demands diﬀerent optimal temperatures for measuring pO2and pCO2, the plasma- based method does not discriminate the two. It is therefore suitable for measuring both under the same condition.

Thirdly, though the PDMS is a breathable material, sweat can still accumulate on the patient’s skin which clogs the fine capillary and causes pressure to drop, subsequently kills the plasma source. This happened during one of the experiments when the test subject tried to hold breath for too long. Under the normal circumstance, this could potentially be avoided by marking the strap with holes, similar to a band-aid or use a hydrotobic coating to keep sweat from the sampling cell.

In short, we have shown evidence of the microplasma-based design being a viable and advantageous alternative to the conventional TCM; and solved some of the issues towards a functional prototype. We encourage future studies to further the development of this method.

(26)

References

1C. Howson, M. Kinney, J. Lawn, et al., “Born too soon: the global action report on preterm birth”, Geneva: World Health Organization, 1–126 (2012).

2C. Gilbert, J. Rahi, M. Eckstein, J. O’sullivan, and A. Foster, “Retinopathy of prematurity in middle- income countries”, The Lancet 350, 12–14 (1997).

3D. Azzopardi, J. Wyatt, E. Cady, D. Delpy, J. Baudin, A. Stewart, P. Hope, P. Hamilton, and E.

Reynolds, “Prognosis of newborn infants with hypoxic-ischemic brain injury assessed by phosphorus magnetic resonance spectroscopy”, Pediatric Research 25, 445 (1989).

4L. Madsen, M. Rasmussen, L. Bjerregaard, S. Nøhr, and F. Ebbesen, “Impact of blood sampling in very preterm infants”, Scandinavian journal of clinical and laboratory investigation 60, 125–132 (2000).

5M. J. Ness, D. M. Davis, and W. A. Carey, “Neonatal skin care: a concise review”, International journal of dermatology 52, 14–22 (2013).

6K. L. Sandberg, H. Brynjarsson, and O. Hjalmarson, “Transcutaneous blood gas monitoring during neonatal intensive care”, Acta Pædiatrica 100, 676–679 (2011).

7D. Quine and B. J. Stenson, “Arterial oxygen tension (pao2) values in infants< 29 weeks of gestation at currently targeted saturations”, Archives of Disease in Childhood-Fetal and Neonatal Edition 94, F51–F53 (2009).

8D. Quine and B. J. Stenson, “Does the monitoring method influence stability of oxygenation in preterm infants? a randomised crossover study of saturation versus transcutaneous monitoring”, Archives of Disease in Childhood-Fetal and Neonatal Edition 93, F347–F350 (2008).

9C. F. Poets and D. P. Southall, “Noninvasive monitoring of oxygenation in infants and children: prac- tical considerations and areas of concern”, Pediatrics 93, 737–746 (1994).

10J. N. Sharma, S. Saxena, and U. Sharma, “Standard curves for mid arm circumference and mid- arm/head circumference ratio in newborns”, The Indian Journal of Pediatrics 57, 389–393 (1990).

11A. Persson, M. Berglund, G. Thornell, G. Possnert, and M. Salehpour, “Stripline split-ring resonator with integrated optogalvanic sample cell”, Laser Physics Letters 11, 045701 (2014).

12J. Severinghaus, “The current status of transcutaneous blood gas analysis and monitoring”, Blood Gas News 7, 4–9 (1998).

13Y. Djafari and N. Abolfathi, “An inexpensive 3d printed amperometric oxygen sensor for transcutaneous oxygen monitoring”, in Biomedical engineering and 2016 1st international iranian conference on biomedical engineering (icbme), 2016 23rd iranian conference on (IEEE, 2016), pp. 281–284.

14R. D. Restrepo, K. R. Hirst, L. Wittnebel, and R. Wettstein, “Aarc clinical practice guideline: transcutaneous monitoring of carbon dioxide and oxygen: 2012”, Respiratory care 57, 1955–1962 (2012).

15SenTec, Neonatalogy, https://www.sentec.com/int/application-areas/neonatology/ (visited on 05/24/2018).

16M. Berglund, G. Thornell, and A. Persson, “Microplasma source for optogalvanic spectroscopy of nanogram samples”, Journal of Applied Physics 114, 033302 (2013).

17S. H. Jeong, Soft intelligence, Liquids matter in compliant microsystems (Uppsala Universitet, 2016).

18S. H. Jeong, S. Zhang, K. Hjort, J. Hilborn, and Z. Wu, “Pdms-based elastomer tuned soft, stretchable, and sticky for epidermal electronics”, Advanced Materials 28, 5830–5836 (2016).

19A. J. Williams, “Abc of oxygen: assessing and interpreting arterial blood gases and acid-base balance”, BMJ: British Medical Journal 317, 1213 (1998).

(27)

Appendix A Spectrum Analysis

Data from the emission spectroscopy in the form of a Matlab structure array was saved to disk and loaded for the data analysis. The structure contains all the spectra from the measurement in a matrix called Values together with arrays specifying wavelengths and time stamps. In the Values matrix the first dimension is the time and the second dimension is the wavelength. The peaks were analyzed by first finding their indices in the matrix and then using the Matlab Curve Fitting Toolbox to construct fitting routines to remove background noise and Gaussian curves to find the peak intensity. The matlab function spectrum_analysis uses these fitting routines to, each time step, perform the background noise fit and cancellation and finding the peak intensity for a number of peaks. The functions also saves a lot of data for visualization purposes.

The fitting routines created were mostly smoothing splines. Two advantages were discovered with using the smoothing spline to remove background noise. Firstly it is based on interpolating between samples and thus follows the spectrum quite well. Secondly the amount of smoothness was easily tuned by changing the smoothing parameter. This combined with careful exclusion of points belonging to the CO2peak allowed for the construction of good fitting routines.

For the 483 nm peak a second order Gaussian curve was used to remove the background noise. This peak is situated close to a large N2 peak with no other environmental noise. The Gaussian adequately modeled this sharp peak.

Matlab source code for the spectrum analysis is included below. The fitting routines were as mentioned created and auto-generated by Matlab Curve Fitting Toolbox. The spectrum_analysis function has capabilities to analyze data from both the spectrometers available in the lab, however only one of them was used in this thesis.

A.1 spectrum_analysis.m

function [ out_data , vis_data ] = spectrum_analysis ( data )

%SPECTRUM_ANALYSIS(DATA)

%

% Analyses t h e g i v e n spectrum f o r CO2 c o n t e n t

wavelength_narrow = data . Specs ( 2 ) . WavelengthDataAdjusted ; wavelength_wide = data . Specs ( 1 ) . WavelengthDataAdjusted ; intensity_narrow = data . Specs ( 2 ) . Values ;

intensity_wide = data . Specs ( 1 ) . Values ;

% Peak l o c a t i o n in data s e t x_560nm_narrow = 2 0 5 0 : 2 1 5 0 ; x_520nm_narrow = 1 7 3 0 : 1 7 9 0 ;

%x_450nm_narrow = 1140:1180;

x_483nm_narrow = 1 4 3 0 : 1 4 5 5 ; x_560nm_wide = 1 6 8 0 : 1 7 7 0 ; x_520nm_wide = 1 5 1 0 : 1 5 7 0 ;

% P r e a l l o c a t e arrays f o r speed

% A l l t h e peaks and g a u s s i a n s are saved f o r v i s u a l i s a t i o n purposes dim_narrow = size ( intensity_narrow ) ;

dim_wide = size ( intensity_wide ) ;

peak_560nm_narrow = zeros ( dim_narrow ( 1 ) , length ( x_560nm_narrow ) ) ; peak_520nm_narrow = zeros ( dim_narrow ( 1 ) , length ( x_520nm_narrow ) ) ;

%peak_450nm_narrow = z e r o s ( dim_narrow ( 1 ) , l e n g t h (x_450nm_narrow ) ) ; peak_483nm_narrow = zeros ( dim_narrow ( 1 ) , length ( x_483nm_narrow ) ) ; peak_560nm_wide = zeros ( dim_wide ( 1 ) , length (x_560nm_wide ) ) ;

(28)

peak_520nm_wide = zeros ( dim_wide ( 1 ) , length (x_520nm_wide ) ) ; gauss_560nm_wide = zeros ( dim_wide ( 1 ) , length (x_560nm_wide ) ) ; gauss_520nm_wide = zeros ( dim_wide ( 1 ) , length (x_520nm_wide ) ) ; gauss_560nm_narrow = zeros ( dim_narrow ( 1 ) , length ( x_560nm_narrow ) ) ; gauss_520nm_narrow = zeros ( dim_narrow ( 1 ) , length ( x_520nm_narrow ) ) ;

%gauss_450nm_narrow = z e r o s ( dim_narrow ( 1 ) , l e n g t h ( x_450nm_narrow ) ) ; gauss_483nm_narrow = zeros ( dim_narrow ( 1 ) , length ( x_483nm_narrow ) ) ; co2_wide = zeros ( dim_wide ( 1 ) , 2 ) ;

o_wide = zeros ( dim_wide ( 1 ) , 1 ) ; n2_o2_wide = zeros ( dim_wide ( 1 ) , 1 ) ; n2_wide = zeros ( dim_wide ( 1 ) , 1 ) ; co2_narrow = zeros ( dim_narrow ( 1 ) , 2 ) ;

% Approximate CO2 f o r every time s t e p by c r e a t i n g a f i t f o r t h e background ,

% remove t h a t background then f i t a gau ssian to i t . Use maximum o f g au ss ian

% as peak h e i g h t . for i = 1 : dim_narrow ( 1 )

bg_560nm_narrow = background_560nm_narrow ( x_560nm_narrow , . . . intensity_narrow ( i , x_560nm_narrow ) ) ;

%bg_450nm_narrow = background_450nm_narrow (x_450nm_narrow , . . .

% intensity_narrow ( i , x_450nm_narrow ) ) ;

peak_560nm_narrow ( i , : ) = intensity_narrow ( i , x_560nm_narrow ) . . . bg_560nm_narrow ( x_560nm_narrow ) ’ ;

%peak_450nm_narrow ( i , : ) = intensity_narrow ( i , x_450nm_narrow ) . . .

% bg_450nm_narrow (x_450nm_narrow ) ’ ;

gauss_560nm_fit_narrow = gaussianfit_560nm_narrow ( x_560nm_narrow , . . . peak_560nm_narrow ( i , : ) ) ;

%gauss_450nm_fit_narrow = gaussianfit_450nm_narrow (x_450nm_narrow , . . .

% peak_450nm_narrow ( i , : ) ) ;

gauss_560nm_narrow ( i , : ) = gauss_560nm_fit_narrow ( x_560nm_narrow ) ; gauss_520nm_narrow ( i , : ) = gauss_520nm_fit_narrow ( x_520nm_narrow ) ;

%gauss_450nm_narrow ( i , : ) = gauss_450nm_fit_narrow ( x_450nm_narrow ) ; gauss_483nm_narrow ( i , : ) = gauss_483nm_fit_narrow ( x_483nm_narrow ) ; co2_narrow ( i , 1) = max( gauss_560nm_narrow ( i , : ) ) ;

co2_narrow ( i , 2) = max( gauss_520nm_narrow ( i , : ) ) ; co2_narrow ( i , 3) = max( gauss_483nm_narrow ( i , : ) ) ;

%co2_narrow ( i , 4) = max( gauss_450nm_narrow ( i , : ) ) ; end

for i = 1 : dim_wide ( 1 )

Toward Non-invasive Neonatal Gas Monitoring with Plasma-based Spectroscopy

Examensarbete 15 hp Juni 2018

Toward non-invasive neonatal

gas monitoring with plasma-based spectroscopy

Hampus Fröjdholm Hedwig Haas

Carmen Lee

Abstract

Toward Non-invasive Neonatal Gas Monitoring with Plasma-based Spectroscopy

Hampus Fröjdholm, Hedwig Haas, Carmen Lee

Populärvetenskaplig sammanfattning av projektet

Acknowledgement

Contents

Glossary

1 Introduction

1.1 Background

1.2 Objectives

1.3 Transcutaneous Gas Monitoring

1.4 Peripheral Devices for the SSRR-based TCM

1.5 Specific Aims

2 Methods

2.1 Experimental Setup

2.2 Circuit Board Design

2.3 Sensor Pad Design

acrylic glass

epoxy coated PLA rim polyimide negative

1 cm

(a) (b) (c)

(d) (e)

2.4 Signal Processing

350 400 450 500 550 600 650 700

0 0.2 0.4 0.6

Wavelength [nm]

In te ns it y

3 Results and Discussion

3.1 Circuit Board

3.2 Sensor Pad

3.3 Signal Processing

4 Conclusions

References

Appendix A Spectrum Analysis

A.1 spectrum_analysis.m