Linköpings universitet SE–581 83 Linköping
Linköping University | Department of Biomedical Engineering
Master thesis, 30 ECTS | Biomedical engineering
2017 | LIU-IMT/LITH-EX-A--17/001--SE
Development of a
photopletysmography based method
for investigating changes in blood
volume pulsations
–
for the purpose of pressure ulcer prevention
Frida Nylund
Tova Persson
Supervisor : Martin Hultman Examiner : Marcus Larsson
Upphovsrätt
Detta dokument hålls tillgängligt på Internet – eller dess framtida ersättare – under 25 år från publiceringsdatum under förutsättning att inga extraordinära omständigheter uppstår. Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner, skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för ickekommersiell forskning och för undervisning. Överföring av upphovsrätten vid en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av dokumentet kräver upphovsmannens medgivande. För att garantera äktheten, säkerheten och tillgängligheten finns lösningar av teknisk och admin-istrativ art. Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i den omfattning som god sed kräver vid användning av dokumentet på ovan beskrivna sätt samt skydd mot att dokumentet ändras eller presenteras i sådan form eller i sådant sam-manhang som är kränkande för upphovsmannenslitterära eller konstnärliga anseende eller egenart. För ytterligare information om Linköping University Electronic Press se förlagets hemsida http://www.ep.liu.se/.
Copyright
The publishers will keep this document online on the Internet – or its possible replacement – for a period of 25 years starting from the date of publication barring exceptional circum-stances. The online availability of the document implies permanent permission for anyone to read, to download, or to print out single copies for his/hers own use and to use it unchanged for non-commercial research and educational purpose. Subsequent transfers of copyright cannot revoke this permission. All other uses of the document are conditional upon the con-sent of the copyright owner. The publisher has taken technical and administrative measures to assure authenticity, security and accessibility. According to intellectual property law the author has the right to be mentioned when his/her work is accessed as described above and to be protected against infringement. For additional information about the Linköping Uni-versity Electronic Press and its procedures for publication and for assurance of document integrity, please refer to its www home page: http://www.ep.liu.se/.
c
Frida Nylund Tova Persson
Abstract
The aim of this thesis is to develop and evaluate a prototype for measuring volumetric changes of the blood with PPG. The prototype consist of a sensor plate controlled in Lab-view and algorithms for signal processing and analysis of results developed in Matlab. A study divided into three parts is performed, where the collected data is used for further development and alternations of the prototype between the study parts.
A measurement in the study is divided into three stages before, during and after pressure, where the subject is changing body position between each stage in order to either apply or relieve pressure from the sensor plate. The amplitude changes of the recorded signals are analysed and the results from the stable parts of the measurements are presented as the ratio between before and during pressure. A ratio separated from 1 either show a decrease or an increase of pulsating blood volume as a response to the applied pressure.
The results from the study show that there are both large spatial variations and large varia-tions over time in the measurements. Today the prototype does not give repeatable results and there are several uncertainties in the measurement method. An optimal sensor plate would be flexible and have several LEDs over a larger area in order to give reliable result despite spatial variations.
Abbreviations and Notations
Abbreviations
AC signal Alternating Current signal bpm Beats Per Minute
DC signal Direct Current signal DAQ Data Acquisition
FWHM Full width at half maximum HP filter High-Pass filter
I/O Input/Output
LED Light Emitting Diode LP filter Low-Pass filter
mmHg Millimeters of Mercury PPG Photoplethysmography PIV Pressure-Induced Vasodilation RMS Root mean square
SNR Signal to noise ratio SD standard deviation
UV Ultraviolet
VI Virtual Instrument
Notations
g Anisotropy m f p Mean free path n Refractive index p Significance r Correlation coefficient r2 Coefficient of determination µa Absorption coefficient µs Scattering coefficient µt Total attenuation coefficient
Contents
Abstract iii
Abbreviations and notations iv
Contents v
List of Figures vii
List of Tables ix 1 Introduction 1 1.1 Motivation . . . 1 1.2 Aim . . . 1 1.3 Research questions . . . 2 1.4 Delimitations . . . 2 2 Theory 3 2.1 Microcirculation . . . 3
2.2 The response of the skin to pressure . . . 4
2.3 Photoplethysmography . . . 5
2.4 Interaction between light and tissue . . . 7
2.5 Signal processing . . . 12 2.6 Statistics . . . 13 3 Method 14 3.1 Planning phase . . . 14 3.2 Setup phase . . . 15 3.3 Study 1 . . . 19
3.4 Development of a new sensor plate . . . 25
3.5 Study 2 . . . 27
3.6 Study 3 . . . 31
3.7 Evaluation . . . 33
4 Results 35 4.1 Interpretation the results . . . 35
4.2 Study 1 . . . 36
4.3 Study 2 . . . 39
4.4 Study 3 . . . 43
4.5 Evaluation . . . 45
5 Discussion 62 5.1 The study results . . . 62
5.2 The mean value of the ratios . . . 63
5.3 Signal processing and the amount of usable signals . . . 64
5.4 Changes in DC level . . . 65
5.5 The difference between the ratio calculated with the AC level, the scaled AC level and the theoretical fraction of pulsating blood . . . 66
5.7 Duration of the measurements . . . 68
5.8 Placement of the sensor plates . . . 68
5.9 Amount of needed LEDs and detectors and placement of these . . . 69
5.10 Differences in the PPG signal for different penetration depths . . . 69
5.11 Reproducibility of measurements with the prototype . . . 70
5.12 Method . . . 72
5.13 The work in a wider context . . . 73
6 Conclusion 74
Bibliography 76
List of Figures
2.1 Skin structure . . . 4
2.2 PPG modes . . . 6
2.3 The components of the PPG signal . . . 7
2.4 Reduced scattering coefficients . . . 8
2.5 Absorption spectra of hemoglobin . . . 9
2.6 Absorption coefficients . . . 10
3.1 The structure of the original sensor plate . . . 15
3.2 A block diagram of the VI . . . 17
3.3 The front panel of the VI . . . 17
3.4 Signals from a measurement with one red LED . . . 18
3.5 Signals from a measurement with one green LED . . . 18
3.6 Placement of the sensor plate, study 1 . . . 20
3.7 Pulse peaks in the original signal, study 1 . . . 21
3.8 The work flow for the signal processing and analysis . . . 21
3.9 The cutting process . . . 22
3.10 The frequency response of a signal . . . 22
3.11 The filtering of a signal . . . 23
3.12 Peak and valley detection . . . 24
3.13 The AC level, scaled AC level and theoretical fraction of pulsating blood . . . 24
3.14 The new sensor plate . . . 25
3.15 Placement of the sensor plates, study 2 . . . 27
3.16 The block diagram of the second version of the VI . . . 29
3.17 The front panel of the second version of the VI . . . 29
3.18 Pulse peaks in the original signal, study 2 . . . 30
3.19 Placements of the sensor plates, study 3 . . . 32
4.1 How to interpret the ratios . . . 36
4.2 Calculated ratios for the back, study 1 . . . 37
4.3 Calculated ratios from a short time interval, study 1 . . . 38
4.4 Calculated ratios on the back, study 2 . . . 39
4.5 Calculated ratios on the shoulders, study 2 . . . 40
4.6 Calculated ratios on the hip, study 2 . . . 40
4.7 Calculated ratios on the back for study 1 with the signal processing from study 2 . 41 4.8 Calculated ratios from a shorter time interval, study 2 . . . 42
4.9 Calculated ratios on the back, study 3 . . . 43
4.10 Calculated AC ratios from study 3 with the signal processing from study 2 . . . 44
4.11 Calculated AC/DC ratios from study 3 with the signal processing from study 2 . . 44
4.12 Correlation of the mean ratios between the two measurements in each study . . . . 48
4.13 Correlation between the different types of ratios, study 2 . . . 49
4.14 Correlation between different measurements, using different types of ratios, study 1 50 4.15 Correlations between placements, study 2 . . . 51
4.16 Correlation between two measurement depths, study 2 . . . 52
4.17 Correlation between two two source-detector distances, study 3 . . . 53
4.18 The correlation for back ratios between detectors, sensor plates and measure-ments, study 2 . . . 54
4.19 The correlation for back ratios between detectors, sensor plates and
measure-ments, study 3 . . . 56
4.20 Correlation between study 2 and study 3 . . . 56
4.21 AC ratios for the back, comparison between having straight and bent legs . . . 57
4.22 AC ratios for the side, comparison between having straight and bent legs . . . 58
4.23 Pressure provocation measurement . . . 59
4.24 Liniment provocation measurement . . . 60
List of Tables
3.1 Wavelength information for the LEDs . . . 16
3.2 Intensities from different plastic covers . . . 19
4.1 Approved signals, study 1 . . . 36
4.2 Approved signals, study 2 . . . 41
4.3 Approved signals, study 1 with the signal processing from study 2 . . . 41
4.4 Approved signals, study 3 . . . 43
4.5 Mean value, mean variance and standard deviation of the variance . . . 45
4.6 Mean values for the green ratios on the back for all measurements and participants 46 4.7 Mean values for the red ratios on the back for all measurements and participants . 46 4.8 Mean value and standard deviation of the mean for each of the measurements in study 2 and 3 . . . 47
4.9 Mean value and standard deviation of all mean for each subject in study 2 and 3 . 47 4.10 Correlation coefficient between the mean ratio of all measurements in study 2 and 3 48 4.11 Correlation coefficient for different types of ratios, study 2 . . . 49
4.12 Correlation coefficient for different types of ratios, study 1 . . . 50
4.13 Correlation coefficients for different placements, study 2 . . . 51
4.14 Correlation coefficients for different depths, study 2 . . . 52
4.15 Correlation between two two source-detector distances, study 3 . . . 53
4.16 Correlation coefficients for the back, study 2 . . . 54
4.17 Correlation coefficients for the back, study 3 with the signal processing from study 3 55 4.18 Correlation coefficients for the back, study 3 with the signal processing from study 2 55 4.19 Correlation coefficients for the back between study 2 and study 3 . . . 56
4.20 Mean and standard deviation of the AC ratio for 8 weekly measurements on two subjects. . . 60
1. Introduction
In this chapter the scope of the master thesis is introduced. The aim and motivation of the thesis work is presented together with research questions and delimitation of this project.
1.1
Motivation
Patients developing pressure ulcers is a worldwide problem, both because of the inconve-nience for the patients and because of the high health care costs [4]. A pressure ulcer is defined as a localized damage in the skin and the underlying tissue caused by pressure [4]. Pressure ulcers are often developed at bony prominences and cause a lot of suffering to the patient. To prevent or reduce pressure ulcers special routines for care and pressure relief equipment is used, but it is difficult to decide which patients have a high risk for developing pressure ulcers. In Sweden today, rating scales are used to decide the risk, but a more precise method is needed.
When the skin is exposed to pressure, the normal response is an increase of blood perfusion in that area to prevent oxygen deficiency. Unfortunately, some people instead have a decrease of the blood perfusion as response to pressure which increase the probability of developing pressure ulcers. Earlier research by Källman et al. has shown that this decrease of blood perfusion as response to pressure is present for approximately 20-25 % of the population. [12]
Photoplethysmography (PPG) is a non-invasive method that can be used to measure volu-metric changes of the pulsating blood, which is correlated to the perfusion [1]. It is an optical method that use light emitting diodes (LED) to illuminate the skin and detectors to collect the light that returns from the tissue [1]. This thesis work is performed in collaboration with the start-up company PUsensor, which intend to develop a product based on PPG for estimating the risk for developing pressure ulcers.
1.2
Aim
The aim of the thesis work is to develop and evaluate a prototype for measuring volumetric changes of the blood with PPG, with and without applied pressure of ones body weight. An algorithm to control an existing sensor plate is developed in Labview together with an algorithm for processing the received signals. A study is performed with approximately 10 individuals and data from the study is used to develop the method of measurement, the prototype design and algorithms for controlling and analysing the signals. The data is also used to evaluate the stability of the prototype over time.
1.3. Research questions
1.3
Research questions
1. What would be an optimal placement of the sensor plate and what is the shortest pos-sible duration time of the measurements that achieve reliable results according to the collected data?
2. Which conclusions can be drawn from the study result with respect to the number of LEDs and detectors and the placement of these?
3. What differences can be found in the PPG signal for measurements with applied pres-sure using different wavelengths and source-detector distances, that is to say for differ-ent penetration depths?
4. What can be decided about the reproducibility and stability of the method of measure-ment with the prototype? Can the same result be repeated for a single individual at different placements of the sensor plate or for two separate measurements?
1.4
Delimitations
The main focus on the thesis will be on the development and evaluation of a specific proto-type.
The participants of the studies are a convenience sample without any relation to a risk for developing pressure ulcers. The study is performed in a non-clinical environment and factors such as room temperature can not be ensured to be constant between different measurement occasions.
2. Theory
In this chapter the theory basis for the thesis work is presented. A biological background of how the body reacts to pressure is followed by a technological description of PPG. Thereafter the interaction between light and tissue is addressed and the necessary signal processing of the PPG signal is briefly discussed. Lastly the statistic terms correlation and significance are explained.
2.1
Microcirculation
In the peripheral parts of our body, one can find the smallest and finest blood vessels, capillar-ies. The blood flows through arterioles that transcend into capillaries and after the capillaries the blood reaches the venous side with venules. The blood flow through arterioles, capil-laries and into venules is called the microcirculation. The larger vessels have diameters of maximum 100 µm [12] and the capillaries have diameters of 5-10 µm [20]. In the capillaries there is an exchange of oxygen, nutrients and metabolites between the blood and the tissue [20][3].
The regulation of blood flow in the microcirculation is through controlling the smooth mus-cle tissue in the blood vessels, which leads to a constriction or dilation of the blood vessels [3]. The system that controls the blood flow works centrally or locally [4]. The central con-trol provides optimal circulatory homeostasis by regulating blood pressure, cardiac output, blood volume, thermoregulation and venous flow [4]. The local control provide optimal nu-tritive blood flow through control of the regional blood flow, blood distribution and capillary exchange [4].
2.1.1
Structure of the skin
Based on the structure, the skin can be divided into two main parts, the superficial epidermis and the deeper dermis [20]. Depending on the body site the thickness of the skin varies between 0.5 to 4.0 mm and the epidermis has a thinner portion compared to the dermis [20]. Figure 2.1 shows a schematic image of the structure of the skin.
The epidermis is composed of epithelial tissue, which in turn contains four main types of cells: keratinocytes, melanocytes, Langerhans cells and Merkel cells. About 90% of the epi-dermal cells are keratinocytes, which produce the protein keratin that protects the skin from heat, microbes and chemicals. Another 8% of the epidermal cells are melanocytes that pro-duce the pigment melanin that colours the skin and protects against UV-light. The Langer-hans cells participate in the immune response and the Merkel cells detect the touch sensa-tions. [20]
col-2.2. The response of the skin to pressure
Figure 2.1: The structure of skin.
elastic fibers and dermal papillae. The dermal papilla contains capillary loops, nerve endings sensitive to touch and free nerve endings. The reticular region makes up about four-fifths of the total thickness of the layer and consists of irregular connective tissue. In the connec-tive tissue fibroblasts, bundles of collagen and coars elastic fibers can be found. In an adult, the dermis contains 8-10% of the total blood amount, due to the extensive network of blood vessels. [20]
2.2
The response of the skin to pressure
When a pressure is applied to tissue the result will be a more compact tissue with a larger blood volume [20]. When the skin is exposed to a pressure that do not stimulate pain recep-tors a sustained vasodilation is induced, called pressure-induced vasodilation (PIV) [17]. PIV protects tissues from pressure induced ischemic damage, which implies a restriction in blood supply to the tissue [17]. PIV has been shown to be present for all healthy individuals, both young and old, at low levels of pressure, such as 32 mmHg over the greater trochanter [21]. If the applied pressure is too high, the blood vessels become occluded and no blood flow occurs [4]. There are also studies that indicate that some people lack the PIV response and the blood flow is weakened at pressures below the occlusion pressure [4].
Reactive hyperemia (RH) is when the blood flow increase above normal levels, due to the release of a brief arterial occlusion [8]. The increased blood flow caused by the RH restores oxygen levels and reduce the levels of vasodilator metabolites to meet the metabolic needs of the tissue [5]. RH should therefore arise after an induced pressure is released.
2.3. Photoplethysmography
Both PIV and RH are present in superficial tissue as skin, as well as in deeper tissue structures [5]. A lack of increased perfusion at pressure can lead to an insufficient blood supply and cause hypoxia, which is associated with the development of pressure ulcers [5].
2.2.1
Pressure ulcers
The development of pressure ulcers depends on both intrinsic factors that describe a per-son’s state of health and extrinsic factors [4]. Intrinsic factors can be morphology, geometry and mechanical properties of tissues, transport and thermal properties, individual physiol-ogy and repair ability [7]. These factors all affect the individual susceptibility and tolerance to develop pressure ulcers [7]. Extrinsic factors are related to the mechanical boundary con-ditions [7]. They are for example magnitude and duration of load, but also which type of load it is; shear, pressure or friction [7]. To prevent pressure ulcers, materials that reduce the extrinsic factors is used, for example mattresses that distribute the support of body weight over a larger area [4].
There is large scientific evidence and clinical resonance for three key direct causal risk fac-tors for pressure ulcer development. These are are immobility, perfusion and skin status. Key indirect causal risk factors are albumin level, sensory perception, diabetes, nutrition and moisture. These all have a good scientific evidence and/or clinical resonance but with some statistical inconsistencies in their association with pressure ulcers. Other potential indirect risk factors are body temperature, age, medication, acute illness, infection, chronic wound and pitting oedema. There are complex interrelationships between these potential indirect risk factors and the key indirect risk factors, but there exists limited scientific evidence so further research is needed to verify these risk factors. [7]
The primary cause of pressure ulcers is unknown, but the most common understanding is that tissue ischemia from an applied pressure is the main cause [4]. Initially, ischemia leads to hypoxia, a decrease in nutrient supply and accumulation of waste products, if the ischemia continues it might lead to tissue inflammation, oedema, thrombus formation and cell death [12]. Ischemia can arise from any condition where the oxygen demand of the tissue is greater than the supply, therefore it is possible to cause ischemia without a total occlusion of the blood supply [12].
The assessment of blood perfusion at different tissue depths could be used as an individual-ized measurement related to the development of pressure ulcers [3]. A non-invasive method of obtaining information about blood perfusion is through photoplethysmography (PPG). [1]
2.3
Photoplethysmography
PPG is a non-invasive optical method for measuring changes of blood volume in tissue [1]. The tissue is illuminated and the light which has passed through the tissue is detected and the signal will change depending on the amount of blood present in the tissue [1]. This section will focus on the PPG device, signal waveform, artifacts and reproducibility. More information about the interaction between light and tissue, which is what gives the PPG signal, will follow in section 2.4.
2.3. Photoplethysmography
2.3.1
The PPG device
The basic components of a PPG device are a light source and a photodetector. The light source should have a relative narrow bandwidth and be as stable as possible both with respect to the average intensity and peak wavelength shift for different temperatures. Additionally it should be compact, have a long operating life and be mechanically robust and reliable. The intensity should be high enough to give a good signal but low enough not to cause excessive tissue heating. Normally a light emitting diode (LED) is used as light source. The photodetector converts light energy into electrical energy which can be registered. It should be compact, have a fast response time and be sensitive to the wavelength the LED emits. After the photodetector has detected the light the signal normally goes through a low noise electronic circuitry for amplification and filtering. [1]
There are two different modes of PPG: transmission mode and reflection mode, these two modes are shown in figure 2.2. In transmission mode the photodetector and LED are on opposite sides of the tissue and the light is transmitted through it . In reflection mode the photodetector and LED are on the same side of the tissue and the detected signal comes from the backscattered light. The site of measurement is limited in transmission mode, since the tissue have to be thin enough to transmit a detectable amount of light, whereas reflection mode measurements can be sited anywhere on the body. [19]
Figure 2.2: The two modes of PPG: transmission mode (left) and reflection mode (right).
2.3.2
The PPG waveform
The detected PPG signal consists of two components, one alternating current (AC) and one direct current (DC) component. The AC component is dependent on the pulsatile blood flow, and is therefore varying with the same frequency as the pulse. Since the AC component is directly connected to the pulsatile blood it can be used to estimate the blood perfusion. The DC component is quasi-constant, and relates to what the the tissue consists of in addition to the pulsatile blood. These two components are shown in figure 2.3. The DC component is not fully constant but has slow variations due to respiration, thermoregulation and sympathetic nervous system activity. All origins of the PPG waveform is not yet fully understood, but it is still a useful diagnostic tool. [1]
2.4. Interaction between light and tissue
Figure 2.3: The components of the PPG signal.
2.3.3
Artifacts
Artifacts in the PPG signal can arise from several different sources, for example from ambient light, patient movement or probe-tissue movement. To minimize movement artifacts the pa-tient should be as still as possible and the probe should be held securely into place. Artifacts from ambient light can be reduced by attaching the probe securely, further shading the area of the measurement, and by electronic filtering. [1]
2.3.4
Reproducibility of PPG measurements
To have confidence in detecting significant responses reproducibility of a clinical physiologi-cal measurement is of utter importance. The reproducibility of PPG measurements is affected by several factors related to the subject, the probe and the environment. Subject related fac-tors are for example posture, relaxation, movement, respiration, wakefulness and acclimati-zation. Probe related factors can be method of attachment, measurement site and pressure between probe and tissue, while environmental factors are for example room temperature. No standards for clinical PPG measurements are yet internationally recognized. [1]
2.4
Interaction between light and tissue
To understand PPG it is important to understand the basic principles of interaction between light and tissue. When light impinges on tissue it can either be reflected, absorbed or scat-tered. The refractive index, n, of the tissue and how it relates to the refractive index of the surrounding decides if the light is reflected or transmitted into the tissue. At transmission
2.4. Interaction between light and tissue
µs, and the anisotropy, g, which often are combined into the reduced scattering coefficient µ1s =µs(1 ´ g). [10]
Scattering
Scattering arises due to refractive index mismatch between scatterers and the surrounding, but is also related to the size of the scatterers. Scattering includes change of both direction, phase and polarization of the light. Filamentous proteins such as collagen in the dermis and keratin in the epidermis are the major contributors to scattering in the skin. In the epidermis also melanosomes are relevant scatterers and structures such as cell walls and nuclei also contribute to the scattering. [14]
Mie theory describes scattering from ideal spheres and can be used to approximate the scat-tering in the skin. The scatscat-tering coefficient, µs, gives the probability of scattering and
the scattering phase function, p(θ), describes the the distribution of the scattering angle.
Since light that passes through skin normally undergoes multiple scattering events, it is more relevant to know the average angle than the distribution and therefore the anisotropy, g=xcos(θ)y, is commonly used instead of the scattering phase function. [10]
Without scattering the light would never change direction and it would be impossible to detect a signal from a detector next to a light source on the skin. It is therefor a necessity for PPG measurements in reflection mode. In a review by Jacques 2013 [10] the mean value of
µ1s of the skin is found to be 46.0 cm´1 at 500 nm with a standard deviation of 13.7 cm´1,
when comparing results from 8 studies. µ1
sthen decreases for longer wavelengths. This is a
much higher value than what is found in most other tissues, for example brain (24.2 cm´1), breast (16.8 cm´1), bone (22.9 cm´1) and fatty tissue (18.4 cm´1) [10]. Results from seven studies of how the scattering coefficient of skin in vivo changes with wavelength is shown in figure 2.4. It is evident from the figure that there are variations between the studies and it can also be seen that the same study found different scattering coefficient on the arm and on the forehead. Some of the variations could be due to true differences in µ1
s, but it has also been
shown that different analysis of the same data can result in very different estimates as well, for example by assuming different phase functions. [14].
350 400 450 500 550 600 650 700 750 Wavelength [nm] 10-1 100 101 102
Reduced scattering coefficient [mm
- 1]
Reduced scattering coefficients of the skin
Simpson 1998 Dognitz 1998 Torricelli 2001 (mean) Doombos 1999 arm Doombos 1999 forehead Svaasand 1995 Zonios 2006 Bosschaart 2011
2.4. Interaction between light and tissue
Absorption
Absorption is a process where a photon transfers its energy to a molecule, most commonly the energy is transferred by exciting an electron [14]. The molecules that absorb visible light are called chromophores and the most dominant one in the skin is hemoglobin [14]. Melanin is also a prominent chromophore in the skin and bilirubin, carotene and lipids act as chro-mophores, but they are less prominent compared to the hemoglobin and melanin [14]. Water is a quite weak absorber in the visible spectrum of light but since it is so abundant in tissue its contribution to the absorption sometimes needs to be considered [10].
Hemoglobin is the protein in the red blood cells that carry the oxygen [14]. It consists of four polypeptide chains bound to one heme each [14]. The heme is responsible both for the ab-sorption and binding to the oxygen [14]. Since the binding of oxygen affects the composition of the hemoglobin the absorption spectrum for oxygenated and deoxygenated hemoglobin is different [14]. The spectra for both oxygenated and deoxygenated hemoglobin are shown in Figure 2.5. In both spectra there is a large peak in the blue range [14]. In the green to yellow range, between 500 and 600 nm, there is a double peak for the oxygenated hemoglobin and a single peak for deoxygenated hemoglobin [14]. The wavelengths where the two curves inter-sect, for example around 800 nm, are called isosbestic points, and at these points there is no difference in absorption for different oxygen saturation [10]. Also other forms of hemoglobin, such as carboxyhemoglobin and methemoglobin have slightly different spectra, but these variants are normally not as abundant in blood [14].
200 300 400 500 600 700 800 900 1000 Wavelength [nm] 102 103 104 105 106 Extinction coefficient [cm -1 /(moles/liter)]
Molar extinction coefficient
Oxygenated hemoglobin Deoxygenated hemoglobin
Figure 2.5: The absorption spectra of oxygenated and deoxygenated hemoglobin. The molar extinction
coefficient is shown and to change to absorption coefficient one needs to multiply it with a factor 2.303 and the molar concentration of hemoglobin. [18]
Melanin absorbs light from the whole visible spectrum to some extent; more strongly for UV light and then decreasingly for the longer wavelengths. This broadband absorbance is not fully understood, but is believed to emerge from a combination of several absorption peaks from the various disordered oligomers and polymers in the melanin. There is large variations in the amount of melanin in the skin of different people, and there is also variations in the structure of melanins, which makes the absorbance complex. [14]
2.4. Interaction between light and tissue 350 400 450 500 550 600 650 700 750 Wavelength [nm] 10-2 10-1 100 101 Absorption coefficient [mm - 1]
Absorption coefficients of the skin
Torricelli 2001 (mean) Svaasand 1995 epidermis Svaasand 1995 dermis Zonios 2006 Bosschaart 2011
Figure 2.6: Absorption coefficients from four different studies. Data from [14]. Beer-Lambert law
The absorption of light in tissue is govern by the Beer-Lambert law.
I =I0¨e´µax, (2.1)
where I0is the original intensity and I is the intensity of the light after having travelled the
distance x through a material with absorption coefficient µa [9]. The absorption coefficient
depends on the extinction coefficients of the chromophores in the tissue and the concentration of these [10].
2.4.1
Wavelength dependency
Since absorption and scattering coefficients change with the wavelength of light it is impor-tant to consider which colour the used light source of a PPG device has. Water is the main component of tissue and it strongly absorbs light in the ultraviolet and far infrared spectrum. Melanin absorbs all visible light, but does so more strongly for the shorter wavelengths. Red and near infrared light therefore can pass into the tissue more easily than other visible light and be largely affected by the blood, why these wavelengths give a stronger AC signal and are most commonly chosen in PPG devices . [1]
2.4.2
Penetration depth and backscattered intensity
As previously mentioned red and near infrared light passes the tissue more easily without be-ing absorbed. This results in a greater penetration depth and larger samplbe-ing volume [1]. The stronger absorption of shorter wavelengths give them a shallower penetration and smaller sampling volume [1]. But also the scattering need to be considered when looking at penetra-tion depth and the intensity of the backscattered light. The Beer-Lambert law (equapenetra-tion 2.1) can be modified to include scattering by replacing µa with µt = µa+µs, which is the total
attenuation coefficient, resulting in the following expression [16]
I= I0¨e´µtx. (2.2)
If looking at individual photons, the pathlength before they interact with tissue will be ran-domly distributed with a probability density function that follows the Beer-Lambert law [16].
2.4. Interaction between light and tissue
The average distance a photon travels before interacting with the tissue is called the mean free path, m f p and is the reciprocal of µt[16]
m f p= 1 µt
= 1
µa+µs. (2.3)
If the interest lies purely in penetration depth along the axis of the incident light beam, equa-tion 2.2 is all that is needed. If there is an interest also in the intensity of the backscattered light one need to have the ratio between µa and µt and the phase function. The ratio gives
the probability of absorption compared to scattering and the scattering phase function gives the probability distribution of the angle in case of scattering [16]. Knowing these parameters one can simulate paths for large quantities of photons, using Monte Carlo simulations, and thereby predict how light will spread in a tissue and for example how much will be backscat-tered at different distances from the light source [16].
2.4.3
How the measured intensity is related to the fraction of pulsating blood
As mentioned earlier the backscattered intensity is affected by the pulsating blood since the hemoglobin is a dominant absorbant. If tissue would be modelled as one part blood and one part static material the absorption coefficient of the tissue would be the following:
µa =εHbcblood+εstatic(1 ´ cblood), (2.4)
where εHband εstatic are the extinction coefficients of hemoglobin and the remaining tissue,
respectively, and cbloodis the fraction of blood in the tissue at diastole. At systole the
frac-tion of blood increases with an amount∆c, which represents the pulsating blood [11]. This pulsating fraction of blood cannot be directly measured but can be calculated by using the Beer-Lambert law (equation 2.1) at diastole and systole, dividing the results, taking the natu-ral logarithm and rearranging as follows:
Idiastole
Isystole
= I0¨e
´(εHbcblood+εstatic(1´cblood))x
I0¨e´(εHb(cblood+∆c)+εstatic(1´cblood´∆c))x
=e(εHb´εstatic)∆cx (2.5) ln Idiastole Isystole ! =(εHb´ εstatic)∆cx (2.6) ∆c= lnIdiastole Isystole (εHb´ εstatic)x (2.7)
In the denominator of equation 2.7 ((εHb´ εstatic)x), the extinction coefficient for the
hemoglobin, εHb, is constant if there is no change in saturation levels. The pulsating blood
should be almost 100% saturated since it comes from the arteries, but to be sure a wavelength without much difference in absorption between oxygenated and deoxygenated blood can be used, for example around 800-810 nm, see figure 2.5.
The extinction coefficient for the remaining tissue, εstatic in equation 2.7, can differ between
different persons and different sites on the body since it depends on what the underlying tissue is composed of. It might also change if there is a structural change in the tissue when a pressure is applied. An assumption that needs to be made to enable a correct comparison
2.5. Signal processing
The distance that the light has travelled between the LED and the detector, x in equation 2.7, is not a constant but a distribution of different lengths for different photons. Assuming the scattering coefficient is approximately constant with or without pressure the distribution of x should also be constant, giving comparable results of∆c with or without an applied pressure. With these discussed assumptions the entire denominator of equation 2.7 can be seen as a constant for a certain measurement site on a certain person. This means that∆c is propor-tional to the natural logarithm of the recorded intensity at diastole divided by that at systole. The AC component of the signal is normally very small in relation to the DC component [1], [11]. The AC component can be expressed as Idiastole´Isystoleand the DC component is
approximately Isystole. Making the assumption that the AC is much smaller then the DC∆c
can be approximated with Maclaurin expansion and simplified to the ratio of the AC and DC levels as shown in equation 2.8.
∆c9 ln Idiastole Isystole ! =ln 1+ Idiastole´Isystole Isystole ! « Idiastole´Isystole Isystole « AC DC (2.8)
2.5
Signal processing
As previously mentioned the PPG signal has a pulsatile AC component, which depends on the blood volume changes originating from the beating of the heart, and a quasi-constant DC component which varies slowly due to for example respiration [1]. There is commonly noise from motion and electronics in the signal as well [1], [13]. The AC component of the PPG can have a frequency of 0.5-4.0 Hz [13]. At rest a normal heart rate is 60-100 beats per minute (bpm) or down to 40 bpm for very physically active people [2]. 40-100 bpm correspond to a frequency of 0.67-1.67 Hz, which is where the AC component is expected to be if the PPG is measured for healthy subjects at rest. For respiration the frequency band is 0.04-1.6 Hz, [13] but the higher frequencies should not be present for healthy subjects at rest. Electrical noise with a frequency of 50 Hz is often present in the PPG signal [1]. Motion artifacts can emerge from the patient’s movement and generally have a frequency of 0.1 Hz or higher [13].
2.5.1
Low-pass and high-pass filters
The high frequency noise can be filtered out with a low-pass (LP) filter and the DC component can be filtered out with a high-pass (HP) filter. An LP filter lets frequencies below a certain limit pass, while the higher frequencies are stopped. The limit frequency is called the cut-off frequency. A high pass filter, on the other hand, stops frequencies below its cut-off and lets frequencies above pass. Using both a high-pass filter and a low-pass filter results in filtering out everything except a band of frequencies, which will be equivalent of a band-pass filter. [15]
The choice of cut-off frequency is of great importance in PPG applications and one has to com-promise between removing enough noise to get a clearly visible signal and keeping enough information so that the waveform is not distorted [1]. Since the frequencies of the AC signal and the motion artifacts overlap it is impossible to filter out all motion artifacts with classical filtering methods [13].
2.6. Statistics
2.6
Statistics
The correlation, r, between data sets and the significance, p, of the correlation is one way of evaluating the reproducibility and reliability of a method. The coefficient of determination, r2, describes the variation and strength of the correlation coefficient.
2.6.1
Correlation
The correlation of two variables show the linear dependency between the variables and can be described by the correlation coefficient which depends on the covariance and the standard deviation of the variables as equation 2.9 shows. [6]
r= cXY
sxsy, (2.9)
cXYis the covariance of X and Y and sxand syis the standard deviation of X and Y respectively
[6].
The covariance of data points in a data set can be calculated with equation 2.10.
cXY= 1 n ´ 1 n ÿ i=1 (xi´x)(yi´y) = 1 n ´ 1 n ÿ i=1 (xiyi´nx y), (2.10)
n is the number of data points, and x and y are the means of X and Y respectively. [6]
The correlation coefficient in a data set is between -1 and 1. If r= +1 the data is presented as straight line with a positive slope and high x values corresponds to high y values. If r=´1 instead, the data is presented as a straight line with a negative slope and and high x values corresponds to low y values. An r value close to zero indicates that there is no or a very small linear relationship between the variables. [6]
2.6.2
Significance of the correlation coefficient
The significance, p, of the correlation coefficient, is a measurement of the uncertainty of the calculated value. The p value gives the probability of the coefficient r to occur, even though there is no correlation in the data set. The smaller the p value is, the more significant the relationship is. There are three commonly used significance levels: 0.05, 0.01 and 0.001, which determines if the result is significant at that level [6].
2.6.3
Coefficient of determination
The coefficient of determination is calculated by taking the square of the correlation coeffi-cient and indicates the strength of the correlation coefficoeffi-cient. The value of r2can be inter-preted as the percentage of the variance in variable Y that can be explained by the linear relationship between X and Y.
3. Method
In this chapter the method is presented. This is done in chronological order starting with the planning phase and setup phase. After these starting phases comes the first study, then the development of a new sensor plate based on experiences from the setup and first study. Thereafter study 2 and 3 are described. Lastly there is an evaluation section describing the evaluation of the prototype.
3.1
Planning phase
Before the study could be performed an ethical application was sent to the Regional Ethical Review Board (Etikprövningsnämnden) in Linköping Sweden. To get an ethical approval the studies had to be planned. This section describes this planning phase.
3.1.1
Planning the studies
The study was planned to be divided into three studies, called study 1, 2 and 3, which were to be performed one by one and have a time gap of a few weeks between them. The plan was to use a convenience sample of 10 to 20 participants and that all participants would be present at three occasions and measurements would be taken with one to four different placements of the sensor plate.
3.1.2
Planned method of measurement
The planned method of measurement was as follows. A subject start with a 15 minute rest lying down. After the rest, a five minute measurement with the subject lying on there stom-ach without applied pressure take place before the subject turns to lie on their back and a 10 minute measurement with the pressure from the body is performed. Afterwards the sub-ject turns around to lay on their stomach and a recovery measurement without pressure is performed for 10 minutes.The age, gender, pulse, blood pressure and body temperature are noted in a measurement protocol.
3.1.3
Ethical approval
Since this master thesis includes measurements on humans an ethical application was needed to execute the study. The ethical application was approved by the Regional Ethical Review Board (Etikprövningsnämnden) in Linköping, Sweden. The approvals reference number is 2017/21-31. All produced data is anonymous, only the subjects name and date of birth is stored with a numerical code corresponding to their data, since the same persons will attend at multiple occasions. This code list is stored separate from the acquired data and both the list and the data is password protected. All participants give both oral and written consent and can at any point withdraw from the study. The written consent is attached as appendix A. The description in the ethical application gives a maximum limit for measurement times and number of participants, but the approval applies also far shorter times and fewer participants.
3.2. Setup phase
3.1.4
Division of the work
The main parts of the thesis work were carried out together, but the areas of responsibility were divided. In each area both contributed with separate assignments and discussed possi-ble solutions together.
3.2
Setup phase
In the setup phase the provided material was investigated and the system was assembled and connected. Control of the system was programmed with Labview and improvements to the original sensor plate were made to enable measurements for the first study.
3.2.1
Material
The material provided for the measurement system was a laptop with Labview, a data acqui-sition (DAQ) device and a sensor plate.
The original sensor plate
The PPG sensor plate is structured with LEDs and sensors in 9 identical modules according to figure 3.1. All components are embedded in a flexible rubber plate to ensure comfort against the skin and the detectors and LEDs are covered with a plastic surface for cleanliness reasons. The dimensions of the plate is 200x130x5 mm.
Figure 3.1: The structure of the original sensor plate. The plate has 9 modules with 2 LEDs and 1
detector in each module.
The optical components of the probe consists of 18 LEDs, 9 of them are red and 9 of them green. The wavelengths are described in table 3.1 together with a yellow LED as an addi-tional alternative. Between each pair of LEDs there is a photodetector. The shortest distance between the LEDs and a detector is 6 mm, from center to center, the rubber barrier being 2 mm. The distances between LEDs and the adjacent detectors from different modules are be-tween 23 and 36 mm. The combination of these wavelengths and detector distances should correspond to a penetration depth of a few mm for the shorter distances and approximately 25 mm for the longer distances.
3.2. Setup phase
Table 3.1: Wavelength information for the LEDs.
Peak wavelength Dominant wavelength FWHM
Red LED 632 nm 625 nm 20nm
Green LED 518 nm 525 nm 35nm
3.2.2
Setup
The sensor plate was wired to the DAQ through multi-conductor cables. The LEDs were connected to a voltage source and controlled by the digital I/O ports of the DAQ device. The detectors were connected via amplifiers to the single ended analog ports of the DAQ device. All signals in and out from the DAQ were handled by Labview. The resolution of the DAQ is 16 bit.
3.2.3
Implementation
The controlling of, and measurements from, the DAQ connected to the sensor plate are as previously mentioned done in Labview. Figure 3.2 show the block diagram of the virtual instrument (VI) that was created and figure 3.3 shows the front panel which is visible to the user. First a while loop starts which displays all the detector signals in a "Waveform Graph" in the front panel of the VI. This makes it possible for the user to see the signals and adjust the sensor plate if they are too high due to poor attachment. The "Case structure" in this while loop turn off all the LEDs when a measurement starts and then turns on all the green LEDs in order to acquire signals in the "Waveform Graph". In the front panel of the VI the user can choose suitable file names and start or stop a measurement.
The measurement is done by using a "DAQ Assistant" block inside a timed while-loop. The DAQ Assistant is set to acquire 1 sample per iteration of the loop from all nine analog input channels. The acquired signals are written to an excel file with a "Write to Measurement File" block every 5000 iterations. The two "case structures" makes it possible to blink the LEDs in different patterns and save the results in separate files for the different patterns.
The controlling of the LEDs is done with another DAQ Assistant block inside the while loop. This DAQ Assistant is set to generate digital line output for the 18 digital I/O ports that are connected to the 18 LEDs.
Source-detector distance
The green and red light reaches different depths and different distances in the skin. To know how far the light from the different LEDs reaches, and thus on which distance from the LEDs the detectors can detect light an experiment was performed. The sensor plate was placed against the skin and a corner LED was lit while signals were detected from all detector chan-nels. This procedure was done for both a green and red LED. The results from these mea-surements are shown in figure 3.4 and 3.5. It can be seen that the red corner LED results in a very strong signal in the closest detector, but there is also a weak signal in the three adjacent detectors. For the green LED there is only a signal in the closest detector, and this signal is about 20 times weaker than the signal from the red LED.
Since the light from the green LEDs does not reach further than the closest detector it was decided to have all green LEDs lit simultaneously. Since the light from the red LEDs does reach further it was decided to only have the four corner LEDs lit simultaneously to enable investigating signals both from the short source-detector distance of the closest detector and the longer source-detector distance of the adjacent detectors.
3.2. Setup phase
Figure 3.2: The block diagram of the VI created in Labview.
3.2. Setup phase 10 15 20 25 30 Time [s] 0 0.5 1 1.5 2 2.5 Voltage [V]
A. Measuremnt with red corner LED
10 15 20 25 30 Time [s] 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 Voltage [V] B. Zoomed in version
Figure 3.4: The signal from the detectors as a red corner LED is lit up at second 14.
10 15 20 25 30 Time [s] 0 0.02 0.04 0.06 0.08 0.1 0.12 Voltage [V]
A. Measuremnt with green corner LED
10 15 20 25 30 Time [s] 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 Voltage [V] B. Zoomed in version
Figure 3.5: The signal from the detectors as a green corner LED is lit up at second 16.
Sampling frequency
The fastest possible iteration time for the main Labview loop (shown in the central part of figure 3.2) is 6 ms, which makes it possible to blink the LEDs in two separate patterns and achieve a sampling frequency of 83 Hz for both signals. One loop iteration corresponds to one LED pattern, so that every other iteration get the same pattern. These two patterns were chosen from previous results about source-detector distance, one with all green LEDs and one with the four red corner LEDs. The acquired signals are saved separately in order to distinguish which samples come from the green and the red pattern respectively.
3.3. Study 1
3.2.4
Improvements of the original plate
The original sensor plate is described in section 3.2.1 and it was put together as a start to this thesis work. There were problems with both the cover material and the probe material that needed to be fixed before the study could start.
Cover material for the surface
The first sensor plate had holes for all LEDs and detectors but did not have a cover material above the LEDs and detectors originally. To find a suitable material to cover the surface two different plastic materials were tested. An optimal material should be non reflective, since reflected light could affect the signal. Measurements were done with the sensor plate facing a black rubber surface, with and without the two different plastic materials in between, and the signals were recorded. The intensities from the measurement show how much light that can travel through the plate material and the plastic material. The result from these measurements are found in table 3.2. It is clear from the table that the second kind of plastic gives less intensity, and that is why this option was chosen to cover the sensor plate.
Table 3.2: The mean intensities from measurements against a black rubber surface with different
plastics covering the surface of the plate. As a reference a normal measurement on the skin gives a DC level of 0.34 for the green LEDs, 1.8 for the near red LEDs and 0.01 for the far red LEDs.
Without plastic surface Plastic surface 1 Plastic surface 2
Mean Green 0.003557 0.03846 0.029992
Mean Red near 0.003508 0.118183 0.078321
Mean Red far 0.003598 0.24442 0.006658
Conductivity of the probe material
During the first tests with the sensor plate many of the received signals were noisy and no pulse could be seen. The noise appeared because the back of the original sensor plate was made out of a black rubber material that was conductive. The easiest solution was to place an isolating material between the rubber back and the electrical circuit.
After further testing, noisy signals still appeared in some of the detectors, due to the front of the probe being made of the same rubber as the back. At the front side of the probe no isolating material could cover the whole circuit board, since it would disturb the detectors and LEDs. Instead electrical tape was used to cover all solder points at the front side of the electrical circuit.
3.3
Study 1
In the fist study the original sensor plate was used and placed on the back for two identical measurements on one subject.
3.3.1
Execution of measurements
In the first study, two identical measurements were performed on each subject without mov-ing the sensor plate in between. The planned method of measurement, described earlier in
3.3. Study 1
To attach the plate at the same position on each subjects back the spine was first found and then by pressing down slightly to the sides of the spine the sacrum could be felt. The plate was attached centered on the sacrum as figure 3.6 describe.
Figure 3.6: The placement of the sensorplate in the sacrum. Materials
The study was performed in a room with a standard bed with a foam mattress and the mea-surement equipment described in section 3.2. To attach the sensor plate to the skin different kinds of surgical tape were used. An ear thermometer was used to measure the subjects’ temperature and the room temperature. An automatic blood cuff was used to measure the blood pressure of the subject.
Participants
The participants were a convenience sample consisting of six females and seven males. 11 of the participants were in the age 20-30 and two of them were 60-80 years old. The participants were healthy and of varying length and weight.
3.3.2
Signal processing
The data from the measurements was processed in Matlab. Firstly the signals were investi-gated manually in order to find out how the signals should be processed to achieve a good analysis. Figure 3.7 show a piece of the original signal where the subjects pulse peaks can be seen, which show that the received signal is measuring the desired changes in blood vol-ume pulsations. Thereafter functions were developed to process the data automatically and preform the process described in figure 3.8.
3.3. Study 1 184 185 186 187 188 189 190 191 192 193 194 Time [s] 0.096 0.098 0.1 0.102 0.104 Amplitude [V]
The original signal
Figure 3.7: The pulse peaks in a section of the original signal.
Figure 3.8: A schematic overview of the work flow for the signal processing and analysis of the data.
Cut signals
The first function cuts the parts of the signal that corresponds to the turning of the subject, since there is a lot of movement artifacts in these parts of the signal. Figure 3.9 shows the original signal to the left and the cut signal to the right.
3.3. Study 1 200 400 600 800 1000 1200 Time [s] 0.06 0.08 0.1 0.12 0.14 Amplitude [V] Original signal 200 400 600 800 1000 Time [s] 0.07 0.08 0.09 0.1 0.11 0.12 0.13 0.14 Amplitude [V] Cut signal
Figure 3.9: A before and after image of the cutting process, the plot to the left is the original signal and
the plot to the right is the cut signal.
Remove signals
Some detector signals are removed from the data set due to disturbances in the signal such as movement artifacts, noise or reflections. They are removed based on their frequency content, seen in the Fourier domain. All signals should have a pulse peak in the frequency range 0.67 - 1.67 Hz, corresponding to a pulse of 40-100 bpm. If a signal lack the pulse peak it is removed, due to the absence of a dominant frequency component that should correspond to the pulse of the subject. The difference between a signal with and without a dominant frequency component in the pulse range is shown in figure 3.10 where the pulse peak is around 1.1 Hz. 1 1.5 2 2.5 Frequency [Hz] 0 5 10 15 20 Amplitude [V] 1 1.5 2 2.5 Frequency [Hz] 0 5 10 15 20 Amplitude [V] Frequency response
Figure 3.10: The frequency response of a signal divided into three phases corresponding to before,
during and after pressure. Left: A signal where all three parts have a dominating frequency component at 1.1 Hz. Right: A signal where two of the parts (blue and yellow, representing before and after) do not have a dominant frequency component in the pulse range.
In a first stage, the signals were manually investigated to evaluate which signals to remove. An algorithm to do the same thing was then developed. The algorithm first locates the max-imum value in the range of 0.67 - 1.67 Hz in the Fourier domain, which is assumed to be the pulse peak. To evaluate if the pulse peak is dominant in the range the root mean square (RMS) is calculated for the samples in a range of 0.15 Hz from the peak. Then the RMS is calculated for the other samples in the interval, to the left and to the right of the peak respectively, and compared to that of the peak. If the RMS for the pulse peak is 1.7 times larger than the RMS of both sides, the signal is classified as having a dominant frequency component.
3.3. Study 1
The specific values of the algorithm, the pulse peak width of 0.15 Hz and the factor 1.7 that the pulse RMS should be larger than the rest, were decided by using three different signals. First the values were set to ensure that no signal would be saved that manually were classified to be removed. Then the values were adjusted to removing as few of the other signals as possible. The algorithm was then tested on several of the other measurements to make sure it worked similarly on them, which it did.
Filter signals
Firstly, the signals are low-pass filtered with an 8th order Butterworth filter which passes normalized frequencies below 0.1, which corresponds to 1.67 Hz. The normalized frequency is defined as the frequency divided with half of the sampling frequency, also known as the Nyquist frequency. Then the signals are high-pass filtered with a 6th order Butterworth filter which passes normalized frequencies over 0.02, which corresponds to 0.67 Hz. Figure 3.11 shows an example of a filtered signal.
155 160 165 170 175 180 Time [s] 0.098 0.1 0.102 0.104 0.106 Amplitude [V] Original signal 155 160 165 170 175 180 Time [s] -6 -4 -2 0 2 4 6 8 Amplitude [V] 10-4 Filtered signal
Figure 3.11: A signal before and after filtration. The left plot is the original signal and the right plot
is the filtered signal.
3.3.3
Analysis of results
After the signal processing is preformed the goal is to determine the changes in blood volume pulsations as a response to applied pressure. The analysis is described in figure 3.8. As the top circle to the right in this figure describes, the analysis is made with the signal divided into the three different phases: a baseline before the pressure is applied, during pressure and after the pressure is released.
Find AC level
The amplitude of the variations in the signal is referred to as the AC level. The AC level is determined by finding the peaks and valleys in the signal as shown in figure 3.12. In order for a sample to be defined as a peak it has to fulfill these criteria: a minimum peak height of 20% of the mean of the absolute value of the filtered signal. To find the valleys of the signal the same criteria is applied to the inverted signal. After peaks and valleys are found, a sorting of these take place. The sorting removes all peaks that comes directly after another peak without a valley in between and vice versa for the valleys. After this sorting the distance in height between neighbouring peak and valley determine the AC level at that point in time.
3.3. Study 1 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 Sample 104 -1 -0.5 0 0.5 1 Amplitude [V]
10-3 Peak & Valley detection
Figure 3.12: The peak and valley detection of a part of the filtered signal, where all valleys and peaks
are marked with a purple dot.
with the DC, by dividing each point in the AC with the same point in the DC signal. The AC level and the scaled AC level is plotted in figure 3.13.
100 110 120 130 140 150 Time [s] -1 -0.5 0 0.5 1 1.5 2 2.5 Amplitude [V]
10-3 AC level, scaled AC level and theoretical fraction of pulsating blood
Filtered signal AC level Scaled AC level
Theoretical fraction of pulsating blood
Figure 3.13: The AC level, scaled AC level and theoretical fraction of pulsating blood plotted above
the filtered signal. Scaled AC level and theoretical fraction of pulsating blood have been additionally scaled to fit into the image.
3.3.4
Calculate the theoretical equivalent of the scaled AC level
As mentioned in section 2.4.3, equation 2.8, the fraction of pulsating blood is proportional to the natural logarithm of the intensity at diastole divided by the intensity at systole. This can be approximated to the AC level divided by the DC level. To see if this approximation is valid the theoretical value was calculated to compare it with the approximation. This was done by finding the peaks in the same manner as when finding the AC level and then extracting the diastole and systole values from the low-pass filtered signal at the peaks and valleys, respectively. When these values were found a result proportional to the fraction of pulsating blood was calculated using lnIdiastole
Isystole
. A normalized fraction of pulsating blood is shown in figure 3.13.
Compare AC levels, scaled AC levels and the theoretical fraction of pulsating blood
In order to compare the blood volume pulsation variations at the three phases, the ratios between the different phases are calculated for both the AC level, the scaled AC level and the theoretical fraction of pulsating blood. The first ratio is between the mean level of the baseline before pressure and the mean level when pressure is applied. If the ratio is above 1,
3.4. Development of a new sensor plate
the subject has an increase in blood volume pulsations and therefore a PIV response. If the ratio, on the other hand, is below 1, the subject lacks the PIV response and has a decrease in blood volume pulsations. The second ratio is between the mean level of baseline before pressure and the mean level after the pressure is released. If the ratio is above 1 during a period of time the subject may have an RH response because of the increase in blood volume. If the subject earlier had a PIV response a value above 1 may be caused by a prolonged PIV response.
3.4
Development of a new sensor plate
After the first study, the decision was made to develop a new sensor plate. The three main reasons for the development of a new plate were the difficulties in attaching the original plate, the number and the placement of the LEDs and detectors and the problems with using a conducting material around the electronics.
3.4.1
The attachment of the sensor plate
A large problem with the measurements is the difficulty to attach the sensor plate so that it closely follows the skin. The original sensor plate is quite large and the attachment would probably be easier if it was smaller, since a sensor plate covering a smaller area of the body would not have to be bent as much as a sensor plate covering a larger area. A smaller sensor plate would also be lighter and thereby not as weighed down by its own weight. The attach-ment could also be better if the sensor plate was more flexible, since it then would follow the contours of the body better.
If the relatively large size should be kept for the new sensor plate, other solutions would be required for making sure it is properly attached. One solution could be to use double sided tape, but the tape need to be skin friendly and able to adhere well to both the skin and the sensor material. The tape would also need to be thin to prevent reflections from the skin surface reaching the detectors directly.
The conclusive solution to the attachment problem was to develop a sensor plate which was as small as possible. To still get information from a larger area four of these sensor plates were used in study 2 and 3.
3.4.2
The placement of the LEDs and photodetectors
The original sensor plate have 18 LEDs, which is unnecessary because the same information can be produced with fewer LEDs and the 18 LEDs warm up the skin. The new sensor plate is developed with only one green and one red LED. Four photodetectors are placed in pairs with a distance of 6 mm and 8 mm from the middle of the source to the middle of the detector. The new plate is described in figure 3.14 and the size of it is 30x36x8 mm.
3.4. Development of a new sensor plate
3.4.3
Material
There are some requirements on the sensor plate material. The material needs to be isolating and have a high absorption of light. Preferably it should be flexible and as thin as possible to follow the contours of the body. The material in contact with the patient needs to be skin friendly. If there is a thin plastic to cover the plate it should adhere well to the sensor plate.
Circuit board
As the new plate was developed to be as small as possible a double-sided circuit board was used, with the LEDs and detectors on one side and all other electronics on the other side. This differed from the circuit board in the original sensor plate by being stiff instead of flexible and also being thicker than the original circuit board.
Casing
A casing to protect the circuit board was constructed. An insulating rubber material was cho-sen for the top of the casing, facing the skin of the subject. Holes for the LEDs and detectors were cut out. The rest of the casing was first printed in a hard plastic material and when assembled the plate had a thickness of 8 mm. Because of the small excess pressure that a thick and hard plate could induce opportunities to make the plate thinner and softer were investigated. One plate was assembled only using electrical tape as backside. There was approximately 1 mm of excess space inside the casing for the electronics, so the casing was made thinner and thereafter the backside was printed in a flexible plastic material.
To investigate if there was a difference between the three different backsides (made of tape, hard plastic and flexible plastic) some test measurements were made. Plates with the different backsides were placed on the back of the subject and their locations were marked with a pen. Then measurements were made in the same manner as in study 1, but with shorter duration. This was done three times while rotating the plates to the different locations in between the measurements. The conclusion made from these test measurements was that the placement of the plate affected the measurement to a greater extent than the backside. Since the difference between the backsides was negligible the flexible plastic was chosen for cleanliness and comfort reasons.
Cover material for the LEDs and detectors
The holes in the rubber top of the casing needed to be covered with something to enable dis-infection and to not leave the skin directly exposed to the electronics during measurements. In the original sensor plate this was done with a plastic sheet over the entire plate, which causes some reflections to reach from the LED to the detector without entering the skin. To avoid this it was decided to fill the holes with silicon instead. Two different two-component silicon for molding were tested, but there was an unexpected reaction with the rubber which spoiled the curing process. Finally a clear silicon on tube was tested, cured properly and chosen to cover the LEDs and detectors.
3.4.4
Wavelength of the LEDs
The original red LEDs have a wavelength of 625 nm. At this wavelength the absorption between oxygenated and deoxygenated blood differs quite a lot, see figure 2.5. This means that the signal would be affected by a change in oxygenation. Since there is a risk that the oxygenation changes between no pressure and with pressure, this introduces an uncertainty to the result. This uncertainty could be reduced by using a different wavelength, where the difference in absorption between oxygenated and deoxygenated blood is smaller. A suitable wavelength would be around 800-810 nm, at the isosbestic point where the absorption is
