The SafeBoosC trial showed that cerebral oximetry combined with a treatment guideline can reduce the the burden of hypoxia in neonates by 50% [Brit. Med. J. 350, g7635 (2015)]. However, guidelines based on oximetry by one oximeter are not directly usable by other oximeters. We made a blood-lipid phantom simulating the neonatal head to determine the relation between oxygenation values obtained by different oximeters. We calculated coefficients for easy conversion from one oximeter to the other. We additionally determined the corresponding SafeBoosC intervention thresholds at which we measured an uncertainty of up to 9.2% when varying hemoglobin content from 25μM to 70μM. In conclusion, this paper makes the comparison of absolute values obtained by different oximeters possible.
© 2016 Optical Society of America
In neonatology, infants born preterm are vulnerable to hypoxic and ischemic insults which lead to long-term disabilities. Diagnostic methods to early detect these conditions and prevent lesions are urgently needed. Optical methods such as near-infrared spectroscopy (NIRS) may be able to fulfill this need by assessing cerebral oxygenation non-invasively . NIRS and visible light spectroscopy (VLS) utilize light to non-invasively and continuously assess tissue oxygen saturation (StO2). Although there are a number of commercial NIRS oximeters available and entering the clinics, the evidence for clinical benefit is weak. A recent randomized clinical trial (SafeBoosC) carried out across Europe in 8 tertiary neonatal intensive care units including 166 extremely preterm infants, aimed to determine if it is possible to stabilize the cerebral oxygenation . Cerebral oximetry was combined with a dedicated treatment guideline , providing a clinical intervention algorithm to assist neonatologists in clinical decision making when StO2 was outside the target range of 55% – 85%. Cerebral oxygenation was acquired by INVOS 5100C with adult SomaSensor or NIRO-200NX with small reusable probe, within 3 hours from birth. These two oximeters had previously been tested for comparability in-vivo . The primary outcome measure of hypoxic and hyperoxic burden, was defined as the integral of the difference between StO2 and the threshold over time outside the target range of 55% – 85% and and compared between the NIRS-visible group and a control group where NIRS was recorded but the data was not available to the clinician. In the trial the hypoxic/hyperoxic burden was reduced by more than 50%, paving the way for a future phase III study to show whether optical spectroscopy prevents brain lesions in preterm infants in the future . However, despite substantial progress, several challenges need to be solved.
One problem is validation, i.e. do these oximeters measure StO2 accurately? How can this be tested? StO2 corresponds to the ratio of concentrations of oxy- and total hemoglobin (StO2 = cO2Hb/ctHb, whereby ctHb = cO2Hb + cHHb). NIRS measures the average concentrations of oxyhemoglobin (cO2Hb) and deoxyhemoglobin (cHHb) in the field of view of the light bundle and is sensitive to the hemoglobin in small blood vessels, i.e. arterioles, capillaries and venules.
A common approach is to validate StO2 measured by the oximeter against co-oximetry of arterial and venous (e.g. jugular) blood samples [5, 6], similar to the validation procedure for pulse oximeters. Thereby often a proportional contribution of 30% arterial and 70% venous blood to the StO2 is assumed. But this proportion cannot easily be measured and is likely to change over time and varies between tissues and subjects.  Moreover, venous samples truly representative of the optically measured tissue are difficult and risky to obtain. Jugular bulb represents global cerebral venous blood while the NIRS sensor samples only a small volume and it is known that brain oxygenation varies with location . Furthermore, jugular blood is possibly contaminated by extracerebral drainage  and 40% of the patients suffer from thrombosis after jugular bulb catherterization . Especially in preterm infants this procedure is not feasible for ethical reasons, which is probably true for adults as well. Therefore, it is impossible to establish the accuracy of the methods in-vivo from blood sample co-oximetry. At best these tests add plausibility .
Since there is no simple test so far and the specific StO2 value depends on several assumptions, it is not surprising that oximeters from different manufacturers provide different StO2 values as established by numerous studies [7, 11–18]. With the lack of a standard, the need arose to translate between different oximeters within and between trials (e.g. intervention thresholds in SafeBoosC). In-vivo vascular occlusions have previously been applied for this, because the induced ischemia allows characterization of the oximeter in a wide range of StO2 . This approach, however, is negatively affected by tissue in-homogeneity and physiological alteration over time.
To overcome this problem, oximeters can be characterized in-vitro in phantoms. This has the advantage of controllable optical properties with minimal variation, and can be constructed for the specific research question. Dye-based phantoms are mainly used for testing absorption measurements of oximeters, but are not suitable to assess oxygenation readings, as real hemoglobin is needed for this. Readings can be evaluated at multiple increments of StO2  or preferably in dynamic phantoms covering a continuous range of StO2 [20–25]. Suzuki et al used a liquid phantom when introducing the NIRO-300 comparing NIRS measurements with co-oximetry as reference . We have used a similar approach to directly compare oximeters as described previously [26–28]. Here we present an improved setup, where we minimized the vertical gradient in the oxygen content of the liquid phantom and improved the alignment of the oximeters which were simultaneously measuring oxygenation of the phantom. This ensures that the different oximeters measure the same oxygenation and absolute StO2 values can truly be compared. In addition we added a second layer to the measurement setup, which simulated the neonatal skull. This setup, thus, is closer to the structure of a neonatal head.
Aims of the current paper are to derive mathematical equations to convert StO2 measured by one oximeter to another one (INVOS adult/neonatal, Nonin neonatal, OxyPrem v1.3, OxiplexTS, and OxyVLS), to calculate oximeter-specific intervention thresholds for a phase III large scale follow-up trial of SafeBoosC, and to measure the inaccuracy of different oximeters due to variation of hemoglobin concentration (ctHb) at these thresholds. The influence of hemoglobin is expected from phantom and piglet studies demonstrating a change in continous wave NIRS measurements as hemoglobin concentration changes .
2. Methods and materials
In the presented phantom experiment we included 4 NIRS oximeters with several sensors, one visible light oximeter and conducted oximetry based on the hemoglobin-oxygen dissociation curve, all measuring StO2. We employed the following NIRS oximeters: OxiplexTS (ISS, Inc., Champaign, IL, USA), INVOS 5100C with Adult SomaSensor SAFB-SM and infant/neonatal OxyAlert NIRSensor CNN/SNN (Medtronic, Inc., Minneapolis, MN, USA), SenSmart Model X-100 Universal Oximetry System with adult 8004CA and non-adhesive neonatal/pediatric sensor 8004CB-NA (Nonin Medical, Inc., Plymouth, MN, USA) and OxyPrem v1.3 (in-house developed NIRS oximeter, University Hospital Zurich, Zurich, Switzerland).
As the intervention thresholds in SafeBoosC were determined from measurements on > 400 preterm infants using INVOS 5100C with the adult SomaSensor , this adult sensor is included in this study. INVOS and Nonin neonatal sensors as well as Oxyprem v1.3 are expected to be used in phase III of SafeBoosC and hence need to be characterized. The Nonin adult sensor was added to see how it compares to its neonatal sensor. The OxiplexTS and OxyVLS (in-house developed VLS oximeter, University Hospital Zurich, Zurich, Switzerland) were included to compare the performance of frequency domain NIRS and oximetry based on visible light spectroscopy to the performance of continuous wave NIRS oximeters, respectively.
2.1.1. NIRS oximeters
The INVOS system utilizes near-infrared light at 730 and 810nm, one source and two detectors at source-detector separations (SDS) of 30 and 40mm, providing two light paths. The INVOS 5100C provides ”real-time data accuracy” which is claimed to be what others call ”absolute” for certain clinical indications in patients > 2.5kg  . The oximeter is approved for clinical use.
SenSmart applies four wavelengths (730, 760, 810 and 870nm) and provides sensors for adults (SDS: 20 and 40mm) and neonates (SDS: 12.5 and 25mm). Both neonatal and adult sensors have two detector and two source locations, giving a total of four light paths providing absolute oxygen saturation for patients < 40kg (neonatal/pediatric sensor)  and > 40kg (adult sensor) . The oximeter is approved for clinical use.
In OxyPrem v1.3 we incorporated four wavelengths (690, 760, 805 and 830nm). It employs a self-calibrating principle  with 8 different light paths derived from two detectors and 4 different SDS (15, 20, 30 and 35mm). By applying this principle, the measured light intensity is independent of the sensitivity of the photo-detectors, light intensity at the source, light coupling factors and the influence of superficial tissue is reduced. The oximeter is not CE-marked but it has passed approval by medical device agencies on several occasions to acquire absolute StO2 in humans for clinical trials.
OxiplexTS is a frequency domain NIRS oximeter with two wavelengths of 692 and 834nm and a probe with SDS of 25, 30, 35 and 40mm and 4 light paths. The oximeter modulates light at an RF frequency of 110MHz which enables measuring absolute absorption coefficient (μa) and reduced scattering coefficient (μ′s) values. Subtraction of background absorbers and water content can be adjusted freely to account for different properties of tissues before calculation of cO2Hb and cHHb. The oximeter is CE-marked for research purposes and acquires absolute StO2 .
2.1.2. Oximetry based on the hemoglobin-oxygen dissociation curve (StO2 derived from pO2)
We calculated StO2 derived from pO2 based on a mathematical model assuming equilibrium binding of O2 with hemoglobin inside red blood cells. This model has the form of a Hill type equation and enables invertible calculation of StO2 based on partial pressure of oxygen (pO2), partial pressure of carbon dioxide (pCO2), pH, temperature, and 2, 3-diphosphoglycerate (2, 3-DPG) concentration [36, 37]. To produce accurate results pO2, pCO2, pH, and temperature were measured and kept in a physiological range. We assumed 2, 3-DPG ≈ 0 as is typical for stored blood . The duration of each experiment was < 4hr, which excludes alterations in the amount of 2, 3-DPG occuring after 4hr due to changes in pH . Since small changes in pH and temperature create large shifts in the oxygen hemoglobin dissociation curve, we kept pH close to the physiologically normal (pHphys = 7.4) value by adding sodium bicarbonate buffer (SBB) and stabilized the temperature between 37 – 38°C by placing the phantom on a heating plate. We altered pO2 between 0 and ≥ 10kPa to cover the full range of StO2 (0 – 100%). Table 1 indicates the sensors employed to derive StO2 from pO2 (manufactured by PreSens -Precision Sensing GmbH, Regensburg, Germany and Metrohm AG, Herisau, Switzerland).
2.1.3. Oximetry based on visible light spectroscopy (OxyVLS)
We employed a Maya2000 Pro (Ocean Optics, Inc., Dunedin, FL, USA) spectrometer (≈ 500 – 930nm spectral range, resolution ≈ 0.2nm) combined with a tungsten halogen source (360 – 2400nm, 7W, Ocean Optics) and a 400μm reflection probe (Ocean Optics), with approximately 2mm distance between light emission and detection. Spectra were acquired every 12s. Based on VLS it is possible to determine the StO2 . Here we applied an improved method to measure StO2. The wavelength range, from 520nm to 600nm, is remarkable by oxyhemoglobin (O2Hb) having two peaks (λ = 542nm and λ = 577nm, data from ) and deoxyhemoglobin (HHb) having only one peak (λ = 556nm, data from ) . Calculating StO2 based on the distance between the peaks as described in  is prone to errors. In normal physiological conditions (70% < StO2 < 85%) 1nm error in measurement of the distance between the peaks creates an error of ≈ 5 – 10% in StO2, which is too unstable. Moreover, when StO2 < 30% no peak is detectable, which previously prevented measurement of StO2 < 30% . Here we apply the “Interval analysis” technique to the spectrum of the liquid phantom in the range from 520 to 600nm to calculate StO2. The details of this technique are depicted elsewhere . As depicted in Fig. 1 we calculated the interval between two data points with the same optical density in the range from λ = 520nm to λ = 600nm. By comparing it to the interval analysis signal of hemoglobin with known StO2 (resolution: 0.1% StO2, data from ), we calculated the StO2 of the liquid phantom. This method led to more stable results and enabled measurement of StO2 < 30%. Figure 2 shows the interval signal calculated for O2Hb, i.e. StO2 = 100%.
2.2. Measurement set-up and liquid phantom
We prepared two liquid phantoms in a phantom container. The set-up is described below.
2.2.1. Phantom container
The phantom container was built in-house. We deployed the CAD tool NX (Siemens PLM Software) to design the phantom container and fabricated it from ABS plastic with a 3D-printer. The inside of the container was covered with bio-compatible epoxy to avoid cytotoxic effects. The phantom container was an irregular octagonal shape with four wide and four narrow side faces as is shown in Fig. 3 and 4. In these figures, the windows are not actually displayed, just the openings for them in the rigid container structure. This geometry enabled placing 4 NIRS oximeters on each wide side of the container. The phantom container had the following features:
Semi-infinite boundaries The side faces of the container and holders pressing the sensors of the oximeters against the windows were built of absorbing material, thus effectively implementing a semi-infinite boundary condition. It also shielded the phantom from ambient light.
No ambient oxygen diffusion The cap of the container completely isolated the phantom from the ambient air. All the openings of the cap for placement of pO2, pCO2, pH, and temperature sensors as well as OxyVLS were sealed with silicone gaskets. Prior to measurement, we scanned the height of the phantom with a pO2 sensor.
High homogeneity and controlled temperature We applied a magnetic stirrer, whose speed was set to 500rpm throughout the measurement, to effectively ensure homogeneity. If a gradient in the phantom exists, then it is due to O2 diffusion from air through the phantom surface. With air on top of the liquid containing yeast, we measured a vertical gradient of less than 40Pa/cm which was negligible. Still we aligned all the sensors at the same height (< ±0.5cm) with their light path being horizontal to exclude the effect of any remaining vertical oxygen gradient. The bottom part of the phantom container constructed from copper enabled to place the phantom container on the hot plate of the stirrer, heating the liquid and thus keeping the phantom temperature constant.
Optical cross-talk The phantom container enabled simultaneous measurement with 4 different NIRS oximeters. The minimum distance between the two closest optodes of two different sensors was 7.5cm, i.e. at least 3.5cm larger than the largest inter-optode distance of any sensor. This effectively prevented cross-talk.
Windows simulating optical properties of the neonatal head We cast 4 windows that reflected the optical properties of skull  with values of μa = 0.10 cm−1 and μ′s = 9.6 cm−1 at 692nm and μa = 0.11 cm−1 and μ′s = 8.3 cm−1 at 834nm, which we expect to be a close approximation to the actual optical properties of the neonatal head. The windows were made from Silpuran 2420 silicone and were colored with 1.15ml/L Elastosil pigment paste FL RAL 9010 (white) (both Wacker Chemie AG, Munich, Germany) and 2.26mg/L carbon black powder (Alfa Aesar, Thermo Fisher (Kandel) GmbH, Karlsruhe, Germany). These windows were placed on each wide face of the container and served as the interface between the NIRS oximeters and the liquid phantom. The thickness of the windows was 2.5mm which approximately reflected the thickness of skull in neonates .
All NIRS oximeters employed in this study claim to have algorithms which reduce the influence of superficial layers. Nonin neonatal as the oximeter with shortest source-detector separation (SDS) in our experiment has a penetration depth of 12.5mm  which is 5 times the thickness of the windows. Based on this, we expect the windows to have only marginal influence on StO2 readings. We nevertheless tried to make them similar to reality. It is obvious that influence of more superficial layers can more effectively be reduced. We therefore neglected the well perfused skin containing hemoglobin. The remaining layers, skull and cerebrospinal fluid, which contain very little hemoglobin are well represented by our single silicone layer which resembles data obtained in-vivo only showing two layers: outer and inner (brain) .We thus expect the phantom to be a good approximation for the neonatal head. Although the skull is per-fused, its hemoglobin content is much lower compared to the brain itself. As a result windows with no blood are still a good estimation of the reality. In the future, the measurement set-up could possibly be further improved: Developing windows which simulate the whole spectrum of hemoglobin by means of mixing blood with the windows seems possible  and might be even a closer approximation to the reality. But stability of blood cells during the production procedure of the windows and long term stability would first have to be verified.
2.2.2. Oxygen supply
A tube from an industrial oxygen tank (O2 ≥ 99.5%, PanGas AG, Dagmersellen, Switzerland) through a flow-meter was immersed to approximately 2cm above the bottom of the container in order to provide oxygen to the phantom when needed.
2.2.3. Liquid phantom
The liquid phantom consisted of phosphate-buffered saline (PBS, after Kreis, pH = 7.4, Kantonsapotheke Zurich, Zurich, Switzerland), human blood from expired human erythrocyte concentrate bags (expiry date < 2 months, total hemoglobin concentration: tHb = 220g/L, hematocrite: htc = 67%), Intralipid 20% solution (IL) (Fresenius Kabi AG, Bad Homburg, Germany), sodium bicarbonate buffer 8.4% (1mmol/ml) (SBB) (B. Braun Medical AG, Sempach, Switzerland), and glucose 50% (AlleMan Pharma GmbH, Reutlingen, Germany). Fresh baker’s yeast was added, when needed, to deoxygenate the hemoglobin. The optical properties of the liquid phantom resembled that of the neonatal brain [45, 46]. We aimed at μ′s of 5.5 (cm−1) and obtained an average μ′s of 6.8 cm−1 (692nm) and 5.4 cm−1 (834nm) (measured by OxiplexTS). Since ctHb is highly variable between neonates, we introduced three mixtures with different levels of ctHb for each phantom: 25, 45, and 70μM. Table 2 indicates the ingredients of the liquid phantom. The phantom contained ≈ 98% water which is only slightly more than up to 95% reported for neonatal brain tissue . The IL content and hence the scattering was not changed. Theoretically this cancels out when calculating StO2 out of cO2Hb and ctHb with only the wavelength dependency remaining . A previous phantom study confirmed this .
For liquid phantoms containing hemoglobin there are two possibilities to reversibly deoxygenate the hemoglobin. The first way is via gas-exchange, either by N2 in-flow or by a membrane oxygenator . The second way is by adding small quantities of respiring yeast into the phantom [20, 27, 28]. Deoxygenating by N2 in-flow is relatively slow for large phantom volumes and also creates in-homogeneity which may invalidate the results when lowering StO2. This method, however, works if the StO2 of the phantom is fixed at a certain level but this fixation requires a sophisticated set-up which prevents any O2 diffusion from the ambient air into the phantom. The inhomogeneity problem is not solved by using a membrane oxygenator because at the inlet, blood with a different level of oxygenation than that of the bulk phantom enters and is a source of inhomogeneity. A simpler alternative for deoxygenating hemoglobin is to add yeast. Yeast in the phantom, if stirred well, causes distributed oxygen consumption and therefore prevents inhomogeneity. This method is also much faster and therefore we decided to deoxygenate the phantom by adding yeast. For oxygenating, however, we used O2 in-flow which might have created inhomogeneity but this is not an issue as we only used the data while deoxygenating the phantom.
2.3. Measurement protocol
We prepared two liquid phantoms with three mixtures each (table 2). Phantom 1 and phantom 2 had the same ingredients (except from the amount of yeast). pO2, pCO2, pH, and temperature sensors, OxyVLS, OxyPrem v1.3 and OxiplexTS were employed in both phantoms. Additionally, Nonin adult and INVOS adult oximeters were employed in phantom 1, whereas Nonin neonatal and INVOS neonatal were employed in phantom 2. We started deoxygenating the hemoglobin by adding yeast and re-oxygenated it by providing O2 into the phantom. The measurement started at pH = 7.4. During the measurement, pH gradually decreased due to CO2 accumulation in the phantom (produced by yeast). In order to keep the pH close to pHphys = 7.4, we gave initially 15ml and additionally two times 10mL SBB to each phantom. We added more glucose (+3mL) to both phantoms when adding more blood (leading to the next mixture). Table 3 summarizes the procedure for phantom 1, and phantom 2. To increase the speed of deoxygenation, we added more yeast (+1.5g) for mixture no. 3 in phantom 1.
We constantly monitored pO2, pCO2, pH, and temperature during the course of the measurement. Table 4 shows the range of pO2, pCO2, pH, and temperature during the measurement for phantom 1 and phantom 2.
2.4. Data processing
We applied a moving average filter over 3 samples on OxyVLS data. For all other oximeters, raw StO2 values were recorded. For INVOS, Nonin, OxyPrem v1.3 and OxiplexTS oximeters we placed several event markers throughout the measurements which were used for alignment of the different time-series. We re-sampled the data from all devices to (sampling rate) and created time-series for all devices on a common time base.We inspected the data visually thereafter to find the synchronization precision. This precision was higher than one sample. This means that the maximum time lag that could occur between samples in repeated measurements was 12s. Obvious artifacts were removed, i.e data with saturated detector for the OxiplexTS. We applied 1st degree polynomial fits (based on lowest least square error) to calculate the relation between StO2 measured by different oximeters during deoxygenations in the range of 16% ≤ StO2 ≤ 94%.
Figures 5 and 6 show the StO2 time series of the oximeters in phantom 1 and phantom 2, respectively. We increased ctHb step-wise: ctHb = 25μM, 45μM and 70μM (table 2) as depicted in Fig. 5 and 6 by the intensity of background red color. The changes in oxygenation are visible.
Figures 7–11 show in scatter plots how each individual oximeter compared to OxiplexTS at ctHb = 25μM, 45μM and 70μM. OxiplexTS was chosen as reference because the StO2 derived from pO2 shows a drift in phantom 2 (Fig. 6) and because it is commercially available (in contrast to OxyVLS). The choice of reference is discussed in detail in section 4.1. Equations of the linear fits and coefficients of determination (R2) are given in the figure captions (StO2,device = a * StO2,OxiplexTS + b). Only data-points while deoxygenating (no oxygen flow) within the gray rectangle (16 ≤ StO2 ≤ 94%) were included for fitting. A dark gray polygon is defined by the fitting lines for 25μM and 70μM. The area of this polygon is different for all oximeters and qualitatively reflects dependence of the StO2 reading on ctHb. For OxyPrem v1.3 and OxyVLS there are more data-points, because these sensors were present in both phantoms. For oximeters from which data of more than one de-oxygenation was available, all available data was used to generate the fitting lines.
We do not report the results obtained from Nonin adult sensor, because we suspect that the sensor was not properly attached to the window of the phantom and data was implausible.
3.1. StO2 conversion table and SafeBoosC intervention threshold
In table 5 we show coefficients for conversion of values recorded by one oximeter to the other (at ctHb = 45μM). For the SafeBoosC and other trials, intervention thresholds were applied to study different interventions. Here it is important to take into consideration the difference in values that oximeters display for a specific StO2. E.g. in the SafeBoosC trial the hypoxic threshold was 55% and the hyperoxic 85%, based on values measured by the INVOS adult oximeter . Table 6 presents what these thresholds correspond to for other NIRS oximeters. The table also contains the uncertainty range of StO2 readings due to variations in ctHb at these thresholds. Figure 12 displays for ctHb = 45μM the StO2 measured by all oximeters compared to the OxiplexTS.
As depicted in Fig. 5 and 6, OxiplexTS, OxyPrem v1.3, Nonin neonatal, INVOS adult and neonatal, StO2 derived from pO2 and OxyVLS responded consistently and linearly correlated to the induced changes in StO2. Rise and decline of StO2 occurred simultaneously, but different oximeters showed different absolute values, dynamic ranges and sensitivities. In phantom 1 curves became narrower because we initially kept the phantom at high StO2 for 20min before adding yeast for the first time. For the second upper plateau in mixture no. 1, we oxygenated the phantom to ≈ 20kPa. The upper plateaus in mixture no. 2 and 3 were oxygenated to only 11kPa. The deoxygenation at mixture no. 3 was faster than the previous ones because the additional yeast (1.5g) outweighed the increased oxygen capacity of the phantom by increased ctHb. In phantom 2 the experiment started at pO2 = 18kPa and we reoxygenated the phantom to 15kPa for mixture no. 2 and 3. The curves become broader because of the increased oxygen capacity by higher ctHb.
4.1. Reference StO2
In several previous phantom experiments co-oximetry has been applied, on samples taken from the phantom, to compare the output of the oximeters to. This is a straightforward procedure for experimental set-ups where ctHb is similar to that of human blood  . Inspired by Suzuki et al.  we tried to use co-oximetry by an ABL800 blood gas analyzer (Radiometer Medical ApS, Brønshøj, Denmark) as reference in a previous study  . This failed due to too low ctHb and too much turbidity (caused by Intralipid) in the sample. Centrifugation of the phantom sample yielded acceptable CtHb, but the saturation values were not trustworthy as contamination by ambient air oxygen was probable.
Instead, in this study we attempted to do oximetry utilizing the hemoglobin-oxygen dissociation curve. We tried to obtain a reference StO2 from pO2, pCO2, pH, temperature and 2, 3-DPG but it was only partially successful. In phantom 1 (Fig. 5) high correlation (0.999%) to OxiplexTS was observed. But in phantom 2 (Fig. 6), there was a large deviation between this StO2 derived from pO2 and StO2 from OxiplexTS, which increased over time. Later we identified the pH sensor as the cause of the problem with drifts of up to ΔpH = ±0.2 when immersed repeatedly into buffer solutions. As pH plays a significant role, we believe that this created the observed drift between the StO2 derived from pO2 and OxiplexTS and OxyVLS values. Moreover, since a correct StO2 measurement employing this approach is dependent on a precise measurement of pO2, pCO2, pH, temperature, and a correct assumption for 2, 3-DPG, we conclude that this approach is not practical for phantom measurements.
The second option to be set as the reference was OxyVLS. But OxyVLS is a built in-house oximeter which is still under development. It is also still not available in the market or for other research institutions. As a result, if it was chosen as the reference, reproducing the results by others would not have been possible.
OxiplexTS was adjusted to subtract the known background absorption of the phantom (98% water) before chromophore calculation and therefore did not show any dependence on ctHb. It was in a good agreement with OxyVLS (correlation coefficient: 0.997), which inherently measures independent of ctHb. In addition it was in good agreement with StO2 derived from pO2 in the first phantom, provides lower-noise StO2 than OxyVLS at high sampling rate and more robust measurements than StO2 derived from pO2. Because of these reasons OxiplexTS was the best option and we set it as the reference for comparison of oximeters in this paper.
4.2. Plateaus, dynamic range, and sensitivity of the oximeters
The upper StO2 plateau was different for all oximeters but it was unaffected by ctHb. In Fig. 5 it seems that the upper plateau of StO2 measured by OxyVLS decreased over time. One reason for this may be in the second upper plateau in mixture no. 1 we oxygenated the phantom to approx 20kPa. The upper plateaus in mixture no. 2 and 3 were oxygenated to only 11kPa. Additionally we added 1.5g more yeast to mixture no. 3. For these reasons we believe the phantom in mixture no. 2 and mixture no. 3 was not at its maximum oxygenation state before deoxygenation was started. Moreover, it may also be explained by the presence of cytochrome C in the medium, which was not included in the analysis and which changes the shape of the reflection spectrum of the turbid phantom, compared to the spectrum of pure hemoglobin in the range from 520nm to 600nm . Yeast contains cytochrome C and was added in two steps . In Fig. 6 this decrease was not observable, because the phantom was always oxygenated to 15kPa and the amount of yeast was also constant throughout the measurement. For all NIRS oximeters there were only marginal changes observable in the upper plateau.
The lower plateau of StO2 (Fig. 5 and 6), in contrast, showed a dependence on ctHb level for all NIRS oximeters. The lower plateau of Oxyprem v1.3, INVOS adult, neonatal, and Nonin neonatal decreased as we increased ctHb from 25μM to 45μM. The change was less pronounced when we increased ctHb from 45μM to 70μM. Accordingly the dynamic range increased. The dynamic range of INVOS oximeters may appear less influenced by the ctHb change, but this is due to the oximeter not displaying values higher than 95% or lower than 15% StO2 (clipping).
In Fig. 7–11 the lowest R2 of the linear fits was > 0.984 which in our opinion is sufficient for comparison among oximeters. The maximum error due to linear regression in the fitting range is less than 4.5% with the highest values at the lower end of the StO2 range which is not relevant for clinical decision-making. Within the range 55 < StO2 < 85% considered normal in SafeBoosC the data points are very well represented by linear regression.
In Fig. 7–11 the slope of each linear fit corresponds to the sensitivity of the oximeter to oxygenation changes with INVOS neonatal being the most and Nonin neonatal being the least sensitive at ctHb = 45μM. There was a pronounced change in sensitivity for ctHb = 25μM compared to 70μM for Nonin neonatal (82%), INVOS adult (51%), and INVOS neonatal (64%), while OxyPrem v1.3 (14%) and OxyVLS (−4%) were much less dependent on ctHb. Such an effect of ctHb, although smaller (INVOS adult: 35%), was also observed in a previous study . The reason for this difference is that in the present study, ctHb was much smaller (phantom htc = 0.52% to 1.36% (corresponding to ctHb = 25μM and 70μM) in the present study compared to htc = 1% to 2% in the previous study). Thus, in the previous study, the influence of background absorbers was smaller . This explains the higher dependence of sensitivity of oximeters on the ctHb level in the current data.
This effect of ctHb on sensitivity, dynamic range, and lower plateaus depends on the technical specifications of the oximeters, such as how many and which wavelengths they incorporate, how other absorbers besides O2Hb and HHb are being treated and if and how the wide spectra of LEDs compared to lasers have been handled. The number of wavelengths and the peak emission wavelengths of the oximeters are often reported by manufacturers, but the latter two points are not publicly available for most oximeters. Generally, the more wavelengths that are incorporated, the more precise the results will be. OxyPrem v1.3 incorporates 4 wavelengths which partially explains its decreased variability to changes in ctHb compared to INVOS sensors having 2 wavelengths only. However, variability of OxyPrem v1.3 is still less than that of the Nonin neonatal employing 4 wavelengths as well. The reason for that may be the contribution of the background absorbers which we reduced in OxyPrem v1.3. When ctHb and hence absorption caused by cO2Hb and cHHb is low, e.g. in neonates, the relative contribution of such background absorbers is large and not negligible, as assumed by many oximeters. For the current phantom (table 2), lipids contributed ≈ 0.5% (being negligible) and water ≈ 98% of the volume. Except for a local absorption minimum of water at 810nm, both, O2Hb and water generally show an increase of absorption in the range 680nm < λ < 920nm. Lipid has low absorption in the wavelength range 600nm ≤ λ ≤ 870nm. Above this wavelength, absorption increases rapidly. Therefore, depending on wavelengths employed, lipid might be relevant in human tissue with higher lipid contents and both, water and lipids, can possibly be mistaken as O2Hb by different oximeters. The result is an cO2Hb offset even when StO2 = 0%. Since this error is wavelength dependent, the choice of wavelengths is crucial in reducing this effect. In OxyPrem v1.3 wavelengths of 690, 760, 805 and 830nm were selected based on simulations in which the wide emission spectrum of the LEDs were modeled by Gaussian distributions and with the aim to reduce the influence of water and lipid to a minimum. Figure 8 shows that we achieved this aim and OxyPrem v1.3 measures StO2 with negligible variation over the wide range of neonatal brain ctHb levels because of the considerations mentioned above.
4.3. Reproducibility of the procedure
One of the aims of this paper was to provide mathematical equations to convert the results obtained by one oximeter to the results of the other oximeters (table 5). This will be valid only if the experiment is reproducible between the two phantoms. Oxyprem v1.3 and OxyVLS were present in both phantoms. In the first phantom we deoxygenated the phantom twice with the same amount of ctHb. This led to two datasets with ctHb = 25μM in phantom 1. In addition we have a dataset in phantom 2. We applied individual linear fits to all three datasets in the range 16 ≤ StO2 ≤ 94%. The standard deviations for slope (relative) and offset (absolute) coefficients were 0.89% and 0.86% for OxyPrem v1.3 while 2.28% and 2.17% for OxyVLS.
In dynamic measurements the correct alignment of the time-series influences the results of the comparison. We therefore aligned the data based on event markers and afterwards visually inspected the time-series. In the presented data, the highest rate of deoxygenation was −8.3% per minute at ctHb = 25μM in phantom 2. This means it takes 8min until hemoglobin reaches from StO2 = 94% to StO2 = 16%. If we assume that alignment is off by one sample (12s), then this corresponds to a worst case error of 8.3%/60s * 12s = 1.66% which is well below the reproducibility of NIRS measurements in-vivo as reported in the literature [4, 7] and we consider this to be acceptable. We therefore conclude that the method was indeed reproducible.
4.4. Comparability of oximeters
The equations for conversion of StO2 from one oximeter to the others assuming a typical neonate with ctHb = 45μM (table 5) enable researchers to quantitatively relate their own findings to data in the literature obtained by other oximeters. This means that clinicians do not have to wait until a specific observation has been made and reported with the same oximeter available to them. Especially for small patient groups such as preterm neonates this allows quicker implementation of new knowledge into practice.
4.5. SafeboosC intervention threshold
In table 6 we calculated oximeter-specific intervention thresholds for the clinical trial SafeBoosC corresponding to StO2 = 55% and 85% as measured by the INVOS adult oximeter for a typical neonate with ctHb = 45μM. Our results show that the same oxygenation leads to quite different StO2 values and intervention thresholds need to be adjusted to the specific sensor and oximeter accordingly. e.g. the hypoxic threshold has to be 11% higher in case of the Nonin neonatal oximeter compared to the INVOS adult. OxyVLS has an SDS ≈ 2mm thus it is not suitable for brain monitoring. For this reason it is not included in table 6.
4.6. Implications of ctHb and background absorbers
A recent study reported that the variation of water content between infants leads to an uncertainty of StO2 readings of up to 8% , which is also supported by [27, 29]. This may systematically flaw clinical decisions. Variation of water content has the same effect on the relative contribution of background absorbers as of variation of ctHb as addressed above. The current experiment revealed a ctHb dependence at the SafeBoosC intervention thresholds (table 6). At the hypoxic threshold StO2, INVOS adult and neonatal (9.2%) showed the largest and OxyPrem v1.3 (1.9%) the lowest dependence on ctHb, i.e. uncertainty range of StO2 readings.
In a trial with intervention thresholds, this means that oximeters with a high uncertainty effectively apply different StO2 thresholds to patients with low ctHb level than for patients with high ctHb level. We expect that the reliability of such studies will be improved by oximeters providing more robust StO2 readings, because they result in a more homogeneous threshold for the patient population.
In this paper we confirmed that different oximeters measured different StO2 values on the same phantom simulating the neonatal head. We provided mathematical equations which translate data obtained from one oximeter to the others. Moreover, we have calculated the effect this has on intervention thresholds. We have additionally measured the dependence of StO2 values on the ctHb, which varies substantially between oximeters.
The authors would like to thank Ranjan K. Dash (Medical College of Wisconsin, Department of Physiology) for valuable explanations to his SHbO2 model and for extending it to the case of several variables deviating from normal values at the same time . The presented work was funded by The Danish Council for Strategic Research (grant number 00603-00482B) and the Nano-Tera projects ParaTex, ObeSense and NewbornCare and the Clinical Research Priority Program Tumor Oxygenation of the University of Zurich.
References and links
1. J. J. Volpe, “Brain injury in premature infants: a complex amalgam of destructive and developmental disturbances,” Lancet Neurol. 8, 110–124 (2009). [CrossRef]
2. S. Hyttel-Sorensen, A. Pellicer, T. Alderliesten, T. Austin, F. van Bel, M. Benders, O. Claris, E. Dempsey, A. R. Franz, M. Fumagalli, C. Gluud, B. Grevstad, C. Hagmann, P. Lemmers, W. van Oeveren, G. Pichler, A. M. Plomgaard, J. Riera, L. Sanchez, P. Winkel, M. Wolf, and G. Greisen, “Cerebral near infrared spectroscopy oximetry in extremely preterm infants: Phase II randomised clinical trial,” Brit. Med. J. 350, g7635 (2015). [CrossRef] [PubMed]
3. A. Pellicer, G. Greisen, M. Benders, O. Claris, E. Dempsey, M. Fumagally, C. Gluud, C. Hagmann, L. Hellstrm-Westas, S. Hyttel-Sorensen, P. Lemmers, G. Naulaers, G. Pichler, C. Roll, F. van Bel, W. van Oeveren, M. Skoog, M. Wolf, and T. Austin, “The SafeBoosC phase II randomised clinical trial: a treatment guideline for targeted near-infrared-derived cerebral tissue oxygenation versus standard treatment in extremely preterm infants,” Neonatology 104, 171–178 (2013). [CrossRef]
4. S. Hyttel-Sorensen, L. C. Sorensen, J. Riera, and G. Greisen, “Tissue oximetry: a comparison of mean values of regional tissue saturation, reproducibility and dynamic range of four NIRS-instruments on the human forearm,” Biomed. Opt. Express 2, 3047–3057 (2011). [CrossRef] [PubMed]
5. N. Nagdyman, P. Ewert, B. Peters, O. Miera, T. Fleck, and F. Berger, “Comparison of different near-infrared spectroscopic cerebral oxygenation indices with central venous and jugular venous oxygenation saturation in children,” Paediatr. Anaesth. 18, 160–166 (2008). [PubMed]
6. R. N. Kreeger, C. Ramamoorthy, S. C. Nicolson, W. A. Ames, R. Hirsch, L. F. Peng, A. C. Glatz, K. D. Hill, J. Hoffman, J. Tomasson, and C. D. Kurth, “Evaluation of pediatric near-infrared cerebral oximeter for cardiac disease,” Ann. Thorac. Surg. 94, 1527–1533 (2012). [CrossRef] [PubMed]
7. M. Wolf, G. Naulaers, F. van Bel, S. Kleiser, and G. Greisen, “A review of near infrared spectroscopy for term and preterm newborns,” J. Near Infrared Spec. 20, 43–55 (2012). [CrossRef]
8. B. Meyer, C. Schaller, C. Frenkel, B. Ebeling, and J. Schramm, “Distributions of local oxygen saturation and its response to changes of mean arterial blood pressure in the cerebral cortex adjacent to arteriovenous malformations,” Stroke 30, 2623–2630 (1999). [CrossRef] [PubMed]
10. W. M. Coplin, G. E. O’Keefe, M. S. Grady, G. A. Grant, K. S. March, H. R. Winn, and A. M. Lam, “Thrombotic, infectious, and procedural complications of the jugular bulb catheter in the intensive care unit,” Neurosurgery 41, 101–107 (1997). [CrossRef] [PubMed]
11. A. Dullenkopf, B. Frey, O. Baenziger, A. Gerber, and M. Weiss, “Measurement of cerebral oxygenation state in anaesthetized children using the INVOS 5100 cerebral oximeter,” Pediatr. Anesth. 13, 384–391 (2003). [CrossRef]
12. L. C. Sorensen and G. Greisen, “Precision of measurement of cerebral tissue oxygenation index using near-infrared spectroscopy in preterm neonates,” J. Biomed. Opt. 11, 054005 (2006). [CrossRef] [PubMed]
13. C. Jenny, M. Biallas, I. Trajkovic, J.-C. Fauchre, H. U. Bucher, and M. Wolf, “Reproducibility of cerebral tissue oxygen saturation measurements by near-infrared spectroscopy in newborn infants,” J. Biomed. Opt. 16, 097004 (2011). [CrossRef] [PubMed]
14. M. Pocivalnik, G. Pichler, H. Zotter, N. Tax, W. Mueller, and B. Urlesberger, “Regional tissue oxygen saturation: comparability and reproducibility of different devices,” J. Biomed. Opt. 16, 057004 (2011). [CrossRef] [PubMed]
15. L. M. Dix, F. van Bel, W. Baerts, and P. M. Lemmers, “Comparing near-infrared spectroscopy devices and their sensors for monitoring regional cerebral oxygen saturation in the neonate,” Pediatr. Res. 74, 557–563 (2013). [CrossRef] [PubMed]
16. T. Szczapa, L. Karpinski, J. Moczko, M. Weindling, A. Kornacka, K. Wroblewska, A. Adamczak, A. Jopek, K. Chojnacka, and J. Gadzinowski, “Comparison of cerebral tissue oxygenation values in full term and preterm newborns by the simultaneous use of two near-infrared spectroscopy devices: an absolute and a relative trending oximeter,” J. Biomed. Opt. 18, 087006 (2013). [CrossRef]
17. T. W. Hessel, S. Hyttel-Sorensen, and G. Greisen, “Cerebral oxygenation after birth - a comparison of INVOS and FORE-SIGHT near-infrared spectroscopy oximeters,” Acta Paediatr. 103, 488–493 (2014). [CrossRef] [PubMed]
18. A. Schneider, B. Minnich, E. Hofstaetter, C. Weisser, E. Hattinger-Jurgenssen, and M. Wald, “Comparison of four near-infrared spectroscopy devices shows that they are only suitable for monitoring cerebral oxygenation trends in preterm infants,” Acta Paediatr. 103, 934–938 (2014). [CrossRef] [PubMed]
19. A. M. De Grand, S. J. Lomnes, D. S. Lee, M. Pietrzykowski, S. Ohnishi, T. G. Morgan, A. Gogbashian, R. G. Laurence, and J. V. Frangioni, “Tissue-like phantoms for near-infrared fluorescence imaging system assessment and the training of surgeons,” J. Biomed. Opt. 11, 014007 (2006). [CrossRef] [PubMed]
20. S. Suzuki, S. Takasaki, T. Ozaki, and Y. Kobayashi, “Tissue oxygenation monitor using NIR spatially resolved spectroscopy,” Proc. SPIE 3597, 582–592 (1999). [CrossRef]
22. J. G. Kim and H. Liu, “Variation of haemoglobin extinction coefficients can cause errors in the determination of haemoglobin concentration measured by near-infrared spectroscopy,” Phys. Med. Biol. 52, 6295–6322 (2007). [CrossRef] [PubMed]
23. E. L. Hull, M. G. Nichols, and T. H. Foster, “Quantitative broadband near-infrared spectroscopy of tissue-simulating phantoms containing erythrocytes,” Phys. Med. Biol. 43, 3381–3404 (1998). [CrossRef] [PubMed]
25. J. Kraitl, U. Timm, and H. Ewald, “Non-invasive measurement of blood and tissue parameters based on vis-nir spectroscopy,” Proc. SPIE 8591, 859105 (2013). [CrossRef]
26. S. Hyttel-Sorensen, S. Kleiser, M. Wolf, and G. Greisen, “Calibration of a prototype NIRS oximeter against two commercial devices on a blood-lipid phantom,” Biomed. Opt. Express 4, 1662–1672 (2013). [CrossRef] [PubMed]
27. S. Kleiser, S. Hyttel-Sorensen, G. Greisen, and M. Wolf, “Comparison of near-infrared oximeters in a liquid optical phantom with varying intralipid and blood content,” Adv. Exp. Med. Biol. 876, 413–418 (2016). [CrossRef] [PubMed]
28. N. Nasseri, S. Kleiser, S. Reidt, and M. Wolf, “Local measurement of flap oxygen saturation; an application of visible light spectroscopy,” Adv. Exp. Med. Biol. 876, 391–397 (2016). [CrossRef]
29. C. Kurth and B. Uher, “Cerebral hemoglobin and optical pathlength influence near-infrared spectroscopy measurement of cerebral oxygen saturation,” Anesth. Analg. 84, 1297–1305 (1997). [PubMed]
30. S. Hyttel-Sorensen, T. Austin, F. van Bel, M. Benders, O. Claris, E. M. Dempsey, M. Fumagalli, C. Gluud, C. Hagmann, L. Hellstrom-Westas, P. Lemmers, G. Naulaers, W. van Oeveren, A. Pellicer, G. Pichler, C. Roll, L. S. Stoy, M. Wolf, and G. Greisen, “Clinical use of cerebral oximetry in extremely preterm infants is feasible,” Dan. Med. J. 60, A4533 (2013). [PubMed]
31. “The INVOS system improving patient outcomes through cerebral/somatic oximetry,” Tech. Rep. 11-PM-0256 MN21010, Medtronic, Inc., Minneapolis, MN, USA (2011).
32. “Model 8004CA, Instructions for use - english,” Tech. Rep. 9539-001-02, Nonin Medical, Inc., 13700 1st Avenue North Plymouth, MN, 55441–55443, USA (2014).
33. “Model 8004CB-NA, Instructions for use - english,” Tech. Rep. 9223-001-02, Nonin Medical, Inc., 13700 1st Avenue North Plymouth, MN, 55441–55443, USA (2014).
34. D. Hueber, S. Fantini, A. Cerussi, and B. Beniamino, “New optical probe designs for absolute (self-calibrating) NIR tissue hemoglobin measurements,” Proc. SPIE 3597, 618–631 (1999). [CrossRef]
35. “Oxiplexts near infrared, non-invasive, tissue spectrometer,” Tech. rep., ISS, Inc., 1602 Newton Drive Champaign, Illinois 61822, USA (2001).
36. R. Dash and J. Bassingthwaighte, “Erratum to: Blood HbO2 and HbCO2 dissociation curves at varied O2, CO2, pH, 2, 3 – DPG and temperature levels,” Ann. Biomed. Eng. 38, 1683–1701 (2010). [CrossRef] [PubMed]
37. R. Dash, B. Korman, and J. Bassingthwaighte, “Simple accurate mathematical models of blood HbO2 and HbCO2 dissociation curves at varied physiological conditions - Evaluation and comparison with other models,” Eur. J. Appl. Physiol. 116, 97–113 (2016). [CrossRef]
38. T. Yoshida and S. S. Shevkoplyas, “Anaerobic storage of red blood cells,” Blood Transfus. 8, 220–236 (2010). [PubMed]
40. W. G. Zijlstra, A. Buursma, and O. W. van Assendelft, Visible and Near Infrared Absorption Spectra of Human and Animal Haemoglobin Determination and Application (VSP Utrecht, 2000).
41. R. Wodick and D. W. Luebbers, “Quantitative analysis of reflection spectra and other spectra of inhomogeneous light paths conducted in multi component systems with the aid of interval analysis part 1,” Biol. Chem. 354, 903–915 (1973).
42. J. Choi, M. Wolf, V. Toronov, U. Wolf, C. Polzonetti, D. Hueber, L. P. Safonova, R. Gupta, A. Michalos, W. Mantulin, and E. Gratton, “Noninvasive determination of the optical properties of adult brain: near-infrared spectroscopy approach,” J. Biomed. Opt. 9, 221–229 (2004). [CrossRef] [PubMed]
43. A. Demel, K. Feilke, M. Wolf, C. F. Poets, and A. R. Franz, “Correlation between skin, bone, and cerebrospinal fluid layer thickness and optical coefficients measured by multidistance frequency-domain near-infrared spectroscopy in term and preterm infants,” J. Biomed. Opt. 19, 017004 (2014). [CrossRef]
44. “SenSmart model X-100 universal oximetry system for pediatric patients,” Tech. Rep. 10392-001-01, Nonin Medical, Inc., 13700 1st Avenue North Plymouth, MN, 55441-5443, USA (2014).
45. S. J. Arri, T. Muehlemann, M. Biallas, H. U. Bucher, and M. Wolf, “Precision of cerebral oxygenation and hemoglobin concentration measurements in neonates measured by near-infrared spectroscopy,” J. Biomed. Opt. 16, 047005 (2011). [CrossRef] [PubMed]
46. S. Ijichi, T. Kusaka, K. Isobe, K. Okubo, K. Kawada, M. Namba, H. Okada, T. Nishida, T. Imai, and S. Itoh, “Developmental changes of optical properties in neonates determined by near-infrared time-resolved spectroscopy,” Pediatr. Res. 58, 568–573 (2005). [CrossRef] [PubMed]
48. S. J. Matcher, P. J. Kirkpatrick, K. Nahid, M. Cope, and D. T. Delpy, “Absolute quantification methods in tissue near-infrared spectroscopy,” Proc. SPIE 2389, 486–495 (1995). [CrossRef]
49. Y. Sakata, M. W. Abajian, M. O. Ripple, and R. Springett, “Measurement of the oxidation state of mitochondrial cytochrome c from the neocortex of the mammalian brain,” Biomed. Opt. Express 3, 1933–1946 (2012). [CrossRef] [PubMed]
51. A. Demel, M. Wolf, C. F. Poets, and A. R. Franz, “Effect of different assumptions for brain water content on absolute measures of cerebral oxygenation determined by frequency-domain near-infrared spectroscopy in preterm infants: an observational study,” BMC Pediatr. 14, 1–6 (2014). [CrossRef]