1. Introduction

 Spectroscopic techniques of various types are now common in both clinical practice and research as non-invasive methods to measure medically relevant factors such as blood perfusion and oxygen saturation (sO₂) [1,2]. These methods rely on the relative transparency of tissue in the so-called optical window between 600 and 1000 nm [3]. Light in this range can be used to probe mm to cm volumes of tissue via transmission, scattering, and/or diffuse reflectance. Differences in the absorption and scattering characteristics of chromophores present in tissue the light passes through will be imprinted in the detected light. In particular, well-characterized differences in the absorption spectra of oxygenated (HbO₂) and deoxygenated (HbR) hemoglobin can be used to calculate sO₂, the relative proportion of oxyhemoglobin (HbO₂) to the total amount of hemoglobin (HbT).

Despite being widely used in everyday medical practice, even the most well-known and robust spectroscopic methods, such as pulse oximetry, have come under scrutiny for accuracy concerns [4–6]. A particular issue arises when measurement conditions differ from typical calibration conditions or testing and validation conditions. Since light skinned people are typically overrepresented in calibration studies [4], pulse oximeters can overestimate sO₂ in patients with a higher skin melanin content leading to a racial dependence in misdiagnosis of hypoxemia. Work to address these racial disparities is ongoing and newly developed spectroscopic techniques, such as hyperspectral imaging (HSI) and diffuse reflectance spectroscopy (DRS), often explicitly attempt to address the variability of melanin in patients [7,8]. While HSI and DRS are not in routine clinical use yet, they are already being used for clinical research, for example, to challenge common recommendations for the use of epinephrine in the surgical field [9,10].

Epinephrine is commonly used as an adjuvant to local anesthetics in order to induce vasoconstriction and reduce intraoperative bleeding. Anesthetics combined with epinephrine have been used to reduce intraoperative bleeding since 1903 [11], but despite this, the optimal waiting time between injection and commencement of surgery is still a subject of debate. Textbooks commonly mention a waiting time around 7-10 minutes [12] and several studies from the face and other parts of the human body have recommended times from 2-10 minutes [13–15], when clinically observing the reduction of intraoperative bleeding. These waiting times have recently been disputed by data from spectroscopic techniques that measure the absorption of HbO₂ and HbR to calculate sO₂ as a proxy for perfusion. For example McKee et al. [10], report that the lowest cutaneous hemoglobin level, measured using DRS, was not observed until 26 minutes after injection into the skin of the forearm. Additionally, a recent study by Thiem et al. [9] concluded that perfusion reduction, measured with HSI, occurred 30 minutes post injection. If the results obtained by McKee et al. and Thiem et al. are representative, it would indicate a need for longer waiting time before commencing surgery than what is used in clinical practice today.

The discrepancies between the observations from spectroscopic techniques to measure sO₂ compared to that when observing intraoperative bleeding, raises the question if these methods are robust to changing environment of the detection volume in the vicinity of an injection site. Despite the explicit attempt to address some confounding factors like melanin variation, questions remain if other factors could be affecting the robustness of sO₂ measurement using spectroscopic techniques in these experiments. For example, the act of injection includes addition of water to the detection volume, affecting both absorption and scattering. Not accounting for this change has the potential to have contributed to these contentious recommendations on the use of epinephrine. In addition, a number of observations have been made that are yet to be fully understood mechanistically. For example, Thiem et al. [9] described observing a distinctive ring pattern, with a marginal circular onset of hypoperfusion and simultaneous hyperaemia around the injection site (center); McKee et al.¹⁰ observe an immediate increase in hemoglobin that they attribute to a local histamine release by mast cells because of tissue trauma from the injection and Bunke et al. observe a previously documented window effect when using DRS [16,17].

DRS and HSI are ideal candidate methods for monitoring the vasoconstrictive effect of epinephrine because they can be used to non-invasively monitor local sO₂. An expected downstream effect of vasoconstriction is a reduction in supply of oxygenated blood, which because of continued metabolism, results in a reduction of the local sO₂. DRS can accurately measure sO₂ at a single location with a defined detection volume [18–20] using a single source detector arrangement at or close to the center of the injection site. Using a single detection fiber allows very high spectral resolution to be acquired from one point. HSI can combine spatial information with spectral information analogous to the point measurement made by DRS. HSI is non-invasive, non-ionizing, contact-free, and can be used to create an anatomical map. While HSI has not yet been implemented in clinical practice applications, it has been demonstrated as a method to monitor sO₂ in animal studies [21], skin flaps [22,23], and during shock and resuscitation [24]. Some clinical pilot studies have been made on humans showing that HSI can create clinically relevant sO₂ maps in diabetic foot ulcers [25,26], peripheral arterial disease [27], cutaneous perforator vessels in the thigh [28], and in flaps during reconstructive surgery [29–31].

The aim of this study was to investigate if DRS and HSI can remain robust in their assessment of sO₂ to the addition of water, as a result of injection into the detection volume. Thus determining the viability of using DRS and HSI to study sO₂ after a subcutaneous injection of a local anesthetic with epinephrine. We also aim to use HSI to shed light on the discussion of the use of epinephrine in local anesthetics by comparing the spatial and temporal distribution of changes in sO₂ post epinephrine injection to saline injection.

2. Methods

2.1 Ethics

The study was approved by the Swedish Ethical Review Authority. The research adhered to the tenets of the Declaration of Helsinki as amended in 2008. All the subjects were thoroughly informed about the study, the voluntary nature of participation, and gave their informed written consent.

2.2 Subjects

Inclusion criterion was skin type I-III on the Fitzpatrick scale [32] in order to reduce melanin-dependent variability in the results. Exclusion criterion was the presence of any advanced medical condition that could contraindicate the injection of local anesthetics containing epinephrine, such as ischemic heart disease, heart arrhythmia, lung disease, asthma, or previous adverse reaction to local anesthetics. Subjects identified as being smokers or having microangiopathy, i.e. resulting from diseases such as diabetes, kidney, or cardiovascular disease, were also excluded. The subjects were asked to refrain from caffeine-containing drinks and food for at least two hours before the measurements. 12 healthy adult volunteers, 6 men and 6 women, were included in the study. The median age of the subjects was 29 years (and ranged from 24 to 40 years). Subjects 1 and 2 were excluded from aggregated DRS data due to equipment fault during acquisition. All subjects were included in aggregated HSI data.

2.3 Diffuse reflectance spectroscopy

A custom-built diffuse reflectance spectroscope based on the Ocean Insight (QE-Pro) spectrometer was used to record absorption spectra from the injection site. A halogen lamp (HL-2000-FHSA, Ocean Insight Inc.) was coupled into the output fibers of a reflectance probe (custom made, Fiberguide Industries Inc., Caldwell, USA) and the return signal coupled into the spectrometer which acquired spectra at 20 Hz. Light was directed into the tissue via the output fibers around the circumference of the probe which was placed in contact with the skin surface (Fig. 1(a)). The light scattered from the skin was collected by the fiber in the center of the probe and detected by the spectrometer in the spectral range between 400–1000 nm. The wavelength and separation between the light source and the detector govern the depth of the measurements; greater separation resulting in a greater depth [20]. The probe used has a radius and source detector separation of 5 mm resulting in a detection volume on the order of a few millimeters into the tissue, which primarily probes the dermal (Fig. 1(b)). Since the signal is an average of the entire volume measured, no spatial information can be obtained from the measurements [33]. The temporal and spectral resolutions of Δt = 0.05 s and Δλ = 0.7 nm respectively were not matched by HSI. In this work, DRS was used to investigate any rapid changes to the absorption of tissue at the injection site and was therefore only applied for 5 minutes post epinephrine injection for comparison to baseline spectra acquired pre-injection.

 

Fig. 1. a) Photograph of epinephrine injection site with placement of DRS probe. Inset shows the front face of the probe with a detection fiber in the center and a ring of illumination fibers at a radius of 5 mm. b) (top) schematic showing a typical photon path through tissue with the primary probed depth with the dimensions of our probe overlapping with the dermal skin layer. (bottom) Schematic of the protocol showing the timing of the measurements in relation to the injection and the frequency at which the data is acquired for DRS (Δt = 0.05s). c) Picture acquired during an HSI acquisition with the injection sites for epinephrine and saline indicated. The illumination is visible to the left with the detection indicated with a black line. The white reference is observed on the right. The inset shows an enhanced contrast image of the two injection sites where the vasoconstrictive effect from the epinephrine is visible compared to the saline injection site. d) (top) Schematic showing the photon paths for the HSI method where more incident photons at different distances from the detection leads to a larger probed volume. With an illumination line width of approximately 2 cm and the detection line in the center, an approximate probe volume includes the epidermal and dermal layers of the skin. (bottom) Schematic of the protocol indicating the reduced frequency at which data is acquired at every single point compared to DRS.

Download Full Size | PDF

2.4 Hyperspectral imaging

A custom-built hyperspectral camera, based on the line-scan HySpex model series (Norsk Elektro Optikk AS, Oslo, Norway), was used to generate a 10 cm wide detection line with a spatial resolution of 150 µm across 640 pixels. The maximum scan length was 25 cm, which is equivalent to roughly 1650 lines. A halogen lamp (operating at 2900 K) was coupled to a line-light generator and provided illumination that could be scanned together with the camera to generate a diffuse reflectance spectrum at every point along the detection line. The camera has a spectral resolution of 3.4 nm, providing 382 unique spectral channels between 400 nm and 1700nm in each spatial pixel. A short pass filter on the illumination was used to block wavelengths longer than 1100 nm to reduce temperature on the patients’ skin. A long-pass spectral filter was used to block light with wavelengths shorter than 500 nm such that the second-order diffraction signal is only expected above 1000 nm. Moreover, due to the peak emission wavelength of the excitation being in the near infrared (NIR) region, shorter wavelengths were omitted due to poor signal-to-noise. Thus, data acquisition was limited to the spectral range between 700 and 1000 nm, which corresponds to 81 spectral channels. Scanning the 10 cm wide line a distance of at least 15 cm was sufficient to cover the region of interest on the forearm including injection sites of both epinephrine and saline (Fig. 1(c)). The duration of a 15 cm scan was roughly 80 s seconds and repeated continuously over 35 min. With an approximate return time of 10s, the spectral information in any given point was updated every Δt ∼ 90 s (Fig. 1(d)).

Prior to scanning, the background dark-signal of the detector was determined and subtracted from every image. A sterile white strip (Tube Holder, 708131, Mölnlycke, Gothenburg, Sweden) was placed in the scan area and acted as “white reference” so that spatial variation of the light intensity could be normalized with reference to a separate reference measurement against a standard calibrated Spectralon diffuse white reference (WS1, Ocean Insight Inc.). Unlike DRS, the incident light is not spatially controlled to yield a single distance to the detection line. The uncontrolled nature of the illumination area (Fig. 1(c)) gives an integration over all possible source detector separation distances, and thus a large undefined detection volume, resulting in an averaged response from the measured absorption (Fig. 1(d)). The hyperspectral camera was mounted on a moveable arm attached to a mobile cart, which enabled positioning of the camera system such that the subject could remain in a comfortable position for an extended period of time. This would also minimize subject motion during acquisition. In this study, HSI was used to map the spatial evolution of the epinephrine effect over a clinically relevant time period of 35 mins necessary to address the question of the necessary time frame of the epinephrine to take effect.

2.5 Study protocol

Ambient temperature was maintained around 22 °C and the subject was placed in supine position during all measurements and rested 10 min before the start of the experiment. The patient’s right arm was stabilized with a vacuum pillow (Germa Protec, Germa AB, Kristianstad, Sweden), such that the forearm was upturned and level. Subjects were instructed not to move the arm during the procedure. Pulse, blood pressure (BP) measurements as well as pulse oximetry measurements were performed at 0 min, 10 min, and at the end of the experiment and were stable during the entire procedure. At 0 min, subjects had a median pulse of 62 bpm, blood pressure of 110/65 mmHg, sO₂ of 98%, at 10 min, median pulse of 62 bpm, blood pressure of 105/65 mmHg, sO₂ of 98% and at 40 min median pulse of 62 bpm, blood pressure of 105/60 mmHg, and sO₂ of 98%.

Prior to any injection, a baseline measurement with HSI was conducted of the subject’s right forearm. Thereafter, 0.5 ml lidocaine (20 mg/ml) + epinephrine (12.5 µg/ml) (Xylocaine Dental Adrenaline, Dentsply Ltd., York, PA, USA) and 0.5 ml saline (9 mg/ml) were injected subcutaneously with spatial separation of at least 7 cm (Fig. 1(c)). Hyperspectral images covering the area of the forearm including both the epinephrine and saline injection sites were acquired at regular intervals for a total period of 35 minutes.

During the HSI measurement period, a baseline DRS measurement was taken from the left arm, after which a subcutaneous injection of 0.5 ml lidocaine (20 mg/ml) + epinephrine (12.5 µg/ml) was performed. The DRS probe was then placed as close as possible to the expected center of influence of the injection as determined by the surgeon administering the injection and data was recorded for 5 minutes continuous. The measurement protocols are schematically demonstrated in Fig. 1(b) and 1(d). To ensure uniformity, all injections were carried out by the same surgeon with the same injection technique (J.B.).

2.6 Data analysis

2.6.1 Preprocessing

DRS data was acquired and saved using a customized LabVIEW (National Instruments Corp.) program. HSI data was acquired and saved using a HySpex third party software. All data were analyzed using Matlab 2022b (The MathWorks Inc. South Natick, MA, USA). Absorption spectra, A, were calculated for both DRS and HSI data as ${\boldsymbol A} = \; - {\log _{10}}({{{\boldsymbol I}_1} - {{\boldsymbol I}_{BG}}} )/({{{\boldsymbol I}_{REF}} - {{\boldsymbol I}_{BG}}} ))\; $where a dark spectrum (I_BG) and a bright spectrum (I_REF) reference measurements were acquired as follows: For DRS, I_REF was determined by recording the response with the probe in contact with a standard calibrated Spectralon diffuse white reference (Ocean Insight, WS1) and multiplying by 1000 to account for difference in detection efficiency in the white reference vs human tissue. Background I_BG was the dark response of the spectrometer measured with the halogen lamp off and the probe covered. For HSI, I_BG was automatically subtracted from each image on a pixel-by-pixel basis using a dark image acquired with the camera shutter closed. I_REF was determined for each spatial position in the scan using the intensity of the grey reference, I_G, in each spatial pixel relative to a recorded reflectance, R_REF, of the grey reference. This reflectance was acquired from a reference image of the grey reference, ${\boldsymbol I}{^{\prime}_G}$, together with a standard calibrated Spectralon diffuse white reference, ${\boldsymbol I}{^{\prime}_W}$, such that ${{\boldsymbol I}_{REF}} = {{\boldsymbol I}_G}/{{\boldsymbol R}_{REF}}$ where ${{\boldsymbol R}_{REF}} = {\boldsymbol I}{^{\prime}_G}/{\boldsymbol I}{^{\prime}_W}$.

In the HSI data, motion between each scan was corrected by manual feature identification (of pen marks and distinctive features) on a single wavelength image. Localized cross correlation on 11-by-11 pixel regions was used to refine feature matching between adjacent scans in time. A rigid affine transform for each time point was calculated from the matched features using the ‘fitgeotform2d’ function in Matlab and applied to all wavelength images at each time point.

2.6.2 sO₂ estimation algorithms

A common method for determining sO₂ from absorption measurements is to generate a calibration lookup table against which the measured absorption can be compared. This is typically done using a single carefully chosen wavelength or mathematical operation on a small number of wavelengths. For example, the absolute absorption at λ = 675 nm will vary with sO₂ because of the large difference in absorption between HbO₂ and HbR at this wavelength. As such, imposing a change in sO₂ and measuring the absorption at λ = 675 nm generates a lookup table from which future measurements of absorption can be compared against, however this calibration can be rendered inaccurate by a changing background absorption. Changing background can be controlled for using a ratio of absorption at two wavelengths, one wavelength with a large difference in absorption between HbO₂ and HbR and one wavelength with no difference. For example, ratio of absorption at 796 nm to 760 nm, or the ratio of absorption at 560 nm to 575 nm. More sophisticated calibration lookup tables exist which attempt to account for wavelength dependent changes in background absorption, for example using the minima of the 2^nd order differential of the absorption spectra at ∼580 nm and ∼760 nm, as published by Holmer et al. [7], helps to correct for the wavelength dependent absorption of melanin. Other techniques for calibration lookup tables exist that are maintained as commercial secrets although unless otherwise published data is available, are not known to perform significantly better than representative methods discussed here.

An alternative to calibration-based methods, spectral unmixing makes use of the full range of the measured absorption spectra to estimate, with a linear decomposition, the relative abundances of known absorption spectra of the expected component chromophores.

2.6.3 Spectral unmixing

Spectral unmixing [34–37] of either the DRS absorption spectra or the pixel-by-pixel HSI absorption spectra was performed as previously described [36] to determine the relative abundance (contribution) of likely absorbing chromophores (endmembers) present in the detection volume. Spectral unmixing assumes a linear combination of a limited set, of endmember spectra, s_i, with fractional abundance, a_i, and spectral noise, w, can model the measured absorption spectra A.

Three endmember models were considered when unmixing:

Spectral unmixing has previously been shown to be sensitive to the spectral range analyzed [37]. The spectral ranges used in this study were determined by maximizing the size of the spectral window until the point at which a large increase in the squared residual norm of the fit model was observed giving the ranges of [500 nm, 950 nm] for DRS and [700 nm, 975 nm] for HSI.

2.6.4 Simulated absorption spectra

A pseudo-simulated series of spectra were generated to demonstrate if, in principle, sO₂ measurement can be influenced by changing the relative magnitude of absorption by water in the detection volume. The simulation was also used to confirm previously discussed effects of changing fractional abundances of melanin and water and to investigate the effect of changing blood volume (total hemoglobin, HbT, defined as the sum of HbR and HbO₂ present).

Spectral unmixing of the baseline DRS absorption spectra from one subject using the “full model”, spectral range [500 nm, 950 nm], provided fractional abundances, ${a_i}$, for chromophores that could be used as the basis of a pseudo-simulated absorption spectrum. A generated spectrum from these derived parameters is a close fit to the measured spectra and enables investigation of the effects of changes in individual endmember fractional abundance. Spectra with sO₂ in the range of 0 to 100%, with constant blood volume were generated first. sO₂ was varied while keeping HbT constant, by varying ${a_{HbR}}$ and ${a_{HbO2}}$ such that the ratio ${a_{HbO2}}/({{a_{HbR}} + {a_{HbO2}}} )$ varied in the range [0, 1]. For subsequent simulations, this ratio was fixed to give a fixed sO₂ of 85% to represent a typical physiological value. In a second simulation, total blood volume was varied by adjusting the fractional abundances ${a_{HbR}}$ and ${a_{HbO2}}$ within the range between zero and two times the value measured from the experimentally acquired spectrum. In the third and fourth simulations, the fractional abundances of water and melanin respectively were varied in the range between zero and two times that of the measured values from the experimentally acquired spectrum.

The simulated spectra with physiological blood volume, water, and melanin but varying sO₂ were used to generate lookup tables for four separate calibration-based sO₂ measurement methods. These were:

The response of these calibrated sO₂ measurement methods was determined on spectra simulated with varying blood volume, water fraction, and melanin fraction at a constant sO₂ = 85% as detailed above. For comparison, spectral unmixing was performed using both the “simple model” and the “full model” detailed above. The spectral range used in the case of the full model was limited to [525 nm, 600 nm] such that the spectral unmixing operation was not the exact, inverse operation of the simulation of the spectra.

2.6.5 Statistics

The aggregated data is displayed as median of all included subjects with 95% confidence intervals indicated by a shaded area.

3. Results

3.1 Simulated absorption spectra

Figure 2(a) demonstrates how the simulation was used to generate an absorption spectrum analogous to DRS measurement at oxygen saturations from 0 to 100%. These spectra were used to calibrate the different methods of extracting oxygen saturation, as described in the methods section. The methods rely on computing lookup tables from example spectra and are analogous to some commercial oxygen saturation monitors.

 

Fig. 2. Simulations of absorption spectra with changing a) sO₂, b) Blood Volume Fraction, c) Water Volume Fraction and d) Melanin Volume Fraction, with a fixed sO₂ of 85%, show how different chromophores have disproportionate impact in different regions of the spectral range of interest. The result of sO₂ measurement on the simulated spectra using two spectral unmixing (SU) models, three calibration methods (Cal.) and a melanin corrected calibration method (Cor.) are shown (see methods section for details). Using SU with all endmembers results in the most accurate measure of sO₂, while SU with too few parameters is insufficient in most conditions. Calibrated methods perform worse when the measurement environment differs from the calibration environment by blood volume (b right), water volume (c right) and melanin (d right). This exposes weaknesses in some techniques used to evaluate sO₂ from spectroscopic data.

Download Full Size | PDF

Using a baseline sO₂ of 85%, fractional contributions in the simulation of the absorption spectrum were altered to mimic changes in volume fractions of Hb, water, and melanin (Fig. 2(b), (c), (d) left). The calibration lookup tables generated from the first simulation were used to estimate sO₂ from the simulated spectra to judge their performance under changing conditions and compared to results from linear spectral unmixing using both the “simple model” and the “full model” (Fig. 2(b), (c), (d) right). Note, the “full model” was modified to use the limited spectral range [525 nm, 600 nm] because using the full range and endmembers matching the simulation trivially results in a perfect recovery of the ground truth. The “simple model” of linear unmixing with an insufficient number of endmembers to accurately characterize the absorption spectrum generates a significant underestimate of sO₂ across all parameters. Increasing the number of endmembers to the set used in the full model improved the fit considerably, although due to the reduced spectral range used in this modified model there was an underestimate of the sO₂ at low blood volumes. In general, the methods used here appear to deal well with changes in one or more of the component chromophores but perform poorly with at least one of the others. For example, intensity ratio calibration at 760:796 nm performs consistently as the blood volume fraction changes (Fig. 2(b) right) but performs very poorly as the water volume fraction changes (Fig. 2(c) right). As seen in Fig. 2(d), methods such as the 2^nd order difference calibration, which are specifically designed to be robust to changes in melanin volume fraction, perform well with respect to simulated changes in melanin fraction. None of the calibration methods however, perform consistently across variation of all parameters in this simulation and in general the negative impact of changing water fraction and blood volume appears to be comparable to the effect of changing melanin.

3.2 Evolution of sO₂ measured by DRS

DRS spectra exhibit notable changes immediately after epinephrine injection and continue to gradually change over the 5 min measurement period (Fig. 3), similar to some of our previous observations [38]. In the representative example shown in Fig. 3(a), a baseline sO₂ value of 90% is calculated from the spectrum using spectral unmixing with the “full model” as detailed in the methods section. Immediately post injection, the absorption appears biased toward longer wavelengths, with increased absorption from 600 nm to 950 nm, and reduced absorption from 500 nm to 600 nm. Measured immediately post injection, sO₂ increases to 98.4%. At 5 min post injection, absorption remains high relative to the baseline at longer wavelengths, although sO₂ has fallen to 56.2% (Fig. 3(a)).

 

Fig. 3. a) Absorption spectra acquired with DRS at baseline, immediately after epinephrine injection (0 min) and after 5 min, and the subsequent drop in sO₂. b) Water volume fraction vs. time after injection extracted from DRS spectra using spectral unmixing with all endmembers compared to spectral unmixing using all endmembers with baseline fixed (see methods section). Median (black line) and 95% confidence interval (shaded area). c) sO₂ vs time after injection extracted from DRS spectra using spectral unmixing with two endmembers (red), baseline fixed endmembers (yellow) and all endmembers (green). When using all endmembers the absolute sO₂ values approach 100% and thus become more clinically relevant. Normalizing to the baseline enhances the decrease of the sO₂ which is in accordance with clinical expectations.

Download Full Size | PDF

Figure 3(b) shows the contribution from water extracted from spectral unmixing averaged across all included subjects. The two traces show the fractional abundance from water measured using the full model and the baseline fixed analysis in which contributions from melanin and fat are fixed to the values obtained at baseline. Using the full model, the contribution from water remains relatively constant with respect to the baseline. This contrasts with a large, post injection increase in the relative contribution from water when the baseline correction is applied. Figure 3(c) shows the calculated sO₂ for these two scenarios, as well as the simple model in which only HbO₂ and HbR are considered. In the latter scenario, two notable features are observed, a post injection increase in sO₂ from baseline, as well as quite low sO₂ baseline values. Using all endmembers in the spectral unmixing analysis, the baseline value becomes more realistic around 95%, after which a reduction is observed during post injection to about 85%. Applying a baseline correction, the only notable change is that the lowest sO₂ value reached after 5 minutes is significantly lower, at 45%. Applying an exponential fit to the traces shown in Fig. 3(c), we can extract a time constant in order to get some characteristic measure of how fast the epinephrine effect acts. In all three cases, we obtain a characteristic time constant of 1-1.5 minutes, which is in line with some of our previous work [16,17], although it also differs from others [9,10]. Critically, despite analysis with the simple model drastically underestimating sO₂ in vivo as well as in simulations, and the baseline fixed result indicating the full model is influenced by changing water content, the time constant of the exponential decay in sO₂ is reliably replicated across the three analysis methods.

3.3 Spatially-resolved sO₂ from HSI

As HSI provides spatial information as well as spectral information, spectral unmixing can be performed at every point in space to map out sO₂ and other fractional components, such as water. Figure 4 (1^st row) shows a representative example of a 2D map of the progression of sO₂, relative to baseline, at the epinephrine injection site. The analysis here only includes the spectral signatures of HbO₂ and HbR (“simple model”). A region around the injection site (white dot) can be seen where there is little to no change in sO₂ over the course of 35 minutes.

 

Fig. 4. Heat maps at 0 min, 5 min and 35 min after injection with epinephrine demonstrating the change in (top row) sO₂ relative to baseline calculated with spectral unmixing using only HbR and HbO₂, (second row) contribution from water calculated with spectral unmixing using the full model, and (third row) sO₂ calculated with spectral unmixing using the full model. (bottom row) schematic showing representation of the diffusion of water in relation to the injection site at the three post injection time instances.

Download Full Size | PDF

Performing the analysis with the full model allows an equivalent map of the contribution from water to be generated (2^nd row). Visual inspection of this map shows that the region in the sO₂ map where no change is observed when applying the simple model (1^st row) coincides with that of water (2^nd row). This is especially apparent in the first few minutes before diffusion of the water into the surrounding tissue. The spatial evolution of sO₂ when taking water into consideration (3^rd row) shows a more spatially homogeneous reduction in sO₂ over time. The bottom row in Fig. 4 schematically depicts the assumed diffusion over time of the injected epinephrine/water mixture. Initially, the rather localized volume with relatively high concentration leads to strong absorption at wavelengths associated with water absorption, disrupting accurate measurement of sO₂. The extent of this disruption being dependent on the method of analysis. Thereafter, the injected fluid gradually diffuses laterally resulting in reduced concentration of water and reduced effect on measurement of sO₂. The time over which this diffusive process occurs appears to be significantly longer (>5 min) than the time constant of the effect observed in the DRS data (1 - 1.5 min). HSI maps of the saline injection site exhibit minimal change in sO₂ while a clear signal from water can still be seen.

3.4 sO₂ at different locations relative to injection site

Using HSI, absorption spectra close to the center of the epinephrine and saline injection sites were compared and analyzed using spectral unmixing with the full model (Fig. 5). In the epinephrine injection site, sO₂ is estimated to 95.1% pre injection (Fig. 5(a)). Immediately post injection, there is increased absorption for all wavelengths in the range [725, 975] and sO₂ is estimated to be 97.1%. After 5 min, absorption remains elevated relative to baseline, but sO₂ has fallen to 94.7% and after 35 min absorption has fallen and sO₂ has fallen again to 92.5%. In the saline injection site (Fig. 5(b)), absorption also increases immediately after injection, however the only difference from baseline after 35 min occurs at longer wavelengths in the region of 950 nm, where we expect to see significant absorption from water (Fig. 2(e)). sO₂ at the saline injection site is 96% at baseline and rises to 98.5% after injection before falling to 97.5% after 5 min and to 95.8% after 35 min.

 

Fig. 5. Time evolution of sO₂ and water with a dependence on distance from the injection site measured using HSI. Representative examples of absorption spectra at the center of the a) epinephrine and b) saline injection site. c) sO₂ and d) water fraction at the center and edge of the epinephrine and saline injection (black line is median values, while shaded area is 95% confidence interval). The red traces represent the exponential fits from which the time constants (t_1/2) have been extracted. In the edge of the injection site the epinephrine effect takes a few minutes, while it appears to take longer in the center where there is a greater influence from water.

Download Full Size | PDF

Figure 5(c) shows the time evolution of sO₂ averaged over all included subjects from these two locations, with the addition of data from the periphery of the epinephrine injection site. An initial, small increase in sO₂ appears to some degree across the entire epinephrine and saline injection sites. However, distinct differences between the center and the periphery of the epinephrine injection site appear immediately after this. The sO₂ measured at the center of both the saline and epinephrine sites appears to decay slowly over the whole time period of measurement and cannot be fit with an exponential. By contrast, sO₂ on the periphery decays quickly and can be fit with an exponential decay with a time constant of 177 (129, 226) s. As seen in Fig. 4, we expect large increases in the concentration of water in the center of the injection site. This is replicated on average across all subjects when comparing the measured water content (Fig. 5(d)); the increase in the periphery is significantly delayed with respect to the center of the epinephrine injection site.

Observations from HSI in the periphery of the injection site replicate the point measurements made when using the full model to analyze the single point DRS data (Fig. 3). A small increase in sO₂ followed by an exponential decrease with a time constant on the order of a few minutes. The DRS data was also analyzed with the baseline fixed model, on the assumption that the melanin and fat content should be invariant with respect to the injection of epinephrine. This was not possible in the HSI data due to imperfect motion correction between frames.

4. Discussion

Spectroscopic techniques are powerful tools in the context of both clinical medicine and clinical research. They have the power to rapidly access otherwise intractable details about the chemical composition of tissue in a non-invasive way. There is always value, however, in addressing the subtleties of both established, highly trusted techniques as well novel techniques with the goal of improving accuracy and reliability. There is an ongoing drive to address shortcomings associated with the use of pulse oximetry in clinical medicine with, for example, the goal of reducing racial disparities in medical outcomes [4]. Here, we discuss an example in which the reliability of DRS and HSI as tools for clinical research could be called into question and suggest techniques to address these concerns. We discuss an example which has the potential to be a determinant in an ongoing debate concerning the time to onset and duration of vasoconstrictive drug combinations used in local anesthetics.

4.1 Baseline sO₂ measurements

A consistent reference value for mean tissue oxygen saturation in the detection volume of either DRS or HSI imaging of the subcutis appears lacking. Bickler et al. [39] tested 5 cerebral oximeters based on spectroscopic methods and demonstrated significant inter-subject and inter-instrument variation. Unlike pulse oximeters, cerebral oximeters, and by extension HSI and DRS have not previously been shown to provide an “actual” measurement of sO₂ in a tissue region of interest. In the absence of an independent measure of sO₂ we make the assumption based on hemoximeter blood gas analyser measurements of venus and arterial blood [39] on healthy patients at rest that sO₂ in arteries is 100% and sO₂ in veins is up to 80%. Assuming a 70:30 ratio of venous to arterial blood, a systemic average baseline sO₂ in healthy volunteers should be 86%. We do not expect the subcutis of the immobilized forearm to contribute significantly to oxygen consumption in this systemic average. As such we are confident the baseline sO₂ measurements between 90% to 95% consistently measured here with both DRS and HSI are physiologically possible. This contrasts with, for example, Thiem et al. [9] who quote a baseline sO₂ value on the order of 45%, which we consider unrealistic from a clinical perspective.

4.2 Hyperaemic effect

McKee et al. [10] and Thiem et al. [9] observe a rapid hyperaemic effect at the center of the injection site. McKee et al. identify this immediate increase in hemoglobin as potentially caused by histamine released by mast cells as a result of tissue trauma. They also attribute the “chemical sympathectomy” effect of lidocaine, resulting in immediate vasodilation. In this study, the same effect is seen with an injection of saline solution, limiting the possible explanations attributable to chemical responses, although tissue trauma remains a possible explanation. In this study we also draw attention to the co-localization of large increases in measured water content with the hyperaemic effect at the center of the injection site. It is possible that the significantly higher concentration of water in the detection volume affects the measured absorption spectrum enough to introduce errors in the assessment of sO₂. At higher water fractions, a method not considering absorption by water in the sO₂ assessment could overestimate sO₂. Hence, one would observe a contradictive rise in sO₂ over time, which may explain some of the observations in previous reports.

4.3 Ring-like presentation of hypoperfusion

The results from HSI data acquired in this study replicate the marginal circular onset of hypoperfusion seen by Thiem et al. [9], which they attribute to a differential and more dynamic diffusion of epinephrine than lidocaine. In this study we were also able to visualize the water content which appears to be anti-correlated with the regions of hypoperfusion. It’s not clear that epinephrine should diffuse faster than the water content of the injected fluid. This leads us to believe that the ring like structure observed may rather be an artefact of the sO₂ measurement, which is sensitive to the changing water fraction and blood volume fraction. Alternatively, if the injection of water causes trauma that results in hypoperfusion that is co-localized with the presence of water, this is common to all injections and the specific effect of epinephrine is only observable in the periphery of the injection site outside this localized trauma site.

4.4 Spectroscopic sO₂ estimation in changing environments

Simulations of absorption spectra under varying fractional abundance of key chromophores indicate that changing environments have differing impacts depending on the method used to calculate sO₂, which is likely attributed to the increasing difficulty in correctly assigning the attenuation of light to either scattering or absorption [1]. Specifically, when adding water in the context of an injection, the measured absorption increases at longer wavelengths [38], both in simulation and in experimental data (Fig. 1(d) and 2(b)). The physical addition of water, a liquid, to the injection site must change the composition of chromophores in the detection volume of the DRS/HSI instrument. If we assume that the detection volume is not significantly changed by the addition of water, displacement of blood (liquid) from this volume is more likely than displacement of other chromophores such as fat and melanin. This would reduce the fractional contribution from blood volume, reducing the absorption at wavelengths shorter than 600 nm. This observation is made in both simulation and experimental data acquired with DRS (Fig. 1(b) and 2(b)).

In this study, the spectral range between 700 nm and 1000 nm was found to be optimal for use in linear unmixing. The spectral features of HbO₂ and Hb absorption across this range are generally increasing and decreasing respectively. To a first order approximation therefore, the measured absorption spectrum, in the range used, becomes more similar to HbO₂ following the addition of water (Fig. 1(c)). Thus, if the addition of water were to adversely affect the accuracy of sO₂ determination, we would predict it would result in an overestimation of sO₂. This is relevant here as we note a co-localization of a hyperaemic effect with the increased water content as a result of injection. We expect that a calibration free method like spectral unmixing, with water absorption included as an endmember, would be robust to this and account for the changing water content. It may however, be necessary to take further steps in the context of such dramatic changes to the environment.

Following the addition of water it may also be that the change in scattering properties alters the shape and size of the DRS/HSI detection volume [20], resulting in an uncalibrated path length difference between baseline measurements and test conditions post injection. We know from HSI data that there is a spatially dependent change in absorption pattern. It can therefore be assumed that there is also a wavelength-dependent change in the depth photons reach [40], which is not accounted for here. Such a path length difference would be more constrained in the context of DRS, where the light-injection and detection points are well defined, but still relevant in the consideration of the depth of the measurement.

4.5 Baseline fixed model

As mentioned above, the addition of water is more likely to displace blood than melanin and fat from the detection volume. On this assumption, it is possible to fix these values to a baseline measurement when performing linear unmixing with the full model. This has the potential to reduce the degrees of freedom in the model and avoid overfitting in the context of an absorption spectrum that has become dominated by water absorption but maintains the contribution from melanin and fat which we know to be important in accurate measurement of sO₂ using spectral unmixing. This was challenging with the HSI data due to motion of the patient between the measurements; however, when applied to the DRS data a significant difference in the measurement of sO₂ was observed (Fig. 3). The hyperaemic effect immediately after injection is removed and the initial increase in water fraction is larger. Thus, a plausible explanation of the unexpected hyperaemia is an artefact of measurement caused by an excess of water in the imaging volume, rather than some unknown physiological phenomenon.

4.6 Clinical impact

For successful clinical translation, it will become important that hyperspectral imaging of sO₂ not only provides accurate relative changes, but also physiologically representative absolute values. We begin to address this question here by comparing the baseline sO₂ measurements obtained with spectral unmixing using the simple model and the full model. We suggest that values on the order of 80% to 95% in healthy tissue at rest is more reasonable than below 50%, which we obtained using, for example, the simple model for spectral unmixing. Moreover, having an approach that does not rely on a set of pre-calibrated values in order to obtain an accurate sO₂ value allows for better flexibility accounting for changes in, for example, melanin and fat content. We have demonstrated that a change in water content certainly affects the assessment of sO₂ and as such, caution should be taken in cases where injections are involved, particularly those inducing vasoconstriction. The broader clinically relevant question is whether the expected variation in water content from patient to patient is sufficient to constitute significant change in sO₂ measurement. Furthermore, other factors not investigated here, such as the patient-to-patient variation in fat content, may have implications for the accuracy of sO₂ measurement with the potential to lead to weight-based disparities in clinical outcomes.

The results presented in this study indicate that caution should be taken when pushing the limits of current spectral techniques for non-invasive blood oxygen saturation and related tissue characterization measurements. With the knowledge that the addition of water could have an artificial hyperaemic effect, it’s not reasonable to compare to a baseline to determine if local hypoperfusion has occurred and thus make recommendations for use of epinephrine in surgical fields. We can therefore conclude that the omission of water from the analysis when observing the change in sO₂ as a result of epinephrine injection may not only yield inaccurate sO₂ values, but can also lead to inaccuracies capturing the epinephrine effect and the time it takes to evolve. However, the difference between epinephrine injection and a control measurement such as saline could be indicative of an epinephrine specific effect. Here we see a significant difference between saline and epinephrine after only a short relative waiting time. There is also a significant difference between articaine only and articaine + epinephrine after only 30 s presented by Thiem et al. [9] Similarly in McKee et al. [10] the epinephrine group returns to baseline in under 7 minutes post injection, while the articane only group remains significantly above baseline. Considering the difference between articane only and articane + epinephrine as being the relevant evidence for effect, these studies add to a growing spectroscopic evidence base supporting the previously received wisdom that a waiting time of a few minutes is sufficient for the effect of epinephrine to be beneficial: the decay in oxygen saturation with a time constant of 145 s from HSI at the periphery of injection site (this study), the exponential time constant of 94 s from DRS (this study), exponential with a time constant of 123 s from photoacoustic imaging [16], exponential with a time constant of 109 s from DRS [16], exponential time constant of 205 s from HSI in eyelids [17].

The potential use of DRS and HSI in clinical practice remains a promising proposition and the two methods provide complementary information. In the context of clinical applicability, both methods are nominally non-invasive, with HSI being completely non-contact. The major drawback of DRS is the lack of spatial information on the surface, however, the depth information is more constrained and can be controlled by lateral separation of excitation and detection fibers to steer the interaction volume to a particular depth [41]. This notion has been expanded on to generate depth resolved spatial information to for example non-invasively detect breast tumors [42]. Due to its non-contact operation, HSI is the preferred method for non-invasive monitoring during in a sterile setting, such as surgery [43]. One drawback with the HSI method employed in this work is that the measurement time is on the order of tens of seconds for a relevant region of interest. However, this is not a fundamental limit but rather an engineering limit. Snap-shot HSI methods exist which essentially acquire a full HSI image across an area in an instant [44], however, this comes at an expense of a lower signal-to-noise ratio, as well as reduced spectral and spatial resolutions. One major benefit of HSI, owing to the spatial information available, is the possibility to develop intra patient measurements of variation in tissue composition via AI algorithms and image processing, which has shown to be useful in the automatic delineation of skin tumors [45].

In regard to which method is preferred when monitoring the effect of an epinephrine injection, the determining factors will be the necessity of non-contact monitoring, the need for a spatial resolution and the ease of access. DRS is likely more suitable in routine clinical practice where a quick check of the progression of epinephrine effect is useful, while HSI is more suitable for use in areas where one needs to ensure that a particular area is vasoconstricted. Moreover, HSI is likely the preferred method for research purposes when aiming to, for example, gain some insight into the epinephrine effect, as demonstrated in the observation of the abovementioned “ring effect”.

5. Conclusion

Without explicit methods to control for changing local environment, spectroscopic methods, especially those based on calibrated lookup tables should be reserved for monitoring small relative changes in stable background environments. We are optimistic that properly constrained calibration free methods such as spectral unmixing can reduce the clinical impact of calibration errors and changing environments when using spectroscopic techniques. Current work only highlights the potential pitfalls in not accounting for relevant absorbing species in tissue when using diffuse reflectance methods to assess sO₂. Addressing the issue of water from an epinephrine injection, we were able to provide some clarity into the time required for the epinephrine effect to take place, which has been a topic of discussion due to its potential impact on clinical/surgical practice. The use of DRS or HSI directly in a clinical context to assess the progress of vasoconstriction and potential for intraoperative bleeding has not been demonstrated here, although this remains a possibility. Further work is needed to identify which chromophores influence the reliability in estimating sO₂, and to what extent, for the successful clinical translation of both HSI and DRS for sO₂ estimation using spectral unmixing.