1.

Introduction

Techniques based on diffuse optical spectroscopic (DOS) principles can be applied to tissue to extract in vivo optical properties from tissue (i.e., absorption and reduced scattering coefficients).^1–3 These techniques may be deployed in a variety of measurement geometries ranging from probe-based point spectroscopy to wide-field imaging approaches and utilize a number of illumination/detection schema to exploit temporal and/or spatial responses from light interactions with tissue.^4–10

Once these approaches characterize and isolate the effects of scattering, the remaining absorption properties are often interpreted as due to a linear superposition of chromophores in tissue, such as hemoglobin species (in either oxygenated or deoxygenated states), melanin, lipids, and water. The decomposition of tissue absorption spectra typically employs fitting methods (least-squares for example) resulting in estimates of chromophore concentrations or volume fractions relative to the total volume of tissue interrogated by the detected photons.

Spatial frequency domain techniques are, in general, based on DOS principles; these are noncontact techniques that entail the projection of structured illumination onto turbid media, such as skin. Camera-based imaging is used to detect the spatial frequency-dependent reflectance from the medium, at wavelengths of interest. Absorption and scattering can then be decoupled from reflectance measurements as each optical property will exhibit a differential response as a function of spatial frequency of illumination. This wide-field imaging approach has been applied for monitoring of port wine stain treatments and tissue transfer flap reperfusion and for assessments of burn wounds and nonmelanoma skin cancers.^11–21

Spatial frequency domain spectroscopy (SFDS) (which we have formerly referred to as spatially modulated quantitative spectroscopy) is a spatial frequency domain method for which the detection scheme employs a fiber coupled spectrometer rather than a two-dimensional camera, thereby trading spatial resolution in favor of high spectral resolution ($\sim 1\mathrm{nm}$) and spectral range ($\sim 400$ to 1100 nm). As with many other diffuse optical measurement methods, SFDS has been validated through homogeneous optical phantoms of known optical properties, demonstrating that this technique is capable of quantifying the absorption and scattering properties of turbid media in both visible and near-infrared (NIR) wavelength regimes.²²

Unlike the homogeneous phantoms used to validate these optical techniques, skin is a highly structured organ, consisting of several layers whose composition and function are differentiated. In the context of typical diffuse spectroscopic measurements, the detected spectral contributions from these stratified layers become a function of the thickness of each layer relative to the total volume of tissue interrogated by the optical technique. In the context of this work, we will refer to this as a partial volume effect.

Melanin is problematic in this case, as it has a relatively featureless absorption spectrum, typically described as a power-law spectral dependence,²³ yet can vary greatly in terms of its concentration and the partial volume it may occupy relative to the total volume interrogated by reflectance-based measurement techniques. These factors can present opportunities for cross-talk in spectral decomposition methods between determinations of melanin concentration and that of hemoglobin species. For example, errors in dermal oxygen saturation have been reported in regions of tissue underlying normal pigmented nevi as a consequence of incorrect accounting for epidermal melanin.²⁴ Similarly, in normal skin, the average local tissue oxygen saturation has been shown to trend inversely with the amount of melanin in the skin as assessed across different skin colors (corresponding to pigmentation differences spanning Fitzpatrick types I to V), providing a clear indication that melanin can directly influence the interpretation of spectral signatures from blood.^25,26

SFDS can also estimate the site-specific melanin layer thickness and the layer-specific concentration of melanin in skin based on its spectral response.²⁷ This approach does not assume any layer thickness, but rather leverages the differential penetration depth of visible and NIR wavelengths.²⁸ Fitting for chromophores in these two regions will produce concentrations of melanin and hemoglobin species, relative to the entire volume interrogated. By determining the absorption and reduced scattering coefficient at each wavelength, however, it is possible to also estimate the site-specific penetration depth, $\delta$, from those spectral regions.²⁹ In the case of skin, visible light typically interrogates $\sim 100\mu \mathrm{m}$ whereas NIR wavelengths can penetrate millimeters through this spatial frequency approach. Assuming that melanin and hemoglobin are respectively confined within upper and lower layer structures, the differential determinations of melanin concentration by visible and NIR regimes can be used to describe the layer thickness in which melanin must reside in order to satisfy its respective concentration “detected per volume interrogated” in each regime. Whereas our previous study demonstrated that SFDS could characterize melanin with respect to multiphoton microscopy methods,²⁷ the question remains as to whether this method could then also produce optical properties that are truly faithful to the underlying physiology (e.g., hemodynamics in skin), once melanin is characterized.

In this study, we used both venous and arterial arm cuff occlusion paradigms to induce hemodynamic perturbations in the forearm. These occlusion paradigms have been used previously to illustrate and/or validate the sensitivity of optical methods in the context of monitoring hemodynamic changes in skin.³⁰ Whereas the occlusion paradigm will modulate tissue hemodynamics, our specific interest in this investigation is to assess the influence and potential sources of error that variations in skin pigmentation may impart in the determination of underlying hemoglobin. The accuracy and limitations of this approach are discussed within the context of quantitative, in vivo studies of skin hemodynamics.

We will interpret data collected by SFDS under these induced hemodynamic perturbations in three distinct ways: (1) homogenous fits (least-squares) in the visible regime (450 to 600 nm), (2) homogenous fits in the NIR regime (650 to 1000 nm), and (3) our own layered model approach (450 to 1000 nm).²⁸ From this single measurement platform, we are able to investigate the spectroscopic-based correlation between melanin and tissue hemodynamics independent from influences of differing measurement geometries or instrumentation sensitivity and performance.

2.

Methods

2.2.

Spatial Frequency Domain Spectroscopy Instrument

The specific instrument used in this study (Fig. 1), an illumination source, a 100-W quartz-tungsten-halogen (QTH) light source (Moritex, MHF-D100LR) that was coupled to a digital micromirror device (DMD) (Texas Instruments, Inc. DLP) with a pattern projection area of $22\times 17\mathrm{mm}$. From this projection configuration, tissue was illuminated with a sequence of sinusoidal intensity patterns that can span across a range of spatial frequencies up to $0.5{\mathrm{mm}}^{-1}$. Remitted light from a 1-mm-diameter spot from tissue, centrally located in the illumination pattern projection area was collected by a fiber (NA 0.39, 1 mm core, Thorlabs, Inc.). Collected light was delivered to an Oriel spectrometer (Oriel 77480). The spectral response from tissue was recorded as a function of the spatial frequency at three evenly spaced phases (0 deg, 120 deg, and 240 deg) as we have previously described.²²

Fig. 1

System diagram and experimental setup: broadband illumination is delivered to the imaging head from a 100-W QTH lamp via a liquid light guide. The digital DMD then encodes the light into sinusoidal intensity patterns which are projected onto the surface of the volar forearm. Diffusely reflected light is delivered to the spectrometer via an optical fiber. An automated blood pressure cuff provides modulation of distal hemodynamics through either venous or arterial occlusions.

Four evenly spaced spatial frequencies were employed, ranging from 0 to $0.3{\mathrm{mm}}^{-1}$. A total of 12 spectral measurements were taken for a given data acquisition time point (four spatial frequencies, acquired at three evenly spaced phases per frequency). An autoexposure routine was performed prior to every occlusion time course series to ensure that the data collected uses the full dynamic range of the 16-bit detector. Additionally, three replicate accumulations were acquired and averaged at each projected pattern to help minimize noise. In its current configuration, this resulted in total acquisition times ranging from 10 to 30 s, depending on the amount of pigmentation present.

3.

Results

Figures 2(a) and 2(c) present the quantitative absorption spectra collected from a single subject (Fitzpatrick skin type III) during both venous and arterial occlusion time courses. Blue spectra represent the baseline absorption prior to inflation of the cuff, red spectra represent the absorption during the occlusion, and green spectra represent the post-cuff-release absorption. In the case of venous occlusion, a gradual increase in absorption is observed during the occlusion phase. This trend is correlated with the expected blood pooling in the extremity (venous blood flow is stopped, yet arteries continue to pump blood into the arm). This increase is most obvious in the 530 to 580 nm region of the spectra, where there are distinct absorption features specific to hemoglobin.³⁵ Upon cuff release, these absorption spectra return to baseline values, consistent with expected physiological behavior.³⁶

Fig. 2

Data from one subject (skin type II/III, plotted at type 2.5). (a) Measured absorption spectra from venous occlusion (90 mmHg), (b) relative, model-based change in melanin concentration determination as a function of time, during venous occlusion experiment, (c) measured absorption spectral from arterial occlusion (200 mmHg), (d) relative, model-based change in melanin concentration determination as a function of time, during arterial occlusion.

In the case of arterial occlusion, total hemoglobin concentration remains constant, as both venous and arterial blood flow is stopped. This is evidenced in the measured spectra, as there is no significant increase in overall absorption during the occlusion phase. Rather, there is an expected, gradual depletion of oxygen from blood. This is most obviously observed in the 530 to 580 nm region where the double absorption band of oxygenated hemoglobin transitions to the single band, characteristic of deoxygenated hemoglobin. Upon release, arterial occlusions are typically characterized by a hyperemic response: a large bolus of oxyhemoglobin to accommodate the high demand for oxygen in tissue that was depleted during the occlusion period.³⁷ This is evidenced in the measured spectra by the dramatic, dual absorption feature in the 530 to 580 nm region, associated with oxyhemoglobin.

Figures 2(b) and 2(d) present the relative change in determined melanin concentration by the three methods described in the previous section [i.e., using only NIR data (650 to 1000 nm), visible wavelength data (450 to 600 nm), and our layered model approach (450 to 1000 nm)]. In the venous occlusion case [Fig. 2(b)], both NIR and visible regime data produced increases in melanin concentration that mimicked trends produced by increases in total hemoglobin. However, when the layered model is applied, the melanin concentration estimation is held to within 2% of baseline. In the arterial occlusion case, there was a negative deviation in melanin concentration estimation using NIR spectra during the occlusion period. There were also negative deviations in melanin for both NIR and visible spectra upon cuff release that followed similar trends to the hyperemic response in blood. The layered model showed only minimal fluctuation in its estimate of melanin throughout all time periods of this occlusion experiment.

3.1.

Baseline Melanin Concentrations Reported by Each Spectral Decomposition Method

As Fig. 3 shows, there are highly disparate values reported for melanin concentration when differing spectral regions/models are employed. This is largely due to the fact that light will interrogate tissues not just containing melanin but also tissues that underlie it. As a result, absorption signatures from melanin will only occupy a fraction of the total spectral signature detected from tissue (i.e., partial volume). Diffusely reflected light in the visible may penetrate hundreds of microns, whereas NIR may reach several millimeters, when planar illumination methods are employed.²⁸ The melanin concentration values determined by the layered model method, however, are consistent with volar arm mesurements reported across skin types in our previous validation study.²⁷

Fig. 3

Total melanin concentration (from baseline) determined by each respective spectral decomposition method. Note that as the magnitudes of the reported values scale differently between homogeneous and layered methods, visible and NIR data are scaled to the left $y$-axis, whereas the layer model data are scaled to the right $y$-axis.

From Fig. 3, it is also evident that the range of reported melanin values are also different, i.e., the differentiation in melanin values are not merely a scale factor. Increases in melanin concentration will additionally reduce the effective penetration depth, thereby increasing the relative contribution from superficial tissues (e.g., melanin) toward the total spectral signal measured.

Although the primary aims of this particular investigation were to compare the performance of three methods for interpreting spectral data, the layer model is unique in that it can also determine melanin layer thickness. This study measured the spectral response from the volar forearm and the layered model determined that the melanin layer thickness was $96\pm 3\mu \mathrm{m}$, averaged across all subjects. This was in agreement with previous studies²⁷ as expectations for epidermal thickness would reside between values from volar upper arm and dorsal forarm (locations measured in previous study and correlated to multiphoton microscopy). Additionally, these thickness values did not exhibit any correlation with phototype. As discussed in that publication, it is important to note that these thickness values are reported in terms of optical pathlength and not physical thickness.

3.2.

$\mathit{\Delta }\mathsf{\text{Melanin}}$ Results

Figures 4(a) and 4(b) show the relative percent changes from baseline melanin concentration for NIR only, Vis only, and our layered model approach across all skin types imaged in this study. The magnitude of errors are larger in the venous occlusion case, given the greater change (increase) in the overall tissue absorption due to the blood pooling that occurs during the occlusion phase of the measurements. Averaging across all skin types, there was $8.1\pm 6.5\%$ cross-talk error derived from using NIR spectra only, $4.4\pm 1.4\%$ from visible spectra, whereas the layered model produced mean errors of $1.4\pm 0.8\%$. As total hemoglobin concentration is kept approximately constant during the arterial occlusion phase, total absorption values did not change significantly. There was a transition from spectral features of oxygenated to deoxygenated hemoglobin. In these arterial occlusions, the overall magnitudes of cross-talk errors were lower; however, the mean errors from the NIR regime decomposition remained relatively higher than the other methods, while the layer model approach provided the minimal error. Here, the mean cross-talk errors were $4.0\pm 2.4\%$, $2.0\pm 1.5\%$, and $0.6\pm 0.3\%$ for the NIR regime, visible regime, and layered model, respectively.

Fig. 4

$\mathrm{\Delta }\text{Melanin}$ metric shown for each individual subject as a function of their assessed Fitzpatrick skin type. (*) depict values determined through homogeneous model fits in the NIR, (o) visible, and ($\mathrm{\Delta }$) depict those resulting from the layered model method.

3.3.

Tissue Oxygenation Results

Figure 5 shows the baseline tissue oxygen saturation percentages (${\mathrm{StO}}_2$) as determined using NIR only, Vis only, and our layered model approach across all skin types imaged in this study. When only the NIR wavelength range is used to determine ${\mathrm{StO}}_2$, not only is there a large variance between specific subjects observed but also there is an apparent correlation between tissue oxygenation and skin type. Here, NIR determined ${\mathrm{StO}}_2$ declines as skin pigmentation increases. When only a visible spectral range is considered, ${\mathrm{StO}}_2$ does not appear to vary significantly as a function of skin type (pigmentation); however, these values are tightly bound between 50% and 60%, which is lower than expected from the literature.^38–40 The layered model results also appear independent from skin type; however, these values are bounded within the range of 80% to 90%, which is slightly higher than previously published, optically derived values reported in the literature. The interpretation of these results will be discussed in the following section.

Fig. 5

Baseline tissue oxygenation (${\mathrm{StO}}_2$) from both venous (o) and arterial (*) occlusion measurements. Error bars represent the standard deviation of ${\mathrm{StO}}_2$ from every time point acquired during baseline. Each measurement for venous occlusions for each subject was not acquired from the same precise tissue location as for the arterial occlusion for the same subject.

4.

Discussion

The goal of this investigation is to evaluate sources of error and cross-talk based on the interpretation of “quantitative” spectral data collected from a single instrument/measurement geometry. Venous and arterial cuff occlusions were used in these experiments as they will induce a perturbation in the hemodynamics that can be temporally controlled and the generalized hemodynamic effects of these two types of occlusions are well understood, i.e., venous occlusions will induce blood pooling (increase total hemoglobin concentration) and arterial occlusions will result in a deoxygenation of hemoglobin followed by a hyperemic response upon cuff release. Whereas the magnitude of these changes is not controlled and is often subject specific, the central hypothesis of the particular study is based on the assertion that both the melanin concentration and spatial distribution remain invariant to these hemodynamic changes, irrespective of the individual magnitudes of these hemodynamic changes.

Based on this, we have developed an initial metric to evaluate whether a choice of spectral regime or layer model would be most effective in minimizing variance in the determination of melanin when deliberate hemodynamic events are produced. The $\mathrm{\Delta }\text{Melanin}$ metric is a descriptor of how homogeneous fitting routines will induce errors (cross-talk) between depth-distributed chromophores in tissue.

In order to account for subject-specific variance, we can look at the relative improvement (i.e., reduction in $\mathrm{\Delta }\text{Melanin}$) on an individual subject basis. As each analysis method uses data collected at the same time points from the same measurement geometry (instrument configuration), differences between these analyses methods for any individual subject or measurement sequence will only be derived from the method alone. In this case, the layered method consistently produced lower $\mathrm{\Delta }\text{Melanin}$ values on a subject by subject basis than those calculated when using only the NIR portion of the spectrum. This includes the $\mathrm{\Delta }\text{Melanin}$ results under both venous and arterial occlusion conditions ($n=18$). The layered method also outperformed the visible regime $\mathrm{\Delta }\text{Melanin}$ results during venous and arterial occlusions with only one exception. This one case occurred with the darkest skin pigmentation subject under the venous occlusion condition. This case represented the most extreme amount of absorption measured in the study due to the combination of highest degree of skin pigmentation along which the greatest increases of hemoglobin concentration. Even in this case, the layered model value for $\mathrm{\Delta }\text{Melanin}$ remained within 1% of that calculated using just the visible spectrum alone.

$\mathrm{\Delta }\text{Melanin}$ illustrates the magnitude of cross-talk error between melanin and hemoglobin; however, it remains unclear from this metric alone as to whether these errors, which were all $<10\%$, would provide any significant impact on the calculation of actual physiologic parameters. To that end, we also examined the estimation of tissue oxygenation, ${\mathrm{StO}}_2$, at baseline as a means to assess whether these errors in chromophore calculation would present source of concern when attempting to characterize tissue physiology through spectroscopic methods.

NIR regime-derived estimations of ${\mathrm{StO}}_2$ values exhibit a negative correlation with skin type (pigmentation). As shown in Fig. 4, baseline ${\mathrm{StO}}_2$ values will decrease when the amount of pigmentation (melanin) increases. This trend is contrary to expected physiology, as there is no correlation between oxygenation and degree of pigmentation. This is an indication that the NIR regime is particularly sensitive to depth dependence, spectral cross-talk artifacts between the decomposition. Given that the spectral features of hemoglobin species as well as melanin in this spectral region are both relatively weak and featureless, it is not surprising that decomposition methods may find it difficult to completely isolate these chromophores from each other. The total volume of interrogation will also change as a function of wavelength in this spectral regime. Given the depth-distributed nature of these chromophores, the resulting measured absorption spectrum is not simply a linear superposition of basis spectra; rather, it also encodes its relative, wavelength-dependent partial volume contribution toward the total signal detected. This demonstrates that, in this case, superficial pigmentation can affect measurement of tissue oxygenation, as determined by diffuse reflectance-based optical methods.

Using visible wavelengths alone, there was significantly less variation in the determined melanin concentration than was observed in the NIR regime. Given that hemoglobin absorption spectra exhibit more distinct features in the visible than in the NIR, the ability to separate these features from the relatively featureless melanin spectrum is far more robust. The amount of cross-talk present in the least-squares decomposition is significantly reduced. Similarly, the variation in interrogation depth in the visible regime is also lower than is observed in the NIR, indicating that there is little volume-based distortion of the detected spectral features in this regime. Visible light, however, is significantly limited in the depth into which it can interrogate tissue. So while it might be more robust in isolating the optical effects of hemodynamics from that of melanin, it will typically only interrogate the most superficial volumes of the dermis.

Unlike homogeneous model-based spectral decomposition methods used in this study, the layered model approach presents an opportunity to directly address the concerns identified in both visible and NIR spectroscopy. This layer model addresses difficulties associated with superficial pigmentation by segmenting tissue in depth based on the distinct spectral properties specific to layered structures in skin. Not only does this tissue segmentation account for the partial volume effects from melanin, but also this method also accounts for the spectral distortion, realizing that the contribution of melanin absorption at every wavelength will change as a function of the total volume of tissue interrogated and not just as a function of its absorption coefficient at that wavelength. As a result of these features, the layered method is able to provide a distribution of ${\mathrm{StO}}_2$ values that have no apparent correlation with skin pigmentation.

It is worth noting, however, that these ${\mathrm{StO}}_2$ values are higher than those derived from visible spectra alone. As noted in the discussion of the visible regime ${\mathrm{StO}}_2$ results, the depth of penetration of detected light in the visible spectral region is significantly limited ($100\mu \mathrm{m}$) due to the relatively high absorption and scattering values in this window. In contrast, the layered model approach utilizes spectral information from both visible and NIR regimes. Hence, when this method determines hemoglobin concentrations, it is derived over tissue volumes several millimeters deep. The respective volume differences probed by these two methods may provide some justification as to why ${\mathrm{StO}}_2$ values appear so disparate (50% to 60% versus 80% to 90%). It has been suggested that oxygenation within dermal tissue may not be a singular value, but rather a gradient as a function of depth.^41,42 Superficial dermal tissue may exhibit a significantly lower ${\mathrm{StO}}_2$ relative to values found within deeper dermal tissue.

Other groups have developed three-layered models of skin as a direct result.^43–45 Spectral reflectance from skin is interpreted as contributions from the epidermal layer (melanin), as well as upper and lower dermal layers, each of which contain differing concentration of hemoglobin species. The resulting calculations of ${\mathrm{StO}}_2$ values from this study support the results presented by these other groups.

The layered model employed in our study presents a significant opportunity to provide noninvasive, quantitative optical measurements of tissue hemodynamics, independent of an individual’s skin pigmentation. The distinct advantage of this particular measurement paradigm and layered model interpretation is that it provides a quantitative measure of melanin concentration and thickness,²⁷ which is necessary to decouple the optical effects of melanin from that of underlying tissue chromophores. From the SFDS spectroscopic platform, we have the additional opportunity to not just interpret deep dermal tissue oxygenation and hemodynamics but also use the layer model segmentation in order to correct the visible regime data for the partial volume imparted by superficial melanin (thereby reducing the slight pigmentation dependence of visible regime calculation of hemodynamics and oxygenation).

However, for these concepts to advance and become rigorously validated, we must also acknowledge the assumptions and limitations of this current layered model. In its present form, this model assumes that the scattering of tissue is homogeneous in depth. In the model employed here, it is only distinct chromophore absorption features that differentiate layered structures. This model, as used here, also assumes that melanin can be described by a single, invariant absorption spectrum.²⁷

5.

Conclusion

Developing optical methods that directly address both the spectral and structural aspects of tissue are critical toward advancing imaging tools for all skin types. There are distinct advantages and significant improvement in isolating melanin from underlying hemodynamics when using a layered model approach over classic homogeneous model interpretation of skin spectra. In this study, we demonstrate that utilizing a simple, semiempirical layered model approach minimizes cross-talk between melanin and hemoglobin concentration, outperforming methods that utilize visible or NIR spectra alone.