3.1.

Measurement Apparatus

Figure 3 schematically shows the experimental configurations for (a) the imaging system, (b) the tissue-like agar gel phantom, and (c) in vivo experiments with upper arm occlusion. A metal halide lamplight (LA-180Me-R, Hayashi, Japan) illuminated the surface of a sample via a light guide with a ring illuminator. The light source covered a range from 380 to 740 nm. Diffusely reflected light was captured by a 24-bit RGB CCD camera (DFK-21BF04, Imaging Source LLC, USA) with a camera lens (Pentax/Cosmica, Japan; f 16 mm, 1:1.4) to acquire an RGB color image of 640 × 480 pixels. The field of view of the imaging system was 360 mm × 270 mm. The lateral resolution of the images was estimated to be 0.56 mm. This means the best resolution with a nonscattering object. An IR-cut filter in the camera rejects unnecessary longer-wavelength light (> 700 nm). A standard white diffuser with 99% reflectance (SRS-99-020, Labsphere Incorporated, New Hampshire) was used to correct instrument to instrument difference in output of the camera and the spatial nonuniformity of the illumination. The RGB image was an average of 16 video frames and it was stored in a personal computer and analyzed according to the visualization process described above. The computation time for analyzing a typical human skin image was 5 s. RGB filters usually have cross-talk. The regression model used in the present study is based on the results of Monte Carlo simulation. We derived the theoretical diffuse reflectance spectra from the Monte Carlo simulation. A set of theoretical tristimulus values in the CIEXYZ color system, X, Y, and Z, were calculated using the diffuse reflectance spectrum, the spectral distribution of the illuminant, and color matching functions based on Eqs. 2, 3, 4. In this case, the color matching functions correspond to the product of the transmittance spectra of RGB color filters and the spectral sensitivity of the CCD. Like RGB filters, the color matching functions also have cross-talk. We derive RGB values from the theoretical XYZ values using the color transformation matrix L₁ in Eq. 1. Since this method estimates the values of C _m, C _ob, and C _db from the XYZ values, the regression model takes the cross-talk of RGB values into account in the estimation of these chromophore concentrations.

Fig. 3

Experimental configurations of (a) the imaging system, (b) the tissue-like agar gel phantom, and (c) the in vivo experiments with upper arm occlusion.

3.2.

Experiments Using Phantoms

The tissue-like agar gel phantom used in the present study consists of an epidermis layer and a dermis layer. We prepared an agar solution by diluting agar powder (Ina Food Industrial Incorporated, Ina, Japan) with saline at a weight ratio of 0.8%. Intralipid 10% stock solution (Fresenius Kabi AB, Sweden) was added to the agar solution to cause scattering in the phantom. The resultant solution was used as a base material for the epidermis and dermis layers. The final volume concentration of Intralipid in each layer was 1%. We tried to make the phantoms with three-layers similar to the theoretical skin model. However, it was difficult to make a thinned layer representing the stratum corneum by the agar gel in the present study. We investigated the effect of the stratum corneum layer on the RGB values numerically. We made a comparison between the RGB values calculated from the Monte Carlo skin model with and without the stratum corneum layer. The difference in the RGB value between the two models was 2.8 on average. This difference may be of minor importance for simulating the transition from oxygenated blood to deoxygenated blood in dermis. Thus, we ignored the stratum corneum layer in the phantom experiments. A coffee solution was introduced as a substitute for melanin into the base material, and this mixture was used to make the epidermis. Therefore, we used the variable C _c for the epidermis layer in the phantom experiments. Several materials, such as India ink, have been used to mimic the absorption properties of melanin.²⁶ We first tried to mimic the absorption spectrum using an India ink solution. We investigated the absorbance spectrum of India ink (R591217, Rotring, Germany) based on the transmittance measurement in the preliminary experiment. Figure 4 shows the measured absorption spectra of India ink solution and coffee solution. The absorption spectrum of melanin²¹ is also presented in Fig. 4. The absorption spectrum of the India ink solution decays linearly as the wavelength increases, whereas that of melanin decays exponentially as the wavelength increases. Therefore, it was difficult to reproduce the diffuse reflectance spectra of human skin over the visible wavelength range using India ink. Thus, we used a coffee solution to mimic the absorption spectrum of melanin. The coffee solution contained a brown pigment called melanoidin.²⁷ The absorption spectrum of melanoidin^{28, 29} has been reported to be similar to that of melanin. We observed the absorbance spectrum of coffee solution within a cuvette based on transmittance measurements. As shown in Fig. 4, the absorption spectrum of the coffee solution decays exponentially as the wavelength increases. An oxygenated dermis was made by adding a small amount of fully oxygenated horse blood having a 44% hematocrit to the base material. We established oxygenation of the horse blood by gassing it with 95% O₂, 5% CO₂.

Fig. 4

Absorption spectra of India ink solution (dashed line) and coffee solution (solid line) obtained from the transmittance measurements, and absorption spectrum of melanin (from Ref. 21). All the absorption spectra are normalized at 500 nm.

The volume concentration of the coffee solution in the epidermis layer was set to be C _c = 5.0%, whereas that of oxygenated blood in the dermis layer was set to be C _tb = 0.4%. Therefore, the volume concentrations of deoxygenated blood and total blood and the oxygen saturation were assumed to be C _db = 0.0%, C _tb = 0.4%, and SO₂ = 100%, respectively. All of these layers were hardened in various molds having the required thickness and size by cooling at approximately 5.5ºC for 30 min. The thickness of the epidermis and dermis layers were made to be 1.0 and 5.0 mm, respectively, whereas the area of each layer was 26 mm × 23 mm. These layers were then stacked to obtain a layered phantom.

Empirical formulas for the phantom should also be prepared. The Monte Carlo simulations for the phantom conditions generated spectral diffuse reflectance samples ranging from 400 to 700 nm. The optical parameters of μ _s(λ) and μ _a(λ) for the phantom were determined by the inverse Monte Carlo method for measured diffuse reflectance and total transmittance.^{8, 30, 31} Optical properties for the skin tissue phantoms estimated by the inverse Monte Carlo method are shown in Figure 5. The refractive index n for each layer was assumed to be 1.33 at three wavelengths. The anisotropy factor g of Intralipid was derived from the equation reported in literature³² as,

Fig. 5

Optical properties for the skin tissue phantoms estimated by the inverse Monte Carlo method.

The proposed method assumes that the scattering coefficients μ _s(λ) and the thickness t of the skin tissue are known, and typical values reported in the literature are used to derive the empirical formulas. However, these values have subject-to-subject variability and may change due to the physiological or structural condition of the tissue. In order to investigate the effect of variation in scattering coefficients and thickness of the epidermis on the measurements, we also performed experiments on the phantoms for various values of μ _s(λ) and t. For test samples, the values of μ _s(λ) in the epidermis layer were set with a 50% decrement or increment of the typical value, whereas those of t in the epidermis layer were set with a 50% decrement or increment.

3.3.

Upper Arm Occlusion Experiments

A pressure cuff was applied to the upper arms of 14 Japanese subjects (10 males and four females, mean age: 25±8 yr), as shown in Fig. 3c. A SPG (EC6, D.E. Hokanson, Washington) and a rapid cuff inflator (E-20, D.E. Hokanson, Washington) were used to measure in vivo forearm volume change ΔV/V ml/100 ml, which has been used to evaluate limb blood flow.^{33, 34} The SPG is a device that measures the volume changes in limbs and digits using a gauge consisting of a mercury-filled silastic rubber tube. The volume change of tissue induced by the blood flow causes the gauge to stretch or contract. The SPG can determine the volume change based on the variation of electrical resistance caused by the stretching or contraction of the gauge.³⁵ Measurements of limb blood flow at 50 mm Hg have been performed in order to estimate physiological indices of blood circulation such as arterial inflow rate, venous capacitance, and maximum venous outflow,³⁵ whereas those at 250 mm Hg have been performed to investigate the relationship between vasodilatation response and endothelial functions for a peripheral blood vessel.^{34, 35}

During the measurements, the subjects sat with their hands placed on a sample stage at approximately heart-level. After resting for 300 s, image acquisition and SPG recording were started and continued for 640 s at 4 s intervals. After 40 s of control, the cuff was inflated to 50 or 250 mm Hg for 300 s by a rapid cuff inflator, which was then deflated for 300 s. Inflation of the cuff to 50 mm Hg prevents blood flow from leaving the measurement site but does not hinder arterial inflow, whereas that at 250 mm Hg blocks both venous outflow and the arterial inflow. Analyses of the RGB images and the forearm volume change ΔV/V were performed offline after the measurements were completed. Three regions of interest (ROIs) of different sizes were placed in different parts of each image, as shown in Figure 3c. The sizes of ROI1, ROI2, and ROI3 were 250 × 25 pixels, 150 × 150 pixels, and 100 × 100 pixels, respectively.

4.1.

Experiments Using Phantoms

Figure 6 shows typical images obtained from a phantom, four sides of the dermis layer of which were immersed in Na₂S₂O₄-added saline, at different time points during the measurement. Note that the edge of the RGB image is dark before the application of Na₂S₂O₄ (0 min). This is primarily due to the light being less diffusive at the edge than in the center. Similar artifacts were also observed in the estimated images of C _c, C _ob, C _db, C _tb, and SO₂ Therefore, the results for the ROIs (white squares) in the images are discussed here. A correction map based on the spatial distribution of diffuse reflected light intensity from the phantom may be promising to correct for the edge effects. This issue should be investigated in a future work. The images of C _db and C _ob altered inversely from the edge to the center because only the sides of the phantom were surrounded by Na₂S₂O₄-added saline. Consequently, the reduction in SO₂ covered most of the image at the end of measurement (t = 90 min). The images of C _tb and C _c remained approximately constant because they are independent of the oxygen state of hemoglobin. The time courses of the average values over ROIs in the images of C _c, C _ob, C _db, C _tb, and SO₂ shown in Fig. 6 are plotted in Fig. 7. At the beginning of measurement, the values of C _c, C _ob, C _db, C _tb, and SO₂ were estimated to be 5.8±2.7%, 0.36±0.07%, 0.09±0.01%, 0.45±0.16%, and 84±19%, whereas the given values were 5.0%, 0.4%, 0.0%, 4.0%, and 100%, respectively. During measurement, C _ob decreased and C _db increased, whereas C _tb and C _c were almost unchanged, as should be expected. The value of SO₂ decreased gradually and finally fell to approximately 15%. The results obtained from the phantom experiment demonstrate the usefulness of the method in visualizing quantitatively the transition from oxygenated blood to deoxygenated blood in the dermis.

Fig. 6

Typical images obtained from a phantom, the dermis layer of which was immersed in Na₂S₂O₄-added saline (from top to bottom: RGB image, C _ob, C _db, C _tb, SO₂, and C _c) for different time points during measurement.

Fig. 7

Time courses of C _c, C _ob, C _db, C _tb, and SO₂ averaged over the ROIs (white squares) in Fig. 6.

Table 1 lists the relative errors in C _c, C _ob, C _db, C _tb, and SO₂ for two different variations in μ _s(λ) and t of the epidermis layer. The estimated values of C _c, C _ob, C _db, and C _tb decrease as the variation in μ _s(λ) changes from negative to positive. This is because the absorption by each chromophore is estimated to be small due to the increase in O(λ) associated with the increase in μ _s(λ). The estimated value of C _c decreases as the variation in t changes from positive to negative, whereas that of SO₂ increases. This is due to the following consideration. As the thickness of the epidermis layer becomes small, the probability that light is absorbed in it will decrease. Thus, the value of C _c will be underestimated if the thickness of the epidermis is smaller than the typical value. On the other hand, more light will penetrate into the dermis layer as the epidermis layer becomes thinner. In this case, the diffuse reflectance will be affected by the absorption of hemoglobin in the dermis layer more strongly than that of the coffee solution in the epidermis. In addition, the hemoglobin strongly absorbs the light at the shorter wavelength. Therefore, the increase in diffuse reflectance at the longer wavelength is larger than that at the shorter wavelength. This may cause over- and underestimation of C _ob and C _db, respectively. Consequently, the oxygen saturation SO₂ will be overestimated if the thickness of the epidermis is smaller than the typical value.

Table 1

Relative errors of C c, C ob, C db, C tb, and SO2 for different variations in μ s(λ) and t of the epidermis layer for the phantom.

4.2.

Upper Arm Occlusion Experiments

Figure 8 shows the forearm volume change ΔV/V measured by the SPG for cuff pressures of (a) 250 mm Hg and (b) 50 mm Hg. In Fig. 8a, ΔV/V first increases as a result of cuff inflation. A second increase due to reactive hyperemia is evident after cuff deflation. In Fig. 8b, ΔV/V rises quickly after the inflation of the cuff, and the rate of increase in ΔV/V then slows. A rapid decrease in ΔV/V occurred after deflation of the cuff. Figure 9 shows an example of the in vivo results obtained from one single subject during cuff occlusion at 250 mm Hg. The time courses of C _m, C _ob, C _db, C _tb, and SO₂ averaged over the area for the ROIs (white squares) in Fig. 9 are shown in Figs. 10a, 10b, 10c, 10d, 10e. Figure 11 shows the time courses of C _m, C _ob, C _db, C _tb, and SO₂ averaged over the three ROIs for all 14 subjects during cuff occlusion at 250 mm Hg. The average values of C _m and C _tb were 4.3±1.2% and 1.1±0.4%, respectively, in pre-occlusion (control), which are close to typical values reported in the literature.^{2, 11, 13} The average value of 71±16% for SO₂ in pre-occlusion is lower than typical arterial SO₂, which ranges from 90% to 98%. Typical value of venous SO₂ is almost 60%. The value of SO₂ measured by this method represents oxygen saturation for the mixture of arterio-venous blood. Almost 75% of the total blood volume in whole body is contained within the veins and venules, whereas 25% of it is contained within the arteries and arterioles. Assuming that the blood volume ratio of venules and arterioles in the skin tissue is similar to that of whole body, and the oxygen saturation of arterial blood under the normal condition is 96%, the oxygen saturation of skin tissue is calculated to be 69%. This value is close to the average values of 71±16% for SO₂ obtained by this method in preocclusion. During cuff occlusion, C _ob and C _db decreased and increased, respectively. The value of SO₂ exhibited the well-known deoxygenation curve, in which the oxygen saturation falls exponentially. The slight increase in C _tb probably has a physiological cause. This is because, during occlusion, the venous outflow is reduced more than the arterial inflow. This is supported by the results for ΔV/V shown in Fig. 8a. After the cuff was deflated, both SO₂ and C _tb increased sharply and then gradually returned to their normal levels. The error ranges for each subject were taken into account for the calculation of the error range for the trace in Fig. 11. Therefore, interindividual differences in response times and levels may have caused some skewing of traces in Fig. 11.

Fig. 8

Time courses of forearm volume changes ΔV/V measured by SPG during upper arm occlusion at (a) 250 mm Hg and (b) 50 mm Hg. The error bars show the standard deviations (n = 14).

Fig. 9

Typical images of in vivo measurements during cuff occlusion at 250 mm Hg (from top to bottom: RGB image, C _ob, C _db, C _tb, SO₂, and C _m).

Fig. 11

Time courses of (a) C _ob, (b) C _db, (c) C _tb, (d) SO₂, and (e) C _m averaged over the three ROIs for all 14 subjects during cuff occlusion at 250 mm Hg. The error bars show the standard deviations (n = 14).

Figure 12 shows an example of in vivo results obtained from one single subject during cuff occlusion at 50 mm Hg. The time courses of C _m, C _ob, C _db, C _tb, and SO₂ averaged over the ROIs (white squares) in Fig. 12 are shown in Figs. 13a through 13e. Figure 14 shows the time courses of C _m, C _ob, C _db, C _tb, and SO₂ averaged over the three ROIs for all 14 subjects during cuff occlusion at 50 mm Hg. The first increase in C _ob appeared after the cuff was inflated, which caused a temporal increase in SO₂, probably due to the blockage of venous outflow and the continuous arterial inflow. After peaking, C _ob became constant, whereas C _db increased during occlusion. These changes in C _ob and C _db indicate the decrease of the arterial inflow rate and the deoxygenation of hemoglobin resulting from the consumption of oxygen by the tissue, respectively. The rapid decreases in both C _ob and C _db immediately after the deflation of the cuff suggest the outflow of venous blood. The tendency of the response in C _tb to the upper arm occlusions at 50 mm Hg corresponds to the results for ΔV/V shown in Fig. 9b. Note that, whereas C _db decreases monotonically, the change in C _ob after deflation of the cuff has three separate phases: an initial rapid decrease, a slight increase, and then a gradual decrease. This means that the reactive response of arterial blood can be caused by the occlusion at not only 250 mm Hg but also 50 mm Hg. However, the reactive response at 50 mm Hg cannot be observed in the results for the total blood volume, i.e., C _tb and ΔV/V, because the slight variation in C _ob was hidden by the change in deoxygenated blood that decreases dramatically. This indicates the possibility of using this method to evaluate the post-occlusion reactive hyperemia at a pressure of 50 mm Hg. Despite the remarkable changes in C _ob, C _db, C _tb, and SO₂, the value of C _m, which is independent of temporary hemodynamics, remained almost unchanged during the measurements, as shown in Figs. 11, 14. The results shown in Figs. 10, 13 demonstrate that the use of ROIs of different sizes and in different parts of the hand would not change the results significantly. Although there are a number of artifacts due to the shadow originating from the curved and irregular surfaces of the hand, the lateral distributions of C _m, C _ob, C _db, C _tb, and SO₂, and their responses to the occlusion were successfully visualized.

Fig. 10

Time courses averaged over the ROIs (white squares) in Fig. 9 of (a) oxygenated blood C _ob, (b) deoxygenated blood C _db, (c) total blood C _tb, (d) oxygen saturation SO₂, and (e) melanin C _m. The error bars show the standard deviations for each ROI.

Fig. 12

Typical images of in vivo measurements during cuff occlusion at 50 mm Hg (from top to bottom: RGB image, C _ob, C _db, C _tb, SO₂, and C _m).

Fig. 13

Time courses averaged over the ROIs (white squares) in Fig. 12 of (a) oxygenated blood C _ob, (b) deoxygenated blood C _db, (c) total blood C _tb, (d) oxygen saturation SO₂, and (e) melanin C _m. The error bars show the standard deviations for each ROI.

Fig. 14

Time courses of (a) C _ob, (b) C _db, (c) C _tb, (d) SO ₂, and(e) C _m averaged over the three ROIs for all 14 subjects during cuff occlusion at 50 mm Hg. The error bars show the standard deviations (n = 14).

Figure 15 shows images obtained from a Japanese subject of the inner forearm skin including a freckle. A high melanin concentration is observed in the area of the freckle and the estimated values of oxygenated and deoxygenated blood are extremely low and high, respectively. The RGB values in high-melanin areas of freckles will be very small at very low resolution, and the conversion to XYZ color space could compound likely artifacts in measurement. In this case, the conversion from RGB color space to XYZ color space may cause misestimation of oxygenated and deoxygenated blood in the dermis. The results shown in Fig. 15 imply a limitation of the method in measuring the hemodynamics of people with darker skin colors. Experiments involving individuals of African or Indian descent should be performed in the future.

Fig. 15

Images of the inner forearm skin including a freckle obtained from a Japanese subject. (a) RGB image, (b) melanin C _m, (c) oxygenated blood C _ob, (d) deoxygenated blood C _db, (e) total blood C _tb, and (f) oxygen saturation SO₂.

We used a single thickness for the epidermis and the dermis and only one typical wavelength-dependent scattering value and anisotropy for skin in the Monte Carlo simulation to derive the regression Eqs. 8, 9, 10 for the concentration of melanin C _m, oxygenated blood C _ob, and deoxygenated blood C _db. We also assumed a single thickness for the epidermis and the dermis in the Monte Carlo simulation. As expected from the results of phantom experiments, changes in these parameters will affect the regression model. Diffusely reflected light from the skin will increase as the absorption of light in skin becomes small and will increase as the scattering coefficient μ _s(λ) becomes small. Thus, if the scattering coefficients μ _s(λ) of a sample are larger than the typical values used in the Monte Carlo simulation, the regression model will underestimate the chromophore concentrations. Diffusely reflected light from the skin will also increase as the anisotropy factors g(λ) decrease. Therefore, if the anisotropy factors g(λ) of a sample are larger than the typical values, the regression model will underestimate the chromophore concentrations. The probability that light is absorbed by melanin in the epidermis will be higher than the probability that light is absorbed by blood in the dermis as the epidermis becomes thicker. Therefore, if the thickness t of the epidermis is larger than the typical values, then the concentration of melanin in the epidermis or that of blood may be under- or overestimated, respectively. Hence, the measurements could be varied with the changes in the scattering coefficients μ _s(λ), the anisotropy factors g(λ), and the thickness t of each layer. However, the possibility of relative measurements of melanin, oxygenated blood, and deoxygenated blood in the skin tissue could still prove valuable. Studies on the Monte Carlo simulation model with the variations in the scattering coefficients, the anisotropy factors, and the thickness of each layer would strength the method in the future. We will investigate the matrix N ₂ involving those variations further in future work.

This method should integrate all information along the depth direction because it relies on diffuse reflection. Therefore, the imaging system does not limit the depth of the measurement. The RGB camera used in the present study has a spectral sensitivity in the visible wavelength region from 400 to 700 nm. Light at longer wavelengths in this region may penetrate the subcutaneous region. Thus, the measurements shown in Figs. 9, 10, 11, 12, 13, 14 in the original manuscript may reflect the hemodynamics not only of skin but also of the overall limb. This is supported by the fact that the time courses of C _tb shown in Figs. 9 through 14 correlate with the SPG reading, which measures the total blood flow primarily through skeletal muscle, skin, and bone. Cuff pressures of 50 and 250 mm Hg used in the in vivo experiments should usually be higher than the venous blood pressures. Occlusions of the forearm with these pressures, therefore, may induce the onset of edema.^{36, 37, 38} Edema will change the volume of the limb, thereby diluting the bulk concentration of the chromophores and some biological scatterers in the skin tissue. In such a case, under- or overestimation of C _m, C _ob, and C _db may have occurred due to the decrease in the bulk absorption or bulk scattering, respectively. The Monte Carlo simulated spectra in this study were noise-free, while the measured spectra always contain some noise in real applications. Since the matrix N ₂ is determined on the basis of the Monte Carlo simulated spectra, the estimated concentrations of chromophores can be affected by the noise. Although we performed the averaging procedure to reduce the noise component in the RGB image, the results presented in the manuscript may still have some errors due to the noise. It would be reasonable to add some noise in the Monte Carlo simulated spectra to simulate reality better and improve the robustness of the method. It may be helpful to apply partial least square (PLS) regression analysis instead of multiple regression analysis to obtain the matrix N ₂, because PLS regression analysis is particularly suited when the matrix of predictors has more variables than observations, and when there is multicollinarity among predictor variables. The imaging system used in the present study does not eliminate specular reflection from the measured reflected image. Thus, there should be some specular components in the measured RGB signals. On the other hand, we did not take the specular reflectance into account in the Monte Carlo simulation model. This specular component included in the measured RGB values could provide the estimation errors in C _m, C _ob, and C _db. The specular reflection may be eliminated by the use of the polarized imaging technique. These considerations should be investigated in the future.