`Printed in Denmark. All rights reserved
`
`Copyright & Blackwell Munksgaard 2005
`
`Skin Research and Technology
`
`Depth profile of diffuse reflectance near-infrared
`spectroscopy for measurement of water content in skin
`
`Hidenobu Arimoto1, Mariko Egawa2 and Yukio Yamada1,3
`1National Institute of Advanced Industrial Science and Technology, Ibaraki, Japan, 2Life Science Research Center, SHISEIDO, Yokohama, Japan and
`3University of Electro Communications, Tokyo, Japan
`
`Background/purpose: The penetration depth of
`in
`light
`diffuse reflectance near-infrared spectroscopy for measur-
`ing water content in skin is assessed both from theoretical
`and experimental points of view.
`Methods: The Monte Carlo simulation was implemented to
`investigate the dependencies of the light penetration depth
`on a source–detector distance. To compare with the simula-
`tion results, an in vivo experiment
`for water contents of
`skin was performed introducing two different optical fiber
`probes.
`Results: It is found that the minimum separation between a
`source and detector fibers influences largely the meas-
`urement depth. The larger separation leads to a deeper
`measurement depth at a particular wavelength. The mea-
`
`surement depth is also influenced fairly by the absorption
`coefficient of the tissue. The larger absorption coefficient
`results in a shallower measurement depth.
`Conclusion: The correlations between the water contents
`measured by the optical and capacitance techniques were
`discussed. The dependencies of the light penetration depth
`on the source-detector geometry and wavelength are pre-
`sented.
`
`Key words: measurement depth – near-infrared spectro-
`scopy – stratum corneum – water content
`
`& Blackwell Munksgaard, 2005
`Accepted for publication 15 July 2004
`
`NEAR-INFRARED (NIR) spectroscopy provides
`
`information on such aspects as constituents
`concentration. NIR spectroscopy was first devel-
`oped in the 1980s by Norris et al. in order to
`analyze agricultural products (1). In the field of
`biomedical engineering, measuring oxygen sa-
`turation of hemoglobin is one of the most suc-
`cessful applications (2–7). There also exists great
`interest in non-invasive blood glucose measure-
`ment (8), cancer detection (9), body fat measure-
`ment (10), and so on although most of them are
`not yet clinically reliable (11–13).
`in the skin
`Measurement of water content
`based on NIR spectroscopy has also long been
`developed. An exact water content measurement
`technique is required for clinical diagnostics and
`evaluating the efficacy of cosmetics products. An
`easy measurement method of water content is
`essential for atopic cases, for example. The most
`popular technique to determine water content in
`skin is based on measuring electrical properties
`such as capacitance and alternating current con-
`ductivity on the skin surface. An advantage of the
`optical techniques for measuring water content is
`
`its flexibility. An interface of probe light does not
`necessarily have to be made to contact the skin
`surface; therefore, non-occlusive measurements
`can be made. In addition, a measurement point
`can be extended to a two-dimensional area by
`using an image sensor and a series of spectral
`filters.
`Walling reported a spectroscopic experiment
`in vitro by using porcine skin (14). Martin pre-
`sented differences of absorption spectra between
`free and bound water. She showed that
`the
`absorption spectra distinguished four types of
`water in skin: water associated with the lipid
`phase within the stratum corneum, bulk water
`below the stratum corneum, secondary water of
`hydration on stratum corneum keratin, and pri-
`mary water of hydration on stratum corneum
`keratin (15, 16). She also presented profiles of the
`measurement depth in diffuse reflectance spec-
`troscopy experimentally (17). Attas proposed the
`two-dimensional skin hydration mapping sys-
`tem, which enables us to obtain the absorption
`distribution over the skin surface by introducing
`the tunable band pass filter (18). Attenuated total
`
`27
`
`Page 1
`
`VALENCELL EXHIBIT 2014
`IPR2017-00321
`
`
`
`Arimoto et al.
`
`reflectance (ATR) spectroscopy in mid-infrared
`range is also an effective technique to evaluate
`water content. The penetration depth is much
`shallower than that of the NIR diffuse reflectance
`because the ATR measurement utilizes evanes-
`cent wave that localizes on the surface of the ATR
`crystal.
`Although many pioneering works on the op-
`tical water content measurement have been pub-
`lished, a method of interpreting a measured
`result
`is still somewhat unclear. One of
`the
`reasons is that the light penetration depth in
`skin tissue is unknown. The light penetration
`depth depends not only on absorption but also
`on the rate of scattering. In addition, absorption
`and scattering depend on the wavelength of the
`probe light. In many cases, diffuse reflectance
`spectroscopy is implemented by using an optical
`fiber probe that consists of an illumination fiber
`and a detection fiber. It is supposed that the
`geometry of the optical fibers also influences the
`light penetration depth. It is essential to investi-
`gate the influence of the geometry of the optical
`fibers and wavelength of probe light on the
`measurement depth in diffuse reflectance NIR
`spectroscopy. However, an in vitro experiment for
`determining the light penetration depth is quite
`difficult because it is almost impossible to make
`sliced tissue samples with successive thickness.
`In addition, the water content in tissue samples
`must be kept the same as the living tissue because
`the light pathway depends largely on the water
`content. Therefore,
`the simulation method of
`light transport in turbid tissue is very useful to
`estimate the measurement depth (19).
`In this paper, we show the relationship between
`the light penetration depth in the skin tissue and
`experimental conditions such as the geometry of
`the optical fiber probes or wavelength of the NIR
`light. First, theoretical estimations are given by a
`computer simulation. Next, experimental results
`are presented to verify the theoretical evaluations
`on the measurement depth.
`
`Simulation of scattering and absorption
`
`Monte Carlo simulation
`First of all, let us briefly review the Monte Carlo
`method that simulates light scattering and ab-
`sorption in turbid media. Monte Carlo simulation
`is, in general, used to solve a wide variety of
`physical problems such as particle transport
`by using computer-generated random variables
`
`28
`
`with certain probability distributions. The physi-
`cal phenomena are treated as stochastic models.
`This is based on the assumption that the expected
`value of the random variable is equivalent to the
`physical quantity considered. In this paper, we
`introduce a photon packet that propagates in the
`turbid tissue depending on optical parameters
`such as absorption, scattering coefficients, and
`the anisotropy parameter. Here, ‘photon packet’
`behaves as a particle that has continuous energy
`and does not have a volume. The energy of the
`photon packet decreases when a scattering event
`occurs, and is finally recorded as detected light
`power if it reaches a detector. The step size of the
`photon packet movement and the deflection angle
`are given as the probability distribution functions.
`Light is not treated as wavefield in the Monte
`Carlo simulation. In other words, the light pro-
`pagation is expressed as radiant energy trans-
`port. This is because we do not consider phase
`and polarization of a wavefield. The scattering
`events occur repeatedly in the turbid media, and
`the polarization direction and phase of the wave-
`field are then randomized. Therefore, informa-
`tion that the photon packet holds at a particular
`moment is only its location and energy as a scalar
`value. Each time a scattering event occurs,
`dropped energy is recorded as absorbed light
`power at
`that coordinate. Propagation of a
`photon packet is terminated when the photon
`packet comes again off the skin surface or when
`energy of the photon packet comes under a
`certain threshold. Then the next photon packet
`is launched from the light source. Finally, the
`spatial distributions of the absorbed light power
`and the detected light power are obtained. The
`stratified skin tissue to be simulated is character-
`ized by optical properties such as the absorption,
`the scattering coefficients, the anisotropy para-
`meter, and the refractive index at each point.
`Since the Monte Carlo simulation is a statistical
`method, a large number of photons and a large
`amount of computation time are required in
`order to obtain reliable calculation results. In
`our simulation, 100,000–1,000,000 photons are
`launched for one model.
`
`Optical arrangement and skin model
`
`The optical arrangement assumed in this simula-
`tion is shown in Fig. 1. Two optical fibers for
`radiating and detecting light are placed vertically
`on the skin surface. The minimum separation
`
`Page 2
`
`
`
`illummatron
`
`Fig. 1. Geometry of optical fibers assumed for the Monte Carlo
`simulation. A pair of optical fibers is used for radiating and detecting
`light. The minimum distance between two fibers is denoted by Ar.
`
`distance between two fibers is denoted by Ar. The
`reason we do not define the fiber separation by
`the distance between middle axes of two fibers is
`
`that the minimum distance influences the light
`pathway in tissue more than the distance be-
`tween the axes as described in the later section.
`
`The diameters of the optical fibers are varied in
`the simulation for estimating the measurement
`depth. The photon packets are launched uni-
`formly inside the illumination circle because a
`multimode option] fiber is usually used in an
`actual measurement and the light power distri-
`bution over a core is assumed to be almost uni-
`form.
`
`Next, let us consider the structure of skin. As
`mentioned before, the structure of skin is char-
`
`acterized by optical properties. We assume that
`the skin consists of three layers:
`the stratum
`corneum, the epidermis, and the dermis with a
`thickness of 20, 300nm, and 3mm, respectively.
`As is well known, a human body consists almost
`fully of water. Therefore, water absorption is
`dominant over other constituents in the NIR
`
`absorption spectrum measured on skin. So we
`characterize the absorption and the scattering
`coefficients with reference to the water content
`
`in each layer. In the stratum corneum, the water
`content is supposed to change from its surface to
`the interface between the stratum comeum and
`
`the epidermis. In this simulation, we assume that
`the water content changes linearly from 10% to
`80% from the top to the bottom of the stratum
`corneum. The water contents in the epidermis
`and the dermis are, in contrast, assumed to be
`
`uniformly 80%. Based on this assumption, we
`determine the distributions of the absorption and
`scattering coefficients. The absorption coefficient
`of each layer is calculated as a product of the
`
`Page 3
`
`Depth profile NIR spectroscopy
`
`120
`
`1 10
`
`8°
`
`Bottom layer of SC
`
`\
`
`Top layer of SC
`
`[cm'1] 8 L
`
`Absorptioncoefficient
`
`1.4
`
`1.6
`
`.8
`
`2
`
`2.2
`
`Wavelength [pm]
`
`Fig.2. The absorption wefficient used in the Monte Carlo simulation
`as a function of wavelength and the depth of skin. The water
`absorption dominates the spectral profile.
`
`water content and the absorption coefficient of
`pure water. The absorption coefficient of pure
`water was measured in advance by a spectro-
`photometer. The absorption coefficients at the top
`(2 = 0 pm) and the bottom (2 = 20 um) planes of
`the stratum comeum are shown in Fig. 2. The
`absorption coefficient at 0<z <20 pm is calcu-
`lated by the linear interpolation based on the
`assumption of the change in the water content as
`mentioned above.
`
`The scattering coefficient as a function of the
`depth 2 is given considering the water content
`also. With the help of previous reports on the
`values of the scattering coefficient, we assume the
`depth-dependent scattering coefficient of
`the
`stratum comeum as shown in Fig. 3a. A value
`of 0<z <20 pm is given by the linear interpola-
`tion in the same manner as for the absorption
`coefficient. As seen in Fig. 3a, the top surface has
`a larger scattering coefficient.
`In addition, a
`longer wavelength yields a smaller value as is
`well known. The refractive index of 1.37 is used
`
`for all layers in this simulation.
`We suppose that the anisotropy parameter (Fig.
`3b) derived from the scattering phase function
`depends on 2 and wavelength. The values are
`also given with the help of previous reports (20).
`
`Spatial distribution of absorption
`
`The first simulation result is spatial absorption
`distributions in turbid tissue. Four results at the
`
`wavelength of 1.45 pm are shown in Figs 4a—d.
`White spots represent absorption, i.e., the amount
`
`29
`
`
`
`Arimoto et al.
`
`(a) 300
`
`250
`
`§
`
`.5 U"0
`
`§
`
`Top layer of SC
`
`Bottom layer of SC
`
`1 .4
`
`1 .6
`
`1 .8
`
`2
`
`2.2
`
`[cm'1] 8 O
`
`Scatteringcoefficient
`0.1
`
`1.0mm
`
`1.0mm
`
`‘—
`
`Wavelength [pm]
`
`Anisotropy parameter
`
`(b) 0.4
`
`0.3
`
`a) 0.2
`
`1 .4
`
`1.6
`
`1 .8
`
`2
`
`2.2
`
`0.63mm
`
`0.63mm
`
`Wavelength [pm]
`
`Fig.3. (a) The scattering coefficient and (b) the anisotropy parameter
`used in the Monte Carlo simulation as functions ofwavelength and the
`depth of skin. The values were estimated based on the Mie theory.
`
`of reduced light power at each point. The gray
`scale is normalized by the maximum and the
`minimum values of the absorption. The dia-
`meters of the optical fibers in a pair of (a) and
`(b) and a pair of (c) and (d) are, respectively,
`(I) = 300 and 133 um, and the fiber separations Ar
`are, respectively, 30 and 32 um. Note that these
`values are chosen based on the actual optical fiber
`probes that are used in the in viva experiment.
`Two left maps, (a) and (c), were calculated by
`tracing all launched photon packets while two
`right maps, (b) and (d), were calulated only by
`detected photon packets. The results of the ex—
`periment will be shown in the next section. Let us
`go back to the simulation results. Both results
`were obtained by launching 1 million photon
`packets. However,
`the numbers of detected
`photons are 27,090 ((1): 300 um) and 12,922
`((I>=133 pm). In this simulation, the energy of
`photon packet is reduced when a scattering event
`
`30 Page 4
`
`Fig.4. Spatial distributions of absorption simulated by the Monte
`Carlo simulation at a wavelength of 1.45 pm. (a) and (b) are results fiir
`optical fibers with a diameter of 0=300pm, and (c) and (d) are
`results for (I) = 133 pm; (a) and (c) represent the absorption concerning
`all launched photon packets, while (17) and (d) show the absorption
`only by detected photon packets.
`
`occurs. Therefore, it is reasonable to evaluate the
`
`detection efficiency by the sum of the detected
`energy rather
`than the number of detected
`photon packets. The ratios of the detected energy
`to the incident one are, respectively, 0.00371
`((1) = 133 um) and 0.00364 ((1) = 133 pm). An inter-
`esting fact is that the light detection efficiencies
`are almost the same in the two cases, although
`the numbers of detected photon packets are very
`different. This means that the average power of
`detected photon packets for the fiber geometry
`(d) is larger than that in (b).
`Another feature found from the result is that
`
`the absorption density is localized around the
`middle of the source and detector fibers. The
`
`reason for this is easily understood. A photon
`packet that propagates in a long path loses a large
`portion of the incident energy, and is more likely
`
`
`
`Depth profile NIR spectroscopy
`
` .__. 0.8
`
`0.6
`
`0.4
`
`3
`.15.
`S-
`
`E8a g
`
`S3
`
`§ 0.2
`0
`
`0
`
`0
`
`200
`
`400
`
`600
`
`800
`
`1000
`
`Depth [p.rn]
`
`Fig.6. Cumulative absorption derived from the distribution plot in
`Fig. 5. The values of the cumulative absorption are normalized by the
`values at a depth of 1.0 mm.
`
`optical fiber of (I) = 300nm is larger than that of
`(I) = 133 pm.
`In the last subsection, it was shown that the
`
`large amount of absorption was seen around the
`middle of the source and detector fibers. This
`
`implies that the photon packets that take shorter
`pathway give larger contributions to the mea-
`sured absorbance. Now, let us focus on the mini-
`
`mum separation distance Ar (see Fig. 1) between
`the source and the detector fibers. To investigate
`the relationship between Ar and the measure-
`ment depth, we assume another pair of optical
`fibers, i.e., a source fiber with a diameter of zero
`
`and a detection fiber with a diameter of 133 pm.
`Here, an ideal simulation should be made by
`introducing a point source and a point detector
`because our interest is dependency of the mea-
`surement depth on the source—detector separa-
`tion. However, any photon packet cannot be
`detected by the point detector in the Monte Carlo
`simulation. Therefore the detector fiber has a
`
`non-zero diameter so that the appropriate detec-
`tion efficiency is obtained. The separation dis-
`tance was changed stepwise from Ar = 0 to
`180 pm. The cumulative absorptions as functions
`of z are plotted in Fig. 7. It can be seen from the
`plots that the measurement depth becomes large
`as two fibers are separated. When Ar=0, the
`value of 2 that gives a cumulative absorption of
`0.5 is only 20pm. In Fig. 8, the measurement
`depths are plotted against Ar. Here, ’depth’ de-
`noted in the vertical axis is defined as the value of
`
`2 that gives the cumulative absorption of 0.5 in
`
`31
`
`to be terminated before being detected. In other
`words, the short distance between the source and
`
`detector fibers results in the high possibility of a
`photon packet to reach the detection area.
`
`Dependence on fiber separation
`
`In order to discuss the relationship between the
`fiber geometry and the measurement depth, let
`us introduce the depth—absorption plot by inte—
`grating Figs. 4a and b along the horizontal direc-
`tion. The plot is shown in Fig. 5. The amount of
`absorption reduces simply from the skin surface
`to the deeper sites. As seen from the plots, the
`maximum measurement depths are,
`in both
`cases, almost limited to 800—1000um. However,
`we should remember that those absorption dis-
`tributions are plotted by focusing only on the
`detected photon packets. If we trace all photon
`packets launched from the illumination area, the
`absorption distribution is extended to the deeper
`site.
`
`Cumulating the absorption values from z = 0
`will help us to find out the difference in the
`measurement depths between two fiber sizes.
`Figure 6 shows the cumulative absorptions that
`are normalized by values at Z: 1 000nm. The
`cumulative absorption of 0.5 is, for example,
`given at depths of 124 and 156 um, respectively,
`for (I) = 300 and 133 pm. This means that half of
`the total absorption occurs in the region of z < 124
`and z<156um for each fiber size. Hence, it is
`clearly understood from this cumulative absorp-
`tion pots that the measurement depth with the
`
`[a.u.]
`
`Absorption
`
`0
`
`200
`
`400
`
`600
`
`800
`
`1000
`
`Fig.5. Absorption distribution as a function of the depth concerning
`only detected photon packets. The difference between 0 = 300 um and
`133 um seen in this plot is mainly because of the detection efficiency.
`
`Page 5
`
`
`
`Arimoto et al.
`
`[a.u.]
`
`Cumulativeabsorption
`
`0
`
`200
`
`400
`
`600
`
`800
`
`1000
`
`1 .4
`
`1.6
`
`1 .8
`
`2
`
`2.2
`
`Depth [um]
`
`Wavelength [pm]
`
`Fig. 7. Cumulative absorption plots obtained by changing the source
`detector separation Ar.
`
`Fig. 9. Measurement depth as a function of wavelength. The spectral
`profile shows the reverse of the absorption coefficient.
`
`120
`
`100
`
`
`
`Depth[pm] 88
`
`S
`
`0
`
`50
`
`100
`
`Ar[pm]
`
`150
`
`200
`
`Fig.8. Measurement depth against the source detector separation.
`The measurement depth is defined by the median of the cumulative
`absorption.
`
`Fig. 7. This plot clearly shows that larger Ar
`results in a deeper measurement.
`Here, we have an important conclusion con-
`ceming the relationship between the geometry of
`the optical fibers and the measurement depth.
`When the minimum separation distance between
`two fibers is fixed, a larger diameter of the optical
`fiber results in deeper measurement (see Fig. 6).
`On the other hand, when the size of the optical
`fiber is fixed, a larger source—detector separation
`results in deeper measurement (see Figs 7 and 8).
`
`Dependence on wavelength
`
`The measurement depth is supposed to depend
`largely on the wavelength because optical prop-
`erties such as the absorption coefficient,
`the
`
`32 Page 6
`
`scattering coefficient, and the anisotropy para-
`meter are functions of wavelength. To investigate
`dependency of the measurement depth on wave-
`length, the fiber geometry is fixed in this simula-
`tion. The construction of the fibers with the
`
`diameter of 133 pm and a separation of 30m,
`which was shown in the simulation result Fig. 4b,
`is used here. The simulation was conducted in
`
`the spectral range from 1.3 to 2.2 pm with a step
`of 0.01 pm. Then a series of 91 calculation results
`were obtained. The estimated measurement
`
`depth as a function of wavelength is shown in
`Fig. 9. Here, the definition of ’depth’ is the same
`as the last subsection. i.e., the value of 2 that gives
`the cumulative absorption of 0.5. Obviously, the
`profile of the measurement depth is influenced
`by the water absorption spectrum (refer to the
`absorption coefficient shown in Fig. 2). The ab-
`sorbance spectrum of water has a feature of two
`significant absorption bands at 1.45 and 1.92 pm,
`corresponding to the first overtone of the hydro-
`xyl stretch and the combination mode of the
`hydroxyl and HOH band,
`respectively. The
`strong absorption results in a short measurement
`depth. This is because a photon packet rapidly
`loses its energy in the spectral range of strong
`water absorption. As a result, photon packets,
`that take long paths are less likely to be detected,
`and therefore the path length of the detected
`photon packets tends to be short.
`In this section, we investigated the influence of
`the fiber geometry and wavelength on the mea-
`surement depth in the diffuse reflectance spectro-
`scopy based on the Monte Carlo simulation. It
`was found that a shorter distance between the
`
`
`
`Depth profile NIR spectroscopy
`
`Absorbance 0
`
`1 .6
`
`1 .8
`
`2
`
`2.2
`
`2.4
`
`1.2
`
`1 .4
`
`source fiber and the detector fiber results in
`
`the larger
`In addition,
`shallower penetration.
`absorption coefficient also gives a shallower pe-
`netration depth.
`
`Experiment of water content
`measurement
`
`In this section, we report the in viva experiment
`for measuring water content in skin based on the
`diffuse reflectance spectroscopy. To verify the
`simulation results presented in the last section,
`two different optical fiber probes were used to
`measure diffuse reflected spectra from skin. The
`geometries of two fiber probes are the same as
`those used in the simulation for Figs. 4b and d,
`i.e., 300 and 133 pm in diameters and 32 and
`30 um in separations.
`The Fourier transform NIR spectrophotometer
`(JASCO, VIR-9600, Tokyo, Japan) was used to
`collect the absorbance spectra. The spectrophot-
`ometer consists of the light source of a tungsten
`halogen lamp, a beam splitter of Can, and a light
`detector of InGaAs. The light beam from the
`Michelson interferometer
`is
`incident on the
`
`source fiber, and then the diffusely reflected light
`is collected by the detection fiber. The measured
`spectral range and the spectral resolution are,
`respectively, 8000—4000/cm (125—25 um) and 8/
`cm. For interpolating the Fourier spectrum, zero
`padding was made to the original interferogram.
`Finally, 1168 points were obtained between
`8000 and 4000/ cm.
`In order to evaluate the
`
`results of the optical method, water content
`was measured by a capacitance method (Cour-
`age+Khazaka, CORNEOMETER CM825, Koln,
`Germany) simultaneously.
`The spectroscopic measurements were per-
`formed with 13 subjects. Diffuse reflected spectra
`with 300 and 133 um fiber probes, and water
`contents with the capacitance method were mea—
`sured at five or six points on the inner arm of
`each subject. Finally, 75 pairs of data (optical and
`capacitance) for both the 300 and 133m fibers
`were obtained. Twelve randomly selected absor-
`bance spectra are shown in Fig. 10. Here, absor-
`bance in the diffuse reflectance measurement is
`
`defined by —log(1/R) where, R represents the
`reflectance. The OH overtone and combination
`
`bands at 1.45 and 1.95 pm are clearly seen, and
`the spectral profile shows that the water absorp-
`tion dominates in the diffuse reflected absorption
`from the skin surface.
`
`Page 7
`
`Wavelength [pm]
`
`Fig.10. Absorbance spectra mazsumd on inner arm by using the
`optical fiber probe and the Fourier transform NR spectrophotometer
`(I-T NIR).
`
`Residual
`
`variance
`
`1 2 3 4 5 6 7 8 91011121314151617181920
`
`Principal component
`
`Fig. 11. Residual variance obtained in the partial least squares regres
`sion (PLS) analysis for water content ofsla'n. The optical fiber with the
`diameter of 133 um shows a good convergence.
`
`To predict the water content from the NIR
`absorbances, the partial least-squares regression
`(PLS) was introduced. Absorbance values at each
`
`wavenumber are used, as the explanatory vari-
`ables while water contents measured by the
`capacitance method are the objective variables.
`Figure 11 shows the residual variances at each
`principal component. It can be seen from the plot
`that the residual variances of the regression result
`with the optical fibers of <I>= 133nm are lower
`than those of the optical fibers of (I) = 300nm at
`principal components larger than seven. This
`means that the correlation between the capaci-
`tance method and the optical method with the
`optical fibers of (I) = 133 pm takes a higher value
`than that with the optical fibers of (I) =300pm
`
`33
`
`
`
`Arimoto et a1.
`
`
`
`
`
`RH(PLSpredicted)[%]
`
`[v.1
`
`
`
`HH(PLSpredicted)
`
`20
`
`30
`
`40
`
`50
`
`RH (Capacitative) [%]
`
`Fig. 12. Correlation plots betwwn the capacitance method and the
`optical method with the optical films of (a) ¢=3wimr and (b)
`133 pm.
`
`when the PLS prediction is performed with
`principal components more than seven. The cor-
`relation plots with 10 principal components with
`optical fibers of (I) = 300 um and of (I) = 133 um are
`shown in Fig. 12a and b, respectively. The ab-
`scissa and ordinate represent water content ob-
`tained by the capacitance and optical methods,
`respectively. The correlation with the optical
`fibers of
`(D = 133 pm is 0.828, while that of
`(I) = 300 pm is 0.799. As pointed out above, the
`correlation with the optical fibers of (I) = 133 pm is
`higher, although the difference is not very large.
`This experimental result verifies the simulation
`result presented in the last section with Fig. 6, i.e.,
`the correlation with optical fibers of (I) = 133 pm is
`higher because the measurement depth is shal-
`lower than that with optical fibers of (I) = 300 um.
`
`34 Page 8
`
`At a wavelength of 1.45 pm, the measurement
`depths with the optical fibers of (I): 300 and
`133m are, respectively, 156 and 124 um, while
`the measurement depth of
`the capacitance
`method is 40m as described before. The mea-
`surement depth of the optical method is much
`deeper than that of the capacitance method;
`however, the optical fiber of (I) = 133 um gives a
`measurement depth closer to that of the capaci-
`tance method than the optical fiber of (I) = 133 pm.
`An ideal experiment should be carried out by
`with changing the separation between the source
`and detector fibers to verify the dependence of
`the measurement depth on the fiber separation
`although this is not presented in this paper.
`However, the results of both the simulation and
`
`experiment show that closer source—detector se-
`paration and thinner optical fiber make the mea-
`surement depth shallower than demonstrated in
`this section. This will enable us to make the
`measurement area localize more at the stratum
`corneum.
`
`Conclusions
`
`The measurement depth of the NIR diffuse re-
`flectance spectroscopy on skin was investigated
`and discussed quantitatively. It was found that
`the measurement depth depended fairly on the
`geometry of the optical fiber probe used for
`radiating and detecting light on skin. The Monte
`Carlo simulation showed that a larger source—
`detector separation and a thicker optical fiber
`resulted in the deeper measurement depth. In
`vivo measurement was also demonstrated in
`
`order to verify the simulation results. The optical
`and the capacitance measurements of water con-
`tent in skin were performed simultaneously. The
`correlation that was calculated by the P15 regres-
`sion analysis showed that the predicted water
`content of skin with the thinner optical fiber was
`closer to the capacitance result than the thicker
`optical fiber.
`
`References
`
`1. Williams P, Norris K. Near-Infrared Technology In the
`Agricultural and Food Industries. Amer Assn of Cereal
`Chemists, 2 edition, 2001.
`2. Riga! Jd, Losch M], Bazin R, Camus C, Sturelle C,
`Dscamps V, Lévéque IL. Near-infrared spectroscopy:
`a new approach to the characterization of dry skin.) Soc
`Cosmet Chem 1993; 44: 197 209.
`
`
`
`3. Cheong WF, Prahl SA, Welch AJ. A review of the optical
`properties of biological tissues. IEEE J Quant Electron
`1990; 26: 2166 2185.
`4. Ya-Xian Z, Suetake T, Tagami H. Number of cell layers of
`the stratum corneum in normal skin-relationship to the
`anatomical location on the body, age, sex, and physical
`parameters. Arch Dermatol Res 1999; 291: 555 559.
`5. Woo YA, Jhii-Weon Chun IK, Kim HJ. Development of a
`method for the determination of human skin moisture
`using a portable near-infrared system. Anal Chem 2001;
`73: 4964 4971.
`6. Khalil OS, Yeh Sj, Lowery MG, Wu X, Hanna CF, Kantor
`S, Jeng TW, Kanger JS, Bolt RA, Mul FFd. Temperature
`modulation of the visible and near infrared absorption
`and scattering coefficients of human skin. J Biol Opt
`2003; 8: 191 205.
`7. Hoshi Y, Tamura M. Dynamic multichannel near-infra-
`red optical imaging of human brain activity. J Appl
`Physiol 1993; 75: 1842 1846.
`8. McNichols RJ, Cote´ CL. Optical glucose sensing in bio-
`logical fluids: an overview. J Biomed Opt 2000; 5: 5 16.
`9. Zeng H, Petek M, Zorman MT, McWilliams A, Palcic B,
`Lam S. Integrated endoscopy system for simultaneous
`imaging and spectroscopy for early lung cancer detec-
`tion. Opt Lett 2004; 29: 587 589.
`10. Conway JM, Norris KH, Bodwell CE. A new approach
`for the estimation of body composition: infrared inter-
`actance. Am J Clin Nutr 1984; 40: 1123 1130.
`11. Arimoto H, Tarumi M, Yamada Y. Temperature-insensi-
`tive measurement of glucose concentration based on
`near infrared spectroscopy and partial
`least squares
`analysis. Opt Rev 2003; 10: 74 76.
`12. Arimoto H, Tarumi M, Yamada Y. Instrumental require-
`ments for non-invasive blood glucose measurement
`using NIR spectroscopy. Opt Rev 2003; 10: 161 165.
`13. Egawa M, Fukuhara T, Takahashi M, Ozaki Y. Determin-
`ing water content in human nails with a portable near-
`infrared spectrometer. Appl Spectrosc 2003; 57: 473 478.
`
`Depth profile NIR spectroscopy
`
`14. Walling PL, Dabney JM. Moisture in skin by near-
`infrared reflectance spectroscopy. J Soc Cosmet Chem
`1989; 40: 151 171.
`15. Martin KA. Direct measurement of moisture in skin by
`NIR spectroscopy. J Soc Cosmet Chem 1993; 44: 249 261.
`16. Martin K. In vivo measurements of water in skin by near-
`infrared reflectance. Appl Spectrosc 1998; 52: 1001 1007.
`17. Martin K. Depth-profiling of skin by near infrared
`spectroscopy. In: Davies AMC, Williams P, eds. Near
`infrared spectroscopy: the future waves. The Proceed-
`ings of the 7th International Conference on Near Infra-
`red Spectroscopy. Montr´eal: NIR Publications, 1995:
`328 333.
`18. Attas M, Schattka TPB, Sowa M, Mantsch H, Zhang S.
`Long-wavelength nearinfrared spectroscopic imaging
`for in-vivo skin hydration measurements. Vibrat Spec-
`trosc 2002; 28: 37 43.
`19. Iino K, Maruo K, Arimoto H, Hyodo K, Nakatani T,
`Yamada Y. Monte carlo simulation of near infrared
`reflectance spectroscopy in the wavelength range from
`1000 nm to 1900 nm. Opt Rev 2003; 10: 600 606.
`20. Staveren HJv, Moes CJM, Marle Jv, Prahl S, Gemert Mv.
`Light scattering in intralipid-10400 1100 nm. Appl Opt
`1991; 30: 4507 4514.
`
`Address:
`Hidenobu Arimoto
`1-2-1 Namiki
`Human-Biomedical Institute
`National Institute of Advanced Industrial Science and Technology
`Tsukuba
`Ibaraki 305-8564
`Japan
`
`Tel: 181-29-861-7880
`Fax: 181-29-861-7275
`e-mail: arimoto-h@aist.go.jp
`
`35
`
`Page 9
`
`