Spatially resolved diffuse reflectance spectroscopy (SRDRS) has been employed to quantify tissue optical properties and its interrogation volume is majorly controlled by the source-to-detector separations (SDSs). To noninvasively quantify properties of dermis, a SRDRS setup that includes SDS shorter than 1 mm is required. It will be demonstrated in this study that Monte Carlo simulations employing the Henyey-Greenstein phase function cannot always precisely predict experimentally measured diffuse reflectance at such short SDSs, and we speculated this could be caused by the non-negligible backward light scattering at short SDSs that cannot be properly modeled by the Henyey-Greenstein phase function. To accurately recover the optical properties and functional information of dermis using SRDRS, we proposed the use of the modified two-layer (MTL) geometry. Monte Carlo simulations and phantom experiment results revealed that the MTL probing geometry was capable of faithfully recovering the optical properties of upper dermis. The capability of the MTL geometry in probing the upper dermis properties was further verified through a swine study, and it was found that the measurement results were reasonably linked to histological findings. Finally, the MTL probe was utilized to study psoriatic lesions. Our results showed that the MTL probe was sensitive to the physiological condition of tissue volumes within the papillary dermis and could be used in studying the physiology of psoriasis.
© 2016 Optical Society of America
Human skin has a complicated structure and is composed of three primary layers including epidermis, dermis, and subcutaneous tissue . The dermis is about 1 to 5 mm thick, making up the majority of the skin, and contains rich fibrous connective tissue elements and vascular networks. There are two horizontal vascular plexuses located at boundaries within the dermis; one of them is in the papillary dermis and the other is at the dermal-subcutaneous junction . The microcirculation in the upper dermis has various functions such as providing dermal and epidermal nutrition, and regulating of heat loss and temperature control; it is also involved in wound healing and numerous immunologic events. The physiological condition of the microcirculation plexus in the upper dermis has been observed to have connection to several diseases, such as diabetes, psoriasis, and coronary artery disease [2–4]. Thus, efficient and noninvasive retrieval of the condition of the microcirculation residing in the upper dermis in-vivo would facilitate understanding its connection to skin related/unrelated diseases and could be beneficial for clinical diagnosis.
Various types of diffuse reflectance spectroscopy (DRS) systems have been devised for measuring the absorption μa and reduced scattering coefficients μs' of skin. To limit the interrogation volume to superficial region of samples, the source-to-detector separations (SDSs) are commonly reduced to less than 1 mm. For example, Kim et al. recovered the optical properties of samples from the reflectance measured at SDSs ranging from 0.26 to 0.78 mm using a fine-tuned inverse algorithm based on the standard diffusion theory . Liu and Ramanujam employed a probe consisted of flat-tip (SDS of 1.5 mm) and angled (SDS of 0.3 mm) fibers to determine the optical properties of two-layer samples using a Monte Carlo based model . It should be noted that many DRS probes designed for superficial tissue investigation utilized a single source-detector pair configuration, and the sample optical properties were deduced from fitting the reflectance to a photon transport model with spectral constraints. In the essence, algorithms involving spectral constraints require predefined wavelength dependent light-tissue interaction assumptions to determine the absorption and scattering properties of samples; for example, scattering power law of samples and/or prerequisite knowledge of light absorbing molecules (chromophores) of samples [5, 6]. However, adding these constraints in the sample optical property recovery would lead to biased recovery results, since the scattering behavior of turbid samples does not necessary follow the scattering power law and to reduce the complexity of the inverse problem, usually an incomplete chromophore set is considered.
Spatially resolved diffuse reflectance spectroscopy (SRDRS) is one type of DRS variants that have been adopted in quantitative sample optical properties recovery. In a typical SRDRS setup, one fiber would be used for light delivery and several other fibers each with different distances apart from the source fiber would be arranged to collect the diffuse reflectance. From the spatially resolved reflectance, sample optical properties could be derived without applying any spectral constraints, and thus the optical properties of samples can be determined independently at each wavelength without prerequisite knowledge of chromophore composition and/or scattering assumptions. SRDRS was initially employed in deep tissue studies with SDSs ranging from a few mm to about 15 mm [7–10]. Later, SRDRS with much smaller SDSs was employed for superficial tissue studies. For example, Bevilacqua et al. measured the spatially resolved reflectance in the SDS range from 0.3 to 1.4 mm to determine the optical properties of brain tissues, and the maximum probing depth was estimated to be less than 2 mm . Cappon et al. utilized an optical fiber probe to collect reflectance at SDSs of 0.23, 0.59, and 1.67 mm for estimating the superficial samples having optical properties comparable to those of brain tissues; they calculated the average probing depth of the probe was roughly less than 0.5 mm . More recently, we employed a Monte Carlo simulation trained Artificial Neural Network (ANN) to recover skin optical properties from diffuse reflectance measured at SDSs of 1 and 2 mm .
To investigate the effect of sample’s scattering phase function on the diffuse reflectance measured at various SDSs, Bevilacqua and Depeursinge preformed a comprehensive comparison between the most widely used Henyey-Greenstein phase function and several other biological tissue relevant scattering phase functions and found that as the dimensionless reduced scattering, defined as the product of reduced scattering coefficient and SDS, was in the range between 0.5 to 10, the diffuse reflectance was substantially affected by the second moment of the sample’s phase function . Similarly, Calabro and Bigio carried out a numerical study and found that as the dimensionless reduced scattering coefficients varied between 3 and 6, which was normally considered as the diffuse regime, the diffuse reflectance still exhibits significant phase function dependence . From the results of these two independent studies, it could be inferred that, for investigating most biological tissue samples, the Henyey-Greenstein phase function lacks the ability in precisely describing large angle scattering events at SDSs shorter than a few mm.
The influence of scattering phase functions on the diffuse reflectance has been rarely explored in the skin study and/or phantom study. We have found that the diffuse reflectance experimental measured at SDS of 1 mm for samples with skin relevant optical properties cannot be reasonably predicted by using a Monte Carlo model employing the Henyey-Greenstein phase function; the results will be shown and discussed in this paper. We speculate that this phenomenon could be caused by the short photon travel distance at SDS of 1 mm at which the diffuse reflectance is still sensitive to the second moment of scattering phase function . However, for skin applications, using SRDRS in the classical probing geometry with SDS comparable to 1 mm is necessary for constraining the probing depth within dermis. This implies that without the precise knowledge of sample’s scattering phase function it is very difficult to accurately recover the optical properties of upper dermis using the classical SRDRS setup.
Previously, we demonstrated an optical fiber probe configured in the modified two-layer (MTL) geometry that could accurately determine skin optical properties using a diffusion model in the wavelength from 600 to 1000 nm at SDSs shorter than 3 mm . Nevertheless, due to the limitations of the diffusion model, the MTL probe worked in conjunction with the diffusion model recovered unreasonable skin optical properties in the 500-600 nm region and therefore it was difficult to make reasonable interpretation of the recovered skin hemoglobin concentration . In this study, we developed an efficient ANN based inverse model for the MTL probing geometry, and its accuracy was validated using phantom measurements. The applicability of the MTL probing geometry in studying the properties of upper dermis was carefully studied using phantom experiments and Monte Carlo simulations. The utility of the MTL probe in investigating the properties of papillary dermis was further verified through the in-vivo swine skin and human psoriasis studies.
2. Materials and methods
2.1 Measurement setups
Our DRS system, which has been described in detail previously, consists of a supercontinuum source (NKT photonics, Denmark), an 1x4 optical switch (Piezosystem Jena, Germany) and a spectrometer equipped with a back-thinned CCD (QE65000, Ocean Optics, FL) capable of collecting light in the wavelength range from 500 to 1000 nm . Two custom made optical fiber probes were employed in this study and the side view of the probes are shown in Fig. 1. One of the probes configured in the classical SRDRS measurement geometry with SDSs of 1 and 2 mm, and the other probe configured in the MTL geometry utilizing a 1.5 mm thick high scattering layer also had SDSs of 1 and 2 mm. All optical fibers employed in the probes were multimode fibers with 480-μm core diameter and 0.22 numerical aperture.
2.2 Modified two layer diffusion model
In this study, we employed the modified two-layer diffusion equation as one of the mathematical models for the sample optical property determination. Its derivation is briefly described as follows. In a two-layer turbid medium system, the diffusion equation can be written as:
2.3 Monte Carlo method and artificial neural network
Alerstam et al. proposed a highly optimized implementation utilizing the Compute Unified Device Architecture (CUDA) to enhance the computation speed of Monte Carlo simulations by about 3 orders of magnitude . Each GPU-MCML simulation was terminated when ten thousand photon packets was collected by the detector to ensure the simulation precision under 0.5%. In this study, we adopted the original code developed by Alerstam et al. which was referred to as GPU-MCML  to model the photon transport and determine the diffuse reflectance in the classical SRDRS and MTL geometries.
In our previous study, we elaborated the key procedures for efficiently obtaining ANN based inverse models . In this study, two different GPU-MCML models were developed for calculating the diffuse reflectance at 1 and 2 mm SDSs in the MTL and classical SRDRS probing geometries. In the Monte Carlo model, the sample was assumed to be homogeneous and the sample optical properties were varied from 0.01 to 1.00 mm−1 in a 0.01 mm−1 step for μa and varied from 0.55 to 5.00 mm−1 in a 0.05 mm−1 step for μs'. The optical properties of the diffusing layer in the MTL geometry at the wavelength of 700 nm were determined to be 0.0001 mm−1 for μa and 20 mm−1 for μs', and these optical properties along with 1.5 mm slab thickness, 2 cm slab diameter, and refractive index of 1.45 were used in the simulations. The refractive index of sample was 1.33 for liquid phantom studies and 1.43 for skin studies. The Henyey-Greenstein phase function with anisotropy factor of 0.9 was employed in all simulations. Each GPU-MCML simulation was terminated when ten thousand photon packets was collected by the detector. The simulation results were consolidated to form databases for the ANN training. Once the databases were ready, we employed the Matlab (Mathworks, MA) training function “trainbr” based on Bayesian regularization to update the weight and bias values according to Levenberg-Marquardt optimization in this study. The Log-sigmoid transfer function “logsig” was chosen for the hidden layer to train these network models. Generally, we used 50-80 neurons in the inverse ANN training for different source-detector separation groups. The numbers of hidden layers and neurons need to be properly adjusted to optimize the ANN performance. Consequently, the ANN models that had 2-in-2-out configuration were obtained. The ANN models could take the diffuse reflectance measured at 2 SDSs as the input and efficiently determine μa and μs' of the sample under investigation.
2.4 Maximum interrogation depth and average photon travel length
The weighted maximum interrogation depth of each Monte Carlo simulation is determined as , where is the maximum penetration depth of a detected photon packet, Wi is the final weight of a detected photon packet, n is the total number of detected photon packets, and . In addition, the average photon travel length of a simulation was defined as , where L is the total path length of a detected photon packet.
Since the optical properties of samples would affect the interrogation depth and photon travel length, we carried out Monte Carlo simulations with nine sample optical properties combinations, covering optical properties of various skin types in the wavelength range from 500 to 1000 nm, as listed in Table 1 to study the interplay between the sample optical properties and the probing characteristics of the two probes employed in this study. The simulation results will be discussed in Section 3.2.
2.5 Liquid phantoms
We used Lipofundin (Intralipid) 20% (B. Braun Melsungen AG, Germany) as the light scattering agent, water-soluble dye Nigrosin (MP Biomedical Inc., Germany) as the light absorber (0.18 gram of Nigrosin powder dissolved in 1000 ml de-ionized water), and de-ionized water as the solvent to fabricate liquid phantoms. The recipe of the liquid phantoms used in this study is listed in Table 2. These phantoms with different optical properties were designed to mimic the optical properties of skin with dark complexion, such as the swine skin employed in this study, in the 500 to 1000 nm region and their benchmark optical property spectra were defined using the inverse adding doubling method . In the phantom study, we utilized LP1 as the calibration phantom to determine the system response. The process of calibration is elaborated in the following. First, GPU-MCML program was utilized to calculate the theoretical reflectance value of LP1 from 500 to 1000 nm. Next, the experimental reflectance of LP1 was divided by the theoretical reflectance to obtain the system response. The measured reflectance of LP2 was assumed to be influenced by a system response that was the same as the one obtained in the previous step. The purpose of the calibration process was to calculate the LP2 reflectance that was independent of the system response and this was achieved by dividing the measured reflectance of LP2 by the system response. The calibrated diffuse reflectance spectra of LP2 could then be sent to the ANN to recover the sample optical property spectra.
2.6 In-vivo swine skin and human skin measurements
To understand the differences between the performance of the MTL and classical probe in recovering skin optical properties, we employed the two probes to measure the optical properties of the back skin of an eleven-weeks-old Lanyu swine. The animal study was approved by the Institutional Animal Care and Use Committee (No.104142). Five measurements were carried out with each optical fiber probe at a same site. Punch biopsies of the measurement site were performed after the optical measurements.
In addition, we measured the skin of two patients with psoriasis using the two probes. The protocol was approved by the Institutional Review Board (No. ER-100-332) and the written consent was obtained from the subject prior to the measurements. For each patient, we measured the psoriatic lesion site and a normal site apart from the lesion site by about 2 cm, and three measurements were carried out for each site.
3. Results and discussion
3.1 Artificial neural network based inverse model
The key procedures for generating ANN-based inverse models for the classical SRDRS probing geometry have been reported in our previous study in detail . In this study, we employed the same procedures for generating inverse models for the two probing geometries, and they are briefly described as follows. First we generated two training databases for the two probing geometries, and each database contained 18,000 diffuse reflectance determined at SDSs of 1 and 2 mm for various optical property combinations. Next, we thoroughly checked the uniqueness of each diffuse reflectance set (2 spectra at SDSs of 1 and 2 mm) in a database to ensure that there was no two or more different optical property combinations that had the same diffuse reflectance values at this set of SDSs. Since that the intensity fluctuation of our light source was within 0.5% in a one-hour time frame, we defined that any two sets of diffuse reflectance that had differences within 0.5% at both 1 and 2 mm SDSs were not distinguishable and therefore not unique. Failure in passing the uniqueness test would result in inaccurate ANN training results, and to remedy this condition one could reduce the optical property range of the database and/or modify the SDS combination as indicated in our previous study . Given that the databases passed the uniqueness test, they were used for constructing ANN-based inverse models that could take the diffuse reflectance measured at 2 SDSs as the input and efficiently determine μa and μs' of the sample under investigation. Finally, the validation of the ANN models was carried out by calculating the optical property recovery error of 5000 samples with randomly selected optical property combinations that were distinct from those employed in the training database. The percent errors of the recovered optical properties in the MTL probing geometry were less than 2.5% for μa and 0.35% for μs'. On the other hand, the recovery error of the ANN model for the classical SRDRS geometry was less than 6% for μa and 1.5% for μs'.
3.2 Recovery of liquid phantom optical properties
To test the experimental performance of the ANN-based inverse models, we carried out a liquid phantom study. In the experiment we utilized LP1 to determine the system response and calibrate the diffuse reflectance spectra of LP2. The calibrated spectra were sent to the ANNs to recover the optical properties of LP2. Measurement results derived from the MTL and classical SRDRS probing geometries are shown as red and blue lines in Fig. 2, respectively. The maximum deviations of the recovered optical properties from the benchmark values for the MTL geometry were 2.61% and 1.75% for μa and μs', respectively, while for the classical SRDRS geometry the recovery error of μa reached 11.17% at 586 nm. In addition, optical property spectra determined using the standard diffusion equation configured in the MTL geometry are demonstrated as grey lines in Fig. 2; the maximum recovery errors were 18.68% and 66.51% for μa and μs', respectively. We also employed the standard diffusion equation configured in the classical SRDRS geometry to recovery phantom optical properties (data not shown), and the maximum recovery errors were 22.38% in μa and 78.76% in μs'· The advantage of the ANN model over the diffusion model can be readily observed here.
In addition, liquid phantom LP3 that had different optical properties from LP1 was employed as the calibration phantom to understand the effect of calibration phantom optical property on the sample optical property recovery. It was found that using the ANN model configured in the MTL geometry would lead to LP2 maximum recovery errors of 9.64% and 8.25% for μa and μs', respectively when employing LP3 as the calibration phantom. In contrast, when utilizing LP3 as the calibration phantom and the MTL diffusion model for optical property recovery, maximum recovery errors of LP2 were 20% and 98% for μa and μs', respectively. A slight influence of the calibration phantom optical properties on the values of the recovered sample optical properties was observed.
It is worth noting in Fig. 2 that the absorption coefficients recovered in the classical SRDRS probing geometry deviate most from the benchmark values at 586 nm where sample absorption coefficient reaches maximum. We suspected that the loss of optical property recovery fidelity of the ANN model was caused by the compromised accuracy of the Monte Carlo simulation in the high absorption region. To investigate this problem, we took the ratio of the experimentally measured diffuse reflectance of the two liquid phantoms at a certain SDS, and similarly the ratio of the diffuse reflectance of the two phantoms determined using Monte Carlo simulations was calculated. The results are shown in Fig. 3. By taking the ratio of diffuse reflectance of two samples measured at a certain SDS, the contribution of system response introduced by instruments or simulation setups to the reflectance would be cancelled, and thus the ratio spectra derived from experiments and Monte Carlo simulations can be reasonably compared. It can be seen in Figs. 3(c) and 3(d) that the Monte Carlo simulation derived ratio spectra for the classical SRDRS geometry at SDS of 1 and 2 mm deviate from those obtained from experiments especially at high absorption wavelengths. At SDS of 0.5 mm, the difference between the simulated ratio spectra and the experimentally derived ones is even greater than that at SDS of 1 mm (data not shown). On the other hand, as displayed in Figs. 3(a) and 3(b), for the MTL geometry at SDS of 1 and 2 mm, the ratio spectra of LP1 and LP2 determined from experiment data and Monte Carlo simulations agree well with each other, and this in turn results in the accurate sample optical property recovery of the MTL geometry as demonstrated in Fig. 2. Our results suggest that the classical SRDRS geometry with SDSs of less than 2 mm is not an optimal probe design for retrieving optical properties of samples with high absorption. We speculated the compromised performance of the Monte Carlo model configured in the classical SRDRS geometry with SDSs of interest in this study was caused by the fact that the Henyey-Greenstein phase function employed in the Monte Carlo model could not properly model photon scattering behavior when the photon travel distance of detected photons was not long enough which was especially the case for samples with elevated absorption. To verify this point, we determined the photon travel length of detected photon packets in the Monte Carlo simulation for the two probing geometries and the results will be shown in the following subsection.
3.3 Photon travel length and maximum interrogation depth
In the Monte Carlo simulations, the travel length in the tissue sample of detected photon packets was recorded and averaged, and the computed results for the 9 samples listed in Table 1 are plotted in Fig. 4. Each data point in Fig. 4(a) stands for the average of three independent simulations, and standard deviation of the three simulations is also illustrated in the figure. It can be seen in Fig. 4(a) that, for a certain SDS, the MTL probing geometry has longer photon travel lengths than does the classical SRDRS probing geometry for all sample optical properties employed here. The optical properties of sample number 7 to 9 are similar to those of human skin in the 600 to 1000 nm range reported earlier [13, 17, 19], and it can be seen in Fig. 4(b) that for these samples the difference between the photon travel length of these two probing geometries is at least 50%. We found that among all samples listed in Table 1, sample number 2 possesses optical properties closest to those of LP2 at 586 nm where the classical SRDRS performed worst. From Fig. 4(b), it can be seen that for SDS of either 1 or 2 mm, the average photon travel length of the classical SRDRS geometry is shorter than that of the MTL geometry by at least 20% for sample number 2. It can be inferred that the difference in the photon travel length for the two probing geometries could be one of the factors leading to their distinct optical property recovery performance particularly for the samples with high absorption. In addition, we noted in Fig. 4(a) that although the average photon travel length for the classical SRDRS geometry with SDS of 2 mm is longer than that for the MTL geometry with SDS of 1 mm for sample number 2, Figs. 3(a) and 3(d) indicates that the Monte Carlo simulation still matches to the experiment better in the MTL geometry than in the classical SRDRS geometry. We speculated that this phenomenon could be caused by the fact that photons had various oblique injection trajectories into the sample due to strong scattering provided by the 1.5 mm thick high scattering slab; thus there could be less large angle scattering events in the sample for the detected photons and the Henyey-Greenstein phase function could be sufficient for describing photon transport in such a scenario. Substantial validation of our speculation requires in-depth analysis on the interplay between various phase functions, SDS, and probing geometry, and this topic will be included in our future study.
Furthermore, Monte Carlo simulations were carried out in an effort to determine the maximum interrogation depths of the two probing geometries for the nine samples listed in Table 1, and the results are plotted in Fig. 5. In Fig. 5(a), each data point and error bar represent the average and standard deviation of three simulations, respectively. It can be seen in Fig. 5(a) that, for both probing geometries, the maximum interrogation depth increases as SDS altered from 1 to 2 mm, despite the differences in the optical property of samples. When the SDS is 1 mm, the maximum interrogation depth of MTL and classical SRDRS probing geometries varies from 166 to 602 μm and from 433 to 1173 μm, respectively. When the SDS is 2 mm, the average maximum interrogation depth varies from 219 to 972 μm for the MTL structure and from 626 to 1557 μm for the classical SRDRS structure. In general, the interrogation depth increases as absorption and reduced scattering coefficients decrease. The percent difference in the maximum interrogation depth for the two probing geometries for the nine samples are illustrated in Fig. 5(b). It can be seen in Fig. 5(b) that the maximum interrogation depth of classical SRDRS probe structure is at least 40% deeper than that of the MTL structure for low absorption samples and the interrogation depth difference can be as high as 243% for sample number 1 that has high absorption. It is noteworthy that the interrogation depth of MTL probe is always shallower than that of the classical SRDRS probe at a same SDS, and this finding is consistent with the results reported previously . The prominent feature of the MTL probing geometry in which it has longer photon travel length and shallower interrogation depth than the classical SRDRS counterpart was also discussed in detail in our earlier study . In addition, human skin in 600-1000 nm has optical properties comparable to those of sample number 4 to 9, and our simulation results indicate that the MTL geometry with SDS of 2 mm would have maximum interrogation depths less than 972 μm. On the other hand, as shown in Fig. 6(b), at SDS of 2 mm, the interrogation depth of the classical SRDRS geometry is larger than that of the MTL geometry by 40% to 192% in this wavelength range. From the Monte Carlo simulation results, it is obvious that MTL geometry has a clear edge on superficial region interrogation, and its shallow probing depth, which is less than 300 μm for sample number 1 to 3, would be suitable for sensitively studying the properties of papillary dermis. It should be noted that, as elaborated in our earlier study, given the same geometry setup, further decreasing the SDS in the MTL geometry would not reduce the interrogation depth due to strong light scattering provided by the top scattering slab .
3.4 Determination of in-vivo swine skin optical properties
We employed our MTL and classical SRDRS probes to measure the back skin of an eleven-weeks-old swine in-vivo. The measured diffuse reflectance at SDSs of 1 and 2 mm with MTL and classical SRDRS probes were then fed to their corresponding ANN models to compute the skin optical properties from 500 to 1000 nm in a 2 nm step. The computed optical properties of back skin of the swine are depicted in Fig. 6. Red and blue lines represent the optical properties recovered from the MTL probe and the classical SRDRS probe, respectively. The error bars representing the standard deviation of five independent measurements are also depicted in the plots. It can be seen in Fig. 6(a) that μa spectra recovered using the MTL and classical SRDRS probe systems are similar except at the 500-570 nm region where the μa coefficients derived from the MTL probe are consistently higher than those obtained from the classical SRDRS probe. This suggests that in the interrogation region of the MTL probe there was more blood content than it was for the classical SRDRS probe. This interesting observation can be reasonably explained. The blood supply network in the human dermis forms a microcirculatory bed . There are two major horizontally distributed plexuses in the dermis, one is located at the dermal-epidermal junction, and the other one is near the dermal-subcutaneous interface. The tissue volume between the two plexuses has relatively low blood vessel density. Since the MTL probe had a shallower interrogation depth than the classical SRDRS probe, it was supposed that the upper vasculature plexus in the dermis filled the probing region of the MTL probe while the probing region of the classical SRDRS probe consisted of not only the upper vasculature plexus but also the tissue volume at which the blood vessel density was low. Therefore, the MTL probe could see more blood content than it was for the classical SRDRS probe.
Reduced scattering spectra of the swine back skin are illustrated in Fig. 6(b), and it can be observed that the μs' spectra derived from the two probes are distinct in trend and magnitude. In general, the μs' spectra recovered from the MTL probe do not decrease with wavelength as those reported in many literature for various biological tissues, and the oscillation of the reduced scattering coefficient across the wavelength range of interest can be perceived. In contrast, the μs' spectra recovered from the classical probe smoothly decrease with wavelength particularly in the 600-1000 nm range. To investigate the cause of the distinct scattering characteristics derived from the two probes, we took skin biopsies at the measurement site to understand the tissue structure. The skin samples were stained with H&E and divided into four equal sections along the skin depth, and they are displayed in Fig. 7. In the figure, S1 represents the most superficial section containing dermal-epidermal junction and S4 represents the deepest section containing dermal-subcutaneous junction. The dermis thickness at the measurement site of the swine was around 1.5 mm which resembles the dermis thickness of human. It can be observed in Fig. 8 that there were abundant extracellular matrix and sparse cells in the dermis in all sections. We also noted the presence of relatively large structures such as smooth muscle bundles and vascular plexus in the sections S1 and S2, which representing upper half of the dermis, but not so much in S3 and S4, which represent the lower dermis. From the H&E sections, we estimated the diameters of extracellular matrix and the relatively large structures were around 15 μm and 50 μm, respectively.
To further understand the contribution of the observed tissue structures on light scattering, we then fit the reduced scattering spectra displayed in Fig. 6(b) to a Mie scattering function to find the average scatterer sizes and their corresponding volume densities. In the Mie theory, we assumed that light scattering came from three kinds of particles of different sizes accounting for smooth muscle, collagen, and structures that have diameter less than 5 μm such as cell nucleus, with indices of refraction of 1.43, 1.43, and 1.44, respectively [20–23]. The background extracellular fluid refractive index was set to be 1.35 . We employed the Matlab function “lsqcurvefit” in the inverse fitting routine to find the best fit scatterer diameters and volume densities of the three kinds of particles mentioned above. Fitting results are listed in Table 3 and the best fit reduced scattering spectra are plotted in Fig. 6(b) as dotted lines. It is conspicuous from the data listed in Table 3 that in the tissue volume interrogated by the MTL probe the relative large size scatterers of diameter of around 70.1 μm played a dominant role in light scattering, while in the interrogation volume of the classical SRDRS probe, light scattering was mainly contributed from the scatterers of diameter of around 10.8 μm.
Since large light scatterers existed mostly in the upper dermis as can be observed in Fig. 7, it can be inferred from the results listed in Table 3 that the interrogation region of our MTL probe primarily contained the upper dermis. In contrast, large light scattering structures had very little contribution to the classical SRDRS probe’s derived reduced scattering spectrum. The results shown in Table 3 implies that main light scatterer in the interrogation region of the classical SRDRS probe could be collagen bundles, and this suggests that the probe’s interrogation region covered both upper and deep dermis.
3.5 Determination of in-vivo human psoriatic skin optical properties
Psoriasis is an inflammatory skin disease and is typically characterized by well delineated reddish and scaly papules and plaques. It is usually treated with UVB, PUVA, retinoids or cytostatic drugs . It has been reported that the reddish color of psoriatic lesion is due to increased blood volume induced by the dilated capillary vessel size ; thus, in the upper dermis of psoriatic skin, the light absorption property is enhanced due to increased blood volume. Moreover, several groups have observed different dermis collagen conditions for patients with psoriasis. Welzel et al. reported less dense arrangement of collagen fibers on psoriatic lesions . Koivukangas et al. discovered increased dermis collagen synthesis rate at both lesion and normal sites of psoriatic patients . These unreconciled findings suggest that, the scattering coefficient value at the psoriatic lesion site could have an undefined relation to that at the normal site.
We applied the MTL and classical SRDRS probes to determine the optical properties of the lesion and normal sites of two psoriatic patients. The normal site was approximately 2 cm away from the lesion site as shown in Fig. 8. Among the two patients, patient 1 had received treatments with Ustekinumab, Cyclosporine, and topical corticosteroids, and patient 2 had received treatments with Acitretin and UVB photo therapy. The absorption and reduced scattering spectra of patient 1 derived from the two probes are illustrated in Fig. 9.
We can see in Fig. 9(a) that the absorption spectra derived from the MTL probe at the lesion site (red solid line) and the normal site (red dashed line) exhibited distinct magnitudes in the region from 500 to 700 nm, while the absorption spectra of lesion and normal sites derived from the classical SRDRS probe did not demonstrate such a good contrast. The skin absorption spectra were linearly fit to a set of chromophore spectra consisting of the extinction coefficients of oxygenated hemoglobin, deoxygenated hemoglobin, collagen, melanin, and water . The fitting results for the two probes are listed in Table 4. In addition, we calculated the maximum percent deviation between the raw μa and the fit μa spectra, and this value indicating the chromophore fitting quality is also included in Table 4. It was found that the total hemoglobin concentration recovered using the MTL probe at the lesion site was about 2.7 times higher than that at the normal site; on the other hand, the total hemoglobin concentrations of lesion and normal sites obtained from the classical SRDRS probe were comparable. This implies that the MTL probe is more sensitive to the microcirculation in the upper dermis than the classical SRDRS probe. It is interesting to note that the difference between the oxygen saturations at the lesion and normal sites recovered from the classical SRDRS probe was not statistically significant (p>0.05) for patient 1; in contrast, oxygen saturation for the lesion site determined from MTL measurement geometry was significantly higher than that for the normal site. Furthermore, both probes recovered higher melanin concentrations at the lesion site than at the normal site; this result seems to agree with our visual observation of gray dry scales on the lesion site.
It can be seen in Fig. 9(b) that the magnitude of the reduced scattering spectrum of the lesion site recovered using the MTL probe is about 40% lower than that at the normal site. Meanwhile, although MTL probe recovered lower average collagen concentration at the lesion site than at the normal site, these values have no statistical significant difference (p>0.05). On the contrary, the reduced scattering spectra recovered from the classical SRDRS probe at the two sites are statistically indistinguishable (p>0.05), while the recovered collagen concentrations of the lesion site are statistically lower than those of normal site (p<0.05). This in-vivo measurement results suggest that collagen bundle is not the main contributor to the reduced scattering coefficient. Both probes found lower average collagen concentration at the lesion site than at the normal site for patient 1; this finding agrees with that reported by Welzel et al. .
The optical properties of the lesion and normal sites of patient 2 recovered using the two probes are illustrated in Fig. 10. Similar to patient 1, patient 2 had higher absorption magnitude at the lesion site than at the normal site in the wavelength range from 500 to 700 nm. From the chromophore fitting results for the patient 2 listed in Table 5, it can be observed that the trend of increased hemoglobin concentration at lesion site of patient 2 were similar to those of patient 1. However, unlike patient 1, patient 2 recovered higher melanin concentration at the normal site using the MTL probe. On the other hand, it was very likely due to relatively large interrogation depth and low skin melanin concentration of patient 2, we could not recover skin melanin concentration using the classical SRDRS probe.
For patient 2, the average reduced scattering spectra for the lesion site from 700 to 1000 nm were lower than those for the normal site derived from either probe, but their differences were statistically insignificant (p>0.05). Nevertheless, both probes recovered higher collagen concentrations at the lesion site than at the normal site for patient 2. It is worth noting that patient 1 and patient 2 have different trends in skin collagen concentration. From these in-vivo measurement results, it can be seen that despite the fact that the MTL probe is more sensitive to the microcirculation than the classical SRDRS probe, the skin collagen concentrations derived from the two probes share similar trends. In the near future, we will employ the MTL probe to measure more psoriatic patients to understand the skin physiology and its variation induced by treatment.
From phantom and simulation results, we found that classical SRDRS probe with SDSs of 1 and 2 mm had compromised accuracy possibly due to the short photon travel length and the lack of precise knowledge of the sample’s light scattering phase function, and the situation was worsened as the SDS further decreased. For the purpose of studying superficial skin, we found that the MTL probe was advantageous over the classical SRDRS probe in that it has shallower interrogation depth at comparable SDSs. Interestingly, we also found that the photon travel length in the MTL geometry was generally longer than that in the classical SRDRS geometry; therefore, the choice of scattering phase function in the MTL geometry is not as critical as it is for the classical SRDRS geometry. We developed an ANN-based inverse model for the MTL geometry and verified through phantom experiments that it was more capable in accurate recovering the optical properties of high absorbing samples than that configured in the classical SRDRS geometry.
In addition, to understand the performance difference of the MTL and classical SRDRS probes in in-vivo applications, we used the two probes to measure the back skin of an eleven-weeks-old swine. We found that the optical properties recovered from the MTL and classical probing geometries showed very different trends, especially in the results of reduced scattering spectra. Notably, we used Mie scattering theory to fit the reduced scattering spectra derived from the two probing geometries, and found that the average scatterer size in the tissue volume sampled by the MTL probe is much larger than that by the classical SRDRS probe. This result was supported by the facts obtained from skin histology that the swine skin had a depth-dependent structure and many large light scattering structures were observed near the dermal-epidermal junction.
Finally, we measured the psoriatic and normal skin of two patients using the two probes. We discovered that the MTL probe is much more sensitive than the classical SRDRS probe in skin microcirculation. The results of our in-vivo skin studies verify that the MTL probe’s interrogation region is shallower than a classical SRDRS probe at SDSs of 1 and 2 mm. Besides, although we consistently found that elevated hemoglobin concentration at the lesion site for the two patients, the contrast between the collagen concentrations of normal and lesion sites were found to be opposite for the two patients. The results shown in this study accentuate the need for in-vivo functional information monitoring of diseased skin. Since the classical SRDRS probe is not sensitive to the microcirculation near the papillary dermis, we expect that that the MTL and classical SRDRS probing geometries could work in conjunction to investigate the optical properties of upper dermis and full dermis, respectively. This would be useful for many skin applications such as studying the two vascular plexuses in skin.
This research was supported by the Ministry of Science and Technology of Taiwan under Grant No. 104-2221-E-006-179-MY3.
1. L. Goldsmith, S. Katz, B. Gilchrest, A. Paller, D. Leffell, and K. Wolff, Fitzpatrick’s Dermatology in General Medicine, Eighth Edition, 2 Volume set (McGraw-Hill Education, 2012).
3. S. Podtaev, R. Stepanov, E. Smirnova, and E. Loran, “Wavelet-analysis of skin temperature oscillations during local heating for revealing endothelial dysfunction,” Microvasc. Res. 97, 109–114 (2015). [CrossRef] [PubMed]
4. E. Tibirica, E. G. Souza, A. De Lorenzo, and G. M. Oliveira, “Reduced systemic microvascular density and reactivity in individuals with early onset coronary artery disease,” Microvasc. Res. 97, 105–108 (2015). [CrossRef] [PubMed]
5. A. Kim, M. Roy, F. Dadani, and B. C. Wilson, “A fiberoptic reflectance probe with multiple source-collector separations to increase the dynamic range of derived tissue optical absorption and scattering coefficients,” Opt. Express 18(6), 5580–5594 (2010). [CrossRef] [PubMed]
6. Q. Liu and N. Ramanujam, “Sequential estimation of optical properties of a two-layered epithelial tissue model from depth-resolved ultraviolet-visible diffuse reflectance spectra,” Appl. Opt. 45(19), 4776–4790 (2006). [CrossRef] [PubMed]
7. R. Bays, G. Wagnières, D. Robert, D. Braichotte, J. F. Savary, P. Monnier, and H. van den Bergh, “Clinical determination of tissue optical properties by endoscopic spatially resolved reflectometry,” Appl. Opt. 35(10), 1756–1766 (1996). [CrossRef] [PubMed]
8. R. M. Doornbos, R. Lang, M. C. Aalders, F. W. Cross, and H. J. Sterenborg, “The determination of in vivo human tissue optical properties and absolute chromophore concentrations using spatially resolved steady-state diffuse reflectance spectroscopy,” Phys. Med. Biol. 44(4), 967–981 (1999). [CrossRef] [PubMed]
9. T. J. Farrell, M. S. Patterson, and B. Wilson, “A Diffusion Theory Model of Spatially Resolved, Steady-State Diffuse Reflectance for the Noninvasive Determination of Tissue Optical Properties Invivo,” Med. Phys. 19(4), 879–888 (1992). [CrossRef] [PubMed]
10. M. Pilz, S. Honold, and A. Kienle, “Determination of the optical properties of turbid media by measurements of the spatially resolved reflectance considering the point-spread function of the camera system,” J. Biomed. Opt. 13(5), 054047 (2008). [CrossRef] [PubMed]
11. F. Bevilacqua, D. Piguet, P. Marquet, J. D. Gross, B. J. Tromberg, and C. Depeursinge, “In vivo local determination of tissue optical properties: applications to human brain,” Appl. Opt. 38(22), 4939–4950 (1999). [CrossRef] [PubMed]
12. D. J. Cappon, T. J. Farrell, Q. Fang, and J. E. Hayward, “Fiber-optic probe design and optical property recovery algorithm for optical biopsy of brain tissue,” J. Biomed. Opt. 18(10), 107004 (2013). [CrossRef] [PubMed]
13. Y. W. Chen and S. H. Tseng, “Efficient construction of robust artificial neural networks for accurate determination of superficial sample optical properties,” Biomed. Opt. Express 6(3), 747–760 (2015). [CrossRef] [PubMed]
14. F. Bevilacqua and C. Depeursinge, “Monte Carlo study of diffuse reflectance at source-detector separations close to one transport mean free path,” J. Opt. Soc. Am. A 16(12), 2935–2945 (1999). [CrossRef]
15. K. W. Calabro and I. J. Bigio, “Influence of the phase function in generalized diffuse reflectance models: review of current formalisms and novel observations,” J. Biomed. Opt. 19(7), 075005 (2014). [CrossRef] [PubMed]
16. S. H. Tseng, C. Hayakawa, J. Spanier, and A. J. Durkin, “Investigation of a probe design for facilitating the uses of the standard photon diffusion equation at short source-detector separations: Monte Carlo simulations,” J. Biomed. Opt. 14(5), 054043 (2009). [CrossRef] [PubMed]
17. C. K. Hsu, S. Y. Tzeng, C. C. Yang, J. Y. Lee, L. L. Huang, W. R. Chen, M. Hughes, Y. W. Chen, Y. K. Liao, and S. H. Tseng, “Non-invasive evaluation of therapeutic response in keloid scar using diffuse reflectance spectroscopy,” Biomed. Opt. Express 6(2), 390–404 (2015). [CrossRef] [PubMed]
18. E. Alerstam, W. C. Lo, T. D. Han, J. Rose, S. Andersson-Engels, and L. Lilge, “Next-generation acceleration and code optimization for light transport in turbid media using GPUs,” Biomed. Opt. Express 1(2), 658–675 (2010). [CrossRef] [PubMed]
19. A. N. Bashkatov, E. A. Genina, V. I. Kochubey, and V. V. Tuchin, “Optical properties of human skin, subcutaneous and mucous tissues in the wavelength range from 400 to 2000 nm,” J. Phys. D Appl. Phys. 38(15), 2543–2555 (2005). [CrossRef]
20. J. Kim, D. P. Davé, C. G. Rylander, J. Oh, and T. E. Milner, “Spatial refractive index measurement of porcine artery using differential phase optical coherence microscopy,” Lasers Surg. Med. 38(10), 955–959 (2006). [CrossRef] [PubMed]
21. L. T. Perelman, V. Backman, M. Wallace, G. Zonios, R. Manoharan, A. Nusrat, S. Shields, M. Seiler, C. Lima, T. Hamano, I. Itzkan, J. Van Dam, J. M. Crawford, and M. S. Feld, “Observation of periodic fine structure in reflectance from biological tissue: A new technique for measuring nuclear size distribution,” Phys. Rev. Lett. 80(3), 627–630 (1998). [CrossRef]
22. J. L. Solan and K. Laden, “Factors affecting the penetration of light through stratum corneum,” J. Soc. Cosmet. Chem. 28, 125–137 (1977).
23. X. J. Wang, T. E. Milner, M. C. Chang, and J. S. Nelson, “Group refractive index measurement of dry and hydrated type I collagen films using optical low-coherence reflectometry,” J. Biomed. Opt. 1(2), 212–216 (1996). [CrossRef] [PubMed]
24. J. S. Maier, S. A. Walker, S. Fantini, M. A. Franceschini, and E. Gratton, “Possible Correlation between Blood Glucose Concentration and the Reduced Scattering Coefficient of Tissues in the near Infrared,” Opt. Lett. 19(24), 2062–2064 (1994). [CrossRef] [PubMed]
25. V. Koivukangas, M. Kallionen, J. Karvonen, H. Autio-Harmainen, J. Risteli, L. Risteli, and A. Oikarinen, “Increased Collagen Synthesis in Psoriasis in vivo,” Arch. Dermatol. Res. 287(2), 171–175 (1995). [CrossRef] [PubMed]