2.2.1.

MPTflex clinical tomograph

The laser-scanning-based clinical multiphoton tomograph, MPTflex (JenLab GmbH, Germany) consists of a compact, turn-key femtosecond laser (MaiTai Ti:Sappire oscillator, sub-100 fs, 80 MHz, tunable 690–1020 nm; Spectra Physics, Mountain View, California), an articulated arm with NIR optics, and a beam scanning module. The system has two photomultiplier tube detectors employed for parallel acquisition of TPEF and second-harmonic generation (SHG) signals. A customized metallic ring taped on the subject’s skin attaches magnetically to the objective holder in the articulated arm, minimizing motion artifacts. The excitation wavelengths used for this study were 790 nm for the acquisition of the cross-sectional ($\mathit{xz}$) images and 880 nm for the acquisition of the z-stacks of horizontal ($\mathit{xz}$) images. The z-stacks were obtained by moving the objective in the z direction, thus scanning at different depths in the skin. The TPEF signal was detected over the spectral range of 410–650 nm, while the SHG signal was detected over a narrow spectral bandwidth of 385–405 nm through emission filters placed in the TPEF and SHG detection channels, respectively. The $\mathit{xz}$ sections were $512\times 512$ pixel images acquired at $\sim 6\mathrm{s}/\mathrm{frame}$. The $\mathit{xz}$ sections were $1024\times 1024$ pixel images acquired at $\sim 30\mathrm{s}/\mathrm{frame}$. We used a Zeiss objective ($40\times$, 1.3 NA, oil immersion) for focusing into the tissue.

We acquired the MPM data using two scanning modalities: (1) $\mathit{xz}$ scanning with z-stacks generated using en-face images from the stratum corneum to the epidermal–dermal junction (EDJ). The excitation wavelength selected for this acquisition mode was 880 nm in order to maximize the melanin contrast against TPEF signals from other components of the epidermis (keratin, NADH/FAD). This was selected due to the fact that melanin has a broad absorption spectrum (maximum around 300 nm gradually decreasing toward the NIR) and a broad TPEF emission spectrum that peaks around 620–680 nm.^19,20 The melanin appears in the MPM images as bright fluorescence spots in the cellular cytoplasm that represents aggregates of melanosomes. We used the data obtained by this scanning modality to measure the melanin distribution as a function of depth and the melanin volume fraction in the epidermis. Optical sections of about $115\times 115\mu {\mathrm{m}}^2$ at different depths ranging from 0 to about $60\mu \mathrm{m}$ ($5\mu \mathrm{m}$ steps) were obtained. As the optical section is limited to a small scan field, to improve the overall characterization of the examined skin, we acquired five stacks of images on each site (volar arm and dorsal forearm) for all subjects. (2) $\mathit{xz}$ scanning which provided cross-sectional, “histology-like” images from the stratum corneum to the superficial dermis. The excitation wavelength selected for this acquisition mode was 790 nm in order to visualize both the epidermal cellular structure and the collagen/elastin fibrillar components of the dermis. We used the data obtained by this scanning modality to measure the epidermal thickness. For each subject, we acquired one cross-sectional image for each site.

2.2.2.

MPM data processing

The quantitative parameters provided by the MPM data were: (A) the melanin distribution as a function of depth in the epidermis, (B) the melanin volume fraction in the epidermis, and (C) the epidermal thickness. For measuring the parameters related to melanin content (items A and B), we used the TPEF images acquired as z-stacks in five different locations on each site. The TPEF images of the epidermis in each acquired image stack were processed by using the following procedure: (1) If present in the image, areas containing TPEF signal from keratin in skin folds were cropped out and (2) The TPEF images were converted into binary images using the same threshold for all images. The threshold value of TPEF signal was defined as

Fig. 1

Multiphoton microscopy (MPM) image processing for estimation of melanin content: (a) raw two-photon excited fluorescence (TPEF) image acquired in the non-sun-exposed site of skin type II, (b) thresholded TPEF image shown in (a), (c) raw TPEF image acquired in the non-sun-exposed site of skin type VI, (d) thresholded TPEF image shown in (c). Scale bar is $20\mu \mathrm{m}$.

The density of bright pixels was calculated for each image in a z-stack of several consecutive image planes ($5\mu \mathrm{m}$ step) starting from the top of the epidermis to the basal layer. The density of bright pixels in each TPEF image was calculated by following the procedure described above and represents an estimate of the melanin concentration in the imaged epidermal layer. (4) In order to calculate the distribution of melanin content as a function of depth in the epidermis, we averaged the values of melanin concentration corresponding to a specific depth in each of the five acquired stacks. (5) For calculating the melanin volume fraction in the epidermis, we averaged the melanin volume fraction calculated for each of the five stacks. For each stack, the melanin volume fraction was calculated as

We used the TPEF/SHG cross-sectional images acquired at 790-nm excitation wavelength (Fig. 2) in order to estimate epidermal thickness in order to provide reference points for calculating the melanin volume fraction in the epidermis. The epidermal thickness was measured in 10 different locations starting from the end of the stratum corneum to the start of the EDJ, as shown in Fig. 2. The epidermal thickness reported here, in each case, represents the average of the ten measurements.

Fig. 2

In vivo MPM cross-sectional image of human skin (xz scan). Merged TPEF (green)/second-harmonic generation (SHG) (blue) cross-sectional image acquired in volar forearm of type II skin. The double-headed arrows show the locations at which the epidermal thickness was measured. The scale bar is $20\mu \mathrm{m}$.

2.3.1.

Instrumentation

The SFDS instrument (also known as spatially modulated quantitative spectroscopy, SMoQS) that was used in this study was developed at the Beckman Laser Institute (Fig. 3).

Fig. 3

System diagram for spatially modulated quantitative spectroscopy (SMoQS) instrument used in initial experiments.

A 100 W Quartz-Tungsten-Halogen light source (Moritex, MHF-D100LR) was coupled to a digital micro-mirror device, DMD, (Texas Instruments, Inc.) in order to project structured light patterns with spatial frequencies from 0 to $0.3{\mathrm{mm}}^{-1}$. The projection field of view for this system is $22\times 17{\mathrm{mm}}^2$. The distal end of a 1-mm core optical fiber (TechSpec, NA 0.39, Thorlabs, Inc.) was imaged at the center of the projection field of view ($1:1$ magnification) and light collected by this fiber is delivered to a spectrometer (Oriel 77480) enabling measurements over the range 450–1000 nm. Crossed 2-inch diameter wire-grid polarizing filters were used to reject specular reflection from the surface of the sample. The polarizer was inserted between the DMD and projection optics and the analyzer was placed between the fiber and collection optics.

In this study, six evenly spaced spatial frequencies were acquired, ranging from 0 to $0.3{\mathrm{mm}}^{-1}$ for a given anatomic location (dorsal forearm and volar upper arm). At each location, three measurements were acquired to ensure measurement consistency, identify any potential motion artifacts during the measurement acquisition, and help to improve the signal to noise. A three-phase demodulation scheme was employed at each spatial frequency, as has been described in the past, to isolate the spatial frequency-dependent reflectance from the skin region of interest from background contributions including those resulting from ambient (room) light illumination. This resulted in a total of 12–18 spectral measurements for each measurement site. Since tissue pigmentation ranged from types I to VI skin, integration time was automatically adjusted (LabView, National Instruments Inc.,) to ensure that the full dynamic range of the 16-bit detector was used for all subjects. Additionally, triplicate data were acquired for each projected pattern to improve the signal-to-noise ratio. With this instrument configuration, acquisition times ranged from 10 to 40 s, depending on the amount of pigmentation in the region of interest.

2.3.2.

SFDS data processing

As previously described in greater detail,⁸ the three phases at each spatial frequency are demodulated and calibrated against a diffuse tissue-simulating phantom of known optical properties to produce diffuse reflectance spectra as a function of spatial frequency. At each wavelength, the reduction in AC reflectance amplitude as a function of spatial frequency can be modeled and analyzed via homogeneous Monte Carlo-based simulations via discrete Hankel transformation of point-source reflectance predictions.⁸ From this model, the contributions of absorption and scattering can be independently identified at each wavelength, resulting in broadband spectra for absorption and scattering without the use of any spectral constraints or assumed power-law dependence for reduced scattering. As this initial step is based on a homogeneous model, the resulting absorption and scattering spectra should be interpreted as a “measured optical response” from tissue as a function of the total volume interrogated at each wavelength. The contribution of layer-specific absorption and scattering will change as a function of the total penetration depth interrogated, and hence as a function of wavelength. Figure 4(a) shows the example (bulk) absorption and reduced scattering spectra from subjects in this study, representing four different skin type categories. Here, the scattering spectrum for skin type VI no longer follows a simple power-law like dependence. In general, scattering from dermal tissue will be greater than that found in the epidermis, even with a high concentration of melanin.²² The combination of a strong absorption coefficient at visible wavelengths and high melanin concentration dramatically limits the depth penetrance at visible wavelengths. Thus, the measured scattering coefficient in the visible wavelength range is primarily related to the scattering properties of epidermis. The penetration depth increases in the NIR resulting in the increased contribution from the relatively higher scattering dermal tissue. For the example presented in Fig. 4(a), we not only measure the wavelength dependence of scattering from epidermal and dermal tissues, but also the depth-dependent transition between the relative contributions of these scattering regimes (as a function of depth penetrance).

Fig. 4

(a) Representative bulk absorption (top) and reduced scattering (bottom) coefficient spectra from Fitzpatrick skin types II, III, IV, and VI as measured by spatial frequency domain spectroscopy (SFDS). Spectra are derived from three measurements taken from individual subjects (dorsal forearm) at each of the four skin types. (b) Example of depth-sensitive spectral decomposition of in vivo skin tissue (volar upper arm of a single subject, skin type III). Blue: tissue absorption measured by SFDS; black (dashed line): depth-specific quantification of melanin relative to total volume of tissue interrogated at each wavelength; red: absorption features specific to the dermis and deeper tissue; green (dotted line): least-squares fit for oxy/deoxy hemoglobin. (Residual absorption features between red and green correspond to carotenoids, lipids, and water).

From these quantitative spectra, we can not only determine the bulk concentration of melanin, hemoglobin, water, lipids, and carotenoids in vivo, but also leverage the differential penetration depths of light between the visible and NIR regimes to isolate chomophore concentrations in depth. While the absorption spectra contain distinct features that can be used to separate and quantify these chromophores relative to the total volume of tissue interrogated by light (up to $\sim 5\mathrm{mm}$ through this specific approach), the combination of the wavelength-specific absorption and scattering provides us with an opportunity to estimate the depth penetrance into tissue of that specific wavelength.^23,24 As the penetrance of visible light into tissue is significantly less than that from the NIR, we can leverage the differential volumes and sensitivities to depth-specific chromophores in these spectral regions to quantify melanin volume fraction, average depth distribution and other chromophores found in tissue.¹⁵ We have tested this model-based, depth segmented approach in layered tissue simulating phantoms²⁵ and in vivo benign pigmented lesions, correlating the mean depths of pigmentation in vivo to high-frequency ultrasound.²⁶

To obtain depth-specific chromophore concentrations and depth distribution thickness, we used the method described by Saager et al.¹⁵ For this study, we used the spectral regions of 450–600 nm and 650–950 nm to represent the visible and NIR spectral regions in the aforementioned method. As this method is based on interpreting measured spectroscopic data in terms of a two-layer model, the outputs of this method are the top layer (optical) thickness (as determined by the absorption signature of melanin, relative to the mean photon path length traveled through the tissue) and the chromophore concentrations specific to the respective layers. In the context of this investigation, we are interested in whether this diffuse optics-based SFDS modality can isolate and assess the top layer properties, namely, layer-specific melanin concentration and melanin layer (mean distribution) thickness that is correlative to microscopy determined results. Figure 4(b) shows an example of this two-layer, depth segmented decomposition of skin absorption from in vivo tissue (skin type III, volar upper arm).

3.1.1.

Multiphoton microscopy

Melanin distribution as a function of depth in the epidermis was measured in TPEF z-stacks from the top of the epidermis to the EDJ. Generally, we measured a higher percentage of melanin in the basal and supra-basal layers ($10–15\mu \mathrm{m}$ from the basal layer) compared to the upper epidermal layers ($15–40\mu \mathrm{m}$ from the basal layer). Type VI dorsal forearm exhibited a more uniform distribution of melanin as a function of epidermal depth compared to other skin types. Distribution of melanin concentration as a function of depth in the epidermis for skin types I and VI is shown in Fig. 5. The position across the epidermis was normalized to the thickness of the epidermis in order to account for variations in epidermal thickness for different skin locations and different individuals. Therefore, fewer points in the plot representing the percentage of melanin content across epidermal layers in Fig. 5 are an indication of a thinner epidermis.

Fig. 5

Melanin distribution as a function of depth in the epidermis in skin types I and VI. Percentage of melanin content in layers across epidermis from the basal layer to the top of the epidermis for volar arm (circle) and dorsal forearm (asterisk) corresponding to skin type I (a) and skin type VI (b). Data represent the average of the melanin content in layers of five different stacks acquired for the same site. The error bars represent standard deviation of the five measurements. The position across the epidermis was normalized to the thickness of the epidermis.

There is a higher percentage of melanin content regardless of depth and a more uniform melanin distribution as a function of depth in type VI dorsal forearm compared to type I skin. These effects are due to differences in melanosome biology; dark (type VI) skin is characterized by intact melanosomes that are not degraded by lysosomal enzymes in the upper epidermal layers (stratum spinosum and granulosum).²⁷ In contrast, light skin melanosomes are degraded and only persist in the upper layers as “melanin dust.”²⁷

The melanin volume fraction estimated from the MPM data for different skin types is shown in Fig. 6(a).

Fig. 6

Melanin volume fraction estimated from MPM and SFDS data. (a) Melanin volume fraction as function of skin type, corresponding to volar upper arm (circle) and dorsal forearm (asterisk) as determined by MPM. Data represent the average of the melanin volume fraction calculated for five stacks acquired for the same site (volar or dorsal) in each of the 12 patients. The error bars represent standard deviation of the five measurements. (b) Average melanin concentration determined by SFDS. Red (asterisks) are estimates from the dorsal forearm as a function of assessed skin type, black (circles) are estimates from the volar upper arm. The error bars represent an estimated model error from SFDS layered decomposition.

For the volar arm, an area generally not exposed to the sun, melanin volume fraction increases with skin type and exhibits an overall change of $\sim 2\times$ from types I to VI. However, dorsal forearm melanin volume fraction was not correlated with the skin type. This is due to the fact that the melanin content in the dorsal forearm, which typically has significant sun exposure, is also related to factors such as frequency and duration of sun exposure, the presence of freckles with increased melanin content,²⁸ and the sun protection routine for each individual. Figure 6(a) shows that the dorsal forearms of skin types I and II are particularly susceptible to this effect.

3.1.2.

Spatial frequency domain spectroscopy

Figure 6(b) shows that higher melanin concentration is also detected using SFDS for the dorsal forearm relative to the volar upper arm for each subject. For the presentation of these data, we assumed a 5% error based on our initial validation work of the empirical model used to isolate layer-specific chromophore concentrations.¹⁵ Although the Fitzpatrick skin type scale is not a direct representation of skin pigmentation, but rather a subjective, scaled assessment of skin response to sun exposure, there was a general trend of increasing detected melanin concentration with increasing skin type. A factor of $\sim 2$-fold increase was observed comparing volar types I to VI, while $\sim 3$-fold higher % melanin values were obtained comparing dorsal types I to VI. These relative ranges of increase are in agreement with other quantification methods of melanin across skin types and sun-exposed skin regions.^29–33 Unlike the melanin volume fraction measured by MPM in the dorsal forearm in skin types I and II, SFDS reported a general increase in melanin content as a function of skin type. The key distinction between these imaging modalities in this particular case is the volume of interrogation. Discussed in more detail in Sec. 4, SFDS interrogates $\sim 50$-fold larger volume than MPM. In this case, it spatially averages heterogeneous pigmentation common in sun-exposed areas of skin for these skin types.

3.2.1.

Multiphoton microscopy

The epidermal thickness measured for different skin types is shown in Fig. 7(a).

Fig. 7

Epidermal and melanin distribution thickness estimated from MPM and SFDS data, respectively. (a) Epidermal thickness as a function of skin type corresponding to volar arm (circle) and dorsal forearm (asterisk) as determined by MPM. Data represent the average of the epidermal thickness calculated in 10 different locations across the MPM images acquired as xz scans as shown in Fig. 2. The error bars represent standard deviation of the 10 measurements. (b) Melanin distribution thickness as determined by SFDS. Red (asterisk) refers to the dorsal forearm and black (circle) the volar upper arm. Note that the thickness values presented are in terms of optical path length and not actual anatomic thickness. Each data point represents a measurement corresponding to each of the 12 patients, acquired from the volar upper arm and dorsal forearm. The error bars represent an estimated model error from SFDS layered decomposition.

MPM measurements showed no correlation between epidermal thickness and skin type. However, we measured a significant difference in the overall epidermal thickness of volar versus dorsal arm ($40\pm \mu \mathrm{m}$ versus $50\pm \mu \mathrm{m}$, respectively, $p=0.007$).

3.2.2.

Spatial frequency domain spectroscopy

Similarly, Fig. 7(b) shows that the melanin distribution thickness was greater for the dorsal forearm that for the volar upper arm. This has been suggested by other literature for two general reasons: (1) epidermal thickness can vary in skin based on anatomical location;³⁴ namely, the epidermal thickness of the dorsal forearm has been reported to be thicker, on average, than that in the volar upper arm. As with the determination of layer-specific melanin concentration, we also assumed a 5% error for distribution thickness based on our initial validation work of the empirical model used to isolate layer-specific chromophore concentrations.¹⁵

It is worth noting that for more heavily pigmented skin (IV-VI), there is an observed correlation between pigmentation and estimated melanin distribution thickness that was not evident in the MPM measurements. We discuss this outcome in Sec. 4.