Full-depth epidermis tomography using a Mirau-based full-field optical coherence tomography.

With a Gaussian-like broadband light source from high brightness Ce(3+):YAG single-clad crystal fiber, a full-field optical coherence tomography using a home-designed Mirau objective realized high quality images of in vivo and excised skin tissues. With a 40 × silicone-oil-immersion Mirau objective, the achieved spatial resolutions in axial and lateral directions were 0.9 and 0.51 μm, respectively. Such a high spatial resolution enables the separation of lamellar structure of the full epidermis in both the cross-sectional and en face planes. The number of layers of stratum corneum and its thickness were quantitatively measured. This label free and non-invasive optical probe could be useful for evaluating the water barrier of skin tissue in clinics. As a preliminary in vivo experiment, the blood vessel in dermis was also observed, and the flowing of the red blood cells and location of the melanocyte were traced.


Introduction
In recent years, optical coherence tomography (OCT) has been widely applied on threedimensional (3-D) image reconstruction of skin tissue [1,2]. In epidermis, to non-invasively probe the layer parameters (LPs), such as average total thickness (a-TT), average number of layers (a-NOLs), and average cellular layer thickness (a-CLT), for stratum corneum (SC) becomes important for evaluating the skin moisturization of epidermis [3]. On this application, axial resolution better than 1.2 μm in tissue is the doorsill to measure LPs of the SC [4]. Besides, the morphology of single 3-D epidermal cell is also important for early detection of normal and abnormal cells of pre-cancer diagnosis [5]. These all require submicron spatial resolution in tissue [6]. Full-field OCT (FF-OCT) [7,8] utilizing twodimensional CCD/CMOS camera has the opportunity to observe the layer structure of SC [9], especially for en face monitoring. Typically, the detection sensitivity of FF-OCT using CCD/CMOS camera is about 80 dB [10], related to the camera area size and en face frame rate.
Keratinocyte and melanocyte are the two major cell types in epidermis, with a normal size from 10 to 50 μm [11]. The epidermis can be divided into several layers, which are stratum basale at the bottom, stratum spinosum, stratum granulosum, stratum lucidum, and SC on the top, through keratinization process within about one month. In epidermis, melanocytes are interspersed at stratum basale with stretching dendrites [12]. For skin care aspect, the proliferation and differentiation of keratinocyte affect the capability of epidermal moisture lock [13,14] and dry skin disease [15].  (35-year-old, male), where yellow and blue arrows indicate the dermis-epidermis junction and blood vessel, respectively. The green arrow heads mark the boundaries of SC. In (a), the SC is much thicker than that of (c) because of z-axial expansion induced by water hydration. In (c), 58% glycerin was used as the index-matching liquid between in vivo human skin and CG. In (a)-(c), red arrows are the boundaries between CGs and index-matching liquids. (d) shows the en face image of (c) at a depth of 46 μm (position of pink dash-dot line in (c)). In (d), the purple arrows point to the melanocyte along its dendrites, traced from melanin caps of the shallower en face images. The white spots in (c) and Typically, LPs of the SC in broad area are related to skin barrier function. In addition to OCT methods, a-TT of the SC was indirectly verified via z-axial population of water content by confocal Raman spectroscopy [13,16] or was directly measured by multiphoton laser tomography [17,18]. For confocal Raman spectroscopy and confocal reflectance microscope [16], they both need the database of a-NOLs from frozen section [19] to verify the a-CLT of SC. For multiphoton laser tomography, such as nonlinear effect (i.e. second/third harmonic generation and coherent anti-Stokes Raman spectroscopy), it needs high power density (about 30-100 mW average power, with transient focal spot size less than 0.5-μm-diameter) and long measurement time (about 15-30 minutes for 100-μm depth) to establish one full 3-D image stack. Figure 1(a) shows the cross-sectional image from a 3-D en face stack scanned by the Mirau-based FF-OCT with a 40 × home-designed Mirau objective; whereas, Fig. 1(b) represents the schematic cross-sectional structure of an anatomical illustration. Via the crosssectional images at different positions (see Media 1), layer-to-layer boundaries become easy to observe. Most of the skin does not have stratum lucidum, except for palm and sole. Figures 1(c) and 1(d) are the in vivo images of human forearm skin in cross-sectional and en face planes, respectively. To compare Figs. 1(a) and 1(c), melanin caps of Fig. 1(a) disappeared because the excised skin tissue is gradually denatured after skin tissue was immersed in phosphate-buffered saline (PBS). Figure 1(e) represents the oblique view of 3-D image, which is the same in vivo tissue of Fig. 1(c) and 1(d). The incident power and 3-D imaging time (about 100-μm-depth) of this system were 5 mW (focal spot size about 220-μmdiameter) and 2 minutes, respectively. The scan speed of en face images is 4.3 frame/sec. Compared with the single-point scanning via Ce 3+ :YAG double-clad crystal fiber light source [20,21], this platform provides high frame speed and low incident power for 3-D reconstruction of skin tissue.
In this study, a Ce 3+ :YAG single-clad crystal fiber (SCF), drawn by laser-heated pedestal growth [22] and cladded by borosilicate glass [23], provides high brightness and high numerical aperture (NA) output light. The diameters of core and cladding are correspondingly 90 and 330 μm. The broadband output power after coupling into multi-mode fiber (Thorlabs, 400-μm-diameter, NA: 0.39, America) is 30 mW, and becomes 5 mW after passing through Mirau objective onto the sample. This light source is eminently suitable for FF-OCT because it is unnecessary to use infrared-cut filter to prevent from thermal problem on sample [24]. With a Gaussian-like broadband spectrum [20,21] the ghost image effect coming from sideband noise is suppressed. Circularly polarized incident light was adopted because it can provide deeper penetration [25]; meanwhile, weak signal light can be strongly enhanced by the reference one. Figure 2(a) shows the schematic diagram of Mirau-based FF-OCT. The Ce 3+ :YAG SCF was pumped by a 1-W, 445-nm laser diode (Nichia, #NDB7875, Japan) through collimating and focusing module, including a 60 × aspheric lens, a band-wave-pass filter (Semrock, #FF01-445/45, America), and a 40 × achromatic lens, where the function of band-wave-pass filter is to reflect the backward broadband light back to the SCF, to enhance the total output power. The broadband light emerging from the output terminal of the SCF was coupling into multimode fiber and was then collimated by an objective lens, where the center wavelength and bandwidth of light after SCF are respectively 560 and 95 nm. Then, this light reflected by broadband polarizing cubic beamsplitter became vertically polarized. After achromatic quarter wave plate, it changed to circular polarization. The circularly polarized light became counter circular polarization when reflected back from reference and sample arms through Mirau objective. Finally, light beams from reference and sample arms both became horizontally polarized after passing achromatic quarter wave plate again (see green arrows in Fig. 2(a)). As a result, the back-reflected light beams from sample and reference arms were combined after broadband polarizing cubic beamsplitter, reflected by mirror, and then projected onto CCD (Imperx, #ICL-B0620, 640 × 480 pixels, America), to generate the interferometric signal with a frame rate of 260 frame/s During one period of interferometric carrier signal, there are 60 sampling frames.

2-1 Experimental setup and system performance
As shown in Fig. 2(b), the home-designed silicone-oil-immersion Mirau objective is composed of a water-immersion objective lens (Olympus, LUMPLFLN 40 × W, NA: 0.8, field-of-view: 550 μm, Japan), a ring holder, two fused silica glass plates (thickness: 150 μm, λ/10 flatness). The diameter of focal field in water is about 220 μm (1/3 field-of-view was used). The Mirau objective was fixed on a z-axial piezo-electric transducer (PZT) (PI, #P-720, Germany). Cover glass (see Fig. 2(a)), the same thickness as fused silica glass plates (see Fig. 2(b)), was laminated under the sample. The total travelling range of the PZT with openloop control is 112 μm. A 500-μm-diameter black ink absorber is used, to well-match the index of first glass plate and to absorb the stray light back to the CCD, for eliminating the DC term of intensity. After coating the interlaced layers by TiO 2 /SiO 2 , T/R = 60/40 (T: transmittance; R: reflectance; n oil = 1.406) broadband beamsplitter coating was coated on the top of second glass plate (GP 2 ). The reflection coating (RC) under first glass plate (GP 1 ) is about 4% as n oil = 1.406.  Figure 2(c) shows the emission spectrum of a Ce 3+ :YAG SCF. The interferometric signal intensity of A-scan reflected from the boundary between water and glass plate was measured by one pixel of CCD (see Fig. 2(d)). The intensity of carrier envelope from carrier signal in Fig. 2(d) was calculated after band-pass filter and Hilbert transform. Following the calculation of literature [7,9], the detection sensitivity is about 81 dB. Referring to the literature [20], the noise floor of this system is substantially suppressed by stronger confocal gate (NA: 0.8) effect, and then the effect of ghost image is further leveled down. The home-designed silicone-oil-immersion Mirau objective provides experimental resolutions of R a = 0.91 μm (see Fig. 2(d)) and R t = 0.56 μm (see Fig. 2(e)) along axial and transversal directions at the surface of water medium (or R a = 0.90 μm and R t = 0.51 μm at the surface of SC (n = 1.47 after water hydration)), respectively [26]; whereas, the theoretical spatial resolutions at the surface of water following diffraction limits are R a = 0.56 μm and R t = 0.43 μm (or R a = 0.55 μm and R t = 0.39 μm at the surface of SC) according to Eq. (1) [ where Δz eff means the effective axial resolution contributed by Δz confocal (confocal gate in water, equal to λ 0 n water /NA 2 , about 1.16 μm for 40 × silicone-oil-immersion objective (NA: 0.8)) and Δz coherence (coherent gate in water, equal to 0.44λ 0 2 /n water Δλ, about 1.09 μm for Ce 3+ :YAG light source with the same objective). n sample and n water are the refractive indices of the sample and the water, respectively. λ 0 and Δλ are the central wavelength and the bandwidth, of the light source. In Fig. 2(b), 40 × OL is used for water-immersion. As the spaces are injected into silicone oil, R a and R t become worse. So, experimental resolution is not as good as theoretical one.

2-2 Signal processing
Typically, FF-OCT takes the en face image from calculating the stack information via phasestepped technique with single-shot CCD at 0°, 90°, 180°, and 270° [27], of which the phase was shifted by triangularly oscillated motion of PZT. As the exposed time of one en face image increases, the detection sensitivity becomes better. Then, 3-D image is reconstructed by piling up the en face images along z-axis. Different from classical FF-OCT, the Miraubased FF-OCT in this study reconstructs the 3-D image stack via sequential interferometric signals.
Assuming a sinusoidal interferometric signal with N sampling points during each carrier, the intensity envelope I(x,y,z 0 ) env at a depth of z 0 during one carrier can be approximately expressed as where z 0 is the central position of the calculated single carrier wave. (x,y) is the pixel location on the CCD. N/P is the number of calculating intervals. I i (x,y) and I j (x,y) are two P-timesaveraged intensities of the N/P intervals.
Equation (3) is consistent with the Eq. (4) in [28], and this equation was also commonly used in fluorescence microscopy for background subtraction [29]. In most applications, the amplitude of sinusoidal wave is amplitude-modulated. Figure  3(a) shows the magnified experimental carrier signal scanned by Mirau-based FF-OCT. As the sinusoidal wave is Gaussian-modulated, the calculated envelopes of N/P = 3, 4, 39, and 156 are calculated in Fig. 3(b) according to Eq. (2). Peak value of N/P = 156 is the nearest result when compared to the raw data using Hilbert transform with band-pass filter; whereas, N/P = 3 has the lowest noise floor. In Fig. 3, the percentages of peak values of N/P = 3, 4, 39, and 156 versus the peak value of raw data using Hilbert transform with band-pass filter are correspondingly 82.7%, 89.0%, 99.3%, and 99.5%. In Fig. 3, carrier signal was measured under lower moving speed of open-loop PZT, and there are 156 sampling point during one period of interferometric carrier signal. For logarithmic image, N/P = 3 has best noise floor, and the calculating process is also much faster than that of N/P = 156. But, N/P = 3 has poor result of peak value. To sum up, N/P = 4 has better benefit for en face image calculation from stack signal. In the experiment, N/P = 4 was used to reconstruct the 3-D morphology of SC.  Fig. 4(e). The noise level after 10-time-average was improved by 2.15 dB. For the whole epidermis (refer to Fig. 4(d)), the signals from other layers will affect the calculation of LPs in SC. In order to directly measure the 3-D LPs of SC, the tested SC was isolated by chemical and enzymatic digestion techniques via 0.1% trypsin at 37°C for 2 hours in an incubator [30]. This SC was immersed in 4% formalin to maintain the structure and was slightly laid on cover glass as shown in Fig. 2(a). Figures 5(a)-5(d) show the cross-sectional images of the isolated SC at y = 45, 56.25, 67.5, and 78.75 μm. In Fig. 5(e), axial LPs of the SC from Fig. 5(a) were measured and then calculated.  Table 1 shows the regional LPs of the SC from Fig. 5. Better than the results from single 2-dimensional image, 3-D image of SC after trypsin can provide more significant LPs of the SC, to evaluate the skin water barrier. Typically, SC immersed in water or PBS after one hour generates about 40% z-axial expansion from hydration [12]. The 3-D measurements of a-TT and a-NOLs from buttock with deducting hydration are 14.56 ± 2.38 μm and 11.72 ± 3.00, respectively. In the literature, a-TT and a-NOLs of the SC at buttock using frozen section are respectively 14.9 ± 3.4 μm (m = 61) [31] and 12 ± 4 (m = 20) [12,22], where m means the number of samples. Actually, histological biopsy is difficult to provide 3-D information of SC. Comparing Figs. 5(a)-5(d) with Table 1, hollow cavities increase the standard deviation of a-CLT. In Table 1, mean value of Fig. 5(a)-5(d) is very close to the result of 3-D vision. In other words, sampling method (mean value of Fig. 5(a)-5(d)) of 3-D information is sufficient enough to be used for evaluating significant LPs of the SC.

Experimental results and discussions
In this study, a desktop system was established [32], which is suitable to take the 3-D skin images of forearm and face for whole epidermis. As deeper information is necessary to image, like collagen, Mirau objective has to be changed into low magnification one. Fulldepth epidermis images of skin tissues for excised and in vivo cases were both reconstructed in detail via the Mirau-based FF-OCT. In Media 2 (in vivo skin tissue), flowing of red blood cells was observed, but the flow speed of red blood cells still cannot be estimated. It becomes capable as the en face frame rate is improved. In contrast, Media 3 (excised skin tissue) did not represent any flowing imprint. In Fig. 1(d), the melanocyte was traced by its melanosomefilled dendrites. It needs to catch the timing that the melanocyte is secreting and transporting the melanosome via dendrites. In this experiment, LPs of the isolated SC were measured by Mirau-based OCT, but the concrete value of this measuring method is for in vivo skin. In vivo skin cannot use trypsin to isolate the SC away, so it is necessary to build image segmentation as boundary between SC and stratum granulosum, to take the in vivo LPs from patients directly. This information will be very helpful for the judgments of water barrier (cosmetics) and hyperkeratosis (skin disease) [33]. Sampling area a-TT (μm) a-NOLs a-CLT (μm) (Average ± SD) Figure 5(a) 14.17 ± 1.42 12.02 ± 2.52 1.19 ± 0.60 Figure 5(

Conclusions
A Mirau-based FF-OCT was demonstrated by reconstructing the 3-D skin tissue and measuring the LPs of the prepared SC. It took about 2 minutes to scan and reconstruct full cuboidal image stack of tissue with 144 (x-axis) × 108 (y-axis) × 100 (z-axis) μm 3 using less than 5 mW incident power. For in vivo skin tissue, 3-D skin morphology was also visualized, but melanin caps will absorb most of the visible light from Ce 3+ :YAG SCF. The differences between in vivo and excised skin tissues coming from different locations, such as hydration of SC, denature of melanin caps, and information of red blood cells, were verified. In the future, this system can directly applied on the clinical observation of deeper position, as the light source is changed to Ti:sapphire SCF broadband light source. Based on this system, the a-TT and a-NOLs from excised buttock with deducting hydration were measured. After the analysis of more patients, it is capable to define the skin ages of water barrier. This is very important to the medical and cosmetic areas of dermatology.