Slide-free clinical imaging of melanin with absolute quantities using label-free third-harmonic-generation enhancement-ratio microscopy

: The capability to image the 3D distribution of melanin in human skin in vivo with absolute quantities and microscopic details will not only enable noninvasive histopathological diagnosis of melanin-related cutaneous disorders, but also make long term treatment assessment possible. In this paper, we demonstrate clinical in vivo imaging of the melanin distribution in human skin with absolute quantities on mass density and with microscopic details by using label-free third-harmonic-generation (THG) enhancement-ratio microscopy. As the dominant absorber in skin, melanin provides the strongest THG nonlinearity in human skin due to resonance enhancement. We show that the THG-enhancement-ratio (erTHG) parameter can be calibrated in vivo and can indicate the melanin mass density. With an unprecedented clinical imaging resolution, our study revealed erTHG-microscopy’s unique capability for long-term treatment assessment and direct clinical observation of melanin’s micro-distribution to shed light into the unknown pathway and regulation mechanism of melanosome transfer and translocation.


Introduction
Melanin is the most important pigment in human and is the determinant of our skin color. In human skin, melanin not only acts as a physical barrier [1,2] to protect epidermal keratinocytes and superficial dermal vessels from ultraviolet light irradiation, but also acts as a free-radical scavenger by interacting with reactive oxygen species. In addition, melanin has been shown to possess metal chelating activity and to act as a physiological redox buffer [3,4]. Malfunctions of melanin production and distribution is a signature of skin pigmentary diseases and melanin should act as the most critical endogenous marker for dermatological diagnosis. For histopathology, the Fontana-Masson (FM) method [5] is a common technique to stain melanin, while humanmelanoma-black-45, Melan-A, or S-100 immunohistochemical stains [6] are exploited to image melanocytes [7], which synthesize melanin in melanosomes and transfer melanosomes to the neighboring keratinocytes through their dendrites. These exogenous stains for melanin or melanocytes are usually combined with the invasive and lengthy formalin-fixed paraffin-embedded (FFPE) immunohistochemistry processes [8], thus precluding their application for noninvasive onsite diagnosis with a histopathological accuracy. The lack of absolute quantity information also makes long term treatment assessment impossible. A noninvasive quantitative imaging tool, not only providing histopathological morphology information but also a three-dimensional (3D) melanin density mapping, will be highly desirable for diagnosis and to serve as the critical part of a treatment evaluation program for melanin-related diseases.
Ex vivo analytical methods to quantify melanin content include high performance liquid chromatography [9], electron paramagnetic resonance [10], Raman [11] and absorption spectroscopy [12], while diffuse reflectance spectroscopy (DRS) [13,14] can achieve quantification of melanin content in human skin in vivo, but lacking the spatial resolution. For in vivo label-free high-resolution imaging of melanin, absorption-based techniques attract much attention. Twophoton-excitation fluorescence (TPEF) microscopy and fluorescent lifetime imaging [15,16] can provide information about relative contents of eumelanin and pheomelanin in human skin in vivo with a subcellular resolution. Photoacoustic microscopy seems to be another attractive choice [17][18][19]. Photoacoustic melanin index (PAMI) [17], optical melanin index (OMI) [15], and global melanin density [16] were proposed to provide quantitative measure of melanin content. For absorption-free techniques, reflection confocal microscopy (RCM) and third harmonic generation (THG) microscopy both show relatively strong contrasts provided by melanin in subjects with darker skin in vivo [20][21][22]. In order to quantify the contribution from melanin, parameters such as image contrast (IC) [20], papillary contrast (PC) [21], or ratio of THG brightness [22] was calculated as a relative quantitative measure. However, no in situ calibration method to connect these indices to the absolute melanin content had been proposed. As a result, no absolute quantification of melanin density in human skin in vivo has been achieved with a high spatial resolution by any microscopy technique.
Since the signal intensity that is related to the absolute quantity of melanin would be easily modified through tissue reabsorption, scattering, and equipment alignment, an internal calibration beacon is thus needed. In this study, due to the combination of the following conditions, we demonstrate the in situ calibration of melanin in human skin with absolute quantities. First, THG is abundant in tissues and can be generated from cytoplasmic organelles and collagen fibers, and thus this background THG signals around the imaged area can serve as internal calibration beacons. Second, THG signals generated from melanin overwhelm all other background THG signals in epidermis, and high signal to background ratio can be obtained. Third, our recent study [23] found that the THG intensity with the presence of melanin will be enhanced out of the background signal through a three-photon resonant mechanism and hyper-Rayleigh scattering, while this enhancement ratio is quantitatively determined by the local melanin mass density.
Here we demonstrate clinical in vivo imaging of the melanin distribution in human skin with absolute quantities on mass density and with microscopic details by using label-free THGenhancement-ratio (erTHG) microscopy. By adopting the resonant enhancement ratio concept to describe the THG generation from melanin and at the same time to deal with the background calibration beacon issue, here we propose to retrieve the erTHG parameter in situ and thus recover the absolute quantity of melanin in term of mass density within a sub-femtoliter volume. In vivo clinical studies were conducted to compare the absolute quantitative measure obtained by erTHG microscopy with those obtained by DRS. Excellent quantitative agreement can be found, thus validating our approach. This quantitative microscopical tool was further applied for treatment assessment of Asian volunteers with solar lentigines. Abnormal melanosome dispersion in diseased cytosols were observed for the first time and treatment does not heal this abnormality. Based on our findings, we conclude the high potential of slide-free label-free erTHG microscopy for absolutely-quantitative melanin imaging in vivo with a subfemtoliter 3D spatial resolution in human skin.

Clinical in vivo THG and second harmonic generation microscope
For our clinical system, a Cr:Forsterite femtosecond laser source (central wavelength of 1262 nm, FWHM bandwidth 92 nm) was collimated and guided into a galvo-resonant scanning head (Thorlabs, MPM-SCAN4) to perform real-time fast 2D-scanning. The scanning pattern was focused onto human skin by a water immersion objective with a NA of 1.15 (Olympus, UApoN340/40X/NA1.15/Working distance 250 µm). Double-chirp mirrors were adopted to control the optical dispersion for efficient excitation. Epi-second-harmonic-generation (SHG) and epi-THG signals were collected by the same objective, divided by a dichroic beam splitter (DBS), and detected by two individual PMTs. The band-pass filters with different center wavelengths and bandwidths (FF02-617/73 for SHG and FF01-417/60 for THG, Semrock) were inserted before the PMTs to filter out the background noise to increase the signal-to-noise ratio. The objective was attached to a 3D step motor so the position of the objective can be adjusted by both manual tuning and remote electrical control. The imaging plane (or the plane of observation) can be moved to different depths by tuning the 3D stage along the optical axis. Signals from two channels were then mapped into 14-bit greyscale images with the size of 512×512 pixels, corresponding to a field of view of 235×235 µm 2 . The acquisition time was approximately 0.38 second per averaged image, after averaging 5 frames at a fixed depth with a 15 Hz frame rate. Images of different depths from skin surface were acquired every 1.8 µm along the optical axis. Afterwards the images can then be stacked into a single .tiff file with multiple sub-images while each sub-image was an image for a certain depth in skin. The 3D imaging depth was within 250 µm limited by the objective working distance. the A similar system was recently reported in Ref [24].

Diffuse reflectance spectroscopy system
A custom-built DRS system was used to acquire spatially-resolved diffuse reflectance spectra from the skin on the ventral forearm. The DRS system shined continuous-wave white light (410-760 nm) on the skin surface through an optical fiber with a power of about 50 µW, which was below the maximum permissible exposure value established by American National Standard Institute. The diffuse reflectance light remitted from the skin was collected by three optical fibers which were separated from the source fiber by distances of 0.22 mm, 0.45 mm and 0.73 mm, respectively. All the fibers had a core diameter of 0.2 mm, a NA of 0.26, their axes normal to the skin surface, and their end faces in gentle contact with the skin during measurements. Diffuse reflectance spectra collected at multiple distances were calibrated with five tissue-mimicking phantoms made of known concentrations of polystyrene beads (Polysciences, Inc., Polybead Microspheres) and hemoglobin (Sigma-Aldrich, ferrous stabilized human hemoglobin). A linear calibration relation was established by comparing the measured reflectance to values predicted using the Monte Carlo method [25]. Scattering coefficients and absorption coefficients of the phantoms were obtained using the Mie theory and the UV-visible absorption spectroscopy measurements, respectively. After calibration the measured tissue reflectance spectra were compared to Monte Carlo simulated values iteratively to find a best set of tissue parameters that minimized the spectral errors.

Clinical protocols
All the following harmonic generation microscopy (HGM) protocols were approved by the Research Ethics Committee of National Taiwan University Hospital (NTUH-REC) under 200903064D, 201403043DINC and 201612113DINC. Before enrollment, each volunteer received information about the investigation procedure, got fully informed about the nature of the study and their rights, and signed an informed consent. None of the subjects reported any discomfort relating to the imaging laser, such as a burn or a sting. When all subjects went to clinic after the image acquisition the dermatologist observed no symptom or reaction concerning inflammation, no change of appearance, no pigment alteration, nor blister and ulceration on the site where images were taken.

Ex vivo epi-HGM imaging of human skin
In the epi-HGM imaging of human skin, the normal human skin was removed during skin surgery in National Taiwan University Hospital. The excised tissues were immersed in normal saline under room temperature and was freshly investigated immediately. The images came from a stack of optical sections obtained at different depths in the freshly-excised normal human skin.

In vivo HGM imaging of healthy volunteers
HGM images were acquired from the skin of 29 normal Asian subjects (11 males and 18 females) aged between 19 and 71 years (median: 32 years). HGM measurements were taken at the locations of ventral forearm, dorsal forearm, ankle, and dorsum of hand. In each location, three 3D stack were acquired.

In vivo DRS measurement of healthy volunteers
The experimental protocol was approved by the Institutional Review Board of National Taiwan University (NTU-REC under 201706HM077), and an informed consent was obtained from each subject. A custom-built DRS system was used to acquire spatially resolved diffuse reflectance spectra from the skin on the ventral forearm of 27 normal Asian subjects aged between 19 and 71 years (median: 33 years). Ten males and 17 females were enrolled. Three DRS measurements were taken from each subject at the same location, and the exposure time for each measurement was about 0.2 s.

In vivo HGM imaging of solar lentigines patients
Seven female Asian volunteers (Fitzpatrick skin phototype III or IV; age range: 46-64 years, median: 48 years) having solar lentigines on the face were enrolled in the study. To evaluate the whitening effects of Q-switched ruby-laser (QSRL) treatment, the volunteers were asked to come for image acquisition before and six-weeks after the treatment. A number of 2D HGM images underneath the same skin location yet of different depths, with at least 100 images, were captured to form one 3D stack. At least three locations of one solar lentigo, and at least three locations on the normal facial skin were imaged. In each location, one 3D stack was acquired.

Histology validation
THG microscopy, with the advantages of non-invasiveness, label-free, bleach-free, high penetration, and sub-femtoliter 3D resolution, has been applied for slide-free morphological and molecular visualizations of 3D sub-cellular structures inside living animals and humans [22], [26][27][28][29]. Under a tight focus, the high THG nonlinearity provides the 3D auto-sectioning capability without the need of a confocal pinhole [30]. As a dominant light absorber in human skin, melanin provides the strongest THG nonlinearity [22,23], through resonance enhancement with the absorption levels [23,[31][32][33] those match the two-photon and three photon frequencies ( Fig. 1(a)), while cytoplasm can be easily observed through melanin-enhanced THG in the basal layer of human skin [22]. To confirm the primary role of melanin in THG enhancement, a histology comparison was carried out. Ex vivo slide-free epi-HGM imaging, using the system previously described in Ref [22]., was performed on the transverse direction of a thin face skin sample. Afterward, transverse histology sections were obtained from the same sample with FM stain. Figure 1(d) and 1(e) show the bright-field images of the stained sections and melanin appears black-colored. As shown in Fig. 1(b), the stronger THG contrasts are found to concentrate in the basal layers (arrows) and have strong histology consistence with the Fontana-Masson stain (arrows in Fig. 1(d)). In addition, the cap-like distribution of brighter THG signals can be observed in the upper layers of epidermis (arrows in Fig. 1(c)), which histologically consists with the supra-nuclear melanin caps [2] shown in the histology section (arrows in Fig. 1(e)). These correspondences initially confirmed the dominance of melanin in the THG contrast in human skin.

Melanin mass density distribution imaging calibration
Our recent study [23] on the origin of the melanin-enhanced THG by using a live cell model has indicated an initial nonlinear process where the THG intensity was enhanced according to the 3.5th power of melanin mass density (MMD). When the MMD is higher than 11 mg/ml, a transition from the resonance-enhanced THG to the high-order hyper-Rayleigh scattering process occurs. This saturation phenomenon of the virtual-transition-based THG nonlinear process is attributed to the multi-melanosome-induced scattering [23] within the sub-femtoliter focal volume. This quantitative relationship describing the THG-enhancement-ratio [23] and MMD in a keratinocyte is ideal for realizing clinical in vivo imaging of the melanin distribution in human skin with absolute quantities.
According to Chen's study [22], the THG brightness ratio was previously analyzed to describe the melanin-induced THG intensity in vivo. Chen [22] defined the THG brightness ratio as the ratio of melanin-rich area THG intensity to the THG intensity of collagen fibers. In vivo epi-THG images of different Fitzpatrick skin types were analyzed to understand the ethnic variation in melanin-induced THG signals. The epi-THG brightness ratios in different skin types are greater than 1 and increase with skin type. Since the THG background intensity in the cytoplasm of a keratinocytes without melanin might not be the same as the THG intensity of collagen fibers, THG brightness ratio is not erTHG.
On the other hand, previous studies suggested a potential beacon light for calibration: THG intensity of collagen fibers. In order to quantify equivalent melanin mass density (MMD) in human skin using erTHG and to calibrate the erTHG with the beacon light, the THG-brightness ratio of collagen fibers and melanin-deficient basal cells, termed intrinsic ratio, needs to be determined. In vivo HGM images of melanin-deficient basal cells in human skin with vitiligo [34] were recently acquired and analyzed. Vitiligo [34] is an acquired chronic depigmentation disorder resulting from progressive loss of epidermal melanocytes. The intrinsic ratio of 1.106 was obtained according to this study, averaged from patients with vitiligo [34]. erTHG value can thus be obtained by dividing the THG brightness ratio with 1.106. With measured erTHG, MMD can be obtained with the following equations from Ref [23], which describe the melanin-induced erTHG inside a live cell as MMD is with the unit of mg/ml. erTHG is with no unit.

Melanin mass density distribution imaging
In order to quantify MMD distribution in the keratinocytes with a sub-micron spatial resolution, THG brightness of collagen fibers and of cytoplasm of cells has to be extracted. Two developed programs, detailed in the previous studies [35,36] were adopted in this study. With the aid of programs, segmentation of collagen fibers and cytoplasm of basal cells could be accomplished, as shown in Fig. 2. For the part of basal cell cytoplasm, at least three qualified sub-images were segmented for the cytoplasm of basal cell based on THG intensity information. If significant amounts of false segmentation occurred in a sub-image, manual removal of apparent false segmentation would be applied. Manual selection would be applied only if there were fewer than ten cells properly segmented by the program. The THG pixel brightness of segmented cytoplasm (called THG Cytoplasm ) for each qualified sub-image was then recorded, so that erTHG can be calculated from the data after normalization with the results from the Vitiligo patients. For the part of collagen fiber, we chose three continuous layers at the depth of basal cell (also the depth of dermal-epidermal junction (DEJ)) to segment collagen fibers on the basis of SHG information. After fiber segmentation, the mean THG brightness of segmented collagen fibers in the region of interest (ROI) was recorded (called THG Collagen ). Afterwards, we can get the THG enhancement ratio (erTHG value) for each pixel as Finally, the ratio was transformed to MMD via Eq. (1) and Eq. (2). Please refer to Fig. 2 for an example. Figure 3 shows the in vivo MMD distribution images underneath the skin of various volunteers. Figure 3 also shows the original unsaturated HGM images with combined THG and SHG images acquired at the DEJ as well as the corresponding erTHG images. The magenta pseudo-colored THG reflects cell distribution in the epidermis and the green pseudo-colored SHG reflects collagen distribution in the dermis. The corresponding MMD images are color-coded to reflect the mass density distribution in the keratinocytes around DEJ in different parts of human skin. Our standard color-coding is limited to 0-25 mg/ml density. We have also provided amplified image (called A-MMD image) in which the color-coding is only limited to 0-10 mg/ml density. The standard deviation values of MMD in Fig. 3 range between 17.5-47% of their mean values. Some of the standard deviations should be contributed by the uncertainty of the Eq. (1) & (2), as will be discussed in Section 3.6. With a high spatial resolution, our proposed methodology also allows us to analyze the MMD distribution inhomogeneity, called inhomogeneity of MMD (IMMD), in an image frame, or inside one basal cell. The IMMD is defined as:

IMMD =
Standard deviation of MMD of segmented cytoplasm Mean MMD of segmented cytoplasm × 100%.
(4) Figure 4 shows the result of the IMMD analysis. The IMMD value in a single cell is also exemplified in Fig. 4. It is also noted that the IMMD values in Fig. 3 are 17.9%, 28.6%, 17.5%, 47.4%, and 38.6%, for ankle, dorsum of hand, ventral forearm, dorsal forearm, and face, respectively.

Melanin mass density verification
To verify our measured absolute MMD values with existing methods, we conducted further observational studies on healthy volunteers. DRS is the only accepted method to apply in vivo clinically to estimate the epidermis MMD. We estimated the average MMD in the epidermis using DRS with Monte Carlo modeling of photon propagation in tissue. We also obtained the average MMD values by using HGM in the same regions of human inner forearm skin to minimize the influence of sunlight exposure. Age distribution, average age, medium age, skin type, and gender distribution were all well controlled to be almost the same. We averaged the MMD value from different subjects after checking the normal distribution and removing the corresponding outliers. As a result, two outliers of the HGM measurements were removed. The average MMD value of the lower epidermis layer taken from DRS measurements was 6.41 mg/ml with a standard deviation of 3.35 mg/ml and a medium value of 7.44 mg/ml. For the HGM value, which was 8.65 mg/ml, it only represented the concentration inside the cytoplasm. After considering the volume nucleus-cytoplasm ratio, which is 0.349 [22], THG obtained the average MMD in keratinocytes around DEJ as 6.41mg/ml with a standard deviation of 0.71 mg/ml and a medium value of 6.42mg/ml. This excellent agreement on the average MMD value indicates the correctness on the HGM obtained absolute MMD quantities and supports our proposed calibration protocol.

Applications
For the purpose of diagnosis, melanin can play as a natural biomarker for pigmentary disorders. For our quantitative imaging, our obtained MMD value can be further treated as the number of melanosomes inside the focal volume. For most of the pigmentary disorders, it is the information regarding chronic progression of abnormalities as well as changes in skin dynamics that helps dermatologists to understand the development of diseases as well as the treatment effects. This is hardly achievable by biopsies because once excised, the specimen no longer remains on the site. To deal with the issue, noninvasive quantitative imaging techniques capable of in vivo pathological diagnosis as well as longterm tracking on the sites are optimally suggested. There are two aspects of the superiority of erTHG microscopy. The first is that our adopted wavelength (1230-1260nm) is capable of penetrating the pigmentary lesional sites, by avoiding the melanin absorption spanning from UV to 1100nm, to characterize the histopathological features with a sub-femtoliter resolution. The second is that erTHG can be used to quantitatively monitor MMD changes induced by treatments.  Solar lentigines (SL), a type of sun-induced hyperpigmentation, can be observed in as many as 90% of whites older than 60 years even for those who diligently used sunscreen [37]. Such a high incidence has brought a huge demand in the field of aesthetic medicine and inspired dozens of research [37][38][39][40]. Despite the amount of research, there is still unrevealed factors with regard to the pathogenesis and development of SL. One of the characteristics of solar lentigines is accumulation of melanin in the basal layer of the epidermis. In the early studies observing H&E stained histology sections, crude melanin content was estimated to support this conclusion [38]. In the report by Andersen et al [39], melanin content was quantified from FM-stained sections of facial SL as a percentage of epidermal area, not by melanin content with absolute values. Using RCM, Pollefliet et al [40] observed significantly increased melanin-containing keratinocyte clusters, a morphological character, but they did not quantify the melanin content or concentration.
By using erTHG, MMD of basal cells concerning absolute melanin contents in SL were quantified (Fig. 5). In vivo erTHG microscopy shows increased MMD in basal cytosols near the DEJ, while after treatment, not all SL patients were with decreased MMD in basal cytosols ( Fig. 5(b)). The change in average MMD value after treatment is rather subtle and is lesion dependent. Our further analysis indicated that not only MMD, the IMMD value in basal cytosols also increased in the lesion, as summarized in Fig. 6. This MMD inhomogeneity does not decrease after treatment. The significant increase of MMD standard deviation cannot be attributed as lesional-dependent estimation error since the imaging of lesional and normal sites were taken under exactly the same condition. With a high recurrence rate after the laser treatment, this IMMD parameter could be a critical one to reveal the pathogenesis and development of the SL lesion. Our further analysis supports high inhomogeneity inside the cytosol of each basal keratinocytes, as exemplified in Fig. 4, rather than between keratinocytes. High IMMD value would thus suggest the increased population of melanosomes with an unusually larger size close to the in vivo THG microscope lateral resolution, which is >450 nm, according to a previous study [41] averaged with data of six volunteers of 20 to 30 years. The increased absolute melanin content indicates an upregulated melanosome uptake, while the enhanced distribution inhomogeneity further indicates the abnormality on the melanosome dispersion mechanism [42], which is not recovered after treatment. Even though more study will be needed, our observation revealed the unique capability of erTHG microscopy for not only longterm treatment assessment but also direct clinical observation to shed light into the unknown pathway and regulation mechanism for melanosome transfer to and translocation in the keratinocytes [43].

Error estimation and photodamage
Another ratio-based metric, redox imaging, is troubled by the depth-dependent variation [44]. Different from redox imaging of NADH and FAD of different wavelengths, erTHG is based on ratio of THG versus THG values, and is thus not affected by the wavelength dependent attenuation which is different from sample to sample. However when applying erTHG to image melanin mass density distribution in vivo, there are possibilities for the estimated value obtained in a single voxel to deviate from the true value of melanin quantity, due to not only the background noise, but especially the non-specific THG background. It is noted that the background noise was measured and deducted from the measured signals when calculating the erTHG value according to Eq. (3). THG background signal could contribute from various sources. The method proposed in this manuscript is based on the fact that melanin provides the strongest THG signal in skin, and we have treated other THG sources combined as our background THG. Even though Figs. 2 & 3  (c). The representative in vivo HGM images acquired at the facial skins of the normal region before laser treatment, the lesioned region before laser treatment, and the lesioned region after laser treatment of three patients. Slide-free images were en face optically sectioned. The corresponding MMD distribution images inside the cytoplasm of basal cells at the DEJ are also provided. In some cases, increased MMD mean value in lesion was observed after the laser treatment. It is noted that low THG intensity in deep layers might not correspond to low melanin density. Epi-SHG and epi-THG are represented by green and magenta pseudo-colors. MMD distribution image is color coded. Imaging depths beneath the skin surface, MMD mean, and IMMD values of the images are provided. show that the melanin-generated THG could be up to 15 times stronger than the THG background, estimation error is unavoidable. Estimation error could also be contributed by the size dependent effect of THG. Cheng & Xie [45] used Green's function formulation to theoretically study THG microscopy and found that the size and shape of the sample in the focal volume will significantly affect the intensity of THG. With most of the melanosomes smaller than the laser focal volume, the THG intensity corresponding to one specific melanin density could be affected by different melanosome size and spatial distributions. As a result, transforming a fixed THG enhancement ratio into a MMD value would result in some error. Since our verification study confirmed the accuracy of the mean value, it is then important to estimate the maximum error from our experimental data. Our study observed the standard deviation values (IMMD) for various images including those in Fig. 3(c), which is 17.5% for ventral forearm and 17.9% for ankle. Under the extreme assumption that all live cells are with uniform MMD distributions, the observed standard deviation will reflect the maximum error of a pixel to be provided by the methodology. In real situation, the cell MMD distribution will never be uniform so that the MMD estimation error on a single pixel should thus be significantly lower than 17.5%. To further improve this estimation error, one might consider to average the obtained values throughout a number of pixels, while the error would decrease with the square root of the involved pixel number. With a high spatial resolution, the pixel number within the cytoplasm of a single basal cell is on the order of 300, which makes the maximum error of the mean MMD of a single cell less than 1%. If the melanin mean value is for one ROI with more than 10 cells, the MMD estimation error would be even lower.
For all clinical erTHG microscopy images shown in this study, basal cells at DEJ with a depth between 47 to 100 µm were in vivo imaged. erTHG ratio is calculated based on cytosol THG and collagen THG. To avoid the depth dependent variation on the ratio-based metric of erTHG, it is advised to acquire the mean THG Collagen value of Eq. (3) based on the collagen within the same field of view (FOV) and similar depth as the melanin, as exemplified in all presented cases in this study. With the same collection wavelength and the same collection path/depth, the tissue reabsorption and scattering factors of THG melanin and THG Collagen can be cancelled. As shown in Fig. 5(a), the basal cells of the normal skin tissue acquired at the depth of 47 µm is with a much stronger THG intensity than those of the SL lesion acquired at a depth of 100µm. However, strong THG intensity might not reflect high melanin content. Through calibration with the THG Collagen of the same FOV and similar depth, different from quantitative THG, erTHG properly reflects the higher melanin content (13.7 mg/ml) of the SL lesion. Since collagen fibril is abundant in the dermis skin layer deeper than the DEJ, to maintain the the reliability of this ratio metric in deep layers, it will then be critical to maintain good THG signal-to-noise ratio not only in melanin-rich area but also in collagen fibers of the same FOV and depth.

Photodamage and laser induced measurement bias
Photodamage, either linear or nonlinear, is a concern for in vivo human study. Extensive studies and discussions have been conducted and previously published to address this concern, especially for the 1230-1260 nm based femtosecond THG microscopy with an average power on the order of 100 mW [22,34,[46][47][48][49][50]. No evidence of photodamage can be found either in clinical trials [22,32,34,35,[51][52][53][54][55] or studies on animal models [22,[47][48][49][50]. This is attributed to the low photon energy (∼1eV) of the adopted excitation light, which avoids the linear melanin absorption [22] and significantly increases the multi-photon ionization threshold [34], when compared with femtosecond excitation light with a higher photon energy. In this study, we concern the photodamage by the increased melanin content in pigemented lesions, which are the primary application areas of the demonstrated erTHG microscopy. In a previous in vivo THG study on pigemented skin lesions [53], including pigmented basal cell carcinoma, melanocytic nevi, seborrheic keratosis, the accumulated photon energy (at 1230 nm) was ∼180 J in each volunteer. Under such an accumulated light dose, no evidence of photodamage, such as coagulation necrosis, was found for all followed-up histological examinations by pathologists on the illuminated specimens after surgical removal. As described in the clinical protocol section, none of the subjects under this study reported any discomfort relating to the imaging laser, such as a burn or a sting. When all subjects went to clinic after the image acquisition the dermatologist observed no symptom or reaction concerning inflammation, no change of appearance, no pigment alteration, nor blister and ulceration on the site where images were taken. No evidence of photodamage was found.
The other concern is the laser-induced measurement bias. There are studies showing that nonlinear imaging using light in the near-infrared (NIR) region can change the melanin distribution in the process of measuring [56,57], due to linear NIR absorption of melanin [57]. To evaluate the possible laser-induced measurement bias, we had repeatedly imaged in vivo the ventral forearm skin of one healthy volunteer at the fixed depth of DEJ for a period of 1 minute. The total accumulated dosage was 6J. The imaging condition is the same as described in Section 2.1 with an average power of 100 mW after the objective. The acquisition time was approximately Fig. 7. In vivo erTHG microscopy images repeatedly measured in the ventral forearm skin of one healthy volunteer at the fixed depth of DEJ for a period of 1 minute. The acquired MMD distribution images inside the cytoplasm of basal cells at the DEJ are provided at the exposure time of 0.77, 1.1, 29.4 and 57.6 seconds. All images were segmented by the program and no manual selection was applied. The differences between the basal cell distribution and the MMD mean value are attributed to the minute movement of the volunteer during the measurement period. MMD distribution image is color coded. MMD mean values of the images are provided. 0.38 second per averaged image, after averaging 5 frames at a fixed depth with a 15 Hz frame rate. To ensure no human bias, the cytosol segmentation was purely performed by the same program and no manual selection was applied. Figure 7 shows the MMD images thus acquired within the first 1.1 seconds, the second and the third images, the 76 th image, and at the end of the one minute, the 149 th image. Stable MMD value can be obtained, indicating negligible laser-induced measurement bias in our proposed method, especially for the practical acquisition condition with the laser dwelling time in one fixed depth of 0.38 second and the laser dwelling time of the whole 3D stack on the order of or less than 40 seconds. We attribute our observation to negligible linear absorption of melanin to our excitation light, which is spectrally located out of the melanin absorption band.

Conclusion
The capability to in vivo image the 3D distribution of melanin in human skin with absolute quantities and microscopic details will not only enable various noninvasive pathogenesis studies and diagnosis on pigmentary disorders, but also make long term treatment assessment possible. Traditional histopathological diagnosis uses exogenous stains to label melanin while the process is invasive, time consuming, labor intensive, and is not able to provide quantitative mass density information. In this paper, we successfully demonstrate clinical in vivo imaging of the melanin distribution in human skin with absolute quantities on mass density and with microscopic details by using label-free erTHG microscopy. As the dominant absorber in skin, melanin provides the strongest THG nonlinearity in human skin through resonance enhancement. Through the calibration of the enhancement ratio, we have successfully established the erTHG parameter which is directly linked to the local melanin mass density in a sub-femtoliter volume. A comparison study with DRS confirmed the correctness of the obtained quantity in term of absolute value of melanin mass density. A repetitive scanning study indicated no laser-induced erTHG measurement bias. Preliminary clinical study on the treatment assessment of patients with solar lentigines was demonstrated to further illustrate the unique information to be provided and revealed by this novel technique.