1.

Introduction

The incidence and mortality rates of skin cancer have increased dramatically during the last decade until it has become the malignancy with the highest incidence in the Brazilian population. In 2012, the estimated incidence of nonmelanoma skin cancer in Brazil is about 134,000 new cases.¹ The high incidence of skin cancer in Brazil is related to the long exposure of workers/farmers to sunlight in the countryside, and of people pursuing leisure (mainly the beach). Despite a permanent government campaign urging people to avoid unprotected exposure to sunlight,² which would ultimately reduce the incidence rate of skin neoplasias, only early detection would help to decrease mortality rates.

Biopsy of suspicious lesions followed by histopathological analysis is considered the gold standard for skin cancer diagnosis. In spite of different clinical presentations for basal cell carcinoma (BCC) lesions, the most pronounced characteristic is an asymptomatic nodule of rosy or translucid lesion, with either a pearly, flat, and shiny aspect or ulcerated appearance, containing telangectasic vessels.³ Skin melanoma (MEL) appears initially as a dark lesion with progressively increasing size, accompanied by alterations in the original color, pigmented dots at the lesion border, ulcerations, and bleeding followed by pain, itching, or inflammation.⁴ Diagnosis is performed by well-trained specialists able to differentiate among common skin pigmented lesions. Up to 50% of early malignant skin lesions may escape detection during clinical routine examinations, while experts achieve an 80% to 90% success rate of detection.⁵

The challenge for modern medicine is to develop an analytical technique that relies upon the morphological and biochemical changes in different tissues, and could give early diagnostic information in real time, noninvasively and nondestructively, in situ. Vibrational spectroscopy, particularly Raman spectroscopy, has the potential to diagnose and study the evolution of human malignancies both in vitro and in vivo in prostate,^6,7 esophagous,⁸ stomach,⁹ lung,^10,11 breast,^12,13 and arteries,^14–16 among others.

Abnormal tissues have differences in their morphology and biochemistry, reflected in the levels of proteins, lipids, and nucleic acids observed compared with normal tissues; these are mainly caused by differences in neoplastic metabolism, or changes in cellular and subcellular structures and functions.^6,17–19 These changes should be detectable by Raman spectroscopy and, therefore, this technique has been considered a promising tool to discriminate the progression from benign to malignant tissues in different pathologies in vivo.¹⁷ The use of near-infrared excitation (typically 785 or 830 nm from diode lasers) has the important advantage of decreased sample fluorescence in biological specimens.^14,20 The use of a fiber-optic Raman probe would provide the capability to perform specific molecular fingerprinting and analysis in in vivo experiments,^13,16,17,21 and detect skin tissue alterations using higher wavenumbers.²² Automated Raman systems for rapid in vivo biological tissue evaluation in less than 1 s have been proposed.^16,21

Spectral differences between nonmelanoma and melanoma cancers in skin tissues have been studied by Raman spectroscopy.^5,22–30 It has been proposed that Raman confocal microspectroscopy be employed to observe spectral differences in tissue ultra structures in biopsy fragments from normal and BCC in order to provide diagnosis of neoplasia.^28–30 Raman spectroscopy has been employed to discriminate BCC from normal skin cancer in vitro through Artificial Neural Network Analysis and Principal Components Analysis algorithms,^23,25 and through a simplified biochemical model using spectra of tissue constituents (collagen and cell fat features).³¹ A multimodal imaging spectroscopic approach (fluorescence, Raman, and 2nd harmonic generation) has been proposed to elucidate the morphochemistry of BCC.³² Recent studies indicated the applicability of Raman spectroscopy to clinically detect secure surgical boundaries in nonmelanoma skin tumor removal procedures.²⁶ In the case of Raman spectra for discriminating MEL, investigators have used FT-Raman (1064 nm excitation) to discriminate between commonly found skin lesions which can be confused with melanoma (BCC, nevi, and keratosis).³³ These studies have indicated a high sensitivity for differential diagnosis among MEL and pigmented nevus in vivo using 785 nm excitation,³⁴ and among MELs from BCC and squamous cell carcinoma, with a maximum sensitivity and specificity of 100% using near-infrared Raman microspectroscopy.²⁹ The amount of water in skin lesions also has been measured, indicating that BCC and MEL have higher water contents compared to normal tissues.³⁵

Recently, a number of studies proposed that the absolute or relative concentrations of most relevant biochemicals and morphological structures present in a particular tissue sample could be estimated using the Raman spectra of such biochemicals or morphological structures linearly fitted to the spectra of tissue samples.^{6,9,11,13,14,18,19,31} Then tissue discrimination and disease grading could be performed by comparing the alterations in these compounds that occur in each tissue type, using a suitable discrimination technique. The application of such a spectral model, based on the biochemical constitution of altered skin compared to normal tissue, would help reveal the most pronounced spectral features responsible for the observed skin tissue spectra, and develop a diagnostic tool based on spectral changes which accompany morphological and biochemical changes.

The objective of this work is to use near-infrared Raman spectroscopy to develop a spectral model based on the estimated Raman contribution of selected basal biochemical compounds including particular proteins, lipids, amino acids, and nucleic acids to the observed Raman signal obtained from normal and neoplastic (BCC and MEL) skin tissues in vitro, obtained by modeling a linear least-square fitting of the selected biochemicals. These contributions are then used in a discriminating model in order to group such tissues according to the differences in the estimated contribution of each biochemical, which can be related to the differences in tissue constitution.

2.

Materials and Methods

This study followed Brazilian guidelines for research with humans/human materials, and was approved by the Council on Ethics in Human Research (CEP)—Unicastelo. Skin tissue fragments of about 2 mm were withdrawn from the center of resected lesions obtained from excisional biopsies, snap frozen, and stored in liquid nitrogen ($-196°\mathrm{C}$) for spectroscopic study.

Before collecting the near-infrared, dispersive Raman spectra, all samples were warmed to room temperature and kept moisturized with 0.9% saline solution. Saline solution is known not to interfere with spectral collection, since the concentration of the NaCl is very low, and the Raman band for the dissolved electrolytes present as mono-atomic ions is not Raman active.³⁶ We used a portable near-infrared, dispersive Raman system (P-1 Raman system, Lambda Solutions, Inc. MA, USA), with an 830 nm excitation, adjustable laser power up to 350 mW, and spectral resolution of about $2{\mathrm{cm}}^{-1}$ in the range of 400 to $1800{\mathrm{cm}}^{-1}$. The spectrometer was connected to a Raman probe (Vector probe, Lambda Solutions, Inc. MA, USA) about 3 m long, with band pass and rejection pass filters. The $1320\times 100\text{pixel}$, back thinned, deep depleted CCD was cooled (Peltier) down to $-75°\mathrm{C}$ to decrease thermal noise.

For spectral collection, each humid tissue fragment was placed in a sample holder made of aluminum; then the probe was placed at a 10 mm distance perpendicular to the tissue surface. The signal scattered by each sample was then collected by the probe and coupled to the signal port of the P-1 spectrometer for dispersion and detection. The gross spectra were then stored. Each sample was referred for reading on the same day and experimental conditions (temperature and humidity). The Raman signal was collected in 5 and 10 s scans for all samples. By increasing the number of scans, we increased the signal-to-noise ratio (SNR) by the root square of the number of scans while not changing spectral resolution.³⁷ The laser power was set to 200 mW for normal and BCC samples, and reduced to 50 mW for melanoma samples, in order to avoid thermal damage due to the strong absorption from melanin.

The procedures of spectra calibration (pixel to Raman shift correlation and spectrometer spectral response correction) were performed using the software Matlab (The Mathworks, version 5.2), following a procedure described elsewhere.³⁸ Background fluorescence was removed by a baseline correction function in commercially available software such as OriginPro. In this approach, straight lines are drawn at selected positions in the spectrum to create a baseline, and then these lines are subtracted from the gross spectrum. Cosmic rays were removed manually.

After calibration and pre-processing, spectra were normalized according to the most intense band around $1450{\mathrm{cm}}^{-1}$, mean-centered, and plotted in the spectral range of 400 to $1800{\mathrm{cm}}^{-1}$.

Two to five spectra were collected from each fragment, for a total of 47 samples with the following diagnostics confirmed by histopathology: 15 normal skin samples (N), 29 BCCs, and 4 MELs. We also collected spectra from one sample of keratosis, one sample of poroma, and one sample of fibrosis, which were withdrawn due to the small number of samples. Some melanoma spectra presented very low SNRs (due to background fluorescence) and were also withdrawn from the study. A total of 30 spectra from normal, 96 spectra from BCC, and 19 spectra from MEL tissues were considered for analysis.

We developed a spectral model based on the relative Raman contribution of selected basal biochemical compounds responsible for Raman bands that are present in N, BCC, and MEL skin tissues. These biochemicals were chosen based on their probable occurrence in skin tissues, and likely responsibility for the spectral features that differentiate both neoplasias from normal sites. The biochemicals (Table 1) were bought from Sigma-Aldrich Brazil or donated from Terapêutica Farmácia de Manipulação, São José dos Campos, SP, Brazil, and most of the spectra were collected from the pure form (as received); some were diluted in water. The time exposure for collecting those spectra varied from 0.01 to 0.1 s for amino acids, 0.1 to 1 s for proteins, nucleic acids and lipids, and 1 to 2 s for carbohydrates and phospholipids. Laser power was set to 300 mW, exception for the pigments, when it was set to 50 mW. These spectra were also submitted to pre-processing and normalization according to the most intense band.

Table 1

List of the biochemicals used in the model and origin of them in the skin structure.

The spectral model was developed by calculating the relative Raman contribution of the selected biochemicals in each tissue fragment using the unique biochemical information (fingerprint) provided by the spectra for each pure compound. Fitting the spectra of tissue constituents to the spectra of skin tissues was performed using ordinary least-square analysis, according to the expression:⁶

Observation of the fitting residual enabled the quality of the fit to be observed. The fitting residual $E$ was calculated as follows:

The Euclidean distance was employed to separate the dataset into classes according to the histopathology (discriminant analysis), by determining the least linear distance from a specific point to the center (mean) of the class to which it was thought to belong, and comparing the distance to the vicinity group. The Euclidean distance $d$ follows the expression:⁴⁰

The least-squares model was developed under Matlab 5.2 (Mathworks, version 5.2), as well as the calculations of the residuals and the Euclidean distances.

3.

Results

Figure 1 shows the average normalized Raman spectra of N, BCC, and MEL skin tissues in the range of 400 to $1800{\mathrm{cm}}^{-1}$. The observable Raman bands of skin can be assigned mainly to proteins/amino acids, lipids/phospholipids, and nucleic acids as shown in Table 2. Important spectral differences were found in the regions between 800 to $1000{\mathrm{cm}}^{-1}$ and 1200 to $1400{\mathrm{cm}}^{-1}$, corresponding to bands assigned to nucleic acids, lipids, and, primarily, proteins. Almost all peaks showed statistically significant differences between tissue types ($p<0.05$, ANOVA).

Fig. 1

Mean normalized Raman spectra of N, BCC, and MEL with vertical dotted lines labeling the peaks described in Table 2. Laser power: 150 mW, wavelength: 830 nm, spectral resolution: $2{\mathrm{cm}}^{-1}$, integration time: 2 s, number of scans: 10. Spectra were offset for clarity.

Table 2

Peak positions of main Raman bands of N, BCC and MEL skin tissues with the respective tentative assignments.

The spectrum of normal skin had features of proteins present in the dermis (mainly actin, collagen, and elastin) that can be identified by bands at 857, 939, 1004, 1248, 1271, 1452, and $1658{\mathrm{cm}}^{-1}$, as well as several other weaker vibrations in the 400 to $1000{\mathrm{cm}}^{-1}$ region. The bands at 1063, 1128, the region from 1270 to 1300, and at $1452{\mathrm{cm}}^{-1}$, could be attributed to the saturated fatty acids of ceramides in the epidermis, and the phospholipids sphyngomyelin and phosphatydilcholine in the cell membrane. The band at $718{\mathrm{cm}}^{-1}$ could also be attributed to phospholipids. Spectral features arising from unsaturated lipids (mainly triolein from adipocytes) in the skin appeared mainly at 1092, 1271, 1301, 1452, and $1658{\mathrm{cm}}^{-1}$.

BCC skin had several spectral features in the same positions as normal skin, indicating similar biochemical constitution, with remarkable differences in the intensities in the 800 to $1000{\mathrm{cm}}^{-1}$ and 1200 to $1400{\mathrm{cm}}^{-1}$ ranges compared with normal. Also, small peaks in the 600 to $1000{\mathrm{cm}}^{-1}$ region, and peaks at 1004, 1092 and $1128{\mathrm{cm}}^{-1}$ were of higher intensity in BCC, indicating a higher lipid content, which could be relevant for spectral diagnosis. The melanoma spectrum was characterized by strong near-infrared fluorescence (not shown) from the melanin. Due to this fact, laser power was reduced to 50 mW to avoid tissue burning and CCD saturation. Despite the higher noise level of melanin spectra due to the strong melanin fluorescence, the spectral features were likely closer to the ones found in N tissues in the regions 600 to $1000{\mathrm{cm}}^{-1}$ and 1200 to $1400{\mathrm{cm}}^{-1}$, with the peaks at 1092 and $1128{\mathrm{cm}}^{-1}$.

In order to evaluate differences in the relative amount of the most relevant tissue biochemicals that were responsible for the Raman bands in normal and malignant skin tissues, a spectral model was developed based on the relative contribution of those biochemicals found in each tissue, by using the Raman spectra of pure compounds and calculating the fit parameter of all biochemicals within the tissue spectrum according to least-squares minimization following Eq. (2). Figure 2 shows the spectra of biochemicals that presented bands close to the ones found in spectra of normal and malignant skin tissues in which the fit parameter resulted in nonnegative contribution in at least two tissue types. Most spectra were collected from the biochemicals as they were purchased; phenylalanine and melanin were diluted in water.

Fig. 2

Raman spectra of the most relevant biochemicals used in the spectral model. Wavelength: 830 nm, spectral resolution: $2{\mathrm{cm}}^{-1}$.

Figure 3 presents the calculated mean relative Raman contribution of the biochemicals according to each tissue type. It has been found a statistically significant difference in the fitting parameter of the selected biochemicals relative to tissue types ($p<0.05$, ANOVA). In Fig. 4, one can observe the resulting spectral model, in which the basal spectra multiplied by the estimated fit parameter, were superimposed upon the mean spectra of N, BCC, and MEL. The corresponding fitting residual for each tissue mean spectra showed good agreement with the spectral model.

Fig. 3

Plot of the mean and standard deviation of the relative Raman contribution (fit parameter $c$) of each basal compound used in the model for each tissue type.

Fig. 4

Plot of the mean Raman tissue spectra and the mean spectra of the model fit, with the mean residual spectrum (left) and the corresponding percentage of the residues of the fitting for each tissue fragment (right) for A: N, B: BCC, and C: MEL.

In order to develop a discrimination model for tissue classification in N, BCC, and MEL based on the estimated Raman contribution of the selected biochemicals, pairs of biochemicals with greater significance in their fit parameters for N and BCC (actin, collagen III, and elastin), for BCC and MEL (actin, melanin, and triolein), and for N and MEL (melanin and nucleic acid), were plotted on a binary scale and the Euclidean distances were calculated. Figure 5 shows the scatter plots for collagen III versus melanin, actin versus melanin and nucleic acid versus actin. Actin versus nucleic acid showed the highest discrimination capability. We also tested combinations of three basal compounds; Fig. 6 shows a 3-D scatter plot of collagen III versus elastin versus melanin. Table 3 shows the sensitivity, specificity, and accuracy of discrimination results using these biochemicals. The combination of collagen III, elastin, and melanin showed better discrimination among groups (Euclidean surfaces not plotted), and higher sensitivity and specificity values.

Fig. 5

Scatter plot of the fit parameter of actin, collagen III, melanin and nucleic acid predicted by the model. Separation of N, BCC, and MEL was done according to the mean Euclidean distance between groups.

Fig. 6

3-D scatter plot of the fit parameter of collagen III, elastin, and melanin estimated by the spectral model. The Euclidean distance separation surface is not shown.

Table 3

Results of the discrimination model in terms of absolute values and sensitivity, specificity, and accuracy.

4.

Discussion

Several authors have proposed the use of spectral models to estimate the relative contribution of important biochemicals in Raman spectra of bio-tissues for cancer diagnosis.^{6,11,13,18,19,31} Haka et al.¹³ used a morphological/biochemical model to correlate changes in the amounts of fat (adipocytes), collagen, cholesterol, and calcium oxalate in the cell nucleus and cytoplasm, aiming at breast cancer diagnosis in vivo. Brennan III et al.¹⁴ used Raman spectra from human coronary arteries for in situ histochemical analysis, accessing the amounts of cholesterol, cholesterol esters, triglycerides, phospholipids, and calcium salts. Motz et al.¹¹ used a Raman probe connected to a near-infrared Raman spectrometer to scan carotid arteries in vivo in order to access the biochemical constitution of plaques, including cholesterol, collagen, and adipocytes (adventitial fat) present in the sample, correlating the plaque composition with plaque vulnerability. Stone et al.⁶ diagnosed urological pathologies (bladder and prostate cancer) through Raman spectroscopy, by quantifying differences in actin, collagen, choline, triolein, oleic acid, cholesterol, and DNA, assessing the gross biochemical changes in each pathology. To biochemically evaluate bladder tissue, Lyng et al.¹⁸ compared the spectral features of cervical cancers to the pure spectra of the most relevant biochemicals, such as amino acids, nucleic acids, proteins, and lipids. De Jong et al.¹⁹ showed that spectra from nontumor tissue showed a higher collagen content, while spectra from tumor tissue were characterized by higher lipid, nucleic acid, protein, and glycogen. Huang et al.⁹ showed that albumin, nucleic acid, phospholipids, and histones were found to be the most significant features for construction of a diagnostic model for epithelial neoplasia of the stomach, giving rise to an overall accuracy of 93.7%. We have also implemented simplified biochemical/morphological models using spectra from tissue constituents obtained directly from tissues to discriminate BCC from normal tissues,³¹ and calcified and noncalcified atherosclerotic plaques from nonatherosclerotic human coronary arteries.⁴⁷ These works demonstrated that major biochemicals could be used to discriminate between tissues according to pathological status.

The spectral differences between BCC and MEL compared with N presented here are in good accord with the recent literature. Spectral changes in N and BCC revealed differences mainly in the amounts of proteins and lipids,^{22,25,26,28–30,33,34} which were confirmed by our recently presented simplified spectral model,³¹ and by the more complete model presented here.

We found changes in the relative contribution of proteins depending on the lesion (collagens I and III were decreased, and elastin and actin were increased in BCC). This indicates a change in the molecular composition of tissue proteins, as suggested by Gniadecka et al.³³ Actin expression plays a role in carcinoma cell organization and growth,⁴⁸ and has been proposed as a marker of invasiveness in BCC.⁴⁹ Telangiectatic vessels could contribute to the increase of elastin in BCC; as could their blood cells, with peaks at 1562 and $1613{\mathrm{cm}}^{-1}$.⁵⁰ We also found an increased contribution of triolein to BCC spectra. Gniadecka et al.³³ demonstrated that BCC and MEL had increases in the Raman bands corresponding to lipids (around $1300{\mathrm{cm}}^{-1}$), and decreases in the bands for proteins (1500 to $1800{\mathrm{cm}}^{-1}$ and 1310 to $1330{\mathrm{cm}}^{-1}$). This could be explained by the fact that growing tumor cells release collagenase that destroys native collagen fibers,³² eroding the extracellular matrix.⁶ Also, BCC tissues present fat reservoirs which might function as a nutritional source for the fast-growing tumor,³¹ reflected in our spectral model as higher triolein content. Stone et al.⁶ found increases in actin and triolein, and a decrease in collagen in urological carcinoma lesions, corroborating our results.

Melanomas exhibited very strong absorption and fluorescence from melanin, which caused tissue burning and a worsening in the SNR [Fig. 4(c)] when using 200 mW power (within the range used in many Raman studies). Due to this, the laser power was reduced to 50 mW to reduce the fluorescence background. However, we observed photobleaching by a visual diminishing in the fluorescence background during the spectrum acquisition of MELs. Neither fluorescence background nor melanin absorption in the near-infrared (800 to 1000 nm) are expected to change the spectral profile of skin and its biochemicals, since melanin absorption does not depend strongly on wavenumber.⁴² Nevertheless, it may affect the estimated fit parameter $c$ since melanin’s high absorption may saturate its Raman signal causing it not to increase linearly with melanin concentration.

It is interesting to consider the behavior of the MEL tissues in terms of tissue proteins (collagen and elastin), which were higher than in BCC tissues, despite the high error bars due to lower SNR. In skin melanomas, the malignant changes in melanocytes start at the membrane level, which undergo degradation of basal membrane and subsequent invasion of dermis. The morphological characteristic of the dermis remains morphochemically unchanged,⁵¹ which could explain this higher protein level. We found that six MEL spectra were misclassified as BCC. These spectra exhibited lower fluorescence than the remaining ones, and presented spectral features intermediate between N and BCC.

In MEL spectra, we found a higher intensity in the Raman band in the region between 1500 to $1700{\mathrm{cm}}^{-1}$, which could be attributed to the intense melanin band at around $1600{\mathrm{cm}}^{-1}$ (Fig. 2).⁴³ The spectral model revealed that melanin was higher in MEL compared with N and BCC, which was expected in these melanin-rich lesions. The relative contribution of nucleic acid was increased in BCC and even higher in MEL, suggesting higher replication rates for carcinoma and melanoma cells.

Our spectral model revealed that actin, collagens I and III, elastin, and triolein were the most important biochemicals for describing overall spectral features in skin tissues; in particular, proteins from the cytoskeleton (actin and collagen III), cell nucleus (nucleic acid), and pigments (melanin) could discriminate BCC and MEL from normal tissues with high sensitivity and specificity. The spectral residuals found in our model (Fig. 4) were on average about 6.7%, 7.4%, and 10.4% for N, BCC, and MEL, respectively. The higher residuals for BCC and MEL could be attributed to differences in the stages of the neoplasia (for BCC) and to the lower SNR for pigmented lesions (for MEL).

The choices of biochemicals used in the model were based mainly on their known presence in the tissue, and the contributions they would give to the observed tissue spectrum⁶ without mean negative fit parameter for at least two tissue types; exception for the phenylalanine, which was included in the model due to its unique spectral feature at $1004{\mathrm{cm}}^{-1}$, presented in all tissue types. Zhao et al.²¹ modeled spectra of normal volar forearm skin collected in vivo using oleic acid, palmitic acid, collagen I, ketatin, and hemoglobin. Either pure amino acids, fatty acids or membrane lipids were shown to be relevant to the model. Instead, use of spectra from proteins and triolein, the main lipid in cellular fat, were shown to better represent tissue characteristics.

A higher residual in the spectral model was found mainly in the region between 800 to $1000{\mathrm{cm}}^{-1}$, and to some extent in the 1200 to $1400{\mathrm{cm}}^{-1}$ region. This region is dominated by peaks of structural proteins (proline/hydroxyproline and amide III). It has been demonstrated that collagen bundles exhibit an anisotropic scattering depending on the alignment of the fibrils relative to the laser polarization, even without the use of an analyzer between the sample and the detector.^52,53 Principally, the Raman bands for amide III (NH deformation of the N-H groups) and proline/hydroxyproline (C-C stretching) become strong when the alignment of the polarized laser is parallel to the fibril’s long axis.⁵³ The band at $1452{\mathrm{cm}}^{-1}$ (deformation of ${\mathrm{CH}}_2$ and ${\mathrm{CH}}_3$ bands) does not show preferential orientation; therefore is not affected by the orientation of the fiber.⁵² These spectral residuals could also be attributed to the lack of an important biochemical in the model, for instance, another protein such as myosin, or other cytoskeletal or structural proteins that could be present in tissue; or even differences between the spectra of pure compounds and ones embedded and (both morphologically and biochemically) functionalized in real tissues.

Despite incremental in terms of Raman measurements of in vitro skin biopsy samples, some biochemicals used in this model have not been used before (such as melanin, actin, and ceramide) and those studies do not show diagnosis of all three tissues (N, BCC, and MEL) in the same model. The advantages of such a spectral model based on pure biochemicals are that it can easily be applied to any spectral data, without the need for a particular pre-processing. A prospective analysis using the new set of samples could be performed without the need to recalculate the parameters from a training dataset. This is because the Raman spectra of tissues can be considered a linear combination of the spectra of the basal biochemicals, and each biochemical is already a basis of the tissue spectral information.^6,14 Different biochemicals have different Raman cross-sections; therefore, their individual spectra would appear with different intensities even if they are present at the same concentration. As a consequence, one cannot take as absolute the contribution or concentration of each compound to the observed tissue spectrum; rather the relative spectral contribution (fitting coefficients) for each constituent should be considered and compared across the different tissue types.

Raman spectroscopy has the potential to become a technique for the biochemical analysis of skin cancer biopsies, or even in real time, in vivo and nondestructively using fiber optic Raman probes, differentially diagnosing BCC from MEL. Studies are under way in order to automate the data collection and evaluate the performance of the model at suspicious skin tissues in vivo, raising the possibility of a local, clinical diagnosis with rapid and reliable results during examinations, or even margin detection during surgery.