**5.1. Skin cancers**

the sample and, with this modification, the application of Raman in clinical fields increases by collecting the scattered light away from the point of laser illumination [7]. Transmission Raman spectroscopy (TRS) is the term used when the collection and illumination points are on opposite sides of the sample, and it is quite useful to analyze opaque materials. Besides, it was proved to be very useful to read many millimeters of tissues [8]. To increase the resolution of Raman microspectroscopy, it is possible to use Coherent anti-Stokes Raman spectroscopy (CARS) or stimulated Raman spectroscopy (SRS), both based on nonlinear optical effects that upgrade the spatial resolution. Coherent Raman spectroscopy techniques are based on nonlinear effects to increase the speed and spatial resolution of Raman spectroscopy by inducing coherent molecular vibrations in the sample increasing resolution. CARS is used to obtain image from cells and tissues by exciting the CH stretching vibrations of lipids and proteins and SRS allows to obtain high-speed images [9, 10]. Raman spectroscopy is also adaptable for fiber-optic probes, making it valuable for medical diagnosis *in vivo*, for instance in hollow organs. The probe should have very reduced dimensions to permit the access to body cavities,

and the spectra acquisition time should be short to allow accurate measurement [11].

Multivariate analysis tools are used to extract information from spectral data set and to assist in biomedical interpretation. Chemometric methods are diverse and offer different approaches to extract specific information from the data. These methods used for data interpretation and to extract information from complex datasets were usually univariate and therefore they were not suitable to describe the sample variation or composition. A range of processing methods can be applied to a Raman data set; usually baseline corrections should be performed as a first approach, prior to normalization methods and then a wide range of multivariate analysis tolls

Multivariate methods include techniques of multivariate classification (or pattern recognition techniques) and multivariate regression. Techniques of multivariate classification can be divided into unsupervised and supervised learning procedures. In unsupervised pattern recognition techniques such as principal-component analysis (PCA) and cluster analysis (CA), there is no need for *a priori* knowledge about the training set samples (spectra). These methods are used to consider differences and similarities between spectra. By contrast, supervised pattern recognition techniques such as linear discriminant analysis (LDA) and artificial neural networks (ANNS) require some *a priori* knowledge, for example, undoubtedly identifying samples from disease and samples from healthy cases. In this way, supervised procedures allow a more precise classification, while unsupervised methods are useful for an exploratory

On the other hand, multivariate regression techniques (or multivariate calibration methods) are usually applied to analyze one or multiple molecules of a complex sample that possess overlapping spectroscopic signals. These techniques include principal components regression

(PCR) and partial least squares regression (PLS) (**Figure 1**).

**4. Multivariate analysis applied to spectroscopic data analysis**

can be applied [12].

278 Raman Spectroscopy

analysis of data.

Raman has been widely investigated to discriminate benign skin lesions from malignant lesions, mainly basal cell carcinoma and squamous cell carcinoma (**Table 1**). Despite the high complexity of skin tissue, studies *in vivo* show huge potential as a routine procedure in hospitals for cancer screening [19]. In 2008, a preliminary report used 289 patients with different

**Cancer type Study** 

Gastric intestinal pre-cancer

**type**

**Statistical analysis**

**Sensitivity and specificity (%)**

Skin *In vivo* PC-GDA, PLS 95, 54 NA [22] Skin *Ex vivo* LDA, MLR 100, 92 NA [44] Skin *Ex vivo* PCA 99.1, 93.3 Proteins, lipids and melanin [24] Skin *In vitro* GDA ~90, ~90 Phenylalanine, tryptophan and

Skin *In vivo* PC-GDA, PLS 90, 64 NA [21] Skin *In vitro* PCA 89, 93 Lipids and proteins [23] Skin *In vivo* PLS, LDA 91, 75 NA [20] Nasopharyngeal *In vivo* PLS 91, 95 NA [33] Oral *Ex vivo* PCA, LDA 80.7, 84.1 Nucleic acids, proteins, lipids [28] Laryngeal *In vivo* PLS-DA, LOPCV 93.3, 90.1 Nucleic acids, proteins, lipids [32] Esophageal *In vivo* PLS-DA, LOPCV 97, 97.4 Lipids, proteins [31] Esophageal *Ex vivo* PCA 90.5, 95 NA [45] Esophageal *Ex vivo* PCA, LDA 93, 95 NA [30] Esophageal *Ex vivo* PCA, LDA 71–81, 81–98 NA [29]

Nasopharyngeal *Ex vivo* PCA, LDA 90.7, 100 Nucleic acids, collagen,

Breast *Ex vivo* Direct analysis NA Carotenoids, lipids,

Gastric *Ex vivo* PCA-DA 90, 90.9 NA [46] Gastric *Ex vivo* PCA, LDA 100, 94.1 RNA bases, ribose [35] Gastric *Ex vivo* PCA, LDA 100, 97 Tyrosine, adenine, coenzyme A [34]

Breast *Ex vivo* PLS-DA, LOPCV 74.2, 86.4 Collagen, amino acids [39] Breast *Ex vivo* LDA, MNLR 95.6, 96.2 Nucleic acids, collagen, lipids [38]

Bladder *In vitro* PCA 98, 95 NA [41]

GDA: General discriminant analysis; LDA: Linear discriminant analysis; LOPCV: Leave-one patient-out cross validation; MLR: Multinomial logistic regression; MNLR: Multinomial logistic regression; NA: Not available; PCA: Principal component analysis; PC-GDA Principal component with generalized discriminant analysis; PLS: Partial least square;

Leukemia *Ex vivo* PCA, LDA 100, 100 Lipids, phospholipids, amino

Colorectal *Ex vivo* PCA, LDA 97.4, 100 Nucleic acids, saccharides,

*In vivo* PCA, LDA, LOPCV

PLS-DA: Partial least square discriminant analysis.

**Table 1.** Raman spectroscopy in cancer diagnosis.

**Spectral differences (cancer** *vs.*

Raman Spectroscopy Applied to Health Sciences http://dx.doi.org/10.5772/intechopen.73087

phospholipids, phenylalanine

carbohydrates, proteins

acids, carotenes

proteins

89.3, 92.2 Collagen, lipids, phenylalanine, proteins

**Refs.**

281

[25]

[27]

[36]

[40]

[43]

[42]

**normal)**

DNA bases

**Figure 2.** Main application areas of Raman spectroscopy in health science.

types of malignant and benign skin lesions [20]. The authors applied real-time *in vivo* Raman spectroscopy to acquire spectra of all samples and analyzed data using PLS and LDA analysis. Direct analysis of spectra showed distinct biomolecular signatures and ROC curves allowed to discriminate skin cancers from benign lesions with an area under curve greater than 0.9 [20]. Later, the same authors published another study *in vivo* using a higher cohort of patients and 518 skin lesions (both malignant and benign conditions that are visually similar to skin cancer) [21]. Using the same instrumentation as before [20], the authors analyzed data using PC-GDA and PLS analysis, once visual inspection of spectra did not show distinctive Raman peaks assigned to skin cancer [21]. Although Raman peaks are almost the same between normal and cancer samples, the intensity of the signals allows the discrimination between different types of lesions and are compatible with those obtained using histopathological methods. To complement and validate these results, in 2015, Zhao et al. [22] added 127 samples to the previous cohort, creating a consolidated group. Using real-time Raman spectroscopy and PC-GDA, an PLS analysis obtained the same discrimination as before, proving the ability of Raman spectroscopy for *in vivo* skin cancer detection. In 2014, other *in vivo* study successfully used Raman microspectroscopy to study 20 skin samples and detect basal cell carcinoma in tissue removed during surgery [21].

Besides *in vivo* studies, there are also some studies *in vitro* to assess the use of Raman spectroscopy for skin cancer detection. In 2010, Bodanese et al. used dispersive Raman spectroscopy and successfully distinguished normal samples from basocellular cell carcinoma [23]. They found spectral differences in the region between 800 and 1000 cm−1 and between 1200 and


GDA: General discriminant analysis; LDA: Linear discriminant analysis; LOPCV: Leave-one patient-out cross validation; MLR: Multinomial logistic regression; MNLR: Multinomial logistic regression; NA: Not available; PCA: Principal component analysis; PC-GDA Principal component with generalized discriminant analysis; PLS: Partial least square; PLS-DA: Partial least square discriminant analysis.

**Table 1.** Raman spectroscopy in cancer diagnosis.

types of malignant and benign skin lesions [20]. The authors applied real-time *in vivo* Raman spectroscopy to acquire spectra of all samples and analyzed data using PLS and LDA analysis. Direct analysis of spectra showed distinct biomolecular signatures and ROC curves allowed to discriminate skin cancers from benign lesions with an area under curve greater than 0.9 [20]. Later, the same authors published another study *in vivo* using a higher cohort of patients and 518 skin lesions (both malignant and benign conditions that are visually similar to skin cancer) [21]. Using the same instrumentation as before [20], the authors analyzed data using PC-GDA and PLS analysis, once visual inspection of spectra did not show distinctive Raman peaks assigned to skin cancer [21]. Although Raman peaks are almost the same between normal and cancer samples, the intensity of the signals allows the discrimination between different types of lesions and are compatible with those obtained using histopathological methods. To complement and validate these results, in 2015, Zhao et al. [22] added 127 samples to the previous cohort, creating a consolidated group. Using real-time Raman spectroscopy and PC-GDA, an PLS analysis obtained the same discrimination as before, proving the ability of Raman spectroscopy for *in vivo* skin cancer detection. In 2014, other *in vivo* study successfully used Raman microspectroscopy to study 20 skin samples and detect basal cell carcinoma in

**Figure 2.** Main application areas of Raman spectroscopy in health science.

280 Raman Spectroscopy

Besides *in vivo* studies, there are also some studies *in vitro* to assess the use of Raman spectroscopy for skin cancer detection. In 2010, Bodanese et al. used dispersive Raman spectroscopy and successfully distinguished normal samples from basocellular cell carcinoma [23]. They found spectral differences in the region between 800 and 1000 cm−1 and between 1200 and

tissue removed during surgery [21].

1300 cm−1, assigned to proteins and lipids, using PCA multivariate analysis, with high sensitivity and specificity [23]. Later, in 2012, the same author reported new results corroborating the previous study [24]. Using a Raman spectrometer attached to a fiber optic and PCA analysis, they analyzed 145 different samples of basocellular cell carcinoma, melanoma and without malignant lesions and were able to discriminate cancer and normal samples with sensitivity and specificity values over 90% [24]. In the same year, a different study was performed by Wang H et al. using HaCaT cells, melanocytes and their malignant derivatives [25]. They tested the ability of micro-Raman spectroscopy to separate different cell lines and found significant spectral differences between HaCaT cells and squamous cell carcinoma, melanocytes and melanoma cells as well as between all normal cells *versus* all tumor cells [25] (**Table 1**). The results of these *in vitro* studies are of extreme importance and can help the interpretation spectra of *in vivo* samples for cancer skin diagnosis.

fingerprint and high wavenumber Raman spectra to extract the maximum biological information and obtained sensitivity and specificity values about 97% for the diagnosis of esophageal squamous cell carcinoma [31]. Similar results were obtained with a probe designed to diagnose laryngeal cancer [32]. Analysis of 2124 Raman spectra of 60 patients during endoscopy showed sensitivity and specificity above 90% for the identification of laryngeal cancer when combined fingerprint and high-wavenumber spectra [32]. Recently, Ming et al. [33] performed a pilot study in 79 patients with and without nasopharyngeal cancer and in postirradiated patients. They detect a specific signature for each one of the three cohorts, which may indicate that Raman could not only be used for diagnostic purposes but also for surveillance in post-treated patients. Furthermore, the authors used a probe with only 1.8 mm, which is the smallest probe used in Raman diagnostics and is more suitable to be used in clinical

Raman Spectroscopy Applied to Health Sciences http://dx.doi.org/10.5772/intechopen.73087 283

Diagnosis of gastric cancer using Raman technologies relies either on the use of fiber optic probes or on SERS (**Table 1**). In fact, SERS was applied to plasma samples to detect gastric cancer in a noninvasive way [34], similar to what was done to diagnose nasopharyngeal cancer [27]. The authors use two cohorts, with a total of 65 samples (32 patients with confirmed gastric cancer and 33 control patients). Using PCA and LDA multivariate analysis, it was observed discrimination between cancer and normal samples with sensitivity and specificity of 100 and 97%, respectively. In 2012, the same methodology was applied to discriminate gastric cancer from normal controls based on serum RNAs, also achieving high sensitivity and specificity (100 and 94.1%, respectively) [35]. SERS seems to be a useful technique to apply in routine clinical diagnosis coupled, for instance, with endoscopy. Fiber optic probes can be used for *in vivo* identification of gastric metaplasia. Lin et al. [36] coupled fingerprinting and high-wavenumber Raman spectroscopy with a fiber optic Raman probe and were able to detect, in real-time, pre-cancerous gastric lesions. They acquired 4520 spectra in real time, during gastroscopy, and by using PCA and LDA analysis, they were able to identify precancerous lesions with high sensitivity and specificity [36]. This can improve early diagnosis of

Breast cancer is the second most prevalent cancer in the world and, it is the most common cancer in women, causing more than 500,000 deaths every year [37]. According to these statistics, it is not surprising that Raman spectroscopy has been used as a diagnostic tool for this disease. Kong et al. used Raman microspectrometry to detect ductal carcinoma in tissue excised during breast-conserving surgery [38]. They developed a model that allowed to discriminate normal and cancerous tissue in approximately 17 min, with sensitivity and specificity above 95% [38]. A different approach was used by Feng et al. in 2015 [39]. Similar to what this group did for other types of cancer (see **Table 1** for detailed information), they applied SERS to saliva proteins of 97 patients and were able to discriminate between control, benign tumors and malignant tumors with sensitivities and specificities between 72.7–75.8% and 81.2– 93.4%, respectively, using PLS-DA analysis [39]. These results give good perspectives for new

neoplasia and significantly improve the efficacy of treatments.

endoscopies [33].

**5.3. Gastric cancers**

**5.4. Breast cancers**
