Micro-Raman spectroscopic study of thyroid tissues

Micro-Raman spectroscopic study of thyroid tissues

Accepted Manuscript Title: Micro-Raman spectroscopic study of thyroid tissues Author: L´azaro Pinto Medeiros Neto Luis Felipe das Chagas e Silva de Ca...

843KB Sizes 1 Downloads 69 Views

Accepted Manuscript Title: Micro-Raman spectroscopic study of thyroid tissues Author: L´azaro Pinto Medeiros Neto Luis Felipe das Chagas e Silva de Carvalho Laurita dos Santos Cl´audio Alberto Tellez Soto Renata de Azevedo Canevari Andr´e Bandiera de Oliveira Santos Evandro Sobroza Mello Marina Aparecida Pereira Cl´audio Roberto Cernea Lenine Garcia Brand˜ao A´ırton Abrah˜ao Martin PII: DOI: Reference:

S1572-1000(16)30064-3 http://dx.doi.org/doi:10.1016/j.pdpdt.2016.11.018 PDPDT 866

To appear in:

Photodiagnosis and Photodynamic Therapy

Received date: Revised date: Accepted date:

18-5-2016 26-10-2016 29-11-2016

Please cite this article as: {http://dx.doi.org/ This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Micro-Raman spectroscopic study of thyroid tissues

Lázaro Pinto Medeiros Netoa, Luis Felipe das Chagas e Silva de Carvalhoa, Laurita dos Santosa, Cláudio Alberto Tellez Sotoa, Renata de Azevedo Canevaria, André Bandiera de Oliveira Santosb, Evandro Sobroza Mellob, Marina Aparecida Pereirab, Cláudio Roberto Cerneab, Lenine Garcia Brandãob, Aírton Abrahão Martina*


Laboratory of Biomedical Vibrational Spectroscopy, Institute for Research and Development (IP&D), Universidade do Vale do Paraíba (UniVap), Av. Shishima Hifumi, 2911, Urbanova, São José dos Campos, 12244-000, São Paulo, SP Brazil b Universidade de São Paulo, Faculdade de Medicina da Universidade de São Paulo, Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo - Av. Dr. Enéas de Carvalho Aguiar, 255, Divisão de Anatomia Patológica, Cerqueira Cesar, 05403000 - São Paulo, SP - Brazil

Address all correspondence to: Airton Abrahão Martin, Universidade do Vale do Paraíba, Instituto de Pesquisa e Diagnóstico, Laboratory of Biomedical Vibrational Spectroscopy, Avenida Shishima Hifumi 2911 - Urbanova, São José dos Campos, Brazil, 12244-000; Tel.: +55 12 39471165; Fax: +55 12 3947 1149; E-mail: [email protected]

Highlights    

Differentiation of normal and diseased thyroid improve patient diagnosis. Application in the diagnosis of cancer of undetermined histology. Raman spectroscopy use in identifying lesions in hard to reach places. Biochemical changes could be observed.

Abstract. Thyroid carcinomas are the most common endocrine malignancy. Inconclusive results for the analysis of malignancies are an issue in the diagnosis of thyroid carcinomas; 20% of thyroid cancer diagnoses are indeterminate or suspicious, resulting in a surgical procedure without immediate need. The use of Raman spectroscopy may help improve the diagnosis of thyroid carcinoma. In this study, 30 thyroid samples, including normal thyroid, goiter and thyroid cancer, were analyzed by confocal Raman spectroscopy. Principal component analysis (PCA), linear discriminant analysis (LDA) with cross validation and binary logistic regression (BLR) analysis were applied to discriminate among tissues. Significant discrimination was observed, with a consistent rate of concordant pairs of 89.2% for normal thyroid versus cancer, 85.7% for goiter versus cancer and 80.6% for normal thyroid versus goiter using just the amide III region. Raman spectroscopy was thus proven to be an important and fast tool for the diagnosis of thyroid tissues. The spectral region of 1200-1400 cm-1 discriminated normal versus goiter tissues despite the great similarity of these tissues.

Keywords: Raman spectroscopy, thyroid, papillary thyroid cancer, goiter, follicular thyroid cancer.

Introduction Thyroid carcinomas are the most common endocrine malignancy. The annual incidence has doubled over the last decade, with approximately 8,050 cases per 100,000 women expected in Brazil in 2014, representing 2.9% of new cancer cases and the fifth most common cancer type, with a rate of 7.91%. The incidence of thyroid carcinoma is lower among men, with 1,150 expected new cases per 100,000 men and a gross rate of 1.15% of all new cases of cancer1. The increased incidence of thyroid cancer has been attributed to the diagnosis of papillary microcarcinoma with a size of ~1 cm. Two hypotheses for this increasing incidence have been described2: a true increase in incidence3 and improvements in screening and diagnostic techniques4. Other studies have argued that the reason for this increase is multifactorial5 and that the real cause of the increased rate of thyroid carcinomas, false-positive results obtained by diagnostic techniques or increased efficiency in screening for small nodules, cannot be determined. The diagnosis of thyroid lesions is mainly performed by the fine-needle aspiration (FNA) technique. This procedure, usually guided by ultrasonography (US), is affordable, quick, and lacks major complications6. Clearly, FNA is essential for the evaluation of non-palpable nodules7. The accuracy of US-guided FNA varies from 65 to 95% for papillary, medullary, and anaplastic cancers. However, this technique is less effective in differentiating between follicular adenoma and carcinoma or between Hurthle cell adenoma and carcinoma8-11. This difficulty of differentiating follicular lesions, such as follicular carcinoma and follicular adenoma, is the main problem in the diagnosis of thyroid lesions. Depending on the clinical parameters of each patient, the sensitivity of FNA may be very low, approximately 66%12. Alterations in the equilibrium of the thyroid gland can lead to thyroidal nodules. These nodules are usually benign injuries (goiter or follicular adenomas). Goiter is an abnormal enlargement of the

thyroid gland. The primary cause of this abnormal enlargement is iodine deficiency. These nodules can be characterized as the simple presence of goiter or even the presence of cancer. Morphological changes, such as size, aspect and cell number, determine the type of pathology. Cellular changes are the main features used to assess and diagnose goiter and the presence of cancer. However, these changes are less useful for the differentiation of thyroid pathologies; for some types of disorders, alterations are minimal and difficult to observe by routine microscopy. Therefore, identifying new techniques enabling faster and more efficient diagnosis associated with histology is extremely important. Raman spectroscopy has shown significant results in studies with different types of cancer as a complement to histological diagnosis. This technique generates a biochemical analysis of the entire tissue by applying laser light scattering to the material, enabling the measurement of differences in intensity vibrational modes related to biochemical components that may be associated with markers of certain types of cancer versus normal tissue13-16. Thus, based on the biochemical characterization of normal tissue, benign lesions and cancer, Raman spectroscopy may eventually provide rapid identification of thyroid lesions and more effective cellular analysis than that obtained by FNA, preventing surgical procedures that are not immediately necessary17. In addition, Raman spectroscopy can be used in clinical practice during the surgical process for in vivo and real-time diagnosis. In the present study, confocal Raman spectroscopy was used to evaluate normal and pathological thyroid tissues, such as goiter and cancer. This evaluation is very important to provide knowledge of thyroid spectra in general and to establish a database of spectra of different thyroid lesions. Raman techniques with or without coupling to an optical fiber may be useful in clinical practice for the differential diagnosis of thyroid lesions, particularly at the time of FNA, to increase the diagnostic accuracy of suspicious or indeterminate lesions.

Methods 2.1 Sample The internal review board (IRB) (221.402/CEP/2013) approved this project. Thyroid tissue samples were obtained from the Hospital das Clínicas da Faculdade de Medicina - HCFM-USP. Each sample was classified by a pathologist following a standard double-blind analysis. The samples were processed by the microdissection technique. The samples were microdissected before slicing and Raman measurements. The microdissection process increased the sensitivity of our test since there were large concentrations of cells classified as the same histological tissue type18 and cryopreserved at -80 °C. For spectral analysis, the samples were sliced with a thickness of 10 μm in a Leica cryostat (Leica CM® Model 1100, Heidelberg, Nussloch, Germany) and placed on CaF2 (calcium fluoride) slides. The data were obtained from 30 tissue samples (size of 1x1 mm) from normal thyroid (10 samples), goiter (10 samples) and thyroid cancer (10 samples). The normal tissue samples were obtained from opposite sides of the lesion to avoid the influence of cancer and goiter. Representative goiter samples were obtained from a cell area of nodules, regardless of type, since the histological pattern of nodes is based on follicular cell hyperplasia. The cancer group included papillary thyroid cancers (PTC 7 samples) and follicular thyroid cancers (FTC 3 samples), both of which are histologically classified as malignant lesions. Despite the morphological differences between these cancers, their cell characteristics allow them to be classified as a single group, cancer. Although distinguishing between these two types of cancer is also important, our initial focus in this study was only differentiating between cancer and goiter compared to normal tissue.

We used tissue sections rather than FNA material because microsections are more useful for discriminating molecular biochemical information from cancer, goiter and normal tissue. In addition, a spectral database of these lesions will be important for future in vivo Raman. 2.2 Raman spectroscopy A confocal Raman spectrometer (River Diagnostics® model 3510 - Netherlands) was used with a 785-nm diode laser as an excitation source. The power was set to 20 mW at the sample, and the integration time was 5 seconds. The laser spot was approximately 2 µm, and the Raman signal was collected with a CCD detector. The thyroid slices were placed in CaF2 slides that were inverted to face the 40X objective. The spot where the laser was focused on the tissues was marked at the opposite site of the CaF2 slides using a wax pencil to ensure that the histopathological analysis was performed at the same region. Four different regions were probed for each sample. Hematoxylineosin (HE) staining was performed for each slice after the Raman measurements. 2.3 Data analysis Background subtraction was performed by fitting the data with a polynomial curve (8 degree) using LabSpec5® software (Horiba Jobin Yvon), and the data were vector normalized through the full spectrum range (400-1800 cm-1) using OPUS® software. The average spectra of each group (normal tissue, goiter and cancer) were obtained and analyzed using the software OriginPro 8 (OriginLab), and standard deviation values were obtained using Minitab®.

2.4 Principal component analysis Principal component analysis (PCA) is a multivariate statistics method broadly used for spectral data analysis, interpretation and classification19-23. PCA allows the number of variables in large spectral datasets to be reduced while retaining most of the variation, thus permitting the identification and differentiation of different spectral groups. The spectra were decomposed into factors, or principal components. Each principal component was related to the spectrum with a variable called a score representing the weight of that particular component to form the basis spectrum. These scores were used to classify the spectra into well-defined groups. The order of the principal components (PCs) denotes their importance to the dataset. PC1 describes the highest amount of variation, PC2 the second highest, etc.24, 25. A PCA scatter plot groups similar datasets (spectra) according to the loadings of the PCs and can be used to distinguish different datasets (samples). The loadings represent the variance for each variable (wavenumber) for a given PC. Analyzing the loadings of a PC can give information about the source of the variability inside a dataset, which, in the case of spectroscopy, are derived from variations in the molecular components contributing to the spectra. The data classification for principal component analysis (PCA) was performed in Minitab®. Spectral loading plot analysis was performed to verify the most significant regions for discriminating between groups. After PCA analysis, the Mann-Whitney test was applied to determine which PC was most representative for classification. The sensitivity and specificity of the tests were calculated considering confidence intervals of 95%, and the PC values were considered significantly different between groups at p ≤ 0.05.

2.5 Linear discriminant analysis with leave-one-out cross-validation The linear discriminant analysis (LDA) used for classification and discrimination purposes between groups was performed using Minitab software. The LDA technique is a supervised analysis because the classes used already have a pre-defined identity. A sample is classified in a group based on the Mahalanobis distance to the center of the group, which is characterized by the average. For each group there is a distance known as the linear discriminant function. Thus, samples with a smaller distance for the linear discriminant function will be classified as belonging to the group. To avoid errors in the classification of samples in each group, the leave-one-out cross-validation technique was applied to the LDA. This technique is used to generate a new classification function in which the Mahalanobis distance is recalculated for each sample. One sample is always omitted, and a new classification function is calculated to determine if the data remain in the same groups classified previously. Thus, the Mahalanobis distance will be recalculated for all samples, reducing error rates. 2.6 Binary logistic regression The binary logistic regression technique (BLR) was applied to estimate the probability that a spectrum was pathologic or healthy. The response variable obtained by BLR assigned a value between 0 and 1, which can be interpreted as the probability of occurrence of a certain event. The concordant pair’s value is the probability that a sample is classified as positive for a pathological condition when it is in fact positive. The statistical indices of Somer's D and Goodman-Kruskal gamma are based on correlation tests and summarize the values obtained for the concordant and discordant pairs. Correlation values close to 0 indicate that the model is not a good predictor, whereas those close to 1 indicate that the model has high predictive power26, 27.

Results 3.1 Spectrum quality analysis Fig. 1 shows the vector-normalized average spectra for each tissue type present in this study, i.e., normal thyroid (normal), goiter and thyroid cancer. The main differences in peak intensity were located at the amide I (1506-1638 cm-1), amide III (1200-1400 cm-1), and CH2 and CH3 spectral regions (1300-1400 cm-1) assigned to proteins and lipids. Intensity differences were also observed in the DNA/RNA region (648-802 cm-1) and at 835-980 cm-1, a region with a predominance of the amino acids proline and hydroxyproline related to collagen protein. Qualitative observation of the mean and standard deviation (μ ± σ) across the spectral range (400-1800 cm-1) revealed significant intra-group variation [Fig. 2 (A-C)]. The variations were more significant in the cancer group, which can be explained by the wide variation in biochemical characteristics among cancers. However, the region with the greatest variation among all spectra was the amide III region (1200-1400 cm-1), which was therefore subsequently used to discriminate between normal and goiter tissues.

3.2 Principal component analysis After the qualitative observation, PCA was performed as an unsupervised analysis for sample discrimination as follows: normal tissue versus goiter tissues, goiter tissues versus cancer tissues and normal tissues versus cancer tissues. Four PCs (principal components) were generated for each group, and a non-parametric statistical test (Mann-Whitney) was used to establish the PC that best discriminated the samples with a value of p ≤ 0.05. The values of the Mann-Whitney test are detailed in Table 1, where each PC has its significance value.

To determine the best discrimination factor, the mean-centered data is typically plotted against variable 1 (horizontal) and variable 2 (vertical). In our case, variable 1 and variable 2 are the pair of PCs. The classification by PCA is performed using one PC against other PC, such as PC1 x PC2 and PC1 x PC3, etc. The advantage of this classification is that the Loading Plot can be used to observe the biochemical content of the sample20. The main PCs were used to create a scatterplot. For the normal group versus cancer group and for the goiter group versus cancer group, PC3 and PC4 were used; however, in the comparison between the normal group and goiter group, there was not a significant p value (p ≥ 0.05). In this case, we used the PCs with values closest to significance: PC1 and PC2. In addition to the scatterplot, the loading plot graph obtained by plotting the main PC against the discrimination among samples enabled the samples to be differentiated based on the peaks along the positive and negative axes. For the analysis, it is crucial to observe how the samples cluster in the scatter plot with 0,0 as a reference point. A black line is used as a visual guide to divide the graph into two groups (i.e., left and right). One group is located on the negative side, and the other group is located on the positive side. The subsequent analysis of the loading plot was based on the scatter plot values. The positive peaks refers to samples on the positive side (the right side of the black line, for example), and the negative peaks refer to the samples grouped on the negative side (the left side of the black line). The plots are presented in Fig. 3 (A-F). The 2-way test enables a more precise correlation of the biochemical differences in the compositions of the samples (normal, goiter or cancer) by analyzing both the scatter and loading plots of PCA. The loading plot of normal tissue versus cancer tissue shows that most of the positive peaks (720, 1084, 1304, 1440, 1592 and 1660 cm-1) are characteristic of cancer tissue, whereas the negative

peaks (816, 856, 940, 1032, 1242 and 1684 cm-1) are characteristic of normal tissue. For the goiter and cancer groups, the positive peaks (760, 816, 856, 940, 1032, 1242, 1636 and 1684 cm-1) are characteristic of goiter tissues, and the negative peaks (720, 1156, 1304, 1440, 1526 and 1660 cm-1) relate to cancer tissues. For the normal group versus goiter group in the region of 400-1800 cm-1, there were no significant results. Therefore, the region of 1200-1400 cm-1, attributed to amide III and chosen by qualitative analysis of the mean and standard deviation (μ ± σ) [Fig. 2 (A-C)], was used for a new analysis. The resultant peaks are listed in Table 2 with their respective assignments. The sensitivity was 73.3%, and the specificity was 86.6%, indicating significant discrimination between normal and cancer tissues. Regarding the differentiation between goiter and cancer, 76.6% sensitivity and 70% specificity were observed. However, no significant differentiation between normal and goiter tissues was obtained, with a sensitivity of 60% and a specificity of 67%. Given the similarity of the tissue, an analysis of the average and standard deviation graph was performed [Fig. 2 (A-C)], and the region of 1200-1400 cm-1 was selected as the most relevant variation between the two groups using PCA analysis. The most discriminating PC was PC3, with a significance value of p = 0.0001 obtained by the Mann-Whitney test. Thus, in a more specific analysis of the region of 1200-1400 cm-1, we obtained the highest values of sensitivity (80%) and specificity (60%), indicating that this region is the most accurate for differentiating among these tissues (Table 3).

In the scatterplot, it is apparent that samples of normal tissue are predominantly on the positive side of the graph, whereas goiter samples are on the negative side. From the loading plot, we noted that the peaks present on the positive side of the graph (1282, 1358, 1370 and 1382 cm -1) are characteristic mainly of the normal group, whereas the peaks present on the negative side (1224,

1248, 1266, 1312, 1322 and 1340 cm-1) are associated with the goiter group [Fig. 4 (A-B)]. The assignment of the peaks found in this analysis is shown in Table 4.

3.3 Linear discriminant analysis with leave-one-out cross-validation LDA with leave-one-out cross-validation analysis for the comparison of normal and cancer tissues in the region of 400-1800 cm-1 resulted in a discriminant value of 78.3%. In the comparison of goiter and cancer tissues in the same region (400-1800 cm-1), the percentage of discrimination was 75%, and for the comparison between normal and goiter tissues in the region of 400-1800 cm-1, we obtained a discriminant value of 65%. To improve the value of the comparison of normal and goiter tissues, we analyzed the region of 1200-1400 cm-1, and the percentage of discrimination increased to 68% (Table 5).

3.4 Binary logistic regression The results obtained by BLR analysis for the comparison of normal and cancer tissues in the region of 400-1800 cm-1 were 89.2% for concordant pairs, with Somer's D = 0.79 and Goodman-Kruskal gamma = 0.79. For the comparison of goiter and cancer tissues in the region of 400-1800 cm-1, the percentage of concordant pairs was 85.7%, with Somer's D index = 0.72 and Goodman-Kruskal gamma = 0.72. For the comparison of normal and goiter tissues in the region of 1200-1400 cm-1, the percentage of concordant pairs was 80.6%, with Somer's D = 0.61 and Goodman Kruskal gamma = 0.61 (Table 6).

Discussion This study assessed the efficiency of Raman spectroscopy for discriminating between normal, benign and malignant thyroid tissues. Changes were evident in important biochemical components related to the presence of cancer, such as increased amounts of DNA, RNA, protein, amino acids, such as proline, hydroxyproline, tryptophan, and phenylalanine, and lipids. Fig. 1 shows the changes in the amide I (~1506 to 1638 cm-1)31,32 and amide III (~1200-1400 cm-1)33 regions, which are expected in tumors due to intense proliferation. In these regions, the normal bending modes of the CH2 and CH3 regions (~1300-1400 cm-1) are present. The increase in the DNA/RNA bands (~648-802 cm-1)34 is explained by the large amount of nuclear material due to the increase in mitosis. Such increases in intensity may be related to the increase in protein and its components in thyroid lesions (mainly carcinomas). Similar results have been reported for a comparison between tumor and normal lung tissues35. Increased intensities of the bands of the amino acids tryptophan and phenylalanine beyond the peaks at 1322 and 1335 cm-1 result in characteristic modes of angular change δ(HCH) in -CH3 and -CH2 in tumors, and these modifications are correlated with structural changes in proteins associated with tumor transformation. Another important cause of the increase in the intensity of the peaks related to protein is enhanced translation of mRNA, reflecting intense cell proliferation. Although the lipid content is decreased in carcinomas, likely due to lipid chain instability, the intensification of lipids observed in some regions in our study (1664, 1368, 1303, and 1271 cm-1) is attributable to the essential role of lipids in hormone production. Changes in the region of 1300-1320 cm-1, related to the presence of lipids36, were observed in our study, suggesting elevated activity in this region for all lesions, particularly for carcinomas, compared with normal tissue.

Another important spectral region is 835-980 cm-1, which contains important peaks for the differentiation of normal tissue. In this region, the peaks 856 and 940 cm -1, assigned to the constituent amino acids of collagen protein37, were increased in normal tissue, indicating a decrease in their intensity in goiter tissue and cancer. Significant variation was observed between the spectra, indicating greater differentiation among tissues [Fig. 2 (A-C)]. The main changes in the spectra were observed in the cancer group. Variations within the same group of samples can be explained by the large biochemical variability within the same type of cancer associated with the physiological characteristics of each individual. The region of 1200-1400 cm-1, attributed to amide III, was the most variable across the spectrum (400-1800 cm-1), indicating its relevance in tissue differentiation. The Mann-Whitney test was important for determining which PC correctly differentiated the tissues (Table 1), enabling improved selection of PCs to avoid false results. The scatterplot and loading plot comparing normal x cancer tissues showed that the most representative peaks of benign tissues were 856 and 940 cm-1, characteristic peaks reported for normal tissue that are increased in other pathologies37, and the peak at 1684 cm-1, attributed to amide I and present only in normal samples in our study. The main peaks in the cancer samples were 1440 cm-1, attributable mainly to lipids38 and present only in pathological samples with a predominance of cancer, the peak at 1084 cm-1, attributable to DNA/RNA, which is known to be elevated in cancer, and the peak at 1304 cm-1, attributable to proteins that are increased due to increased cell proliferation in cancer35. Regarding the goiter and cancer groups, the main peaks related to goiter were 816 and 1636 cm -1, attributed to collagen, and peak 1032 cm-1, attributed to the amino acid phenylalanine and collagen, which was present in large quantities in the goiter tissues in the present study. The differences between normal and goiter tissues were also consistent, as the normal samples were mostly on the

negative side of the graph and exhibited peaks at 856 and 940 cm-1, both characteristic of normal tissue37. Moreover, the peaks for the goiter samples occurred at 1004 and 1032 cm-1 and are attributable to the amino acid phenylalanine, which is present in some cases of goiter and suggests a higher amount of protein39. The sensitivity and specificity for the cancer and normal groups were greater than 70%, demonstrating the capacity to correctly classify normal and pathological samples. However, for the comparison of normal and goiter samples, sensitivity and specificity were insufficient. One possible explanation is that the normal tissues used in this study were obtained from adjacent areas containing cancer or goiter of the thyroid gland and therefore also had undergone similar biochemical changes. To refine the discrimination among these tissues, a reanalysis of the normal versus goiter groups in the region of 1200-1400 cm-1 was performed. Furthermore, compatibility was observed when the scatterplot and loading plot information were crossed [Fig. 4 (A-B)]. Peaks were observed in normal samples at 1282 cm-1, attributable to collagen, and 1358 cm-1, attributable to the amino acid tryptophan. The main peaks were attributed to collagen (1322 cm-1) and nucleic acids (1340 cm-1), which may be explained by the increase in cell proliferation in goiter compared to normal tissue (Table 4). The vibrational modes in the 1200-1400 cm-1 region are mainly related to collagen information; thus, we can conclude that collagen spectral regions can be used to discriminate goiter and normal samples. It is difficult to analyze the full spectrum because overlap with other biological components might affect the classification. Goiter does not exhibit malignant features, and the tissue growth follows a sequential pattern. Consequently, it is not trivial to discriminate this type of sample using the full spectrum. A specific

spectral window might therefore provide greater discrimination as a “spectral biomarker” of such lesions in future studies. The BLR results confirmed those observed by the confusion matrix, resulting in a good sample classification index (Table 6). According to the values of discordant and concordant pairs that predict the likelihood of our model, the present study obtained significant values of greater than 80% for all groups. These values, together with the statistical indices of Somer's D and Goodman-Kruskal gamma, support the effectiveness of our data. The discrimination values obtained in this study are higher than those obtained in a previous study conducted by our group31 employing Fourier transform Raman spectroscopy (FT-Raman). In that previous study, the discriminant values of the normal group and goiter were only 58.3% and 64.9% compared to the goiter tissue and PTC tissue groups, respectively, and 72.5% in normal tissue compared to the tumor tissue groups. The confocal Raman technique permits rapid analysis of tissue, and an increased sensitivity of Raman spectroscopy with respect to thyroid tissues was observed. The FNA technique is important for the diagnosis of thyroid lesions and may be used in combination with other techniques, such as Raman spectroscopy, to improve the diagnostic results40. The differentiation of the types of thyroid lesions by FNA varies according to the histological pattern41. In our study, the positive predictive value obtained by comparison of cancer and normal tissues was 82.6%, and the specificity value was 86.6%. These values are similar to the values obtained by the combination of FNA with molecular tests. The difference between the discriminant values of conventional techniques and our optical techniques is mainly related to the reading protocol of the pathologist. In conventional techniques, the sample is read as a whole, whereas Raman spectroscopy is a punctate technique. Therefore, the lack of coincidence between the regions analyzed by Raman spectroscopy and the microdissected samples read by the pathologist may lead to divergent results.

The routine gold standard technique is histology, which requires a surgical procedure, such as biopsy. The aim of Raman spectroscopy is to improve thyroid lesion diagnosis in real time, simultaneously with FNA, to obviate surgical procedures that are not immediately required. Knowledge of molecular features is very important. Histopathology is currently necessary to obtain this information, but in the near future, spectral “biomarkers” of thyroid pathology may be employed in FNA procedures. An important factor for consideration is the faster speed of diagnosis by Raman spectroscopy compared with that of the currently used technique. Furthermore, the application of the optical technique in real time in the near future will provide precise biochemical information about the sample, ensuring greater accuracy in the diagnosis of thyroid lesions. Other studies should be performed to improve the sensitivity of the technique and to evaluate additional pathologies involving the thyroid gland because our results indicate that confocal Raman spectroscopy can be a useful tool in the diagnosis of thyroid diseases. These results provide an important basis for further steps in improving the diagnosis of thyroid cancer by Raman spectroscopy.

Conclusion The results presented here demonstrate that Raman spectroscopy can differentiate thyroid tissues based on the biochemical modifications present. Binary logistic regression analysis based on PCA data demonstrated that the accuracy in the comparisons was significant, with good predictive value and correctly classified spectra. The spectral region of 1200-1400 cm-1 can discriminate normal versus goiter tissues despite their great similarity.

First autor is a doctor’s student in Biomedical Engineering. He received his master’s degree in Biomedical Engineering at the Universidade do Vale do Paraíba - Brazil in 2014, bachelor's degree in Biomedicine and specialization in Public Health at the Centro Universitário do Sul de Minas Gerais - Brazil in 2008 and 2010, respectively. His current research interests include Raman spectroscopy in detecting carcinomas.

Acknowledgements This research was supported in part by Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq (307809/2013-7) and co-funded by FINEP (01.10.0661.0). Luis Felipe CS Carvalho was funded Fundação de Amparo a Pesquisa do Estado de São Paulo - FAPESP for post doctoral research (2014/05978-1) and Laurita dos Santos was funded by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – CAPES for post doctoral research (88881.062862/2014-01).

References 1. Instituto Nacional do Câncer – INCA, Estimativa 2014 - Incidência de câncer no Brasil, 1th ed., INCA, Rio de Janeiro (2014). 2. L. G. T. Morris, D. Myssiorek, “Improved detection does not fully explain the rising incidence of welldifferentiated thyroid cancer: a population-based analysis,” Am. J. Surg. 200, 454-461 (2010) [doi:10.1016/j.amjsurg.2009.11.008]. 3. J. D. Cramer, P. Fu, K. C. Harth, S. Margevicius, S. M. Wilhelm, “Analysis of the rising incidence of thyroid cancer using the surveillance, epidemiology and end results national cancer data registry,” Surgery 148, 1147-1153 (2010) [doi: 10.1016/j.surg.2010.10.016]. 4. J. A. Sipos, E. L. Mazzaferri, “Thyroid cancer epidemiology and prognostic variables,” Clin. Oncol. 22, 395-404 (2010) [doi: 10.1016/j.clon.2010.05.004]. 5. E. M. Gomes, F. Vaisman, A. P. Vidal, R. Corbo, M. D. Cruz, P. de F. Teixeira, A. Buescu, M. Vaisman, “Frequency of thyroid carcinoma and thyroid autoimmunity in first degree relatives of patients with papillary thyroid carcinoma - A single center experience,” Arq. Bras. Endocrinol. Metabol. 55, 326-330 (2011) [doi: http://dx.doi.org/10.1590/S0004-27302011000500005]. 6. H. G. C. Rodrigues, A. B. N. Pontes, L. F. F. Adan, “Use of molecular markers in samples obtained from preoperative










http://doi.org/10.1507/endocrj.EJ11-0410]. 7.

J. D. Lin, “Thyroid cancer in thyroid nodules diagnosed using ultrasonography and fine needle aspiration cytology,” J. Med. Ultrasound 18, 91-104 (2010) [doi: http://dx.doi.org/10.1016/S09296441(10)60014-8].

8. C. Ravetto, L. Colombo, M. E. Dottorini, “Usefulness of fine-needle aspiration in the diagnosis of thyroid carcinoma: a retrospective study in 37895 patients,” Cancer 90, 357-363 (2000) [PMID: 11156519].

9. A. A. Renshaw, “Accuracy of thyroid fine-needle aspiration using receiver operator characteristic curves,” Am. J. Clin. Pathol. 116, 477-482 (2001) [doi: http://dx.doi.org/10.1309/M3K5-23C2-455E0HB5]. 10. J. A. Blansfield, M. J. Sack, J. S. Kukora, “Recent experience with preoperative fine-needle aspiration biopsy of thyroid nodules in a community hospital,” Arch. Surg. 137, 818-821 (2002) [PMID: 12093339]. 11. M. R. Castro, H. Gharib, “Thyroid fine-needle aspiration biopsy: progress, practice, and pitfalls,” Endocr. Pract. 9, 128-136 (2003) [doi: http://dx.doi.org/10.4158/EP.9.2.128]. 12. Y. Y. Tee, A. J. Lowe, C. A. Brand, R. T. Judson, “Fine-needle aspiration may miss a third of all malignancy in palpable thyroid nodules: a comprehensive literature review,” Ann. Surg. 246, 714-720 (2007) [PMID: 17968160]. 13. A. T. Harris, A. Rennie, H. Waqar-Uddin, S. R. Wheatley, S. K. Ghosh, D. P. Martin-Hirsch, S. E. Fisher, A. S. High, J. Kirkham, T. Upile, “Raman spectroscopy in head and neck cancer,” Head Neck Oncol. 2, 2-6 (2010) [doi: 10.1186/1758-3284-2-26]. 14. A. Saha, I. Barman, N. C. Dingari, S. McGee, Z. Volynskaya, L. H. Galindo, W. Liu, D. Plecha, N. Klein, R. R. Dasari, M. Fitzmaurice, “Raman spectroscopy: a real-time tool for identifying microcalcifications during stereotactic breast core needle biopsies,” Biomed. Opt. Express 2, 2792-2803 (2011) [doi: 10.1364/BOE.2.002792]. 15. D. G. Leslie, R. E. Kast, M. J. Poulik, R. Rabah, S. Sood, G. W. Auner, M. D. Klein, “Identification of pediatric brain neoplasms using Raman spectroscopy,” Pediatr. Neurosurg. 48, 109-117 (2012) [doi: 10.1159/000343285]. 16. M. Diem, A. Mazur, K. Lenau, J. Schubert, B. Bird, M. Miljković, C. Krafft, J. Popp, “Molecular pathology via IR and Raman spectral imaging,” J. Biophotonics 6, 855-886 (2013) [doi: 10.1002/jbio.201300131].

17. G. Riesco-Eizaguirre, P. Santisteban, “New insights in thyroid follicular cell biology and its impact in thyroid cancer therapy,” Endocr. Relat. Cancer 14, 957-77 (2007) [doi: 10.1677/ERC-07-0085]. 18. S. Hetz, A. Acikgoez, C. Moll, H. G. Jahnke, A. A. Robitzki, R. Metzger, M. Metzger, “Age-related gene expression analysis in enteric ganglia of human colon after laser microdissection,” Front Aging Neurosci. 6, 1-12 (2014) [doi: 10.3389/fnagi.2014.00276]. 19. L. F. C. S. Carvalho, E. T. Sato, J. D. Almeida, H. S. Martinho, “Raman micro-spectroscopy for rapid screening of oral squamous cell carcinoma,” Exp. Mol. Pathol. 98, 502-509 (2015) [doi: 10.1016/j.yexmp.2015.03.027]. 20. F. Bonnier, H. J. Byrne, “Understanding the molecular information contained in principal component analysis of vibrational spectra of biological systems,” Analyst 137, 322-332 (2012) [doi: 10.1039/c1an15821j]. 21. Taylor & Francis Group, Introduction to multivariate statistical analysis in chemometrics, CRC Press, New York (2009). 22. T. Korenius, J. Laurikkala, M. Juhola, “On principal component analysis, cosine and Euclidean measures in information retrieval,” Inf. Sci. 177, 4893-4905 (2007) [doi:10.1016/j.ins.2007.05.027]. 23. M. J. German, A. Hammiche, N. Ragavan, M. J. Tobin, L. J. Cooper, S. S. Matanhelia, A. C. Hindley, C. M. Nicholson, N. J. Fullwood, H. M. Pollock, F. L. Martin, “Infrared spectroscopy with multivariate analysis potentially facilitates the segregation of different types of prostate cell,” Biophys. J. 90, 37833795 (2006) [doi: 10.1529/biophysj.105.077255]. 24. J. G. Kelly, J. Trevisan, A. D. Scott, P. L. Carmichael, H. M. Pollock, P. L. Martin-Hirsch, F. L. Martin, “Biospectroscopy to metabolically profile biomolecular structure: a multistage approach linking computational analysis with biomarkers,” 10.1021/pr101067u].

J. Proteome Res. 10, 1437-1448 (2011) [doi:

25. F. J. Martin, M. J. German, E. Wit, T. Fearn, N. Ragavan, H. M. Pollock, “Identifying variables responsible for clustering in discriminant analysis of data from infrared microspectroscopy of a biological sample,” J. Comput. Biol. 14, 1176-1184 (2007) [doi:10.1089/cmb.2007.0057]. 26. V. Bewick, L. Cheek, J. Ball, “Statistics review 14: logistic regression,” Crit. Care 9, 112-118 (2005) [doi: 10.1186/cc3045]. 27. L. F. C. S. Carvalho, E. T. Sato, J. D. Almeida, H. S. Martinho, “Diagnosis of inflammatory lesions by high-wavenumber FT-Raman spectroscopy,” Theor. Chem. Acc. 130, 1221-1229 (2011) [doi: 10.1007/s00214-011-0972-2]. 28. Z. Movasaghi, S. Rehman, I. U. Rehman, “Fourier Transform Infrared (FTIR) Spectroscopy of Biological Tissues,” Appl. Spectrosc. Reviews 43, 34-179 (2008) [doi:10.1080/05704920701829043]. 29. N. Stone, C. Kendall, J. Smith, P. Crow, H. Barr, “Raman spectroscopy for identification of epithelial cancers,” Faraday Discuss 126, 141-157 (2004) [PMID: 14992404]. 30. G. Shetty, C. Kendall, N. Shepherd, N. Stone, H. Barr, “Raman spectroscopy: Elucidation of biochemical changes












10.1038/sj.bjc.6603102]. 31. C. S. Teixeira, R. A. Bitar, H. S. Martinho, A. B. Santos, M. A. Kulcsar, C. U. Friguglietti, R. B. da Costa, E. A. Arisawa, A. A. Martin, “Thyroid tissue analysis through Raman spectroscopy,” Analyst 134, 2361-2370 (2009) [doi: 10.1039/b822578h]. 32. S. D. Ruchita, Y. K. Agrawal, “Raman spectroscopy: recent advancements, techniques and applications,” Vib. Spec. 57, 163-176 (2011) [doi:10.1016/j.vibspec.2011.08.003]. 33. K. Z. Liu, , C. P. Schultzb, E. A. Salamonc, A. Mana, H. H. Mantscha, “Infrared spectroscopy diagnosis of thyroid tumors,” J. Mol. Struct. 661-662, 397-404 (2003) [doi:10.1016/j.molstruc.2003.07.021]. 34. P. R. T. Jess, D. D. W. Smith, M. Mazilu, K. Dholakia, A. C. Riches, C. S. Herrington, “Early detection of cervical neoplasia by Raman spectroscopy,” Int. J. Cancer 121, 2723-2728 (2007) [doi: 10.1002/ijc.23046].

35. Z. Huang, A. McWilliams, H. Lui, D. I. McLean, S. Lam, H. Zeng, “Near-infrared Raman spectroscopy for optical diagnosis of lung cancer,” Int. J. Cancer 107, 1047-1052 (2003) [doi: 10.1002/ijc.11500]. 36. N. Stone, P. Stavroulaki, V. Kendall, M. Birchall, H. Barr, “Raman spectroscopy for early detection of laryngeal








10.1097/00005537-200010000-00037]. 37. A. T. Harris, M. Garg, X. B. Yang, S. E. Fisher, J. Kirkham, D. A. Smith, D. P. Martin-Hirsch, A. S. High, “Raman spectroscopy and advanced mathematical modelling in the discrimination of human thyroid cell lines,” Head Neck Oncol. 1 (2009) [doi: 10.1186/1758-3284-1-38]. 38. L. E. Kamemoto, A. K. Misra, S. K. Sharma, M. T. Goodman, H. Luk, A. C. Dykes, T. Acosta, “NearInfrared Micro-Raman Spectroscopy for in Vitro Detection of Cervical Cancer,” Appl. Spectrosc. 64, 255-261 (2010) [doi: 10.1366/000370210790918364]. 39. C. J. Frank, R. L. McCreery, D. C. B. Redd, “Raman spectroscopy of normal and diseases human breast tissues,” Anal. Chem. 67, 777-783 (1995) [doi: 10.1021/ac00101a001]. 40. P. Pujary, K. Maheedhar, C. M. Krishna, K. Pujary, “Raman spectroscopic methods for classification of normal and malignant hypopharyngeal tissues: an exploratory study,” Patholog. Res. Int. 2011 (2011) [doi: 10.4061/2011/632493]. 41. M. Becker-Putsche, T. Bocklitz, J. Clement, P. Rösch, J. Popp, “Toward improving fine needle aspiration cytology by applying Raman microspectroscopy,” J. Biomed. Opt. 18, 047001-047007 (2013) [doi: 10.1117/1.JBO.18.4.047001]. 42. K. Alexander, G. C. Kennedy, Z. W. Baloch, E. S. Cibas, D. Chudova, J. Diggans, L. Friedman, R. T. Kloos, V. A. LiVolsi, S. J. Mandel, S. S. Raab, J. Rosai, D. L. Steward, P. S. Walsh, J. I. Wilde, M. A. Zeiger, R. B. Lanman, B. R. Haugen, “Preoperative diagnosis of benign thyroid nodules with indeterminate cytology,” N. Engl. J. Med. 367, 705-715 (2012) [doi: 10.1056/NEJMoa1203208]. 43. V. A. LiVolsi, “Papillary thyroid carcinoma: an update,” Mod. Pathol. 24, S1-S9 (2011) [doi: 10.1038/modpathol.2010.129].

44. M. Sobrinho-Simoes, C. Catarina Eloy, J. Magalhães, C. Lobo, T. Amaro, “Follicular thyroid carcinoma,” Mod. Pathol. 24, S10-S18 (2011) [doi:10.1038/modpathol.2010.133]. 45. A. C. Meier, “Thyroid nodules: pathogenesis, diagnosis and treatment,” Baillieres Best Pract. Res. Clin. Endocrinol. Metab. 14, 559-75 (2000) [doi:10.1053/beem.2000.0103]. 46. D. Chudova, J. I. Wilde, E. T. Wang, H. Wang, N. Rabbee, C. M. Egidio, J. Reynolds, E. Tom, M. Pagan, C. T. Rigl, L. Friedman, C. C. Wang, R. B. Lanman, M. Zeiger, E. Kebebew, J. Rosai, G. Fellegara, V. A. LiVolsi, G. C. Kennedy, “Molecular classification of thyroid nodules using highdimensionality genomic data,” J. Clin. Endocrinol. Metab. 95, 5296-5304 (2010) [doi: 10.1210/jc.20101087].

Caption List

Fig. 1 Spectral averages for normal tissues, goiter and cancer with the main variant regions. The specific regions of greatest variability among the spectra include 648-802, 835-980, 1200-1400 and 1506-1638 cm-1.

Fig. 2 (A-C) Mean and standard deviation (μ ± σ) spectra for the studied tissues. A: normal tissue; B: goiter tissue; C: cancer tissue. The demarcated region exhibited the most variation among the groups (1200-1400 cm-1).

Fig. 3 (A-F). Scatterplot and loading plot graph for the normal, goiter and cancer tissues in the 400-1800 cm-1 region.

Fig. 4 (A-B). Scatterplot (A) and loading plot (B) for the normal x goiter tissue comparison in the region of 1200-1400 cm-1.

Table 1 Significance values (p ≤ 0.05) for the main PCs. PCs

Normal x Cancer

Normal x Goiter

Goiter x Cancer

















PC: Principal component

Table 2 Main peaks of thyroid tissues (normal, goiter and cancer). Peaks (cm-1)



Nucleotides and DNA


C-C stretching (collagen assignment)


Peaks of collagen protein


Proline, hydroxyproline


Proline, hydroxyproline, ν(C-C) skeletal of collagen backbone


νs(C-C), symmetric assignment)


δ(HCH) (CH3) twist, δ(HCH)(CH2) twist modes of collagen & phospholipids, phenylalanine


Phosphodiester groups in nucleic acids


ν(C-N) stretching (proteins), ν(C-O) stretching (carbohydrates)


ν(C-C), ν(C-N) stretching (protein), chain ν(C-C) stretching (lipids)


Amide III, CH2 wagging vibrations from glycine backbone and proline side chains, Collagen


Amide III of collagen [CH2 wagging, ν(C-N) stretching] and pyrimidine bases


δ(HCH) (CH3) twist, δ(HCH)(CH2) twist (collagen assignment)


(CH3,CH2) twisting mode of collagen/lipid


δ(CH2), δ(CH3) deformation, δ(C-H) deformation, cholesterol, fatty acid band, δ(CH2) (lipids), δ(CH2) bending (lipids)


Guanine, adenine (ring breathing modes in the DNA bases)


In-plane vibrations of conjugated (C=C)




δ(C=C), phenylalanine


Amide I band, differences in collagen content


Amide I (α-helix)


Amide I, amide I (collagen), amide I (protein), lipids, DNA


Amide I





ν: stretching vibrations; s: symmetric; a: asymmetric; δ: bending. Wavenumber with a confidence interval of ± 2 cm -1. Assignment of peaks based on Ref. 28. *This wavenumber was also assigned as C-C backbone stretching at 938 cm-1 by Ref. 29 and as skeletal modes (polysaccharides, amylose) by Ref. 30.

Table 3 Sensitivity and specificity values calculated from the main PCs. Groups

Region (cm-1)



Normal x Cancer

400-1800 cm-1



Goiter x Cancer

400-1800 cm-1



Normal x Goiter

1200-1400 cm-1



Table 4 Vibrational modes in the region of 1200-1400 cm-1 for normal x goiter tissues. Peaks (cm-1)



Amide III (β sheet structure)


Amide III (collagen)


Amide III (proteins in the α-helix conformation), ν(C-N), δ(N-H) amide III, α-helix, collagen


Collagen, Nucleic acids and phosphates


(CH3,CH2) twisting mode of collagen/lipid


δ(CH3) δ(CH2) twisting and wagging in collagen


Nucleic acid mode, collagen, guanine deformation (protein and carbohydrates)




Saccharide band


CH3 band



ν: stretching vibrations; s: symmetric; a: asymmetric; δ: bending. Wavenumber with a confidence interval of ± 2 cm-1. Assignment of peaks based on Ref. 20.

Table 5 Results of linear discriminant analysis with leave-one-out cross-validation. Groups

Region (cm-1)

Discrimination %

Normal x Cancer



Goiter x Cancer



Normal x Goiter



Normal x Goiter



Table 6 Results of the binary logistic regression model. Groups

Regions (cm-1)

% Concordant



Normal x Cancer





Goiter x Cancer





Normal x Goiter





SD: Somer’s D; GKG: Goodman Kruskal gamma.