Raman spectroscopy studies for diagnosis of cancers in human uterine cervix

Raman spectroscopy studies for diagnosis of cancers in human uterine cervix

Vibrational Spectroscopy 41 (2006) 136–141 www.elsevier.com/locate/vibspec Raman spectroscopy studies for diagnosis of cancers in human uterine cervi...

293KB Sizes 0 Downloads 16 Views

Vibrational Spectroscopy 41 (2006) 136–141 www.elsevier.com/locate/vibspec

Raman spectroscopy studies for diagnosis of cancers in human uterine cervix C. Murali Krishna a,*, N.B. Prathima a, R. Malini a, B.M. Vadhiraja b, Rani A. Bhatt c, Donald J. Fernandes b, Pralhad Kushtagi c, M.S. Vidyasagar b, V.B. Kartha a a Centre for Laser Spectroscopy, Manipal Academy of Higher Education, Manipal 576104, Karnataka, India Department of Radiation Oncology, Shirdi Sai Baba Cancer Hospital, Manipal Academy of Higher Education, Manipal 576104, Karnataka, India c Department of Obstetrics and Gynecology, Kasturba Medical College, Manipal Academy of Higher Education, Manipal 576104, Karnataka, India b

Received 1 August 2005; received in revised form 11 January 2006; accepted 31 January 2006 Available online 17 April 2006

Abstract Cancer of uterine cervix is one of the leading cancers among women in both developed and developing countries. Optical spectroscopy methods mostly Fourier transform infrared (FTIR) and fluorescence have widely been used to diagnose cervix cancer using cells as well as tissues. Raman spectra of normal and malignant tissues were recorded in fingerprint region. Normal cervix tissues are characterized by strong, broad amide I, broader amide III and strong peaks at 853 and 938 cm1 which can be attributed to structural proteins such as collagen. Prominent features of malignant tissue spectra with respect to normal tissue are—relatively weaker and sharper amide I, minor red shift in DCH2 and sharper amide III which indicate the presence of Deoxyribonucleic acid (DNA), lipids and non-collagenous proteins. In order to develop highly objective discrimination methods, very elaborate data analysis was carried out using Principal Components Analysis (PCA). Standard sets for normal and malignant were prepared and tested retrospectively and prospectively. Several parameters such as (scores of factor, Mahalanobis distance and spectral residuals) were explored for discrimination and very clean clustering of normal and malignant spectra was achieved. A multiparametric approach (limit test) combining all the above discrimination parameters was also considered, in order to develop unambiguous discrimination. This analysis has produced very high, 99.5%, sensitivity and specificity. Results obtained in this study thus validate Raman spectroscopy methods for discrimination of normal and malignant tissues in cervical cancers. # 2006 Elsevier B.V. All rights reserved. Keywords: Cervix cancer; Raman spectroscopy; PCA; Cancer diagnosis

1. Introduction Cervical cancer is the second most common cancer in women worldwide and the leading cause of cancer mortality in women. It accounts for 25% of new cases and 23% of deaths due to cancer among Indian women [1]. An estimated 370,000 new cases are diagnosed yearly. The existing screening and detection techniques have several deficiencies that preclude the effective and efficient management of cancer. The primary screening tool is Pap smear, which involves examination of exfoliated cells under a microscope for abnormal cells.

* Corresponding author. Tel.: +91 820 257 1201x22526/22596; fax: +91 820 257 1931/0061/0062/0063. E-mail addresses: [email protected], [email protected] (C.M. Krishna). 0924-2031/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.vibspec.2006.01.011

Colposcopy is another widely used tool, which usually follows an abnormal Pap smear. The incidence and mortality of cervical cancer has been shown to be on decline in developed countries where these screening methods are widely practiced. However, these methods have several shortcomings such as high false negative/positive results that could be due to the subjective interpretations of the pathologist who diagnoses the disease based on morphological abnormalities [2,3]. Cervical malignancy is preceded by Cervical intraepithelial neoplasia (CIN) I, CIN II, CIN III and carcinoma in situ before becoming invasive. Radiotherapy, surgery and chemotherapy, alone or in combination, are the accepted modalities of treatment for cervical malignancy. It is estimated that 90% of deaths can be prevented on early diagnosis, which emphasizes the need for effective screening methods. Tissue optical spectroscopy methods such as Raman, FTIR and fluorescence, have been shown as potential alternatives to

C.M. Krishna et al. / Vibrational Spectroscopy 41 (2006) 136–141

detect neoplastic changes [4–26]. These techniques are rapid, objective and can be developed for in vivo screening of the disease, thus avoiding the need of a painful biopsy. These noninvasive techniques can then be exploited for population screening, early diagnosis, prognosis, monitoring of therapy and subsequent follow-up. One of the main attractions of these methods is that they provide information about the biochemical, structural and patho-physiological changes occurring in the tissues. Cervical cancer has been one of the well-studied forms of malignancy by optical spectroscopic methods, mostly by FTIR [8–16], fluorescence [17–27] and a combination of reflectance and fluorescence spectroscopy [28]. Raman spectroscopy offers certain distinct advantages over these techniques which include high spatial resolution (down to 1 mm), use of less harmful NIR radiation, requires less sample preparation, not influenced by water bands and can be used for in vivo/in situ measurements. Despite the above-mentioned advantages, very few studies using Raman spectroscopy for diagnosis of cervical cancer are reported in the literature [4–7]. In an earlier study [6], two different methods of algorithms were developed for tissue differentiation to explore the feasibility of using Raman spectroscopy for detecting cervical precancers. The first method used empirically selected peak intensities and ratios of peak intensities to differentiate precancers from other tissue categories. The second method employed multivariate statistical method (based on Principal Component Analysis and Fisher Discriminant Analysis) to differentiate precancers from other tissues. In a more recent study by the same group [7], peaks and peak ratios (corresponding to the most significant differences between tissue types) were identified by a student’s t-test. In our recent studies, PCA combined with multiparametric limit tests was used for discriminating different pathological conditions in oral cancer [31] and the same approach has been extended to cervix cancer in the present study. In the case of biomedical applications, such as diagnosis, it is necessary that several groups confirm the findings to establish the method for routine use. It is also necessary that studies be carried out on different populations. In view of this, Raman spectroscopy investigations were carried out on Indian population. A very elaborate data analysis was carried out using PCA. Different discrimination parameters such as scores of factors, Mahalanobis distance and spectral residuals were explored for discrimination. A multiparametric approach (limit test) was also considered in order to develop more objective and unambiguous discrimination. The results obtained in the study are discussed in the paper.

137

Table 1 Sample details Spectrum no. 1–79 80–149

Nature

Histopathology

Malignant

Squamous cell carcinoma. FIGO – STAGE III Normal

Normal

both, cervix and uterus were employed in the study. At least six spectra were recorded on each sample at different locations and each spectral point was treated as a sample. In each case, immediate sections (mirror image) of the tissue samples were sent for pathological verification. Only certified normal and malignant tissues were employed in the study. A total of 70 normal and 80 malignant spectra were employed in data analysis. Details of samples used in the study are shown in Table 1. 2.1. Laser Raman spectroscopy Raman spectra were recorded with a set up assembled by us. This set up consists of a diode laser; SDL-8530 (785 nm, 100 mW) for excitation and combination of HR 320 spectrograph (600 g/mm blazed at 900 nm) and Spectrum One liquid N2 cooled CCD was employed for detection of scattering. A holographic filter (Kaiser Optics) was used to filter the excitation source. A notch filter (HSPF-5812, Kaiser Optics) was used to reject the Raleigh scattering. At least six spectra were recorded on each sample at different sites. An integration time of 30 s and 20 accumulations were the parameters used for spectral recording. These conditions were kept constant for all measurements. The tissue samples were kept moist with saline during spectral acquisition. A schematic of Raman setup used in the present study is shown in Fig. 1.The recorded spectra were calibrated with a cubic order fit to known frequencies of Tylenol (4-acetamidophenol) using diode adjust algorithms in GRAMS 32 (Galactic Industries Corporation, USA). 2.2. Data analysis Spectral corrections, baseline, smoothening, calibration and normalization (DCH2 peak) of spectra were carried out using algorithms of Grams 32. Baseline subtraction was performed by

2. Materials and methods Specimens (biopsies or surgical resections) used in the present investigation were collected from Department of Radiation Oncology, Shirdi Sai Baba Cancer Hospital, MAHE, Manipal and Dept. of OBG, Kasturba Medical College, MAHE, Manipal. 80 biopsy sites from malignant subjects considered for radiotherapy and 70 control sites obtained in saline from regular hysterectomy, with no indications of malignancy in

Fig. 1. Schematic of Laser Raman Instrumentation.

138

C.M. Krishna et al. / Vibrational Spectroscopy 41 (2006) 136–141

fitting a third order polynomial and the spectra were vector normalized to the highest peak (DCH2) peak. PCA was carried out under different conditions: entire spectrum (800– 1800 cm1) and selected regions using Grams PLS plus/IQ (Galactic Industries Corporation, USA). PCA was standardized by employing 15, 10 and 8 factors. Total % Variance, Eigen values and factor profiles were employed for standardizing the number of factors for PCA and in our analysis, entire spectrum with 8 factors gave best results and further data analysis was carried out under these conditions. Parameters such as Scores of factors, Mahalanobis distance, spectral residuals and matchmismatch tables computed using limit tests were employed in classification. PCA for standard sets were also carried out under similar conditions. Mahalanobis distance, spectral residuals and match-mismatch tables were computed by comparing each spectrum against standard sets of spectra corresponding to normal and malignant tissues. In the present analysis, randomly selected 32 normal and 37 malignant spectra were used for the construction of standard sets. Discrimination was carried out both retrospectively (by rotating out constituent spectra of standard sets) and prospectively using all other spectra. 3. Results and discussion Cancers of cervix are one of the well-studied malignant conditions by optical spectroscopy, mostly by FTIR and fluorescence as can be seen from the literature [8–27]. Raman spectroscopy studies of cervical cancer tissues is characterized by a decrease in collagen/elastin levels and relative increase in lipids, nucleic acids and carbohydrates as reported in the literature [4–7]. Typical Raman spectra of normal and malignant tissue samples, indicated by dark and broken line, respectively, are shown in Fig. 2A. Very prominent differences in spectral profiles for both conditions are observed. As can be seen from the figure, the normal spectrum is characterized by a broader amide I, and peaks at 1384, 1269, 939 and 854 cm1, which are missing in malignant spectra. Further, malignant spectra show a sharper amide I, minor red shift in 1450 cm1 and sharper amide III band. The differences can be more clearly illustrated by difference spectrum which was computed by subtracting mean malignant spectrum from mean normal spectrum shown in Fig. 2B. The strong positive peaks seen in the difference spectrum (Fig. 2B), 810, 854, 939, 1165, 1196, 1245, 1267, 1385, 1633 and 1671 cm1, are contributed by the normal spectrum. These spectral bands based on peak positions and relative intensities can be tentatively assigned to structural proteins such as collagen and elastin using available literature data [29,30]. The negative peaks, 1086, 1123, 1305, 1331, 1440 and 1654 cm1 are contributed by malignant spectrum and these bands can be assigned to biomolecules such as lipids, DNA and non-collagenous proteins [28,29]. In the present investigation, macro or conventional Raman system is used for obtaining the Raman signal. In this approach, it is not possible to carry out section-wise studies (epithelial or sub epithelial sections). The specimen is mounted such that the epithelial sections face the probing laser beam. In this mode, Raman

Fig. 2. (A) Mean Raman spectra of normal (dotted line) and malignant (thin line) cervical tissues. (B) Difference spectrum of normal and malignant tissue. (C) Mean and standard deviation spectra of cervix tissue. (i) Normal; (ii) malignant (solid line, mean spectra; dashed line, mean  standard deviation).

spectrum can be assumed to originate mostly from the surface layers [31,32]. Normal cervical epithelial tissue consists of squamous/columnar cell layer at the top exposing a continuous bilayer membrane surface. This is followed by the epithelial layer consisting of the normal cells for about 100–200 mm, at the base of which we have few cells in the various stages of cell division. Below this comes the basal membrane made up of Type IV collagen, followed by connective tissue consisting of mostly Type II collagen, with fibroblasts, capillaries and others.

C.M. Krishna et al. / Vibrational Spectroscopy 41 (2006) 136–141

In Raman macrospectroscopy with near infrared laser, the laser beam can, and do, penetrate the sub-epithelial layer. However, the Raman signal from the lower layers undergoes multiple scattering resulting in the loss of signal. Thus the signal from the top layer of cells reaching the collection optics is very high. It is known that during the reproducible age, cervical tissue undergoes inflammation associated by fibrous changes in the tissue which may explain presence of structural proteins like collagen in normal Raman spectrum whereas hypercellularity is reflected in the malignant spectrum. The findings of difference spectrum are corroborated by spectral profile of factor 1 and profile of spectral residual between the measured and simulated mean normal spectrum using the factors of standard malignant set (data not shown). Objective discrimination is one of the major advantages of optical spectroscopy methods. Multivariate statistical tools like PCA, Hierarchical Cluster Analysis, and artificial neural networks are some of the widely used approaches for this purpose. In the present study, PCA was employed. In PCA, large spectral data are compressed into small number of independent variations known as Factors or Principal Components and contributions of these factors are known as Scores. Scores of factors is one of the widely used parameter for classification. As can be seen from Fig. 3, contribution of factor 1 is positive for normal and negative for malignant spectra. Thus, score of factor gives a reasonable discrimination of normal and malignant spectra. The mean score for normal and malignant are 0.074 and 0.058, respectively. A minor overlap between two clusters corresponding to normal and malignant spectra can be seen up to mean 1 standard deviation. Thus almost 75% sensitivity/specificity in classification of tissue type was achieved by this approach. Though classification based on scores of factors is widely used, it can be subjective as discrimination factors are selected. Usually, algorithms of discriminant analysis [33] are employed for identifying discriminating factors or principal components. However, this may not provide a very good classification under all circumstances. In the present study, other parameters like Mahalanobis distance, a measure of proximity of two spectra and Spectral residual, squared error

Fig. 3. PCA of cervical tissue spectra: score vs. sample number: (*) normal (~) malignant.

139

sum of difference between recorded and simulated spectra, have also been explored to achieve better discrimination between normal and malignant tissue type. The Mahalanobis distance is normally expressed in units of standard deviation. In the present analysis it is given by D2 ¼ ðStest ÞM 1 ðStest Þ0 where Stest is the vector of the scores and sum of squared spectral residuals for a given test sample, and M given by M = ((S0 S)/(n  1)), where S contains the corresponding parameters for the calibration set (n standards) [33,34]. There are two main advantages in using D2 as a discriminating parameter. As seen from the equation, D2 explicitly accounts for any correlations between the variables, namely scores of factors. Secondly when sufficiently large number of certified samples is available for various stages of malignancy, standard calibration sets can be prepared for each class. In the present study, these parameters were computed by comparing all spectra against normal and malignant standard sets. Randomly selected 32 normal and 37 malignant spectra were used to develop standard sets for normal and malignant conditions, respectively. Mahalanobis distance and spectral residual of these constituent spectra of standard sets were computed by comparing against both normal and malignant standard sets retrospectively. By fixing an upper limit for inclusion in any class represented by the standard calibration set for that class we could possibly achieve any desired level of discrimination for staging [32,35]. Values for these parameters will be very small when spectra and standard set correspond to same class and very high when spectra of different classes are compared. As an example, values for malignant spectra were very small when compared against malignant standard set and higher when compared against normal standard set and vice versa. Plots of Mahalanobis distance versus sample number and Mahalanobis distance versus Spectral residual for retrospective studies using normal standard set are shown in Fig. 4A and B, respectively. Mahalanobis distance for all normal spectra is around 0.85  0.5 and malignant spectra show values around 7.68  1.89. As mentioned earlier, Mahalanobis distance is the standard deviation and spectrum having values more than 3 will have probability of 0.5% or less as being classified as the same class against which the spectrum is compared. Since all malignant spectra have much higher Mahalanobis distance values, the probability of their belonging to the normal group against which all spectra were compared, is practically zero, Fig. 4A and B. Similarly, spectral residuals for all normal spectra were around 3.07  1.8 and higher values, around 19.76  4.22 were observed for malignant spectra, Fig. 4A. Thus 99.5% classification of normal and malignant tissues could be achieved by this methodology, Fig. 4 A and B. Validity of these standard sets as well as the approach was evaluated by comparing all other normal and malignant spectra, prospectively, against normal and malignant standard sets. As expected, all spectra corresponding to malignant tissues gave very low values for M distance (1.28  1.02) and Spectral Residuals (4.32  3.22) when compared against malignant standard set and higher values for M distance (6.84  1.87) and spectral

140

C.M. Krishna et al. / Vibrational Spectroscopy 41 (2006) 136–141 Table 2 Multiparametric match/mismatch table compared against malignant standard set

Fig. 4. PCA of cervical tissue spectra: retrospective study. (A) Plot of M distance vs. sample number. (B) Plot of Spectral Residual vs. Mahalanobis distance (*) normal (~) malignant.

residuals (18.33  4.18)) when compared against normal standard set. Results obtained in prospective study using malignant standard set are shown in Fig. 5. In this study also very good classification between normal and malignant spectra was observed as shown in Fig. 5. Thus the results obtained in the study not only validate the standard sets, but also indicate that Raman spectroscopy methods can be used for discrimination of tissue type in cervical malignancy. This approach of computing Mahalanobis distance and spectral residual was further extended to construct matchmismatch tables. In this analysis, spectra were compared against standard sets of data with fixed values for Mahalanobis distance and spectral residual for inclusion into that particular group. This methodology is known as Limit Test. In this analysis,

Fig. 5. PCA of cervical tissue spectra: prospective study. (A) Plot of Spectral Residual vs. Mahalanobis distance (*) normal (~) malignant.

Sample number

Match

Limit tests

1 2 3 4 5 6–79 80 81 82 83 84 85–149

Possible Possible Yes Possible Possible Yes No No No No No No

Pass (PP?#) Pass (PP?#) Pass (PPP#) Pass (PP?#) Pass (PP?#) Pass (PPP#) Fail (P?F#) Fail (PFF#) Fail (FFF#) Fail (FFF#) Fail (PFF#) Fail (PFF#)

Mahalanobis distance value was set at 3 for acceptance. If the values for a given spectrum fall within the limits, that spectrum will be labeled as Yes (match); all other spectra will be labeled as No (no match). In this study, all the test spectra were compared against malignant standard set. All the normal spectra were labeled as No (no match) and malignant spectra were labeled as Yes (match) as shown in Table 2. Exactly opposite results were obtained when normal standard set was used (data not shown). These results thus further validate the standard normal and malignant sets developed in the study as well as Raman spectroscopy approach for discrimination of malignant and normal tissues in cervical malignancy. 4. Conclusions Raman spectra of normal and malignant cervix tissue spectra exhibit very prominent differences. Spectral features like broader amide I, and peaks at 1384, 1269, 939 and 854 cm1 are characteristics of normal spectra which indicate presence of structural proteins like collagen. On the other hand several features in malignant spectra such as sharper amide I, a minor red shift in 1450 cm1 and a sharper amide III band can be assigned to proteins, lipids and DNA based on available literature data. These observations are further supported by spectral profiles of difference spectrum (normal  malignant spectra), load of factor 1, spectral residual between actual and simulated normal spectrum using factors of malignant standard sets. PCA produced a very good discrimination of normal and malignant spectra via scores of factors, Mahalanobis distance, and spectral residuals. Parameters such as Mahalanobis distance, spectral residual were tested retrospectively and prospectively, thus validating the standards sets. Further, a multiparametric limit tests approach was also explored and was found to provide best classification. In this methodology pathological status of an unknown spectrum can be determined by comparing with standard sets. Thus this methodology provides an opportunity to achieve desired level of discrimination for normal and malignant condition by developing standard sets for each category. The present study thus validates application of Raman spectroscopy methods in diagnosis of cervical malignancy.

C.M. Krishna et al. / Vibrational Spectroscopy 41 (2006) 136–141

Acknowledgements The work was carried out under the Department of Atomic Energy, Board of Research in Nuclear Sciences, Govt. of India project entitled ‘Laser spectroscopy as predictor of tumor response to radiation therapy in cervical cancer’, No. 2003/34/ 17/BRNS/1903. One of the authors (NBP) is grateful to DAE-BRNS for providing the research fellowship to carry out this work. Miss Keerthi is acknowledged for her technical support in sample collection, storage and data acquisition. References [1] [2] [3] [4] [5] [6]

[7] [8]

[9]

[10] [11]

[12] [13]

[14]

P.N. Notani, Curr. Sci. 81 (2001) 465. L.G. Koss, J. Am. Med. Assoc. 261 (1989) 737. M.F. Mitchell, Consult. Obstet. Gynecol. 6 (70) (1994). R.R. Alfano, C.H. Liu, W.L. Sha, H.R. Zhu, D.L. Akins, J. Cleary, R. Prudente, E. Cellmer, Lasers Life Sci. 4 (1991) 23. A. Mahadevan-Jansen, M.F. Mitchell, N. Ramanujam, U. Utzinger, R. Richards-Kortum, Photochem. Photobiol. 68 (1998) 427. A. Mahadevan-Jansen, M.F. Mitchell, N. Ramanujam, A. Malpica, S. Thomsen, U. Utzinger, R. Richards-Kortum, Photochem. Photobiol. 68 (1998) 123. U. Utzinger, D.L. Heintzelman, A. Mahadevan-Jansen, A. Malpica, M. Follen, R. Richards-Kortum, Appl. Spectr. 55 (2001) 955. S. Mordechai, R.K. Sahu, Z. Hammody, S. Mark, K. Kantarovich, H. Guterman, A. Podshyvalov, J. Goldstein, S. Argov, J. Microsc. 215 (2004) 86. S. Mark, R.K. Sahu, K. Kantarovich, A. Podshyvalov, H. Guterman, J. Goldstein, R. Jagannathan, S. Argov, S. Mordechai, J. Biomed. Opt. 9 (2004) 558. J.I. Chang, Y.B. Huang, P.C. Wu, C.C. Chen, S.C. Huang, Y.H. Tsai, Gynecol. Oncol. 91 (2003) 577. R. Sindhuphak, S. Issaravanich, V. Udomprasertgul, P. Srisookho, S. Warakamin, S. Sindhuphak, R. Boonbundarlchai, N. Dusitsin, Gynecol. Oncol. 90 (2003) 10. M.J. Romeo, B.R. Wood, M.A. Quinn, D. McNaughton, Biopolymers 72 (2003) 69. P.T. Wong, M.K. Senterman, P. Jackli, R.K. Wong, S. Salib, C.E. Campbell, R. Feigel, W. Faught, M. Fung Kee Fung, Biopolymers 67 (2002) 376. L. Chiriboga, P. Xie, H. Yee, D. Zarou, D. Zakim, M. Diem, Cell. Mol. Biol. 44 (1998) 219 (Noisy-le-grand).

141

[15] L. Chiriboga, P. Xie, H. Yee, V. Vigorita, D. Zarou, D. Zakim, M. Diem, Biospectroscopy 4 (1998) 47. [16] B.R. Wood, L. Chiriboga, H. Yee, M.A. Quinn, D. McNaughton, M. Diem, Gynecol. Oncol. 93 (2004) 59. [17] B.J. Morris, C. Lee, B.N. Nightingale, E. Molodysky, L.J. Morris, R. Appio, S. Sternhell, M. Cardona, D. Mackerras, L.M. Irwig, Gynecol. Oncol. 56 (1995) 245. [18] E.M. Gill, A. Malpica, R.E. Alford, A.R. Nath, M. Follen, R. RichardsKortum, N. Ramanujam, Photochem. Photobiol. 77 (2003) 653. [19] A. Nath, K. Rivoire, S. Chang, D. Cox, E.N. Atkinson, M. Follen, R. Richards-Kortum, J. Biomed. Opt. 9 (2004) 523. [20] W.K. Huh, R.M. Cestero, F.A. Garcia, M.A. Gold, R.S. Guido, K. McIntyre-Seltman, D.M. Harper, L. Burke, S.T. Sum, R.F. Flewelling, R.D. Alvarez, Am. J. Obstet. Gynecol. 190 (2004) 1249. [21] K. Badizadegan, V. Backman, C.W. Boone, C.P. Crum, R.R. Dasari, I. Georgakoudi, K. Keefe, K. Munger, S.M. Shapshay, E.E. Sheetse, M.S. Feld, Faraday Discuss. 126 (2004) 265. [22] C. Brookner, U. Utzinger, M. Follen, R. Richards-Kortum, D. Cox, E.N. Atkinson, J. Biomed. Opt. 8 (2003) 479. [23] I. Pavlova, K. Sokolov, R. Drezek, A. Malpica, M. Follen, R. RichardsKortum, Photochem. Photobiol. 77 (2003) 550. [24] I. Georgakoudi, B.C. Jacobson, M.G. Muller, E.E. Sheets, K. Badizadegan, D.L. Carr-Locke, C.P. Crum, C.W. Boone, R.R. Dasari, J. Van Dam, M.S. Feld, Cancer Res. 1 (62) (2002) 682. [25] I. Georgakoudi, E.E. Sheets, M.G. Muller, V. Backman, C.P. Crum, K. Badizadegan, R.R. Dasari, M.S. Feld, Am. J. Obstet. Gynecol. 186 (2002) 374. [26] M.C. Skala, J.M. Squirrell, K.M. vrotos, J.C. Eickhoff, A. GendronFitzpatrick, K.W. Eliceiri, N. Ramanujam, Cancer Res. 65 (2005) 1180. [27] S.K. Majumdar, et al. Curr. Sci. 70 (1996) 833. [28] S.K. Chang, Y.N. Mirabal, E.N. Atkinson, D. Cox, A. Malipica, M. Follen, R. Richards-Kortum, J. Biomed. Opt. 10 (2005) 024031. [29] E.S. Parker, Applications of infrared and Raman and resonance Raman spectroscopy in biochemistry, Plenum Press, New York, 1983. [30] P.J. Tonge, P.R. Carey, Raman, Resonance Raman and FTIR spectroscopic studies of enzyme-substrate complexes, in: R.J.H. Clark, R.E. Hester (Eds.), Biomolecular Spectroscopy Part A, Advances in Spectroscopy, vol. 20, John Wiley & Sons, Chchester, p. 129 (Chapter 3). [31] K. Venkatakrishna, J. Kurein, Keerthilatha M. Pai, C. Murali Krishna, G. Ullas, V.B. Kartha, Curr. Sci. 80 (2001) 101–105. [32] R. Malini, K. Venkatakrishna, J. Kurien, K.M. Pai, L. Rao, V.B. Kartha, C. Murali Krishna, Biopolymers 81 (2006) 179. [33] PLS plus/IQ, Galactic Industries Corporation 1991–1999. [34] P.C. Mahalanobis, Proc. Natl. Inst. Sci. Ind. 12 (1936) 49–55. [35] B.K. Manjunath, J. Kurein, L. Rao, C. Murali Krishna, M.S. Chidananada, K. Venkatakrishna, V.B. Kartha, J. Photochem. Photobiol. B Biol. 73 (2004) 49.