- Email: [email protected]
Detection of benign epithelia, prostatic intraepithelial neoplasia, and cancer regions in radical prostatectomy tissues using Raman spectroscopy
Vibrational Spectroscopy 53 (2010) 227–232
Contents lists available at ScienceDirect
Vibrational Spectroscopy journal homepage: www.elsevier.com/locate/vibspec
Detection of benign epithelia, prostatic intraepithelial neoplasia, and cancer regions in radical prostatectomy tissues using Raman spectroscopy Suneetha Devpura a , Jagdish S. Thakur a,∗ , Fazlul H. Sarkar b , Wael A. Sakr b , Vaman M. Naik c , Ratna Naik a a b c
Department of Physics and Astronomy, Wayne State University, 666 W Hancock, Detroit, MI 48201, USA Karmanos Cancer Institute, Detroit, MI 48201, USA Department of Natural Sciences, University of Michigan-Dearborn, Dearborn, MI 48128, USA
a r t i c l e
i n f o
Article history: Received 5 January 2010 Received in revised form 3 March 2010 Accepted 9 March 2010 Available online 16 March 2010 Keywords: Raman spectroscopy Prostate cancer diagnosis Benign epithelia Prostatic intraepithelial neoplasia Bulk tissues Gleason scores
a b s t r a c t In this paper we have studied benign epithelia (BE), prostatic intraepithelial neoplasia (PIN), adenocarcinoma, and different Gleason scores in human prostate bulk tissues using Raman spectroscopy. A careful investigation of the data shows that two main differences in the Raman spectral features of BE, PIN, and cancerous tissues: (i) a strong variations in the band intensities of certain bands and (ii) shift in certain band positions. In order to quantify these variations, Raman data were further analyzed using chemometric methods of principal component analysis (PCA) and discriminant function analysis (DFA). The PCA and DFA clearly separated the data into three main distinct pathological groups representing BE, PIN, and cancerous state in tissue. Similarly the analysis of the Raman data of tissues with different Gleason scores shows that the data can be categorized into three distinct groups representing Gleason scores 6, 7, and 8. The results of this study demonstrate that Raman spectroscopy can diagnose different stages of the prostate cancer due to the differences in their biochemical compositions. Published by Elsevier B.V.
1. Introduction Prostate cancer is the second leading cause of cancer related deaths in men in the United States [1]. The initial detection of prostate cancer involves a digital rectal examination (DRE) and a prostate specific antigen (PSA) test [2] which measures the level of PSA in patients’ blood. The presence of high levels of PSA is normally taken as an indication of prostate cancer [3]. Normal procedure for confirming the diagnosis of prostate cancer requires a prostate biopsy, where a radiologist/urologist takes six or more samples of prostate tissue for pathological examination. The interpretations of histological examination of biopsies, considered as the “gold” standard for diagnosis, are often quite subjective and vary significantly from one pathologist to another. Another widely used procedure for assessing the severity of prostate cancer is the Gleason grading [4,5]. Under the Gleason grading procedure, there are five grades ranging from 1 to 5 indicating the degree to which the cancerous tissue looks compared to normal prostate tissue. Grade 1 looks more towards the normal prostate which contains proliferated microacinar structures contoured by prostatic luminal cells without a basal cell layer [6]. On the contrary, a grade 5 contains irregular, poorly differentiated infiltrating individual cells and grades 2–4 fall in between. A given Gleason score is a combination of two Gleason
∗ Corresponding author. Tel.: +1 313 577 2104; fax: +1 313 577 3932. E-mail address: [email protected] (J.S. Thakur). 0924-2031/$ – see front matter. Published by Elsevier B.V. doi:10.1016/j.vibspec.2010.03.009
grades. Therefore the Gleason score may vary between 2 and 10. However, a considerable difference in the Gleason scores among pathologists for a given type of tissue has been observed. Interestingly, it is still difficult to predict which cancer is likely to behave more aggressively even with a mid-grade tumor, which raises the possibility that reliable method for predicting aggressive biology of prostate cancer would be important. Other than the histopathological examination and PSA test, no other reliable and objective clinical tests are available for confirming the diagnosis of prostate cancer, especially when the cancer foci are very small. These diagnostic procedures are generally quite resource extensive and time consuming. With the advancement in optical technologies, it is becoming possible to have non-invasive and reliable optical detection methods whose results can be interpreted objectively. Therefore, new diagnostic procedures and tests are continually being developed for assessing the aggressiveness and stages of the disease. Raman spectroscopy by its very nature is capable of providing highly detailed chemical information about a tissue sample and is gaining a wide recognition as an objective method for the diagnosis of diseases in tissues. In the Stokes Raman scattering process, an incident laser light interacting with the molecules gets inelastically scattered, transferring some of its energy to the molecular vibrational excitations in the sample. Since biological tissue contains a large number of different molecules, its Raman spectrum, which contains detailed chemical, compositional and conformational information about the molecules in the tissue under investigation, will be rich in information content. A biolog-
228
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
ical tissue from an organ contains certain types of biomolecules in a certain molecular composition unique to the health condition of that organ. This unique composition of the biomolecules also has a unique set of vibrational states representing the vibrational motions of these biomolecules. Occurrence of any disease in an organ results in changes in the normal biochemical composition of the organ which is reflected in the vibrational spectrum of the tissue. Thus the Raman spectroscopy could be useful for detecting these changes, which may be able to predict the health condition of a tissue. Raman spectroscopy has been successfully tested in a variety of diseases in organs, like skin [7], mammary glands [8], brain [9], and gastrointestinal track. Recently, formalin-fixed paraffin preserved cervical tissue samples of normal, pre-malignant (cervical intraepithelial neoplasia), and invasive carcinoma have been investigated to detect biochemical changes accompanying cervical cancer progression [10]. Raman spectroscopy has also been used to study the differences between various prostatic adenocarcinoma cell lines [11], and for in vitro identification and grading of prostatic adenocarcinoma [12]. Raman spectral data in conjunction with PCA [13] have been used to determine the variations in the glycogen and nucleic acid contents in malignant and benign pathologies in prostatic tissues. The basic aim of the present study was to test the potential of Raman spectroscopy as a diagnostic tool for the detection of BE [14], PIN [15], and different Gleason scores of cancer in human prostate tissues. In this study, we have collected Raman spectra from different pathological regions of bulk human prostate tissue samples. We have carefully analyzed these spectra through visual inspection and found that the average spectrum for each pathological state contained different spectral characteristics strong enough to differentiate each pathological state from others. We have further analyzed the data using PCA and DFA [16] statistical procedures to determine different pathological groups in our data set. 2. Materials and methods In this study, ten prostatic bulk tissues were used (protocol #: 0705004871). These tissues were obtained from the archives of the Department of Pathology at Karmanos Cancer Institute in Detroit, MI. 2.1. Preparation of tissues for Raman spectroscopy Two parallel tissue sections (one 5 m and the other 10 m thick) were cut from paraffin-embedded tissue samples, using a microtome. The 5-m-thick tissue section was placed on a glass slide, dewaxed [17,18] by washing for 5 min each in xylene and pure ethanol, and used for hemotoxyilin and eosin (H&E) stain-
ing for subsequent pathological examination by an experienced pathologist. The 10-m-thick section was placed on a stainless steel substrate, dewaxed and dried in air for the Raman studies. All the tissues were processed at the University Pathology Research Services, Karmanos Cancer Institute. When the tissue was prepared for the Raman spectroscopy study, it was dewaxed by treating it again with xylene and ethanol baths. In this process, sometimes wax residuals are left in the tissue which can be detected through the presence of sharp wax bands at 1063, 1130, 1296, 1436, and 1465 cm−1 . However, if tissue is dewaxed properly then the wax residuals can be completely removed from the tissue [17]. The Raman spectra were recorded from unstained tissue sections using corresponding regions marked as BE, PIN, and different cancer categories on the parallel H&E stained tissue section (see Fig. 1). 2.2. Raman spectroscopic measurements We have used a Renishaw RM1000 Raman microscopespectrometer to record the Raman spectra from prostate tissues. The spectrometer is equipped with a 578 × 385 pixel CCD and a 785 nm diode laser. A 50× objective lens was used to focus the excitation laser beam to a spot size of ∼4 m × 30 m on the sample and collect the back-scattered light. The spectral resolution is ∼3.5 cm−1 . The laser power at the sample was about 3 mW. The spectra were recorded in the 500–1900 cm−1 range using an integration time of 20 s per spectrum, and three accumulations were averaged to improve the signal-to-noise ratio. The Raman spectra were recorded from the unstained tissue section. The appropriate regions on the unstained tissue section were identified with the help of the adjacent stained section with regions marked by a pathologist as BE, PIN, and cancer. The identification of these regions is done with an optical microscope to make sure the data were collected only from the marked regions. The Raman spectra were collected from different spots from each of the marked regions. The separation distance between each of the data collection spot is about 10 times the laser spot diameter. Each collected spectrum represents a datum from a physically different location on the tissue. Care was taken to ensure that the spots from which the spectra were collected are located well inside the appropriate regions marked by the pathologist. A total of 366 spectra were recorded from regions marked as BE, PIN, and cancer of different Gleason scores such as 6, 7, and 8. 3. Results and discussion After collecting the data, we have visually inspected each spectrum and discarded those which contain any trivial non-standard Raman features. Although these types of spectra are few; how-
Fig. 1. Stars represent the locations of the Raman measurements (a) tissue section used for the Raman measurement and (b) adjacent (H&E) stained tissue section used for pathological examination.
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
229
ever, mixing them with others can introduce errors in the results of statistical analysis. From the remaining good spectra we removed any spurious bands due to cosmic radiation. The noise was filtered using wavelets method [19] while the fluorescence background in the spectra was removed with minmax adaptive algorithm that requires no a priori knowledge of the spectra [20]. Finally, each spectrum was normalized using the highest intensity band in the spectrum. It is a well known that due to the morphological complexity in diseased tissues, misrepresentation of the observed morphological features happens particularly at the border regions of cancer with other pathologies (PIN, BE or cancer) leading to serious errors in the diagnosis. However, with Raman spectroscopy being an objective and quantitative optical method which can determine biochemical composition of tissue, the chances of misrepresentation are small. To find pathologically misrepresented regions of the tissues, we visually examined all the spectra in each of the pathological category. We found that spectra from each of the pathological categories have some distinct features, though in some categories the distinct and unique features are quite subtle and small. Interestingly, we found that the spectral features of some of the spectra collected from a region assigned to a particular pathology did not match with the unique spectral features of that pathological state but they matched with the spectral features of the pathological state assigned to the adjacent region. However, the observed mismatches in the visual inspection were not corrected in the data analysis using PCA and DFA. It is interesting to note that the data analysis did associate all the misrepresented spectra with pathological states to which their spectral features matched. The average spectra from BE, PIN, and cancer regions of bulk tissues are shown in Fig. 2. These spectra are typical of biological tissues exhibiting a combination of sharp and broad Raman bands arising from the molecular vibrations of proteins, nucleic acids, lipids, and other constituent molecules. In our samples we have not observed any bands which can be associated with wax. The low intensity broad bands observed at 1063 and 1129 cm−1 are close to the Raman bands of wax. However, they do not originate from wax, as the Raman bands of wax are normally very sharp and intense [17]. The most prominent sharp band around 1002 cm−1 is seen in all the cases having similar spectral features. This well known band is attributed to ring deformation mode associated
with aromatic amino acids; phenyl alanine and proline, and is often observed in the Raman spectra of proteins [9]. Careful examination of Fig. 2 shows that the average spectrum of each region (e.g. BE, PIN, and cancer) of the bulk tissues has some characteristic spectral features that differentiates it from the others. These different characteristic features basically highlight the fact that each region has different biochemical compositions unique to that region. In addition to the subtle differences in the spectral features, the differences in the BE, PIN, and cancer regions can also be seen in the intensity of the bands at 726, 759, 782, 1063 and 1097 cm−1 . Especially, the difference in the intensities is quite large for the band at 782 cm−1 in the PIN and cancer regions compared to the BE region. This mode represents breathing of U, C, T rings of DNA bases which can be interpreted as an enhancement of DNA content in PIN and cancer compared to the BE region. This observation is in agreement with the results of Teba et al. (2007) [21]. However, it should be emphasized that there is a considerable overlap in Raman vibrational frequencies from other biological molecules and each observed band cannot be uniquely assigned to a particular molecule. To understand all the spectral variations in terms of chemical composition/conformational changes, one needs to know the vibrational assignment for each of the observed Raman frequencies. Fig. 2 shows that most of the spectral changes occur in the region 700–1200 cm−1 of the Raman spectra. This region of a spectrum contains numerous bands with varying intensities. Fig. 3 shows an enlarged region covering the 630–850 cm−1 region of the spectrum to highlight the differences. In fact, the intensity ratios of such characteristic bands have often been used to interpret the relative molecular compositions in tissues [22,23]. There are two significant changes observed in the data which could be used to differentiate the prostate cancer tissue from BE and PIN: (i) we have observed a definite correlation in the variations of band intensities at 726, 759, 782, and 831 cm−1 . In order to quantify this correlation, we chose a reference band at 643 cm−1 whose maximum intensity did not change for all the three main pathological states. With reference to this band, variations in other bands are shown in Fig. 3. Table 1 lists the percentage variations in the bands at 726, 759, 782, and 831 cm−1 with respect to the reference band at 643 cm−1 . The ratios of 726/643 and 782/643 of cancer increased significantly compared to BE and PIN states. However, if we consider the bands at 759 and 782 cm−1 in BE and PIN states, the relative intensities were reversed. A strong variation in the intensity bands at 726 and 782 cm−1 for cancerous regions is an important observation since these bands are related to the breathing modes of DNA whose content increases significantly in the cancerous region due to cell
Fig. 2. Average Raman spectrum for BE, PIN and cancer (here the black, green, and red curves represent BE, PIN, and cancer, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
Fig. 3. Variations in Raman intensity and shift in certain bands for BE, PIN and cancer spectra (here the black, green, and red curves represent BE, PIN, and cancer, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
230
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
Table 1 Percentage increase in the peak intensities of selected Raman bands with respect to the intensity of the 643 cm−1 peak. Raman band (cm−1 )
Cancer
PIN
BE
726 759 782 831
39 40 110 141
7 15 52 114
5 35 32 130
Fig. 6. First three PC loadings spectra for BE, PIN, and cancer. The Raman shift for the main positive and negative bands are marked against each loading curve (the black solid curve represents PCL1, the red dash-dot curve represents PCL2, and the dotted blue curve represents PCL3). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
show the spectral ranges where the variations between different Gleason scores can easily be seen. 3.1. Statistical analysis of the data
Fig. 4. The average Raman spectrum for G6, G7, and G8 Gleason scores (Here the pink, green, and blue curves represent G6, G7, and G8, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
proliferation. (ii) We observe shift in the band positions of 726 and 782 cm−1 bands in BE state compared to PIN and cancer states (see Fig. 3). Although the shift is small (about 2.0 cm−1 ), we believe it is significant since it is observed consistently in all the spectra in this group and is reflected in the average spectrum shown in Fig. 3. We have also studied the spectra from different Gleason scores of cancer. In Fig. 4, we have shown the mean spectrum from each Gleason score. Raman spectra of Gleason scores 6, 7, and 8 show small and subtle differences in some of the bands. In Fig. 5, we
For the purpose of quantification of the variances of the data from BE, PIN, and cancer regions, we have performed statistical analysis of the data using widely used statistical techniques such as PCA and DFA. For the analysis of the bulk prostate data, we considered spectral range 600–1800 cm−1 , the “finger print” region containing the fundamental vibrations of molecules. This region of the vibrational spectrum is often used for identification of molecules and conformational changes. First, we applied PCA method to reduce the dimensionality of the original data set having 601 variables. The transformed data set consists of 23 eigenvectors capturing more than 94% of variance in the data. Fig. 6 shows the results for the principal component loadings (PCLi, where i = 1, 2, 3). Unlike a Raman spectrum, loading spectrum contains positive and negative bands and their corresponding frequencies can be correlated with the variations in the molecular composition through the molecular vibrational assignments. The PCL1 contains maximum variance within the data set and it also captures the trend of band variances. From the PCL1 graph, one can see the enhancement in the concentrations of DNA content (670 and 782 cm−1 ), proteins (726 cm−1 ), phospholipids (1330 and 1368 cm−1 ), and nucleic acids (1576 cm−1 ) [24]. While, PCL2 represents increase in proteins (726 cm−1 ), DNA (1068 and 1094 cm−1 ), fatty acids
Fig. 5. Two narrow spectral ranges where significant differences in the Raman spectra of G6, G7, and G8 are observed (here the pink, green, and blue curves represent G6, G7, and G8, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
231
Fig. 8. First three PC loadings spectra for Gleason scores 6, 7, and 8 (the black solid curve represents PCL1, the red dash-dot curve represents PCL2, and the dotted blue curve represents PCL3). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.) Fig. 7. Classification results from DFA for BE, PIN, and cancer (the black triangles, green squares, and red circles represent BE, PIN, and cancer (CA)). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
(1132 and 1658 cm−1 ), phospholipids (1454 cm−1 ), and decrease in collagen (1280 cm−1 ), tryptophan (1550 cm−1 ), and phenylalanine (1605 cm−1 ). PCL3 shows an increase in phenylalanine 1000 cm−1 and decrease in fatty acids (1064, 1132, 1297, and 1440 cm−1 ) [24]. In Fig. 3, we have shown the variation in the intensities of the bands at 726, 759, 782, and 831 cm−1 . In the PCL1 spectrum we see these bands represent increase in proteins (726 cm−1 ) and DNA (784 cm−1 ), while from the PCL2 spectrum, we see that the bands 759 cm−1 and 831 cm−1 represent decrease in tryptophan and amino acids, respectively. Although, it is difficult to assess the concentrations of various components such as nucleic acids and proteins in a tissue under investigation, the Raman spectra in conjunction with known vibrational assignments for molecules do permit qualitative inference of their relative concentrations. In order to determine the presence of different groups in the data, we have performed DFA of the data with input taken from the PCA calculations. The results are shown in Fig. 7. It is encouraging to note that even though there is a huge heterogeneity in the tissues, the DFA classified the spectral data as BE, PIN, and cancer into three very distinct groups. However, some of the data points of one region are found to overlap with the others, e.g. some of PIN data are mixed with cancer data. This may be due to the fact that both cancer and PIN exhibit similar type of luminal cells [21]. When the spectra are being collected, given the size of laser spot (∼4 m × 30 m) it is difficult to determine whether the laser hits on luminal or basal cells or both within the epithelial cells. Thus the measured data can be a mixture of both. The prediction of the pathological state of the tissues using Raman studies and DFA is compared with that of pathology, and the results are shown in Table 2. To test the validity of our predicted classifications, we have performed leave-one-out cross-validation. With this method the association of each datum with one of the known pathological states is determined while the remaining data
are used as the training data set. The results of cross-validation are compared with the classification results in Table 2. The prediction accuracy (Table 2) becomes slightly smaller for cancer when data are tested with cross-validation method due to the smaller data size available for the analysis. The results for the detection of cancer in terms of sensitivity and specificity using the cross-validated data are 94% and 82%, respectively. The higher values of sensitivity and specificity are encouraging, which may guide in-depth future studies. The cancer tissues we studied contain Gleason scores 6, 7, and 8 only. We have performed the PC loadings of the Gleason score data and the PC loading spectra are shown in Fig. 8. PCL1 clearly shows decrease in the fatty acids contents (1064, 1132 and 1296 cm−1 ) and increase in other components that give rise to the bands at 1206, 1282, 1352 and 1396 cm−1 in the range from 1180 to 1400 cm−1 consistent with PCL3 as shown in Fig. 6. The energy of various positive and negative bands of PCL1 is marked against them. The spectra of PCL2 and PCL3 are also shown in the graph. The DFA clearly separates Gleason scores of 6, 7, and 8 into three distinct groups [25] as shown in Fig. 9. The overlapping in different Gleason scores is expected, since Gleason score is a combination of two Gleason grades, i.e. certain spectra may be common to one of the Gleason scores. The individual accuracies of cross-validation in Gleason scores 6, 7, and 8 are more than 81% as shown in Table 3. This study
Table 2 Comparison of Raman diagnostic results with pathologic diagnosis. Raman diagnosis
BE PIN Cancer
Pathological diagnosis BE
PIN
Cancer
Total
Original
Crossvalidation
123 2 1
11 46 5
9 0 169
143 48 175
86.0% 95.8% 96.6%
78.3% 95.8% 94.3%
Fig. 9. Classification results from DFA for Gleason scores 6, 7, and 8 (here the pink circles, green squares, and blue triangles represent G6, G7, and G8, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
232
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
Table 3 Comparison of Raman diagnostic results of Gleason scores with pathologic diagnosis. Raman diagnosis
Cancer G6 Cancer G7 Cancer G8
Pathological diagnosis Cancer G6
Cancer G7
Cancer G8
Total
Original
Cross-validation
60 0 0
3 59 1
0 1 51
63 60 52
95.2% 98.3% 98.1%
81.0% 90.0% 86.5%
demonstrates that Raman spectroscopy can be used to differentiate different Gleason gradings in cancers. 4. Conclusions In this study we have investigated BE, PIN, and cancer regions of dewaxed human prostate tissues using Raman spectroscopy. The bulk tissues for this study were obtained from prostate cancer patients who underwent radical prostatectomy. We found that Raman spectroscopy can detect cancerous regions in bulk tissues in the presence of PIN and BE regions. Investigation of the Raman data shows that the average spectra of BE, PIN, and cancer regions have distinguishable unique Raman spectral features highlighting the fact that the biochemical composition of the tissue region changes when its pathological state changes. Although taking into consideration the variations in the spectral features over the full range (600–1800 cm−1 ) increases the discrimination ability of the analysis procedure; the bands at 726, 759, 782, and 831 cm−1 are also found to be key contributors for distinguishing the prostate cancer from PIN and BE. We also found that cancerous regions having different Gleason scores (Gleason scores 6, 7, and 8) can be differentiated, suggesting that Raman spectroscopy can detect different degrees of aggressiveness of the prostate cancer in tissues. Acknowledgement We would like to acknowledge the support from the Richard J. Barber Foundation for Interdisciplinary Research. References [1] A. Jemal, R. Siegel, E. Ward, Y. Hao, J. Xu, M.J. Thun, Cancer J. Clin. 59 (2009) 225–249, doi:10.3322/caac.20006. [2] A.W. Partin, M.W. Kattan, E.N. Subong, P.C. Walsh, K.J. Wojno, L.E. Oesterling, P.T. Scardino, J.D. Pearson, J. Am. Med. Assoc. 277 (1997) 1445–1451. [3] P.A. Humphrey, Modern Pathol. 17 (2004) 292–306, doi:10.1038/ modpathol.3800054.
[4] W.C. Allsbrook, K.A. Mangold, M.H. Johnson, R.B. Lane, C.G. Lane, J.I. Epstein, Hum. Pathol. 32 (1) (2001) 81–88, doi:10.1053/hupa.2001.21135. [5] G.L. Andriole, T.L. Bullock, J.S. Belani, E. Traxel, Y. Yan, D.G. Bostwick, P.A. Humphrey, Urology 70 (Suppl. 6A) (2007) 22–26, doi:10.1016/j.urology.2007.06.1128. [6] H. Hricak, P.T. Scardino, R.H. Reznek, Prostate Cancer, Cambridge University Press, New York, 2009, pp. 1–6. [7] P.J. Caspers, G.W. Lucassen, R. Wolthuis, H.A. Bruining, G.J. Puppels, Biospectroscopy 4 (1998) S31–S39. [8] K.E. Shafer-Peltier, A.S. Haka, M. Fitzmaurice, J. Crowe, J. Myles, R.R. Dasari, M.S. Feld, J. Raman Spectrosc. 33 (2002) 552–563, doi:10.1002/jrs.877. [9] C. Krafft, S.B. Sobottka, G. Schackert, R. Salzer, Analyst 130 (2005) 1070–1077, doi:10.1039/b419232j. [10] F.M. Lyng, E.Ó. Faoláin, J. Conroy, A.D. Meade, P. Knief, B. Duffy, M.B. Hunter, J.M. Byrne, P. Kelehan, H.J. Byrne, Exp. Mol. Pathol. 82 (2007) 121–129, doi:10.1016/j.yexmp.2007.01.001. [11] P. Crow, B. Barrass, C.A. Kendall, M. Hart-Prieto, M. Wright, R. Persad, N. Stone, Br. J. Cancer 92 (2005) 2166–2170, doi:10.1038/sj.bjc.6602638. [12] P. Crow, N. Stone, C.A. Kendall, J.S. Uff, J.A.M. Farmer, H. Barr, M.P.J. Wright, Br. J. Cancer 89 (2003) 106–108, doi:10.1038/sj.bjc.6601059. [13] I.T. Jolliffe, Principal Components Analysis, 2nd ed., Springer-Verlag, New York, 2002. [14] N. Stone, C. Kendall, N. Shepherd, P. Crow, H.J. Barr, Raman Spectrosc. 33 (2002) 564–573, doi:10.1002/jrs.882. [15] D.W. Keetch, P. Humphrey, D. Stahl, D.S. Smith, W.J.J. Catalona, Urology 154 (2) (1995) 347–351. [16] W.R. Klecka, Discriminant Analysis, Sage Publications, Newbury Park, 1980. [17] E.Ó. Faoláin, M.B. Hunter, J.M. Byrne, P. Kelehan, H.A. Lambkin, H.J. Byrne, F.M. Lyng, J. Histochem. Cytochem. 53 (1) (2005) 121–129, doi:10.1369/jhc.4A6536.2005. [18] E.Ó. Faoláin, M.B. Hunter, J.M. Byrne, P. Kelehan, M. Mcnamara, H.J. Byrne, F.M. Lyng, Vib. Spectrosc. 38 (2005) 121–127, doi:10.1016/j.vibspec.2005.02.013. [19] F. Keinert, Wavelets and Multiwavelets, Chapman and Hall/CRC Press, Boca Raton, FL, 2004. [20] A. Cao, A.K. Pandya, G.K. Serhatkulu, R.E. Weber, H. Dai, J.S. Thakur, V.M. Naik, R. Naik, G.W. Auner, R.J. Rabah, Raman Spectrosc. 38 (2007) 1199–1205, doi:10.1002/jrs.1753. [21] F. Teba, R. Martin, V. Gomez, L.M. Herranz, L. Santamaria, Image Anal. Stereol. 26 (2007) 93–99. [22] G. Shetty, C. Kendall, N. Shepherd, N. Stone, H. Barr, Br. J. Cancer 94 (2006) 1460–1464, doi:10.1038/sj.bjc.6603102. [23] D. Pappas, B.W. Smith, J.D. Winefordner, Talanta 51 (2000) 131–144. [24] Z. Movasaghi, S. Rehman, I.U. Rehman, Appl. Spectrosc. 42 (2007) 493–541, doi:10.1080/05704920701551530. [25] M.J. Baker, E. Gazi, M.D. Brown, J.H. Shanks, P. Gardner, N.W. Clarke, Br. J. Cancer 99 (2008) 1859–1866, doi:10.1038/sj.bjc.6604753.
Contents lists available at ScienceDirect
Vibrational Spectroscopy journal homepage: www.elsevier.com/locate/vibspec
Detection of benign epithelia, prostatic intraepithelial neoplasia, and cancer regions in radical prostatectomy tissues using Raman spectroscopy Suneetha Devpura a , Jagdish S. Thakur a,∗ , Fazlul H. Sarkar b , Wael A. Sakr b , Vaman M. Naik c , Ratna Naik a a b c
Department of Physics and Astronomy, Wayne State University, 666 W Hancock, Detroit, MI 48201, USA Karmanos Cancer Institute, Detroit, MI 48201, USA Department of Natural Sciences, University of Michigan-Dearborn, Dearborn, MI 48128, USA
a r t i c l e
i n f o
Article history: Received 5 January 2010 Received in revised form 3 March 2010 Accepted 9 March 2010 Available online 16 March 2010 Keywords: Raman spectroscopy Prostate cancer diagnosis Benign epithelia Prostatic intraepithelial neoplasia Bulk tissues Gleason scores
a b s t r a c t In this paper we have studied benign epithelia (BE), prostatic intraepithelial neoplasia (PIN), adenocarcinoma, and different Gleason scores in human prostate bulk tissues using Raman spectroscopy. A careful investigation of the data shows that two main differences in the Raman spectral features of BE, PIN, and cancerous tissues: (i) a strong variations in the band intensities of certain bands and (ii) shift in certain band positions. In order to quantify these variations, Raman data were further analyzed using chemometric methods of principal component analysis (PCA) and discriminant function analysis (DFA). The PCA and DFA clearly separated the data into three main distinct pathological groups representing BE, PIN, and cancerous state in tissue. Similarly the analysis of the Raman data of tissues with different Gleason scores shows that the data can be categorized into three distinct groups representing Gleason scores 6, 7, and 8. The results of this study demonstrate that Raman spectroscopy can diagnose different stages of the prostate cancer due to the differences in their biochemical compositions. Published by Elsevier B.V.
1. Introduction Prostate cancer is the second leading cause of cancer related deaths in men in the United States [1]. The initial detection of prostate cancer involves a digital rectal examination (DRE) and a prostate specific antigen (PSA) test [2] which measures the level of PSA in patients’ blood. The presence of high levels of PSA is normally taken as an indication of prostate cancer [3]. Normal procedure for confirming the diagnosis of prostate cancer requires a prostate biopsy, where a radiologist/urologist takes six or more samples of prostate tissue for pathological examination. The interpretations of histological examination of biopsies, considered as the “gold” standard for diagnosis, are often quite subjective and vary significantly from one pathologist to another. Another widely used procedure for assessing the severity of prostate cancer is the Gleason grading [4,5]. Under the Gleason grading procedure, there are five grades ranging from 1 to 5 indicating the degree to which the cancerous tissue looks compared to normal prostate tissue. Grade 1 looks more towards the normal prostate which contains proliferated microacinar structures contoured by prostatic luminal cells without a basal cell layer [6]. On the contrary, a grade 5 contains irregular, poorly differentiated infiltrating individual cells and grades 2–4 fall in between. A given Gleason score is a combination of two Gleason
∗ Corresponding author. Tel.: +1 313 577 2104; fax: +1 313 577 3932. E-mail address: [email protected] (J.S. Thakur). 0924-2031/$ – see front matter. Published by Elsevier B.V. doi:10.1016/j.vibspec.2010.03.009
grades. Therefore the Gleason score may vary between 2 and 10. However, a considerable difference in the Gleason scores among pathologists for a given type of tissue has been observed. Interestingly, it is still difficult to predict which cancer is likely to behave more aggressively even with a mid-grade tumor, which raises the possibility that reliable method for predicting aggressive biology of prostate cancer would be important. Other than the histopathological examination and PSA test, no other reliable and objective clinical tests are available for confirming the diagnosis of prostate cancer, especially when the cancer foci are very small. These diagnostic procedures are generally quite resource extensive and time consuming. With the advancement in optical technologies, it is becoming possible to have non-invasive and reliable optical detection methods whose results can be interpreted objectively. Therefore, new diagnostic procedures and tests are continually being developed for assessing the aggressiveness and stages of the disease. Raman spectroscopy by its very nature is capable of providing highly detailed chemical information about a tissue sample and is gaining a wide recognition as an objective method for the diagnosis of diseases in tissues. In the Stokes Raman scattering process, an incident laser light interacting with the molecules gets inelastically scattered, transferring some of its energy to the molecular vibrational excitations in the sample. Since biological tissue contains a large number of different molecules, its Raman spectrum, which contains detailed chemical, compositional and conformational information about the molecules in the tissue under investigation, will be rich in information content. A biolog-
228
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
ical tissue from an organ contains certain types of biomolecules in a certain molecular composition unique to the health condition of that organ. This unique composition of the biomolecules also has a unique set of vibrational states representing the vibrational motions of these biomolecules. Occurrence of any disease in an organ results in changes in the normal biochemical composition of the organ which is reflected in the vibrational spectrum of the tissue. Thus the Raman spectroscopy could be useful for detecting these changes, which may be able to predict the health condition of a tissue. Raman spectroscopy has been successfully tested in a variety of diseases in organs, like skin [7], mammary glands [8], brain [9], and gastrointestinal track. Recently, formalin-fixed paraffin preserved cervical tissue samples of normal, pre-malignant (cervical intraepithelial neoplasia), and invasive carcinoma have been investigated to detect biochemical changes accompanying cervical cancer progression [10]. Raman spectroscopy has also been used to study the differences between various prostatic adenocarcinoma cell lines [11], and for in vitro identification and grading of prostatic adenocarcinoma [12]. Raman spectral data in conjunction with PCA [13] have been used to determine the variations in the glycogen and nucleic acid contents in malignant and benign pathologies in prostatic tissues. The basic aim of the present study was to test the potential of Raman spectroscopy as a diagnostic tool for the detection of BE [14], PIN [15], and different Gleason scores of cancer in human prostate tissues. In this study, we have collected Raman spectra from different pathological regions of bulk human prostate tissue samples. We have carefully analyzed these spectra through visual inspection and found that the average spectrum for each pathological state contained different spectral characteristics strong enough to differentiate each pathological state from others. We have further analyzed the data using PCA and DFA [16] statistical procedures to determine different pathological groups in our data set. 2. Materials and methods In this study, ten prostatic bulk tissues were used (protocol #: 0705004871). These tissues were obtained from the archives of the Department of Pathology at Karmanos Cancer Institute in Detroit, MI. 2.1. Preparation of tissues for Raman spectroscopy Two parallel tissue sections (one 5 m and the other 10 m thick) were cut from paraffin-embedded tissue samples, using a microtome. The 5-m-thick tissue section was placed on a glass slide, dewaxed [17,18] by washing for 5 min each in xylene and pure ethanol, and used for hemotoxyilin and eosin (H&E) stain-
ing for subsequent pathological examination by an experienced pathologist. The 10-m-thick section was placed on a stainless steel substrate, dewaxed and dried in air for the Raman studies. All the tissues were processed at the University Pathology Research Services, Karmanos Cancer Institute. When the tissue was prepared for the Raman spectroscopy study, it was dewaxed by treating it again with xylene and ethanol baths. In this process, sometimes wax residuals are left in the tissue which can be detected through the presence of sharp wax bands at 1063, 1130, 1296, 1436, and 1465 cm−1 . However, if tissue is dewaxed properly then the wax residuals can be completely removed from the tissue [17]. The Raman spectra were recorded from unstained tissue sections using corresponding regions marked as BE, PIN, and different cancer categories on the parallel H&E stained tissue section (see Fig. 1). 2.2. Raman spectroscopic measurements We have used a Renishaw RM1000 Raman microscopespectrometer to record the Raman spectra from prostate tissues. The spectrometer is equipped with a 578 × 385 pixel CCD and a 785 nm diode laser. A 50× objective lens was used to focus the excitation laser beam to a spot size of ∼4 m × 30 m on the sample and collect the back-scattered light. The spectral resolution is ∼3.5 cm−1 . The laser power at the sample was about 3 mW. The spectra were recorded in the 500–1900 cm−1 range using an integration time of 20 s per spectrum, and three accumulations were averaged to improve the signal-to-noise ratio. The Raman spectra were recorded from the unstained tissue section. The appropriate regions on the unstained tissue section were identified with the help of the adjacent stained section with regions marked by a pathologist as BE, PIN, and cancer. The identification of these regions is done with an optical microscope to make sure the data were collected only from the marked regions. The Raman spectra were collected from different spots from each of the marked regions. The separation distance between each of the data collection spot is about 10 times the laser spot diameter. Each collected spectrum represents a datum from a physically different location on the tissue. Care was taken to ensure that the spots from which the spectra were collected are located well inside the appropriate regions marked by the pathologist. A total of 366 spectra were recorded from regions marked as BE, PIN, and cancer of different Gleason scores such as 6, 7, and 8. 3. Results and discussion After collecting the data, we have visually inspected each spectrum and discarded those which contain any trivial non-standard Raman features. Although these types of spectra are few; how-
Fig. 1. Stars represent the locations of the Raman measurements (a) tissue section used for the Raman measurement and (b) adjacent (H&E) stained tissue section used for pathological examination.
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
229
ever, mixing them with others can introduce errors in the results of statistical analysis. From the remaining good spectra we removed any spurious bands due to cosmic radiation. The noise was filtered using wavelets method [19] while the fluorescence background in the spectra was removed with minmax adaptive algorithm that requires no a priori knowledge of the spectra [20]. Finally, each spectrum was normalized using the highest intensity band in the spectrum. It is a well known that due to the morphological complexity in diseased tissues, misrepresentation of the observed morphological features happens particularly at the border regions of cancer with other pathologies (PIN, BE or cancer) leading to serious errors in the diagnosis. However, with Raman spectroscopy being an objective and quantitative optical method which can determine biochemical composition of tissue, the chances of misrepresentation are small. To find pathologically misrepresented regions of the tissues, we visually examined all the spectra in each of the pathological category. We found that spectra from each of the pathological categories have some distinct features, though in some categories the distinct and unique features are quite subtle and small. Interestingly, we found that the spectral features of some of the spectra collected from a region assigned to a particular pathology did not match with the unique spectral features of that pathological state but they matched with the spectral features of the pathological state assigned to the adjacent region. However, the observed mismatches in the visual inspection were not corrected in the data analysis using PCA and DFA. It is interesting to note that the data analysis did associate all the misrepresented spectra with pathological states to which their spectral features matched. The average spectra from BE, PIN, and cancer regions of bulk tissues are shown in Fig. 2. These spectra are typical of biological tissues exhibiting a combination of sharp and broad Raman bands arising from the molecular vibrations of proteins, nucleic acids, lipids, and other constituent molecules. In our samples we have not observed any bands which can be associated with wax. The low intensity broad bands observed at 1063 and 1129 cm−1 are close to the Raman bands of wax. However, they do not originate from wax, as the Raman bands of wax are normally very sharp and intense [17]. The most prominent sharp band around 1002 cm−1 is seen in all the cases having similar spectral features. This well known band is attributed to ring deformation mode associated
with aromatic amino acids; phenyl alanine and proline, and is often observed in the Raman spectra of proteins [9]. Careful examination of Fig. 2 shows that the average spectrum of each region (e.g. BE, PIN, and cancer) of the bulk tissues has some characteristic spectral features that differentiates it from the others. These different characteristic features basically highlight the fact that each region has different biochemical compositions unique to that region. In addition to the subtle differences in the spectral features, the differences in the BE, PIN, and cancer regions can also be seen in the intensity of the bands at 726, 759, 782, 1063 and 1097 cm−1 . Especially, the difference in the intensities is quite large for the band at 782 cm−1 in the PIN and cancer regions compared to the BE region. This mode represents breathing of U, C, T rings of DNA bases which can be interpreted as an enhancement of DNA content in PIN and cancer compared to the BE region. This observation is in agreement with the results of Teba et al. (2007) [21]. However, it should be emphasized that there is a considerable overlap in Raman vibrational frequencies from other biological molecules and each observed band cannot be uniquely assigned to a particular molecule. To understand all the spectral variations in terms of chemical composition/conformational changes, one needs to know the vibrational assignment for each of the observed Raman frequencies. Fig. 2 shows that most of the spectral changes occur in the region 700–1200 cm−1 of the Raman spectra. This region of a spectrum contains numerous bands with varying intensities. Fig. 3 shows an enlarged region covering the 630–850 cm−1 region of the spectrum to highlight the differences. In fact, the intensity ratios of such characteristic bands have often been used to interpret the relative molecular compositions in tissues [22,23]. There are two significant changes observed in the data which could be used to differentiate the prostate cancer tissue from BE and PIN: (i) we have observed a definite correlation in the variations of band intensities at 726, 759, 782, and 831 cm−1 . In order to quantify this correlation, we chose a reference band at 643 cm−1 whose maximum intensity did not change for all the three main pathological states. With reference to this band, variations in other bands are shown in Fig. 3. Table 1 lists the percentage variations in the bands at 726, 759, 782, and 831 cm−1 with respect to the reference band at 643 cm−1 . The ratios of 726/643 and 782/643 of cancer increased significantly compared to BE and PIN states. However, if we consider the bands at 759 and 782 cm−1 in BE and PIN states, the relative intensities were reversed. A strong variation in the intensity bands at 726 and 782 cm−1 for cancerous regions is an important observation since these bands are related to the breathing modes of DNA whose content increases significantly in the cancerous region due to cell
Fig. 2. Average Raman spectrum for BE, PIN and cancer (here the black, green, and red curves represent BE, PIN, and cancer, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
Fig. 3. Variations in Raman intensity and shift in certain bands for BE, PIN and cancer spectra (here the black, green, and red curves represent BE, PIN, and cancer, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
230
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
Table 1 Percentage increase in the peak intensities of selected Raman bands with respect to the intensity of the 643 cm−1 peak. Raman band (cm−1 )
Cancer
PIN
BE
726 759 782 831
39 40 110 141
7 15 52 114
5 35 32 130
Fig. 6. First three PC loadings spectra for BE, PIN, and cancer. The Raman shift for the main positive and negative bands are marked against each loading curve (the black solid curve represents PCL1, the red dash-dot curve represents PCL2, and the dotted blue curve represents PCL3). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
show the spectral ranges where the variations between different Gleason scores can easily be seen. 3.1. Statistical analysis of the data
Fig. 4. The average Raman spectrum for G6, G7, and G8 Gleason scores (Here the pink, green, and blue curves represent G6, G7, and G8, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
proliferation. (ii) We observe shift in the band positions of 726 and 782 cm−1 bands in BE state compared to PIN and cancer states (see Fig. 3). Although the shift is small (about 2.0 cm−1 ), we believe it is significant since it is observed consistently in all the spectra in this group and is reflected in the average spectrum shown in Fig. 3. We have also studied the spectra from different Gleason scores of cancer. In Fig. 4, we have shown the mean spectrum from each Gleason score. Raman spectra of Gleason scores 6, 7, and 8 show small and subtle differences in some of the bands. In Fig. 5, we
For the purpose of quantification of the variances of the data from BE, PIN, and cancer regions, we have performed statistical analysis of the data using widely used statistical techniques such as PCA and DFA. For the analysis of the bulk prostate data, we considered spectral range 600–1800 cm−1 , the “finger print” region containing the fundamental vibrations of molecules. This region of the vibrational spectrum is often used for identification of molecules and conformational changes. First, we applied PCA method to reduce the dimensionality of the original data set having 601 variables. The transformed data set consists of 23 eigenvectors capturing more than 94% of variance in the data. Fig. 6 shows the results for the principal component loadings (PCLi, where i = 1, 2, 3). Unlike a Raman spectrum, loading spectrum contains positive and negative bands and their corresponding frequencies can be correlated with the variations in the molecular composition through the molecular vibrational assignments. The PCL1 contains maximum variance within the data set and it also captures the trend of band variances. From the PCL1 graph, one can see the enhancement in the concentrations of DNA content (670 and 782 cm−1 ), proteins (726 cm−1 ), phospholipids (1330 and 1368 cm−1 ), and nucleic acids (1576 cm−1 ) [24]. While, PCL2 represents increase in proteins (726 cm−1 ), DNA (1068 and 1094 cm−1 ), fatty acids
Fig. 5. Two narrow spectral ranges where significant differences in the Raman spectra of G6, G7, and G8 are observed (here the pink, green, and blue curves represent G6, G7, and G8, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
231
Fig. 8. First three PC loadings spectra for Gleason scores 6, 7, and 8 (the black solid curve represents PCL1, the red dash-dot curve represents PCL2, and the dotted blue curve represents PCL3). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.) Fig. 7. Classification results from DFA for BE, PIN, and cancer (the black triangles, green squares, and red circles represent BE, PIN, and cancer (CA)). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
(1132 and 1658 cm−1 ), phospholipids (1454 cm−1 ), and decrease in collagen (1280 cm−1 ), tryptophan (1550 cm−1 ), and phenylalanine (1605 cm−1 ). PCL3 shows an increase in phenylalanine 1000 cm−1 and decrease in fatty acids (1064, 1132, 1297, and 1440 cm−1 ) [24]. In Fig. 3, we have shown the variation in the intensities of the bands at 726, 759, 782, and 831 cm−1 . In the PCL1 spectrum we see these bands represent increase in proteins (726 cm−1 ) and DNA (784 cm−1 ), while from the PCL2 spectrum, we see that the bands 759 cm−1 and 831 cm−1 represent decrease in tryptophan and amino acids, respectively. Although, it is difficult to assess the concentrations of various components such as nucleic acids and proteins in a tissue under investigation, the Raman spectra in conjunction with known vibrational assignments for molecules do permit qualitative inference of their relative concentrations. In order to determine the presence of different groups in the data, we have performed DFA of the data with input taken from the PCA calculations. The results are shown in Fig. 7. It is encouraging to note that even though there is a huge heterogeneity in the tissues, the DFA classified the spectral data as BE, PIN, and cancer into three very distinct groups. However, some of the data points of one region are found to overlap with the others, e.g. some of PIN data are mixed with cancer data. This may be due to the fact that both cancer and PIN exhibit similar type of luminal cells [21]. When the spectra are being collected, given the size of laser spot (∼4 m × 30 m) it is difficult to determine whether the laser hits on luminal or basal cells or both within the epithelial cells. Thus the measured data can be a mixture of both. The prediction of the pathological state of the tissues using Raman studies and DFA is compared with that of pathology, and the results are shown in Table 2. To test the validity of our predicted classifications, we have performed leave-one-out cross-validation. With this method the association of each datum with one of the known pathological states is determined while the remaining data
are used as the training data set. The results of cross-validation are compared with the classification results in Table 2. The prediction accuracy (Table 2) becomes slightly smaller for cancer when data are tested with cross-validation method due to the smaller data size available for the analysis. The results for the detection of cancer in terms of sensitivity and specificity using the cross-validated data are 94% and 82%, respectively. The higher values of sensitivity and specificity are encouraging, which may guide in-depth future studies. The cancer tissues we studied contain Gleason scores 6, 7, and 8 only. We have performed the PC loadings of the Gleason score data and the PC loading spectra are shown in Fig. 8. PCL1 clearly shows decrease in the fatty acids contents (1064, 1132 and 1296 cm−1 ) and increase in other components that give rise to the bands at 1206, 1282, 1352 and 1396 cm−1 in the range from 1180 to 1400 cm−1 consistent with PCL3 as shown in Fig. 6. The energy of various positive and negative bands of PCL1 is marked against them. The spectra of PCL2 and PCL3 are also shown in the graph. The DFA clearly separates Gleason scores of 6, 7, and 8 into three distinct groups [25] as shown in Fig. 9. The overlapping in different Gleason scores is expected, since Gleason score is a combination of two Gleason grades, i.e. certain spectra may be common to one of the Gleason scores. The individual accuracies of cross-validation in Gleason scores 6, 7, and 8 are more than 81% as shown in Table 3. This study
Table 2 Comparison of Raman diagnostic results with pathologic diagnosis. Raman diagnosis
BE PIN Cancer
Pathological diagnosis BE
PIN
Cancer
Total
Original
Crossvalidation
123 2 1
11 46 5
9 0 169
143 48 175
86.0% 95.8% 96.6%
78.3% 95.8% 94.3%
Fig. 9. Classification results from DFA for Gleason scores 6, 7, and 8 (here the pink circles, green squares, and blue triangles represent G6, G7, and G8, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
232
S. Devpura et al. / Vibrational Spectroscopy 53 (2010) 227–232
Table 3 Comparison of Raman diagnostic results of Gleason scores with pathologic diagnosis. Raman diagnosis
Cancer G6 Cancer G7 Cancer G8
Pathological diagnosis Cancer G6
Cancer G7
Cancer G8
Total
Original
Cross-validation
60 0 0
3 59 1
0 1 51
63 60 52
95.2% 98.3% 98.1%
81.0% 90.0% 86.5%
demonstrates that Raman spectroscopy can be used to differentiate different Gleason gradings in cancers. 4. Conclusions In this study we have investigated BE, PIN, and cancer regions of dewaxed human prostate tissues using Raman spectroscopy. The bulk tissues for this study were obtained from prostate cancer patients who underwent radical prostatectomy. We found that Raman spectroscopy can detect cancerous regions in bulk tissues in the presence of PIN and BE regions. Investigation of the Raman data shows that the average spectra of BE, PIN, and cancer regions have distinguishable unique Raman spectral features highlighting the fact that the biochemical composition of the tissue region changes when its pathological state changes. Although taking into consideration the variations in the spectral features over the full range (600–1800 cm−1 ) increases the discrimination ability of the analysis procedure; the bands at 726, 759, 782, and 831 cm−1 are also found to be key contributors for distinguishing the prostate cancer from PIN and BE. We also found that cancerous regions having different Gleason scores (Gleason scores 6, 7, and 8) can be differentiated, suggesting that Raman spectroscopy can detect different degrees of aggressiveness of the prostate cancer in tissues. Acknowledgement We would like to acknowledge the support from the Richard J. Barber Foundation for Interdisciplinary Research. References [1] A. Jemal, R. Siegel, E. Ward, Y. Hao, J. Xu, M.J. Thun, Cancer J. Clin. 59 (2009) 225–249, doi:10.3322/caac.20006. [2] A.W. Partin, M.W. Kattan, E.N. Subong, P.C. Walsh, K.J. Wojno, L.E. Oesterling, P.T. Scardino, J.D. Pearson, J. Am. Med. Assoc. 277 (1997) 1445–1451. [3] P.A. Humphrey, Modern Pathol. 17 (2004) 292–306, doi:10.1038/ modpathol.3800054.
[4] W.C. Allsbrook, K.A. Mangold, M.H. Johnson, R.B. Lane, C.G. Lane, J.I. Epstein, Hum. Pathol. 32 (1) (2001) 81–88, doi:10.1053/hupa.2001.21135. [5] G.L. Andriole, T.L. Bullock, J.S. Belani, E. Traxel, Y. Yan, D.G. Bostwick, P.A. Humphrey, Urology 70 (Suppl. 6A) (2007) 22–26, doi:10.1016/j.urology.2007.06.1128. [6] H. Hricak, P.T. Scardino, R.H. Reznek, Prostate Cancer, Cambridge University Press, New York, 2009, pp. 1–6. [7] P.J. Caspers, G.W. Lucassen, R. Wolthuis, H.A. Bruining, G.J. Puppels, Biospectroscopy 4 (1998) S31–S39. [8] K.E. Shafer-Peltier, A.S. Haka, M. Fitzmaurice, J. Crowe, J. Myles, R.R. Dasari, M.S. Feld, J. Raman Spectrosc. 33 (2002) 552–563, doi:10.1002/jrs.877. [9] C. Krafft, S.B. Sobottka, G. Schackert, R. Salzer, Analyst 130 (2005) 1070–1077, doi:10.1039/b419232j. [10] F.M. Lyng, E.Ó. Faoláin, J. Conroy, A.D. Meade, P. Knief, B. Duffy, M.B. Hunter, J.M. Byrne, P. Kelehan, H.J. Byrne, Exp. Mol. Pathol. 82 (2007) 121–129, doi:10.1016/j.yexmp.2007.01.001. [11] P. Crow, B. Barrass, C.A. Kendall, M. Hart-Prieto, M. Wright, R. Persad, N. Stone, Br. J. Cancer 92 (2005) 2166–2170, doi:10.1038/sj.bjc.6602638. [12] P. Crow, N. Stone, C.A. Kendall, J.S. Uff, J.A.M. Farmer, H. Barr, M.P.J. Wright, Br. J. Cancer 89 (2003) 106–108, doi:10.1038/sj.bjc.6601059. [13] I.T. Jolliffe, Principal Components Analysis, 2nd ed., Springer-Verlag, New York, 2002. [14] N. Stone, C. Kendall, N. Shepherd, P. Crow, H.J. Barr, Raman Spectrosc. 33 (2002) 564–573, doi:10.1002/jrs.882. [15] D.W. Keetch, P. Humphrey, D. Stahl, D.S. Smith, W.J.J. Catalona, Urology 154 (2) (1995) 347–351. [16] W.R. Klecka, Discriminant Analysis, Sage Publications, Newbury Park, 1980. [17] E.Ó. Faoláin, M.B. Hunter, J.M. Byrne, P. Kelehan, H.A. Lambkin, H.J. Byrne, F.M. Lyng, J. Histochem. Cytochem. 53 (1) (2005) 121–129, doi:10.1369/jhc.4A6536.2005. [18] E.Ó. Faoláin, M.B. Hunter, J.M. Byrne, P. Kelehan, M. Mcnamara, H.J. Byrne, F.M. Lyng, Vib. Spectrosc. 38 (2005) 121–127, doi:10.1016/j.vibspec.2005.02.013. [19] F. Keinert, Wavelets and Multiwavelets, Chapman and Hall/CRC Press, Boca Raton, FL, 2004. [20] A. Cao, A.K. Pandya, G.K. Serhatkulu, R.E. Weber, H. Dai, J.S. Thakur, V.M. Naik, R. Naik, G.W. Auner, R.J. Rabah, Raman Spectrosc. 38 (2007) 1199–1205, doi:10.1002/jrs.1753. [21] F. Teba, R. Martin, V. Gomez, L.M. Herranz, L. Santamaria, Image Anal. Stereol. 26 (2007) 93–99. [22] G. Shetty, C. Kendall, N. Shepherd, N. Stone, H. Barr, Br. J. Cancer 94 (2006) 1460–1464, doi:10.1038/sj.bjc.6603102. [23] D. Pappas, B.W. Smith, J.D. Winefordner, Talanta 51 (2000) 131–144. [24] Z. Movasaghi, S. Rehman, I.U. Rehman, Appl. Spectrosc. 42 (2007) 493–541, doi:10.1080/05704920701551530. [25] M.J. Baker, E. Gazi, M.D. Brown, J.H. Shanks, P. Gardner, N.W. Clarke, Br. J. Cancer 99 (2008) 1859–1866, doi:10.1038/sj.bjc.6604753.