Raman systems have tremendous potential as adjunct devices for endoscopes to improve the identification of early colon cancers. However, the traditional low frequency (LF) measurement range has several obstacles that make it challenging to develop a routine clinical tool. An alternative is to use high frequency (HF) range. To test this idea Raman spectra were obtained in both the LF and HF ranges from the same colon lesions. Multivariate analyses predicted the pathology with high sensitivity and specificity for both the LF and HF data sets. This suggests that Raman systems that measure HF spectra, and are simpler to adopt into the clinic, could be used in vivo to improve the identification of neoplastic lesions.
© 2013 OSA
Like all cancers, the earlier colon cancers are found the better the prognosis for the patients . Several advances have taken place in endoscopic imaging technologies that improve the ability of clinicians to localize small or occult lesions [2,3]. However only a fraction of the suspicious sites identified during a colonoscopy are neoplastic lesions, leading to a significant number of false positives, and potential early cancers at some sites are missed because they were considered benign and not requiring a biopsy by the clinician . To address these deficiencies, reliable, highly specific, optical techniques are required which function as optical biopsy tools to objectively classify lesions in vivo [4,5].
Research on the use of Raman emissions in the low frequency range (0 to 1800 cm−1) for differentiating between normal and neoplastic tissue in vivo, has significantly increased over the last decade [5–10]. Raman is a powerful analytical technique, but the inherently weak emission had prohibited its use as a fast medical diagnostic method until relatively recent advances in lasers, spectrometers, detectors and optical fibers made it possible. Despite these technological improvements, only a device that differentiates between skin lesions has been approved as a routine clinical tool . For endoscopic applications the technical challenges in collecting good quality Raman spectra are much greater. Furthermore Raman has to be combined with other fast, low specificity, optical modalities like white light, narrow band, or autofluorescence video imaging . Here the clinician uses one or more video modality to locate suspicious tissue sites, and then collects point Raman spectra of these sites with a fiber optic catheter passed down the instrument channel of the endoscope. These spectra can then be used to predict the tissue pathology in real time. Raman systems have been tested for this purpose on several organs including the colon, larynx, oral cavity, cervix, stomach, and skin [5–10].
The excitation wavelength of choice for clinical Raman systems is 785 nm since it penetrates deeper into tissue, stimulates less emission in the fiber optic catheter [12,13], and less tissue autofluorescence than visible wavelengths. However, these confounding emissions remain problematic because both can be strong in the low frequency range (LF) which coincides with the Raman emission that is traditionally measured from tissue. Simply subtracting off the contribution to the spectrum from the fiber is not a good solution since it is difficult to reliably remove. Another approach is to reduce the fiber emission by elaborate and expensive optical filters placed at the distal end of the fiber optic catheter . The drawback is that optically filtered endoscopic catheters have to be reprocessed for reuse after each procedure multiple times to make them economically viable. Unfortunately the long, narrow and flexible catheters can be damaged during the procedure or the subsequent reprocessing, significantly reducing the average number of reuse cycles. Even if these expensive catheters could be made to survive multiple procedures, the high tissue autofluorescence in the LF range remains and cannot be optically filtered out, which results in a reduction in the diagnostic accuracy.
An alternative to measuring LF Raman is to measure emission from tissue in a range that has substantially less fiber optic noise and tissue autofluorescence. This would enable the use of fiber optic catheters without expensive optical filters, perhaps even single use products. High frequency (HF) Raman measurements have these desired properties; with the right choice of fiber, and although HF Raman emissions contain fewer spectral bands, tests on several organs including the colon have been very promising [13–17]. To further investigate this possibility a comparison study was conducted between LF and HF Raman emissions obtained from the same ex vivo colonic tissue sites to determine the sensitivity and specificity of each range at predicting the tissue pathology. This study was approved by the Research Ethics Board of the University of British Columbia (Certificate #: H08-02693).
2. Samples and methods
For this study a total of 47 colon tissue samples were collected from 18 patients. Excised tissue was collected from 8 patients during surgery to remove a previously identified malignant lesion. Samples were obtained from the lesion itself, and from the surrounding tissue visually free of disease. For some sites two tissue fragments were obtained, and these were treated separately giving a total of 11 lesion and 9 normal samples. The average volume of the samples was approximately 10 mm3. All excised lesion samples were classified by histology as adenocarcinomas, and those from the surrounding tissue were normal. The remaining samples were biopsies obtained from 10 polyps (with matched normal epithelium) during a routine colonoscopy. Eight polyps were classified by histological evaluation as adenomatous (with varying grades of dysplasia), and two were invasive adenocarcinomas. Two tissue fragments were obtained from 4 of the polyps and 3 of the normal epithelium sites and these were treated separately giving a total of 14 polyp and 13 normal samples. The average volume of a biopsy sample used for Raman measurements was approximately 2 mm3. All samples were immediately placed in saline (@ 4°C) after collection and the Raman spectra measured within 30 minutes. Each sample was placed on an aluminum foil covered glass microscope slide, placed under the fiber optic catheter fixed above the sample, and 10 spectra obtained from the same site with a 1 second integration time. In addition to the colon samples, some preliminary in vivo tests of the Raman system were done on the palm skin of a volunteer.
2.2. Raman system
The Raman system used to take measurements was described previously , and was modeled on similar systems used for other organs by our group [10,15,16]. The excitation source used was a 785 nm diode laser. Emission was analyzed with a spectrograph incorporating a manually tunable grating and a charge coupled device (CCD) detector. One of two detachable fiber optic catheters was used to deliver excitation light to the sample and collect emission from it. These catheters contained ultra low OH impurity fibers, and gold coated excitation fibers. One catheter incorporated optical filters at the distal end to filter out laser noise, fiber emission, and to sharply attenuate all collected light with wavelengths ≤ 820 nm (≤ 540 cm−1 relative to 785nm excitation) . This catheter attached at its proximal end to a second set of optical filters with similar transmission characteristics which further reduce the unwanted emissions. The second catheter was identical to the first except with no filters at the distal end. A fixed catheter to tissue distance of ≈7.5 mm was used which generates a tissue spot size of ≈3.5 mm in diameter. Together the spectrograph, CCD and optical filters allow reliable spectra to be obtained from: 540 to 1800 cm−1 (LF) and from 2050 to 3100 cm−1 (HF) at a spectral resolution of ≈10 cm−1. The maximum excitation power was 150 mW. Custom designed software was used to subtract the fluorescence background in real time using a modified 5th order polynomial fit .
The raw Raman spectra from the colon samples had to be standardized before they can be analyzed or compared with published spectra. This was accomplished with the following steps: removing the ambient background, averaging the 10 spectra from each site, calibrating, smoothing, fluorescence removal, and normalizing to reduce the effect of intensity variations from different tissue sites with the same pathology. The normalization was accomplished by summing the area under each curve and dividing each variable in the smoothed spectrum by this sum. Simple Raman peak ratios were calculated using two peaks from each range. These peaks were selected on the basis of a student’s t-test which indicated which peaks were the most significantly different between normal and diseased tissue. Commercial software was used for the multivariate statistical analyses (Statistica 10.0, StatSoft Inc. Tulsa, OK) where a spectral range is used rather than single Raman peaks (univariate). Principal components (PCs) for all the spectra in the data set were computed to reduce the number of variables for the multivariate analyses. Student’s t-tests were used on the PCs that accounted for 0.1% or more of the variance to determine those most significant at separating spectra into two pathology groups: normal and diseased tissue. A linear discrimination analysis (LDA) with leave-one-out cross validation was used on the most significant PCs. To avoid over fitting the data, the number of PCs used in the LDA were limited to one third of the total number of cases of the smallest subgroup. The excised tissue and the biopsy samples were analyzed separately.
The initial in vivo test spectra of palm skin in the LF taken with the optically filtered and unfiltered catheters are shown in Fig. 1(a) and 1(c). The difference spectrum was dominated by 3 broad peaks at 590, 780, and 1030 cm−1. Below 500 cm−1 the emission collected with the unfiltered catheter saturated the detector. In the HF range the skin spectra were very similar for both catheters apart from a small decrease in the intensity below 2400 cm−1 and above 2850 cm−1 with the optically filtered catheter as shown in Fig. 1(b). Removing the background fluorescence results in very similar Raman spectra obtained with either catheter as shown in Fig. 1(d).
For the more reflective colon tissue, the LF emission at 590, 780, and 1030 cm−1 saturated the detector with the optically unfiltered catheter (not shown). Even with the optically filtered catheter, these peaks were still clearly present in all excised tissue and biopsy samples as shown in Fig. (2) , and very strong in a few. There was a substantial fluorescence contribution to the LF spectra for all samples as shown in Fig. 2(a) and 2(b). Generally the Raman emission intensity from the biopsies was weaker but the fluorescence was stronger than from the excised tissue. Furthermore the fluorescence was stronger in diseased excised tissue than normal tissue, but the reverse of this for the biopsy samples. After the fluorescence contribution was subtracted Raman peaks were clearly visible around 1340, 1450, 1650 and 1730 cm−1 as well as a number of smaller peaks most notably around 1000, 1150 and 1550 cm−1 as shown in Fig. 2(c) and 2(d). There were also differences in the shape of the LF Raman spectra for excised and biopsy tissue.
For the HF range, a comparison of spectra taken of colon tissue with the optically filtered and unfiltered catheters (not shown) demonstrated similar differences below 2400 cm−1 and at around 2850 cm−1 as for the palm skin measurements as shown in Fig. 1(b) and 1(d). Comparisons in the HF spectra obtained from diseased and normal tissues with the optically filtered catheter are shown in Fig. 3(a) and 3(c) for excised tissues and Fig. 3(b) and 3(d) for biopsy tissues. The HF tissue fluorescence intensity drops to less than half that of the LF range, and was stronger for the biopsy samples than the excised tissue as in the LF case as shown in Fig. 3(b). Furthermore the fluorescence was again stronger in diseased excised tissue than normal tissue, but the reverse of this for the biopsy samples. Raman emissions were observed in the raw data clustered around 2900 cm−1, and low intensity broad peak centered at 2150 cm−1. Between 2200 and 2800 cm−1 of the emission range there were a number of weak peaks that did not seem to vary for different samples. Above 3000 cm−1 the intensity rises steeply. Figures 3(c) and 3(d) show the Raman range from 2825 to 3025 cm−1 after fluorescence removal, differences in the shape of the spectra between excised and biopsy tissue are seen.
Due to saturation of most LF spectra obtained with the optically unfiltered catheter, statistical analyses were carried out on spectra obtained with the optically filtered catheter only. Even with this catheter, 4 spectra out of the 47 had to be rejected because of partial detector saturation. The rejected spectra were from 3 excised tissue samples (2 malignant and one normal), and from one adenomatous polyp. Simple Raman peak ratios for the remaining LF, and all the HF, spectra are shown in Fig. 4 . Note: different Raman peaks were used in the ratio calculation for the excised and biopsy samples. There were various degrees of clustering in the Raman peak ratios. Ratios calculated from the peaks in the LF spectra of excised tissue produced the best cluster separation and those from the peaks in the HF spectra of biopsy tissue produced the worst.
The results of the multivariate analyses on the LF spectra are shown in Fig. 5 . Figures 5(a) and 5(b) are two dimensional (2D) scatter plots of two principal components that were highly correlated with tissue pathology for the excised tissue and biopsy samples respectively. Figures 5(c) and 5(d) show the posterior probabilities derived from the leave-one-out LDA for the excised and biopsy tissue respectively. From these plots it can be seen that all lesions in the LF range were identified at the cost of one false positive (88% specificity) for the excised tissue, and three false positives (77% specificity) for the biopsies.
The results of the multivariate analyses on the HF spectra are shown in Fig. 6 . Figures 6(a) and 6(b) are 2D scatter plots of the two principal components that are the most highly correlated with tissue pathology for the excised tissue and biopsy samples respectively. The pathology classes were again clearly clustered in two groups, and this clustering appears better for the excised tissue samples. Figures 6(c) and 6(d) show the posterior probabilities derived from the leave-one-out LDA for the excised and biopsy tissue respectively. From these plots it can be seen that all lesions in the HF range, all lesions were identified at the cost of one false positive (89% specificity) for the excised tissue, and two false positives (85% specificity) for the biopsies.
The initial test spectra obtained with the optically filtered and unfiltered catheters from the human palm skin were consistent with what others had found for various lipids, proteins, amino acids, and water molecules [10,14,19], contaminated by various amounts of fiber emission [12,13]. In the LF range the fiber emission was large with the optically unfiltered catheter despite the use of ultra low impurity materials. Catheters used to measure spectra in this range have to include expensive optical filters at their distal end to reduce fiber emission, avert detector saturation and increase the signal to noise ratio. In contrast, the HF Raman spectra showed only small differences in spectra obtained with the two different catheters. It is unclear what causes the divergence in the spectra below 2400 cm−1 as shown in Fig. 1(b). The difference cannot be explained by greater fiber fluorescence tail since this is inconsistent with the LF spectra, although it could be caused by a broad fiber emission peak. In contrast the difference above 2850 cm−1 was most likely due to a drop in transmission of the catheter filters in this range (from data provided by filter manufacturer).
The LF Raman spectra from the colon samples were also consistent with what others had found for various lipids, proteins, amino acids, and water molecules [5,19]. There was a significant fluorescence contribution to the spectra, and despite extensive optical filtering, fiber emission peaks were clearly evident as well. From this, it can be deduced that the measured fluorescence is in part coming from the fiber catheter in addition to autofluorescence from the tissue. The LF Raman spectra of biopsy tissues were generally less intense, and riding on a higher fluorescence background compared to the excised tissues. Similar differences have been seen before between small ex vivo colon samples and in vivo measurements . The malignant excised tissue shows increases in Raman peaks around 1340 (lipids), and 1450 (lipids), and 1650 cm−1 (amide I, and H20) and decreases at 1150 (proteins), 1540 (amino acids), and 1735 cm−1 (lipids) compared to the normal excised tissue spectra. The changes in the spectra from the biopsy tissue seem to be largely the reverse of this apart from the 1735 cm−1 peak. These differences are probably caused by variations in sample composition, where the excised tissue has more contribution from deeper tissue layers. It should be noted that in vivo spectra will likely be more similar to the excised tissue spectra shown here.
For the HF range the spectra were similar to those obtained from other organs [14–16]. The fluorescence was still significant, but it was about half of that found in the LF range. It can be deduced from these data that only a small fraction of this fluorescence was coming from the delivery and collection fibers consistent with the trend in the LF spectra. The biopsy Raman spectra were again generally less intense, and riding on a higher fluorescence background compared to the excised tissue. The dominant Raman emissions near 2900 cm−1 were mainly due to a combination of lipid (C-H) peaks at 2853 and 2866 cm−1 and generic protein vibrations at 2930 cm−1 . The low intensity broad peak centered at 2150 cm−1 and the steep intensity increase above 3000 cm−1 were assigned to water molecule vibrational modes . The weak peaks occurring between 2200 and 2800 cm−1 may be due to small CCD etaloning effects . For the malignant excised tissue the intensity of the lipid peaks at 2853 and 2866 cm−1 dropped, but the intensity of the generic protein peak at 2930 cm−1 increased compared to normal samples. In contrast it is hard to see really significant changes in the HF spectra from the biopsy samples. These differences probably occur because of variations in sample composition as seen in the LF spectra.
The peak ratios calculated from the spectra were surprisingly good at separating the tissue into two pathology groups. For the LF excised tissue, one can clearly see the 1340 and 1735 cm−1 peaks change significantly in the average spectra shown in Fig. 2(c), and the ratio of these peaks clearly separates normal and malignant tissue as shown in Fig. 4(a). For the LF biopsy tissue, the 1340 cm−1 peak no longer shows significant change between normal and diseased tissue as shown in Fig. 2(d), instead the peak at 1445 cm−1 was used with the 1735 cm−1 peak to calculate the ratio (Fig. 4(b)). However in this case the two pathology groups are not as well separated. For the HF spectra using a simple peak ratio to predict the pathology was not so reliable (Fig. 4(c) and 4(d) generating lower diagnostic accuracies. This is not too surprising since the relative changes in peak heights between normal and diseased tissue as indicated by the HF average spectra shown in Fig. 3(c) and 3(d) were less than those found in the LF.
The multivariate analyses produced much better diagnostic statistics for separating normal and diseased tissue compared to the simple peak ratios. Furthermore the HF spectra were slightly better at predicting the pathology than the LF spectra despite there being fewer Raman peaks in the HF measurement range. This result was not so surprising for the HF spectra from excised tissue, since one can clearly see changes in the shape of the average spectra for different pathologies. However the HF spectra from the biopsy samples were also good at predicting the pathology, although in this case the average spectra of diseased and normal tissue were quite similar. This highlights the power of multivariate analyses where multiple small changes in a spectral range can be diagnostically very significant.
The main objective of this study was to test whether HF Raman spectra can be used as a diagnostic tool to discriminate between different colon tissue pathologies or not. The Raman peak ratios calculated from the HF spectra were good at predicting the pathology but worse than ratios obtained from the LF spectra. In contrast, multivariate analyses indicated that HF spectra were very good at predicting the pathology and slightly better than the LF spectra. Although the spectra analyzed in this study were obtained with catheters incorporating expensive optical filtering at the distal end, it was shown here, and by others , that such filtering can be omitted for HF spectroscopy with the right choice of fiber. Utilizing optically unfiltered catheters to obtain HF Raman spectra has several key advantages mainly around the design, reliability, cost and longevity of the catheters. More work is needed to follow up on this pilot study and to test the conclusions drawn, but the evidence so far points to HF Raman spectroscopy as having great potential for improving the in vivo detection of early neoplastic lesions of the colon.
The authors would like to thank Calum MacAulay, Demissie Desta, Hanna Pawluk, Ruby Flores, Jagoda Korbelik and Sylvia Lam. This work was supported by the Canadian Institutes of Health Research (Grant # CST-85477).
References and links
1. E. P. Whitlock, J. S. Lin, E. Liles, T. L. Beil, and R. Fu, “Screening for colorectal cancer: a targeted, updated systematic review for the U.S. Preventive Services Task Force,” Ann. Intern. Med. 149(9), 638–658 (2008). [PubMed]
2. A. Adler, J. Aschenbeck, T. Yenerim, M. Mayr, A. Aminalai, R. Drossel, A. Schröder, M. Scheel, B. Wiedenmann, and T. Rösch, “Narrow-band versus white-light high definition television endoscopic imaging for screening colonoscopy: a prospective randomized trial,” Gastroenterology 136(2), 410–416 quiz 715 (2009). [CrossRef] [PubMed]
5. A. Molckovsky, L. M. Song, M. G. Shim, N. E. Marcon, and B. C. Wilson, “Diagnostic potential of near-infrared Raman spectroscopy in the colon: differentiating adenomatous from hyperplastic polyps,” Gastrointest. Endosc. 57(3), 396–402 (2003). [CrossRef] [PubMed]
6. D. P. Lau, Z. Huang, H. Lui, D. W. Anderson, K. Berean, M. D. Morrison, L. Shen, and H. Zeng, “Raman spectroscopy for optical diagnosis in the larynx: Preliminary findings,” Lasers Surg. Med. 37(3), 192–200 (2005). [CrossRef] [PubMed]
7. R. Malini, K. Venkatakrishna, J. Kurien, K. M. Pai, L. Rao, V. B. Kartha, and C. M. Krishna, “Discrimination of normal, inflammatory, premalignant, and malignant oral tissue: a Raman spectroscopy study,” Biopolymers 81(3), 179–193 (2006). [CrossRef] [PubMed]
8. A. Robichaux-Viehoever, E. Kanter, H. Shappell, D. Billheimer, H. Jones 3rd, and A. Mahadevan-Jansen, “Characterization of Raman spectra measured in vivo for the detection of cervical dysplasia,” Appl. Spectrosc. 61(9), 986–993 (2007). [CrossRef] [PubMed]
9. S. K. Teh, W. Zheng, K. Y. Ho, M. Teh, K. G. Yeoh, and Z. Huang, “Diagnostic potential of near-infrared Raman spectroscopy in the stomach: differentiating dysplasia from normal tissue,” Br. J. Cancer 98(2), 457–465 (2008). [CrossRef] [PubMed]
11. H. Zeng, Y. S. Fawzy, M. A. Short, M. Tercelj, A. McWilliams, M. Petek, B. Palcic, J. Zhao, H. Lui, and S. Lam, “Combining field imaging endoscopy with point analysis spectroscopy for improving early lung cancer detection,” in Proceedings of IEEE Conference on Engineering in Medicine and Biology (Institute of Electrical and Electronics Engineers, New York, 2008), pp.1849–1850.
12. M. G. Shim, B. C. Wilson, E. C. Marple, and M. Wach, “Study of fiber optic probes for in vivo medical Raman spectroscopy,” Appl. Spectrosc. 53(6), 619–627 (1999). [CrossRef]
13. L. F. Santos, R. Wolthuis, S. Koljenović, R. M. Almeida, and G. J. Puppels, “Fiber-optic probes for in vivo Raman spectroscopy in the high-wavenumber region,” Anal. Chem. 77(20), 6747–6752 (2005). [CrossRef] [PubMed]
14. A. Nijssen, K. Marquelin, L. F. Santos, P. J. Caspers, T. C. Bakker Schut, J. C. den Hollander, M. H. Neumann, and G. J. Puppels, “Discriminating basal cell carcinoma from perilesional skin using high wavenumber Raman spectroscopy,” J. Biomed. Opt. 12(3), 034004 (2007). [CrossRef]
15. M. A. Short, S. Lam, A. McWilliams, J. Zhao, H. Lui, and H. Zeng, “Development and preliminary results of an endoscopic Raman probe for potential in vivo diagnosis of lung cancers,” Opt. Lett. 33(7), 711–713 (2008). [CrossRef] [PubMed]
16. M. A. Short, S. Lam, A. M. McWilliams, D. N. Ionescu, and H. Zeng, “Using laser Raman spectroscopy to reduce false positives of autofluorescence bronchoscopies: A pilot study,” J. Thorac. Oncol. 6(7), 1206–1214 (2011). [CrossRef] [PubMed]
17. M. A. Short, S. F. Lam, D. A. Owen, C. Macaulay, I. T. Tai, and H. Zeng, “Development of an endoscopic Raman spectroscopy system for the improving detection of early colonic neoplasias,” Gastrointest. Endosc. 73(4), AB371–AB372 (2011). [CrossRef]
18. J. Zhao, H. Lui, D. I. McLean, and H. Zeng, “Automated Autofluorescence Background Subtraction Algorithm for Biomedical Raman Spectroscopy,” Appl. Spectrosc. 61(11), 1225–1232 (2007). [CrossRef] [PubMed]
19. Z. Movasaghi, S. Rehman, and I. U. Rehman, “Raman spectroscopy of biological tissues,” Appl. Spectrosc. Rev. 42(5), 493–541 (2007). [CrossRef]