White light bronchoscopy (WLB) is the standard method for the localization of central lung cancers, but its limitation is poor sensitivity for the detection of early lung cancers, severe dysplasia, and carcinoma in situ (CIS).1 Autofluorescence bronchoscopy (AFB) on the basis of tissue autofluorescence emission differences between normal and malignant tissue2–5 has been shown to be capable of significantly improving the detection of CIS and other preneoplastic lesions.6 The addition of AFB to WLB has been found to increase CIS detection sensitivity several-fold,7,8 although not all studies have found AFB to result in a statistically significant increase in sensitivity.9 However, AFB is associated with a large increase in the number of benign conditions appearing suspicious, thereby reducing specificity, primarily because many other conditions can result in erythema, which presents a strong signal in AFB imaging.
It has been postulated that quantitative autofluorescence and/or reflectance spectroscopic measurements could be used to improve the detection and classification accuracy of suspected lesions identified during tissue imaging, and could thereby reduce the number of unnecessary biopsies while maintaining as much as possible the sensitivity of AFB. In other tissues, investigators have identified intrinsic differences in optical and vascular properties between malignant and nonmalignant tissues, and have related these directly to the changes in tissue physiology and morphology that occur during cancer progression.10,11 Spectroscopic measurements and analyses have been performed on lesions of the lung and have also shown potential for improving the specificity of autofluorescence imaging.12,13 Other studies found that properties of the vascular system, such as erythema (measured as the volume fraction of the tissue composed of blood) or tissue hypoxia (desaturated hemoglobin fraction), were most significant.14,15 The blood volume fraction in particular is expected to be useful for premalignant lesions, especially for angiogenic squamous dysplasia, which is associated with a vascularized epithelium and an increase in mucosal blood content.16
Hypoxia is also a promising candidate feature, being common in tumors resulting from increased oxygen consumption through an elevated metabolic rate, increased microvessel separation, a reduced quality of the microcirculation, and a reduced blood flow rate because of increased vessel resistance and increased blood viscosity.17,18 Hypoxia could also be a useful measure for premalignant lesions if malignancy-induced angiogenesis were a response to microenvironmental hypoxia. However, this seems unlikely for the thin epithelium of the lung, and other factors, such as alterations of oncogenes,19,20 may be better candidate explanations. One differential path-length spectroscopy study of the lung15 found the oxygen-desaturated hemoglobin fraction to be elevated in cancer, but no difference was observed in metaplasia or mild dysplasia versus normal tissue.
This study was performed to identify spectrally estimable tissue properties with the potential for specificity improvements by quantitative spectroscopy in the lung using a prototype WLB and AFB imaging device with an integrated noncontact spectroscopy system.21 The device was evaluated earlier in a small pilot study, where it showed potential for obtaining useful spectral measurements during bronchscopy.22 Spectral analysis algorithms for estimating tissue optical and microvascular properties had been developed in other related studies,23,24 and these algorithms were applied to visually suspicious biopsy specimens taken in this study and investigated for their ability to discriminate high-grade (severe dysplasia—cancer) from lower-grade histologies. It was hypothesized that the area under the receiver operating characteristic (ROC) curves for spectrally estimated tissue parameters of visually suspicious lesions would exceed the area expected by chance [area under the curve (AUC): 0.5]. Classification algorithms developed in this investigative phase were intended to be evaluated prospectively in a subsequent study for hypothesis testing of the impact on clinical performance.
This was an exploratory clinical investigation conducted at 4 institutions located in Canada, Russia, Slovenia, and the United Kingdom. Patients were enrolled between January 2006 and February 2007. The objective of this study was to gather data by noncontact spectral measurement, and from these data to identify discriminating parameters relating to cancer progression, with the final aim of improving the specificity of autofluorescence imaging. The protocol was approved by local ethics committees and written informed consent was obtained from all patients. To be included in this study, patients were required to have been referred for bronchoscopy to confirm the presence or absence of lung cancer as part of a diagnostic workup. Patients were selected on the basis of positive results of other tests (such as chest x-ray, computed tomography, automated quantitative cytometry, or cytology), or because of the presence of suspicious symptoms. Enrollment was not permitted if the patients had unstable angina or were otherwise unable to undergo bronchoscopy with biopsy. Patients were to be enrolled at the maximal rate from each site during the study period, up to a combined total of 550. This sample size had been projected to yield up to 2000 biopsy specimens, which was expected to provide 100 to 200 specimens for lesion types of interest (eg, early cancer, CIS, dysplasia) on the assumption of a 30% prevalence rate for early cancer cases. This quantity was heuristically considered sufficient to prevent overfitting of the originally envisioned discriminant function for analyzing spectral signatures—a method that was superseded by the more parsimonious modeling approach developed in the pilot study from this protocol24 and used in the analysis here.
White light (WL) and autofluorescence imaging and spectral acquisition were performed with the ClearVu Elite system (Perceptronix). This system uses a Xenon arc light source to provide either white light (400 to 700 nm, 10 mW) for WLB imaging and reflectance spectral measurements, or strong blue light (400 to 460 nm, >50 mW) with weak near-infrared (NIR) light (720 to 800 nm, 4 mW) for AFB imaging and fluorescence spectral measurements. The illumination fiber bundle of the endoscope is interfaced with the light source to illuminate the bronchial tree, and the imaging bundle of the endoscope collects and relays the reflected and autofluorescent light from the tissue surface. The spectral measurements were performed using a specially designed spectral attachment between the endoscope eyepiece and the camera. The spectral acquisition system is described in detail by Zeng et al.21 The system is capable of real-time (video rate: 60 frames/s) imaging by either WLB or AFB modes, whereas corresponding spectroscopy measurements of sites of interest could be obtained by inserting a fiber-mirror into an interim imaging plane inside the camera unit. The region of light diverted for spectral capture is visible to the operator as a black spot in the image, which is positioned over the target (Fig. 1). Once located properly, a button is pressed to capture the spectral data, along with the associated image and any descriptive information the operator enters. A commercially available, flexible bronchoscope (Olympus BF-60) was used in conjunction with this device.
Bronchoscopy was performed under local, topical anesthesia, with additional sedation administered as required for standard outpatient bronchoscopy examination. Inspection of the lung for potential sites for biopsy was performed using both WLB and AFB imaging modes, with the choice of biopsy sites being at the discretion of the bronchoscopist on the condition that control biopsies of normal appearing mucosa were not permitted under the study protocol. Where sites of interest warranting biopsy were identified, spectra were acquired before obtaining the tissue sample. A minimum of 1 spectral measurement each was captured under both imaging modes. Where time and subject tolerance would permit, additional replicate spectral measurements were obtained. During spectral capture, biopsy sites were classified as visually suspicious for severe dysplasia to cancer, or as abnormal in appearance but not specifically suspicious for malignancy (ie, suggestive of inflammation, fibrosis, mild dysplasia, etc., and therefore representing a more flexible category in recognition of the fact that relatively speculative specimens may also need to be taken). Bronchoscopists were not aware of the spectral results before making the visual classification or the decision to biopsy.
Biopsy specimens were fixed with formalin and embedded in paraffin, and 5-μm sections were stained with hematoxylin and eosin. Review of biopsy slides was initially conducted by an onsite pathologist, and later by a central reference pathologist. The presence of bronchial epithelium was required for specimens to be considered adequate for analysis. All pathology determinations were made blinded to clinical and bronchoscopy results and to the findings of earlier histology examinations. Specimens were graded according to the consensus classification of the WHO/IASCL grading system.25 Approximate agreement between site and reference pathologist evaluations was required for data analysis of a collected biopsy. Specifically, biopsy results were considered to disagree if histopathology grading disagreed by more than 1 grade. In such cases, the sample was sent to a third, independent pathologist. If the third pathologist was within 1 level of agreement with the site or reference pathologist, then the third pathologist's result was used; otherwise the sample was excluded from analysis. For disputes with regard to unsatisfactory sample status or for nonsignificant pathology disagreements (1 level difference or less), the reference pathologist's result was used.
WL reflectance spectra were not analyzed for discrimination of high-grade lesions directly, but were instead assessed using a mathematical model to generate spectrally based estimates of physiologic parameters by modeling light transport through tissue. WL entering tissue is scattered by the tissue microstructures (eg, cell membranes, mitochondria) and/or absorbed by chromaphores (eg, hemoglobin) within the tissue.26 That light that is ultimately scattered back out of the tissue contributes to the diffuse reflectance spectrum, and it has its spectral characteristics altered by the scattering process and by absorption by the tissue that the light has passed through before exit. The diffuse reflectance spectrum is therefore the sum of light passing through many possible scattering paths within the tissue, and a mathematical model must be used to predict the final spectrum on the basis of the assumed chromaphores present (eg, hemoglobin), the distribution and density of scattering microstructures, and any geometric variations of these factors. The model used in this study had as input parameters the tissue blood volume fraction (BV), the fraction of hemoglobin that is desaturated from oxygen (dO2), and the scattering particle volume fraction. The details of the developed model and the algorithm used for extracting the microvascular tissue, and morphologic parameters from the measured spectra have been described in detail by Fawzy et al.23,24 To summarize the fitting process, a predicted spectrum is produced from an initial parameter set estimate. The parameter values are then adjusted until an optimal match (defined by the integrated minimum least squared difference) is obtained with the empirically observed spectra (Fig. 2). The parameter values from the optimal match form the estimates of the tissue properties to be analyzed for discrimination potential. Before analysis, all spectra were screened for acceptable signal-noise levels and for spectrometer saturation events. The instrument response was corrected for by dividing the raw spectra by a reference spectrum acquired against a white reflectance standard (SRS-99; Lab-sphere). The accuracy of model estimates has not been tested in vivo, but testing against known tissue phantoms has been performed to confirm the absolute accuracy of model estimates.
Fluorescence spectra were not analyzed by using the above model; instead an estimate of the fluorescence intensity was obtained directly. The raw fluorescence spectra were first preprocessed to remove the instrument response. The NIR-G value, the ratio of the average intensity of the NIR (715 to 835 nm) reflectance band to the average intensity of the green fluorescence band (495 to 515 nm), was then calculated. The NIR response is approximately constant in the lung, and therefore the NIR-G ratio provides a distance-compensated estimate of the autofluorescence response.
The goal of data analysis was to assess the ability of the estimated summary statistics of the WL and fluorescence spectra (BV, dO2, NIR-G) to potentially improve the discrimination of high-grade (severe dysplasia or worse) versus lower-grade visually suspicious lesions. Discrimination was assessed using ROC curves, obtained by plotting specificity against sensitivity for high-grade versus lower-grade biopsies. The AUC was calculated, and 95% confidence intervals accounting for intrasubject correlations among biopsies were obtained.27 Where multiple spectra were obtained for a single biopsy site, the average of the model results was used. ROC curves were also obtained for secondary response variables created by multiplying the estimated microvascular parameters (yielding the volume fraction of oxygen-saturated and desaturated blood). To evaluate the performance of BV, dO2, and NIR-G in combination, binomial logistic regression with backward stepwise selection was used (P value of 0.05 required for variables to be retained), and the ROC curve of the final logistic fit was determined.
Results were also evaluated using a linear mixed model with fixed site, histology grade, and interaction effects. Many patients provided more than 1 specimen, which could introduce the risk of pseudoreplication if tissue optical properties are correlated within individuals. An example would be where a general state of inflammation increases the BV and NIR-G measurements of any spectral measurements of that patient. If such individual variations occur, there is a risk of overestimating the statistical significance of model effects were specimens to be treated as independent. The linear model therefore included an additive, normally distributed random effect term along with the fixed effects (ie, a mixed model) to compensate for possible variations between individuals. Model fitting was performed by the maximum likelihood method using the “nlme” package of the statistical language R.28,29 Histology grades were aggregated into simplified groups of normal, hyperplasia, metaplasia, mild/moderate dysplasia (mid-grade), and severe dysplasia/CIS/cancer. This grouping was chosen so that each grade category would have an adequate sample size, such that any trends among early grades could be evaluated if present. Both BV and NIR-G results were strongly skewed, and a natural logarithm transformation was applied to both to better approximate the normal distribution before analysis by the linear mixed model.
The target study size consisted of 550 patients; however, at the end of the enrollment period only 485 patients were able to be recruited, with at least 1 biopsy specimen from 399 of these patients (representing a total of 970 specimens). For patients with specimens taken, the median age was 62 years (range: 20 to 86 y), and 67% were male. Of these, there were 767 specimens available from 367 patients where there existed both analyzable WL/fluorescence spectra and biopsy pathology results. From this total analyzable set, there were 352 specimens from 235 patients that were indicated as suspicious for the presence of cancer. The breakdown of specimens and patients is shown in Figure 3. An average of 1.8 WL and fluorescence spectra per biopsy (maximum of 5) were acquired from suspicious, analyzable lesions. The proportion of positive specimens from visually suspicious lesions was significantly higher (21%) than for abnormal lesions (4.5%). In practice, the abnormal classification category was also found to apply to lesions of low suspiciousness for malignancy and to nonsuspicious but abnormal appearing lesions, resulting in the high yield and the large quantity of such sampled lesions. Sites differed in total subject recruitment (45, 171, 156, and 113 from Canada, Russia, Slovenia, and the United Kingdom, respectively), which, along with a differing propensity to biopsy, contributed to large variations in analyzable, suspicious specimens per site (Table 1). The yield of high-grade (severe dysplasia or worse) specimens from analyzable, suspicious lesions was also site-dependent, with yields observed as low as 13% for Russia or as high as 50% for Slovenia. Investigators indicated that suspicious lesions were identified primarily by the presence of redness under AFB or WLB (excluding that caused by contact with the bronchoscope); however, other subtle and subjective features such as irregularities of the mucosa and chromaphore patterns were also used. Study centers inexperienced with this fluorescence bronchoscopy system (ie, all but Slovenia) reported greater confidence in discerning between suspicious and abnormal lesions as the study progressed, and in particular discrimination was increasingly made between “abnormal” and “suspicious” redness.
Empirical ROC curves for discriminating higher from lower histology grades were the primary performance measure, and are shown in Figure 4 for NIR-G, BV, and dO2. The NIR-G ratio showed the highest discrimination, followed by dO2, and BV had the lowest ratio; the AUCs were 0.83, 0.80, and 0.74, respectively. All measures had an AUC significantly better than that expected for chance performance. Figure 4 also shows the ROC curves for severe dysplasia and CIS only, excluding cancer. Despite the presence of only 8 such severe dysplasia/CIS biopsies, ROC curve areas were also significantly (P<0.05) better than chance, with AUCs of 0.77, 0.85, and 0.69 for NIR-G, BV, and dO2, respectively. For visually abnormal lesions, NIR-G and dO2 had AUCs significantly greater than chance (0.76 and 0.75, respectively). For all response variables, the effect of histology grade in the linear mixed model analysis was highly significant (P<0.0001). Study center was also found to be a significant effect for all responses (P=0.004, <0.0001, and 0.001 for NIR-G, BV, and dO2, respectively). Furthermore, there was evidence of an interaction between site and histology grade for NIR-G and BV (P=0.01 for both, vs. 0.31 for dO2), suggesting that the performance of these measures may be site-dependent. There were 12 specimens graded as benign but abnormal (ie, inflammation, fibrosis, granulomatous, etc.), which was considered an insufficient quantity for separate analysis, and thus these histopathologies were excluded from the linear mixed model analysis.
When BV is multiplied by dO2, the result is a secondary feature estimating the oxygen-desaturated BV (the saturated BV is similarly estimated). It was found that the desaturated BV yielded better discrimination (Fig. 5) than BV or dO2 alone, whereas the discrimination for the saturated BV was relatively poor (AUC of 0.83 and 0.63 for desaturated and saturated BV, respectively). Both BV and NIR-G were well correlated (Spearman ρ: 0.54), and in the logistic regression performed using NIR-G, BV, and dO2 they were redundant features, with NIR-G being retained in backward stepwise selection. In the reduced logistic model, both dO2 and NIR-G were highly significant (P<10−6) terms, and the AUC of the combined logistic regression score was 0.88. Discrimination was similarly observed for the logistic model and the desaturated BV when applied to severe dysplasia and CIS only (AUCs of 0.79 and 0.87, respectively).
The ROC curves of all response variables were markedly better than chance for discriminating high-grade from lower-grade histologies. For example, the result for dO2 would imply that this feature, if applied to visually suspicious lesions representative of those in this study, may reduce the number of unnecessary biopsies by approximately 60%, but at the expense of failing to biopsy approximately 10% of high-grade lesions (sensitivity of 90%). This performance for tissue oxygen-desaturated fraction is similar to that reported for spectral discrimination of tumors of the colon.30 The fact that these factors show discrimination with regard to suspicious lesions selected for biopsy suggests that quantitative spectral analysis may have some potential to improve the specificity of the AFB/WLB bronchoscopy procedure with only a small loss of sensitivity. It is encouraging that when only severe dysplasia/CIS was considered, the ROC curve area was still significantly better than for chance—although with only 8 such biopsy specimens, this finding must be considered highly tentative. Apparent performance may also have been underestimated by the occasional difficulty in acquiring both the biopsy and the spectral data from the exact location of interest.
This study found significant discrimination of high-grade lesions to be present using spectrally estimated tissue parameters. However, whether the addition of spectral analysis to WLB/AFB would be beneficial for clinical practice depends on several considerations. One is the extent to which AFB specificity has been limiting its clinical adoption, and what level of reduction in high-grade lesion detection as a result of avoiding biopsies of certain lesions (false negatives by spectral analysis) would be clinically acceptable to improve specificity. Another consideration is whether the performance of a combined procedure (WLB/AFB and spectral analysis) would meet clinician requirements. Although this study has provided data for identifying promising factors, it was not designed to estimate combined performance, in that control biopsies of normal appearing tissue were not permitted under the protocol and long-term follow-up of patients was not performed. Consequently, the performance of the WLB/AFB procedure with this device in isolation is not known, and thus the implied combined performance by adding quantitative spectral analysis also cannot be estimated.
In principle, the ROC curve of the spectral parameters could be applied to a corresponding WLB/AFB performance point to infer a combined performance. However, for the result to be generalizable to other clinical settings, it is required that the spectral results be independent of the criteria used by the bronchoscopist for classifying lesions as suspicious—but this is not expected by theory (where BV and NIR-G are known to be already qualitatively visible in the video image and used by the bronchoscopist), nor supported by the results (eg, the significance of study center effects). The finding of significant site and site by grade interactions may relate to subject population differences, but is more likely linked to differences between bronchoscopists, particularly given that large differences in positive yield on visually suspicious biopsies between sites were found, indicating that there were differences in how selection criteria were being applied. It is interesting that site by grade differences were not apparent for dO2, given that the nature of this measurement makes it the only one not directly qualitatively visible in the video image and hence possibly the least affected by operator biopsy selection differences. Another consideration is that the results were calculated for all suspicious lesions, but where spectral data would likely be most useful is in those lesions that were only borderline suspicious—highly suspicious lesions would likely have been biopsied regardless of the spectral result. The observed performance of NIR-G and dO2 on lesions visually classified as abnormal suggests that these features could still be useful on such borderline cases. Furthermore, results inferred from a post hoc spectral analysis might not be representative if the spectral results were available in real time to provide immediate feedback to the operator. Therefore, although these results suggest that there may be benefits to the use of these spectral features with AFB/WLB, the above issues suggest that caution is warranted in extrapolating to infer the benefit in a clinical setting.
We found that the volume fraction of oxygen-desaturated blood (multiplying BV and dO2) was significantly more discriminatory than the volume fraction of saturated blood (the combination of BV and the oxygen saturated fraction). However, both AFB and WLB imaging modes are only able to allow qualitative visualization of the total BV. This suggests that a functional imaging approach to visualizing the volume fraction of hypoxic blood may potentially yield improvements with regard to AFB or WLB. It is conceivable that subject-level sensitivity and specificity might be improved in this way by allowing the detection of high-grade lesions not found to be suspicious earlier. However, if hypoxic conditions are only present in already suspicious or advanced lesions, then the advantages of such a system may be limited. An alternative application for spectral data could involve the use of spectral analysis for the characterization of lesions already identified as tumors. Tissue hypoxia has been found to provide important information with regard to metastatic potential, resistance to radiotherapy, and patient survival,31–33 and spectrally estimated tissue hypoxia could be of use in informing patient prognosis or the choice of treatment method.
Noncontact spectral measurements of suspicious lesions in the bronchial tree have been made to characterize differences between high-grade (severe dysplasia—cancer) and lower-grade histologies. Biopsy specimens with high-grade histology were found to have significant increases in the estimated BV of the tissue and a reduction in the oxygen saturation fraction of hemoglobin. The autofluorescence intensity was also significantly reduced in high-grade lesions. It was found that the discrimination of the oxygen-desaturated BV was significantly higher than that of the saturated fraction. The results indicate that the addition of spectral analysis to WLB/AFB may be of benefit, but that the gain in performance is expected to be affected by bronchoscopist technique in identifying suspicious lesions. Further study would be required to understand how real-time spectral information would perform if used in conjunction with WLB/AFB in a clinical context.
The authors thank the following investigators and institutions for their participation in the study: Dr Andrey Arseniev, N. N. Petrov Research Institute of Oncology, St. Petersburg, Russia; Dr Marjeta Tercelj, Ljubljana University Medical Center, Slovenia; and Dr Pearce Wilcox, St. Paul's Hospital, Vancouver, Canada.
1. Woolner LB, ed. Pathology of cancer detected cytologically. In: Atlas of Early Lung Cancer
: National Cancer Institute. National Institutes of Health and Human Services. Tokyo: Igaku Shoin; 1983;107–213.
2. Palcic B, Lam S, Hung J, et al. Detection and localization of early lung cancer
by imaging techniques. Chest. 1991;99:742–743.
3. Hung J, Lam S, LeRiche JC, et al. Autofluorescence
of normal and malignant bronchial tissue. Laser Surg Med. 1991;11:99–105.
4. Lam S, MacAulay C, Palcic B. Detection and localization of early lung cancer
by imaging techniques. Chest. 1993;103:12S–14S.
5. Lam S, Palcic B. Autofluorescence bronchoscopy
in the detection of squamous metaplasia and dysplasia in current and former smokers. J Natl Cancer Inst. 1999;91:561–562.
6. Lam S, MacAulay C, LeRiche JC, et al. Detection and localization of early lung cancer
by fluorescence bronchoscopy
. Cancer. 2000;89S:2468–2473.
7. Lam S, Lam B, Petty TL. Early detection for lung cancer
. New tools for case finding. Can Fam Physician. 2001;47:537–544.
8. Lam S, Kennedy T, Unger M, et al. Localization of bronchial intraepithelial neoplastic lesions by fluorescence bronchoscopy
. Chest. 1998;113:696–702.
9. Haubinger K, Becker H, Stanzel F, et al. Autofluorescence bronchoscopy
with white light bronchoscopy
compared with white light bronchoscopy
alone for the detection of precancerous lesions: a European randomized controlled multicentre trial. Thorax. 2005;60:496–503.
10. Mourant JR, Hielscher AH, Eick AA, et al. Evidence of intrinsic differences in the light scattering properties of tumorigenic and nontumorigenic cells. Cancer. 1998;84:366–374.
11. Zonios G, Perelman LT, Backman V, et al. Diffuse reflectance spectroscopy
of human adenomatous colon polyps in vivo. Appl Opt. 1998;38:6628–6636.
12. Bard M, Amelink A, Skurichina M, et al. Improving the specificity of fluorescence bronchoscopy
for the analysis of neoplastic lesions of the bronchial tree by combination with optical spectroscopy
: preliminary communication. Lung Cancer
13. Bard M, Amelink A, Skurichina M, et al. Optical spectroscopy
for the classification of malignant lesions of the bronchial tree. Chest. 2006;129:995–1001.
14. Amelink A, Sterenberg H, Bard M, et al. In vivo measurement of the local optical properties of tissue by use of differential path-length spectroscopy
. Optic Lett. 2004;29:1087–1089.
15. Bard M, Amelink A, Hegt V, et al. Measurement of hypoxia
-related parameters in bronchial mucosa by use of optical spectroscopy
. Am J Respir Crit Care Med. 2005;171:1178–1184.
16. Keith R, Miller Y, Gemmill R, et al. Angiogenic squamous dysplasia in bronchi of individuals at high risk for lung cancer
. Clin Cancer Res. 2000;6:1616–1625.
17. Jain R. Determinants of tumor blood flow: a review. Cancer Res. 1988;48:2641–2658.
18. Vaupel P, Kallinowski F, Okunieff P. Blood flow, oxygen and nutrient supply, and metabolic microenvironment of human tumors: a review. Cancer Res. 1989;49:6449–6561.
19. Cao Y, Li C, Moeller B, et al. Observation of incipient tumor angiogenesis that is independent of hypoxia
inducible factor-1 activation. Cancer Res. 2005;65:5498–5505.
20. Gazdar A, Minna J. Angiogenesis and the multistage development of lung cancers. Clin Cancer Res. 2000;6:1611–1612.
21. Zeng H, Petek M, Tercelj M, et al. Integrated endoscopy system for simultaneous imaging and spectroscopy
for early lung cancer
detection. Optic Lett. 2004;29:587–589.
22. Tercelj M, Zeng H, Petek M, et al. Acquisition of fluorescence and reflectance spectra during routine bronchoscopy
examinations using the ClearVu Elite device: pilot study. Lung Cancer
23. Fawzy Y, Zeng H. Determination of scattering volume fraction and particle size distribution in the superficial layer of a turbid medium by using diffuse reflectance spectroscopy
. Appl Optic. 2006;45:3902–3912.
24. Fawzy Y, Petek M, Tercelj M, et al. In vivo assessment and evaluation of lung tissue morphologic and physiological changes from non-contact endoscopic reflectance spectroscopy
for improving lung cancer
detection. J Biomed Optic. 2006;11:044003–044012.
25. Travis WD, Colby TS, Shimosato Y, et al. WHO/IASLC Classification of Lung and Pleural Tumours. 3rd ed. Berlin: Springer; 1999.
26. Patterson M, Wilson B, Wyman D. The propagation of optical radiation in tissue I. Models of radiation transport and their application. Laser Med Sci. 1991;6:155–168.
27. Obuchowski NA. Nonparametric analysis of clustered ROC curve data. Biometrics. 1997;53:567–578.
28. Pinheiro J, Bates D, DebRoy S, et al. R Core team. nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1-86, 2007.
29. R Development Core Team (2007). R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http:
30. Maxim P, Carson J, Benaron D, et al. Optical detection of tumors in vivo by visible light tissue oximetry. Tech Cancer Res Treat. 2005;4:227–234.
31. Kim S, Rabbani Z, Vollmer R, et al. Carbonic anhydrase IX in early-stage non-small cell lung cancer
. Clin Cancer Res. 2004;10:7925–7933.
32. Swinson D, Jones J, Richardson D, et al. Carbonic anydrase IX expression, a novel surrogate marker of tumor hypoxia
, is associated with a poor prognosis in non-small-cell lung cancer
. J Clin Oncol 2003;21:473–482.
33. Hockel M, Vaupel P. Tumor hypoxia
: definitions and current clinical, biologic and molecular aspects. J Natl Cancer Inst. 2001;93:266–276.
Keywords:© 2009 Lippincott Williams & Wilkins, Inc.
bronchoscopy; lung cancer; spectroscopy; autofluorescence; hypoxia