- Email: [email protected]
Endoscopic Raman spectroscopy enables objective diagnosis of dysplasia in Barrett's esophagus
ORIGINAL ARTICLE
Endoscopic Raman spectroscopy enables objective diagnosis of dysplasia in Barrett’s esophagus L. Max Almond, MB, ChB, MRCS, DM,1,2 Jo Hutchings, BSc, PhD,1 Gavin Lloyd, BSc, PhD,1 Hugh Barr, MD, ChM, FRCS, FRCE, FHEA,1,2 Neil Shepherd, DM, FRCPath,3 John Day, BSc, MSc, PhD,4 Oliver Stevens, BA, MEng,4 Scott Sanders, MD, FRCPath,5 Martin Wadley, MB, ChB, MD, FRCS,2,6 Nick Stone, BSc (Hons), MSc (Dist), MSc (Dist), PGDip (IPSM), PhD, MBA, CSci, FIPEM,7 Catherine Kendall, BSc, MSc, PhD1 Gloucester, Warwick, Bristol, Exeter, United Kingdom
Background: Early detection and targeted endoscopic resection of Barrett’s esophagus–associated high-grade dysplasia (HGD) can prevent progression to invasive esophageal malignancy. Raman spectroscopy, a highly sophisticated analytical technique, has been translated into an endoscopic tool to facilitate rapid, objective diagnosis of dysplasia in the esophagus. Objective: To evaluate the ability of endoscopic Raman spectroscopy (ERS) to objectively detect esophageal HGD and adenocarcinoma. Design: A total of 798 one-second spectra were measured from 673 ex vivo esophageal tissue samples, collected from patients with Barrett’s esophagus by using a novel endoscopic Raman probe. Spectra were correlated with consensus histopathology. Multivariate analysis was used to evaluate the classification accuracy of ERS ex vivo. Setting: Probe measurements were conducted in the laboratory. Tissue specimens were collected from the operating theatre and endoscopy unit. Patients: Tissue from 62 patients was included in the study. Interventions: Endoscopic biopsy/resection or esophagectomy was performed where indicated clinically. Main Outcome Measurement: Diagnostic performance of ERS for detection of HGD and esophageal adenocarcinoma. Results: ERS demonstrated a sensitivity of 86% and a specificity of 88% for detecting HGD and adenocarcinoma. The ability to grade dysplasia and differentiate intestinal metaplasia from nonintestinal metaplasia columnar-lined esophagus was also demonstrated. Diagnostic classification was based on objective measurement of the biochemical profile of different tissue types. The potential for combination ERS and narrow-band imaging was also demonstrated. Limitations: Measurements were taken from ex vivo tissue. Conclusion: ERS enables rapid, accurate, objective diagnosis of superficial esophageal disease (metaplasia, dysplasia, intramucosal cancer) in clinically applicable time scales. (Gastrointest Endosc 2013;-:1-9.)
Abbreviations: ER, endoscopic resection; ERS, endoscopic Raman spectroscopy; HGD, high-grade dysplasia; IM, intestinal metaplasia; LGD, low-grade dysplasia; NBI, narrow-band imaging; NDBE, nondysplastic Barrett’s esophagus; PC, principal component; RS, Raman spectroscopy; WLE, white light endoscopy. DISCLOSURE: The authors disclosed no financial relationships relevant to this publication. This work was funded by the National Institute for Health Research (NIHR) (N.S. J.H.), the Royal College of Surgeons of England (L.M.A.), and the McIntyre Research Fund (L.M.A.).
www.giejournal.org
Copyright ª 2013 by the American Society for Gastrointestinal Endoscopy 0016-5107/$36.00 http://dx.doi.org/10.1016/j.gie.2013.05.028 Received February 7, 2013. Accepted May 22, 2013. Current affiliations: Biophotonics Research Unit (1), Departments of Esophagogastric Surgery (2) and Histopathology (3), Gloucestershire Hospitals NHS Foundation Trust, Gloucester; Interface Analysis Centre, (footnotes continued on last page of article)
Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 1
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
Barrett’s esophagus represents a metaplastic transformation from a normal squamous esophageal epithelium to a columnar-lined esophagus, which occurs in response to chronic pulsatile reflux of gastric acid and bile into the distal esophagus. Barrett’s esophagus confers an increased risk of malignant degeneration via precancerous dysplasia. However, despite this potentially curable premalignant phase, the incidence of esophageal adenocarcinoma has increased by 500% over the past 40 years, although its prognosis has remained relatively constant with current 5-year survival rates of just 8% to 15%.1-3 Currently, patients with nondysplastic Barrett’s esophagus (NDBE) are surveyed endoscopically at intervals of 2 to 3 years with adherence to a strict biopsy protocol consisting of quadrantic biopsies every 2 cm of Barrett’s mucosa.4 The need for this random-sampling approach reflects difficulties in the identification of early neoplastic lesions in Barrett’s segments, causing missed dysplasia in as many as 57% of patients and limiting our ability to perform targeted R0 endoscopic resection of suspicious lesions.5 A novel, custom-built, Raman spectroscopic probe that can be passed down the instrument channel of a gastroscope was developed in an attempt to improve endoscopic detection and targeted treatment of dysplasia and intramucosal (T1a) cancer of the esophagus. The probe uses a well-established analytical technique termed Raman spectroscopy (RS) to interrogate the biochemical composition of its target. RS is highly suitable for translation into clinical use because Raman signals are highly specific to molecular constituents of cells and tissues.6-12 Furthermore, they can be acquired in clinically applicable time scales from fresh, unprocessed tissue without the need for exogenous contrast. In vitro data have demonstrated that RS is capable of objective diagnosis of 7 pathologies in the distal esophagus including high-grade dysplasia (HGD), adenocarcinoma, and NDBE.13 The custom-built probe transmits a low-power, monochromatic, near-infrared laser beam that excites biochemical vibrations in the esophageal epithelium. Inelastically scattered laser photons are preferentially detected and analyzed to derive an objective biochemical profile of the tissue being interrogated. Several groups are in the process of translating RS into an endoscopic tool for use in the esophagus.9,14,15 However, previous probe designs have concentrated on maximizing signal uptake rather than ensuring precise interrogation of the epithelium without spectral contributions from deeper tissues. The novel probe described here is the first of its type with a confocal design ensuring a sampling depth on the order of 150 mm in an effort to detect early superficial lesions that are potentially curable by endoscopic therapy and reduce spectral contamination by basal cells that have been shown to have a spectral signature similar to that of dysplasia.16 The ability for a rapid, objective diagnosis of early Barrett’s esophagus–associated neoplasia using a Raman 2 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
Take-home Message A novel confocal Raman probe can differentiate epithelial and mucosal disease including high-grade dysplasia (HGD) and intramucosal cancer from low-grade dysplasia and nondysplastic Barrett’s esophagus with high diagnostic accuracy in clinically applicable time scales (1 second). Detection of subepithelial “buried” neoplasia is possible by using endoscopic Raman spectroscopy (ERS). ERS could enable targeted endoscopic resection of HGD and intramucosal cancer associated with Barrett’s esophagus before whole-segment ablation therapy.
probe ex vivo is evaluated against a consensus histological criterion standard. The potential for multimodality imaging by using endoscopic RS (ERS), narrow-band imaging (NBI), and white light endoscopy (WLE) is also addressed. In addition, the biochemical basis of spectral classification between benign and neoplastic esophageal tissues is evaluated.
MATERIALS AND METHODS Raman probe system The custom-built fiberoptic endoscopic Raman probe transmits 830 nm of light generated by an Innovative Photonics System (Princeton, NJ) laser source. After tissue interaction, inelastically scattered laser photons are transmitted back along the probe to a Renishaw system 100 spectrometer (Renishaw PLC, Wotton-under-Edge, Gloucestershire, UK) optimized for near-infrared signals. The probe has a diameter of 2.7 mm, enabling easy passage through the instrument channel of an endoscope. It is coupled to an optical cable containing the input and output fibers. Built-in optical filters remove elastically scattered signal and tissue fluorescence to optimize collection of inelastic Raman signals. The probe is designed for direct tissue contact and has confocal delivery and collection optics, ensuring interrogation of the epithelium to a depth of approximately 150 mm. Power output at the probe tip was regulated to approximately 60 mW during all measurements. An expert histopathologist verified that no histological damage was visible at this power level.
Sample preparation and spectral measurements Ethical approval for the study was granted by Gloucestershire Local Research Ethics Committee, Gloucestershire, UK. All patients provided informed consent to permit the use of their esophageal tissue for ex vivo spectral measurements. Eligible patients included those undergoing esophagectomy for esophageal adenocarcinoma or multifocal HGD or surveillance endoscopy for Barrett’s esophagus. Patients with squamous cell carcinoma or squamous dysplasia were www.giejournal.org
Almond et al
Spectroscopic diagnosis of esophageal dysplasia
Figure 1. Unprocessed 1-second Raman probe spectra measured from Barrett’s esophagus, adenocarcinoma, and normal squamous esophagus. Spectra have been offset for clarity. An air background spectrum is shown for comparison, and prominent spectral features have been identified.
excluded. Spectra measured from nonhomogeneous tissue samples containing more than 1 pathology type were excluded from the subsequent analysis. After Ivor-Lewis esophagectomy, the surgical specimen was opened within minutes of resection, and a Perspex grid was fixed across the mucosal surface. Each grid hole was 3 mm in diameter, enabling spectral measurements to be taken by passing the probe through the grid at specified positions and gently apposing it to the esophageal mucosa to mimic the clinical environment at endoscopy. Raman spectra were acquired using 5-, 1-, and 0.1second acquisition times across a wave number range of 400 to 1850 cm 1. Biopsy specimens were then taken at each grid position by using endoscopy biopsy forceps passed through the grid so that spectral signatures could be correlated with the histopathological diagnosis. Biopsy specimens were immediately fixed in formalin and sent for routine histological processing and expert reporting. To augment the spectral dataset, endoscopic point biopsy and endoscopic resection (ER) specimens were also used for spectral analysis. After endoscopic collection, these specimens were immediately snap-frozen in liquid nitrogen and transported to the laboratory. After thawing, 3 spectra were measured per ER specimen, and 1 spectrum per point biopsy specimen. Care was taken to ensure that specimens were oriented with their epithelial surface upward. Samples were then formalin fixed, processed, and reported. The histology of all tissue samples was reported by 2 independent expert GI pathologists from different centers. Tissue specimens with a consensus histological diagnosis were used as the criterion standard for comparison with Raman probe performance.
Trimodality imaging feasibility assessment The potential for combining ERS, NBI, and WLE was also investigated. One-second Raman spectra were measured www.giejournal.org
from a small subset of fresh esophageal biopsy specimens in conjunction with NBI. For comparison, spectra were also acquired in the presence of standard WLE and after extrusion of all ambient light (dark control).
Statistical analysis Multivariate analyses were used to extract the subtle spectral differences between esophageal tissue types. Spectral data were initially normalized and mean centered. Both green glass correction and air background subtraction algorithms were also used initially, although neither significantly enhanced diagnostic classification performance and were therefore excluded from subsequent analyses. Principal component fed linear discriminant analysis was used to develop a classification model in Matlab (Mathworks Inc, Natick, Mass). Principal components were generated that simplified the data while retaining the spectral information required for diagnostic classification. Linear discriminant analysis incorporated histological information to describe functions that enabled spectral classification.16 Leave-one-tissue-site-out cross-validation was used to test the diagnostic performance of the probe against the consensus pathological criterion standard diagnosis.
RESULTS A total of 2226 spectra were measured from esophageal tissue from 62 patients over a 20-month period. Consensus histological diagnoses were available for 798 one-second spectra acquired from 673 ex vivo esophageal tissue samples. The majority (651/798 [82%]) of these spectra were measured directly from fresh esophageal mucosa after Ivor-Lewis esophagectomy. Of 798 spectra, 147 (18%) were acquired from endoscopic biopsy or endoscopic Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 3
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
TABLE 1. The diagnostic performance of the Raman probe is indicated Sensitivity, %
Specificity, %
NSq (251)
87
96
LGD/NDBE (135)
72
91
HGD/adenocarcinoma (199)
86
88
NSq (225)
80
96
LGD/NDBE (80)
71
86
HGD/adenocarcinoma (189)
83
90
NSq (26)
96
98
LGD/NDBE (55)
96
98
HGD/adenocarcinoma (10)
96
97
IM (100)
91
80
Non-IM CLO (CB/FB) (14)
80
91
NSq (251)
90
78
Buried adenocarcinoma (74)
78
90
LGD (21)
100
100
HGD/IMC (10)
100
100
HGD/adenocarcinoma (199)
94
91
NSq (225)
91
94
NDBE (114)
96
92
Gastric (12)
92
96
NDBE/LGD (49)
98
86
HGD/adenocarcinoma (89)
86
98
3-group model (spectra acquired in 1 s), whole dataset
3-group model (spectra acquired in 1 s); fresh, surgically resected samples only
3-group model (spectra acquired in 1 s); snap-frozen, endoscopic samples only
2-group models (spectra acquired in 1 s)
2-group models (spectra acquired in 5 s)
NSq, Normal squamous; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett’s esophagus; HGD, high-grade dysplasia; IM, intestinal metaplasia; CLO, columnar-lined esophagus; CB, cardiac Barrett’s; FB, fundic Barrett’s; IMC, intramucosal cancer (number of homogeneous samples of each pathology). All results were statistically cross-validated by using leave-one-tissue-site-out cross-validation.
resection specimens that had been snap-frozen in the endoscopy department. Figure 1 illustrates 1-s Raman spectra acquired from normal squamous esophagus, NDBE, and adenocarcinoma. An air background spectrum is included for comparison, demonstrating unwanted spectral artifacts caused by elastic scattering and fluorescence generated within the probe fibers. A 3-group principal component (PC)–fed linear discriminant classification model was developed by using consensus normal squamous, NDBE/low-grade dysplasia (LGD), and high-grade dysplasia (HGD)/adenocarcinoma samples. Leave-one-tissue-site-out cross-validation demonstrated a Raman probe sensitivity of 86% and specificity of 88% for correctly identifying HGD/adenocarcinoma samples from the other pathologies (Table 1). Figure 2 4 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
illustrates a scatterplot of cross-validated linear discriminant scores for each Raman probe prediction, which visually demonstrates the diagnostic capability of the Raman probe to separate these pathology groups in clinically applicable (1-second) time scales. The diagnostic accuracies attained using 5-second spectral acquisition times are illustrated in Table 1 for comparison. Table 1 also illustrates the diagnostic classification of the 18% of spectra acquired from endoscopic biopsy/resection tissue samples. Sensitivities and specificities of 96% to 98% were attained for all pathologies included in this dataset. Analysis of spectra measured solely from esophagectomy specimens is also illustrated. Two-group diagnostic classification of LGD and HGD/ intramucosal cancer demonstrated a sensitivity and specificity of 100% for the diagnosis of advanced neoplasia www.giejournal.org
Almond et al
Figure 2. Three-group linear discriminant score prediction scatterplot identifying samples that were measured from fresh (stars) and from fresh-frozen (circles) tissue samples.
because classification of all spectra matched the consensus criterion standard diagnosis (Table 1). The absence of any misclassification in this model likely reflects the relatively small numbers of homogeneous LGD and HGD/intramucosal cancer samples. However, these data certainly demonstrate the ability of ERS to detect and discriminate epithelial pathology. In addition, Raman probe measurements were able to discriminate intestinal metaplasia (IM) from cardiac metaplasia and fundic metaplasia with a sensitivity and specificity of 91% and 80%, respectively. IM, which is characterized histologically by goblet cells, has a greater malignant potential than cardic or fundic metaplasia and is considered necessary for a diagnosis of Barrett’s esophagus in the United States. Similarly, a sensitivity of 96% and specificity of 92% were demonstrated for correctly identifying Barrett’s esophagus from gastric mucosa (Table 1). Table 1 also illustrates that buried cancerous elements beneath normal squamous mucosa were detectable for various possible reasons including changes in localized biochemical expression in the surface epithelium covering the adenocarcinoma and potentially the transmission and collection of some light from beneath the surface epithelium. Diagnostic performance was shown to increase when using consensus pathology diagnoses to train the classification models (Table 2). Sensitivity for the detection of HGD/ intramucosal cancer increased from 70%-76% to 86%, and specificity increased from 85% to 88%. The level of agreement between Raman predictions and criterion standard (consensus) histopathological diagnosis for the 2-group model (NDBE/LGD and HGD/adenocarcinoma) was k Z 0.70. The level of agreement between the 2 independent pathologists for all samples was k Z 0.67. Further analysis was undertaken to evaluate the biochemical basis of diagnostic spectral classification. Linear www.giejournal.org
Spectroscopic diagnosis of esophageal dysplasia
discriminant loads generated after linear discriminant analysis identified 4 PCs as being particularly responsible for spectral classification: PCs 8 and 9 were most influential for differentiating normal squamous tissue from metaplastic and neoplastic tissue (Fig. 3), and PCs 6 and 11 were important for differentiating NDBE and LGD from advanced neoplasia (Fig. 4). Further examination of PCs 8 and 9 (Fig. 3) reveals prominent peaks at 490, 933, 1086 to 1110, and 1128 cm 1, all of which are consistent with a higher glycogen concentration in normal squamous epithelium compared with both Barrett’s and neoplastic tissue. This is entirely consistent with previous in vitro biomolecular modeling studies.9,10,15,20 Reference to peak assignment tables enables assignment of other prominent peaks at 638, 731, 782, 885, 1021, 1095, 1334, and 1621 cm 1 to nucleic acids and protein structures, including histone, which imply alterations in RNA and DNA concentrations between normal squamous and metaplastic/ neoplastic tissue. Plotting of PCs 6 and 11 also reveals prominent glycogen peaks at 492, 574, and 937 cm 1. The peaks at 771, 782, 1018, 1360, and 1511 cm 1 are indicative of nucleic acids, which again reflects increases in DNA and RNA concentrations between metaplastic and malignant tissues. Although the PCs illustrated partially account for the ability of the model to classify spectra, it must be remembered that a significant amount of important information is described by the other PCs. There are therefore many other biochemical peaks that contribute to spectral classification. Many of these were identified previously through preliminary proof-of-concept in vitro work. The potential for trimodality (ERS, NBI, and WLE) imaging was investigated using a small subset of endoscopic biopsy samples. One-second Raman spectra were measured from 11 fresh esophageal biopsy specimens in conjunction with NBI, WLE, and after extrusion of all ambient light (dark control). Figure 5 illustrates that all spectra acquired in conjunction with NBI lie well within a 95% confidence interval of the dark control measurements, demonstrating feasibility for a trimodality ERS-NBI-WLE diagnosis.
DISCUSSION Patients with Barrett’s esophagus, particularly those with dysplasia, are at a significantly increased risk of progression to adenocarcinoma. Detection of early neoplastic disease in these patients could facilitate prompt endoscopic intervention to eradicate HGD and intramucosal (T1a) cancer, preventing disease progression.21 A recent consensus statement from an international consortium of Barrett’s experts has supported an approach of targeted ER of focal HGD lesions followed by whole Barrett’s segment eradication by using radiofrequency ablation.22 This approach aims to ensure accurate disease staging as well as resection of all microscopic disease (R0) followed by eradication of any malignant field change. However, in Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 5
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
TABLE 2. The impact on diagnostic probe performance when using consensus pathology as the criterion standard diagnosis Sensitivity, %
Specificity, %
NSq
84
96
NDBE/LGD
68
86
HGD/adenocarcinoma
76
85
NSq
84
95
NDBE/LGD
72
86
HGD/adenocarcinoma
70
85
NSq
87
96
NDBE/LGD
72
91
HGD/adenocarcinoma
86
88
Pathologist 1
Pathologist 2
Consensus pathology
NSq, Normal squamous; NDBE, nondysplastic Barrett’s esophagus; LGD, low-grade dysplasia; HGD, high-grade dysplasia.
Figure 3. Principal component (PC) 8 and PC 9 loadings after principal component analysis on spectra measured from normal squamous, nondysplastic Barrett’s esophagus, low-grade dysplasia, high-grade dysplasia, and adenocarcinoma tissue.
Figure 4. Principal component (PC) 6 and PC 11 loadings after principal component analysis on spectra measured from normal squamous, nondysplastic Barrett’s esophagus, low-grade dysplasia, high-grade dysplasia, and adenocarcinoma tissue.
6 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
www.giejournal.org
Almond et al
Figure 5. Variation in tissue spectra acquired in the presence of white light endoscopy and narrow-band imaging (NBI) compared with those acquired after exclusion of all ambient light. Confidence intervals (95% and 99%) for the dark control are illustrated. PC, principal component.
a recent study of 742 ERs from 326 patients with Barrett’s neoplasia, an R0 resection was attained in only 26.8% of patients after initial therapy.22 Subsequent repeated ER achieved R0 resections in 74.5%, but only after a delay of several months between treatments. The study concluded that if ER is to be used as a viable therapeutic strategy in HGD and intramucosal cancer, improved mechanisms of lesion detection and margin definition must be developed. Despite a variety of advanced endoscopic imaging tools including NBI, chromoendoscopy, and autofluorescence imaging, the detection of dysplastic lesions in the esophagus remains difficult and extremely subjective. As many as 57% of Barrett’s esophagus–associated dysplasia may be missed by standard biopsy protocols in conjunction with WLE, and high-quality (R0) targeted endoscopic treatment is difficult.22 Development of improved techniques for endoscopic detection of esophageal neoplasia re-mains an essential ongoing area of translational research. This paper describes the application of a novel fiberoptic Raman probe that could enable rapid, objective diagnosis and targeting of Barrett’s esophagus–associated dysplasia and intramucosal adenocarcinoma without the need for exogenous contrast. The probe has demonstrated the ability to differentiate advanced neoplasia (HGD/ adenocarcinoma) from LGD, NDBE, and normal squamous esophagus with a sensitivity of 86% and specificity of 88%. These results are based on 1-second spectral measurements and were statistically cross-validated to avoid reporting of overly optimistic classification performance. It must also be considered that progression from NDBE to adenocarcinoma represents a progressive series of molecular and biochemical changes that manifest as progressive histological grades of dysplasia. There may be a considerable www.giejournal.org
Spectroscopic diagnosis of esophageal dysplasia
spectrum of molecular abnormalities that are consistent with a particular histological grade of dysplasia, and therefore spectral classification performance may be underestimated when based on correlation with histological appearances. Adenocarcinoma could also be differentiated from normal squamous esophagus with a sensitivity of 94% and a specificity of 91%, demonstrating potential for use in differentiating benign and malignant esophageal strictures. In addition, ERS could distinguish Barrett’s esophagus from gastric mucosa with a sensitivity of 96% and specificity of 92%. This distinction may be of particular importance in patients with an irregular z-line who may be incorrectly labeled with a diagnosis of Barrett’s esophagus. An ability to distinguish Barrett’s esophagus and gastric mucosa may also be of use in patients with hiatal hernias in whom the exact position of the esophagogastric junction is unclear. This is the first time that a Raman probe has demonstrated the ability to detect superficial neoplastic disease, as well as to distinguish IM from non-IM columnar-lined esophagus and Barrett’s esophagus from gastric mucosa. In addition, the ability to detect submucosally invasive cancer has also been reported, suggesting potential for detection of buried metaplasia during postablation surveillance. Similarly, there may also be a potential role in immediate objective intraoperative margin assessment during esophagectomy, although this will require further clinical evaluation. Histological diagnosis of dysplastic Barrett’s esophagus known to have considerable interreporter variation.12,17-19 Because incorrect histological diagnoses could confuse the assessment of probe performance, the analyses were repeated by using the pathology results provided by each of the independent pathologists and a consensus of the 2. Diagnostic performance was found to increase when using consensus pathology diagnoses to train the classification models, intimating that performance could have improved further by inclusion of an additional pathologist to further optimize the criterion standard (Table 2). It is possible that 1-second Raman probe measurements may be capable of at least matching the performance of an individual pathologist. It is also notable that diagnostic accuracy for detecting advanced neoplasia increased when LGD samples were removed from the analysis, a group for which it is known that pathologists have particular difficulty agreeing on. Spectral diagnosis was based on objective detection of identifiable biochemical differences between benign and neoplastic tissues that concur with the published literature. This avoids the subjectivity associated with histological reporting; in fact, spectral classification was shown to increase when consensus histological diagnoses were used as the criterion standard.13 Current advanced endoscopic tools such as confocal microscopy and NBI lack this objectivity. Additional data not presented here have confirmed no significant difference in spectra acquired by 2 independent operators using 2 identical probes.23 The ability to record Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 7
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
consistent reproducible spectra by using a single probe over a 20-month period has also been demonstrated.23 The feasibility of combination ERS, NBI, and WLE has been highlighted for the first time, although more data will be necessary to clearly evaluate this potential. NBI is a widely available imaging adjunct that enhances the resolution of the mucosal surface, potentially improving detection of dysplasia and intramucosal cancer in the hands of experienced users. However, considerable interuser variation and high false-positive rates are widely reported.24,25 ERS would complement a wide-field imaging technique such as NBI, which has subjectivity and poor specificity.26 A dual-modality (ERS-NBI) imaging system could potentially limit the high false-positive rate associated with NBI and provide objective diagnostic assessment to reduce interuser variation. A limitation of this study is that spectra were acquired from ex vivo tissue samples. These samples had been rendered ischemic, which may have had a small impact on their spectral signal, although it should be noted that previous in vivo studies at other body sites have not noted significant spectral changes between ex vivo and in vivo measurements.27 In addition, most of the esophagectomy specimens were from patients who had received neoadjuvant chemotherapy before resection. Shrinkage and fibrosis of the tumor, which may have altered the biochemical composition of the mucosa, could account for the lower diagnostic accuracies obtained from surgically resected tissue compared with those reported from endoscopy specimens, suggesting that the true diagnostic accuracies of HGD and adenocarcinoma might be 96% to 98%, as demonstrated after analysis of endoscopy samples alone (Table 1). In vivo trials will be required to address some of these issues, and preparation for this is under way. ERS is a highly sophisticated analytical technique that could provide rapid objective in vivo diagnosis of a range of esophageal pathologies during endoscopy without the need for staining or exogenous contrast administration. A Raman probe could be used to confirm dysplastic areas within Barrett’s segments. In addition, it could facilitate careful lesion margin assessment to ensure high-quality R0 endoscopic resections. Multiple probes are currently undergoing construction, and engineering tolerance levels are being carefully defined before a clinical trial of ERS in patients with Barrett’s esophagus. If successful, it is hoped that wider clinical implementation in tertiary units with facilities for esophageal ER will be possible in the near future.
3.
4.
5.
6. 7.
8.
9.
10.
11.
12. 13.
14.
15.
16.
17. 18.
19. 20.
21. 22.
REFERENCES 1. National Cancer Institute’s Surveillance, Epidemiology, and End Results database. 2011. Available at: http://seer.cancer.gov/statistics/. Accessed February 4, 2013. 2. National Oesophago-Gastric Cancer Audit. Third Annual Report, 2010 An audit of the care received by patients with Oesophago-Gastric Cancer
8 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
23.
24.
in England and Wales. https://catalogue.ic.nhs.uk/publications/clinical/ oesophago-gastric/nati-clin-audi-supp-prog-oeso-gast-canc-2010/clinaudi-supp-prog-oeso-gast-2010-rep1.pdf. Accessed January 25, 2013. Office for National Statistics. Cancer Survival by Cancer Network, England: Patients diagnosed 1997-2010 and followed up to 2011. http://www.ons.gov.uk/ons/dcp171778_292519.pdf. Accessed January 25, 2013. Watson A, Heading RC, Shepherd NA, et al. A report of the Working Party of the British Society of Gastroenterology. Guidelines for the diagnosis and management of Barrett’s columnar-lined esophagus. 2005. http://www.bsg.org.uk/images/stories/docs/clinical/ guidelines/oesophageal/Barretts_Oes.pdf. Accessed January 6, 2013. Vieth M, Ell C, Gossner L, et al. Histological analysis of endoscopic resection specimens from 326 patients with Barrett’s esophagus and early neoplasia. Endoscopy 2004;36:776-81. Kendall C, Isabelle M, Bazant-Hegemark F, et al. Vibrational spectroscopy: a clinical tool for cancer diagnostics. Analyst 2009;134:1029-45. Stone N, Kendall C, Smith J, et al. Raman spectroscopy for identification of epithelial cancers. Faraday Discuss 2004;126:141-57; discussion 169-183. Teh SK, Zheng W, Ho KY, et al. Near-infrared Raman spectroscopy for early diagnosis and typing of adenocarcinoma in the stomach. Br J Surg 2010;97:550-7. Almond LM, Hutchings J, Shepherd N, et al. Raman spectroscopy: a potential tool for early objective diagnosis of neoplasia in the oesophagus. J Biophotonics 2011;4:685-95. Shetty G, Kendall C, Shepherd N, et al. Raman spectroscopy: elucidation of biochemical changes in carcinogenesis of esophagus. Br J Cancer 2006;94:1460-4. Crow P, Barrass B, Kendall C, et al. The use of Raman spectroscopy to differentiate between different prostatic adenocarcinoma cell lines. Br J Cancer 2005;92:2166-70. Jerjes WK, Upile T, Wong BJ, et al. The future of medical diagnostics. Head Neck Oncol 2011;23:3-38. Kendall C, Stone N, Shepherd N, et al. Raman spectroscopy, a potential tool for the objective identification and classification of neoplasia in Barrett’s esophagus. J Pathol 2003;200:602-9. Wong Kee Song LM, Molckovsky A, Wang KK, et al. Diagnostic potential of Raman spectroscopy in Barrett’s esophagus. SPIE 5692, Advanced Biomedical and Clinical Diagnostic Systems III 2005;140. doi:10.1117/12.584986. Bergholt MS, Zheng W, Lin K, et al. In vivo diagnosis of esophageal cancer using image-guided Raman endoscopy and biomolecular modeling. Technol Cancer Res Treat 2011;10:103-12. Hutchings J, Kendall C, Shepherd N, et al. Evaluation of linear discriminant analysis for automated Raman histological mapping of esophageal high-grade dysplasia. J Biomed Opt 2010;15:066015. Shaheen NJ, Richter JE. Barrett’s esophagus. Lancet 2009;373:850-61. Kerkhof M, van Dekken H, Steyerberg EW, et al. Grading of dysplasia in Barrett’s esophagus: substantial interobserver variation between general and gastrointestinal pathologists. Histopathology 2007;50: 920-7. Reid BJ, Haggitt RC, Rubin CE, et al. Observer variation in the diagnosis of dysplasia in Barrett’s esophagus. Hum Pathol 1988;19:166-78. Bergholt MS, Zheng W, Lin K, et al. Characterizing variability in in vivo Raman spectra of different anatomical locations in the upper gastrointestinal tract toward cancer detection. J Biomed Opt 2011;16:037003. Almond LM, Barr H. Advanced endoscopic imaging in Barrett’s esophagus. Int J Surg 2012;10:236-41. Bennett C, Vakil N, Bergman J, et al. Consensus statements for management of Barrett’s dysplasia and early-stage esophageal adenocarcinoma, based on a Delphi process. Gastroenterology 2012;143:336-46. Almond LM, Hutchings J, Kendall C. Preclinical evaluation of a Raman spectroscopic probe for endoscopic classification of oesophageal pathologies. Proc SPIE 2012;8219:90-7. Curvers W, Baak L, Kiesslich R, et al. Chromoendoscopy and narrow-band imaging compared with high-resolution magnification
www.giejournal.org
Almond et al endoscopy in Barrett’s esophagus. Gastroenterology 2008;134: 670-9. 25. Sharma P, Bansal A, Mathur S, et al. The utility of a novel narrow band imaging endoscopy system in patients with Barrett’s esophagus. Gastrointest Endosc 2006;64:167-75. 26. Huang Z, Bergholt MS, Zheng W, et al. In vivo early diagnosis of gastric dysplasia using narrow-band image-guided Raman endoscopy. J Biomed Opt 2010;15:037017. 27. Crow P, Molckovsky A, Stone N, et al. Assessment of fiberoptic nearinfrared Raman spectroscopy for diagnosis of bladder and prostate cancer. Urology 2005;65:1126-30.
www.giejournal.org
Spectroscopic diagnosis of esophageal dysplasia
University of Bristol, Bristol (4), Department of Histopathology, Warwick Hospital, Warwick (5), Department of Upper Gastrointestinal Surgery, Worcestershire Hospitals NHS Trust, Worcester (6), Department of Physics, Exeter University, Exeter, United Kingdom (7). Reprint requests: L. Max Almond, Biophotonics Research Unit, Leadon House, Gloucestershire Royal Hospital, Great Western Road, Gloucester GL1 3NN, United Kingdom. If you would like to chat with an author of this article, you may contact Dr Almond at [email protected].
Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 9
Endoscopic Raman spectroscopy enables objective diagnosis of dysplasia in Barrett’s esophagus L. Max Almond, MB, ChB, MRCS, DM,1,2 Jo Hutchings, BSc, PhD,1 Gavin Lloyd, BSc, PhD,1 Hugh Barr, MD, ChM, FRCS, FRCE, FHEA,1,2 Neil Shepherd, DM, FRCPath,3 John Day, BSc, MSc, PhD,4 Oliver Stevens, BA, MEng,4 Scott Sanders, MD, FRCPath,5 Martin Wadley, MB, ChB, MD, FRCS,2,6 Nick Stone, BSc (Hons), MSc (Dist), MSc (Dist), PGDip (IPSM), PhD, MBA, CSci, FIPEM,7 Catherine Kendall, BSc, MSc, PhD1 Gloucester, Warwick, Bristol, Exeter, United Kingdom
Background: Early detection and targeted endoscopic resection of Barrett’s esophagus–associated high-grade dysplasia (HGD) can prevent progression to invasive esophageal malignancy. Raman spectroscopy, a highly sophisticated analytical technique, has been translated into an endoscopic tool to facilitate rapid, objective diagnosis of dysplasia in the esophagus. Objective: To evaluate the ability of endoscopic Raman spectroscopy (ERS) to objectively detect esophageal HGD and adenocarcinoma. Design: A total of 798 one-second spectra were measured from 673 ex vivo esophageal tissue samples, collected from patients with Barrett’s esophagus by using a novel endoscopic Raman probe. Spectra were correlated with consensus histopathology. Multivariate analysis was used to evaluate the classification accuracy of ERS ex vivo. Setting: Probe measurements were conducted in the laboratory. Tissue specimens were collected from the operating theatre and endoscopy unit. Patients: Tissue from 62 patients was included in the study. Interventions: Endoscopic biopsy/resection or esophagectomy was performed where indicated clinically. Main Outcome Measurement: Diagnostic performance of ERS for detection of HGD and esophageal adenocarcinoma. Results: ERS demonstrated a sensitivity of 86% and a specificity of 88% for detecting HGD and adenocarcinoma. The ability to grade dysplasia and differentiate intestinal metaplasia from nonintestinal metaplasia columnar-lined esophagus was also demonstrated. Diagnostic classification was based on objective measurement of the biochemical profile of different tissue types. The potential for combination ERS and narrow-band imaging was also demonstrated. Limitations: Measurements were taken from ex vivo tissue. Conclusion: ERS enables rapid, accurate, objective diagnosis of superficial esophageal disease (metaplasia, dysplasia, intramucosal cancer) in clinically applicable time scales. (Gastrointest Endosc 2013;-:1-9.)
Abbreviations: ER, endoscopic resection; ERS, endoscopic Raman spectroscopy; HGD, high-grade dysplasia; IM, intestinal metaplasia; LGD, low-grade dysplasia; NBI, narrow-band imaging; NDBE, nondysplastic Barrett’s esophagus; PC, principal component; RS, Raman spectroscopy; WLE, white light endoscopy. DISCLOSURE: The authors disclosed no financial relationships relevant to this publication. This work was funded by the National Institute for Health Research (NIHR) (N.S. J.H.), the Royal College of Surgeons of England (L.M.A.), and the McIntyre Research Fund (L.M.A.).
www.giejournal.org
Copyright ª 2013 by the American Society for Gastrointestinal Endoscopy 0016-5107/$36.00 http://dx.doi.org/10.1016/j.gie.2013.05.028 Received February 7, 2013. Accepted May 22, 2013. Current affiliations: Biophotonics Research Unit (1), Departments of Esophagogastric Surgery (2) and Histopathology (3), Gloucestershire Hospitals NHS Foundation Trust, Gloucester; Interface Analysis Centre, (footnotes continued on last page of article)
Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 1
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
Barrett’s esophagus represents a metaplastic transformation from a normal squamous esophageal epithelium to a columnar-lined esophagus, which occurs in response to chronic pulsatile reflux of gastric acid and bile into the distal esophagus. Barrett’s esophagus confers an increased risk of malignant degeneration via precancerous dysplasia. However, despite this potentially curable premalignant phase, the incidence of esophageal adenocarcinoma has increased by 500% over the past 40 years, although its prognosis has remained relatively constant with current 5-year survival rates of just 8% to 15%.1-3 Currently, patients with nondysplastic Barrett’s esophagus (NDBE) are surveyed endoscopically at intervals of 2 to 3 years with adherence to a strict biopsy protocol consisting of quadrantic biopsies every 2 cm of Barrett’s mucosa.4 The need for this random-sampling approach reflects difficulties in the identification of early neoplastic lesions in Barrett’s segments, causing missed dysplasia in as many as 57% of patients and limiting our ability to perform targeted R0 endoscopic resection of suspicious lesions.5 A novel, custom-built, Raman spectroscopic probe that can be passed down the instrument channel of a gastroscope was developed in an attempt to improve endoscopic detection and targeted treatment of dysplasia and intramucosal (T1a) cancer of the esophagus. The probe uses a well-established analytical technique termed Raman spectroscopy (RS) to interrogate the biochemical composition of its target. RS is highly suitable for translation into clinical use because Raman signals are highly specific to molecular constituents of cells and tissues.6-12 Furthermore, they can be acquired in clinically applicable time scales from fresh, unprocessed tissue without the need for exogenous contrast. In vitro data have demonstrated that RS is capable of objective diagnosis of 7 pathologies in the distal esophagus including high-grade dysplasia (HGD), adenocarcinoma, and NDBE.13 The custom-built probe transmits a low-power, monochromatic, near-infrared laser beam that excites biochemical vibrations in the esophageal epithelium. Inelastically scattered laser photons are preferentially detected and analyzed to derive an objective biochemical profile of the tissue being interrogated. Several groups are in the process of translating RS into an endoscopic tool for use in the esophagus.9,14,15 However, previous probe designs have concentrated on maximizing signal uptake rather than ensuring precise interrogation of the epithelium without spectral contributions from deeper tissues. The novel probe described here is the first of its type with a confocal design ensuring a sampling depth on the order of 150 mm in an effort to detect early superficial lesions that are potentially curable by endoscopic therapy and reduce spectral contamination by basal cells that have been shown to have a spectral signature similar to that of dysplasia.16 The ability for a rapid, objective diagnosis of early Barrett’s esophagus–associated neoplasia using a Raman 2 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
Take-home Message A novel confocal Raman probe can differentiate epithelial and mucosal disease including high-grade dysplasia (HGD) and intramucosal cancer from low-grade dysplasia and nondysplastic Barrett’s esophagus with high diagnostic accuracy in clinically applicable time scales (1 second). Detection of subepithelial “buried” neoplasia is possible by using endoscopic Raman spectroscopy (ERS). ERS could enable targeted endoscopic resection of HGD and intramucosal cancer associated with Barrett’s esophagus before whole-segment ablation therapy.
probe ex vivo is evaluated against a consensus histological criterion standard. The potential for multimodality imaging by using endoscopic RS (ERS), narrow-band imaging (NBI), and white light endoscopy (WLE) is also addressed. In addition, the biochemical basis of spectral classification between benign and neoplastic esophageal tissues is evaluated.
MATERIALS AND METHODS Raman probe system The custom-built fiberoptic endoscopic Raman probe transmits 830 nm of light generated by an Innovative Photonics System (Princeton, NJ) laser source. After tissue interaction, inelastically scattered laser photons are transmitted back along the probe to a Renishaw system 100 spectrometer (Renishaw PLC, Wotton-under-Edge, Gloucestershire, UK) optimized for near-infrared signals. The probe has a diameter of 2.7 mm, enabling easy passage through the instrument channel of an endoscope. It is coupled to an optical cable containing the input and output fibers. Built-in optical filters remove elastically scattered signal and tissue fluorescence to optimize collection of inelastic Raman signals. The probe is designed for direct tissue contact and has confocal delivery and collection optics, ensuring interrogation of the epithelium to a depth of approximately 150 mm. Power output at the probe tip was regulated to approximately 60 mW during all measurements. An expert histopathologist verified that no histological damage was visible at this power level.
Sample preparation and spectral measurements Ethical approval for the study was granted by Gloucestershire Local Research Ethics Committee, Gloucestershire, UK. All patients provided informed consent to permit the use of their esophageal tissue for ex vivo spectral measurements. Eligible patients included those undergoing esophagectomy for esophageal adenocarcinoma or multifocal HGD or surveillance endoscopy for Barrett’s esophagus. Patients with squamous cell carcinoma or squamous dysplasia were www.giejournal.org
Almond et al
Spectroscopic diagnosis of esophageal dysplasia
Figure 1. Unprocessed 1-second Raman probe spectra measured from Barrett’s esophagus, adenocarcinoma, and normal squamous esophagus. Spectra have been offset for clarity. An air background spectrum is shown for comparison, and prominent spectral features have been identified.
excluded. Spectra measured from nonhomogeneous tissue samples containing more than 1 pathology type were excluded from the subsequent analysis. After Ivor-Lewis esophagectomy, the surgical specimen was opened within minutes of resection, and a Perspex grid was fixed across the mucosal surface. Each grid hole was 3 mm in diameter, enabling spectral measurements to be taken by passing the probe through the grid at specified positions and gently apposing it to the esophageal mucosa to mimic the clinical environment at endoscopy. Raman spectra were acquired using 5-, 1-, and 0.1second acquisition times across a wave number range of 400 to 1850 cm 1. Biopsy specimens were then taken at each grid position by using endoscopy biopsy forceps passed through the grid so that spectral signatures could be correlated with the histopathological diagnosis. Biopsy specimens were immediately fixed in formalin and sent for routine histological processing and expert reporting. To augment the spectral dataset, endoscopic point biopsy and endoscopic resection (ER) specimens were also used for spectral analysis. After endoscopic collection, these specimens were immediately snap-frozen in liquid nitrogen and transported to the laboratory. After thawing, 3 spectra were measured per ER specimen, and 1 spectrum per point biopsy specimen. Care was taken to ensure that specimens were oriented with their epithelial surface upward. Samples were then formalin fixed, processed, and reported. The histology of all tissue samples was reported by 2 independent expert GI pathologists from different centers. Tissue specimens with a consensus histological diagnosis were used as the criterion standard for comparison with Raman probe performance.
Trimodality imaging feasibility assessment The potential for combining ERS, NBI, and WLE was also investigated. One-second Raman spectra were measured www.giejournal.org
from a small subset of fresh esophageal biopsy specimens in conjunction with NBI. For comparison, spectra were also acquired in the presence of standard WLE and after extrusion of all ambient light (dark control).
Statistical analysis Multivariate analyses were used to extract the subtle spectral differences between esophageal tissue types. Spectral data were initially normalized and mean centered. Both green glass correction and air background subtraction algorithms were also used initially, although neither significantly enhanced diagnostic classification performance and were therefore excluded from subsequent analyses. Principal component fed linear discriminant analysis was used to develop a classification model in Matlab (Mathworks Inc, Natick, Mass). Principal components were generated that simplified the data while retaining the spectral information required for diagnostic classification. Linear discriminant analysis incorporated histological information to describe functions that enabled spectral classification.16 Leave-one-tissue-site-out cross-validation was used to test the diagnostic performance of the probe against the consensus pathological criterion standard diagnosis.
RESULTS A total of 2226 spectra were measured from esophageal tissue from 62 patients over a 20-month period. Consensus histological diagnoses were available for 798 one-second spectra acquired from 673 ex vivo esophageal tissue samples. The majority (651/798 [82%]) of these spectra were measured directly from fresh esophageal mucosa after Ivor-Lewis esophagectomy. Of 798 spectra, 147 (18%) were acquired from endoscopic biopsy or endoscopic Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 3
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
TABLE 1. The diagnostic performance of the Raman probe is indicated Sensitivity, %
Specificity, %
NSq (251)
87
96
LGD/NDBE (135)
72
91
HGD/adenocarcinoma (199)
86
88
NSq (225)
80
96
LGD/NDBE (80)
71
86
HGD/adenocarcinoma (189)
83
90
NSq (26)
96
98
LGD/NDBE (55)
96
98
HGD/adenocarcinoma (10)
96
97
IM (100)
91
80
Non-IM CLO (CB/FB) (14)
80
91
NSq (251)
90
78
Buried adenocarcinoma (74)
78
90
LGD (21)
100
100
HGD/IMC (10)
100
100
HGD/adenocarcinoma (199)
94
91
NSq (225)
91
94
NDBE (114)
96
92
Gastric (12)
92
96
NDBE/LGD (49)
98
86
HGD/adenocarcinoma (89)
86
98
3-group model (spectra acquired in 1 s), whole dataset
3-group model (spectra acquired in 1 s); fresh, surgically resected samples only
3-group model (spectra acquired in 1 s); snap-frozen, endoscopic samples only
2-group models (spectra acquired in 1 s)
2-group models (spectra acquired in 5 s)
NSq, Normal squamous; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett’s esophagus; HGD, high-grade dysplasia; IM, intestinal metaplasia; CLO, columnar-lined esophagus; CB, cardiac Barrett’s; FB, fundic Barrett’s; IMC, intramucosal cancer (number of homogeneous samples of each pathology). All results were statistically cross-validated by using leave-one-tissue-site-out cross-validation.
resection specimens that had been snap-frozen in the endoscopy department. Figure 1 illustrates 1-s Raman spectra acquired from normal squamous esophagus, NDBE, and adenocarcinoma. An air background spectrum is included for comparison, demonstrating unwanted spectral artifacts caused by elastic scattering and fluorescence generated within the probe fibers. A 3-group principal component (PC)–fed linear discriminant classification model was developed by using consensus normal squamous, NDBE/low-grade dysplasia (LGD), and high-grade dysplasia (HGD)/adenocarcinoma samples. Leave-one-tissue-site-out cross-validation demonstrated a Raman probe sensitivity of 86% and specificity of 88% for correctly identifying HGD/adenocarcinoma samples from the other pathologies (Table 1). Figure 2 4 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
illustrates a scatterplot of cross-validated linear discriminant scores for each Raman probe prediction, which visually demonstrates the diagnostic capability of the Raman probe to separate these pathology groups in clinically applicable (1-second) time scales. The diagnostic accuracies attained using 5-second spectral acquisition times are illustrated in Table 1 for comparison. Table 1 also illustrates the diagnostic classification of the 18% of spectra acquired from endoscopic biopsy/resection tissue samples. Sensitivities and specificities of 96% to 98% were attained for all pathologies included in this dataset. Analysis of spectra measured solely from esophagectomy specimens is also illustrated. Two-group diagnostic classification of LGD and HGD/ intramucosal cancer demonstrated a sensitivity and specificity of 100% for the diagnosis of advanced neoplasia www.giejournal.org
Almond et al
Figure 2. Three-group linear discriminant score prediction scatterplot identifying samples that were measured from fresh (stars) and from fresh-frozen (circles) tissue samples.
because classification of all spectra matched the consensus criterion standard diagnosis (Table 1). The absence of any misclassification in this model likely reflects the relatively small numbers of homogeneous LGD and HGD/intramucosal cancer samples. However, these data certainly demonstrate the ability of ERS to detect and discriminate epithelial pathology. In addition, Raman probe measurements were able to discriminate intestinal metaplasia (IM) from cardiac metaplasia and fundic metaplasia with a sensitivity and specificity of 91% and 80%, respectively. IM, which is characterized histologically by goblet cells, has a greater malignant potential than cardic or fundic metaplasia and is considered necessary for a diagnosis of Barrett’s esophagus in the United States. Similarly, a sensitivity of 96% and specificity of 92% were demonstrated for correctly identifying Barrett’s esophagus from gastric mucosa (Table 1). Table 1 also illustrates that buried cancerous elements beneath normal squamous mucosa were detectable for various possible reasons including changes in localized biochemical expression in the surface epithelium covering the adenocarcinoma and potentially the transmission and collection of some light from beneath the surface epithelium. Diagnostic performance was shown to increase when using consensus pathology diagnoses to train the classification models (Table 2). Sensitivity for the detection of HGD/ intramucosal cancer increased from 70%-76% to 86%, and specificity increased from 85% to 88%. The level of agreement between Raman predictions and criterion standard (consensus) histopathological diagnosis for the 2-group model (NDBE/LGD and HGD/adenocarcinoma) was k Z 0.70. The level of agreement between the 2 independent pathologists for all samples was k Z 0.67. Further analysis was undertaken to evaluate the biochemical basis of diagnostic spectral classification. Linear www.giejournal.org
Spectroscopic diagnosis of esophageal dysplasia
discriminant loads generated after linear discriminant analysis identified 4 PCs as being particularly responsible for spectral classification: PCs 8 and 9 were most influential for differentiating normal squamous tissue from metaplastic and neoplastic tissue (Fig. 3), and PCs 6 and 11 were important for differentiating NDBE and LGD from advanced neoplasia (Fig. 4). Further examination of PCs 8 and 9 (Fig. 3) reveals prominent peaks at 490, 933, 1086 to 1110, and 1128 cm 1, all of which are consistent with a higher glycogen concentration in normal squamous epithelium compared with both Barrett’s and neoplastic tissue. This is entirely consistent with previous in vitro biomolecular modeling studies.9,10,15,20 Reference to peak assignment tables enables assignment of other prominent peaks at 638, 731, 782, 885, 1021, 1095, 1334, and 1621 cm 1 to nucleic acids and protein structures, including histone, which imply alterations in RNA and DNA concentrations between normal squamous and metaplastic/ neoplastic tissue. Plotting of PCs 6 and 11 also reveals prominent glycogen peaks at 492, 574, and 937 cm 1. The peaks at 771, 782, 1018, 1360, and 1511 cm 1 are indicative of nucleic acids, which again reflects increases in DNA and RNA concentrations between metaplastic and malignant tissues. Although the PCs illustrated partially account for the ability of the model to classify spectra, it must be remembered that a significant amount of important information is described by the other PCs. There are therefore many other biochemical peaks that contribute to spectral classification. Many of these were identified previously through preliminary proof-of-concept in vitro work. The potential for trimodality (ERS, NBI, and WLE) imaging was investigated using a small subset of endoscopic biopsy samples. One-second Raman spectra were measured from 11 fresh esophageal biopsy specimens in conjunction with NBI, WLE, and after extrusion of all ambient light (dark control). Figure 5 illustrates that all spectra acquired in conjunction with NBI lie well within a 95% confidence interval of the dark control measurements, demonstrating feasibility for a trimodality ERS-NBI-WLE diagnosis.
DISCUSSION Patients with Barrett’s esophagus, particularly those with dysplasia, are at a significantly increased risk of progression to adenocarcinoma. Detection of early neoplastic disease in these patients could facilitate prompt endoscopic intervention to eradicate HGD and intramucosal (T1a) cancer, preventing disease progression.21 A recent consensus statement from an international consortium of Barrett’s experts has supported an approach of targeted ER of focal HGD lesions followed by whole Barrett’s segment eradication by using radiofrequency ablation.22 This approach aims to ensure accurate disease staging as well as resection of all microscopic disease (R0) followed by eradication of any malignant field change. However, in Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 5
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
TABLE 2. The impact on diagnostic probe performance when using consensus pathology as the criterion standard diagnosis Sensitivity, %
Specificity, %
NSq
84
96
NDBE/LGD
68
86
HGD/adenocarcinoma
76
85
NSq
84
95
NDBE/LGD
72
86
HGD/adenocarcinoma
70
85
NSq
87
96
NDBE/LGD
72
91
HGD/adenocarcinoma
86
88
Pathologist 1
Pathologist 2
Consensus pathology
NSq, Normal squamous; NDBE, nondysplastic Barrett’s esophagus; LGD, low-grade dysplasia; HGD, high-grade dysplasia.
Figure 3. Principal component (PC) 8 and PC 9 loadings after principal component analysis on spectra measured from normal squamous, nondysplastic Barrett’s esophagus, low-grade dysplasia, high-grade dysplasia, and adenocarcinoma tissue.
Figure 4. Principal component (PC) 6 and PC 11 loadings after principal component analysis on spectra measured from normal squamous, nondysplastic Barrett’s esophagus, low-grade dysplasia, high-grade dysplasia, and adenocarcinoma tissue.
6 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
www.giejournal.org
Almond et al
Figure 5. Variation in tissue spectra acquired in the presence of white light endoscopy and narrow-band imaging (NBI) compared with those acquired after exclusion of all ambient light. Confidence intervals (95% and 99%) for the dark control are illustrated. PC, principal component.
a recent study of 742 ERs from 326 patients with Barrett’s neoplasia, an R0 resection was attained in only 26.8% of patients after initial therapy.22 Subsequent repeated ER achieved R0 resections in 74.5%, but only after a delay of several months between treatments. The study concluded that if ER is to be used as a viable therapeutic strategy in HGD and intramucosal cancer, improved mechanisms of lesion detection and margin definition must be developed. Despite a variety of advanced endoscopic imaging tools including NBI, chromoendoscopy, and autofluorescence imaging, the detection of dysplastic lesions in the esophagus remains difficult and extremely subjective. As many as 57% of Barrett’s esophagus–associated dysplasia may be missed by standard biopsy protocols in conjunction with WLE, and high-quality (R0) targeted endoscopic treatment is difficult.22 Development of improved techniques for endoscopic detection of esophageal neoplasia re-mains an essential ongoing area of translational research. This paper describes the application of a novel fiberoptic Raman probe that could enable rapid, objective diagnosis and targeting of Barrett’s esophagus–associated dysplasia and intramucosal adenocarcinoma without the need for exogenous contrast. The probe has demonstrated the ability to differentiate advanced neoplasia (HGD/ adenocarcinoma) from LGD, NDBE, and normal squamous esophagus with a sensitivity of 86% and specificity of 88%. These results are based on 1-second spectral measurements and were statistically cross-validated to avoid reporting of overly optimistic classification performance. It must also be considered that progression from NDBE to adenocarcinoma represents a progressive series of molecular and biochemical changes that manifest as progressive histological grades of dysplasia. There may be a considerable www.giejournal.org
Spectroscopic diagnosis of esophageal dysplasia
spectrum of molecular abnormalities that are consistent with a particular histological grade of dysplasia, and therefore spectral classification performance may be underestimated when based on correlation with histological appearances. Adenocarcinoma could also be differentiated from normal squamous esophagus with a sensitivity of 94% and a specificity of 91%, demonstrating potential for use in differentiating benign and malignant esophageal strictures. In addition, ERS could distinguish Barrett’s esophagus from gastric mucosa with a sensitivity of 96% and specificity of 92%. This distinction may be of particular importance in patients with an irregular z-line who may be incorrectly labeled with a diagnosis of Barrett’s esophagus. An ability to distinguish Barrett’s esophagus and gastric mucosa may also be of use in patients with hiatal hernias in whom the exact position of the esophagogastric junction is unclear. This is the first time that a Raman probe has demonstrated the ability to detect superficial neoplastic disease, as well as to distinguish IM from non-IM columnar-lined esophagus and Barrett’s esophagus from gastric mucosa. In addition, the ability to detect submucosally invasive cancer has also been reported, suggesting potential for detection of buried metaplasia during postablation surveillance. Similarly, there may also be a potential role in immediate objective intraoperative margin assessment during esophagectomy, although this will require further clinical evaluation. Histological diagnosis of dysplastic Barrett’s esophagus known to have considerable interreporter variation.12,17-19 Because incorrect histological diagnoses could confuse the assessment of probe performance, the analyses were repeated by using the pathology results provided by each of the independent pathologists and a consensus of the 2. Diagnostic performance was found to increase when using consensus pathology diagnoses to train the classification models, intimating that performance could have improved further by inclusion of an additional pathologist to further optimize the criterion standard (Table 2). It is possible that 1-second Raman probe measurements may be capable of at least matching the performance of an individual pathologist. It is also notable that diagnostic accuracy for detecting advanced neoplasia increased when LGD samples were removed from the analysis, a group for which it is known that pathologists have particular difficulty agreeing on. Spectral diagnosis was based on objective detection of identifiable biochemical differences between benign and neoplastic tissues that concur with the published literature. This avoids the subjectivity associated with histological reporting; in fact, spectral classification was shown to increase when consensus histological diagnoses were used as the criterion standard.13 Current advanced endoscopic tools such as confocal microscopy and NBI lack this objectivity. Additional data not presented here have confirmed no significant difference in spectra acquired by 2 independent operators using 2 identical probes.23 The ability to record Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 7
Spectroscopic diagnosis of esophageal dysplasia
Almond et al
consistent reproducible spectra by using a single probe over a 20-month period has also been demonstrated.23 The feasibility of combination ERS, NBI, and WLE has been highlighted for the first time, although more data will be necessary to clearly evaluate this potential. NBI is a widely available imaging adjunct that enhances the resolution of the mucosal surface, potentially improving detection of dysplasia and intramucosal cancer in the hands of experienced users. However, considerable interuser variation and high false-positive rates are widely reported.24,25 ERS would complement a wide-field imaging technique such as NBI, which has subjectivity and poor specificity.26 A dual-modality (ERS-NBI) imaging system could potentially limit the high false-positive rate associated with NBI and provide objective diagnostic assessment to reduce interuser variation. A limitation of this study is that spectra were acquired from ex vivo tissue samples. These samples had been rendered ischemic, which may have had a small impact on their spectral signal, although it should be noted that previous in vivo studies at other body sites have not noted significant spectral changes between ex vivo and in vivo measurements.27 In addition, most of the esophagectomy specimens were from patients who had received neoadjuvant chemotherapy before resection. Shrinkage and fibrosis of the tumor, which may have altered the biochemical composition of the mucosa, could account for the lower diagnostic accuracies obtained from surgically resected tissue compared with those reported from endoscopy specimens, suggesting that the true diagnostic accuracies of HGD and adenocarcinoma might be 96% to 98%, as demonstrated after analysis of endoscopy samples alone (Table 1). In vivo trials will be required to address some of these issues, and preparation for this is under way. ERS is a highly sophisticated analytical technique that could provide rapid objective in vivo diagnosis of a range of esophageal pathologies during endoscopy without the need for staining or exogenous contrast administration. A Raman probe could be used to confirm dysplastic areas within Barrett’s segments. In addition, it could facilitate careful lesion margin assessment to ensure high-quality R0 endoscopic resections. Multiple probes are currently undergoing construction, and engineering tolerance levels are being carefully defined before a clinical trial of ERS in patients with Barrett’s esophagus. If successful, it is hoped that wider clinical implementation in tertiary units with facilities for esophageal ER will be possible in the near future.
3.
4.
5.
6. 7.
8.
9.
10.
11.
12. 13.
14.
15.
16.
17. 18.
19. 20.
21. 22.
REFERENCES 1. National Cancer Institute’s Surveillance, Epidemiology, and End Results database. 2011. Available at: http://seer.cancer.gov/statistics/. Accessed February 4, 2013. 2. National Oesophago-Gastric Cancer Audit. Third Annual Report, 2010 An audit of the care received by patients with Oesophago-Gastric Cancer
8 GASTROINTESTINAL ENDOSCOPY Volume
-,
No.
-
: 2013
23.
24.
in England and Wales. https://catalogue.ic.nhs.uk/publications/clinical/ oesophago-gastric/nati-clin-audi-supp-prog-oeso-gast-canc-2010/clinaudi-supp-prog-oeso-gast-2010-rep1.pdf. Accessed January 25, 2013. Office for National Statistics. Cancer Survival by Cancer Network, England: Patients diagnosed 1997-2010 and followed up to 2011. http://www.ons.gov.uk/ons/dcp171778_292519.pdf. Accessed January 25, 2013. Watson A, Heading RC, Shepherd NA, et al. A report of the Working Party of the British Society of Gastroenterology. Guidelines for the diagnosis and management of Barrett’s columnar-lined esophagus. 2005. http://www.bsg.org.uk/images/stories/docs/clinical/ guidelines/oesophageal/Barretts_Oes.pdf. Accessed January 6, 2013. Vieth M, Ell C, Gossner L, et al. Histological analysis of endoscopic resection specimens from 326 patients with Barrett’s esophagus and early neoplasia. Endoscopy 2004;36:776-81. Kendall C, Isabelle M, Bazant-Hegemark F, et al. Vibrational spectroscopy: a clinical tool for cancer diagnostics. Analyst 2009;134:1029-45. Stone N, Kendall C, Smith J, et al. Raman spectroscopy for identification of epithelial cancers. Faraday Discuss 2004;126:141-57; discussion 169-183. Teh SK, Zheng W, Ho KY, et al. Near-infrared Raman spectroscopy for early diagnosis and typing of adenocarcinoma in the stomach. Br J Surg 2010;97:550-7. Almond LM, Hutchings J, Shepherd N, et al. Raman spectroscopy: a potential tool for early objective diagnosis of neoplasia in the oesophagus. J Biophotonics 2011;4:685-95. Shetty G, Kendall C, Shepherd N, et al. Raman spectroscopy: elucidation of biochemical changes in carcinogenesis of esophagus. Br J Cancer 2006;94:1460-4. Crow P, Barrass B, Kendall C, et al. The use of Raman spectroscopy to differentiate between different prostatic adenocarcinoma cell lines. Br J Cancer 2005;92:2166-70. Jerjes WK, Upile T, Wong BJ, et al. The future of medical diagnostics. Head Neck Oncol 2011;23:3-38. Kendall C, Stone N, Shepherd N, et al. Raman spectroscopy, a potential tool for the objective identification and classification of neoplasia in Barrett’s esophagus. J Pathol 2003;200:602-9. Wong Kee Song LM, Molckovsky A, Wang KK, et al. Diagnostic potential of Raman spectroscopy in Barrett’s esophagus. SPIE 5692, Advanced Biomedical and Clinical Diagnostic Systems III 2005;140. doi:10.1117/12.584986. Bergholt MS, Zheng W, Lin K, et al. In vivo diagnosis of esophageal cancer using image-guided Raman endoscopy and biomolecular modeling. Technol Cancer Res Treat 2011;10:103-12. Hutchings J, Kendall C, Shepherd N, et al. Evaluation of linear discriminant analysis for automated Raman histological mapping of esophageal high-grade dysplasia. J Biomed Opt 2010;15:066015. Shaheen NJ, Richter JE. Barrett’s esophagus. Lancet 2009;373:850-61. Kerkhof M, van Dekken H, Steyerberg EW, et al. Grading of dysplasia in Barrett’s esophagus: substantial interobserver variation between general and gastrointestinal pathologists. Histopathology 2007;50: 920-7. Reid BJ, Haggitt RC, Rubin CE, et al. Observer variation in the diagnosis of dysplasia in Barrett’s esophagus. Hum Pathol 1988;19:166-78. Bergholt MS, Zheng W, Lin K, et al. Characterizing variability in in vivo Raman spectra of different anatomical locations in the upper gastrointestinal tract toward cancer detection. J Biomed Opt 2011;16:037003. Almond LM, Barr H. Advanced endoscopic imaging in Barrett’s esophagus. Int J Surg 2012;10:236-41. Bennett C, Vakil N, Bergman J, et al. Consensus statements for management of Barrett’s dysplasia and early-stage esophageal adenocarcinoma, based on a Delphi process. Gastroenterology 2012;143:336-46. Almond LM, Hutchings J, Kendall C. Preclinical evaluation of a Raman spectroscopic probe for endoscopic classification of oesophageal pathologies. Proc SPIE 2012;8219:90-7. Curvers W, Baak L, Kiesslich R, et al. Chromoendoscopy and narrow-band imaging compared with high-resolution magnification
www.giejournal.org
Almond et al endoscopy in Barrett’s esophagus. Gastroenterology 2008;134: 670-9. 25. Sharma P, Bansal A, Mathur S, et al. The utility of a novel narrow band imaging endoscopy system in patients with Barrett’s esophagus. Gastrointest Endosc 2006;64:167-75. 26. Huang Z, Bergholt MS, Zheng W, et al. In vivo early diagnosis of gastric dysplasia using narrow-band image-guided Raman endoscopy. J Biomed Opt 2010;15:037017. 27. Crow P, Molckovsky A, Stone N, et al. Assessment of fiberoptic nearinfrared Raman spectroscopy for diagnosis of bladder and prostate cancer. Urology 2005;65:1126-30.
www.giejournal.org
Spectroscopic diagnosis of esophageal dysplasia
University of Bristol, Bristol (4), Department of Histopathology, Warwick Hospital, Warwick (5), Department of Upper Gastrointestinal Surgery, Worcestershire Hospitals NHS Trust, Worcester (6), Department of Physics, Exeter University, Exeter, United Kingdom (7). Reprint requests: L. Max Almond, Biophotonics Research Unit, Leadon House, Gloucestershire Royal Hospital, Great Western Road, Gloucester GL1 3NN, United Kingdom. If you would like to chat with an author of this article, you may contact Dr Almond at [email protected].
Volume
-,
No.
-
: 2013 GASTROINTESTINAL ENDOSCOPY 9