Raman endoscopy for in vivo differentiation between benign and malignant ulcers in the stomach.

Abstract

The aim of this study was to evaluate the clinical utility of an image-guided Raman endoscopy technique for in vivo differential diagnosis of benign and malignant ulcerous lesions in the stomach. A rapid-acquisition image-guided Raman endoscopy system with 785 nm laser excitation has been developed to acquire in vivo gastric tissue Raman spectra within 0.5 s during clinical gastroscopic examinations. A total of 1102 in vivo Raman spectra were acquired from 71 gastric patients, in which 924 Raman spectra were from normal tissue, 111 Raman spectra were from benign ulcers whereas 67 Raman spectra were from ulcerated adenocarcinoma. There were distinctive spectral differences in Raman spectra among normal mucosa, benign ulcers and malignant ulcers, particularly in the spectral ranges of 800-900, 1000-1100, 1245-1335, 1440-1450 and 1500-1800 cm(-1), which primarily contain signals related to proteins, DNA, lipids and blood. The malignant ulcerous lesions showed Raman signals to be mainly associated with abnormal nuclear activity and decrease in lipids as compared to benign ulcers. Partial least squares-discriminant analysis (PLS-DA) was employed to generate multi-class diagnostic algorithms for classification of Raman spectra of different gastric tissue types. The PLS-DA algorithms together with leave-one tissue site-out, cross validation technique yielded diagnostic sensitivities of 90.8%, 84.7%, 82.1%, and specificities of 93.8%, 94.5%, 95.3%, respectively, for classification of normal mucosa, benign and malignant ulcerous lesions in the stomach. This work demonstrates that image-guided Raman endoscopy technique associated with PLS-DA diagnostic algorithms has for the first time promising clinical potential for rapid, in vivo diagnosis and detection of malignant ulcerous gastric lesions at the molecular level.

Full text

Page 1

Raman endoscopy for in vivo differentiation between benign and malignant
ulcers in the stomach†
Mads Sylvest Bergholt,aWei Zheng,aKan Lin,aKhek Yu Ho,bMing Teh,cKhay Guan Yeoh,b
Jimmy Bok Yan Sodand Zhiwei Huang*a
Received 20th May 2010, Accepted 21st September 2010
DOI: 10.1039/c0an00336k
The aim of this study was to evaluate the clinical utility of an image-guided Raman endoscopy
technique for in vivo differential diagnosis of benign and malignant ulcerous lesions in the stomach. A
rapid-acquisition image-guided Raman endoscopy system with 785 nm laser excitation has been
developed to acquire in vivo gastric tissue Raman spectra within 0.5 s during clinical gastroscopic
examinations. A total of 1102 in vivo Raman spectra were acquired from 71 gastric patients, in which
924 Raman spectra were from normal tissue, 111 Raman spectra were from benign ulcers whereas 67
Raman spectra were from ulcerated adenocarcinoma. There were distinctive spectral differences in
Raman spectra among normal mucosa, benign ulcers and malignant ulcers, particularly in the spectral
ranges of 800–900, 1000–1100, 1245–1335, 1440–1450 and 1500–1800 cm?1, which primarily contain
signals related to proteins, DNA, lipids and blood. The malignant ulcerous lesions showed Raman
signals to be mainly associated with abnormal nuclear activity and decrease in lipids as compared to
benign ulcers. Partial least squares-discriminant analysis (PLS-DA) was employed to generate multi-
class diagnostic algorithms for classification of Raman spectra of different gastric tissue types. The
PLS-DA algorithms together with leave-one tissue site-out, cross validation technique yielded
diagnostic sensitivities of 90.8%, 84.7%, 82.1%, and specificities of 93.8%, 94.5%, 95.3%, respectively,
for classification of normal mucosa, benign and malignant ulcerous lesions in the stomach. This work
demonstrates that image-guided Raman endoscopy technique associated with PLS-DA diagnostic
algorithms has for the first time promising clinical potential for rapid, in vivo diagnosis and detection of
malignant ulcerous gastric lesions at the molecular level.
1. Introduction
Gastric malignancies remain one of the major causes of cancer-
associated death in humans1and are currently one of the most
common malignancies in the world with particularly high inci-
dence rates in East Asia.2Patients with gastric cancer can present
with endoscopic appearance being indistinguishable from benign
peptic ulcer disease.3It has been reported that approximately
1–5% of benign-looking ulcers will develop into gastric cancers.4,5
Due to the fact that the therapeutic outcome of gastric cancer is
highly related to the stage of the disease, early identification of
malignant ulcers is crucial to improving the survival rate of
patients. The visual identification and localization of gastric
ulcers using conventional white-light reflectance (WLR) imaging
is adequate, but the differential diagnosis of malignant and
benign gastric ulcers remains difficult as WLR endoscopic
imaging heavily relies on visualization of gross morphological
tissue changes (e.g., irregularity, nodularity, anatomical loca-
tions etc.)4,6associated with neoplastic transformation. Subtle
and focal cancer tissue in ulcerated lesions may not be apparent,
making it necessary to perform numerous random endoscopic
biopsies of ulcer bases and margins for assessment of tissue
pathology. The malignant ulcer disease is often diagnosed at an
advanced stage even after repeated endoscopic intrusion.7
Gastric tissue biopsies can be small in diameter (?1–3 mm), and
some histopathological features (e.g., the elements of regenerat-
ing epithelium of benign ulcers) can be confused with cancer
cells;5and thus the pathologic examinations might be inconclu-
sive in making the objective diagnostics. As such, the conven-
tional endoscopic routine for assessing the pathology of ulcers is
invasive and impractical for real-time screening of high-risk
patients who may have multiple suspicious lesions in the
stomach.
In recent years, optical spectroscopic imaging methods, such
as autofluorescence imaging (AFI) technique capable of detect-
ing the changes of endogenous fluorophores and morphological
architectures of tissue, and the narrow-band imaging (NBI)
technique which enhances visualization of irregular mucosal and
vascular patterns, have shown promising diagnostic potential
aOpticalBioimaging Laboratory,DepartmentofBioengineering,Facultyof
Engineering, National University of Singapore, 9, Engineering Drive 1,
Singapore 117576. E-mail: biehzw@nus.edu.sg; Fax: +65-6872-3069;
Tel: +65-6516-8856
bDepartment of Medicine, Yoo Loo Lin School of Medicine, National
University of Singapore and National University Hospital, Singapore
119260
cDepartment of Pathology, Yoo Loo Lin School of Medicine, National
University of Singapore and National University Hospital, Singapore
119074
dDepartment of Surgery, Yong Loo Lin School of Medicine, National
University of Singapore and National University Hospital, Singapore
119074
† This article is part of a themed issue on Optical Diagnosis. This issue
includes work presented at SPEC 2010 Shedding Light on Disease:
Optical Diagnosis for the New Millennium, which was held in
Manchester, UK June 26th – July 1st 2010.
3162 | Analyst, 2010, 135, 3162–3168This journal is ª The Royal Society of Chemistry 2010
PAPER www.rsc.org/analyst | Analyst

Page 2

for in vivo detection of preneoplastic and early neoplastic lesions
at endoscopy.8Although AFI and NBI imaging techniques
provide high-detection sensitivities, these wide-field endoscopic
imaging modalities still suffer from moderate diagnostic speci-
ficities, owing to the inter-observer dependence and the lack of
ability to reveal specific biomolecular information about the
tissue. Thus, the development of the rapid and non-invasive
optical diagnostic techniques providing the direct point-wise
assessment of biochemical and histological information from
ulcerated lesions in vivo to complement the wide-field endoscopic
imaging modalities (WLR, NBI and AFI) would be of significant
medical relevance in the early diagnosis of benign and malignant
ulcers at gastroscopy. Optical spectroscopic techniques such as
fluorescence spectroscopy, light scattering spectroscopy and
Raman spectroscopy have been comprehensively investigated for
the evaluation of malignancies in tissues.9–28Raman spectros-
copy is a nondestructive, inelastic light scattering technique in
which the scattered photon is shifted to another wavelength with
respect to the incident excitation light, depending on the specific
vibrational motions of molecules in tissue and cells. Thus Raman
spectroscopy reveals specific biochemical and biomolecular
structures and conformation (i.e., biomolecular signatures) of
tissue, providing the unique opportunity to distinguish between
different pathological tissue types at the molecular level. Under
near-infrared (NIR) laser light excitation (e.g., 785 nm), NIR
Raman spectroscopy has shown great promise for detecting
alterations in diseased tissue in vitro and in vivo with high
biomolecular specificity.17–22,24–26For instance, the diagnostic
sensitivities and specificities in the range of ?85–95% and ?90–
98% respectively have been reported for differentiation between
different pathologic types (e.g., intestinal metaplasia, heli-
cobactor pylori infection, dysplasia and adenocarcinoma types)
of gastric tissue using NIR Raman spectroscopy associated with
multivariate analysis.17,18,21,24–27However, in vivo Raman clinical
applications have been limited not only by the difficulty in
capturing inherently weak tissue Raman signals, but also by the
relatively slow speed of spectral acquisitions.20,23The effective
collection of fiber optic tissue Raman photons presents another
great challenge in clinical Raman endoscopy. Very recently, we
have successfully developed a flexible 1.8 mm bifurcated fiber
optic Raman probe with dual filter coatings deposited on the
fiber tip, that can pass down the instrument channel of most
medical endoscopes for effective tissue excitation and Raman
signal collections at endoscopy.23By integrating the Raman
endoscopic probe with a high throughput dispersion-type
Raman spectrometer,28in vivo tissue Raman measurements in the
stomach during clinical gastroscopy have been realized in real-
time under the guidance of multimodal wide-field imaging
modalities (i.e., WLR, AFI, and NBI). Consequently, the clinical
utility of the in vivo Raman endoscopic diagnosis of gastric
dysplasia24or neoplasia25,26have been demonstrated with
success. However, to date, the in vivo Raman study of gastric
benign and malignant ulcers has not yet been evaluated in detail.
The mucosal degradation (e.g., absence of mucous membrane)
and altered tissue characteristics of ulcer lesions may potentially
develop into malignant tumors in the stomach.4–6The main aim
of this study thus was to investigate the Raman spectral prop-
erties of peptic ulcerous lesions and to assess the clinical utility of
the image-guided NIR Raman endoscopy for distinguishing
between normal mucosa, benign and malignant ulcers in the
gastric in vivo. The multivariate statistical techniques including
partial least squares-discriminant analysis (PLS-DA) were
employed to generate multi-class diagnostic algorithms for
classification of in vivo Raman spectra of different gastric tissue
types.
2. Materials and methods
2.1 Raman endoscopy instrumentation
The integrated Raman spectroscopy and trimodal wide-field
imaging technique developed for in vivo tissue measurements at
endoscopy has been described in detail elsewhere.23,28Briefly, the
Raman spectroscopy system consists of a spectrum stabilized 785
nm diode laser (maximum output: 300 mW, B&W TEK Inc.,
Newark, DE), a transmissive imaging spectrograph (Holospec
f/1.8, Kaiser Optical Systems Inc., Ann Arbor, MI) equipped
with a HSG-785-LF grating, a liquid nitrogen-cooled, NIR-
optimized, back-illuminated and deep depletion charged coupled
device (CCD) camera (1340 ? 400 pixels at 20 ? 20 mm per pixel;
Spec-10: 400BR/LN, Princeton Instruments, Trenton, NJ), and
a specially designed Raman endoscopic probe for both laser light
delivery and in vivo tissue Raman signals collection. The Raman
endoscopic probe is composed of 32 collection fibers surrounding
the central light delivery fiber (200 mm in diameter, N.A. ¼ 0.22)
with two stages of optical filtering incorporated at the proximal
and distal ends of the probe for maximizing the collection of
tissue Raman signals while reducing the interference of Rayleigh
scattered light, fiber fluorescence and silica Raman signals.
Control of the system was implemented by a personal computer
(PC) using a custom-designed program that triggered data
acquisition and analysis (e.g., CCD dark-noise subtraction,
wavelength calibration, system spectral response calibration,
signal saturation detection, cosmic ray rejection, first order
5-pixel Savitzky-Golay smoothing,
background subtraction (5th-order polynomial fit), etc),23as well
as real-time display of in vivo tissue Raman spectra during clin-
ical endoscopic measurements. The system acquired Raman
spectra over the wavenumber range of 800–1800 cm?1, and each
spectrum was measured within an integration time of 0.5 s under
the 785 nm laser excitation power of 1.5 W cm?2. The spectral
resolution of the system is 9 cm?1. All wavelength-calibrated
spectra were corrected for the wavelength-dependence of the
system using a standard lamp (RS-10, EG&G Gamma Scientific,
San Diego, CA). The trimodal endoscopy imaging system
primarily comprises a 300 W short-arc xenon light source,
agastrointestinal(GI)videoendoscope
Olympus), and a video system processor (CV-260SL, Olympus).
The light reflected or fluorescence emitted from tissue are
detected by two monochrome CCD chips mounted behind the
two objective lens placed next to each other at the distal tip of the
GI videoendoscope: one CCD for WLR/NBI and the other one
for AFI. The video system processor converts the signal received
from the CCD in the endoscope into RGB video image for
display on a video monitor. With this unique Raman endoscopy
system developed, wide-field endoscopic images (WLR/AFI/
NBI) and the corresponding real-time in vivo Raman spectra of
the tissue imaged can be simultaneously acquired, displayed and
tissue autofluorescence
(GIF-FQ260Z,
This journal is ª The Royal Society of Chemistry 2010Analyst, 2010, 135, 3162–3168 | 3163

Page 3

recorded in real-time in the video system processor and the PC,
respectively.
2.2 Patients
The present study is part of an ongoing nationwide program
aimed at early diagnosis and treatment of gastric malignancies
run by the Singapore gastric cancer epidemiology, clinical and
genetic program (GCEP).29This study was conducted with
approval by the Institutional Review Board (IRB) of the
National Healthcare Group (NHG) of Singapore. All patients
signed an informed consent permitting the investigative collec-
tion of in vivo gastric Raman spectra in the endoscope centre at
the National University Hospital (NUH), Singapore. The in vivo
Raman spectra were acquired from gastric patients under the
guidance of multimodal wide-field endoscopic imaging modali-
ties (WLR/AFI/NBI). The Raman probe was passed down to the
instrument channel of the endoscope and visible approximately
2 mm in front of the endoscope tip and could precisely be placed
in gentle contact with the tissue surface as verified on the endo-
scope monitor by the clinicians in-charge during gastroscopic
examinations. Note that necrotic debris was displaced when the
Raman probe was touched on the ulcerated surface to reveal the
underlying tissue. Early malignancies are typically located in
the margin of peptic ulcers.
In this study, to establish the implications of ulceration, we
investigate and evaluate the Raman endoscopic differential
diagnosis between benign and malignant ulcerous tissues. The
Raman spectra were acquired from the central bases of all the
active ulcerous lesions extending into muscularis mucosa from
different anatomical sites without visible signs of active bleeding
or blood clotting upon the Raman measurements. All the
malignant ulcerous lesions measured were ulcerated adenocar-
cinomas with an approximate wall thickening of 6 mm (which is
much thicker than the 785 nm laser penetration depth of
?800 mm in gastric tissue19) ensuring that the ulcerous lesions
contained prominent neoplastic transformation (as confirmed by
histopathology). To include inter- and/or intra-tissue variations
caused by the probe positioning for data analysis, multiple
Raman spectra (?6 to 8 spectra) were obtained from each tissue
site. As a result, a total of 1102 in vivo Raman spectra were
obtained from 71 patients (37 men and 34 women with a median
age of 67 years old), in which 924 Raman spectra were from
normal mucosal tissue; while 111 Raman spectra were from
benign ulcerous tissue, and 67 Raman spectra were from malig-
nant ulcerous tissue (adenocarcinoma (intestinal and diffuse
types) as confirmed by histopathology). Immediately after in vivo
Raman acquisitions, the biopsy samples were taken from the
tissue sites measured (with suction markings) and fixed in 10%
formalin solution for histopathological examination by a senior
gastrointestinal pathologist. For the assessment of diagnostic
sensitivity and specificity of Raman endoscopy for tissue classi-
fication in the stomach, histopathological results served as the
gold standard.
2.3 Multivariate statistical analysis
The high dimensionality of the Raman spectral space (each
Raman spectrum ranging from 800–1800 cm?1with a set of
685 intensities) will result in inefficiency in implementation and
optimization of conventional clustering algorithms (e.g., linear
discriminant analysis (LDA), support vector machines (SVM),
etc.).13,26To reduce the dimension of the spectral data, principal
component analysis (PCA) is usually employed to extract a set of
orthogonal principal components (PCs) that account for the
maximum variance in the Raman spectral dataset for tissue
diagnosis and characterization.19,24,30Alternatively, the much
practised partial least squares (PLS)-discriminant analysis
(DA)31–33can advantageously be applied for multi-class classifi-
cation problems by encoding the class membership of zeros and
ones, representing group affinities in an appropriate Y-indicator
matrix. PLS-DA employs the fundamental principle of PCA but
further rotates the components (latent variables (LVs)) by
maximizing the covariance between the spectral variation and
group affinity so that the LVs explain the diagnostic relevant
variations rather than the most prominent variations in the
spectral dataset.31–33In most cases, this ensures that the diag-
nostic significant spectral variations are retained in the first few
LVs. In this study, the performance of the PLS-DA diagnostic
algorithm was validated in an unbiased manner using the leave-
one-tissue site-out, cross validation methodology.34In the vali-
dation procedure, one tissue site (?6–8 Raman spectra) was left
out and the PLS-DA modeling was redeveloped using the
remaining Raman spectra. The redeveloped PLS-DA diagnostic
algorithm was then used to classify the withheld Raman spectra.
This process was repeated iteratively until all withheld Raman
spectra were classified. The number of retained LVs was deter-
mined based on the minimal root mean square error of cross
validation (RMSECV) curves. Multivariate statistical analysis
was performed using the PLS toolbox (Eigenvector Research,
Wenatchee, WA) in the Matlab (Mathworks Inc., Natick, MA)
programming environment.
3.Results
Fig. 1A shows in vivo mean Raman spectra ?1 standard devia-
tions (SD) of normal (n ¼ 924), benign ulcers (n ¼ 111) and
malignant ulcers (n ¼ 67) in the stomach. Prominent Raman
bands are observed in both normal and ulcerated gastric tissue at
the following peak positions with tentative biochemical assign-
ments:10,14,17855 cm?1(v(C–C) of proline), 936 cm?1(v(C–C) of
a-helix conformation for proteins), 1004 cm?1(n(C–C) ring
breathing of phenylalanine), 1080 cm?1(n(C–C) of lipids),
1245–1265 cm?1(amide III n(C–N) and d(N–H) of proteins),
1302 (d(CH2) deformations of lipids and proteins), 1335 cm?1
(CH3CH2 twisting of nucleic acids), 1445 cm?1(d(CH2) of
proteins and lipids), 1575 cm?1(guanine, adenine, heme)
1620 cm?1(n(C]C) of porphyrins), 1655 cm?1(amide I v(C]O)
of proteins, a-helix conformation) and 1745 cm?1(v(C]O) of
phospholipids). The corresponding difference spectra of different
tissue types (Fig. 1B) reveal the significant Raman spectral
changes (i.e., Raman peak intensities, Raman peak positions and
spectral bandwidths), particularly in the spectral ranges of
800–900, 1000–1100, 1245–1335, 1440–1450 and 1500–1800 cm?1
which primarily contain signals related to proteins, DNA, lipids
and blood. Particularly, 6 important sub-regions peaking at
around 875, 1004, 1335, 1620, 1665 and 1745 cm?1in the Raman
spectra were identified among the three tissue types (p < 0.0001,
3164 | Analyst, 2010, 135, 3162–3168This journal is ª The Royal Society of Chemistry 2010

Page 4

one-way ANOVA). Comparison of the Raman intensities ?1 SD
at each of the six identified spectral sub-regions is illustrated in
the bar-chart (Fig. 2). For instance, benign and malignant
ulcerous gastric tissue differ from normal mucosa at 875, 1004,
1620, 1665 and 1745 cm?1; whereas malignant ulcers show lower
intensities at 875 and 1745 cm?1but exhibit much increased
Raman signals at 1302, 1335, 1445 and 1665 cm?1as compared to
benign ulcers (Fig. 1B). These observations indicate that there is
a significant increase and decrease in the percentage of distinctive
biomolecules relative to the total Raman-active constituents in
different tissue types, suggesting the diagnostic potential of
Raman endoscopy for invivo identification of malignant ulcerous
lesions in the stomach at gastroscopy.
To develop diagnostic algorithms for differentiation between
normal mucosa, benign ulcers and malignant ulcers, the stan-
dardized Raman spectra were assembled into a dataset with
wavenumber columns and individual case rows. Prior to data
analysis, the constructed dataset was mean centered to remove
common variance.35The PLS-DA with leave-one-tissue site-out,
cross validation was subsequently employed to generate diag-
nostic algorithms, such that 5 LVs were found to be the optimal
numbers of retained components as defined by the local
minimum of RMSECV indicated in Fig. 3, accounting for 53.8%
of the total Raman spectral variances. Fig. 4 displays the first
three diagnostic significant LV weights (i.e., LV1, LV2 and LV3)
accounting for the largest Raman spectral variance (16.5%,
17.1% and 10.5%) and generally represented variations in the
major Raman peaks (i.e., 875, 936, 1004, 1245, 1335, 1445, 1620,
1655 and 1745 cm?1). Successive components accounted for
distinctive amount of spectral variations (i.e., LV4, 6.8%; LV5,
2.9%).
Fig. 5 displays the three-dimensional score value plot for the
first three LVs illustrating good clustering with respect to tissue
types. A ternary plot36of the Raman cross validated prediction
results was also generated (Fig. 6). This depicts the probabilistic
outcome for each tissue type, providing a three-class diagnostic
model for tissue classification. The final diagnostic category of
each Raman spectrum is determined by the nearest proximity of
the data-point to the diagnostic category related to the vertex
of the ternary plot, representing the 100 percent posterior
probability belonging to either normal mucosa, benign ulcer or
Fig. 1
normal mucosa (n ¼ 924), benign ulcers (n ¼ 111) and malignant ulcers
(n ¼ 67) in the stomach. Note that the mean Raman spectra of ulcer
tissues are vertically shifted for better visualization. (B) The corre-
sponding difference spectra calculated from the mean Raman spectra
between the three different gastric tissue types.
(A) In vivo mean Raman spectra ?1 standard deviations (SD) of
Fig. 2
ANOVA p < 0.0001) of Raman peak intensities (mean ?1 SD) for
normal mucosa (white), benign ulcer (striped), and malignant ulcer
(black).
Histogram displaying the most prominent differences (one-way Fig. 3
variables (LVs) and the root mean square error of cross validation
(RMSECV) for correct classification of normal mucosa, benign ulcers
and malignant ulcers.
The relationship between the number of PLS factors-latent
This journal is ª The Royal Society of Chemistry 2010 Analyst, 2010, 135, 3162–3168 | 3165

Page 5

malignant ulcer. The leave-one-tissue site out, cross validated
diagnostic sensitivities of 90.8%, 84.7%, and 82.1%; and speci-
ficities of 93.8%, 94.5%, and 95.3%, respectively, were achieved
for differentiation among normal mucosa, benign ulcers and
malignant ulcers (Table 1), confirming that the PLS-DA-based
diagnostic algorithm developed based on Raman biomolecular
signals is powerful for classification of in vivo Raman spectra of
different gastric tissue types during clinical gastroscopic exami-
nations.
4. Discussion
Current gold standard for the differential diagnosis of benign
and malignant ulcers relies on histological examination (e.g.,
cytologicaland architectural abnormalities) ofmultiple
endoscopic specimens by the pathologist; however, random
biopsy sampling of ulcerated lesions under WLR endoscopic
procedures is highly invasive and impractical for screening
patients who may have multiple suspicious gastric lesions. The
point-wise optical diagnostic technology which can provide
biochemical and histological information for identifying malig-
nant ulcerated tissue in vivo could be of great clinical value to
assist the clinicians in targeted biopsies and surgery operations.
Raman spectroscopy is a unique vibrational spectroscopic tech-
nique, which can noninvasively capture specific molecular
information for tissue diagnosis and characterization, allowing
the study of tissue in its native state without requiring tissue
preparation or treatment (i.e., free of artifacts introduced by
mechanically cutting, chemical processing). In addition, the good
compatibility of fiber-optic technology with the Raman tech-
nique increases the possibility of Raman spectroscopy to be
endoscopically utilized in clinic. Our very recent development of
a novel integrated endoscopic Raman system greatly facilitates
analysis of in situ gastric tissue Raman signals of different
pathologies, thereby bringing Raman technology into clinical
endoscopic applications.
In this work, we investigate, for the first time, the in vivo NIR
Raman spectral properties of benign and malignant peptic
Fig. 4
44.2% of the total variation in the Raman spectral dataset revealing the
diagnostically significant Ramanspectralfeatures fortissueclassification.
The first three diagnostically significant LVs accounting for
Fig. 5
LV3 illustrating the intrinsic clustering of normal mucosa, benign ulcers
and malignant ulcers.
Three-dimensional score value plot spanned by LV1, LV2 and
Fig. 6
belonging to normal mucosa, benign ulcers and malignant ulcers, illus-
trating the good clustering of the three distinctive gastric tissue types
achieved by the partial least squares-discriminant analysis (PLS-DA).
Two-dimensional ternary plot of the posterior probabilities
Table 1
tissue types using partial least squares-discriminant analysis (PLS-DA)
together with the leave-one tissue site out, cross validation
Classification results of Raman prediction of the three gastric
Tissue type
Raman prediction
Normal Benign ulcerMalignant ulcer
Normal
Benign ulcer
Malignant ulcer
Sensitivity
Specificity
839
8
3
90.8%
93.8%
45
94
9
84.7%
94.5%
40
9
55
82.1%
95.3%
3166 | Analyst, 2010, 135, 3162–3168 This journal is ª The Royal Society of Chemistry 2010

Page 6

ulcerous lesions in the stomach. To further characterize gastric
ulcerated lesions, we also compared with Raman spectra of
normal gastric tissue. Subtle inter- and/or intra- tissue variability
but significant differences in Raman spectra between normal
mucosa, benign ulcers and malignant ulcers are observed (Fig. 1),
confirming the promising potential of Raman endoscopy for in
vivo differentiation between benign and malignant peptic
ulcerous lesions in the stomach during clinical gastroscopic
examinations. For instance, the Raman peak intensities at 875,
936, 1004, 1245, 1445, 1620, 1665 and 1745 cm?1appear to be
unique for ulcerated lesions with a certain degree of similar
alterations (increase/decrease) of tissue Raman signals in benign
and malignant ulcers as compared to normal mucosa (Figs. 1 and
2). This indicates that benign and malignant ulcerous lesions still
contain some similar tissue constituents (e.g., due to inflamma-
tion, loss of mucous membrane, necrotic debris, increased blood
content, etc.). One notes that the diagnostic abilities of the
Raman peak intensities 875, 1335, 1655 and 1745 cm?1together
with the difference spectrum (malignant-benign) of Fig. 1B, are
directly comparable to the findings in other in vivo studies of
gastric precancer and cancer.24–26Despite the presence of peptic
ulceration with varying invasive degrees, the in vivo Raman
signals of cancer tissues still reflect a distinctive composition of
biochemical information (e.g., collagen, DNA and lipids) for
tissue diagnosis. For instance, the decrease in the Raman peak at
875 cm?1for malignant ulcers has also been observed in Raman
studies of non-ulcerated precancer and cancer.21,24–26Further-
more, peptic ulcers were also associated with a decrease in
Raman signals primarily originating from lipids (1302, 1440 and
1745 cm?1). A decrease in Raman signals of the carbonyl
n(C]O) vibration was found in cancerous tissue (Fig. 2), which
also has been observed in other studies (e.g., larynx,12gastric24–26
and the lung10). Besides, the variations in protein-related Raman
signals at 875, 936, 1004, 1245, 1445 and 1665 cm?1generally
associated with ulceration (Fig. 1B) likely reflect different
architectural abnormalities, such as decrease or absence of the
mucous membrane (i.e., epithelium and glandular secreting
structures) and various degrees of necrotic debris and inflam-
matory mediators (including neutrophiles, plasma cells and
lymphocytes) associated with disease transformation.37,38In fact,
the increase/decrease of prominent Raman intensities at 936,
1245, 1302 and 1665 cm?1for benign ulcers as compared to
normal mucosa suggests that the Raman signals detected mostly
arise from the connective tissues (i.e., large spectral contribution
from collagen) in the stomach. This is due to the effective Raman
excitation and collection of the granulation tissue, reflecting the
absence of the mucous membrane in benign ulcers.39–41
Our further semi-quantitative non-negative least squares
(NNLS)25fitting analysis of in vivo gastric Raman spectra using
basis spectra of the six reference biochemicals (i.e., actin,
collagen, DNA, histones, phosphatidylcholine and triolein)
confirms the significantly larger contributions from collagen and
actin in benign ulcers compared to normal mucosa (data not
shown). The increase in Raman signals of actin (a major
component of muscle cells) could be related to the Raman signal
detected from the deeper muscularis mucosa layer together with
the presence of actin-rich myofibroblasts within the granulation
tissue.40Hence, Raman endoscopy can reveal the absence of
mucus membrane (Fig. 6), confirming the ability of in vivo
Raman technique for achieving a high diagnostic specificity for
tissue diagnosis and characterization. On the other hand,
distinctive Raman intensity changes (e.g., 855, 875, 1004, 1245,
1302, 1445 and 1665 cm?1) associated with proteins in malignant
ulcers are likely related to biochemical changes in the extracel-
lular matrixas well as theincrease in themetabolic activities (e.g.,
increased mitotic activities which include enzymes, hormones
etc.)5,24,25as compared to benign ulcers. For example, the
significant Raman peak increase at 1655 cm?1(amide I of
proteins in a-helical conformation) for malignant ulcerous tissue
spectra as compared to benign ulcers correlates well with gastric
cytological studies that relate the hyperchromatic state of cells
with neoplastic transformation.22,24–26,42In addition, the distinct
Raman bands associated with nucleic acids (1335 and 1575 cm?1)
were more intense for malignant ulcers. The increase in these
peaks as compared to benign ulcers could also be linked with an
abnormal DNA content in the gastric neoplastic cells, which is
one of the main characteristics of cell carcinogenesis process.25,42
On top of these, we also found a significant increase in Raman
peaks related to blood content of malignant and benign ulcers
compared to normal gastric tissue. For instance, the intense
Raman bands in the regions 1575–1620 cm?1(heme and C–C
stretching of porphyrins) could signify the increased Raman
photon collection of vascular-rich granulation tissue for mucosal
regeneration in the stomach,40,43reconfirming the absence of
mucous membrane in ulcerous tissue. Therefore, the Raman
spectral differences observed between normal gastric mucosa,
benign and malignant ulcerous tissue suggest that Raman
endoscopy can be used to elucidate architectural and cellular
changes of gastric tissue associated with malignant trans-
formation in vivo.
The incorporation of the entire Raman spectra for PLS-DA
modeling showed that in vivo Raman spectra can provide clini-
cally critical diagnostic information (approximately 53.8% of the
spectral variation utilizing 5 LVs) for differentiation between
normal mucosa, benign ulcers and malignant ulcers. The good
clustering (Fig. 5) by using 5 LVs in the cross validation sug-
gested that the diagnostic algorithms rendered were robust for
Raman spectral analysis. Furthermore, this PLS-DA analysis
(Fig. 4 and 5) showed that LV1 predominantly represents the
differences between normal tissue and benign/malignant ulcers
(i.e., 875, 936, 1004, 1245, 1445, 1620, 1655 and 1745 cm?1). In
contrast, LV2 mainly distinguishes malignant ulcers from benign
ulcers/normal mucosa, whereas LV3 primarily differentiates
benign ulcers from normal mucosa/malignant ulcers. The Raman
prediction of the constructed multi-class PLS-DA diagnostic
model (Fig. 6) shows that malignant ulcerated tissue can be
differentiated from benign ulcers in vivo. The cross validated
diagnostic sensitivities of 90.8%, 84.7%, 82.1%; and specificities
of 93.8%, 94.5%, 95.3%, respectively, were achieved for classifi-
cation of normal mucosa, benign ulcers and malignant ulcers,
suggesting the clinical potential of Raman endoscopy for in vivo
diagnosis of gastric malignancies. Hence, the intrinsic tissue
biomolecular Raman signals identified in this study can be
advantageously utilized in real-time for guiding the clinician to
biopsy suspicious lesions in ulcer bases that are likely to yield
significant pathology.
Overall, Raman endoscopyoffers several advantagesincluding
objective, real-time, and noninvasive detection under multimodal
This journal is ª The Royal Society of Chemistry 2010Analyst, 2010, 135, 3162–3168 | 3167

Page 7

imaging guidance as a complementary diagnostic tool, which
potentially can improve efficiency of screening procedures (i.e., in
situ diagnosis) and assist in determining the margins of tumors
for surgical procedures. The reduction in biopsies of ulcers would
be very beneficial as routine multiple biopsies are likely to induce
massive hemorrhage in the ulcerated lesions, which may result in
the spread of cancer cells in the stomach. The clinicians partici-
pating in in vivo Raman studies in this work found the learning
curve of Raman endoscopic screening to be very straightforward
as compared to endoscopic biopsies (i.e., reducing tissue
sampling numbers, avoiding vascular puncturing etc.). It should
be noted that the Raman endoscopic characterization of the
bleeding ulcers and early malignancies primarily located in the
ulcer margins warrants further investigations as the tissue
properties could be markedly different from the ulcer bases (e.g.,
tissue architecture, blood flow, collagen distribution and tissue
regenerative repair, etc.).40,41,43
In summary, this work demonstrates for the first time that the
image-guided Raman endoscopy technique developed can be
used to reveal significant Raman spectral differences between
normal mucosa, benign and malignant peptic ulcers in vivo
during clinical gastroscopy. The Raman spectroscopic properties
of gastric tissue can be effectively translated into a great wealth
of diagnostic information providing new insights into biochem-
ical and architectural changes of benign and malignant gastric
ulcers in vivo. The PLS-DA algorithms were employed to classify
Raman spectra of different gastric tissue types with good diag-
nostic accuracy. It is expected that the image-guided Raman
endoscopic technology has great potential for rapid, in vivo
differentiation between benign and malignant gastric ulcers at
the molecular level.
Acknowledgements
This research was supported by the National Medical Research
Council, the Biomedical Research Council, and the Faculty
Research Fund from the National University of Singapore.
References
1 C. J. Clark, R. C. Thirlby, V. Picozzi Jr, D. B. Schembre,
F. P. Cummings and E. Lin, Curr. Probl. Surg., 2006, 43, 566–570.
2 D. M. Parkin, F. Bray, J. Ferlay and P. Pisani, Ca–Cancer J. Clin.,
2002, 55, 74–108.
3 I. Podolsky, P. R. Storms, C. T. Richardson, W. L. Peterson and
J. S. Fordtran, Dig. Dis. Sci., 1988, 33, 1057–1063.
4 C. Y. Chen, D. C. Wu, Y. T. Kuo, C. H. Lee, T. S. Jaw, W. Y. Kang
and J. S. Hsu, AJR, Am. J. Roentgenol., 2007, 190, 1505–1511.
5 N. Woolf, Pathology: Basic and systemic, WB Saunders Company,
London, 1998.
6 M. Classen, G. N. J. Tytgat, C. J. Lightdale, Gastroenterological
endoscopy, Georgh Thieme Verlag, Stuttgart, 2002.
7 S. Yalamarthi, P. Witherspoon, D. McCole and C. D. Auld,
Endoscopy, 2004, 36, 874–879.
8 A. J. Overhiser and P. Sharma, Rev. Gastroenterol. Disord., 2008, 8,
186–193.
9 I. Georgakoudi, B. C. Jacobson, J. Van Dam, V. Backman,
M. B. Wallace, M. G. M€ uller, Q. Zhang, K. Badizadegan, D. Sun,
G. A. Thomas, L. T. Perelman and M. S. Feld, Gastroenterology,
2001, 120, 1620–1629.
10 Z. Huang, A. Mcwilliams, H. Lui, D. I. Mclean, S. Lam and H. Zeng,
Int. J. Cancer, 2003, 107, 1047–1052.
11 Z. Huang, H. Lui, D. I. McLean, M. Korbelik and H. Zeng,
Photochem. Photobiol., 2005, 81, 1219–1226.
12 S. K. Teh, W. Zheng, D. P. Lau and Z. Huang, Analyst, 2009, 134,
1232–1239.
13 E. Widjaja, W. Zheng and Z. Huang, Int. J. Onc., 2008, 32, 653–662.
14 A. Mahadevan-Jansen and R. Richards-Kortum, J. Biomed. Opt.,
1996, 1, 31–70.
15 J. Hung, S. Lam, J. C. Leriche and B. Palcic, Lasers Surg. Med., 1991,
11, 99–105.
16 R. Drezek, K. Sokolov, U. Utzinger, I. Boiko, A. Malpica, M. Follen
and R. Richards-Kortum, J. Biomed. Opt., 2001, 6, 385–396.
17 N. Stone, C. Kendall, J. Smith, P. Crow and H. Barr, Faraday
Discuss., 2004, 126, 141–157.
18 T. Kawabata, T. Mizuno, S. Okazaki, M. Hiramatsu, T. Setoguchi,
H. Kikuchi, M. Yamamoto, Y. Hiramitsu, K. Kondo, M. Baba,
M. Ohta, K. Kamiya, T. Tanaka, S. Suzuki and H. Konno,
J. Gastroenterol., 2008, 43, 283–290.
19 S. K. Teh, W. Zheng, K. Y. Ho, M. Teh, K. G. Yeoh and Z. Huang,
Br. J. Cancer, 2008, 98, 457–465.
20 A. Nijssen, S. Koljenovic, T. C. Bakker Schut, P. J. Caspers and
G. J. Puppels, J. Biophotonics, 2009, 2, 29–36.
21 S. K. Teh, W. Zheng, K. Y. Ho, M. Teh, K. G. Yeoh and Z. Huang,
J. Biomed. Opt., 2008, 13, 034013.
22 S. K. Teh, W. Zheng, K. Y. Ho, M. Teh, K. G. Yeoh and Z. Huang,
Br. J. Surg., 2010, 97, 550–557.
23 Z. Huang, S. K. Teh, W. Zheng, J. Mo, K. Lin, X. Shao, K. Y. Ho,
M. Teh and K. G. Yeoh, Opt. Lett., 2009, 34, 758–760.
24 Z. Huang, M. Bergholt, W. Zheng, K. Lin, K. Y. Ho, M. Teh and
K. G. Yeoh, J. Biomed. Opt., 2010, 15, 037017.
25 Z. Huang, S. K. Teh, W. Zheng, K. Lin, K. Y. Ho, M. Teh and
K.G. Yeoh,
Biosens. Bioelectron.,
j.bios.2010.07.125.
26 M. S. Bergholt, W. Zheng, K. Lin, K. Y. Ho, M. Teh, K. G. Yeoh,
J. B. Y. So and Z. Huang, Int. J. Cancer, 2010, DOI: 10.1002/
ijc.25618.
27 S. K. Teh, W. Zheng, K. Y. Ho, M. Teh, K. G. Yeoh and Z. Huang,
Int. J. Cancer, 2010, 126, 1920–1927.
28 Z. Huang, H. Zeng, I. Hamzavi, D. I. Mclean and H. Lui, Opt. Lett.,
2001, 26, 1782–1784.
29 K. G. Yeoh, J. Gastroenterol. Hepatol., 2007, 22, 970–972.
30 T. C. Bakker Schut, M. J. H. Witjes, H. J. C. M. Sterenborg,
O.C.Speelman,J.L.N.
H. A. Bruining and G. J. Puppels, Anal. Chem., 2000, 72, 6010–6018.
31 P. Geladi and B. R. Kowalski, Anal. Chim. Acta, 1986, 185, 1–17.
32 A. Lorber, L. E. Wangen and B. R. Kowalski, J. Chemom., 1987, 1,
19–31.
33 M. Barker and W. Rayens, J. Chemom., 2003, 17, 166–173.
34 R. Picard and D. Cook, J. Am. Stat. Assoc., 1984, 79, 575–583.
35 A. Smilde, R. Bro, P. Geladi, Multi-way analysis with applications in
the chemical sciences, John Wiley & Sons Ltd, Chichester, 2004.
36 T. E. Graedel and R. McGill, Science, 1982, 215, 1191–1198.
37 N. Kunapareddy, J. P. Freyer and J. R. Mourant, J. Biomed. Opt.,
2008, 13, 054002.
38 G. J. Puppels, H. S. Garritsen, G. M. Segers-Nolten, F. F. de Mul and
J. Greve, Biophys. J., 1991, 60, 1046–1056.
39 C. Krafft, D. Codrich, G. Pelizzo and V. Sergo, J. Biophotonics, 2008,
2, 154–169.
40 A. Gillessen, B. Voß, J. Rauterberg and W. Domschke, Scand.
J. Gastroenterol., 1993, 28, 688–689.
41 A. Tarnawski, K. Tanoue, A. M. Santos and I. J. Sarfeh, Scand.
J. Gastroenterol. Suppl., 1995, 210, 9–14.
42 J. H. Hughes, C. J. Leigh, S. S. Raab, S. Y. Hook, M. B. Cohen and
M. J. Suhrland, Cancer, 1998, 84, 289–294.
43 U. K. Franzeck, R. M€ unch, M. W€ achter, B. Vesti, R. Ammann and
A. Bollinger, Gastrointest. Endosc., 1993, 39, 806–809.
2010,DOI:10.1016/
Roodenburg,E.T. Marple,
3168 | Analyst, 2010, 135, 3162–3168This journal is ª The Royal Society of Chemistry 2010