Optical diagnosis of laryngeal cancer using high wavenumber Raman spectroscopy

Abstract

We report the implementation of the transnasal image-guided high wavenumber (HW) Raman spectroscopy to differentiate tumor from normal laryngeal tissue at endoscopy. A rapid-acquisition Raman spectroscopy system coupled with a miniaturized fiber-optic Raman probe was utilized to realize real-time HW Raman (2800-3020 cm(-1)) measurements in the larynx. A total of 94 HW Raman spectra (22 normal sites, 72 tumor sites) were acquired from 39 patients who underwent laryngoscopic screening. Significant differences in Raman intensities of prominent Raman bands at 2845, 2880 and 2920 cm(-1) (CH(2) stretching of lipids), and 2940 cm(-1) (CH(3) stretching of proteins) were observed between normal and cancer laryngeal tissue. The diagnostic algorithms based on principal components analysis (PCA) and linear discriminant analysis (LDA) together with the leave-one subject-out, cross-validation method on HW Raman spectra yielded a diagnostic sensitivity of 90.3% (65/72) and specificity of 90.9% (20/22) for laryngeal cancer identification. This study demonstrates that HW Raman spectroscopy has the potential for the noninvasive, real-time diagnosis and detection of laryngeal cancer at the molecular level.

Full text

This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright

Author's personal copy
Biosensors
and
Bioelectronics
35 (2012) 213–
217
Contents
lists
available
at
SciVerse
ScienceDirect
Biosensors
and
Bioelectronics
j
our
na
l
ho
me
page:
www.elsevier.com/locate/bios
Optical
diagnosis
of
laryngeal
cancer
using
high
wavenumber
Raman
spectroscopy
Kan
Lina,
David
Lau
Pang
Chengb,
Zhiwei
Huanga,∗
aOptical
Bioimaging
Laboratory,
Department
of
Bioengineering,
Faculty
of
Engineering,
National
University
of
Singapore,
Singapore
117576,
Singapore
bDepartment
of
Otolaryngology,
Singapore
General
Hospital,
Singapore
169608,
Singapore
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
26
December
2011
Received
in
revised
form
21
February
2012
Accepted
23
February
2012
Available online 15 March 2012
Keywords:
High
wavenumber
Raman
spectroscopy
Laryngeal
cancer
Optical
diagnosis
a
b
s
t
r
a
c
t
We
report
the
implementation
of
the
transnasal
image-guided
high
wavenumber
(HW)
Raman
spec-
troscopy
to
differentiate
tumor
from
normal
laryngeal
tissue
at
endoscopy.
A
rapid-acquisition
Raman
spectroscopy
system
coupled
with
a
miniaturized
fiber-optic
Raman
probe
was
utilized
to
realize
real-
time
HW
Raman
(2800–3020
cm−1)
measurements
in
the
larynx.
A
total
of
94
HW
Raman
spectra
(22
normal
sites,
72
tumor
sites)
were
acquired
from
39
patients
who
underwent
laryngoscopic
screening.
Significant
differences
in
Raman
intensities
of
prominent
Raman
bands
at
2845,
2880
and
2920
cm−1
(CH2stretching
of
lipids),
and
2940
cm−1(CH3stretching
of
proteins)
were
observed
between
normal
and
cancer
laryngeal
tissue.
The
diagnostic
algorithms
based
on
principal
components
analysis
(PCA)
and
linear
discriminant
analysis
(LDA)
together
with
the
leave-one
subject-out,
cross-validation
method
on
HW
Raman
spectra
yielded
a
diagnostic
sensitivity
of
90.3%
(65/72)
and
specificity
of
90.9%
(20/22)
for
laryngeal
cancer
identification.
This
study
demonstrates
that
HW
Raman
spectroscopy
has
the
potential
for
the
noninvasive,
real-time
diagnosis
and
detection
of
laryngeal
cancer
at
the
molecular
level.
© 2012 Elsevier B.V. All rights reserved.
1.
Introduction
Laryngeal
cancer
is
one
of
the
most
common
malignancies
in
humans
worldwide
due
to
its
high
incidence
rate
and
mortality
(Chu
and
Kim,
2008).
For
instance,
in
Southeast
Asia,
the
rates
of
incidence
and
mortality
of
laryngeal
cancer
are
significantly
higher
than
other
areas
of
the
world
(Cao
et
al.,
2007;
Döbróssy,
2005).
Early
cancer
diagnosis
in
the
larynx
with
effective
treatment
(e.g.,
surgery,
radiotherapy
or
chemotherapy
alone
or
in
combi-
nation)
is
crucial
to
improving
the
5-year
survival
rate
(Chu
and
Kim,
2008;
Lee
et
al.,
2008).
Positive
endoscopic
biopsy
currently
is
the
gold
standard
for
cancer
diagnosis,
but
it
is
invasive
and
impractical
for
screening
high
risk
patients,
which
might
affect
the
quality
of
the
voice
due
to
multiple
biopsies
(Lee
et
al.,
2008).
Fiber-
optic
laryngoscopy
is
the
primary
physical
examination
tool
for
now
(Chu
and
Kim,
2008),
which
relies
on
white-light
illumina-
tion
while
requires
highly
experienced
skills
of
recognization
and
locating
pathologic
tissues
(Beser
et
al.,
2009).
Raman
spectroscopy
is
a
unique
vibrational
technique
capable
of
probing
biomolecular
changes
in
tissue,
which
has
shown
great
promise
for
early
diag-
nosis
and
detection
of
precancer
and
cancer
diagnosis
in
a
variety
∗Corresponding
author
at:
Optical
Bioimaging
Laboratory,
Department
of
Bio-
engineering,
Faculty
of
Engineering,
National
University
of
Singapore,
9
Engineering
Drive
1,
Singapore
117576,
Singapore.
Tel.:
+65
6516
8856;
fax:
+65
6872
3069.
E-mail
address:
biehzw@nus.edu.sg
(Z.
Huang).
of
organs
(e.g.,
skin,
cervix,
lung,
esophagus,
stomach,
colon,
kid-
ney,
bladder,
breast,
nasopharynx
and
the
larynx)
(Nijssen
et
al.,
2002;
Stone
et
al.,
2000;
Lau
et
al.,
2003,
2005;
Almond
et
al.,
2011;
Bergholt
et
al.,
2011a,b;
Draga
et
al.,
2010;
Gniadecka
et
al.,
2004;
Haka
et
al.,
2006;
Hale
et
al.,
2009;
Huang
et
al.,
2003,
2005;
Mo
et
al.,
2009;
Kanter
et
al.,
2009;
Magee
et
al.,
2009;
Short
et
al.,
2006;
Teh
et
al.,
2008,
2009a;
Widjaja
et
al.,
2008;
Mahadevan-
Jansen
and
Richards-Kortum,
1996;
Wong
Kee
Song
et
al.,
2005;
Shim
et
al.,
2000).
Current
Raman
research
in
diagnosing
laryngeal
cancer
is
mostly
focused
on
the
so-called
fingerprint
region
(i.e.,
800–1800
cm−1)
that
contains
rich
biochemical
information
about
the
tissue
(Stone
et
al.,
2000;
Mahadevan-Jansen
and
Richards-
Kortum,
1996;
Lau
et
al.,
2003,
2005;
Teh
et
al.,
2009b).
However,
the
strong
fluorescence
background
and
Raman
signals
attributed
to
the
silica
fiber
severely
interfere
with
the
detection
of
the
inher-
ently
weak
tissue
Raman
signal,
leading
to
a
complex
fiber
probe
filtering
design
as
well
as
signal
analysis
in
the
fingerprint
region.
On
the
other
hand,
the
high
wavenumber
(HW)
(2800–3800
cm−1)
Raman
spectroscopy
can
also
provide
complementary
biochemical
information
for
tissue
diagnosis
and
characterization
with
much
stronger
tissue
Raman
signals
but
reduced
tissue/fiber
fluorescence
background
(Koljenovic
et
al.,
2005;
Nijssen
et
al.,
2007;
Mo
et
al.,
2009),
as
compared
to
the
fingerprint
Raman
spectroscopy.
To
date,
HW
Raman
spectroscopy
for
laryngeal
tissue
diagnosis
and
characterization
has
yet
been
reported
in
literature.
In
this
work,
we
study
the
implementation
of
the
transnasal
image-guided
HW
Raman
spectroscopy
developed
to
differentiate
tumor
from
normal
0956-5663/$
–
see
front
matter ©
2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.bios.2012.02.050

Author's personal copy
214 K.
Lin
et
al.
/
Biosensors
and
Bioelectronics
35 (2012) 213–
217
laryngeal
tissue.
A
rapid-acquisition
Raman
spectroscopy
system
coupled
with
a
miniaturized
fiber-optic
Raman
probe
was
utilized
to
realize
real-time
HW
Raman
(2800–3020
cm−1)
measurements
in
the
larynx.
Multivariate
statistical
techniques,
including
prin-
cipal
components
analysis
(PCA)
and
linear
discriminant
analysis
(LDA),
are
utilized
to
develop
diagnostic
algorithms
for
differentia-
tion
between
normal
and
cancerous
laryngeal
tissue.
The
receiver
operating
characteristic
(ROC)
curve
is
also
conducted
to
further
evaluate
the
performance
of
PCA–LDA
algorithms
on
HW
Raman
spectroscopy
for
laryngeal
cancer
diagnosis.
2.
Materials
and
methods
2.1.
Raman
endoscopic
instrumentation
Fig.
1
shows
the
schematic
diagram
of
the
image-guided
Raman
spectroscopy
developed
for
tissue
measurements
and
characteriza-
tion
(Huang
et
al.,
2009).
Briefly,
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)
equipped
with
a
liquid
nitrogen
cooled
(−120 ◦C),
NIR-optimized,
back-illuminated
and
deep
depletion
charge-coupled
device
(CCD)
camera
(1340
×400
pixels
at
20
!m
×
20
!m
per
pixel;
Spec-10:
400BR/LN,
Princeton
Instruments).
We
have
constructed
a
1.8
mm
fiber-optic
Raman
endoscopic
probe
with
dual
coatings
on
the
fiber
tip
for
optimizing
both
the
tissue
excitation
and
Raman
col-
lections
(Huang
et
al.,
2009).
The
785
nm
laser
is
coupled
into
the
central
delivery
fiber
(200
!m,
NA
=
0.22)
of
the
Raman
probe
for
tissue
excitation,
while
the
backscattered
tissue
Raman
pho-
tons
from
the
laryngeal
tissue
are
collected
by
the
surrounding
fibers
(32
!m
×
200
!m,
NA
=
0.22).
The
Raman
fiber
probe
fits
into
the
instrument
channel
of
laryngoscope
and
can
be
safely
tar-
geted
to
different
locations
in
the
larynx
under
the
multimodal
wide-field
imaging
(i.e.,
white-light
reflectance
(WLR)
and
narrow-
band
imaging
(NBI))
guidance.
The
system
acquires
HW
Raman
spectra
over
the
range
of
2800–3020
cm−1,
and
each
raw
spec-
trum
is
acquired
within
1
s
with
light
irradiance
of
1.5
W/cm2.
The
spectral
resolution
of
the
system
is
about
9
cm−1,
and
all
wavelength-calibrated
HW
Raman
spectra
are
also
corrected
for
the
wavelength-dependence
of
the
system
using
a
standard
lamp
(RS-
10,
EG&G
Gamma
Scientific,
San
Diego,
CA).
HW
Raman
spectra
are
then
extracted
from
the
raw
tissue
spectra
using
established
pre-
processing
methods
including
smoothing,
baseline
subtraction,
etc.
(Bergholt
et
al.,
2011a).
All
the
spectral
pre-processing
is
completed
on-line,
and
the
Raman
spectra
and
the
outcome
of
decision
algo-
rithms
can
be
displayed
in
real-time
in
a
comprehensible
graphical
user
interface
(GUI)
during
clinical
transnasal
Raman
endoscopy.
2.2.
Subjects
This
study
was
approved
by
the
SingHealth
Centralized
Institu-
tional
Review
Board
(IRB),
Singapore.
A
total
of
39
different
patients
with
a
mean
age
of
60
who
underwent
surgical
resection
due
to
laryngeal
malignancies
were
recruited
for
this
study.
All
patients
preoperatively
signed
an
informed
consent
permitting
Raman
mea-
surements
on
laryngeal
tissue.
HW
Raman
spectra
were
directly
acquired
from
the
suspicious
lesion
sites
for
each
patient
through
gently
placing
the
fiber-optic
Raman
probe
on
the
tissue
with
sig-
nal
acquisition
time
of
<1
s.
HW
Raman
spectra
were
also
measured
from
the
surrounding
normal
sites
that
appear
completely
normal
in
the
laryngoscopist’s
opinion
(i.e.,
normal
tissue
does
not
exhibit
colored
patterned
changes
that
only
accompany
precursor
lesions)
(Arens
et
al.,
2004),
but
no
biopsies
were
taken
from
normal
appear-
ing
tissue.
Only
highly
abnormal
sites
measured
were
biopsied
and
then
submitted
for
histopathologic
examination.
A
total
of
94
Raman
spectra
(22
normal,
72
tumor
as
confirmed
by
histopathol-
ogy)
from
different
tissue
sites
were
collected.
For
the
assessment
of
diagnostic
sensitivity
and
specificity
of
Raman
endoscopy
for
tissue
classification,
histopathological
results
served
as
the
gold
standard.
2.3.
Data
preprocessing
All
raw
spectral
data
were
processed
on-line
with
software
developed
in
the
Matlab
environment
(The
MathWorks,
Inc.,
Nat-
ick,
MA)
(Bergholt
et
al.,
2011a).
Raw
spectra
are
first
pre-processed
by
a
first-order
Savitsky–Golay
filter
to
reduce
background
noise
(Savitzky
and
Golay,
1964).
A
first-order
polynomial
was
used
to
fit
tissue
autofluorescence
background
and
then
subtracted
from
the
raw
spectra
to
obtain
the
pure
HW
Raman
spectra.
The
HW
Raman
spectra
are
then
normalized
over
the
integrated
area
under
the
curve
from
2800
to
3020
cm−1to
allow
a
better
comparison
of
the
spectral
shapes
and
relative
Raman
band
intensities
among
different
subjects/tissue
sites
(Huang
et
al.,
2003).
2.4.
Multivariate
statistical
analysis
Principal
components
analysis
(PCA)
was
used
to
reduce
dimen-
sionality
of
the
Raman
data
(each
Raman
spectrum
ranging
from
2800
to
3020
cm−1with
set
of
255
intensities),
retaining
the
most
diagnostically
significant
information
for
effective
tissue
classifica-
tion.
The
spectra
were
first
standardized
to
ensure
that
mean
of
the
spectra
was
zero
and
the
standard
deviation
(SD)
was
one,
eliminat-
ing
the
influence
of
inter-
and/or
intra-subject
spectral
variability
on
PCA.
Mean
centering
ensured
that
the
principal
components
(PCs)
form
an
orthogonal
basis
(Devore,
2009;
Lachenbruch
and
Mickey,
1968).
Thus,
PCA
were
employed
to
extract
a
set
of
orthog-
onal
PCs
comprising
loadings
and
scores
that
accounted
for
most
of
the
total
variance
in
original
spectra.
Each
loading
vector
is
related
to
the
original
spectrum
by
a
variable
called
the
PC
score,
which
rep-
resents
the
weight
of
that
particular
component
against
the
basis
spectrum.
The
most
diagnostically
significant
PCs
(p
<
0.05)
were
determined
by
Student’s
t-test
(Devore,
2009)
and
then
selected
as
input
for
the
development
of
linear
discriminant
analysis
algo-
rithms
for
classification.
LDA
determines
the
discriminant
function
that
maximizes
the
variances
in
the
dataset
between
groups
while
minimizing
the
variances
between
members
of
the
same
group.
The
performance
of
the
diagnostic
algorithms
rendered
by
the
PCA–LDA
models
for
correctly
predicting
the
tissue
groups
(e.g.,
normal
vs.
cancer)
was
estimated
in
an
unbiased
manner
using
the
leave-
one
subject-out,
cross-validation
method
(Dillon
and
Goldstein,
1984;
Lachenbruch
and
Mickey,
1968)
on
all
model
spectra.
In
this
method,
the
spectra
from
each
same
patient
were
held
out
from
the
data
set
and
the
PCA–LDA
modeling
was
redeveloped
using
the
remaining
HW
Raman
spectra.
The
redeveloped
PCA–LDA
diag-
nostic
algorithm
was
then
used
to
classify
the
withheld
spectra.
This
process
was
repeated
until
all
withheld
spectra
were
classi-
fied.
Receiver
operating
characteristic
(ROC)
curves
were
generated
by
successively
changing
the
thresholds
to
determine
correct
and
incorrect
classifications
for
all
tissues.
Multivariate
statistical
anal-
ysis
was
performed
online
using
in-house
written
scripts
in
the
Matlab
(Mathworks
Inc.,
Natick,
MA)
programming
environment
(Bergholt
et
al.,
2011a).
3.
Results
Fig.
2A
shows
the
comparison
of
mean
HW
Raman
spectra
±1
SD
of
normal
(n
=
22)
and
cancer
(n
=
72)
laryngeal
tissue.
Prominent
Raman
bands
such
as
2845,
2880,
and
2920
cm−1(CH2stretch-
ing
of
lipids),
and
2940
cm−1(CH3stretching
of
proteins)
(Eikje
et
al.,
2005;
Mo
et
al.,
2009;
Santos
et
al.,
2005)
are
found
in
both

Author's personal copy
K.
Lin
et
al.
/
Biosensors
and
Bioelectronics
35 (2012) 213–
217 215
Fig.
1.
Schematic
of
the
integrated
Raman
spectroscopy
and
trimodal
endoscopic
imaging
system
developed
for
in
vivo
tissue
Raman
measurements.
normal
and
tumor
laryngeal
tissues.
As
shown
in
the
difference
spectrum
(Fig.
2B),
the
intensities
of
Raman
band
between
2810
and
2900
cm−1in
cancer
tissue
is
obviously
greater
than
normal
tissue,
while
the
Raman
band
between
2900
and
3020
cm−1the
normal
is
higher.
This
suggests
that
there
is
an
increase
or
decrease
in
particular
types
of
biomolecules
relative
to
the
total
Raman-
active
biomolecules
in
cancer
tissue
as
compared
to
normal
tissue,
demonstrating
the
potential
role
of
HW
Raman
spectroscopy
for
cancer
diagnosis
in
the
larynx.
To
determine
the
most
significant
Raman
features
for
tissue
analysis
and
classification,
the
multivariate
statistical
technique
(e.g.,
PCA–LDA)
coupled
with
Student’s
t-test
are
performed
by
incorporating
the
entire
HW
Raman
spectra.
Fig.
3
shows
the
first
five
dominant
principal
components
(PCs)
accounting
for
about
99.2%
(PC1:
89.1%;
PC2:
7.41%;
PC3:
1.52%;
PC4:
1.08%;
PC5:
0.07%)
of
the
total
variance
calculated
from
HW
Raman
spectra
of
laryngeal
tissues.
Overall,
the
PC
features
among
different
PCs
are
different,
but
some
PC
features
roughly
correspond
to
HW
Raman
spectra,
with
peaks
and
troughs
at
positions
(e.g.,
CH2stretch
band
(lipids)
near
2845
cm−1,
2880
cm−1,
2920
cm−1and
CH3stretch
band
(pro-
teins)
near
2940
cm−1)
similar
to
those
of
tissue
HW
Raman
spectra.
The
first
PC
accounts
for
the
largest
variance
within
the
spectral
data
sets
(e.g.,
89.1%),
whereas
successive
PCs
describe
features
that
contribute
progressively
smaller
variances.
Unpaired
two-sided
Fig.
2.
(A)
Comparison
of
the
mean
HW
Raman
spectra
±1
standard
deviations
(SD)
of
normal
(n
=
22)
and
cancer
(n
=
72)
laryngeal
tissue.
(B)
Difference
spectrum
±1
SD
between
cancer
(n
=
72)
and
normal
laryngeal
tissue
(n
=
22).
Note
that
the
mean
normalized
HW
Raman
spectrum
of
normal
tissue
was
shifted
vertically
for
better
visualization
(panel
A);
the
shaded
areas
indicate
the
respective
standard
deviations.
The
picture
shown
is
the
Raman
acquisitions
from
the
larynx
using
endoscopic
fiber-optic
Raman
probe.

Author's personal copy
216 K.
Lin
et
al.
/
Biosensors
and
Bioelectronics
35 (2012) 213–
217
302030002980296029402920290028802860284028202800
PC5
PC4
PC3
PC2
Loading of PCs
Raman shift(cm-1
)
PC1
Fig.
3.
The
first
five
principal
components
(PCs)
accounting
for
about
99.2%
of
the
total
variance
calculated
from
HW
Raman
spectra
of
laryngeal
tissue
(PC1
=
89.1%;
PC2
=
7.41%;
PC3
=
1.52%;
PC4
=
1.08%;
PC5
=
0.07%).
Student’s
t-tests
on
the
first
five
PCs
show
that
only
three
PCs
(PC1,
PC2
and
PC3,
p
<
0.05)
are
diagnostically
significant.
To
develop
effective
diagnostic
algorithms
for
tissue
classifi-
cation,
the
three
diagnostically
significant
PCs
are
fed
into
the
LDA
model
together
with
leave-one
subject-out,
cross-validation
technique
for
tissue
classification.
PCA–LDA
algorithms
on
the
HW
tissue
Raman
data
provide
the
diagnostic
sensitivity
of
90.3%
(65/72)
and
specificity
of
90.9%
(20/22)
for
laryngeal
cancer
identifi-
cation
(Fig.
4).
In
addition,
the
ROC
curve
(Fig.
5)
was
also
generated
from
the
posterior
probability
plot
in
Fig.
4
at
different
threshold
levels,
displaying
the
performance
of
PCA–LDA-based
diagnostic
algorithms
derived
for
laryngeal
cancer
detection.
The
integration
area
under
the
ROC
curve
is
0.97,
further
confirming
that
HW
Raman
technique
coupled
with
PCA–LDA-based
diagnostic
algo-
rithms
is
robust
for
laryngeal
cancer
diagnosis.
4.
Discussion
Raman
spectroscopy
holds
a
great
promise
for
clinical
appli-
cation,
as
it
can
be
used
as
a
non-invasive
technique
for
early
1009080706050403020100
0.00
0.25
0.50
0.75
1.00
Cancer (n=72)
Normal (n=22)
Site Number
Posterior Probablity
Fig.
4.
Scatter
plot
of
the
posterior
probability
belonging
to
normal
and
can-
cer
categories
using
the
PCA–LDA
method
together
with
leave-one
subject-out,
cross-validation
method.
The
algorithm
yields
a
diagnostic
sensitivity
of
90.3%
and
specificity
of
90.9%
for
differentiation
between
normal
and
tumor
tissues.
1.00.80.60.40.20.0
0.0
0.2
0.4
0.6
0.8
1.0
Sensitivity
1 - Specificity
Fig.
5.
ROC
curve
of
discrimination
results
for
HW
Raman
spectra
utilizing
the
PCA–LDA-based
spectral
classification
with
leave-one
subject-out,
cross
validation.
The
integration
area
under
the
ROC
curves
is
0.97
for
PCA–LDA-based
diagnostic
algorithm.
detection
of
biomolecular
changes
associated
with
tissue
pathol-
ogy.
Our
study
demonstrates
that
HW
Raman
technique
is
capable
of
generating
spectral
differences
between
normal
and
cancer
tis-
sue
in
the
larynx
as
shown
in
Fig.
2.
The
empirical
analysis
based
on
the
intensity
ratio
measurements,
which
relate
to
the
changes
in
protein-to-lipid
contents,
has
already
been
reported
in
liter-
ature
(Huang
et
al.,
2003;
Short
et
al.,
2005).
Student’s
t-test
(p
<
0.0001)
is
used
to
test
the
significant
difference
of
Raman
inten-
sity
ratios
of
normal
and
cancer
tissues.
Intensity
at
2920
cm−1and
2940
cm−1is
higher
in
normal
tissues,
while
at
2845
and
2890
cm−1
are
higher
in
cancer
tissues.
The
three
significant
Raman
peak
intensity
ratios
of
I2940/I2845 ,
I2940/I2890 and
I2940/I2920 correlated
with
their
histopathologic
findings
were
also
evaluated,
and
the
decision
lines
(i.e.,
diagnostic
algorithms)
separated
cancer
from
normal
tissues
with
a
sensitivity
of
72.2%
(52/72),
45.8%
(33/72)
and
44.4%
(32/72),
while
a
sensitivity
of
81.8%
(18/22),
72.7%
(16/22)
and
63.6%
(14/22),
respectively,
for
laryngeal
cancer
identification.
These
indicate
that
different
ratios
of
Raman
band
intensities
only
give
a
certain
levels
of
accuracy
for
tissue
classification.
The
Raman
intensity
ratios
of
I2940/I2845 and
I2940/I2890 are
both
lower
for
can-
cer
tissue,
this
may
be
due
to
the
decrease
in
content
of
collagen.
In
cancer
progression,
from
genetic
mutation
to
invasive
cancer
in
the
laryngeal,
the
epithelium
turns
thicker
and
thus
obstructs
the
colla-
gen
Raman
emission
from
deep
collagen
basal
membrane,
thereby
decreases
the
overall
ratio
of
I2940/I2845 in
laryngeal
cancer
tissue
(Teh
et
al.,
2008).
Besides,
the
contribution
of
collagen
in
cancer
tis-
sue
should
be
reduced
due
to
proliferation
of
cancerous
cells
and
express
as
a
class
of
metalloportease,
which
in
turn
decrease
the
content
of
collagen
level.
Raman
band
at
2845
cm−1is
tentatively
ascribed
to
CH2lipids
and
it
seems
to
correspond
to
Raman
peak
at
1450
cm−1in
the
fingerprint
region,
which
is
assigned
to
CH2pro-
tein/lipids.
CH2protein/lipids
are
found
to
be
higher
in
cancer
than
in
normal
laryngeal
tissues,
which
can
be
explained
by
the
increase
of
mitotic
activity
in
the
nucleus
(Stone
et
al.,
2000;
Haka
et
al.,
2006;
Mo
et
al.,
2009;
Teh
et
al.,
2008,
2009b).
The
distinctive
differ-
ences
in
HW
Raman
spectra
between
normal
and
cancer
laryngeal
tissues
reinforce
that
HW
Raman
spectroscopy
can
be
used
to
reveal
molecular
changes
associated
with
carcinogenesis
progression.
Considering
the
complexity
of
biological
tissue,
multivariate
sta-
tistical
analysis
(PCA–LDA)
which
incorporates
the
entire
Raman
spectra
data
for
analysis
is
more
robust
and
rigorous
to
differenti-
ate
spectra
that
represent
either
normal
or
cancer
tissue.
Compared

Author's personal copy
K.
Lin
et
al.
/
Biosensors
and
Bioelectronics
35 (2012) 213–
217 217
with
intensity
ratio
approach,
there
is
a
great
improvement
in
diagnostic
sensitivity
(∼25%)
and
specificity
(∼11%)
of
PCA–LDA
algorithms
(Fig.
4).
The
ROC
curve
(Fig.
5)
of
PCA–LDA
model-
ing
(AUC
=
0.97)
further
verifies
a
better
diagnostic
efficacy
of
HW
Raman
spectroscopy
integrated
with
PCA–LDA
algorithm
as
com-
pared
to
the
intensity
ratio
diagnostic
algorithms.
One
note
that
PCA
is
primarily
for
data
reduction
rather
than
identification
of
biochemical
or
biomolecular
components
of
tissue,
it
is
difficult
to
interpret
the
physical
meanings
from
the
component
spectra.
This
could
be
tackled
by
using
more
powerful
diagnostic
algo-
rithms
such
as
genetic
algorithms
(Mountford
et
al.,
2001),
the
distinctive
spectral
regions
that
are
optimal
for
tissue
differen-
tiation
may
be
indentified
and
related
to
particular
biochemical
and
biomolecular
changes
(e.g.,
proteins,
lipids
and
nucleic
acid)
associated
with
neoplastic
changes.
It
is
also
crucial
to
further
understanding
the
relationship
between
the
neoplastic-related
morphologic/biochemical
changes
and
tissue
HW
Raman
spec-
tra
for
laryngeal
precancer/cancer
diagnosis
(Stone
et
al.,
2000;
Bergholt
et
al.,
2011a;
Teh
et
al.,
2008).
We
are
currently
working
on
this
direction
by
recruiting
more
patients
for
developing
more
powerful
HW
Raman
genetic
diagnostic
algorithms
for
real-time
laryngeal
tissue
diagnosis
and
characterization.
In
summary,
this
work
demonstrates
that
transnasal
image-
guided
real-time
HW
Raman
spectroscopic
technique
integrated
with
an
endoscope-based
fiber-optic
Raman
probe
can
be
used
to
acquire
HW
Raman
spectra
from
laryngeal
tissue
in
the
range
2800–3020
cm−1during
clinical
endoscopic
examination.
The
sig-
nificant
differences
in
HW
Raman
spectra
are
observed
in
normal
and
cancer
laryngeal
tissue.
The
PCA–LDA
modeling
on
HW
Raman
spectra
provides
good
tissue
classification,
illustrating
the
potential
of
HW
Raman
spectroscopy
for
real-time
laryngeal
cancer
detection
during
clinical
endoscopic
examination.
Acknowledgement
This
research
was
supported
by
the
Biomedical
Research
Coun-
cil,
and
the
National
Medical
Research
Council,
Singapore.
References
Almond,
L.M.,
Hutchings,
J.,
Shepherd,
N.,
Barr,
H.,
Stone,
N.,
Kendall,
C.,
2011.
J.
Biophotonics
4,
685–695.
Arens,
C.,
Dreyer,
T.,
Glanz,
H.,
Malzahn,
K.,
2004.
Eur.
Arch.
Otorhinolaryngol.
261,
71–76.
Bergholt,
M.S.,
Zheng,
W.,
Lin,
K.,
Ho,
K.Y.,
Teh,
M.,
Yeoh,
K.G.,
So,
J.B.Y.,
Huang,
Z.,
2011a.
Technol.
Cancer
Res.
Treat.
10,
103–112.
Bergholt,
M.S.,
Zheng,
W.,
Lin,
K.,
Ho,
K.Y.,
Teh,
M.,
Yeoh,
K.G.,
Yan
So,
J.B.,
Huang,
Z.,
2011b.
Int.
J.
Cancer
128
(11),
2673–2680.
Beser,
M.,
Gultekin,
E.,
Yener,
M.,
Zeybek,
M.,
Öner,
B.,
Topc¸
u,
V.,
2009.
Eur.
Arch.
Otorhinolaryngol.
266,
1953–1958.
Cao,
W.F.,
Zhang,
L.Y.,
Liu,
M.B.,
Tang,
P.Z.,
Liu,
Z.H.,
Sun,
B.C.,
2007.
Hum.
Pathol.
38,
747–752.
Chu,
E.A.,
Kim,
Y.J.,
2008.
Otolaryngol.
Clin.
North
Am.
41,
673–695.
Devore,
J.L.,
2009.
Probability
and
Statistics
for
Engineering
and
the
Sciences.
Thom-
son/Brooks/Cole,
Belmont,
CA.
Dillon,
W.R.,
Goldstein,
M.,
1984.
Nueva
York.
EUA:
Wiley.
Döbróssy,
L.,
2005.
Cancer
Metastasis
Rev.
24,
9–17.
Draga,
R.O.P.,
Grimbergen,
M.C.M.,
Vijverberg,
P.L.M.,
Swol,
C.F.P.v.,
Jonges,
T.G.N.,
Kummer,
J.A.,
Ruud
Bosch,
J.L.H.,
2010.
Anal.
Chem.
82,
5993–5999.
Eikje,
N.S.,
Ozaki,
Y.,
Aizawa,
K.,
Arase,
S.,
2005.
J.
Biomed.
Opt.
10,
014013.
Gniadecka,
M.,
Philipsen,
P.A.,
Sigurdsson,
S.,
Wessel,
S.,
Nielsen,
O.F.,
Christensen,
D.H.,
Hercogova,
J.,
Rossen,
K.,
Thomsen,
H.K.,
Gniadecki,
R.,
Hansen,
L.K.,
Wulf,
H.C.,
2004.
J.
Invest.
Dermatol.
122,
443–449.
Haka,
A.S.,
Volynskaya,
Z.,
Gardecki,
J.A.,
Nazemi,
J.,
Lyons,
J.,
Hicks,
D.,
Fitzmaurice,
M.,
Dasari,
R.R.,
Crowe,
J.P.,
Feld,
M.S.,
2006.
Cancer
Res.
66,
3317–3322.
Hale,
W.,
Rachel,
K.,
Cory,
S.,
Brian,
S.,
Raja,
R.,
Janet,
P.,
Abhilash,
P.,
Greg,
A.,
Michael,
D.K.,
2009.
J.
Pediatr.
Surg.
44,
1152–1158.
Huang,
Z.,
McWilliams,
A.,
Lui,
H.,
McLean,
D.I.,
Lam,
S.,
Zeng,
H.,
2003.
Int.
J.
Cancer
107,
1047–1052.
Huang,
Z.,
Lui,
H.,
McLean,
D.I.,
Korbelik,
M.,
Zeng,
H.,
2005.
Photochem.
Photobiol.
81,
1219–1226.
Huang,
Z.,
Teh,
S.K.,
Zheng,
W.,
Mo,
J.,
Lin,
K.,
Shao,
X.,
Ho,
K.Y.,
Teh,
M.,
Yeoh,
K.G.,
2009.
Opt.
Lett.
34,
758–760.
Kanter,
E.M.,
Vargis,
E.,
Majumder,
S.,
Keller,
M.D.,
Woeste,
E.,
Rao,
G.G.,
Mahadevan-
Jansen,
A.,
2009.
J.
Biophotonics
2,
81–90.
Koljenovic,
S.,
Schut,
T.C.B.,
Wolthuis,
R.,
de
Jong,
B.,
Santos,
L.,
Caspers,
P.J.,
Kros,
J.M.,
Puppels,
G.J.,
2005.
J.
Biomed.
Opt.
10,
031111–031116.
Lachenbruch,
P.A.,
Mickey,
M.R.,
1968.
Technometrics
10,
1–11.
Lau,
D.P.,
Huang,
Z.,
Lui,
H.,
Man,
C.S.,
Berean,
K.,
Morrison,
M.D.,
Zeng,
H.,
2003.
Lasers
Surg.
Med.
32
(3),
210–214.
Lau,
D.P.,
Huang,
Z.,
Lui,
H.,
Anderson,
D.W.,
Berean,
K.,
Morrison,
M.D.,
Shen,
L.,
Zeng,
H.,
2005.
Lasers
Surg.
Med.
37
(3),
192–200.
Lee,
L.-A.,
Cheng,
A.-J.,
Fang,
T.-J.,
Huang,
C.-G.,
Liao,
C.-T.,
Chang,
J.T.-C.,
Li,
H.-Y.,
2008.
Laryngoscope
118,
50–55.
Magee,
N.D.,
Villaumie,
J.S.,
Marple,
E.T.,
Ennis,
M.,
Elborn,
J.S.,
McGarvey,
J.J.,
2009.
J.
Phys.
Chem.
B
113,
8137–8141.
Mahadevan-Jansen,
A.,
Richards-Kortum,
R.,
1996.
J.
Biomed.
Opt.
1,
31–70.
Mo,
J.,
Zheng,
W.,
Low,
J.J.H.,
Ng,
J.,
Ilancheran,
A.,
Huang,
Z.,
2009.
Anal.
Chem.
81,
8908–8915.
Mountford,
C.E.,
Somorjai,
R.L.,
Malycha,
P.,
Gluch,
L.,
Lean,
C.,
Russell,
P.,
Barraclough,
B.,
Gillett,
D.,
Himmelreich,
U.,
Dolenko,
B.,
Nikulin,
A.E.,
Smith,
I.C.P.,
2001.
Br.
J.
Surg.
88,
1234–1240.
Nijssen,
A.,
Bakker
Schut,
T.C.,
Heule,
F.,
Caspers,
P.J.,
Hayes,
D.P.,
Neumann,
M.H.A.,
Puppels,
G.J.,
2002.
J.
Invest.
Dermatol.
119,
64–69.
Nijssen,
A.,
Maquelin,
K.,
Santos,
L.F.,
Caspers,
P.J.,
Schut,
T.C.B.,
den
Hollander,
J.C.,
Neumann,
M.H.A.,
Puppels,
G.J.,
2007.
J.
Biomed.
Opt.
12,
034004–034007.
Savitzky,
A.,
Golay,
M.J.E.,
1964.
Anal.
Chem.
36,
1627–1639.
Santos,
L.F.,
Wolthuis,
R.,
Koljenovi´
c,
S.,
Almeida,
R.M.,
Puppels,
G.J.,
2005.
Anal.
Chem.
77,
6747–6752.
Shim,
M.G.,
Wong
Kee
Song,
L.M.,
Marcon,
N.E.,
Wilson,
B.C.,
2000.
Photochem.
Pho-
tobiol.
72,
146–150.
Short,
K.W.,
Carpenter,
S.,
Freyer,
J.P.,
Mourant,
J.R.,
2005.
Biophys.
J.
88,
4274–4288.
Short,
M.A.,
Lui,
H.,
McLean,
D.,
Zeng,
H.,
Alajlan,
A.,
Chen,
X.K.,
2006.
J.
Biomed.
Opt.
11,
034004–034009.
Stone,
N.,
Stavroulaki,
P.,
Kendall,
C.,
Birchall,
M.,
Barr,
H.,
2000.
Laryngoscope
110,
1756–1763.
Teh,
S.K.,
Zheng,
W.,
Ho,
K.Y.,
Teh,
M.,
Yeoh,
K.G.,
Huang,
Z.,
2008.
Br.
J.
Cancer
98,
457–465.
Teh,
S.K.,
Zheng,
W.,
Lau,
D.P.,
Huang,
Z.,
2009a.
Analyst
134,
1232–1239.
Teh,
S.K.,
Zheng,
W.,
Ho,
K.Y.,
Teh,
M.,
Yeoh,
K.G.,
Huang,
Z.,
2009b.
J.
Raman
Spectrosc.
40,
908–914.
Widjaja,
E.,
Zheng,
W.,
Huang,
Z.,
2008.
Int.
J.
Oncol.
32,
653–662.
Wong
Kee
Song,
L.M.,
Molckovsky,
A.,
Wang,
K.K.,
Burgart,
L.J.,
Dolenko,
B.,
Somorjai,
R.L.,
Wilson,
B.C.,
2005.
Proc.
SPIE
5692,
140–146.