Fourier transform infrared for noninvasive optical

diagnosis of oral, oropharyngeal, and laryngeal cancer

GEORGINA E. MENZIES, HANNAH R. FOX, CONOR MARNANE, LAYSAN POPE, VINOD PRABHU,
STUART WINTER, ANNA V. DERRICK, and PAUL D. LEWIS
SINGLETON PARK, SWANSEA, AND OXFORD, UNITED KINGDOM

The 5-year survival rate for advanced head and neck cancers is 50%. There is cur-
rently no noninvasive method or effective screening procedure available to diag-
nose head and neck cancer at the earliest stages when it is still highly curable. This
study aims to show how Fourier transform infrared (FTIR) spectroscopy could be
used as a sensitive, noninvasive, low cost technique to diagnose head and neck
cancer at an earlier stage and, thus, increase the likelihood of survival. Sputum sam-
ples were collected from 16 cases with oral or oropharyngeal cancer, 8 cases with
laryngeal cancer patients and 15 normal controls. Cell pellets were produced from
each of these samples and used to generate FTIR spectra within the ‘biochemical fin-
gerprint’ wavenumber region of 1800 to 950 cm21. Discrimination between cancer
and normal sputum was achieved using infrared wavenumbers 1650 cm21,
1550 cm21, and 1042 cm21 determined by robust feature selection. These 3 wave-
numbers were used to develop potential models to discriminate both oropharyngeal
and laryngeal cancer from normal control. In cancer cases, the absorbance levels
for 1550 cm21 were increased relative to controls, whereas 1042 cm21 absorbance
was decreased suggesting changes to protein and glycoprotein structure within
sputa cells. This preliminary study shows potential for how FTIR could be developed
into a simplistic diagnostic tool that could easily be implemented by a nonspecialist
to diagnose and monitor head and neck cancer. The method could especially pro-
vide a means for detecting laryngeal cancer hidden from noninvasive observation.
(Translational Research 2014;163:19–26)

Abbreviations: FTIR ¼ fourier transform infrared; IR ¼ infrared; PLS ¼ partial least squares; RR ¼ rel-
ative risk; SCC ¼ squamous cell carcinoma; SD ¼ standard deviation; SlimPLS ¼ PLS method;
UK ¼ United Kingdom

H
ead and neck cancer represents the fifth most cancers are of squamous origin, arising from any part
common cancer in men and eighth most com- of the upper respiratory tract. Tumors of the larynx
mon in women worldwide with about have represented the largest proportion of cancers his-
600,000 new cases each year.1 More than 90% of the torically, but over the last few decades, the incidence

From the College of Medicine, Swansea University, Singleton Park, Reprint requests: Paul D. Lewis, College of Medicine, Swansea Uni-
United Kingdom; Department of Otolaryngology, Singleton versity, Singleton Park, SA2 8PP, United Kingdom; e-mail: p.d.
Hospital, Swansea, United Kingdom; West Wing, John Radcliffe, [email protected].
Oxford Cancer Center, Churchill Hospital, Oxford, United Kingdom. 1931-5244/$ - see front matter
Conflict of interest: None. Ó 2014 Mosby, Inc. All rights reserved.
This work was funded through a grant provided by Abertawe Bro- http://dx.doi.org/10.1016/j.trsl.2013.09.006
Morgannwg University Health Board.
Submitted for publication April 4, 2013; revision submitted
September 6, 2013; accepted for publication September 10, 2013.

19
 Translational Research
20 Menzies et al January 2014

opment of optical diagnostic modalities for head and

AT A GLANCE COMMENTARY neck cancer.6 Optical diagnostic devices use light of
varying wavelength to detect differences between dis-
Menzies GE, et al
ease, and abnormal tissue that phenotypically might
Background be similar. A recent review by Jerjes et al highlights
The 5-year survival rate for advanced head and a range of minimally invasive in vitro and in vivo optical
neck cancers is 50%. This study shows how optical diagnostic techniques that show promise to distinguish
spectroscopy, based on a small panel of infrared pathologic stages of head and neck disease including
wavelengths, could be used as a sensitive, low elastic scattering spectroscopy, differential pathlength
cost technique to diagnose head and neck cancer spectroscopy, near-infrared spectroscopy, Raman spec-
at an earlier stage and thus increase the likelihood troscopy, confocal reflectance microscopy, fluorescence
of survival. imaging, microendoscopy, and optical coherence to-
mography.7 One optical diagnostic method that has
Translational Significance been evaluated in a number of studies for head and
Fourier transform infrared (FTIR) spectroscopy neck tissue is Raman spectroscopy.8
could be developed into a simplistic diagnostic, Raman spectroscopy relies on in-elastic light scatter-
tool that could easily be implemented by a nonspe- ing. A Raman spectrum results from a shift in frequency
cialist to predict head and neck cancer especially in the incident excitation light above and below the
for detecting laryngeal cancer hidden from nonin- wavelength of the incident photons. For biological tis-
vasive observation. sue, the shift in frequency is due to the vibrational fre-
quencies of the molecules, and structural change to
these molecules could serve as a biomarker. The tech-
nique is able to distinguish between premalignant, ma-
of oral/oropharyngeal disease has been rapidly increas- lignant, inflammatory, and normal biopsy tissue.9
ing. In the United Kingdom (UK), oral/oropharyngeal Devpura et al have also used Raman spectroscopy to
disease now represents the highest burden of disease. distinguish tissue, in situ, of normal and squamous
Historically, smoking and alcohol have represented cell carcinoma (SCC).10 Harris et al have evaluated
the main risk factors for head and neck cancer.2 Most the sensitivity and specificity of Raman spectroscopy
head and neck cancers associated with these risk factors to detect head and neck cancer using peripheral blood
arise in the 601 group. There is a rising incidence in samples with promising results.11 Interestingly, 3 stud-
oral/oropharyngeal disease world-wide, which has ies have demonstrated how Raman spectroscopy is
been attributed to human papillomavirus infection. Re- able to differentiate between laryngeal and abnormal
cent studies, using postoperative radiotherapy and con- or normal biopsy tissue with high sensitivity and speci-
comitant chemotherapy have been able to show a small ficity.12-14 Another study by Lau et al showed how
improvement in survival,3 but the 5-year survival of Raman spectroscopy was able to classify tissue
head and neck cancer patients, including oral and laryn- obtained from cancer and control cases in the
geal, is still around 63%,4 and is associated with high nasopharynx.15
rates of morbidity and a severe impact on patients qual- Any diagnostic method that is noninvasive has obvi-
ity of life. Late presentation of lesions, lack of suitable ous benefits, and for head and neck cancer, a technology
markers for early detection, and failure of advanced le- providing analysis of sputum would present an ideal di-
sions to respond to conventional treatments contribute agnostic modality. Surface enhanced Raman spectros-
to a poor outcome. The diagnosis of head and neck can- copy has been previously used to determine
cer is usually made after cytologic and/or histologic biochemical changes in saliva of oral cancer cases.
evaluation of biopsy specimens in symptomatic pa- Whereas this technique was successful, it required the
tients. use of gold nanoparticles to enhance the Raman spec-
There is currently no method or effective screening troscopy signal as otherwise the spectra are weak pre-
procedure available to diagnose head and neck cancer venting adequate analysis.16
at the earliest stages when it is still highly curable. A Another optical diagnostic technology with potential
large-scale cluster-randomized controlled trial in India for head and neck cancer tissue discrimination is Fourier
concluded that oral visual screening could reduce oral transform infrared (FTIR) spectroscopy. The principles
cancer mortality in high-risk cases.5 An early detection of FTIR and how infrared (IR) light is absorbed by bio-
for laryngeal cancer, where the tumor is hidden from molecules are as follows. All atoms in a molecule are in
noninvasive observation, is particularly desirable. In re- continuous vibration relative to each other. If a certain
cent years, there has been increased interest in the devel- frequency of IR radiation is directed onto a molecule
Translational Research
Volume 163, Number 1 Menzies et al 21

and this frequency is the same as a vibration frequency at the time of sampling (mean age 60.2 6 8.4 years; 11
within the molecule, then the radiation will be absorbed. males, 4 females). We were not able to obtain sputum
Thus, the total range of IR frequencies that a molecule from nonsmoker volunteers. All control cases had not
will absorb is characteristic of the molecular structure. previously had a diagnosis of any cancer type.
An IR spectrum for a biological sample charts the quan- Sputum collection and processing. Each subject took
tity of absorbance across a whole range of frequencies a deep cough and expectorated sputum into a multispeci-
that correlates with the amount of molecular functional men container. After collection, sputum samples were
groups. Such a spectral pattern is often referred to as kept frozen at 280 C. Prior to processing, sputum
a biochemical ‘‘fingerprint.’’ An FTIR spectrometer is was defrosted at room temperature for 12–24 hours.
the instrument used to detect IR light absorbed by mol- Cells within the sputum were isolated by breaking
ecules. down the expectorate with a solution consisting of
Preliminary studies have shown how FTIR is able to 2.5 g dithiothreitol in 31 mL Cytolyt (Fluka Biochemika
measure biochemical changes in tissue specimens and Sigma-Aldrich Chemie GmbH, Buchs, Switzerland),
discriminate oral and oropharyngeal tumor from normal which digested mucus and other large glycoproteins in
tissue.17-19 We have previously evaluated FTIR for the specimen. Samples then underwent centrifugation
diagnosing lung cancer using sputum with putative at 3000 rpm for 10 minutes and supernatant poured
high sensitivity and specificity.20 Therefore, we per- off leaving a cell pellet. Residual pellets were freeze-
formed a pilot study to evaluate the ability of FTIR, us- dried overnight, diluted in 200 mL of sterile distilled
ing a portable and simple-to-use device, to discriminate water, agitated for 5 minutes, and split into 20 mL
cancer patients from normal patients using sputum by aliquots.
biochemical changes. Here we provide the findings of
this study and show how FTIR and feature selection, us- FTIR spectroscopy. A portable Bruker Alpha-FTIR
ing a partial least squares (PLS) method (SlimPLS),21 mid-infrared spectrometer (Bruker UK Limited,
could provide a sensitive and specific optical diagnostic Coventry, UK) was used to generate infrared spectra.
technology for the detection of oral, oropharyngeal, and For each sample, a 5 mL aliquot of cell pellet was
laryngeal cancer. spotted directly onto the sampling module and air-
dried at room temperature. Infrared absorbance
spectra were generated at a resolution of 2 cm21 from
METHODS the 4000 to 450 cm21 wavenumber range and the
Study subjects. This study had approval from the loco- frequency region between 950 to 1800 cm21 retained
regional ethical committee (10/WMW02/6), and for analysis. Each spectrum represented the average of
informed consent was gained in each case prior to 24 scans for improved signal to noise ratio. This
inclusion. Sputum was collected from 24 patients with process was repeated until 6 replicates were collected
a final diagnosis of head and neck cancer: 7 tonsil for each sample.
SCC (5 right tonsil, 2 left tonsil); 1 floor of the mouth Data processing and analysis. All data processing,
SCC; 2 tongue SCC (1 base of tongue, 1 left tongue); analysis and visualization were performed using Bruker
6 pharyngeal SCC; 8 laryngeal SCC (mean age Opus 6.5 software (Bruker UK Limited) and the R sta-
61.9 6 8.6 years); 21 males, 3 females; 7 current tistical computing environment22 using in-built
smokers, 6 exsmokers, 3 never-smokers; 17 with algorithms and code developed by our group. We
a weekly alcohol consumption, 4 nondrinkers, 3 assessed the ‘fingerprint’ spectral region so
unknown status. All tumors with a known grading absorbance values between wavenumbers 950 and
were moderately differentiated with the exception of 1 1800 cm21 (601 data points) were preprocessed prior
poorly differentiated case. Data for p16 status of oral/ to further analysis. Spectra were baseline corrected
oropharyngeal tumors were not available. All samples and smoothed using a 9-point Savitzky-Golay
from cancer cases were taken prior to investigative algorithm and then vector normalized. We applied the
surgery in the Ear Nose and Throat Department of Shapiro-Wilk test to confirm that absorbance values
Singleton Hospital, Swansea, and informed consent to for each wavenumber followed a normal distribution.
provide a sputum sample was obtained from each Feature selection of discriminatory wavenumbers using
patient at a previous clinical appointment. SlimPLS. Feature selection for infrared wavenumbers
Sputum was also collected from a control group of 15 that could significantly discriminate between 2 groups
smokers, staff members at Swansea University, with low of samples was performed using the SlimPLS algo-
to moderate weekly alcohol consumption who had not rithm in R.21 Briefly, SlimPLS is a feature selection
had a previous diagnosis of cancer and had no symp- method based on PLS. PLS generalizes and combines
toms to be considered high risk for head and neck cancer features from principal components analysis and
 Translational Research
22 Menzies et al January 2014

Fig 1. Mean spectra of cancer (solid line) and normal (dotted line) sputum samples; from the 1800 cm21 to
950 cm21 wavenumber regions.

multiple linear regression. It models the relationships cases and 15 normal control cases. Spectra repeated
between sets of observed features (eg, wavenumbers) for each case were highly reproducible and the genera-
using orthogonal score vectors, or latent variables, tion time per spectrum using the Bruker Alpha (Bruker
called components. The components account for as UK Limited) was approximately 10 minutes. Mean
much of the variation in the data as possible, while spectra for each group are shown in Figure 1. The
PLS also simultaneously models response based on prominent absorbance bands observed in both cancer
predictor variables. PLS has, thus, been used widely and normal spectra are characteristic of vibration
as a classification method to construct predictive modes assumed to represent functional groups in
models from large multidimensional data sets. cellular molecules including proteins, carbohydrates,
The SlimPLS algorithm adapts PLS as a robust and nucleic acids.23-25
method for extracting a small subset of features In all spectra, major peaks were observed for absorp-
from a much larger set that best discriminate 2 cate- tions in the amide I and II regions at 1650 cm21 and
gories of dependent variables. It is well suited for fea- 1540 cm21, respectively. A second prominent set of
ture selection when the number of features (in this peaks was observed in the carbohydrate and nucleic
case wavenumbers) far exceeds the number of sam- acid associated regions between 1000 cm21 and
ples. Although it does not provide a predictive model 1150 cm21. Differences in IR absorbance levels be-
like conventional PLS, SlimPLS provides the best set tween cancer and normal can be seen throughout the
of feature variables for further predictive model devel- spectra particularly within the 1000 cm21 and
opment. The SlimPLS algorithm determines the sig- 1150 cm21 region.
nificance level of each feature assessed, and in this
Discriminatory wavenumbers for cancer and normal
study, only wavenumbers having a P value less than sputum. The SlimPLS algorithm was used to determine
0.05 were retained. wavenumbers that were statistically significant in
As this research is a small pilot study, we have not at- discriminating cancer from normal sputum. The 3
tempted to derive robust predictive models of cancer. most significant discriminatory wavenumbers were
However, relative risk for each status was calculated us- 1650 cm21 (mean absorbance and standard deviation
ing normalized absorbance of key discriminatory wave- [SD]: cancer 5 2.69 [0.12]; normal 5 2.50 [0.14]);
numbers along with estimated levels of sensitivity and 1550 cm21 (mean/SD: cancer 5 3.53 [0.21];
specificity. normal 5 3.50 [0.35]); 1042 cm21 (mean/SD:
cancer 5 0.71 [0.12]; normal 5 0.96 [0.22]). Boxplots
RESULTS representing the absorbance distribution for each of
Infrared absorbance spectra. FTIR was used to gener- these wavenumbers for cancer and normal cases are
ate absorbance spectra in the frequency region 950 to shown in Figure 2. Absorbance bands at 1650 cm21
1800 cm21 to establish biochemical differences in cells and 1550 cm21 characteristically reflect stretching of
extracted from sputum for 24 head and neck cancer C 5 O bonds and bending of N-H bonds (in addition
Translational Research
Volume 163, Number 1 Menzies et al 23

Fig 2. Boxplots to show the distribution of absorbance for the 3 wavenumbers 1650 cm21, 1550 cm21, and
1042 cm21 determined significantly between the cancer (C) and normal (N) groups by the SlimPLS algorithm.
The median and quartiles are enclosed by the box. The bars data still within 1.5 times either the upper or lower
interquartile ranges. Circles represent outliers.

to stretching of C-N bonds) in amide linkages of greater than 3 has a RR for being high-risk of 3.75 (95%
proteins.24 The absorbance band at 1042 cm21 reflects confidence interval 5 1.29, 10.86).
C-O stretching and C-O bending which are tentatively
assigned to caborhydrates.24
The pattern of absorbance levels for these 3 wave- DISCUSSION
numbers combined, between cancer and normal spu- The objective of this study was to determine signifi-
tum, can be visualized using a 3-dimensional scatter cant wavenumbers that could be used to distinguish be-
plot (Fig 3). Cancer and normal cases are separated di- tween cancer and control groups using FTIR
agonally through the plot (with the exception of 1 can- spectroscopy applied to sputum. Previously, we have
cer case) with an apparent high sensitivity and shown, using multivariate statistical analysis, that spu-
specificity. The scatterplot showed no separation be- tum can be used to show biochemical changes between
tween histologic subtypes (ie, SCC and non-SCC) of cohorts.20 In that study, we used the common ‘filter’
cancer cases. method of applying a series of univariate differential
We then calculated a ratio of absorbance for the statistics to determine wavenumbers that significantly
1550 cm21 and 1042 cm21 wavenumbers to derive differentiated cancer from control cases. In this study,
a simple model using a ratio cut-off of 3 that optimally we have used a feature selection method based on par-
discriminates cancer from normal cases. This model is tial least squares, SlilmPLS, to determine and evaluate
shown graphically in Figure 4 where each ratio is ranked significant wavenumbers between sputum in disease
from lowest to highest and a cut-off line drawn at a ratio state groups. This method was chosen for 2 reasons.
of 3 for 1550 cm21 and 1042 cm21. In cancer cases, the First, SlimPLS has been shown to have a number of ad-
absorbance levels for 1550 cm21 are increased relative vantages over other filter methods when applied to large
to controls, whereas for 1042 cm21 they are decreased datasets with hundreds to thousands of features.21 Sec-
(Fig 4). We were then able to determine the relative ond, in this present study, we compared the wavenum-
risk (RR) for cancer dependent on the proportions of bers returned as significant by SlimPLS with those
cancer and normal cases having ratio values above or obtained by univariate statistics after correcting for
below the cut-off. A case having a ratio greater than 3 multiple testing (results not shown). That method re-
has an estimated RR for cancer of 6.24 (95% confidence turned only 2 significant wavenumbers where 1 was
interval 5 1.81, 21.70). Similarly, a case having a ratio not part of an absorbance peak and, thus, difficult to
 Translational Research
24 Menzies et al January 2014

Fig 3. Scatter plot for the absorbance levels for the 3 significant wave-
numbers 1650 cm21, 1550 cm21, and 1042 cm21 showing the differ- Fig 4. Plot showing the absorbance ratio scores for 1550 cm21 and
ences between cancer (solid circles) and normal (clear circles) sputum. 1042 cm21 wavenumbers in each case ranked from lowest to highest
ratio along the x-axis: cancer (solid circles), high risk (grey circles),
and normal sputum (clear circles).

interpret. Thus, we view as SlimPLS as a superior

method. changes to glycans in secreted mucins, possible in re-
The overall distribution of differences in absorbance sponse to inflammation. A similar strategy for sputum
observed between spectra of sputum from cancer and would permit insight into exact molecular changes ob-
normal control cases was markedly less than that ob- served in head and neck cancer.
served for a similar comparison between lung cancer An important finding of our study is that FTIR was
and normal control sputa.20 Absorbance differences able to identify biochemical changes in laryngeal as
were most predominant between head and neck cancer well as oral and oropharyngeal cancer sputa relative to
and normal sputum in both the 1000 cm21 and the normal controls. Preliminary detection of laryngeal tu-
1150 cm21 wavenumber regions. When discriminating mors is more difficult than that for the majority of oral
between cancer and normal control sputum samples, 3 and oropharyngeal tumors. Similar absorbance levels
significant wavenumbers allowed visualization of across wavenumbers 1650 cm21 and 1550 cm21 and
groupings of cases. Wavenumbers 1650 cm21 and 1042 cm21 would suggest that certain protein and gly-
1550 cm21, occur in the Amide I and Amide II regions coprotein structural changes are shared across these tu-
where change in absorbance levels reflect structural mor types and are unrelated to location. Stone et al have
change to proteins as well as intramolecular or intermo- previously shown using Raman spectroscopy that, in bi-
lecular interactions of proteins with other molecules. opsy specimens, protein structural changes are detect-
Such changes can discriminate between cancer and con- able between normal controls and patients with
trol tissue in other cancer types.25 The third wavenum- laryngeal cancer or dysplasia.12 A further study is re-
ber (1042 cm21), associated with C-O stretching in quired to determine whether the biochemical changes
carbohydrates, shows an absorbance decrease in cancer we detect by FTIR in cancer sputa can be detected in pa-
cases relative to control. Previous studies have shown tients with dysplasia. Indeed, in a further study with in-
carbohydrate related absorbance levels to increase or creased sample size, patients deemed high risk should
decrease dependent on the cancer type.26 We have be followed prospectively to determine cancer develop-
recently determined exact infrared spectral profiles for ment and sputum samples monitored temporally for
monosaccharides and carbohydrate structures associ- such biochemical changes.
ated with glycoprotein structural changes in tumours.27 Jerjes et al show the need for advancements in optical
Using this method, we were able to correlate molecule- techniques as new, noninvasive, cost-effective diagnosis
specific biochemical changes that corresponded with in- methods for oral cancers.7 As many as 40% of oral can-
frared absorbance patterns in sputum. It is likely that cers are detected at a late stage (T3/T4) lowering the
these biochemical changes in sputum relate to structural survival rate significantly.7 Pending the results of future
Translational Research
Volume 163, Number 1 Menzies et al 25

large-scale studies, FTIR may prove to be a sensitive di- 5. Sankaranarayanan R, Ramadas K, Thomas G, et al. Trivandrum
agnostic modality for head and neck cancer, and further Oral Cancer Screening Study Group. Effect of screening on oral
cancer mortality in Kerala, India: a cluster-randomised controlled
evaluation could lead to a predictive model of disease. trial. Lancet 2005;365:1927–33.
Although we were not able to obtain cytology data for 6. Swinson B, Jerjes W, El-Maaytah M, Norris P, Hopper C. Optical
the sputum samples in this study, a future large-scale techniques in diagnosis of head and neck malignancy. Oral Oncol
evaluation could compare the sensitivity and specificity 2006;42:221–8.
of FTIR against a second diagnostic procedure such as 7. Jerjes WK, Upile T, Wong BJ, et al. The future of medical diag-
nostics: review paper. Head Neck Oncol 2011;3:38.
cytologic examination.28 As with our previous study, 8. Harris AT, Rennie A, Waqar-Uddin H, et al. Raman spectroscopy
we show FTIR to be able to determine biochemical in head and neck cancer. Head Neck Oncol 2010;2:26.
changes in cancer patients in a noninvasive sample 9. Malini R, Venkatakrishna K, Kurien J, et al. Discrimination
source.20 FTIR offers 2 further advantages over other di- of normal, inflammatory, premalignant, and malignant oral
agnostic modalities. First, the whole process of sam- tissue: a Raman spectroscopy study. Biopolymers 2006;81:
179–93.
pling through to generation of an infrared spectrum is 10. Devpura S, Thakur JS, Sethi S, Naikc VM, Naika R. Diagnosis of
extremely cheap and only requires basic training. Sec- head and neck squamous cell carcinoma using Raman spectros-
ond, modern spectrometers, such as that used in this copy: tongue tissues. J Raman Spec 2011;43:490–6.
study, are portable and may be transported between 11. Harris AT, Lungari A, Needham J, et al. Potential for Raman spec-
clinics. Thus, there is much optimism that FTIR could troscopy to provide cancer screening using a peripheral blood
sample. Head Neck Oncol 2009;1:38.
be developed as a noninvasive and cost-effective moni- 12. Stone N, Stavroulaki P, Kendall C, Birchall M, Barr H. Raman
toring system for a number of cancers. spectroscopy for early detection of laryngeal malignancy: prelim-
inary results. Laryngoscope 2000;10:1756–63.
CONCLUSION 13. Lau DP, Huang Z, Lui H, et al. Raman spectroscopy for optical di-
agnosis in the larynx: preliminary findings. Lasers Surg Med
This preliminary study has shown the possibility for 2005;3:192–200.
using FTIR as a noninvasive diagnostic technique for 14. Teh SK, Zheng W, Lau DP, Huang Z. Spectroscopic diagnosis of
head and neck cancer for the detection of biochemical laryngeal carcinoma using near-infrared Raman spectroscopy and
changes in sputum. Importantly, the method detects la- random recursive partitioning ensemble techniques. Analyst
2009;134:1232–9.
ryngeal tumors, hidden from noninvasive observation, 15. Lau DP, Huang Z, Lui H, et al. Raman spectroscopy for optical di-
in addition to oral and oropharyngeal tumors providing agnosis in normal and cancerous tissue of the nasopharynx-
a single set of biomarkers for multiple tumor types. preliminary findings. Lasers Surg Med 2003;32:210–4.
FTIR could be developed into a simplistic diagnostic 16. Kah JC, Kho KW, Lee CG, et al. Early diagnosis of oral cancer
tool that could be easily implemented by a nonspecialist. based on the surface plasmon resonance of gold nanoparticles.
Int J Nanomedicine 2007;2:785–98.
The biochemical changes observed in sputum between 17. Pallua JD, Pezzei C, Zelger B, et al. Fourier transform infrared im-
normal and cancer cases in this study could be a used aging analysis in discrimination studies of squamous cell carci-
as a noninvasive way to monitor disease progression. noma. Analyst 2012;137:3965–74.
Further work needs to be done to fully evaluate the sen- 18. Fukuyama Y, Yoshida S, Yanagisawa S, Shimizu M. A study on
sitivity and specificity of the minimal set of infrared the differences between oral squamous cell carcinomas and nor-
mal oral mucosas measured by Fourier transform infrared spec-
wavenumbers to predict head and neck cancer. Future troscopy. Biospectroscopy 1999;5:117–26.
work should also attempt to evaluate whether these bio- 19. Schultz CP, Liu KZ, Kerr PD, Mantsch HH. In situ infrared histo-
chemical changes can be detected in patients deemed at pathology of keratinization in human oral/oropharyngeal squa-
risk of head and neck cancer for the potential of early mous cell carcinoma. Oncol Res 1998;10:277–86.
detection. 20. Lewis PD, Lewis KE, Ghosal R, et al. Evaluation of FTIR spec-
troscopy as a diagnostic tool for lung cancer using sputum.
The authors acknowledge Aaran Lewis for his assistance with BMC Cancer 2010;10:640.
FTIR. This work was funded through a grant provided by Abertawe 21. Gutkin M, Shamir R, Dror G. SlimPLS: a method for feature se-
Bro-Morgannwg University Health Board. lection in gene expression-based disease classification. PLoS One
2009;4:6416.
22. R Development Core Team. R. A language and environment for
REFERENCES
statistical computing. R Foundation for Statistical Computing,
1. IARC. Cancer incidence in five continents, vol 1X. Available at: Vienna, Austria. 2009. Available at: http://www.R-project.org.
http://www-dep.iarc.fr. Accessed December 8, 2013. Accessed August 8, 2013.
2. La Vecchia C, Tavani A, Franceschi S, Levi F, Corrao G, Negri E. 23. Wang HP, Wang HC, Huang YJ. Microscopic FT-IR studies of lung
Epidemiology and prevention of oral cancer. Oral Oncol 1997;5: cancer cells in pleural fluid. Sci Total Environ 1997;204:283–7.
302–12. 24. Stuart BH. Infrared spectroscopy: fundamentals and applications.
3. Warnakulasuriya S. Global epidemiology of oral and oropharyn- New York: John Wiley & Sons, Inc, 2004.
geal cancer. Oral Oncol 2009;45:309–16. 25. Maziak DE, Do MT, Shamji FM, Sundaresan SR,
4. Siegel R, Naishadham D, Jemal A. Cancer statistics, 2012. CA Perkins DG, Wong PT. Fourier-transform infrared spectro-
Cancer J Clin 2012;62:10–29. scopic study of characteristic molecular structure in cancer
 Translational Research
26 Menzies et al January 2014

cells of esophagus: an exploratory study. Cancer Detect Prev 27. Lewis AT, Jones K, Lewis KE, Jones S, Lewis PD. Detection of
2007;31:244–53. Lewis antigen structural change by FTIR spectroscopy. Carbohy-
26. Yano K, Ohoshima S, Gotou Y, Kumaido K, Moriguchi T, drate Polymers 2013;92:1294–301.
Katayama H. Direct measurement of human lung cancerous and 28. El-Tawil SG, Adnan R, Muhamed ZN, Othman NH. Compara-
noncancerous tissues by Fourier transform infrared microscopy: tive study between Pap smear cytology and FTIR spectroscopy:
can an infrared microscope be used as a clinical tool? Anal Bio- a new tool for screening for cervical cancer. Pathology 2008;40:
chem 2000;15:218–25. 600–3.