Biosensors and Bioelectronics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Biosensors and Bioelectronics

journal homepage: www.elsevier.com/locate/bios

Development and application of a new electronic nose instrument

for the detection of colorectal cancer
E. Westenbrink a,n, R.P. Arasaradnam b,c, N. O'Connell c, C. Bailey c, C. Nwokolo c,
K.D. Bardhan b,d, J.A. Covington a
a
School of Engineering, University of Warwick, Coventry CV4 7AL, UK
b
Clinical Sciences Research Institute, University of Warwick, Coventry CV2 2DX, UK
c
Department of Gastroenterology, University Hospital Coventry and Warwickshire, Coventry CV2 2DX, UK
d
Rotherham General Hospital, Rotherham S60 2UD, UK

art ic l e i nf o a b s t r a c t

Article history: Colorectal cancer is a leading cause of cancer death in the USA and Europe with symptoms that mimick
Received 19 June 2014 other far more common lower gastrointestinal (GI) disorders. This difficulty in separating colorectal
Received in revised form cancer from these other diseases has driven researchers to search for an effective, non-invasive screening
26 September 2014
technique. Current state-of-the-art method of Faecal Immunochemical Testing achieving sensitivity
Accepted 8 October 2014
 90%, unfortunately the take-up in the western world is low due to the low patient acceptability of stool
samples. However, a wide range of cancers have been distinguished from each-other and healthy controls
Keywords: by detecting the gas/volatile content emanating patient biological media. Dysbiosis afforded by certain
Electronic nose disease states may be expressed in the volatile content of urine – a reflection of the gut bacteria′s me-
Gas sensors
tabolic processes. A new electronic nose instrument was developed at the University of Warwick to
Volatile organic compounds
measure the gas/volatile content of urine headspace, based on an array of 13 commercial electro-che-
Colorectal cancer
Urine mical and optical sensors. An experimental setup was arranged for a cohort of 92 urine samples from
patients of colorectal cancer (CRC), irritable bowel syndrome (IBS) and controls to be run through the
machine. Features were extracted from response data and used in Linear Discriminant Analysis (LDA)
plots, including a full 3-disease classification and one focussing on distinguishing CRC from IBS. The latter
case was tested by the success of re-classification using an (n 1) K-nearest neighbour algorithm,
showing 78% sensitivity and 79% specificity to CRC.
& 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-SA
license (http://creativecommons.org/licenses/by-nc-sa/3.0/).

1. Introduction invasive, has an associated morbidity and is not sufficiently cost

1.1. Disease background
Colorectal cancer (CRC) remains one of the leading causes of
cancer-related death in Europe and the USA (Siegel et al., 2012;
Ferlay et al., 2010). For this reason there has been significant re-
search focus on attempting to find an effective and unobtrusive
method for screening at risk individuals. The challenge is the
broad overlapping symptoms exhibited from patients suffering
from CRC and from those with other lower gastro-intestinal (GI)
diseases, especially very common disorders such as irritable bowel
syndrome (IBS). The gold standard diagnostic test for this disease
is colonoscopy with biopsy. However, this technique is highly
n covered by to be a malignant melanoma (Williams and Pembroke,
Correspondence to: School of Engineering, University of Warwick, Gibbet Hill
Road, CV4 7AL, UK.
E-mail address: e.w.westenbrink@warwick.ac.uk (E. Westenbrink).
effective for large-scale screening purposes. Current non-invasive
options for CRC screening has shifted from measuring faecal occult
blood to Faecal Immunochemical Testing (FIT) for haemoglobin at
specific cut-off thresholds. This method shows an impressive
specificity of 87–96% towards the disease, but has a high variation
in sensitivity with the potential to produce reasonably low values
(66–88%) (De Meij et al., 2014).
One non-invasive method, which is gaining interested for the
diagnosis of a variety of cancers (e.g. lung, breast, colorectal,
prostate), measures the volatiles/gases that emanate from human
biological media in order to find distinct bio-markers indicating a
disease state. The basis for this thinking began with an early study
that highlighted the ability of a canine to smell a discrepancy in a
legion on the hand of its owner, which was subsequently dis-
1989). Further investigations into canine olfaction have led to
discrimination of patients of prostate, breast, ovary and lung

http://dx.doi.org/10.1016/j.bios.2014.10.044
0956-5663/& 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-SA license (http://creativecommons.org/licenses/by-nc-sa/3.0/).

Please cite this article as: Westenbrink, E., et al., Biosensors and Bioelectronics (2014), http://dx.doi.org/10.1016/j.bios.2014.10.044i
2 E. Westenbrink et al. / Biosensors and Bioelectronics ∎ (∎∎∎∎) ∎∎∎–∎∎∎
cancers from healthy individuals (Lippi and Cervellin, 2012; So-
noda et al., 2011; Arasaradnam et al., 2011). This concept was ta-
ken forward but with high-end analytical Gas chromatography–
mass spectrometry instruments (GC–MS). Such instruments have
also been used to analyse patient samples of breath and urine to
distinguish between a range of different types of cancer by ana-
lysing the volatile compounds present (Peng et al., 2010, Altomare
et al., 2013, Silva et al., 2011).
Many metabolic bi-products are present in all biological waste
media, such as exhaled air, sweat, urine and faeces (Buszewski
et al., 2007). The exact mechanisms behind the generation of
particular chemical groups are extremely varied and are affected
by a broad spectrum of diet- and disease-related factors. A com-
plex interaction of colonic cells, human gut microflora and in-
vading pathogens produce the variety of gases and volatiles within
the lower GI tract (Garner et al., 2007; Probert et al., 2009). This
group has previously hypothesised that the resultant products of
this process could be measured in urine – so called urine meta-
bolomics and have found evidence to support this (Arasaradnam
et al., 2012). A potential reason for this is the ability of digestive
disease to alter the permeability of the affected GI area (Arasar-
adnam and Bardhan, 2010). This study aims to support evidence
showing that the volatile and gas groups within a patient’s urine
can be used as a bio-signature, which contains information re-
garding the disease state by successfully distinguishing colon
cancer from controls using urine samples.
1.2. Electronic nose technology
An electronic nose can be considered an umbrella term that can
be used to describe any instrument formed from an array of sen-
sors with overlapping sensitivity. First developed in the 1980’s
(Persaud and Dodd, 1982), it is an attempt to mimic the biological
olfactory system by evaluating the total chemical profile of a
complex mixture of chemical compounds, instead of detecting
each individually. A traditional electronic nose is formed from an
array of chemical sensors broadly tuned to different chemical
groups. When a sample is presented to the array, as each sensor is
different, its response to that complex odour is unique. This pat-
tern can then be learnt using a pattern recognition algorithm. One
sensing technology that can be used in such instruments are based
on electrochemical cells to oxidise or reduce a target analyte. This
process produces an ampero-metric response at the electrodes by
a subsequent ionic current flow through the sensor. Such sensors
are less affected by long term drift than a number of other sensing
technologies and the sensitivity of the newest generation of
4-electrode designs have brought down detection limits to the
parts per billion-level. These sensors do suffer from some degree
of cross-sensitivity, although some individual selectivity can be
improved by the diffusion layer. However, this ′limitation′ is key to
their successful application for an electronic nose, where the focus
is on maximising the richness of information included in each
sensor response rather than targeting individual chemicals. In
comparison, single gas detection can be achieved using specific
wavelengths of light that it emits when it is excited by absorbed
infra-red light. This technology is capable of ppm or even ppb-
level sensitivity with good reproducibility, but is severely limited
by the number of gases and volatiles that have a re-emission
wavelength narrow and specific enough to be measured. However,
it is possible to combine both specific and non-specific arrays of
sensors to enhance the overall performance on the electronic nose.
Commercial electronic noses based on electro-resistive sensing
technologies, such as metal oxides and conducting polymers, have
been previously been used to detect lung cancers from healthy
controls (Machado et al., 2005). Urine headspace has also been
shown to be a suitable medium to measure by electronic nose for
Please cite this article as: Westenbrink, E., et al., Biosensors and Bioelectronics (2014), http://dx.doi.org/10.1016/j.bios.2014.10.044i
distinguishing prostate cancer (Asimakopoulos et al., 2014). Some
instruments using ampero-metric sensors have also been em-
ployed to measure gas and volatile products in other biomedical
applications, such as the discrimination of bacteria (McEntegart
et al., 2000). There have been several studies that have employed
electronic nose technology to distinguish between gastro-
enterological disease states using patient faeces (Probert et al.,
2009) these have been limited by Faecal samples from CRC
patients and healthy controls have been run through a metal
oxide-based Cyranose 320 electronic nose in recent work. A sen-
sitivity and specificity of 85% and 87% were achieved for distinc-
tion of cancer in this study along with a reduced success in dis-
criminating advanced adenomatous polyps (a condition which
leads to risk of developing CRC) (De Meij et al., 2014). However,
there is a very low acceptability rate among patients for taking
faecal samples, which makes it a difficult medium for collecting
large cohorts. This is shown in a number of studies that employ a
screening strategy for CRC by Faecal Immunochemical Testing,
where participation rates were found to have a maximum of ap-
proximately 35% (Quintero et al., 2012, Salas et al., 2014). In this
respect, urine is much more convenient and appropriate to be
employed in a wider clinical setting as it can be collected on
demand.
2. Materials and methods
2.1. WOLF (Warwick olfaction) system
The electronic nose system used in this study has been devel-
oped and built in-house at the University of Warwick, and is the
result of a large body of research conducted in this area. It consists
of an array of 13 sensors, employing a range of different sensing
technologies. This total includes eight amperometric electro-che-
mical sensors (Alphasense Ltd., UK), two non-dispersive infra-red
(NDIR) optical devices (Clairair Ltd., UK) and a single photo-ioni-
sation detector (Mocon, USA). These sensors are each constructed
within a cylindrical package that includes all the physical com-
ponents required for the detection method. Whilst each of these
sensors are designed to target a single gas (detailed in Table 1),
they are also capable of detecting a variety of different gas and
volatile compounds. The effect of this cross-sensitivity is twofold:
there cannot be any significant specificity in response to any gas,
but the appropriate combination of sensors can be employed to
maximise the number of relevant gases that can be incorporated
into their response. This set of sensors has been chosen to combine
the detection of a number of specific gases known to be affected by
the body (such as CO2, O2 and methane) and those shown to be
linked to lower gastro-intestinal disease (NH3, SO2, and H2S). They
also have a broad range of overlapping cross-sensitivities to a
number of higher molecular weight volatile organic compounds
(VOCs). In this manner, the WOLF system attempts to maximise
the amount of information may be collected on the bio-signature
of an individual, thereby making it more applicable to the char-
acterisation and distinction of disease states.
Table 1
List of sensor manufacturers, mechanisms, and target gases included in the array.
Manufacturer Sensing method Target gases
Alphasense Ltd. Electro-chemical O2, NH3, ETOa, SO2, O3, NO, NO2, SO2, H2S,
H2
Clairair Ltd. Infra-red Optical CO2, CH4
MOCON Photo-ionisation All
a
ETO abbreviates Ethylene Oxide.
 E. Westenbrink et al. / Biosensors and Bioelectronics ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Each single sensor is plugged into an individual printed circuit
board (PCB) that contains control electronics for converting the
raw detection response into an analogue voltage level, so that it
can be easily integrated into the data acquisition architecture. For
the electro-chemical sensors this is completed by commercial In-
dividual Sensor Boards (ISBs) produced by Alphasense. These
control boards maintain the potential between the working and
counter electrodes, the biasing on a reference electrode to induce
reduction/oxidation of the target, and a current loop transceiver
for conversion of the ampero-metric signal into an analogue vol-
tage. The Clairair optical sensors are also controlled using a com-
mercially-produced PCB called Cirius X, which has an on-board
temperature/pressure monitor and drive for the infra-red sources.
The Mocon photo-ionisation detector includes a circuit for con-
verting to analogue voltage within the physical cylindrical pack-
age, so that an in-house constructed break-out PCB was all that
was required. The above control measures remove the influence of
some environmental effects that the detectors are sensitive to,
thus helping to allow for ppm- and ppb-level resolution.
The analogue voltage signals produced by the control boards
were then connected to an interface PCB that amplifies them
based on a simple op-amp design with individually-tuned gain
resistors, as well as transferring power to the breakout boards.
There are also separate environmental monitors to measure flow
rate through the machine (AWM3300V mass flow meter, Honey-
well) and temperature/humidity of the sample chamber (SHT15,
Sensirion) whose data are converted and transferred through this
PCB. The amplified channels are sent to NI USB-6009 and NI USB-
6211 OEM data acquisition (DAQ) boards produced by National
Instruments Ltd. These communicate via universal serial bus (USB)
to an all-in-one desktop PC board running LabVIEW 2012, built
into the electronic nose.
A program has been developed in LabVIEW for this application
that reads the data from the DAQs and writes it all (along with
labelling headers) into separate text files for each individual test.
The front panel of the program allows users to change testing
variables such as the duration of sample tests and the name of the
resulting text file, as well as to monitor the raw sensor outputs and
environmental factors both during and between experiments.
Photographs of the exterior of the system and user interface for
the LabVIEW software are included in Supplementary Figures,
which also display the interior of the PC case with the sensor
chamber, ISBs and desktop PC board visible.
2.2. Experimental method
To test the efficacy of the WOLF, a number of chemical standards
were first used. Here, an experimental method was setup using a
Dri-Blocks DB-2D (Techne) heater to allow the release of headspace
from test samples based on aqueous solutions. A flow of clean dry air
from a laboratory supply was regulated and then split into two
channels, one of which was sent into the sample with a valve to cut
off flow in between sample runs. A cheque valve was included in the
sample channel before re-joining with the other “make-up” flow
path, which prevents a sudden increase in air pressure and flow rate
upon introduction of a sample to minimise the effect on sensor re-
sponse. A custom heating block was machined to hold 30 mL uni-
versal sample containers, and fitted into the heater before being
brought to a temperature of 4070.1 °C. Sample aliquots were to be
held in 30 mL universal sample containers (in conjunction with
standard containers used in the medical field) with customised caps
that included bulkhead fittings to allow air flow.
Initially, a variety of different short-chain organic compounds
with different functional groups (such as ketone, ester, alcohol,
alkane and aromatic) were dissolved in aqueous solution at a
range of concentrations on the ppm scale. The full list of these
Please cite this article as: Westenbrink, E., et al., Biosensors and Bioelectronics (2014), http://dx.doi.org/10.1016/j.bios.2014.10.044i
3
Table 2
Patient demographics for urine sample cohort run through the WOLF.
CRC IBS Controls
Number 39 35 18
Mean age 70 48 41
Male % 70 11 70
Mean BMI 27.1 28 24
Current smokers (%) 6.7 10 6.7
Alcohol-average units per week 7.3 2 9.2
compounds is given in a Supplementary Table. These were used to
determine the cross-sensitivities of the WOLF sensors to these
chemical standards. The 5 mL samples were heated to 4070.1 °C
for 5 min in order to build up headspace and the injected into the
WOLF. The sample headspace flowed into the WOLF for 5 min and
then replaced with clean air (repeated 3 times). The sequence of
solution concentrations tested was randomised to mitigate the
chances that a long-term response drift would allow for fabrica-
tion of response change with increasing concentration.
After gaining insight on how known volatile groups affected
the response of the WOLF sensors, testing of the machine’s diag-
nostic capabilities on biological samples was undertaken. Urine
samples were collected from pre-diagnosed patients recruited at
the University Hospital of Coventry and Warwickshire, including a
total of 39 suffering from colorectal cancer (CRC) and 35 from ir-
ritable bowel syndrome (IBS). In addition, 18 additional healthy
controls were tested (Ethical Approval Number: 09/H1211/38).
Table 2 below shows the demographics of each group. All cases of
CRC (adenocarcinoma) were confirmed on colonoscopy and his-
tology. The demographics do show a variation in some factors,
including gender and age. However, these are indicative of the
average demographics of patients found within the disease groups.
For example IBS sufferers are much more likely to be female, and
CRC patients are only very rarely under the age of 60. All samples
were collected as standard spot urine early in the morning to
ensure a common fasting period, and urine “dipstick” tests were
conducted to rule out the presence of infection, diabetes or renal
disease. Urine samples were stored frozen at 80 70.1 °C within
2 h of being collected, and then defrosted overnight at 5 70.1 °C
before experimental testing. These were run through an identical
experimental method to the volatile solutions, with 5 mL aliquots
running with initial headspace build-up at 407 0.1 °C for 5 min
followed by another 5 min of sample introduction. This was also
repeated three times per sample, again to ensure that a full profile
of volatile and gaseous content of each was being measured.
2.3. Statistical analysis
Statistical analysis was undertaken using multivariate techni-
ques common to electronic noses. First, features were extracted
from the raw data based on the baseline-corrected voltage chan-
ges averaged over 3 points (Sig-Base3), the response integrals from
the start of response to the maximum response (AreaMax) and the
times for the sensor responses to return from maximum to 50% of
that value (T50). Eqs. (1)–(3) were used to calculate these three
features in respective order:
⎛ ∑max + 1 x ∑min + 1 x ⎞⎟ ⎛⎜ ∑3x = 1 x ⎞⎟
Sig − Base3(x) = ⎜⎜ max − 1 − min − 1 ⎟−⎜ ⎟
⎝ 3 3 ⎠ ⎝ 3 ⎠ (1)
4 E. Westenbrink et al. / Biosensors and Bioelectronics ∎ (∎∎∎∎) ∎∎∎–∎∎∎

imax
AreaMax (x) = ∑ xi
i=1 (2)

T 50(x) = t(xmax /2) , where t > t xmax (3)

Once extracted, these features were processed by Linear Dis-

criminant Analysis (a pre-classified method), using a commercial
package (Multisens Analyser, JLM Innovation, Germany). The
method initially involved assigning a disease-state class to each
sample and extracting a set of individual features from the re-
sponses of the sensors to it. These features were then ranked by a
control algorithm to maximise the separation between samples of
different class while minimising spread within classes and com-
bined into one or more discriminant functions (bound by the
constraint of the number of classes). The general equation for
discriminant function g(x) is illustrated below in Eq. (4), where wi
and xi are respectively the weights and values of the ‘ith′ extracted
feature (ranging from 1 to d), and w0 is a constant threshold Fig. 1. Ethylene oxide sensor response to a range of concentrations of aqueous
toluene solution.
weight. Classifications were made both for comparing the data of
all three groups, as well as for a two-group CRC and IBS compar-
ison that more accurately simulates a clinical situation where
a patient arrives complaining of symptoms linked to lower GI
disease.
d
g (x) = w0 + ∑ wi xi
i=1 (4)

Once a full classification of the latter case was constructed

graphically, single samples were removed from the pre-classified
“training set” and re-introduced as unknowns. The sensitivity and
specificity for this study were calculated using the success rate of
the K-nearest neighbour algorithm in re-classifying individual
blind samples, given the rest of the samples as a training set. Each
individual sample was re-classified, with the result being com-
pared to the actual group it belonged to. A tally of true and false
positives/negatives was compiled, with overall numbers being
used to calculate sensitivity and specificity using Eqs. (5) and (6)
shown below (TP is true positive, FP is false positive, TN is true
negative and FN is false negative).
TP
Sensitivity =
TP + FN (5)

TN
Specificity = Fig. 2. Radial plot of the average normalised response of all 13 WOLF sensors to
TN + FP (6)
CRC (full line) and IBS (dashed line) urine samples.

distinction between these two disease groups. There is generally a

3. Results and discussion
The results of the initial experiments with aqueous solutions of
individual volatile compounds revealed various relationships be-
tween sensor responses and concentration. However, the presence
of each individual compound evoked response from a different set
of sensors within the array, confirming a degree of selectivity to be
gained from using this combination. Fig. 1 shows the total voltage
response signal generated by the ethylene oxide sensor ISB in-
tegrated over the entire sample time, to toluene solution at a range
of concentrations. The figure illustrates an approximately linear
relationship between sensor response and volatile concentration
within samples. This initial testing allowed for individual tuning of
sensor gains before introduction of actual urine samples, in order
to maximise the potential sensor response.
The normalised change in output voltage of the WOLF sensors
to IBS and CRC samples (with all responses averaged) is shown in
Fig. 2. This highlights both the overall similarities in response to all
urine samples and the more subtle variations used to find
positive change in current produced from these sensors, which is
expected considering an increase in analytes can only cause a
larger ionic current within the cells. These responses are at least
partially produced by an increase in humidity from the headspace
of the sample (which approximately changes from 6.8% to around
14%). However, individual sensor responses vary between different
samples independently of humidity levels, proving the presence of
gases and volatile groups being detected.
Fig. 3 shows a full classification of all three sample groups from
which a degree of class distinction can be seen. However, there is
also a large amount of inter-class overlap and intra-class spreading
in this diagram. The spread shown in IBS is in agreement with its
pathological circumstances as a widely-variable collection of
symptoms rather than a well-characterised disease state. The
healthy control group also shows a larger variation in response,
again supporting the results of previous investigations where this
feature is particularly highlighted as compared to any disease
group (Covington et al., 2013). A potential area for further study

Please cite this article as: Westenbrink, E., et al., Biosensors and Bioelectronics (2014), http://dx.doi.org/10.1016/j.bios.2014.10.044i
 E. Westenbrink et al. / Biosensors and Bioelectronics ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Fig. 3. LDA classification separating all three sample groups of CRC, IBS and healthy
controls (Ctl).
Fig. 4. Box plot of LDA classification separating IBS and CRC sample groups.
would be to investigate the three common subsets of IBS: diar-
rhoea-based (IBS-D), constipation-based (IBS-C), and alternating
(IBS-A) to evaluate if differences could be observed here.
The LDA classification using only CRC and IBS groups is shown
in Fig. 4. This is the classification that most accurately represents
the problem to be solved when patients enter a primary health-
care centre complaining of symptoms. Reasonable separation be-
tween the two groups is demonstrated as well as good clustering
of similarly-classed samples, particularly in the case of CRC.
The basis for measuring the diagnostic merit of this method
was re-classification of individual samples that were taken out of
the main “training set” one-by-one and re-introduced as un-
knowns. The system attempted to successfully choose the class for
each unknown sample using an (n 1) K-nearest-neighbour algo-
rithm for estimating from which it likely belonged. Employing this
technique, a sensitivity and specificity from this classification were
found to be 78% and 79%, respectively. These compare favourably
to other non-invasive screening techniques such as faecal occult
blood and FIT, with success measures that are similar or higher
than what has been discovered in diagnostic review studies (De
Meij et al., 2014).
Please cite this article as: Westenbrink, E., et al., Biosensors and Bioelectronics (2014), http://dx.doi.org/10.1016/j.bios.2014.10.044i
5
This initial training set of samples has established an effective
set of LDA discriminant functions for distinguishing between CRC
and IBS within this pilot cohort. These can now be used in a larger
prospective study that presents a larger cohort of samples as un-
knowns, to be classified prior to confirmation by colonoscopy. The
cross-sensitivity of the chosen (and in fact any) gas sensors to a
large range of volatiles limits the effectiveness of this system to
link any specific species to any disease state. However, the mul-
tivariate statistical method employed does not allow this limita-
tion to affect its ability to distinguish between the groups in the
cohort. A degree of individual variation was found in the profile of
responses for urine samples in the same disease group has been
found in this study, which correlates with findings in other similar
investigations (Arasaradnam et al., 2014b, in press). The above
factors will make this system much more useful as an initial triage
for detecting likely candidates for colon cancer before moving on
to a definitive diagnosis and staging by colonoscopy, rather than
competing with the current ‘gold standard′ techniques directly.
4. Conclusions
A cohort of 92 total urine samples have been run through a
newly-built electronic nose, the ‘WOLF′, based on state-of-the-art
sensors and control hardware being driven by custom LabVIEW
software interface. This new system was successfully tested to
prove the sensitivity of the sensors to common volatile organic
groups. LDA plots produced from the WOLF urine data show a
reasonable separation of all three groups included in this study,
although there is a degree of inter-group overlapping due to a
significant number of outliers. However, particularly well-clus-
tered groups are shown when distinguishing colorectal cancer
from irritable bowel syndrome. Re-classification of single un-
known samples into this latter case produced 78% sensitivity and
79% specificity for detecting CRC, which accurately represents a
situation requiring this new diagnostic tool within the clinical
setting. These results are comparable to those produced clinically
from current techniques, with a potential increase in sensitivity
using a test that only requires 10 min to complete and is practical
within a primary care environment.
These results have highlighted the need for further work, in-
cluding a comparison of different subsets of IBS samples to see if
better-defined groups can be obtained. Another interesting future
study would be to follow-up on patients whose samples were at
the barrier between groups, to see if any disease progression can
be marked using this method. Regardless of the need for future
work, this investigation has strengthened the case for the potential
of electronic nose technologies to be used as diagnostic tools to
detect colorectal cancer from both another common disease and
healthy controls.
Acknowledgements
The authors would like to acknowledge WPH, the Eveson Trust,
and the Bardhan Research Education Trust who provided funding
for this research.
Appendix A. Supplementary Information
Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.bios.2014.10.044.
6 E. Westenbrink et al. / Biosensors and Bioelectronics ∎ (∎∎∎∎) ∎∎∎–∎∎∎
References
Altomare, D.F., Di Lena, M., Porcelli, F., Trizio, L., Travaglio, E., Tutino, M., Dragonieri,
S., Memeo, V., de Gennaro, G., 2013. Exhaled volatile organic compounds
identify patients with colorectal cancer. Br. J. Surg. 100 (1), 144–150.
Arasaradnam, R.P., Bardhan, K.D., 2010. Bioactive Foods and Extracts – Cancer
Treatment and Prevention. Taylor Francis, New York.
Arasaradnam, R.P., Covington, J.A., Harmston, C., Nwokolo, C.U., 2014a. Review ar-
ticle: next generation diagnostic modalities in gastroenterology – gas phase
volatile compound biomarker detection. Alim. Pharmacol. Ther. 39, 780–789.
Arasaradnam, R.P., McFarlane, M., Ryan-Fisher, C., Westenbrink, E.W., Hodges, P.,
Thomas, M.G., Chambers, S., Bailey, C., O’Connell, N., Harmston, C., Nwokolo, C.,
Bardhan, K.D., Covington, J.A. Detection of colorectal cancer (CRC) by urinary
volatile organic compound analysis. 2014b, Plos One (in press).
Arasaradnam, R.P., Nwokolo, C.U., Bardhan, K.D., Covington, J.A., 2011. Electronic
nose versus canine nose: clash of the titans. Gut 60, 1768.
Arasaradnam, R.P., Ouaret, N., Thomas, M.G., Gold, P., Quraishi, M.N., Nwokolo, C.U.,
Bardhan, K.D., Covington, J.A., 2012. Evaluation of gut bacterial populations
using an electronic e-nose and field asymmetric ion mobility spectrometry:
further insights into fermentonomics. J. Med. Eng. Technol., 36 pp. 333–337.
Asimakopoulos, A.D., Del Fabbro, D., Miano, R., Santonico, M., Capuano, R., Pen-
nazza, G., D'Amico, A., Finazzi-Agro, E., 2014. Prostate cancer diagnosis through
electronic nose in the urine headspace setting: a pilot study. Prostate Cancer
Prostatic Dis. 17, 206–211.
Buszewski, B., Kesy, M., Ligor, T., Amann, A., 2007. Human exhaled air analytics:
biomarkers of disease. Biomed. Chromatogr. 21, 533–566.
Covington, J.A., Westenbrink, E.W., Ouaret, N., Harbord, R., Bailey, C., O'Connell, N.,
Cullis, J., Williams, N., Nwokolo, C.U., Bardhan, K.D., Arasaradnam, R.P., 2013.
Application of a novel tool for diagnosing bile acid diarrhoea. Sensors, 13; , pp.
11899–11912.
De Meij, T.G., Ben Larbi, I., van der Schee, M.P., Lentferink, Y.E., Paff, T., Terhaar Sive
Droste, J.S., Mulder, C.J., van Bodegraven, A.A., de Boer, N.K., 2014. Electronic
nose can discriminate colorectal carcinoma and advanced adenomas by fecal
volatile biomarker analysis: proof of principle study. Int. J. Cancer 134,
1132–1138.
Ferlay, J., Parkin, D.M., Steliarova-Foucher, E., 2010. Estimates of cancer incidence
and mortality in Europe in 2008. Eur. J. Cancer 46, 765–781.
Please cite this article as: Westenbrink, E., et al., Biosensors and Bioelectronics (2014), http://dx.doi.org/10.1016/j.bios.2014.10.044i
Garner, C.E., Smith, S., de Lacy Costello, B., White, P., Spencer, R., Probert, C.S.,
Ratcliffe, N.M., 2007. Volatile organic compounds from feces and their potential
for diagnosis of gastrointestinal disease. FASEB J. 21, 1675–1688.
Lippi, G., Cervellin, G., 2012. Canine olfactory detection of cancer versus laboratory
testing: myth or opportunity? Clin. Chem. Lab. Med. 50, 435–439.
Machado, R.F., Laskowski, D., Deffenderfer, O., Burch, T., Zheng, S., Mazzone, P.J.,
Mekhail, T., Jennings, C., Stoller, J.K., Pyle, J., Duncan, J., Dweik, R.A., Erzurum, S.
C., 2005. Detection of lung cancer by sensor array analyses of exhaled breath.
Am. J. Respir. Crit. Care Med. 171, 1286–1291.
McEntegart, C.M., Penrose, W.R., Strathmann, S., Stetter, J.R., 2000. Detection and
discrimination of coliform bacteria with gas sensor arrays. Sens. Actuators
B-Chem. 70, 170–176.
Peng, G., Hakim, M., Broza, Y.Y., Billan, S., Abdah-Bortnyak, R., Kuten, A., Tisch, U.,
Haick, H., 2010. Detection of lung, breast, colorectal, and prostate cancers from
exhaled breath using a single array of nanosensors. Br. J. Cancer 103, 542–551.
Persaud, K., Dodd, G.H., 1982. Analysis of discrimination mechanisms of the
mammalian olfactory system using a model nose. Nature 299, 352–355.
Probert, C.S., Ahmed, I., Khalid, T., Johnson, E., Smith, S., Ratcliffe, N., 2009. Volatile
organic compunds as diagnostic biomarkers in gastronintestinal and liver dis-
ease. J Gastroninest. Liver Dis. 18, 337–343.
Quintero, E., Castells, A., Bujanda, L., et al., 2012. Colonoscopy versus fecal im-
munochemical testing in coloretal-cancer screening. N. Engl. J. Med. 366,
697–706.
Salas, D., Vanaclocha, M., Ibanez, J., Molina-Barcelo, A., et al., 2014. Participation and
detection rates by age and sex for colonoscopy versus fecal immunochemical
testing in colorectal cancer screening. Cancer Causes Control 25, 985–997.
Siegel, R., Naishadham, D., Jemal, A., 2012. Cancer statistics. CA Cancer J. Clin. 62,
10–29.
Silva, C.L., Passos, M., Câmara, J.S., 2011. Investigation of urinary volatile organic
metabolites as potential cancer biomarkers by solid-phase microextraction in
combination with gas chromatography-mass spectrometry. Br. J. Cancer 105,
1894–1904.
Sonoda, H., Kohnoe, S., Yamazato, T., Satoh, Y., Morizono, G., Shikata, K., Morita, M.,
Watanabe, A., Morita, M., Kakeji, Y., Inoue, F., Maehara, Y., 2011. Colorectal
cancer screening with odour material by canine scent detection. Gut 60,
814–819.
Williams, H., Pembroke, A., 1989. Sniffer dogs in the melanoma clinic? Lancet 1,
734.