Design of a Breath Analysis Device for Self-monitoring and Remote
Health-care
D. Germanese
1
, M. D’Acunto,
1,2
and O. Salvetti
1
1
CNR-ISTI, Institute of Information Science and Technology, National Research Council, via Moruzzi 1, Pisa, Italy
2
CNR-ISM, Institute of Structure of Matter, National Research Council, via Fosso del Cavaliere 100, Roma, Italy
Keywords: Self-monitoring, Breath Analysis, Home-care, Gas Sensors, Portable Device, Signal Processing.
Abstract: Technique as new as promising, breath analysis enables the monitoring of biochemical processes in human
body in a non-invasive way. This is why it is drawing, more and more, the attention of scientific
community: many studies have been addressed in order to find a correlation between breath volatile organic
compounds (VOCs) and several diseases. Despite its potential, breath analysis is still far from being used in
clinical practice. These are some of the principal reasons: (i)high costs for the standard analytical
instrumentation; (ii)need of specialized personnel for the interpretation of the results; (iii)lack of
standardized procedures to collect breath samples. Our aim is to develop a device, which we call Wize
Sniffer (WS), based on commercial gas sensors, which is: (i)able to analyse breath gases in real time;
(ii)portable; (iii)low-cost; (iv)easy-to-use also for non-specialized personnel. Another aim is to foster
homecare, that means promote the purchase and the use, also in home environment, of such device. The
Wize Sniffer is composed of three modules: signal measurement, signal conditioning and signal processing.
To satisfy the goal of developing a device by using low-cost technology, its core is composed of an array of
commercial, low cost, semiconductor-based gas sensors, and a widely employed open source controller: an
Arduino board. To promote the use of such device also in home environment, and foster its daily use, it is
programmed in order to send breath test results also to a remote pc: the pc of user’s physician, for example.
In addition, the design of the Wize Sniffer is based on a modular configuration, thus enabling to change the
type of the gas sensors according to the breath molecules to be detected. In this case, we focus our attention
to the prevention of cardio-metabolic risk, for which the healthcare systems are registering an exponential
growth of social costs, by monitoring those dangerous habits for cardio-metabolic risk itself.
1 INTRODUCTION
Nowadays, the gold standard methods for gas
analysis are techniques such as Gas Chromatography
(GC), or Proton Transfer Reaction- Mass
Spectrometry (PTR-MS), and they can be used to
analyze also breath gases with high accuracy and
sensitivity (D. Guo et al., 2010). On the other hand,
these methods are very expensive; moreover, the
analysis, very time consuming, can be performed
only by specialized personnel.
More recent approaches exploit e-noses to
analyze the breath. E-noses, based on gas sensors,
may lose in sensitivity and accuracy with respect
GC/PTR-MS, but they are able to detect in real time
some specific molecules. The most commonly used
sensors for e-noses are solid state gas sensors. Their
working principle is based on a reversible interaction
of the gas with the surface of a solid state material,
which results in a physical eect, depending on the
sensing material, used to achieve the detection of
gases in solid state gas sensors. For example, optical
processes employ infra-red absorption of gases,
whilst chemical processes allow detecting the gas by
means of a selective chemical reaction with a
reagent. The detection of this reaction can be
performed by measuring the conductivity change of
gas-sensing material, or the change of capacitance,
work function, mass, optical characteristics or
reaction energy released by the gas/solid interaction.
E-noses are designed for broader applications
(environmental, industrial ones), rather than for
medical eld. Nowadays, the e-noses that are used
for breath analysis exploit very expensive approach
to detect breath compounds, and they are suitable for
one (or, at most, two) molecule only; we can
mention Bedfont’s PiCO+Smokerlyzer (able to
Germanese, D., D’Acunto, M. and Salvetti, O.
Design of a Breath Analysis Device for Self-monitoring and Remote Health-care.
In Doctoral Consortium (DCBIOSTEC 2016), pages 9-14
9
detect the exhaled Carbon Monoxide) and NOBreath
(able to detect the exhaled Nitric Oxide),
(www.bedfont.com/shop/smokerlyzer,
www.bedfont.com/shop/nobreath), Toshiba’s
research prototype Breathalyzer (able to detect the
exhaled Acetone)
(www.toshiba.co.jp/about/press/2014_03/pr1801.ht
m).
In this work we present the first prototype of the
Wize Sniffer (WS), a portable device based on
chemical semiconductor-based gas sensor array
which represents a low-cost effort to analyze
exhaled breath in real time. In particular, the WS is
able to monitor in real time a specific number of
breath molecules related to noxious habits for
cardio-metabolic risk and oxidative stress. Not only:
an Arduino board is programmed to read sensors’
output and send breath analysis results to a remote
personal computer, which might be the one’s own,
or the physician’s one. In addition, the modular
configuration of the WS enables to change the gas
sensors to detect other types of breath molecules
thus personalizing the device. The use of a low-cost
technology, the compactness of the device and the
possibility to send the results also to a remote
personal computer, allow for a user’s daily
screening, also in home environment.
2 BREATH COMPOUNDS
DETECTED BY THE WIZE
SNIFFER AND CLINICAL
IMPLICATIONS
Breath is composed of oxygen, carbon dioxide,
water vapor, nitric oxide, and a large number of
volatile organic compounds (VOCs) which origin
can be endogenous (that means, they originate from
metabolic processes that occurs in human body and
participate to alveolar exchanges) or exogenous (that
means, they derive from food, or beverages, or
dermal adsorption) (W. Miekisch et al., 2004).
As a consequence, we can affirm that each breath
contains fundamental information about the internal
state of a person. Indeed, more than 35 of the VOCs
present in our breath have been assessed as
biomarkers for particular diseases or metabolic
disorders: for example, increased level of ammonia
in breath may be related to renal diseases (D. Guo et
al., 2010); ethane and pentane derive from lipid per-
oxygenation in case of oxidative stress (M. Phillips
et al., 2003; F. Pabst et al., 2007).
We focus our attention on a set of breath
molecules, some of which related to those noxious
habits for cardio-metabolic risk, such as smoking
and alcohol intake. The molecules detected by the
Wize Sniffer are listed here:
Carbon monoxide (CO): it is naturally
produced by the action of heme oxygenase on
the heme for haemoglobin breakdown. This
produces carboxyhemoglobin, which is more
stable than oxyhemoglobin. Indeed, an
increase of CO leads haemoglobin to carry
less oxygen through the vessels. CO is
present in cigarette smoke, very dangerous
for cardio-metabolic risk. Its baseline value in
a healthy subject is round about 3.5ppm (up
to 14-30ppm in smokers);
Hydrogen (H
2
): it derives from the
breakdown of the carbohydrates in the
intestine and in the oral cavity by anaerobic
bacteria. Its baseline value is round about
9.1ppm;
Ammonia (NH
3
): an increase of NH
3
in
blood may be caused by cigarette smoke,
renal failure, cardiac failure, changes in
cardio-circulatory system. Its baseline value
is round about 0.42ppm;
Ethanol (C
2
H
6
O): exhaled ethanol can be
classied as endogenous or exogenous.
Exogenous Ethanol comes from alcoholic
drink. It is recognized that ethanol breakdown
leads to an accumulation of free radicals into
the cells, a clear example of oxidative stress.
Ethanol may cause arrhythmias and depresses
the contractility of cardiac muscle. Its
baseline value is round about 0.62ppm;
Carbon dioxide (CO
2
) and Oxygen (O
2
):
Their variations show how much O
2
is
retained in the body, and how much CO
2
is
produced as a by-product of cellular
metabolism. In most forms of lung diseases
and some of congenital heart disease
(cyanotic lesions-bluish-grey discoloration of
the skin, lack of O
2
in the body), a decrease of
CO
2
exhaled rate is commonly observed. It
must be noted that the breathing rate
inuences the level of CO
2
in the blood: slow
breathing rates cause Respiratory Acidosis
(i.e., increase of blood CO
2
partial pressure,
which may stimulate hypertension or heart
rate acceleration). On the contrary, too rapid
breathing rate leads to hyperventilation,
which may provoke Respiratory Alkalosis
(i.e., decrease of blood CO
2
partial pressure,
no longer ts its role of vasodilator, leading
to possible arrhythmia or heart trouble). Their
DCBIOSTEC 2016 - Doctoral Consortium on Biomedical Engineering Systems and Technologies
10
baseline values are round about 40000ppm
for CO
2
and 13-15% for O
2
;
Hydrogen Sulde (H
2
S): it is a vascular
relaxant agent, and has a therapeutic eect in
various cardiovascular diseases (myocardial
injury, hypertension). In general, H
2
S could
have therapeutic eect against oxidative
stress due to its capability to neutralize the
action of free radicals. Its baseline value is
round about 0.33ppm.
3 WIZE SNIFFER’S HARDWARE
AND SOFTWARE
ARCHITECTURE
A block scheme of WS is shown in Figure 1. Basing
on this scheme, we can describe the framework of
WS as composed of three modules: signal
measurement, signal conditioning and signal
acquisition.
Figure 1: Schematic sketch of the WS’s architecture. The
core is an acquiring device which includes a gas sampling
box (of 600ml according to the tidal volume (D. Shier et
al., 2007) and made up of ABS and Delrin) where six gas
sensors are placed, and a micro-controller board. Since the
sensors’ output is affected by the water vapour present in
exhaled gases, a HME lter is placed at the beginning of a
corrugated tube reducing the humidity from initial 90% to
60-70%. In addition, the humidity percentage is monitored
within the sampling box, as well as the temperature. Other
two gas sensors having shorter response time work in
owing-regime by means of a sampling pump, which
works at 120 ml/s. A flow-meter monitors the exhaled
breath volume. A flushing pump purges the chamber and
recovery the sensors’ steady state between two
consecutive measures.
The measurement circuit (store chamber + gas
sensor array) detects the breath molecules and
transforms gas signals into electronic signals, which
are processed by the microcontroller board. The
microcontroller used in our system is a low cost,
widely employed open source controller: Arduino
Mega 2560 with Ethernet module. Table 1
summarizes all the VOCs detected by the WS, and
the gas sensors used. Most of the gas sensors are
manufactured by Figaro Engineering, and they are
not expensive at all. They are based on a metal-
oxide semiconductor sensing element, that is, a
variation of sensing element’s internal resistance
occurs when it detect gas particles.
Table 1: VOCs to be detected and gas sensors used.
Breath molecule Sensor and its sensitivity (ppm)
Carbon
monoxide
TGS2442 (50-1000ppm),
TGS2620 (50-5000ppm)
Ethanol
TGS2602 (1-10ppm) and
TGS2620 (50-5000ppm)
Carbon dioxide TGS4161, 0-40000
Oxygen MOX20, 0-16%
Hydrogen sulfide TGS2602, 1-10
Ammonia
TGS2444 (10-100ppm) and
TGS2602 (1-10ppm)
Hydrogen
TGS821 (10-5000ppm), TGS2602
(1-10ppm) and TGS2620 (50-
5000ppm)
As briefly described in Section 4, in order to receive
breath test results from the WS even on a remote
Personal Computer (for example, the physician’s
one), a client-server architecture is implemented. It
means, the Arduino Mega2560 processes sensors’
raw data, executes a daemon on port 23, waits a
command line from the PC and provides the data. A
measure is considered valid if the user’s exhaled
volume equals at least 600ml (store chamber’s
volume, see Figure 1). How the WS microcontroller
board analyzes row data is described in the next
section. In gure 2 the nal conguration of the
Wize Snier is shown.
Figure 2: Final conguration of the rst prototype of the
Wize Sniffer.
Design of a Breath Analysis Device for Self-monitoring and Remote Health-care
11
4 WS FUNCTIONALITY TESTS
AND DATA PROCESSING
If, on one hand, developing a device by using low-
cost technology means facilitating its purchase and
use, on the other hand, especially in this case, this
may represent a challenge. A challenge in terms of
data processing, because, on one side,
semiconductor-based gas sensors are, of course, very
low cost, very easy to be integrated and very
sensitive; on the other side, they are not selective
and they are affected by cross-sensitivity (in Table 1
we can note that the sensors TGS2602 and TGS2620
detect more than one molecule). Their non-
selectivity and their cross-sensitivity have
implications on data analysis, making it very
complicated, especially if a quantitative analysis of
the detected molecules should be done. Moreover,
the behaviour of such type of gas sensors is not
linear.
For our purposes, we aim to make a quantitative
analysis of the detected molecules. It means we aim
to calculate the concentration (in ppm) of the
detected breath molecules. Such approach requires:
an accurate reconstruction of the sensors’
sensitivity curves under our measurement
conditions (32°C+/-10% and 70%RH+/-10%,
it means, conditions that are very close to the
breath physiological ones);
a model to describe how each input (it means,
each molecule) influences the output (it
means, the variation in sensors’ internal
resistance). The simplest model may be
represented by a linear regression;
an accurate evaluation of sensors’ cross
sensitivity;
an accurate evaluation on how temperature
and humidity affect sensors’ outputs;
a model to calculate the concentration of each
molecule.
Meanwhile, since this approach is taking
considerable time, we are exploiting a more
traditional approach for data analysis based on
features extraction (by means of Principal
Component Analysis) and classification (by means
of K-Nearest Neighbour algorithm). In any case, a
signal pre-processing is needed to compensate drifts
and sample-to-sample variations. This more
traditional approach has been implemented first
using a synthetic database of 114 subjects containing
individuals of dierent age (in the range 30-60 years
old), habits (moderate/heavy smokers, non-smokers,
teetotal, moderate/heavy drinkers), lifestyles
(sedentary/ sporty, etc.), and body type. This
database was implemented as if all the gas sensors
worked correctly, in ideal conditions. Plotting the
results of the analysis of the synthetic database by
the PCA (Figure 3), several clusters, highlighted by
means of an algorithm based on KNN (Figure 4) can
be identied.
Figure 3: Biplot of the synthetic database in the rst two
principal components. Several clusters can be
distinguished.
Figure 4: Score plot of the synthetic database analysed by
Principal Component Analysis; LsHd = Light Smokers,
heavy drinkers; LsVHd = Light Smokers, very heavy
drinkers; HsHd= Heavy Smokers, heavy drinkers; HsVHd
= Heavy Smokers, very heavy drinkers. For “Healthy
class” we intend subjects with low cardio-metabolic risk.
Then, a measuring protocol has been draft in order
to test the Wize Snier on a population of 26 healthy
individuals, with dierent age (range 30-60 years
old), habits, lifestyles, body type. Why just healthy?
Because in this case we are focusing on cardio-
metabolic risk, not on a disease. By detecting
molecules related to those noxious habits for cardio-
metabolic risk, the Wize Sniffer should not make a
diagnosis, but should only help the user to monitor
his/her well-being and lifestyle. Indeed, as shown in
DCBIOSTEC 2016 - Doctoral Consortium on Biomedical Engineering Systems and Technologies
12
Figure 4, the classifier classifies the subjects
according to their habits. For a single-subject
monitoring, further statistical analysis over long time
periods should be carried out to evaluate if the
subject increases/decreases his/her cardio-metabolic
risk basing on a change of class.
For the measuring protocol, the methodological
issues about breath sampling procedure have been
taken into account (W. Miekisch et al., 2008). In
practice are used three methods of sampling:
“alveolar (end-tidal) sampling”, if only systemic
volatile biomarkers are to be assessed, “mixed
expiratory air sampling” (which corresponds to a
whole breath sample), “time-controlled sampling”
(which corresponds to a part of exhaled air sampled
after the start of expiration; this method shows large
variations of samples compositions because of wide
variations of individual breathing manoeuvers). For
our purposes, mixed expiratory air sampling method
was chosen, since our interest was focused on both
endogenous and exogenous biomarkers. In addition,
since the composition of single breaths may vary
considerably from each other, because of dierent
modes and depth of breathing, in order to average
breath-by-breath fluctuations in composition due to
the irreproducible lung emptying and flow variations
(F. Di Francesco et al., 2008), we preferred a
sampling of multiple (three) quite breaths.
The KNN classier was able to correctly classify
20/26 subjects, as shown in Table 2. While an
alcohol consumption up to 1-2 Alcohol unit/ day is
often considered not dangerous (in healthy subjects),
smoking is considered very noxious in any case.
We can note that the classifier seems to be not able
to recognize smoker subjects. Actually this may be
due to Carbon Monoxide sensor which has a
minimum LOD of 50ppm (very high), thus resulting
not able to detect Carbon Monoxide even in heavy
smoker subjects. Indeed, we are testing the
performances of another Carbon Monoxide
semiconductor-based gas sensor, MQ-7, which
should be more sensitive and it has a lower LOD
(round about 20ppm, but we are evaluating its
response also to lower concentration, for example 2-
5ppm). In addition, its cost is lower than the one of
CO Figaro gas sensor.
5 CONCLUSIONS
Here we described the first prototype of a low-cost,
portable device for the self-monitoring of those
noxious habits for cardio-metabolic risk by detecting
in the breath molecules principally related to
Table 2: Outcome of the KNN classifier used for classify
WS data. In the I column, the subject’s ID; in the II
column subject’s habits are reported; in the III column the
outcome of the KNN classifier; in the IV column the
right/wrong classification is commented.
AU= Alcohol unit; CIG.= cigarette; D= day; W= week
Subj
ID
Habits Classification Comment
215
1 AU/D
No risk ok
218
1 AU/twice a W
No risk ok
211
12-13 CIG / D; 1
AU / 4-5 times a W
No risk
heavy
smoker
201
1 AU /once a W
No risk ok
207
1 AU/1-2times a W
No risk ok
213
teetotal, non smoker
No risk ok
208
1 AU / 3 times a W
No risk ok
221
teetotal, non smoker
Heavy Drinker No risk
220
1 AU /2 times a W Heavy Drinker
No risk
214
1 AU < once a W
No risk ok
206
1 AU/5-6times a W
No risk ok
205
1 AU 3 times a W
No risk ok
223
2 CIG /D; 1 AU /
twice a W
No risk ok
212
1 AU / D
No risk ok
216
1 AU / once a W
No risk ok
217
teetotal, non smoker
No risk ok
203
1 AU < once a W
No risk ok
209
1 AU < once a W
No risk ok
202
1 AU/5-6times a W
No risk ok
210
1-2 AU / D
No risk ok
225
4 CIG /D; 1 AU <
once a W
No risk
Light
smoker
224
1 AU < twice a W
No risk ok
204
1-2 CIG /D; 1 AU <
2 times a W
Heavy Drinker
Light
smoker
219 1 AU/2 times a W No risk ok
226 1 AU /once a W No risk ok
222
10-15 CIG /
D; 1
AU / once a
W
No risk
heavy
smoker
smoking habit and alcohol intake. Anyway, its
modular conguration and its ease of use allow
changing the type of gas sensors according to the
breath molecule to be detected, and the disorder to
be monitored. The low cost and compactness of the
device allow for a daily screening that, even if
without a real diagnostic meaning, could represent a
pre-monitoring, useful for an optimal selection of
more sophisticated and standard medical analysis.
Further studies will be addressed in order to
improve the performances of the WS, and to
investigate the criticalities of such type of device
and of breath analysis in general. In particular:
the metabolic pathway of the breath molecules
to be detected will be studied in depth;
Design of a Breath Analysis Device for Self-monitoring and Remote Health-care
13
breath sampling methods will be further
investigated. A standardized procedure for
breath sampling may be very useful, since the
composition of each breath is largely
influenced by many factors, such as lung
volume (Jones J.G., 1967), posture
(Anthonisen N.R. et al., 1970), flow rate
(Jones J.G. and Clarke S.W., 1969), ambient
air (F. Di Francesco et al., 2008);
as described in Subsection 5.3, a model to
calculate the concentration of the detected
molecules, based on the non-linear behaviour
of semiconductor gas sensors, will be
implemented; then, the quantitative analysis
performed by the Wize Sniffer may be
compared to the one performed by the gold
standard (Mass Spectrometry, for example);
the number of functionality tests will be
increased, and more experimental results will
be provided;
the possibility to develop ad-hoc gas sensors
based on semiconductor polymers as sensing
element will be investigated (Ding B. et al.,
2009).
ACKNOWLEDGEMENTS
This work was funded in the framework of the
Collaborative European Project SEMEOTICONS
(SEMEiotic Oriented Technology for Individuals
CardiOmetabolic risk self-assessmeNt and Self-
monitoring), grant N. 611516.
Massimo Magrini, Paolo Paradisi, Marco Righi,
and COSMED s.r.l. are warmly acknowledged for
useful support.
REFERENCES
Anthonisen N. R., Robertson P. C. and Ross W. R., 1970.
Gravity-dependent sequential emptying of lung
regions. In J. Appl. Physiol., 589-95.
Di Francesco F., Fuoco R., Trivella M., Ceccarini A.,
2005. Breath Analysis: trends in techniques and
clinical applications. In Microchemical Journal,
vol.79, no 1-2, pp.405-410.
Di Francesco F., Loccioni C., Fioravanti M., Russo A.,
Pioggia G., Ferro M., Roehrer I., Tabucchi S., Onor
M., 2008. Implementation of Fowler’s method for
end-tidal air sampling. In Journal of Breath Research.
Ding B., Wang M., Yu J., Sun G., 2009. Gas sensors based
on electrospun nanobers. In Sensors 2009.
Guo D., Zhang D., Li N., Zhang L., Yang J., 2010. A
novel breath analysis system based on electronic
olfaction. In IEEE Transaction on Biomedical
Engineering.
Jones J.G.., 1967. The effect of pre-inspiratory lung
volumes on the result of the single breath O
2
test. In
Respiratory Physiology, 375-85.
Jones J.G., Clarke S.W., 1969. Effect of expiratory flow
rate on regional lung emptying. In Clin. Sci. 343-56.
Miekisch W., Kischkel S., Sawacki A., Lieban T., Mieth
M., Schubert J.K., 2008. Impact of sampling
procedures on the result of breath analysis. In Journal
of Breath Research.
Miekisch W., Schubert J.K., Noeldge-Schomburg G.F.E.,
2004. Diagnostic potential of breath analysis- focus on
volatile organic compounds. In Clinica Chimica Acta,
347, pp.25-39.
Pabst F., Miekisch W., Fuchs P., Kischkel S., Schubert
J.K., 2005. Monitoring of oxidative and metabolic
stress during cardiac surgery by means of breath
biomarkers: an observational study. In Journal of
Cardiothoracic Surgery, 2:37.
Phillips M., Cataneo R. N., Greenberg J., Grodman R.,
Salasar M., 2003. Breath Markers in oxidative stress in
patients with unstable angina. In Heart Disease, vol.5.
Risby T.H. and Solga S.F., 2006. Current status of clinical
breath analysis. In Applied Physics B 85, pp.421-426.
Shier D., Butler J., Lewis R., 2007. Hole’s Human
Anatomy & Physiology . 11
th
Ed., McGraw-Hill.
DCBIOSTEC 2016 - Doctoral Consortium on Biomedical Engineering Systems and Technologies
14