Using Embedded Sensors in Smartphones to Monitor and Detect
Early Symptoms of Exercise-induced Asthma
Chinazunwa Uwaoma and Gunjan Mansingh
Department of Computing, The University of the West Indies, Kingston, Jamaica
Keywords: Mobile Computing, Sensor Fusion, Health Monitoring, Smartphone, Asthma, Context Recognition.
Abstract: This paper describes work in progress on integrated design architecture for monitoring and detecting early
symptoms of asthma attack using smartphone as sensors’ platform for data capturing, processing,
presentation and feedback. We present an application scenario of exercise-induced asthma where a patient
wears a smartphone equipped with built-in sensors which are capable of providing clinical data and context
on detection of any anomaly in the monitored vital signs. Our design architecture extends the functionality
of “Nine-degree of Freedom” (9-DoF) sensor fusion model and context recognition using expert system
frameworks. The design centers on the idea of creating a simple and portable asthma monitoring system that
is able to detect asthma vital signs, perform signal analysis and context generation; and also send
information to other mobile devices worn by caregivers and physicians. This approach removes the need to
have external monitoring sensors patched on the user’s body, thereby enhancing the usability and reliability
of the system in providing timely information on the state of a patient’s health.
1 INTRODUCTION
Asthma is of one the long-term respiratory
conditions that require real-time and continuous
monitoring, as an attack could occur anytime and
anywhere. The ailment has considerable social and
economic impact on the life of an individual sufferer
and the society (Braman, 2006). The growing
concern to reduce the cost of healthcare utilization
due to asthma, the burden of personal management
of the disease and the workload both on the patients
and healthcare providers, has necessitated the
development of asthma healthcare monitoring and
delivery systems.
Combination of mobile computing technologies
and wireless body sensor networks presents
opportunities for unobstructive monitoring of
patient’s clinical data and seamless communication
of the monitored data to healthcare professionals.
Recent studies have focused on utilizing the
potentials of intelligent mobile devices and wireless
body sensors for data acquisition, signal processing
and context recognition in health applications (Seto
et al., 2009; Zhou and Zhang 2011). However, many
heterogeneous wireless body sensors used for health
monitoring do not always provide precise
information on the sensed signals. Oberoi (2011)
argues that these sensors come with different
Operating Systems as well as proprietary network
and middleware protocols, which makes it difficult
to have a single sensing system that can
comprehensively monitor and detect events of
interest. Furthermore, the inability of patients to
choose and setup different sensors and monitors; and
also, difficulty in analyzing the circumstances
surrounding the sensed data may result in delayed
communication and uncertainty of data presented to
healthcare professionals. A design platform that will
allow integration of multiple sensor modalities and
implementation of context recognition frameworks
on a Smartphone could help overcome these issues.
As mobile computing evolves, mobile phones are
being turned into high performance computer
systems which are handy, discreet and pocket fit
devices (Fernandes and Afonso, 2011). Modern
smartphones come readily equipped with advanced
sensors for detection and recognition of events,
internet access, improved processing and memory
units as well as flexible communication system and
connectivity. Weghorn (2013) notes that these
intelligent mobile devices also run open software
system (e.g. Android) with better User Interface
features compared to stand-alone sensors and
monitors which use ‘closed’ system, specialized
145
Uwaoma C. and Mansingh G..
Using Embedded Sensors in Smartphones to Monitor and Detect Early Symptoms of Exercise-induced Asthma.
DOI: 10.5220/0004806901450150
In Proceedings of the 3rd International Conference on Sensor Networks (SENSORNETS-2014), pages 145-150
ISBN: 978-989-758-001-7
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
hardware and also, bulky in size. The enhanced
capabilities provided in smartphones allow for
development and use of third-party software such as
asthma monitoring applications. Premised on the
above considerations, our design objective centers
on integrated system architecture that allows the use
of personal mobile devices to monitor, detect and
alert early signs of asthma attack.
2 RELATED WORK
Research has focused on the use of advance
information and communications technologies to
improve overall asthma management and control
ranging from electronic peak-flowmetry and asthma
diaries through asthma web-based tools to mobile
phone applications (Seto et al., 2009). Seto et al.
(2009) discuss several approaches that have been
used in the design of asthma e-health systems. These
solution methods however, are not quite sufficient to
provide comprehensive monitoring of asthma patient
without the inclusion of lung sound monitoring.
A new paradigm in asthma management and
control is monitoring wheeze rate which involves
signal analysis of asthma breath sounds. Mobile
phones are capable of recording breath sounds and
performing analysis on the recorded signal
(Anderson et al., 2001; Zhou and Zhang
2011).Wheeze is one the frequent symptoms
experienced by patients and doctors have described
it as a clinical predictor of asthma exacerbation .
Wheeze detection and analysis provides medical
doctors with critical information on how to adjust
treatment of patients with asthma condition. There
are several computer algorithms which can detect
and evaluate percentage of wheeze on normal
breath. Wisniewski and Zielinski (2010) provide
detailed review on these algorithms. Nonetheless,
acoustic signal analysis is only one of many tasks in
preventing asthma attack. There is need to measure
and analyze other vital signs including
environmental effects as well as patient’s level of
activity, in order to provide accurate and reliable
information in controlling asthma exacerbation.
Leveraging the potentials of the internal sensors,
Smartphones can be configured to provide reliable
and timely medical intervention for patients dealing
with asthma crises. Smartphones are able to
correlate patterns in the user’s movement using
embedded sensors like GPS, gyroscope and
accelerometer. They can also recognize sudden
changes in bodily position and decide on the severity
of patient’s condition by evaluating sensors’
measurements (Kiss, 2011). The embedded
microphone can record patient’s breath sound, while
ambient data such as air pressure, humidity and
temperature can be captured by barometer,
hygrometer and thermometer in the smartphone, to
provided context on detected events.
2.1 Monitoring Exercise-Induced
Asthma
Exercised-Induced Asthma (EIA) is triggered by
exercise and physical exertion. Asthmatics with
chronic conditions manifest signs of asthma attack
during exercise. However, there are many people
without asthma who develop symptoms only during
rigorous exercises like sporting activities. An
important characteristic feature of asthma condition
is the hypersensitivity of the airways to
environmental factors. McFadden and Gilbert (1994)
observe that persons with EIA have airways that are
very sensitive to changes in temperature, humidity,
and altitude. Common symptoms of EIA are wheeze
and shortness of breath. Besides wheeze and
difficulty in breathing, another vital sign of EIA
attack is change in patients’ posture. Persons
experiencing asthma attack tend to lean forward in
an effort to get sufficient air into the lungs, as shown
in the presentation (Signs of a Pending Asthma
Attack n.d).
There are preventive measures against EIA
attack such as inhaled medications and Peak–
Expiratory Flow (PEF) measurements (Casa et al.,
2012). However, asthma patients at times find it
tasking to adhere to prescribed medications and
action plans which results in poor management and
control of the condition. Hence the need for a
personal assistive monitoring tool to alert patients
and their care givers on signs of pending asthma
attack as the patients engage in rigorous physical
activities. Figure 1 depicts how a smartphone could
be worn to monitor asthma vital signs during
exercise.
Figure 1: Exercising with Smartphone strapped on the
neck to monitor asthma vital signs (Source: Google
Image).
SENSORNETS2014-InternationalConferenceonSensorNetworks
146
2.2 Sensor Fusion and Context
Recognition
A new development in sensor applications is multi-
dimensional sensing which enhances the user’s
experience through sensor fusion. Sensor fusion is a
technique that uses special filtering algorithms to
compensate limitations of individual sensors in order
to generate a desired output.
Integrating inputs from
multiple sensors using sensor fusion allows for more
accurate and reliable sensing, which consequently
produces much higher levels of recognition for
appropriate response. Pattern-matching algorithms
can identify and correlate events that will assist with
delivering awareness to the user or the monitoring
system. Through training and experience, the
monitoring system adapts to changing circumstances
and responses based on the generated context (Lee et
al., 2008).
The importance of context awareness has
continued to gain wider recognition in healthcare
monitoring applications. By combining sensor
fusion and context recognition techniques, we can
develop smart systems that can help manage patients
with chronic conditions such as asthma; and alert the
patient or the healthcare providers of any anomaly.
3 THE EXTENDED 9-DOF
SENSOR FUSION MODEL
Our design methodology builds on the platform of
sensor fusion solution that combines triaxial-
accelerometer, triaxial-gyroscope and triaxial-
magnetometer.
These 3D-sensors have the capability of
performing basic sensing (Ristic, 2012). An
accelerometer measures linear motion, a gyroscope
defines orientation while the magnetometer provides
direction sensing. Their capabilities notwithstanding,
these sensors are characterized by certain limitations
that affect accuracy in applications. However, sensor
fusion can be used to overcome the shortcomings of
individual sensors. It eliminates deficiencies in
separate devices by using intelligent algorithms and
unique filtering techniques to produce synthesized
and more sophisticated output; thus satisfying the
impression that “The whole is greater than the sum
of its parts” (Karimi, 2013).
An important feature of 9-DoF sensor fusion
solution is its flexibility to accommodate several
sensors. More sensors can be added on need basis
and this automatically transforms the sensor fusion
solution to an m-DoF solution where ‘m’ implies
multiple’ (see Table 1).
The design of the proposed asthma monitoring
system includes activity and motion sensors,
location sensor, ambient temperature and humidity
sensors, air pressure sensor, and an embedded
MEMS microphone for recording and monitoring
asthma wheeze rate. The design is based on m-DoF
solution (see Figure 2). The model allows two data
paths to be used in processing the sensed data
(Ristic,
2012).
1. Pass Through data path - sends raw data
directly to the application.
2. Sensor Fusion path - allows initial raw data to
be processed and synthesized into a smart
output.
In our design, the monitored wheeze signal will
require a separate application for analysis and need
not to go through the sensor fusion path. Also data
from ambient sensors may have to be sent ‘raw’ for
analysis. This scenario, thus, makes the model
suitable for the proposed design.
Table 1: m-DoF Sensor Fusion Models.
m-DoF Sensors Mobile Platforms
6-DoF
3D-gyroscope, 3D-
compass
Android, Linux
and Windows
9-DoF
3D-gyroscope, 3D-
compass, 3D-
accelerometer
Android, Linux
and Windows
12-DoF
3D- gyroscope, 3D-
compass, 3D-
accelerometer,
Barometer,
Thermometer, ALS
Windows and
Android
4 INTEGRATED DESIGN
ARCHITECTURE
The integrated system architecture consists of three
main layers: Data Acquisition Layer, Data
Processing Layer, and Data Analysis Layer (see
Figure 3). Implementing m-DoF sensor fusion
algorithms at the Data Acquisition Layer allows
multi-modal data collection. Data Processing
components enable extraction and representation of
important features of the data while Data Analysis
components provide analysis and description of
context data sets.
UsingEmbeddedSensorsinSmartphonestoMonitorandDetectEarlySymptomsofExercise-inducedAsthma
147
Figure 2: An extended 9-DoF sensor fusion model for asthma monitoring system using embedded sensors in Smartphones
(Ristic, 2012).
Figure 3: Layered architecture of the asthma monitoring
system.
5 DESCRIPTION OF THE
ARCHITECTURAL
COMPONENTS
In this section, we explain the functionality of the
integrated system architecture.
5.1 Data Collection and Signal
Processing
Embedded sensors in the Smartphone collect signals
and quantities of interest - physiological, activity,
ambient and location data. Sensed data can be
distorted due to noise, interference, or other
environmental factors; hence the need for signal
processing to obtain meaningful data. Readers and
Filters algorithms are data processing components
that can interface with the sensors and perform
signal cleansing on the recorded data respectively.
However, these components may not be used for the
fused data since we are applying the m-DoF sensor
fusion solution that performs the two tasks on the
firmware at the Runtime (Ristic, 2012).
5.2 Data Fusion
The described Sensor Fusion model in section 3
provides flexibility for certain aspects of the sensed
signals to be combined to extract desired features
while other data “pass through” as calibrated direct
input to the application processor as shown in Figure
2. With exercise-induced asthma in view, candidate
parameters for fusion are linear motion, gravity,
orientation and direction; derivable from
accelerometer, gyroscope and compass sensors.
Combining these features gives better description of
patient’s level of activity in terms of motion and
position of the body. To obtain these key features,
the accelerometer compensates for gravity loss in
gyroscope which provides the orientation estimates,
while the compass corrects the heading error.
5.3 Wheeze Detection
Wheeze detection systems are designed based on the
idea and principles of stethoscope. Previous designs
involve the use of external sensitive microphones
placed on the chest to record lung sound or on the
neck to record tracheal sound (Sen and Kahya 2006).
However, our design uses an embedded MEMS
microphone located at the mouthpiece of the
Smartphone.
Sovijarvi et al. (as cited in Wisniewski &
Zielinski 2010) note that asthma wheeze signal has a
dominant frequency usually above 100 Hz and with
duration of about 80ms to 100ms. The breath sounds
is pre-processed by applying High pass filters to
remove noise or other low-frequency signals that
may manifest similar morphologies. Characteristic
features used for determining the presence of
Output 2
Sensor 1
Sensor 2
-
-
-
Sensor m
Output1
Output 1
Output 2
Alerts
Diagnosis
Feedback
Alarm
Context/
Knowledge
Generation
Raw Data
Pre-processed Data
Fused Data
Pass
Through
Sensor
Fusion
USER INTERFACE
SENSORNETS2014-InternationalConferenceonSensorNetworks
148
wheeze include the tone, location, duration and
number of wheeze in a breath cycle.
Frequency domain algorithms are considered to
be more efficient for wheeze detection and
classification. However time-domain analysis using
classification features like Kurtosis and Renyi
entropy will be faster for real-time applications; and
also, computationally effective for low-power
mobile devices like smartphones.
5.3.1 Kurtosis
Kurtosis (1) shows the degree of peakedness of a
probability distribution for a random variable X and
it is defined as:
k =

(1)
Kurtosis value k around or larger than 3 indicates
normal distribution for non-wheeze signals while a
value less than 3 shows that the breath sound
contains wheezes and thus would have a uniform or
sub-gaussian distribution (Aydore et al., 2009).
5.3.2 Renyi Entropy
This feature, also known as generalized Shannon
entropy is a measure of uncertainty of signals and
uniformity of distributions (2). For random variable
X, Renyi entropy is defined as:


log 

)
(2)
Study in (Aydore et al., 2009) shows that wheeze
signal distribution has higher degree of uniformity
when compared to non-wheeze signal.
By applying filter algorithms, the signal features
are extracted into the temporary facts/ context
database for further classification.
5.4 Context Generation, Analysis and
Classification
It is important to select algorithms that will allow the
system to generate intelligent information on the
reported event or sensed data. Thus, we consider
classification method based on decision rules
obtained from medical experts and literature, as has
been used in previous studies (Aydore et al., 2009;
Basilakis et al., 2010; Borgohain and Sanyal, 2012;
Bae et al., 2013). Rule-based classifiers label records
or facts using a collection of “condition-action” rules
defined as follows;
Rule: (condition) l, where ‘condition’ is a
conjunction of attributes and l is the class label.
Processed data from different sensor modes can be
grouped into four categories namely: Physiological
data, Activity data, Ambient data and Location data.
For the purpose of analysis however, we use only
two categories as summarized in Table 2. In the
exemplary classification and analysis of the sensed
data, we illustrate how sensor fusion technique and
rule-based expert system can be combined to handle
multi-parameter sensing and context recognition.
Table 2: Classification of Processed Sensor Data.
Category A: Physiological Data
Sensor Measurement
Duratio
n
Kurtosis
(k)
Class
Label
MEMS
Micro-
phone
Breath Sound
Signal
80ms < 3 Wheeze
< 80ms 3
Non-
wheeze
Category B: Activity Data
Sensor
(Fused)
Measure-
ment
Motion Orientatio
n
Class
Label
Accelero-
meter,
Gyroscope
and
Compass
Activity
Level
Fast Upright Very
active
Slow Inclined Active
None Inclined Non-
active
The generated context is annotated using JSON-
based description. The annotations can be displayed
directly to the Smartphone console or stored
temporarily in the context database for subsequent
analysis. Context or situational analysis and pattern
recognition can be performed by using Rete
Algorithm (Figure 4) to combine decisions and
confidence levels from experts in the application
domain.
Figure 4: Rete Decision Tree - Network for Context
Recognition.
Knowledge inferred from context analysis
provides classes of services: alerts, diagnosis,
feedback and alarm. The information can be stored
as patient’s health history on the SD card or shared
with healthcare providers for further diagnosis and
treatment. Alarm and Alert messages are sent in the
event of emergency or detection of any anomaly on
the monitored signals.
UsingEmbeddedSensorsinSmartphonestoMonitorandDetectEarlySymptomsofExercise-inducedAsthma
149
6 CONCLUDING REMARKS AND
FUTURE WORK
The emerging IC MEMS sensory technology in
Smartphones defines a new path in healthcare
monitoring. Benefits of using mobile phone as
multi-parameter sensing device include the ability of
an asthma patient to carry an accurate all-in-one
monitor anywhere; and the ability to make baseline
measurements at anytime thereby generating a
database that could allow for improved detection of
disease state and control.
Our emphasis has been on sensor fusion and
context modelling given that data fusion and context
awareness are critical for optimal performance of
any health monitoring system. Whereas data fusion
provides accurate value, context recognition
provides knowledge on what to do with the data.
The sensor fusion requires substantial MCU power
which may not be fully provided by Smartphones
given the limited power source for these devices.
Using a dedicated sensor processor may be an
efficient way of performing sensor data computing.
Optimal performance and cost however, are major
considerations for independent sensor processor.
The wheeze signal also, needs to be analyzed in a
separate application. We are investigating how the
application can run concurrently with data fusion in
order to have a synchronized output.
REFERENCES
Anderson, K., Qiu, Y., Whittaker, A. R. & Lucas, M.,
2001. Breath sounds, asthma, and the mobile phone.
The Lancet, 35(9290), 1343-1344.
Aydore, S., Sen, I., Kahya, Y. P. & Mihcak, M. K., 2009.
Classification of respiratory signals by linear analysis.
In EMBC 2009, Annual International Conference of
the Engineering in Medicine and Biology Society.
IEEE. 2617-2620.
Bae, W. D., Alkobaisi, S., Narayanappa, S. & Liu, C. C.,
2013. A Real-time Health Monitoring System for
Evaluating Environmental Exposures. Journal of
Software, 8(4), 791-801.
Basilakis, J., Lovell, N. H., Redmond, S. J. & Celler, B.
G., 2010. Design of a decision-support architecture for
management of remotely monitored patients. IEEE
Transactions on Information Technology in
Biomedicine, 14(5), 1216-1226.
Borgohain, R. & Sanyal S., 2012. Rule Base Expert
System for Cerebral Palsy Diagnosis. International
Journal of Advance Networking and Application,
viewed 01 October 2013. http://arxiv.org/ftp/arxiv/
papers/1207/1207.0117.pdf.
Braman, S. S., 2006. The global burden of asthma. CHEST
Journal, 130, suppl. 1, 4-12.
Casa, D. J., et al. 2012. National Athletic Trainers'
Association position statement: preventing sudden
death in sports. Journal of Athletic Training, 47(1), 96.
Fernandes, B., Afonso, J.A. & Simões, R., 2011. Vital
signs monitoring and management using mobile
devices. In 6th Iberian Conference on Information
Systems and Technologies (CISTI). IEEE. 1-6.
Karimi, K., 2013. Sensor Fusion and Internet of Things
(IoT). The Embedded Beat, 14 May, viewed 01
October 2013. https://community.freescale.com/
community/the-embedded-beat/blog/2013/05/14/
sensor-fusion-and-the-internet-of-things-iot.
Kiss, G., 2011. Using smartphones in healthcare and to
save lives. In Internet of Things (iThings/CPSCom),
International Conference on Cyber, Physical and
Social Computing. IEEE. 614-619.
Lee, H., Park, K., Lee, B., Choi, J. & Elmasri, R., 2008.
Issues in data fusion for healthcare monitoring. In
Proceedings of the 1st international conference on
Pervasive Technologies Related to Assistive
Environments. ACM. 3.
McFadden Jr, E. R. & Gilbert, I. A., 1994. Exercise-
induced asthma. New England Journal of Medicine,
330(19), 1362-1367.
Oberoi, S., 2011. Sensor fusion brings situational
awareness to health device. Embedded, 01 June,
viewed 01 October 2013. http://www.embedded.com/
design/embedded/4216526/2/Sensor-fusion-brings-
situational-awareness-to-health-devices.
Ristic, L. J., 2012. Sensor Fusion and MEMS for 10-DoF
Solutions. EE Times, 03 September, viewed 01
0ctober 2013. http://www.eetimes.com/
document.asp?doc_id=1279870.
Sen, I. & Kahya, Y. P., 2006. A multi-channel device for
respiratory sound data acquisition and transient
detection. In EMBS 2005, 27th Annual International
Conference of the Engineering in Medicine and
Biology Society. IEEE. 6658-6661.
Seto, E. Y. W. et al. 2009. A wireless body sensor network
for the prevention and management of asthma. In
SIES'09, International Symposium on Industrial
Embedded Systems. IEEE. 120-123.
Signs of a Pending Asthma Attack n.d. WebMD, viewed
01 October 2013. http://www.webmd.com/asthma/
asthma-symptoms-7/slideshow-asthma-attack.
Weghorn, H., 2013. Applying mobile phone technology
for making health and rehabilitation monitoring more
affordable. In 2013 ISSNIP Biosignals and
Biorobotics Conference (BRC), IEEE. 1-5.
Wisniewski, M. & Zielinski, T., 2010. Digital analysis
methods of wheezes in asthma. In International
Conference on Signals and Electronic Systems
(ICSES). IEEE. 69-72.
Zhou, S., Zhang, Z. & Gu, J., 2011. Time-domain ECG
signal analysis based on smart-phone. In EMBC 2011,
Annual International Conference of the Engineering in
Medicine and Biology Society. IEEE. 2582-2585.
SENSORNETS2014-InternationalConferenceonSensorNetworks
150