MobileECG: An Ubiquitous Heart Health Guardian

Jos

e Maria Monteiro

, Jo

ao Paulo do Vale Madeiro

, Angelo Brayner

and Narciso Arruda

Department of Computing Science, Federal University of Cear

a, Fortaleza, Cear

a, 60440-900, Brazil

Institute for Engineering and Sustainable Development,

University for the International Integration of the Afro-Brazilian Lusophony, Redenc¸

ao, Cear

a, 62790-970, Brazil

Keywords: ECG Signal Processing, Pervasive and Mobile Platform, ECG Feature Extraction, Prediagnosis, Data

Integration.

Abstract:

Electrocardiogram (ECG) is a widespread and efﬁcient medical procedure for monitoring heart health. ECG is

a fast, low-cost and non-invasive examination. Its output allows anomaly analysis by health experts. Despite

its application in clinical environments, ECG acquisition and analysis as a daily routine is far from being a

reality for a large part of the world’s population. In this context, we present here a mobile and pervasive

platform, named MobileECG, which provides ECG signal acquisition, automatic feature extraction and real-

time prediagnosis. Furthermore, MobileECG implements the ubiquitous computing features. Hence, it runs

on mobile devices (smartphone or tablet), assuring this way anytime and anywhere access for anyone to its

functionalities. MobileECG is in fact an ubiquitous Heart Health Guardian. Besides, MobileECG supports

ECG data integration and publication using Linked Data technology, providing a public knowledge base,

which may be used to support complex queries, run mining algorithms and yield collaboration among experts.

1 INTRODUCTION

Electrocardiography is a technique used to record

electrical potential oscillations produced by cardiac

activity. Temporal evolution of such oscillations is

called the electrocardiogram (ECG) signal, which is

the most widespread test in cardiology for the diag-

nosis of cardiac diseases and anomalies. The human

body by itself acts as a conductor of electrical cur-

rent, and any two points on the surface can be con-

nected by electrodes to record an ECG or monitor

the heart’s rhythm. The time-series obtained by the

electrocardiographic record contains a series of wave-

forms and complexes, which are called the P-wave,

the QRS complex, and the T-wave, which are sepa-

rated by regular intervals. The analysis of ECG signal

behavior (waveform pattern, duration and interval) al-

lows the extraction of a variety of information, which

can subsidize the identiﬁcation of a great variety of

heart diseases.

Chronic non-communicable diseases (NCD), in-

cluding heart disease and cancer, are responsible for

approximately 70% of all deaths worldwide, accord-

ing to the World Health Organization (Daar et al.,

2007). Allied to this fact, the world population is ag-

ing. Such a context requires a signiﬁcant need to mon-

itor patient’s health status, while she/he is in her/his

personal environment. Consequently, a plethora of

wearable systems for health monitoring has been de-

veloped and applied for providing real-time feedback

information on individual’s health condition, either to

the user himself, to a medical center, or directly to a

health professional.

Over the last decades, ubiquitous computing has

gained critical importance for implementing health

care applications. Ubiquity provides the means to

make health care services available for anyone at any-

time and anywhere. Thus, patient data may be ac-

quired by biosensors containing embedded micropro-

cessors, coupled up with bluetooth modules. Further-

more, a mobile application may execute signal pro-

cessing or data analytic techniques on collected data,

or even simply forward collected data to a physician’s

or a clinic’s computers.

In this paper, an innovative pervasive and mobile

platform for ECG signal acquisition, processing and

prediagnostic extraction is presented. The proposed

platform, denoted MobileECG, is able to collect and

preprocess ECG raw data in a mobile device (smart-

phone or tablet). Thereafter, prediagnosis algorithms

are executed in order to trigger alerting messages to

the patient, physician, or emergency station, which

should be sent by means of a WiFi or cellular net-

work.

Monteiro, J., Madeiro, J., Brayner, A. and Arruda, N.

MobileECG: An Ubiquitous Heart Health Guardian.

DOI: 10.5220/0007703705790586

In Proceedings of the 21st International Conference on Enterprise Information Systems (ICEIS 2019), pages 579-586

ISBN: 978-989-758-372-8

579

Besides, MobileECG maintains an integrated

ECG data repository, which stores an integrated view

of ECG data coming from heterogeneous sources

(with different ECG data standards) and data of pa-

tient, treatment and drug data sources. Such a repo-

sitory can be used by machine learning algorithms to

automatize prediagnosis.

Empirical experiments on real ECG datasets have

been conducted to evaluate MobileECG efﬁciency

and effectiveness. The results reveal that MobileECG

can be an important tool for helping health experts to

identify heart diseases more quickly in a more reliable

way.

The paper is structured as follows. Section 2 an-

alyzes approaches similar to the one presented in this

paper. Section 3 details the different steps of the pro-

posed platform. In Section 4, results of empirical ex-

periments are presented to highlight the beneﬁts and

feasibility of the proposed approach. Finally, section

5 concludes the paper and points out future work.

2 RELATED WORK

A literature review on pervasive health care systems

is presented in (Orwat et al., 2008). According to

the authors, 60% of investigated systems provide ana-

lytical and diagnostic functionality. Only 46% im-

plement automatic alerting, and from this, only 14%

support an alerting function generated by a mobile

device. Since pervasive health care systems encom-

pass a wide variety of components, one can classify

them by the following criteria: ECG data acquisi-

tion systems (Venkatachalam et al., 2011), real-time

ECG digital signal processing (Madeiro et al., 2012;

do Vale Madeiro et al., 2017), feature extraction (El-

haj et al., 2016; Martis et al., 2014), ECG remote

monitoring systems (Worringham et al., 2011), col-

laborative databases (Gonc¸alves et al., 2011) and ma-

chine learning techniques (Luz et al., 2016; Elhaj

et al., 2016).

In (da Silva et al., 2015), the authors describe

a biosensor for sensing ECG signals, with minimal

electrical contact points with users, a so-called off-

person ECG. The key goal is to support ECG signal

acquisition on palms or ﬁngers. Thus, the proposed

biosensor can be embedded in any object with which

the user interacts.

In (Wen et al., 2008), the authors propose an ECG

telemonitoring system. According to their approach,

the Holter Monitor will record the ECG signal of the

patient continuously up to 48 h. The monitored data

is transmitted to the server through the Internet when

a wired network is available. The Holter also con-

tains a software program performing real-time ECG

classiﬁcation. When speciﬁc abnormal heartbeats are

detected, the Holter transmits them with the GPS

(global positioning system) information to the server

via MMS (multimedia messaging service) in real-

time. The physician at the server side can communi-

cate with the patient also by using MMS. In the server,

a GIS (geographic information system) is used for lo-

cating the patient in an emergency case by using the

GPS information packaged in the MMS message.

In (Worringham et al., 2011), the authors propose

a system to enable walking-based cardiac rehabili-

tation in which the patient’s single-lead ECG, heart

rate, and GPS-based speed and location are transmit-

ted by a programmed smartphone to a secure server

for real-time monitoring by physicians. According to

the authors, the feasibility of this approach was evalu-

ated in 134 remotely-monitored exercise assessment

and exercise sessions in cardiac patients unable to

undertake hospital-based rehabilitation. Completion

rates, rates of technical problems, detection of ECG

changes, pre- and post-intervention six minute walk

test (6 MWT), cardiac depression and Quality of Life

(QOL) could be measured. Several exercise and post-

exercise ECG changes were detected.

In (Ngo and Veeravalli, 2014), DuyHoa Ngo et

al. have proposed a platform based on Web Semantic

technologies which provides the storage of features

extracted from the ECG signal within a database fol-

lowing Linked Data patterns. The authors emphasize

that the proposed platform is part of a healthcare sys-

tem based on cloud data storage, which also captures

ECG signals and other vital signs through biosensors

located on the surface of the patient’s body.

Thus, one can enumerate two important strengths

of the proposed platform: the reproducibility of the

methodology is notorious due to the fact that the pro-

posed hardware and software are open-source; Mo-

bileECG provides basis for integrating multiple ECG

data sources, including data from biosensors and data

from public or institutional databases which may po-

tentialize the accuraccy of pattern recognition pro-

cesses.

3 THE MobileECG PLATFORM

The MobileECG architecture presents the following

components: acquisition module, mobile application

module, ECG signal processing and feature extraction

module, ECG data extraction module, and ﬁnally, the

data integration and publishing module. Fig. 1 de-

picts an abstract model of the proposed architecture,

emphasizing the physiological data ﬂow, from the pa-

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

580

tient to the ECG data repository. For the sake of clar-

ity, the MobileECG components are grouped by their

behavior in a client-server architecture. Thus, the ac-

quisition and mobile application modules are located

in the client area, whereas the other modules are in the

server area, as web servers. Each one of the aforemen-

tioned components is described in detail and analyzed

next.

Figure 1: Overview of the MobileECG platform.

3.1 Acquisition Module

The acquisition module is composed of a biosensor

coupled with an Arduino microcontroller and it is lo-

cated in the client area. The capture of the ECG

biopotentials is performed by the biosensor, which

measures and ampliﬁes electrical potentials derived

from the electrical activity of the heart, and then

transmits the analog signal to Arduino. An open-

source hardware that allows direct communication

(SHIELD) with an Arduino microcontroller has been

deployed as biosensor. The electrodes used for sig-

nal acquisition consist of three leads which should be

placed on the wrists and right ankle of the patient. The

shield converts the analog differential signal (ECG

biopotentials) into a single stream of data as output.

A third order “Besselworth” analog ﬁlter’s cutoff fre-

quency is set to f

= 40Hz, and the output analog sig-

nal is discretized via a dedicated ADC embedded in

the Arduino board. Default values for discretizing pa-

rameters are: 10-bit ADC with 256 Hz sampling rate.

The Arduino board is responsible for receiving the

analog signal and for converting it into a digital sig-

nal. Each discretized sample from each lead or chan-

nel is split into two bytes, named high byte and low

byte, and each byte is stored in packets. Thereafter,

each individual byte from the packet is serially trans-

mitted to the bluetooth module, which in turn sends

ECG raw data to a mobile device.

3.2 Mobile Device Application Module

The mobile application is in charge of establishing

connections with a paired device (bluetooth module

associated to a given biosensor and Arduino board)

and for searching for unpaired devices. Whenever a

device is selected, the mobile application collects the

bluetooth identiﬁcation of the connected device and

its Mac address.

As soon as the mobile application receives ECG

raw data from a given device, it is able to plot ECG

signal buffers on screen as long as a 5-second buffer

of ECG signal is completely ﬁlled. A 5-second

buffer of ECG signal contains 21,760 bytes. The mo-

bile application triggers a thread, named Connection-

Thread, which receives the Mac address of the paired

device. Within the Connection-Thread, we deﬁne a

number of bytes (or packet length) transferred from

the Arduino bluetooth module to the mobile device

on each transfer. Thus, an Android device continu-

ously receives packets of 289 bytes from the Arduino

bluetooth module. Each received packet is stored into

a byte array (length of 21, 760 bytes). Thereafter, the

mobile application triggers the ECG data buffer de-

coding process. This process searches for each pair

of synchronizing bytes (A5 and 5A) and for the pair

of bytes corresponding to the ECG sample of a given

channel. Since the second channel has been deﬁned as

default, high and low bytes are located, respectively,

ﬁve bytes and six bytes after each pair of synchroniz-

ing bytes. Thus, a given ECG sample is computed as

the sum of the low byte and high byte shifted by 8

bits left. Each computed ECG sample is stored in a

buffer array which will contain a 5-second ECG win-

dow for subsequent display on screen and subsequent

sending to the Web Service. After ﬁlling a 5-second

ECG signal buffer, the mobile application forwards

the pre-processed ECG buffer to the Web Service. Af-

ter sending each ECG buffer to the Web Service, the

class Plot-Real-Time ﬁnally performs the plot of the

buffer.

3.3 ECG Signal Processing and ECG

Feature Extraction Module

As soon as the mobile app ﬁnishes the ECG buffer

transmission, the Web Service (Apache Maven) joins

all received signal buffers and stores them in a sin-

gle ﬁle. A set of algorithms for digital signal pro-

cessing is performed, including signal ﬁltering for

denoising, wavelet transform (time-frequency analy-

sis) for selective enhancement of the QRS complex,

P and T waves, wave peak detection and waveform

delineation. The main goal is to extract the following

MobileECG: An Ubiquitous Heart Health Guardian

581

parameters: P-wave amplitude, P-wave duration, PR

interval, QRS amplitude, QRS duration, T-wave am-

plitude, T-wave duration, QT and ST segments, and

intervals between beats. Those algorithms have been

implemented using Matlab, which is an IDE devel-

opment environment widely applied for digital signal

processing (Islam et al., 2012).

The ﬁrst stage for ECG feature extraction is re-

sponsible for detecting the QRS complex, which is

the most expressive waveform of the ECG signal re-

garding amplitude and period of oscillation. The cor-

rect detection of the QRS complex and its precise de-

lineation are fundamental conditions for efﬁcient de-

tection and segmentation of the other waves. Addi-

tionally, it provides the necessary support for algo-

rithms able to recognize patterns of cardiac arrhyth-

mias. After proceeding with the denoising process, a

sequence of ﬁltering routines are established for en-

hancing the QRS complex and attenuating artifacts

and other physiologic waves. The Wavelet Trans-

form, the First-Derivative function, and the Hilbert

Transform are sequentially computed, and the mod-

ule of the resultant analytic signal is used at a decision

stage. An adaptive threshold algorithm is applied over

the resultant ﬁltered signal for individual detection of

each QRS (Madeiro et al., 2012). Then, the computa-

tion of an indicator related to the area covered by the

QRS complex provides the detection of QRS onset

and offset (Madeiro et al., 2012).

Concerning P-wave and T-wave detection and seg-

mentation, ﬁrstly, we apply Wavelet Transform over

signal windows established between segmented QRS

complexes for estimating T-wave and P-wave peak lo-

cations. Then, we ﬁt the parameters of a synthetic

function, computed as a composition of two Gaussian

functions, in order to model each waveform (T-wave

and P-wave) by evaluating the normalized root mean

square (RMS) error. This approach provides the de-

lineation of each waveform (by detecting onsets and

ends) and the computing of parameters related to the

wave width and to a distortion factor of the Gaussian

functions which also characterize their morphologies

(do Vale Madeiro et al., 2017). All the above ex-

plained algorithms, built as Matlab language scripts,

are compiled and converted in a single executable ﬁle.

After the Web Service receives all the ECG signal

buffers and converts them into a single ﬁle, the Web

Service itself calls a Java method that runs the re-

ferred executable ﬁle. Then, the set of all ECG ex-

tracted parameters (waveform peaks and widths, seg-

ments and intervals) are converted into text ﬁles. Af-

ter obtaining these text ﬁles, the Web Service calls a

Java method responsible for storing the extracted pa-

rameters within a relational database.

As an illustrative example, Fig. 2 presents a set of

results showing ECG beat cycles with all the char-

acteristic waves properly segmented, and two time

series of the extracted parameters intervals between

beats and QRS widths.

Figure 2: Practical example of ECG feature extraction by

applying algorithms over the signal sent by Android app.

3.4 Data Extraction Module

This module consists of storing features extracted

from an ECG signal in a relational database. By do-

ing this, it is possible to formulate SQL queries in

order to relate ECG signals to properties of a given

diagnosis. Figure 3 brings the schema of the pro-

posed database. The ECG table contains an identi-

ﬁcation of a given ECG record (exam) and also some

clinic/diagnosis information from the corresponding

patient, such as his/her systolic and diastolic pres-

sure at the time of ECG acquisition. The table Pa-

tient contains some identiﬁcation data of a given pa-

tient, such as document number, birth date and birth

place. The Prescription table stores information on

current medicines taken by a given patient. Drug is

fed with medicines related to treatments for the most

common cardiac diseases. The table Cycle relates the

PQRST complex to a unique cardiac beat. Tables re-

lated to P-wave, QRS complex and T-wave contain

the identiﬁer of the cycle to which they belong. More-

over, they contain information derived directly from

the process for waveform detection and segmentation:

onsets, offsets and the corresponding amplitudes.

Tables Cycle

iagnosis, ECG

iagnosis and

ECG

ardiopathy provides to associate ECG wave-

form patterns to known cardiac conditions and/or

cardiac diseases, such as: conduction disorders,

arrhythmias, cardiomyopathies, and other events.

The content of such tables may result from Machine

Learning algorithms, or may derive from ECG public

databases, in the case of ECG signals originated from

those databases, or even may derive from manual

specialists’ annotations.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

582

3.5 Data Integration and Publishing

Advances in heart disease research have been ham-

pered in part by the fragmented gathering and stor-

ing of data. Among the major ECG standards, we

can highlight: (i) AHA/MIT-BIH (Physionet) (Gold-

berger et al., 2000a), an ECG data format extensively

used in cardiac physiology research; (ii) SCP-ECG

(Mandellos et al., 2010), which is an ECG European

standard that speciﬁes a data format and a transmis-

sion protocol for ECG records; and (iii) HL7 aECG

(Bond et al., 2011), which is an ECG data American

standard adopted by the Food and Drug Administra-

tion (FDA), from the USA, for clinical research.

This knowledge needs to be integrated. To make

matters worse, typically heart disease knowledge is

expressed in ambiguous terminology. Without the aid

of a well-deﬁned knowledge representation, the ad

hoc data integration is hard work. In addition, another

fundamental challenge to heart research is querying

complex heart data.

In this context, we propose the designing of an

integrated repository of ECG data whose purpose

is: (i) to integrate data in heterogeneous ECG stan-

dards, that is, to store ECG data coming from vari-

ous sources with different data standards (AHA/MIT-

BIH, SCP-ECG and HL7 aECG); (ii) to integrate data

about cardiac diseases, cardiomyopathies, treatments

and drugs with LOD

(Linked Open Data); (iii) to

store ECG data coming from Patient’s smartphones;

(iv) To make possible the querying of complex heart

data; and (v) to enable running data mining algo-

rithms over integrated data.

Therefore, to support the design of the integrated

ECG repository, we propose the use of Semantic

Web technologies, such as: RDF, OWL, SKOS and

SPARQL. These technologies enable the explicit re-

presentation of knowledge and its processing to de-

duce new information. Besides, they provide an envi-

ronment where heterogeneous data can be combined

based on a common knowledge representation, the

schemata can be extended in an easily and dynamic

way, and applications can query that integrated data

and draw inferences using vocabularies.

In (Gonc¸alves et al., 2011), the authors proposed

an ECG reference ontology, that is, an ontology re-

sulting from an application-independent representa-

tion of the ECG domain. This reference ontology can

be used to support the design of interoperable versions

of ECG data formats like AHA/MIT-BIH, SCP-ECG,

and HL7 aECG.

This ECG reference ontology can be used to sup-

port the redesign and the possible uniﬁcation of ex-

https://lod-cloud.net/

isting ECG data standards. In this sense, it can be

used to support the design of an interoperable and

collaborative ECG data repository, which can store

ECG signals from various data sources with differ-

ent ECG data standards. However, in order to build a

wide ECG repository that makes it possible to answer

complex queries, it is necessary to extend the refer-

ence ontology proposed in (Gonc¸alves et al., 2011),

adding concepts related to drugs, treatments, and dis-

ease causes, among others, which is one of the goals

of the MobileECG platform. In this way, we use the

vocabularies ecg

, available in the work (Gonc¸alves

et al., 2007), and heal th

(additional terms).

The ecgo vocabulary, contains terms created by

the proposed approach. Fig. 4 presents the ontology

applied in this step. As we can observe, it contains

proposed classes for providing inference and classi-

fying of cardiac rhythms (e.g. ecgo : SlowRhythm,

ecgo : NormalRhythm, ecgo : FastRhythm), types of

cardiac beats/cycles (e.g. ecgo : NormalBeat and

ecgo : JunctionalEscapeBeat) and possible ECG di-

agnosis (e.g. ecgo : PersistentAtrialFibrillation and

ecgo : LongQT Syndrome).

The mappings between the relational database

schema and the extended ontology have been exe-

cuted by means of the R2RML language and the

D2RQ tool (Priyatna et al., 2014). It is important to

emphasize that it is not necessary to use an ontology

to perform mappings with the D2RQ tool; however its

use greatly facilitates the mapping process.

The MobileECG platform allows you to import

ECG signals from any repository using one of the

following standards: AHA/MIT-BIH, SCP-ECG and

HL7 aECG. To support this import process, we have

implemented a wrapper for each one of these three

formats. Thus, to import a particular ECG signal,

from a public database, it is necessary to follow three

steps: (i) initially, download the ﬁles containing the

signal, from a public database; (ii) then, apply the set

of algorithms described in the Section 3.3 for extract-

ing the features of the signal; (iii) ﬁnally, the data of

the signal must be extracted, similar to the process

described in the Section 3.4.

The data integration and publishing module con-

sists of a Web Service which uses Java methods for

accessing the relational database containing the infor-

mation extracted from the ECG signals and exports

the relational data to the RDF format. We split the

process into two steps. In the ﬁrst step, we create a

dump of the relational data in RDF format. To per-

form this step, we use the D2RQ tool together with the

mappings in the R2ML3 language, which associates

http://nemo.inf.ufes.br/biomedicine/ecg.html

Available at https://health-lifesci.schema.org

MobileECG: An Ubiquitous Heart Health Guardian

583

Pwave

- pid

pid

cyid

eid

- chid

- onsetTime

- peakTime

peakT

- offsetTime

- onsetAmp

onsetAmp

- peakAmp

peakAmp

- offsetAmp

Patient

- paid

paid

name

- CPF

- gender

gender

birthDate

- deathDate

- birthPlace

ECG

- eid

- paid

paid

sinal

- datetime

- BMI

- pacemaker

pacemaker

systolicPressure

- diastolicPressure

- cancer

- tobaccoism

- alcoholism

- syncope

syncope

sedentarism

- flutter

- channelsNumber

- CVA

Pwave

Prescription

- pri

pri

- date

Drug

- d

- description

Cycle

- cyid

cyid

eid

- chid

- peid

peid

- cdid

Cycle

did

QRSWave

- qrsid

qrsid

cyid

eid

- chid

- onsetTime

- peakTime

peakT

- offsetTime

- onsetAmp

onsetAmp

- peakAmp

peakAmp

- offsetAmp

TWave

- tid

- cyid

cyid

eid

- chid

- onset

onset

Time

- peakTime

peakT

offsetTime

- onsetAmp

onsetAmp

- peakAmp

peakAmp

- offsetAmp

PrescriptionDrug

- pi

did

- dosage

ECG_Drug

- ei

- did

- dosage

Diagnosis

- daid

- description

ECG_Diagnosis

- eid

- daid

sis

Cardiopathy

- caid

- description

ECG_Cardiopathy

- ei

- caid

thy

ECG_Cardiopathy

- pai

pai

caid

ECG_Channel

- eid

- chid

chid

Channel

- chid

- description

Period

- peid

peid

startTime

Cycle_Diagnosis

- cd

- description

Institution

- ins

ins

- description

Patient_Institution

- pai

insid

description

ins

Figure 3: The relational database schema.

the relational database schema with the ontology vo-

cabulary adopted to represent the ECG signals. In the

second step, the RDF data present in the dump gen-

erated previously are materialized in an RDF triple-

store, more speciﬁcally in Virtuoso, in a semiauto-

matic way. Virtuoso provides a SPARQL endpoint

that is provided to perform semantic queries.

The RDF Repository is a collaborative database

for heart research. It makes possible the identiﬁcation

of data cohorts (including patients, medicines, treat-

ments, and ECGs) for study, and the export of data

for statistical analysis. Besides, the RDF Repository

will make it possible to infer new knowledge through

the use of machine learning algorithms.

4 EMPIRICAL EVALUATION

In order to assess the potentials of the proposed ap-

proach, empirical evaluation have been carried out.

Firstly, we evaluate the performance of the Arduino

code for ECG signal acquisition, conditioning and for

sending ECG data to the bluetooth module.

In terms of memory consumption, using Olimex

SHIELD-EKG-EMG for analog signal acquisition,

Arduino Uno and IDE Arduino 1.8.5, we obtain that

sketch (code) uses 2272 bytes (7%) of program stor-

age space and global variables use 223 bytes (10%) of

dynamic memory, leaving 1825 bytes for local vari-

ables. Regarding the mobile application, we con-

duct the experiments using smartphone model LG-

M250ds and Android version 7.0. We have measured

the smartphone memory comsumption, immediately

before and immediately after the smartphone appli-

cation starts to run. We can observe a signiﬁcant

increasing for memory consumption while the class

Plot-Real-Time performs the plot of the 5s-ECG sig-

nal buffer.

According to Fig. 5, the total use of memory is

37.99 MB immediately before the ﬁrst ECG signal

buffer is exhibited on the mobile screen, increasing to

56.07 MB (see Fig. 6) while the ECG signal buffer is

being plotted, during 5 seconds, and again decreasing

to 37.99 MB.

As already mentioned, after the mobile app con-

cludes the ECG buffer transmission, the Web Service

joins all the received signal buffers and converts them

in a single ﬁle. We have measured the average time re-

quired for speciﬁc ECG signal processing tasks which

runs in the Web Service, in terms of the total duration

of a given record. Thus, for those experiments, we

have used a 2.70 GHz Intel(R) Core(TM) i7-7500U

CPU, 8 GB of RAM and Windows 10 Home, version

1803, for running Web Service. We have obtained the

results illustrated in Table 1.

Table 1: Performance evaluation (processing time) for ex-

tracting ECG signal features within Web Service (percent-

age of the signal duration).

ECG Signal processing

task

Processing time

ECG signal ﬁltering

for denoising

0.129%

QRS complex detection

and delineation

15.43%

T-wave and P-wave detection

and delineation

26.638%

Querying over ECG signal data is a rather com-

plex task, since the data is stored in unstructured bi-

nary ﬁles. On the other hand, the MobileECG plat-

form enables the execution of semantic queries. To

demonstrate this fact, we present a query which was

executed on an RDF database generated through the

MobileECG approach.

Sparql Query 01: Which male patients over 60

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

584

ecg:Cycle

health:Drug

+ rdfs:label

ecg:Waveform

ecg:Patient

+ foaf:name

+ foaf:gender

gender

+ ecgo:CPF

ecgo

+ schema:birthDate

+ schema:birthPlace

+ schema:deathData

part_of

ecg

Patient

nder

CPF

birthDate

birthPla

ecgo:hasECG

ecgo:NormalBeat

ecg:Pwave

ecgo:ProlongQTSyndrome

ecg:Twave

ecg:Wave

+ ecg:peakTime

+ ecg:onsetAmp

+ ecg:offsetAmp

+ ecg:peakAmp

ecg:Segment

ecg:ElementaryForm

+ ecg:onsetTime

ecg

+ ecg:offsetTime

ecg:PRsegment ecg:STsegment

ecg:QRSwave

ecg

Wave

ecg

peakTime

ecg

onsetAmp

ecg

offsetAmp

ecg

pe p

peakAmp

men

Seg

cle

Cardiopathy

HypertrophicCardiomyopathy

ARVC

ecg

Prescription

+ ecgo:prescriptionDate

prescriptionDate

ecgo

prescriptionDate

+ ecgo:dosage

health:drug

ecgo:PatientECG

+ ecgo:BMI

ecgo

+ ecgo:pacemaker

ecgo

pacemaker

+ ecgo:systolicPressure

ecgo

systolicPressure

+ ecgo:diastolicPressure

ecgo

+ ecgo:channelsNumber

Cardiopat

ecgo:Tobaccoism

ecgo:Alcoholism

ecgo

ntECG

ecgo

hasECG

ecgo:Canal1

ecgo:Canal12

ecgo:LongQTSyndrome

ecgo:SlowRhythm

ecgo:Beat

+ ecgo:beatRate

Bea

ecgo:hasBeat

ecgo:NormalRhythm ecgo:FastRhythm

ecg

ecgo

ecg

avefor

ecgo

Canal1

ecgo:PersistentAtrialFibrillation

Figure 4: Scheme of the ontology applied in the data integration and publishing module.

Figure 5: Memory consumption immediately before the

ﬁrst ECG signal buffer is exhibited on mobile screen.

Figure 6: Memory consumption immediately after the ﬁrst

ECG signal buffer starts to be exhibited on mobile screen.

years old have had accelerated beats (over 100 beats

per minute) within the ECG?

SELECT ?patient

WHERE {?patient a health:Patient ;

health:age ?age ;

health:gerner ?gender ;

ecga:hasECG ?ecg .

?ciclo ecgo:part of ?ecg .

?ciclo ecgo:hasBeat ?bat .

?bat a ecgo:FastBeat.

FILTER(?age >60 && ?gender = "male")}

Figure 7: Sparql Query 01.

Finally, we have computed the number of gener-

ated RDF triples as we increase the number of full-

Figure 8: Amount of generated RDF triples per number of

full-processed ECG records.

processed ECG records, e.g. ECG signals for which

we have published the corresponding data in the RDF

repository. For this experiment, we have considered

applying 30-min ECG records from MIT-BIH Ar-

rhythmia Database (Goldberger et al., 2000b) at the

following scenarios: 1, 5, 10, 20, 30, 40, 46 full-

processed ECG signals. The obtained results are il-

lustrated in Fig. 8, evidencing a linear increase in the

number of RDF triples, and, therefore, in the use of

memory as we increase the number of full-processed

ECG signals (RDF repository).

5 CONCLUSION

In this work, we introduced MobileECG, a complete

cardiac activity monitoring and prediagnosis solu-

tion. The platform comprises everything from the

MobileECG: An Ubiquitous Heart Health Guardian

585

stage of signal acquisition, through biosensors, Ana-

log/Digital conversion through the Arduino, digital

signal transmission executed by a mobile applica-

tion, to the stage of ECG feature and data extrac-

tion, integrating and publishing, through Web Service

technology. Additionally, a collaborative database

is maintained to integrate and store ECG data com-

ing from heterogeneous sources and data on patients,

treatments, and drugs. The database is published us-

ing Linked Data standards. A case study was ana-

lyzed aiming at to demonstrate MobileECG proper-

ties. As future work, anonymizing patient data tech-

niques should be implemented, aiming at the protec-

tion of individual information. Moreover, machine

learning algorithms will be deployed for classiﬁca-

tion/recognition of arrhythmia and other events.

ACKNOWLEDGEMENTS

Sources of funding: The authors acknowledge the

supports of the Brazilian Research Council, CNPq

(Grant n. 426002/2016-4), and Cear

a State Founda-

tion for the Support of Scientiﬁc and Technological

Development (BP3-0139-00284.01.00/18).

REFERENCES

Bond, R. R., Finlay, D. D., Nugent, C. D., and Moore, G.

(2011). A review of ECG storage formats. Int. journal

of medical informatics, 80(10):681–697.

da Silva, H. P., Carreiras, C., Lourenc¸o, A., Fred, A.,

das Neves, R. C., and Ferreira, R. (2015). Off-

the-person electrocardiography: performance assess-

ment and clinical correlation. Health and Technology,

4(4):309–318.

Daar, A. S., Singer, P. A., Persad, D. L., Pramming,

S. K., Matthews, D. R., Beaglehole, R., Bernstein,

A., Borysiewicz, L. K., Colagiuri, S., Ganguly, N.,

et al. (2007). Grand challenges in chronic non-

communicable diseases. Nature, 450(7169):494–496.

do Vale Madeiro, J. P., dos Santos, E. M. B. E., Cortez, P. C.,

da Silva Felix, J. H., and Schlindwein, F. S. (2017).

Evaluating gaussian and rayleigh-based mathematical

models for t and p-waves in ecg. IEEE Latin America

Transactions, 15(5):843–853.

Elhaj, F. A., Salim, N., Harris, A. R., Swee, T. T., and

Ahmed, T. (2016). Arrhythmia recognition and classi-

ﬁcation using combined linear and nonlinear features

of ecg signals. Computer methods and programs in

biomedicine, 127:52–63.

Goldberger, A. L., Amaral, L. A., Glass, L., Hausdorff,

J. M., Ivanov, P. C., Mark, R. G., Mietus, J. E., Moody,

G. B., Peng, C.-K., and Stanley, H. E. (2000a). Phys-

iobank, physiotoolkit, and physionet. Circulation,

101(23):e215–e220.

Goldberger, A. L., Amaral, L. A., Glass, L., Hausdorff,

J. M., Ivanov, P. C., Mark, R. G., Mietus, J. E., Moody,

G. B., Peng, C.-K., and Stanley, H. E. (2000b). Phys-

iobank, physiotoolkit, and physionet. Circulation,

101(23):e215–e220.

Gonc¸alves, B., Guizzardi, G., and Pereira Filho, J. G.

(2007). An electrocardiogram (ECG) domain ontol-

ogy. In Workshop on Ontologies and Metamodels for

Software and Data Engineering, Jo

ao Pessoa, Brazil.

Gonc¸alves, B., Guizzardi, G., and Pereira Filho, J. G.

(2011). Using an ecg reference ontology for seman-

tic interoperability of ecg data. Journal of Biomedical

Informatics, 44(1):126–136.

Islam, M., Tangim, G., Ahammad, T., Khondokar, M., et al.

(2012). Study and analysis of ecg signal using matlab

&labview as effective tools. International journal of

Computer and Electrical engineering, 4(3):404.

Luz, E. J. d. S., Schwartz, W. R., C

amara-Ch

avez, G., and

Menotti, D. (2016). Ecg-based heartbeat classiﬁcation

for arrhythmia detection: A survey. Computer meth-

ods and programs in biomedicine, 127:144–164.

Madeiro, J. P., Cortez, P. C., Marques, J. A., Seisdedos,

C. R., and Sobrinho, C. R. (2012). An innovative ap-

proach of QRS segmentation based on ﬁrst-derivative,

hilbert and wavelet transforms. Medical engineering

& physics, 34(9):1236–1246.

Mandellos, G. J., Koukias, M. N., Styliadis, I. S., and Lym-

beropoulos, D. K. (2010). e-scp-ecg+ protocol: An

expansion on SCP-ECG protocol for health telemon-

itoring—pilot implementation. International journal

of telemedicine and applications, 2010:1.

Martis, R. J., Acharya, U. R., and Adeli, H. (2014). Current

methods in electrocardiogram characterization. Com-

puters in biology and medicine, 48:133–149.

Ngo, D. and Veeravalli, B. (2014). Applied semantic tech-

nologies in ecg interpretation and cardiovascular di-

agnosis. In Bioinformatics and Biomedicine (BIBM),

2014 IEEE Int. Conf. on, pages 17–24. IEEE.

Orwat, C., Graefe, A., and Faulwasser, T. (2008). Towards

pervasive computing in health care – a literature re-

view. BMC Medical Informatics and Decision Mak-

ing, 8(1):26.

Priyatna, F., Corcho, O., and Sequeda, J. (2014). Formal-

isation and experiences of r2rml-based sparql to sql

query translation using morph. In Proc. of the 23rd

Int. Conf. on World wide web, pages 479–490. ACM.

Venkatachalam, K., Herbrandson, J. E., and Asirvatham,

S. J. (2011). Signals and signal processing for the

electrophysiologist. Circulation: Arrhythmia and

Electrophysiology, 4(6):965–973.

Wen, C., Yeh, M.-F., Chang, K.-C., and Lee, R.-G. (2008).

Real-time ecg telemonitoring system design with mo-

bile phone platform. Measurement, 41(4):463–470.

Worringham, C., Rojek, A., and Stewart, I. (2011). Devel-

opment and feasibility of a smartphone, ecg and gps

based system for remotely monitoring exercise in car-

diac rehabilitation. PloS one, 6(2):e14669.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

586