ECGMultilead QT IntervalEstimationUsingSupport...

Research ArticleECG Multilead QT Interval Estimation Using SupportVector Machines

Jhosmary Cuadros 1 Nelson Dugarte2 Sara Wong3 Pablo Vanegas3 Villie Morocho 3

and Ruben Medina 2

1Department of Electronics Engineering Universidad Tecnica Federico Santa Maria Valparaiso Chile2Research Center for Biomedical Engineering and Telemedicine Electrical Engineering DepartmentUniversidad de Los Andes Merida Venezuela3Computer Science Department Engineering School Universidad de Cuenca Cuenca Ecuador

Correspondence should be addressed to Jhosmary Cuadros jhosmarycuadrossansanousmcl

Received 31 October 2018 Revised 27 January 2019 Accepted 12 February 2019 Published 15 April 2019

Academic Editor Jose Joaquin Rieta

Copyright copy 2019 Jhosmary Cuadros et al )is is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

)is work reports a multilead QT interval measurement algorithm for a high-resolution digital electrocardiograph )esoftware enables off-line ECG processing including QRS detection as well as an accurate multilead QT interval detectionalgorithm using support vector machines (SVMs) Two fiducial points (Qini and Tend) are estimated using the SVM algorithmon each incoming beat )is enables segmentation of the current beat for obtaining the P QRS and Twaves )e QT interval isestimated by updating the QT interval on each lead considering shifting techniques with respect to a valid beat template )evalidation of the QT interval measurement algorithm is attained using the Physionet PTB diagnostic ECG database showing apercent error of 260 plusmn 225msec with respect to the database annotations )e usefulness of this software tool is also tested byconsidering the analysis of the ECG signals for a group of 60 patients acquired using our digital electrocardiograph In this casethe validation is performed by comparing the estimated QT interval with respect to the estimation obtained using theCardiosoft software providing a percent error of 249 plusmn 199msec

1 Introduction

Cardiovascular diseases are the leading cause of deathglobally and they are also the first motivation for medicalconsultation )is health problem increases the need formedical service and has an important impact on health costs[1] Among the different cardiovascular diseases the arterialhypertension (HTA) [2] the coronary artery disease (CAD)[3] and the ischemic cardiomyopathy [4] are the mostcommon In particular Chagas heart disease [5] is a parasitedisease caused by the protozoan Trypanosoma cruzi leadingto a progressive cardiac damage such as cardiomyopathy [6])e disease is characterized by heart failure disease ar-rhythmias that may predispose to systemic thromboem-bolism or sudden death and atrioventricular block )emost frequent electrocardiographic abnormalities in pa-tients with Chagas disease are ventricular extrasystole

complete block of right branch of His bundle atrioven-tricular block and ventricular fibrillation [7] Chagas diseaseis also related to other cardiovascular diseases such as ar-terial hypertension and coronary artery disease [8] How-ever the abnormalities vary depending on geographicalregion and status of the disease in each patient )e dis-tribution of Chagas has changed in recent years due topermanent migrations [9]

In Chagas disease one of the alterations that affect theelectrocardiogram (ECG) signal is the abnormal variation ofintervals such as the QT interval and morphology of the Twave expressed by their residuum [10] )e QT intervalrepresents the ventricular depolarization and repolarizationtime interval )is interval is measured from the Q onset tothe T end As the QT interval decreases while the heart rateincreases the QT interval is usually normalized with respectto the heart rate and it is denoted asQTc)e accuracy in the

HindawiJournal of Healthcare EngineeringVolume 2019 Article ID 6371871 14 pageshttpsdoiorg10115520196371871

estimation of the QT interval depends on the accuracy inidentification of Q onset and T end [11] Accurate mea-surement of the QT interval could be difficult for some T-wave morphologies and noisy electrocardiogram signals[12] Several research groups have shown that advancedhigh-resolution ECG (HRECG) processing enables the es-timation of several advanced indexes related to cardiovas-cular diseases )e study of HRECG started in the seventieswith the goal of noninvasive detection of electrical activity ofthe HisndashPurkinje system [13] Later other cardiac zones orECG intervals included micropotentials related with path-ological conduction of the heart [14]

Starc and Schlegel [15] have developed a system thatenables real-time assessment of the QT interval variabilityin the eight independent leads of a standard electrocar-diogram )is system includes hardware that performs theECG signal acquisition at 1000 samples per second in eachlead with a resolution of 12 bits per sample )e system alsoincludes advanced HRECG software for estimating thescore of reduced amplitude zones (RAZ) as well as heartrate variability (HRV) analysis either in time domain or infrequency domain Among the advanced tools for ECGanalysis it includes the T-wave morphological multileadanalysis based on principal component analysis (PCA)Based on this system Schlegel et al [16] have derived anadvanced ECG score for prediction of coronary arterydisease (CAD) which is based on a set of parameters es-timated using the advanced software analysis )e softwaresystem known as Cardiosoft [17] runs on the windowsoperating system

Karjalainen et al [18] developed a method for quanti-fying the QT interval variability )e method is based on atechnique known as principal component regression (PCR))is method allows reducing the dimensionality of data )evariation of the QT interval is proportional to the variationin a parameter representing the third eigenvalue of thecorrelation matrix )e algorithm divides the ECG signalinto segments including the QRS and the T wave and thenthe PCR algorithm is applied)is algorithm allows studyingthe QT interval variability without detecting the T wave

Christov and Simova [19] proposed an automatedmethodthat scored as second in the PhysionetCinC Challenge 2006)e method is based on analyzing the presence of peaks orslopes in two consecutive segments of 10msec )e searchis performed first for estimating the Q onset and then for theT end A revision of several classical methods for QT de-tection is presented by Xue and Reddy [20]

)e analysis of the ECG signal is becoming an im-portant tool for the assessment of several cardiac diseasesIn this context we are proposing a system includinghardware and software for HRECG analysis where supportvector machine (SVM) techniques [21ndash23] are used withinthe framework for QT interval estimation )e proposedsystem includes the signal acquisition electronics hardwareas well as a software system for ECG signal analysis )esoftware system is based on open-source libraries and runson the Linux operating system )e proposed systemperforms multilead QT interval estimation )e proposedmethodology is based on the work reported by Starc and

Schlegel [15] but we have used a different approach forpreprocessing and for segmentation of theQRS template onthe eight independent leads of the 12-lead standard surfaceelectrocardiogram In our algorithm rather than relying onthe QRS detector for determining the beginning of the QRS(Qonset) and a biparabolic curve modeling for de-termination at the end of the T wave (Tend) we performthese tasks using SVM techniques )e SVM learningtechnique is a very useful tool for classifying ECG patterns[24] )is has motivated their application for accurateestimation of this interval )e QT interval estimationmethod is validated using the Physionet PTB diagnosticECG database [25ndash27] as well as a dataset including agroup of 60 patients acquired using our digital electro-cardiograph system )e article is organized as followsfirst a brief introduction and state of the art is given as wellas a description of the theoretical bases and then themethodological framework is described Subsequently theresults are presented and finally conclusions are given

2 Methods

21 DIGICARDIAC ECG Acquisition System )e DIG-ICARDIAC system enables acquisition processing and han-dling of the electrocardiogram as well as the clinical patientinformation )e proposed system incorporates advanced al-gorithms for signal processing and interval estimation as wellas telemedicine tools for handling the patient electronic record)e DIGICARDIAC instrument has two types of operationmodes First of all it can be used as any electrocardiograph in aclinical consultation Secondly it works as an instrument ofdetailed analysis normally used in cardiac clinical research

)e electronic hardware of the DIGICARDIAC system[28] is used for acquiring the standard 12-lead electrocar-diogram )is hardware system was designed to work as awhole connected with a personal computer as shown inFigure 1 )e hardware starts working when it gets pluggedinto the USB port )e high-resolution electrocardiogramsignals are transmitted constantly to the computer )edeveloped software tools running on the personal computerallow the user to initiate the electrocardiogram data re-cording otherwise the transmitted information from thehardware is discarded )e acquisition is performed usingthe standard 12-lead system at 2000 samples per second witha resolution of 12 bits per sample An example of acquisitionis shown in Figure 2 where the V4 recorded lead is shown

211 Acquisition Data )e protocol included three groupsof subjects 20 patients with Chagas disease 20 patients withhypertension and 20 healthy subjects Patients underwent ahigh-resolution 12-lead electrocardiogram (ECG) recordingusing DIGICARDIAC A standard supine 12-lead ECG wasrecorded for 5minutes in each of the 60 patients duringnormal breathing

22 Multilead QT Interval Measurement )e algorithmreported for estimation of the QT interval is multilead in thesense that the eight independent leads of the standard

2 Journal of Healthcare Engineering

electrocardiogram are considered for estimating the QTinterval Particularly the segmentation of theQRS templateas well as the incoming beat involves stages where pa-rameters are extracted from estimations performed on eachof the leads Additionally even when an estimation of theQT interval is provided as a result for each of the leadstheir combination allows one to provide a representativeQT interval estimation for all leads A description of thedifferent processing stages is presented in the followingparagraphs

221 Signal Preprocessing )e signal is preprocessed with amoving averaging filter for attenuating the power line in-terference )e baseline wandering was attenuated using ahigh-pass recursive filter and a FIR filter for attenuating theelectromyographic noise as reported in [19] )e phasecharacteristic is constant and the filter was applied forwardand then backwards for minimizing the phase distortion)e high-pass filter response is given by

y(n) C1(x(n)minus x(nminus 1)) + C2y(nminus 1) (1)

where x(n) is the input signal sequence and y(n) is theoutput filtered sequence and coefficients C1 and C2 arecalculated as follows

C1 1

1 + tan fcπT( 1113857

C2 1minus tan fcπT( 1113857

1 + tan fcπT( 1113857

(2)

where fc is a cutoff frequency of 064Hz and T is samplingperiod )e next step is using the Savitzky and Golay filter[29] for attenuating the electromyographic noise)e filter isbased on performing a least-square approximation of thesignal given by

Y(n) 1

M1113944

N

jminusNCjX(n + j) (3)

where Y(n) and X(n) are the signal after and before theleast-square approximation respectively N is the length ofthe approximation interval at both sides of the sample andCj are the weighting approximation coefficients given by

Cj 3N2

+ 3Nminus 1minus 5j2 (4)

and the normalization coefficient M is given by

M (2N + 1) 4n2 + 4Nminus 3( 1113857

3 (5)

)e procedure is applied for a total interval (2N + 1) of31milliseconds )e next step during the preprocessing isthe QRS detection which is performed according to anadaptation of the Pan and Tompkins algorithm proposed byMneimneh et al [30] which uses a methodology based onthresholding)en a segmentation of the signal is made thistask consists in taking samples present in the ECG signalwhere the QRS complexes are present )us the ECG recordis divided into their component waves (a heartbeat is therepresentation of a cardiac cycle) )e segmented ECGsignals are used for constructing the QRS templates

222 Templates A histogram for the RR interval is con-structed based on the first twenty beats extracted from theECG signal )en an average QRS signal is obtained con-sidering the ten beats with RR intervals closest to themaximum of the histogram )is averaged signal is a QRStemplate )e procedure is repeated for each of the eightindependent leads of the standard electrocardiogram [15] InFigure 3 the template for each independent lead is shown

Patient RefIsolation

ADconvert

USBinterface Computer

Sowarestage

Hardware stage

DigitalizationAnalog

acquisition

Multichanneldif amp

Figure 1 DIGICARDIAC system components

Figure 2 )e DIGICARDIAC user interface window showing agraphical representation of the V4 ECG lead

Journal of Healthcare Engineering 3

100 200 300Time (ms)

400 500 600 7000ndash02

ndash01

0

01

02

03

04

05

06A

mpl

itude

(mv)

(a)

Time (ms)100 200 300 400 500 600 7000

ndash01

ndash005

0

005

01

015

02

025

03

Am

plitu

de (m

v)

(b)

Time (ms)100 200 300 400 500 600 7000

ndash03ndash02

ndash07

ndash05ndash04

ndash010

0102

ndash06

Am

plitu

de (m

v)

(c)

Time (ms)100 200 300 400 500 600 7000

ndash01

0

ndash04

ndash02

01

02

03

ndash03

Am

plitu

de (m

v)

(d)

Time (ms)100 200 300 400 500 600 7000

03

ndash03

ndash01

0504

ndash02

02

001

ndash04

Am

plitu

de (m

v)

(e)

Time (ms)100 200 300 400 500 600 7000

02

ndash03

ndash01

03

04

05

ndash02

01

0

Am

plitu

de (m

v)

(f )

Time (ms)100 200 300 400 500 600 7000

05

02

06

03

04

0

ndash01

01Am

plitu

de (m

v)

(g)

Time (ms)100 200 300 400 500 600 7000

Am

plitu

de (m

v)

05

02

06

03

04

0

ndash01

01

(h)

Figure 3 Templates for each of the 8 independent derivations of the ECG (a) Deviation template I (b) Deviation template II (c) Deviationtemplate V1 (d) Deviation template V2 (e) Deviation template V3 (f ) Deviation template V4 (g) Deviation template V5 (h) Deviationtemplate V6


223 QRS Template Segmentation e ducial point of theQRS is determined using the SVM technique in each leadtemplate (this point is denoted as Qini) e QRS detectionalgorithm provides the point QRS1 corresponding to theQRSwave onset the pointQRS2 corresponding to the R peakof the QRS and the point QRSend corresponding to the endof the QRS A representative pair of values QRS1 and QRS2for all leads is calculated using the procedure proposed byStarc and Schlegel [15] that enables calculation of the widthof theQRS denoted asΔQRS e averageQini is obtained forthe eight independent leads and Qini0 is selected as the valuefor the lead that occurs earliest with respect to the averagee point QRSend0 is selected using a similar procedure butconsidering the average for QRSend points in the eight leadse points Qini0 and QRSend0 represent the initial and endpoints of the QRS wave However as suggested in [15] alarger interval is used by extending 20ms before and 30msafter the end dening the points PQbreak and QTbreak asshown in Figure 4

224 T-Wave Template Segmentation e end of the Twave is estimated for each independent template lead usingthe SVM algorithm is point is denoted as Tend e pointT2 is located 30ms after the Tend point and T1 is located20ms before the maximum of the absolute value of the Twave Location for the T wave on each template is denedbetween points QTbreak and T2 as shown in Figure 5

225 Template QT Interval e QT interval in each tem-plate lead is dened using the estimation for Qini0 and Tende calculation is performed as follows

QTk0 Tkend minusQkini0 (6)

where k is the lead considered

23 Beat-to-Beat QT Interval Estimation After consideringthe rst twenty beats necessary for template calculation andQT interval initialization our algorithm performs the beat-to-beat QT interval calculation Each incoming beat iscompared with respect to the lead template and acceptedwhen the Pearson correlation coecient is larger than apredened threshold e accepted beat is segmented byobtainingQini and Tend using SVM techniques and then thepoints QRS1 QRS2 QRSend T1 T2 PQbreak and QTbreak areobtained using a procedure similar to the template seg-mentation e calculation of the QT interval for the in-coming beat is performed by shifting the incoming signalwith respect to the template to obtain the maximal corre-lation within a particular beat regione amount of shiftingis used for updating the QT interval on each lead using theprocedure proposed by Starc and Schlegel [15] e beat-to-beat QT interval estimation is calculated as follows

QTki QTk0 + ΔQRSi + ΔTki (7)

where i is the incoming beat and k is the lead consideredeupdating of the QT interval is the addition of the amount ofshifting in the QRS template denoted as ΔQRSi and the

T-wave template denoted as ΔTki e amount of shifting iscalculated as explained in Section 233 e system providesas a result the sequence of QT intervals for each of the leadsas well as their average

231 Calculation of the Shifting for the QRS Templatee total amount of shifting for the QRS template is cal-culated considering two stages in the rst stage the shiftingof the segment PQbreak minusQTbreak in the incoming beat withrespect to the similar interval in the QRS template is cal-culated is process is illustrated in Figure 6 During thesecond stage the amount of shifting of the segment

PQbreakQTbreak

QRSend0QRS1QRS2Qini0

QRS templateSignal

ndash02

ndash01

0

01

02

03

04

05

06

07

08

Am

plitu

de (m

v)

100 200 300 400 500 600 7000Time (ms)

Figure 4 Template segmentation for obtaining PQbreak andQTbreak

700600500400Time (ms)

3002001000ndash02

ndash01

0

01

02

03

Am

plitu

de (m

v)

04

05

06

07

08

SignalT templateQTbreak

TendT1

T2

Figure 5 T-wave template segmentation between QTbreak and T2


QRS1 minusQRS2 in the incoming beat with respect to the similarsegment in theQRS template is calculatede alignment forthis segment is shown in Figure 7 is procedure is per-formed for each lead for obtaining the array ΔQRSki wherek denotes the lead and i is the incoming beat From thisarray a representative ΔQRSi is calculated as described forthe QRS template segmentation

232 Calculation of the Shifting for the T-Wave Templatee total amount of shifting for the T-wave template is alsoperformed using two stages in the rst stage the shifting iscalculated for the segment between QTbreak and T2 in theincoming beat with respect to the similar interval in the T-wave template e alignment process is shown in Figure 8During the second stage the shifting is calculated for the Twave considering the segment between T1 and Tend in thesegmented incoming beat with respect to the T-wave tem-plate is alignment process is illustrated in Figure 9 isprocess is performed for each lead for obtaining the array ofshifting for the T wave ΔTki

233 Temporal Shifting between Two Signals e amount ofshifting between a template denoted as x(n) and the in-coming signal denoted as y(n) is calculated using the fol-lowing functional

J(θ) minθsumN

n1(x(n)minusy(n plusmn θ))2 (8)

where θ is the shifting in samples between signals and N isthe length of the segmente incoming signal is normalizedwith respect to the template

234 Qini and Tend Calculation Using SVM Two supportvector machine algorithms were trained for detecting thesepoints e rst SVM is trained for detecting the Qini pointand the second SVM was trained for detecting the Tendpoint e training for the rst SVM is performed using asfeature ECG signal intervals that are centered at the actualQini location (dened by an expert)ese intervals are calledmarkers and other ECG signal intervals selected from otherECG regions are called nonmarkers e second SVM istrained similarly for detecting the Tend point

(1) Training A total of 1800 beats are used for extractingfeatures for training 600 beats are extracted from hyper-tensive patients 600 beats are extracted from chagasic pa-tients and 600 beats are extracted from control subjects Anadditional set of 300 beats (100 from each patient group) isused for validation

e annotation of the actual location of Qini and Tend wasperformed manually on each beat by a group of cardiologists

(2) Features Used for the SVM e SVM training is donebased on selected reference points called markers andnonmarkers For each of the selected points the featuresused are a vector including signal samples located aroundthe selected point as well as signal samples extracted fromthe discrete wavelet decomposition [31] of the beat selected

e choice of the mother wavelet is based on researchstudies about the application of dierent wavelet families forQRS complex and T-wave detection [32 33] ese researchworks recommend the utilization of the wavelet familyDaubechies-4 (level 5) for QRS complex detection andSymlet-7 (level 6) for the T wave because these waveletsshow coecients with greater amplitude and a pattern thatroughly identies the starting and the end point of the QRS

ndash04

ndash02

0

02

04

06

08

1A

mpl

itude

(mv)

1501000 50Time (ms)

PQbreak ndash QTbreak beatPQbreak ndash QTbreak template

(a)

0 15010050Time (ms)

ndash04

ndash02

0

02

04

06

08

1

Am

plitu

de (m

v)


(b)

Figure 6 First stage for the alignment of the QRS template and the incoming beat In (a) the initial alignment between the QRS templateand the incoming beat is shown In (b) both signals are aligned considering the procedure described in Section 233 for the interval betweenPQbreak and QTbreak


complex as well as the start and end point of the Twave Inconsequence this type of wavelets is used in the presentwork

In this case a level 5 reconstruction of the Daubechies-4wavelet is used for the Qini detection A level 6 re-construction using the Symlet-7 is used for the Tend de-tection e level of detail d6 oers relevant informationabout the T wave by characterizing this pattern with twopeaks one positive and one negative including a zerocrossing (Figure 10) Both reconstructed signals are obtained

with the same size as the QRS template In Figure 11 thetime matching between the original QRS signal and itswavelet reconstruction is shown In this case the level 5detail of the Daubechies-4 wavelet decomposition isreconstructed using the inverse discrete wavelet transform(IDWT) where the zero crossings and peaks correlate withthe Q R and S ducial points

(3) Features Vector Each feature vector for markers andnonmarkers points is a vector including 41 samples where

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300

Template QTbreak ndash T2

Beat QTbreak ndash T2

(a)

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(b)

Figure 8 First stage for the alignment of the T-wave template and the incoming beat In (a) the initial alignment between the T-wavetemplate and the incoming beat is shown In (b) both signals are aligned considering the procedure described in Section 233 for the intervalbetween QTbreak and T2

12

1

08

Template QRS1 ndash QRS2

Beat QRS1 ndash QRS2

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(a)



12

1

08

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(b)

Figure 7 Second stage for the alignment of the QRS template and the incoming beat In (a) the incoming beat is shifted with respect to theQRS template In (b) both signals are aligned for the interval between QRS1 and QRS2


samples 1 to 10 are the samples of the reconstructed waveletdetail located at the left of the selected Qini ducial pointSamples 11 to 30 are the samples in the ECG beat located ona window centered on the selected point and samples 31 to41 are the samples of the reconstructed wavelet detail locatedat right of the selectedQini ducial pointe construction ofeach marker vector is illustrated in Figure 12 At the left ofthe gure the features vector content is shown At the rightthe window of the data source used for constructing thefeatures vector is shown e top and bottom are thereconstructed wavelet details of the beat and at the center is

the data from the corresponding QRS beat e markers forthe Tend are constructed similarly however the waveletdecomposition uses a Symlet-7 wavelet and the recon-structed detail 6 is considered for representing the markere length of the marker in this case is longer (51 samples)but the dierence is that samples from 11 to 40 are extractedfrom the incoming beat e rest of samples comes from thereconstructed wavelet detail

(4) Nonmarker Selection e nonmarker features vector isconstructed with the same size and composition as the

08060402

s

0

(a)

04

ndash02

02

d60

(b)

Figure 10 (a) A QRS signal template (b) e IDWT reconstructed detail level 6 using Symlet-7 Two peaks are shown at the end of the Twave one positive and one negative including a zero crossing

ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)

Figure 11 (a) AQRS signal template (b)e reconstructed Daubechies-4 detail level 5e zero crossings and peaks correlate with theQ Rand S ducial points

ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend

Template T1 ndash Tend

(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)

Figure 9 Second stage for the alignment of the T-wave template and the T wave in the incoming beat In (a) the incoming T wave isshifted with respect to the T-wave template In (b) both signals are aligned for the interval between T1 and Tend


marker feature vector However in this case the samplesfrom the ECG beat and reconstructed wavelet detail arelocated outside from the marker regions )e location isselected at random from the beat and far from the fiducialpoint (Qini or Tend) For eachmarker vector a number of fivenonmarker vectors are selected )is procedure was per-formed for each of the 1800 beats considered for training

(5) Parameters Tuning Parameters gamma and sigma of theSVMwere set using the following procedure the detection ofthe fiducial point was performed in incoming beats extractedfrom the test set for classification accuracy estimation )eparameters were varied for obtaining a classification accu-racy larger than 95 )e selected parameters for the Qiniwere gamma 900 and sigma 25 and for the Tend weregamma 10 and sigma 60

(6) Detection of Qini and Tend Using the Trained SVM )edetection of Qini and Tend points is performed using thefollowing procedure

(1) In each incoming beat the level 5 reconstructionusing the IDWT is obtained using the db4 waveletSimilarly the level 6 reconstruction using theSymlet-7 wavelet is calculated

(2) A search window is established taking as referencethe fiducial points calculated for the template on eachlead Qkini0 and Tkend )us the search window forQini is centered in the approximated initial fiducialpoint for the current lead and it has a length of sevensamples in the interval [Qkini0 minus 3 Qkini0 Qkini0 + 3]Similarly for the end of the T wave the searchwindow is located in the following interval[Tkend minus 3 Tkend Tkend + 3]

(3) For each of the points in the search window thefeatures vector is constructed and classified using theSVM )e points classified as markers are labeledwith minus1 and nonmarker with +1

(4) )e points with a predicted label of minus1 are selected Ifthere is more than one point with a predicted label

minus1 the first point of the group is selected as Qkini Inthe event that no point is labeled with minus1 during theQini detection the parameter is set as Qkini Qkini0using the value calculated for the initial templateConcerning the Twave when there is more than onepoint with predicted label minus1 the last point of thegroup is selected as Tkend Similarly in the event ofno points labeled with minus1 the parameter is set asTkend Tkend0 using the value calculated for theinitial template

Figure 13 shows a set of beats with different morphol-ogies and its corresponding Qini and Tend points detectedwith the SVM

3 Results and Validation of the Algorithm

31 Validation Using Synthetic Signals A test using simu-lated signals was performed based on the work reported in[34] A heartbeat was selected and the amplitude of the Twave was varied by reducing their amplitude to 80 50and 30)isQRS was replicated to obtain four signals with300 beats and white noise was added with zero mean andstandard deviation of 3 of the original T-wave amplitudeEach signal was replicated to complete the eight independentleads )e set of signals have a constant QT of 391msec inconsequence the expected QT variability is 0msec )eestimated QT for this experiment attained a standard de-viation of 14msec

32 Validation Using the Physionet Database )e multileadQT estimation method is validated using a subset of the PTBDiagnostic ECG annotated database [25 35] )is subsetincludes a group of 97 subjects where 80 belong to thecontrol group 7 are patients with hypertrophy 6 are patientswith valvular disease and 4 are patients with myocarditis)e database contains the standard 12-lead ECG and thesimultaneously recorded 3 Frank lead ECG )e signals aresampled at 1000Hz with a resolution of 05 μV and theyhave variable duration

)e validation process was performed by comparingthe values of the QT interval calculated by our algorithmwith respect to the manually estimated values of the QTparameter obtained by a group of Physionet experts (thisinformation is provided in a text file for this dataset) )ecomparison is expressed in terms of the percent error thepercentage error obtained for each patient is reported as wellas the mean and standard deviation of each of these mea-surements Results of the validation using the Physionet PTBdiagnostic ECG dataset are shown in the first row of Table 1)e first column corresponds to the group analyzed thesecond column corresponds to the value (mean plusmn std (minmax)) of the QT parameter in the Physionet reference file(QT phy) the third column represents the value of QTcalculated with our QT interval estimation algorithm (QTest) and the last column shows the percentage error for QTinterval (error ())

)e annotated QT interval considering the 97 patients ofthe Physionet database is 38783 plusmn 3257msec with a

DWT samples locatedat the right of the

reference point (lowast)

DWT samples locatedat the le of the


[times1

times10

times11

times30

31times4

1]

Samples from theincoming QRS beat

Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)

Figure 12 Feature vector for training the SVM used for detectingthe Qini


100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)

ndash06ndash04ndash02

002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )

100 200 300 500400 600 7000ndash07ndash06ndash05ndash04ndash03ndash02ndash01

0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)

Figure 13 Detection of Qonset and Tend points for dierent morphologies of the heartbeat


minimum of 31700msec and a maximum of 477msec )eestimated QT interval is 38474 plusmn 3261msec with a mini-mum of 31690msec and a maximum of 47460msec )eaverage percentage error is less than 5 260 plusmn 225 )eresults obtained with this subset of the Physionet database areclose to results reported by participants of the 2006 QT es-timation challenge as Christov and Simova [19] Howeverour approach is multilead and intended for QT variabilityanalysis

)e second row of Table 1 reports the values corre-sponding to the QT interval obtained when processing the80 electrocardiography records of control subjects using ouralgorithm compared with respect to the values provided inthe Physionet reference file )e calculation of the intervalQT indicates a good correlation with respect to the esti-mation provided in the Physionet dataset presenting anaverage error of 235)is set of control subjects belongs tothe test set and the percent of error is low demonstrating thegood performance of the algorithm proposed )e third rowof Table 1 reports the results obtained for the QT intervalestimation considering the 7 electrocardiography records ofpatients with myocardial hypertrophy )e results arecompared with respect to the values provided in thePhysionet reference file In this case the percentage of errorfor the mean QT interval is less than 5 presenting amaximum error of 82 and a minimum error of 142

Results of QT intervals obtained when processing the 6electrocardiographic records of patients with valvular heartdisease using our algorithm with respect to the valuesprovided in the Physionet reference file are presented in thefourth row of Table 1 )e automatic estimation obtained bythe algorithm provides good results with an average errorfor the parameter QT less than 4

)e fifth row of Table 1 reports the estimated QT in-terval values obtained when processing the 4 electrocar-diography records of patients with myocarditis with respectto the values contained in the Physionet reference file )evalues for the average percent error are 311 and thestandard deviation is 248

)e results shown in Table 1 are satisfactory since thisvalidation group (test data) is not part of the SVM trainingdataset and even though the average percent error for themean and standard deviation is lower than 5 )ese resultsshow the good performance of the algorithm proposed inthis work Similarly it is worth to mention there is a goodcorrelation with respect to the values obtained for the QTinterval reported in [36] and using the same Physionetdataset [26] Results in Table 1 show globally an un-derestimation with respect to manual annotations of thePhysionet dataset of 309msec Such underestimation could

be explained by the fact that annotators performed themanual estimation of the QT interval using only the lead IIfor the Physionet dataset [35] used in this validation Incontrast our algorithm performs the QT estimation using amultilead approach where all leads have a similar weight aswe are simply averaging theQT intervals measured in each ofthe leads A bias in the estimation using this dataset has alsobeen reported by other researchers [37 38] during thevalidation of multilead QT estimation algorithms

33 Validation with a Dataset Acquired UsingDIGICARDIAC Results of the validation using 60 patientsacquired with the DIGICARDIAC system are shown inTable 2 )e validation for this dataset is performed bycomparing the average QT interval in the eight leads withrespect to the QT interval calculated using the Cardiosoftsoftware [17] )e average QT interval estimated by theCardiosoft software is 38953 plusmn 3037msec with a minimumof 32000msec and a maximum of 45000msec )e averageQT interval estimated using our algorithm is39162 plusmn 3094msec with a minimum of 31840msec and amaximum of 45330msec )e mean and standard de-viations of the QT differences between our method and theCardiosoft estimation are 249 plusmn 199msec)eminimum is008msec and the maximum is 823msec

Table 3 shows the numerical results corresponding to theparameter QT (msec) and QTc (msec) obtained by pro-cessing 20 electrocardiography records of hypertensivepatients using our algorithm and the CardioSoft application)e average percentage error for QT is 174 with aminimum value of 007 and a maximum value of 438Regarding the QTc the average error of 266 is obtainedwith a minimum of 007 and a maximum of 520 In thiscase the error is below 6 which is an acceptable measure

)e numerical results corresponding to the QT interval(msec) and QTc (msec) obtained by processing 20 electro-cardiographic records of chagasic patients using the algo-rithm proposed and the CardioSoft application are shown inTable 4 )e results provided by the proposed algorithmshow a good accuracy as the average error for the parametersQT and QTc is less than 4

Table 5 shows the numerical results corresponding to theQT and QTc parameter in msec In this case the estimation isperformed for 20 electrocardiography records of controlsubjects (healthy) using the proposed algorithm and comparedwith respect to the CardioSoft application Concerning theQTinterval the mean value of the error was 254 which is anacceptable measure Regarding the QTc the average error is187 )ese low error values show that the algorithm pro-posed could be useful for clinical application research

Table 1 Validation of QT estimation using the Physionet database )e parameter is represented as mean plusmn std (min max)

Group QT phy (msec) QT est (msec) Error ()Physionet dataset 38783 plusmn 3257 (317 477) 38474 plusmn 3261 (3169 4746) 260 plusmn 225 (0 1411)Control subjects 38696 plusmn 3266 (317 477) 38464 plusmn 3236 (3169 4746) 235 plusmn 185 (0 922)Myocardial hypertrophy 38014 plusmn 3936 (317 426) 37564 plusmn 4398 (3213 4381) 436 plusmn 217 (142 82)Valvular heart disease 39458 plusmn 3079 (346 426) 38386 plusmn 3267 (3443 4381) 364 plusmn 520 (049 1411)Myocarditis 40837 plusmn 1879 (3855 4270) 40385 plusmn 1017 (3951 4185) 311 plusmn 248 (027 592)


)e results shown in Tables 2ndash5 reflect the good per-formance of the methodology proposed in this research )ealgorithm offers a good accuracy for the estimation of theQTand QTc intervals with respect to the estimated values ob-tained using the commercial software provided by Car-dioSoft )e comparison shows an average error of less than4

4 Statistical Comparison between Patients andControl Subjects

In this section we report the results of using one-wayanalysis of variance to determine statistically significantchanges in the QT and QTc parameters between groups ofpatients One-way analysis of variance was used to de-termine statistically significant changes (plt 005 and Fgt 1)between groups of patients )e use of this type of statisticalanalysis is justified because there are independent samples oftwo groups and we want to contrast the null hypothesis(equality of means) with the hypothesis that the means arenot equal

Table 6 summarizes the results obtained for the QT andQTc parameter between hypertensive patients and controlsubjects in meanplusmn SD

As can be seen the average QTc and QT intervals arelarger for the group of hypertensive patients in comparisonwith the control group It is also important to note that thereare statistically significant differences between hypertensivepatients and control subjects represented by a value ofplt 005 In general the QT and QTc interval is prolonged inhypertensive patients with respect to healthy subjects asreported in [39ndash41]

Results obtained for the comparison of chagasic patientsand healthy subjects concerning the QTand QTc interval arereported in Table 7 Results show statistically significantdifferences for QT and QTc intervals between control sub-jects and chagasic patients (plt 005) Concerning this dis-ease further research is still necessary considering a larger

Table 2 Validation considering 60 patients acquired with DIGICARDIAC )e validation is performed by comparing with respect to theCardiosoft estimation

QT Cardiosoft (msec) QT est (msec) Error ()Mean 38953 39162 249Std dev 3037 3094 199Min 32000 31840 008Max 45000 45330 823

Table 3 QT interval (msec) and corrected QT (msec) for hypertensive patients with respect to the Cardiosoft application

Parameter QT est QT card QT Er () QTc est QTc card QTc Er ()Mean 38992 3912 174 42131 42395 266Std dev 2663 2963 136 2158 2049 133Min 3488 348 007 3797 380 007Max 4476 450 438 452 464 520

Table 4 QT interval (msec) and corrected QTc (msec) for chagasic patients compared with respect to the Cardiosoft application


Table 5 QT interval (msec) and corrected QTc (msec) for healthy subjects with respect to the Cardiosoft application


Table 6 Comparison for the QT and QTc interval for hypertensivepatients and control subjects

Parameter Hypertensive(meanplusmn std dev)

Healthy(meanplusmn std dev) plt 005

QT(msec) 38929plusmn 2706 38356plusmn 3118 0

QTc(msec) 42251plusmn 2531 43456plusmn 2911 0


group of patients and handling exclusion criteria concerningother diseases affecting the left ventricular repolarization[42]

5 Conclusions

)e DIGICARDIAC system incorporates a software toolthat enables accurate estimation of the multilead QT in-terval variability )e QT interval estimation algorithmuses high-performance machine learning techniques suchas support vector machines for accurate estimation of fi-ducial points Qini and Tend on a multilead QT estimationframework )e algorithm was validated using the Physi-onet QT interval dataset and additionally a second vali-dation was performed with respect to other commerciallyavailable software tools such as CardioSoft Results ob-tained during the validation process are promising as themean percentage error and deviation standard are low )evalidation performed on hypertensive patients comparedwith respect to healthy subjects confirms the presence ofQT interval prolongation in HTA patients with respect tothe normal subjects In contrast the validation consideringChagas disease patients suggests the need of improving theclinical protocol considering a more strict selection ofpatients

Data Availability

)is research used a third-party dataset available online forvalidation purposes )e dataset is the PTB Diagnostic ECGDatabase available at httpsphysionetorgphysiobankdatabaseptbdb

Conflicts of Interest

)e authors declare that they have no conflicts of interest

Acknowledgments

)is work was supported by the Prometeo Project of theMinistry of Higher Education Science Technology andInnovation of the Republic of Ecuador )e work was alsofunded by the Research Coordination (CDCHT) fromUniversidad de los Andes in Venezuela and the ResearchDepartment from University of Cuenca (DIUC) in Ecuador

References

[1] G A Roth C Johnson A Abajobir et al ldquoGlobal regionaland national burden of cardiovascular diseases for 10 causes

1990 to 2015rdquo Journal of the American College of Cardiologyvol 70 no 1 pp 1ndash25 2017

[2] B Williams G Mancia W Spiering et al ldquoTen command-ments of the 2018 ESCESH guidelines for the management ofarterial hypertensionrdquo European Heart Journal vol 39no 33 pp 3021ndash3104 2018

[3] F A Ehlert D S Cannom E G Renfroe et al ldquoComparisonof dilated cardiomyopathy and coronary artery disease inpatients with life-threatening ventricular arrhythmias dif-ferences in presentation and outcome in the avid registryrdquoAmerican Heart Journal vol 142 no 5 pp 816ndash822 2001

[4] E Braunwald ldquoCardiomyopathiesrdquo Circulation researchvol 121 no 7 pp 711ndash721 2017

[5] C Bern ldquoChagasrsquo diseaserdquo New England Journal of Medicinevol 373 no 5 pp 456ndash466 2015

[6] D F Davila J H Donis A Torres and J A Ferrer ldquoAmodified and unifying neurogenic hypothesis can explain thenatural history of chronic Chagas heart diseaserdquo InternationalJournal of Cardiology vol 96 no 2 pp 191ndash195 2004

[7] P C Pereira and E Navarro ldquoChallenges and perspectives ofchagas disease a reviewrdquo Journal of Venomous Animals andToxins including Tropical Diseases vol 19 no 1 p 34 2013

[8] J F Silva L S A Capettini J F P da Silva et al ldquoMech-anisms of vascular dysfunction in acute phase of Trypanosomacruzi infection in micerdquo Vascular Pharmacology vol 82pp 73ndash81 2016

[9] J Perez and I Molina ldquoChagas diseaserdquo gte Lancet vol 391no 10115 pp 82ndash94 2018

[10] S Mangini M d L Higuchi J T Kawakami et al ldquoInfectiousagents and inflammation in donated hearts and dilated car-diomyopathies related to cardiovascular diseases Chagasrsquo heartdisease primary and secondary dilated cardiomyopathiesrdquoInternational Journal of Cardiology vol 178 pp 55ndash62 2015

[11] M Baumert A Porta M A Vos et al ldquoQT interval variabilityin body surface ECG measurement physiological basis andclinical value position statement and consensus guidanceendorsed by the european heart rhythm association jointlywith the esc working group on cardiac cellular electrophys-iologyrdquo Europace vol 18 no 6 pp 925ndash944 2016

[12] P Postema and A Wilde ldquo)e measurement of the QT in-tervalrdquo Current Cardiology Reviews vol 10 no 3 pp 287ndash294 2014

[13] E de la Rosa P Paglini-Oliva L B Prato et al ldquoEarly de-tection of chronic asymptomatic chagas infectionrdquo MedicalScience Monitor vol 24 pp 4567ndash4571 2018

[14] J Bronzino and D Peterson ldquoBiomedical signals imagingand informatics serrdquo in gte Biomedical EngineeringHandbook CRC Taylor and Francis Boca Raton FL USA4th edition 2014

[15] V Starc and T T Schlegel ldquoReal-time multichannel systemfor beat-to-beat QT interval variabilityrdquo Journal of Electro-cardiology vol 39 no 4 pp 358ndash367 2006

[16] T Schlegel W Kulecz A Feiveson et al ldquoAccuracy of ad-vanced versus strictly conventional 12-lead ECG for detectionand screening of coronary artery disease left ventricularhypertrophy and left ventricular systolic dysfunctionrdquo BMCCardiovascular Disorders vol 10 no 1 p 28 2010

[17] Cardiosoft ldquoHigh frequency QRS electrocardiographrdquo 2009httpwwwcardiosoftcom

[18] P Karjalainen M Tarvainen and T Laitinen ldquoPrincipalcomponent regression approach for QT variability estima-tionrdquo in Proceedings of 27th Annual International ConferenceIEEE-EMBS Shanghai China 2005

Table 7 Comparison for the QT and QTc interval variability forchagasic patients with respect to control subjects

Parameter Chagasic(medianplusmn SD)

Healthy(medianplusmn SD) plt 005

QT (msec) 39890plusmn 350 38356plusmn 3118 0QTc(msec) 42060plusmn 2701 43456plusmn 2911 0


[19] I Christov and I Simova ldquoFully automated method for QTinterval measurement in ECGrdquo Computers in Cardiologyvol 33 pp 321ndash324 2006

[20] Q Xue and S Reddy ldquoAlgorithms for computerized QTanalysisrdquo Journal of Electrocardiology vol 30 pp 181ndash1861998

[21] K P Murphy Machine Learning A Probabilistic PerspectiveMIT Press Cambridge MA USA 2012

[22] S Huang N Cai P P Pacheco S Narandes Y Wang andW Xu ldquoApplications of support vector machine (SVM)learning in cancer genomicsrdquo Cancer Genomics-Proteomicsvol 15 no 1 pp 41ndash51 2018

[23] J A K Suykens J De Brabanter L Lukas and J VandewalleldquoWeighted least squares support vector machines robustnessand sparse approximationrdquo Neurocomputing vol 48 no 1ndash4pp 85ndash105 2002

[24] S Berkaya A Uysal E Gunal S Ergin S Gunal andM Gulmezoglu ldquoA survey on ECG analysisrdquo BiomedicalSignal Processing and Control vol 43 pp 216ndash235 2018

[25] M Oeff H Koch R Bousseljot and D Kreiseler gte PTBdiagnostic ECG database National Metrology Institute ofGermany Braunschweig Germany 2012 httpwwwphysionetorgphysiobankdatabaseptbdb

[26] A L Goldberger L A N Amaral L Glass et al ldquoPhysioBankPhysioToolkit and PhysioNetrdquo Circulation vol 101 no 23pp e215ndashe220 2000

[27] P Laguna R G Mark A Goldberg and G B Moody ldquoAdatabase for evaluation of algorithms for measurement of QTand other waveform intervals in the ECGrdquo in Computers inCardiology 1997 pp 673ndash676 IEEE Piscataway NJ USA1997

[28] N Dugarte R Medina L Huiracocha and R Rojas ldquoOpensource cardiology electronic health record development forDIGICARDIAC implementationrdquo in Proceedings of 11thInternational Symposium on Medical Information Processingand Analysis (SIPAIM 2015) p 96810Y International Societyfor Optics and Photonics Cuenca Ecuador November 2015

[29] A Savitzky andM J E Golay ldquoSmoothing and differentiationof data by simplified least squares proceduresrdquo Analyticalchemistry vol 36 no 8 pp 1627ndash1639 1964

[30] M Mneimneh R Povinelli and M Johnson ldquoIntegrativetechnique for the determination of QT intervalrdquo in Com-puters in Cardiology pp 329ndash332 IEEE Piscataway NJUSA 2006

[31] R Haddadi E Abdelmounim M El Hanine and A BelaguidldquoDiscrete wavelet transform based algorithm for recognitionof QRS complexesrdquoWorld of Computer Science amp InformationTechnology Journal vol 4 no 9 2014

[32] L Sornmo and P Laguna Bioelectrical Signal Processing inCardiac and Neurological Applications Elsevier AmsterdamNetherlands 2005

[33] P Addison gte Ilustrated Wavelet Transform HandbookCRC Press Boca Raton FL USA 2017

[34] M Baumert V Starc and A Porta ldquoConventional QT var-iability measurement vs template matching techniquescomparison of performance using simulated and real ECGrdquoPLoS One vol 7 no 7 Article ID e41920 2012

[35] G B Moody H Koch and U Steinhoff ldquo)e physionetcomputers in cardiology challenge 2006 QT interval mea-surementrdquo in Computers in Cardiology pp 313ndash316 IEEEPiscataway NJ USA 2006

[36] M Mneimneh R Povinelli and M Johnson ldquoIntegratetechnique for the determination of QT intervalrdquo in

Proceedings of Computer in Cardiology Challenge vol 33pp 329ndash332 Valencia Espantildea 2006

[37] R Gregg S Babaeizadeh D Feild E Helfenbein J Lindauerand S Zhou ldquoComparison of two automated methods for QTinterval measurementrdquo in Computers in Cardiologypp 427ndash430 IEEE Piscataway NJ USA 2007

[38] P Langley F Smith S King D Zheng A Haigh andA Murray ldquoFully automated computer measurement of QTinterval from the 12-lead electrocardiogramrdquo in Computers inCardiology pp 345ndash348 IEEE Piscataway NJ USA 2006

[39] A A Akintunde O E Ayodele O G Opadijo A T Oyedejiand O B Familoni ldquoQT interval prolongation and dispersionepidemiology and clinical correlates in subjects with newlydiagnosed systemic hypertension in Nigeriardquo Journal ofCardiovascular Disease Research vol 3 no 4 pp 290ndash2952012

[40] C Passino F Franzoni A Gabutti R Poletti F Galetta andM Emdin ldquoAbnormal ventricular repolarization in hyper-tensive patients role of sympatho-vagal imbalance and leftventricular hypertrophyrdquo International Journal of Cardiologyvol 97 no 1 pp 57ndash62 2004

[41] J Klimas T Stankovicova J Kyselovic and L BacharovaldquoProlonged QT interval is associated with blood pressurerather than left ventricular mass in spontaneously hyper-tensive ratsrdquo Clinical and Experimental Hypertension vol 30no 7 pp 475ndash485 2008

[42] T T Schlegel R Medina D Jugo et al ldquoCardiac re-polarization abnormalities and potential evidence for loss ofcardiac sodium currents on electrocardiograms of patientswith chagas heart diseaserdquo Journal of Electrocardiologyvol 40 no 6 pp S132ndashS133 2007


International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018


Active and Passive Electronic Components

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of



Journal ofEngineeringVolume 2018

SensorsJournal of



RotatingMachinery


Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom

estimation of the QT interval depends on the accuracy inidentification of Q onset and T end [11] Accurate mea-surement of the QT interval could be difficult for some T-wave morphologies and noisy electrocardiogram signals[12] Several research groups have shown that advancedhigh-resolution ECG (HRECG) processing enables the es-timation of several advanced indexes related to cardiovas-cular diseases )e study of HRECG started in the seventieswith the goal of noninvasive detection of electrical activity ofthe HisndashPurkinje system [13] Later other cardiac zones orECG intervals included micropotentials related with path-ological conduction of the heart [14]

Starc and Schlegel [15] have developed a system thatenables real-time assessment of the QT interval variabilityin the eight independent leads of a standard electrocar-diogram )is system includes hardware that performs theECG signal acquisition at 1000 samples per second in eachlead with a resolution of 12 bits per sample )e system alsoincludes advanced HRECG software for estimating thescore of reduced amplitude zones (RAZ) as well as heartrate variability (HRV) analysis either in time domain or infrequency domain Among the advanced tools for ECGanalysis it includes the T-wave morphological multileadanalysis based on principal component analysis (PCA)Based on this system Schlegel et al [16] have derived anadvanced ECG score for prediction of coronary arterydisease (CAD) which is based on a set of parameters es-timated using the advanced software analysis )e softwaresystem known as Cardiosoft [17] runs on the windowsoperating system

Karjalainen et al [18] developed a method for quanti-fying the QT interval variability )e method is based on atechnique known as principal component regression (PCR))is method allows reducing the dimensionality of data )evariation of the QT interval is proportional to the variationin a parameter representing the third eigenvalue of thecorrelation matrix )e algorithm divides the ECG signalinto segments including the QRS and the T wave and thenthe PCR algorithm is applied)is algorithm allows studyingthe QT interval variability without detecting the T wave

Christov and Simova [19] proposed an automatedmethodthat scored as second in the PhysionetCinC Challenge 2006)e method is based on analyzing the presence of peaks orslopes in two consecutive segments of 10msec )e searchis performed first for estimating the Q onset and then for theT end A revision of several classical methods for QT de-tection is presented by Xue and Reddy [20]

)e analysis of the ECG signal is becoming an im-portant tool for the assessment of several cardiac diseasesIn this context we are proposing a system includinghardware and software for HRECG analysis where supportvector machine (SVM) techniques [21ndash23] are used withinthe framework for QT interval estimation )e proposedsystem includes the signal acquisition electronics hardwareas well as a software system for ECG signal analysis )esoftware system is based on open-source libraries and runson the Linux operating system )e proposed systemperforms multilead QT interval estimation )e proposedmethodology is based on the work reported by Starc and

Schlegel [15] but we have used a different approach forpreprocessing and for segmentation of theQRS template onthe eight independent leads of the 12-lead standard surfaceelectrocardiogram In our algorithm rather than relying onthe QRS detector for determining the beginning of the QRS(Qonset) and a biparabolic curve modeling for de-termination at the end of the T wave (Tend) we performthese tasks using SVM techniques )e SVM learningtechnique is a very useful tool for classifying ECG patterns[24] )is has motivated their application for accurateestimation of this interval )e QT interval estimationmethod is validated using the Physionet PTB diagnosticECG database [25ndash27] as well as a dataset including agroup of 60 patients acquired using our digital electro-cardiograph system )e article is organized as followsfirst a brief introduction and state of the art is given as wellas a description of the theoretical bases and then themethodological framework is described Subsequently theresults are presented and finally conclusions are given

2 Methods

21 DIGICARDIAC ECG Acquisition System )e DIG-ICARDIAC system enables acquisition processing and han-dling of the electrocardiogram as well as the clinical patientinformation )e proposed system incorporates advanced al-gorithms for signal processing and interval estimation as wellas telemedicine tools for handling the patient electronic record)e DIGICARDIAC instrument has two types of operationmodes First of all it can be used as any electrocardiograph in aclinical consultation Secondly it works as an instrument ofdetailed analysis normally used in cardiac clinical research

)e electronic hardware of the DIGICARDIAC system[28] is used for acquiring the standard 12-lead electrocar-diogram )is hardware system was designed to work as awhole connected with a personal computer as shown inFigure 1 )e hardware starts working when it gets pluggedinto the USB port )e high-resolution electrocardiogramsignals are transmitted constantly to the computer )edeveloped software tools running on the personal computerallow the user to initiate the electrocardiogram data re-cording otherwise the transmitted information from thehardware is discarded )e acquisition is performed usingthe standard 12-lead system at 2000 samples per second witha resolution of 12 bits per sample An example of acquisitionis shown in Figure 2 where the V4 recorded lead is shown

211 Acquisition Data )e protocol included three groupsof subjects 20 patients with Chagas disease 20 patients withhypertension and 20 healthy subjects Patients underwent ahigh-resolution 12-lead electrocardiogram (ECG) recordingusing DIGICARDIAC A standard supine 12-lead ECG wasrecorded for 5minutes in each of the 60 patients duringnormal breathing

22 Multilead QT Interval Measurement )e algorithmreported for estimation of the QT interval is multilead in thesense that the eight independent leads of the standard






C1 1

1 + tan fcπT( 1113857


1 + tan fcπT( 1113857

(2)


Y(n) 1

M1113944

N



Cj 3N2



M (2N + 1) 4n2 + 4Nminus 3( 1113857

3 (5)




ADconvert


Sowarestage

Hardware stage


acquisition

Multichanneldif amp




100 200 300Time (ms)

400 500 600 7000ndash02

ndash01

0

01

02

03

04

05

06A

mpl

itude

(mv)

(a)

Time (ms)100 200 300 400 500 600 7000

ndash01

ndash005

0

005

01

015

02

025

03

Am

plitu

de (m

v)

(b)

Time (ms)100 200 300 400 500 600 7000

ndash03ndash02

ndash07

ndash05ndash04

ndash010

0102

ndash06

Am

plitu

de (m

v)

(c)

Time (ms)100 200 300 400 500 600 7000

ndash01

0

ndash04

ndash02

01

02

03

ndash03

Am

plitu

de (m

v)

(d)

Time (ms)100 200 300 400 500 600 7000

03

ndash03

ndash01

0504

ndash02

02

001

ndash04

Am

plitu

de (m

v)

(e)

Time (ms)100 200 300 400 500 600 7000

02

ndash03

ndash01

03

04

05

ndash02

01

0

Am

plitu

de (m

v)

(f )

Time (ms)100 200 300 400 500 600 7000

05

02

06

03

04

0

ndash01

01Am

plitu

de (m

v)

(g)

Time (ms)100 200 300 400 500 600 7000

Am

plitu

de (m

v)

05

02

06

03

04

0

ndash01

01

(h)













PQbreakQTbreak


QRS templateSignal

ndash02

ndash01

0

01

02

03

04

05

06

07

08

Am

plitu

de (m

v)

100 200 300 400 500 600 7000Time (ms)


700600500400Time (ms)

3002001000ndash02

ndash01

0

01

02

03

Am

plitu

de (m

v)

04

05

06

07

08


TendT1

T2






J(θ) minθsumN








ndash04

ndash02

0

02

04

06

08

1A

mpl

itude

(mv)

1501000 50Time (ms)


(a)

0 15010050Time (ms)

ndash04

ndash02

0

02

04

06

08

1

Am

plitu

de (m

v)


(b)







1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(a)

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(b)


12

1

08



06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(a)



12

1

08

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(b)






08060402

s

0

(a)

04

ndash02

02

d60

(b)


ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)


ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend


(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)





















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia






C1 1

1 + tan fcπT( 1113857


1 + tan fcπT( 1113857

(2)


Y(n) 1

M1113944

N



Cj 3N2



M (2N + 1) 4n2 + 4Nminus 3( 1113857

3 (5)




ADconvert


Sowarestage

Hardware stage


acquisition

Multichanneldif amp




100 200 300Time (ms)

400 500 600 7000ndash02

ndash01

0

01

02

03

04

05

06A

mpl

itude

(mv)

(a)

Time (ms)100 200 300 400 500 600 7000

ndash01

ndash005

0

005

01

015

02

025

03

Am

plitu

de (m

v)

(b)

Time (ms)100 200 300 400 500 600 7000

ndash03ndash02

ndash07

ndash05ndash04

ndash010

0102

ndash06

Am

plitu

de (m

v)

(c)

Time (ms)100 200 300 400 500 600 7000

ndash01

0

ndash04

ndash02

01

02

03

ndash03

Am

plitu

de (m

v)

(d)

Time (ms)100 200 300 400 500 600 7000

03

ndash03

ndash01

0504

ndash02

02

001

ndash04

Am

plitu

de (m

v)

(e)

Time (ms)100 200 300 400 500 600 7000

02

ndash03

ndash01

03

04

05

ndash02

01

0

Am

plitu

de (m

v)

(f )

Time (ms)100 200 300 400 500 600 7000

05

02

06

03

04

0

ndash01

01Am

plitu

de (m

v)

(g)

Time (ms)100 200 300 400 500 600 7000

Am

plitu

de (m

v)

05

02

06

03

04

0

ndash01

01

(h)













PQbreakQTbreak


QRS templateSignal

ndash02

ndash01

0

01

02

03

04

05

06

07

08

Am

plitu

de (m

v)

100 200 300 400 500 600 7000Time (ms)


700600500400Time (ms)

3002001000ndash02

ndash01

0

01

02

03

Am

plitu

de (m

v)

04

05

06

07

08


TendT1

T2






J(θ) minθsumN








ndash04

ndash02

0

02

04

06

08

1A

mpl

itude

(mv)

1501000 50Time (ms)


(a)

0 15010050Time (ms)

ndash04

ndash02

0

02

04

06

08

1

Am

plitu

de (m

v)


(b)







1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(a)

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(b)


12

1

08



06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(a)



12

1

08

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(b)






08060402

s

0

(a)

04

ndash02

02

d60

(b)


ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)


ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend


(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)





















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


100 200 300Time (ms)

400 500 600 7000ndash02

ndash01

0

01

02

03

04

05

06A

mpl

itude

(mv)

(a)

Time (ms)100 200 300 400 500 600 7000

ndash01

ndash005

0

005

01

015

02

025

03

Am

plitu

de (m

v)

(b)

Time (ms)100 200 300 400 500 600 7000

ndash03ndash02

ndash07

ndash05ndash04

ndash010

0102

ndash06

Am

plitu

de (m

v)

(c)

Time (ms)100 200 300 400 500 600 7000

ndash01

0

ndash04

ndash02

01

02

03

ndash03

Am

plitu

de (m

v)

(d)

Time (ms)100 200 300 400 500 600 7000

03

ndash03

ndash01

0504

ndash02

02

001

ndash04

Am

plitu

de (m

v)

(e)

Time (ms)100 200 300 400 500 600 7000

02

ndash03

ndash01

03

04

05

ndash02

01

0

Am

plitu

de (m

v)

(f )

Time (ms)100 200 300 400 500 600 7000

05

02

06

03

04

0

ndash01

01Am

plitu

de (m

v)

(g)

Time (ms)100 200 300 400 500 600 7000

Am

plitu

de (m

v)

05

02

06

03

04

0

ndash01

01

(h)













PQbreakQTbreak


QRS templateSignal

ndash02

ndash01

0

01

02

03

04

05

06

07

08

Am

plitu

de (m

v)

100 200 300 400 500 600 7000Time (ms)


700600500400Time (ms)

3002001000ndash02

ndash01

0

01

02

03

Am

plitu

de (m

v)

04

05

06

07

08


TendT1

T2






J(θ) minθsumN








ndash04

ndash02

0

02

04

06

08

1A

mpl

itude

(mv)

1501000 50Time (ms)


(a)

0 15010050Time (ms)

ndash04

ndash02

0

02

04

06

08

1

Am

plitu

de (m

v)


(b)







1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(a)

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(b)


12

1

08



06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(a)



12

1

08

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(b)






08060402

s

0

(a)

04

ndash02

02

d60

(b)


ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)


ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend


(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)





















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia












PQbreakQTbreak


QRS templateSignal

ndash02

ndash01

0

01

02

03

04

05

06

07

08

Am

plitu

de (m

v)

100 200 300 400 500 600 7000Time (ms)


700600500400Time (ms)

3002001000ndash02

ndash01

0

01

02

03

Am

plitu

de (m

v)

04

05

06

07

08


TendT1

T2






J(θ) minθsumN








ndash04

ndash02

0

02

04

06

08

1A

mpl

itude

(mv)

1501000 50Time (ms)


(a)

0 15010050Time (ms)

ndash04

ndash02

0

02

04

06

08

1

Am

plitu

de (m

v)


(b)







1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(a)

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(b)


12

1

08



06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(a)



12

1

08

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(b)






08060402

s

0

(a)

04

ndash02

02

d60

(b)


ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)


ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend


(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)





















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





J(θ) minθsumN








ndash04

ndash02

0

02

04

06

08

1A

mpl

itude

(mv)

1501000 50Time (ms)


(a)

0 15010050Time (ms)

ndash04

ndash02

0

02

04

06

08

1

Am

plitu

de (m

v)


(b)







1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(a)

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(b)


12

1

08



06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(a)



12

1

08

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(b)






08060402

s

0

(a)

04

ndash02

02

d60

(b)


ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)


ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend


(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)





















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia






1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(a)

1

05

Am

plitu

de (m

v)

0

ndash050 50 100 150

Time (ms)200 250 300



(b)


12

1

08



06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(a)



12

1

08

06

04

Am

plitu

de (m

v)

02

0

ndash020 5 10 15

Time (ms)20 25 30

(b)






08060402

s

0

(a)

04

ndash02

02

d60

(b)


ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)


ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend


(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)





















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





08060402

s

0

(a)

04

ndash02

02

d60

(b)


ndash01

010203040506

0

100 200 300 400 500 600

s

SQ

R

(a)

100

ndash015ndash01

ndash0050

00501

015

200 300 400 500 600

d5

Min R

Max

(b)


ndash05

0

05

1A

mpl

itude

(mv)

8020 40 1000 12060Time (ms)

Beat T1 ndash Tend


(a)

Beat T1 ndash Tend


ndash05

0

05

1

Am

plitu

de (m

v)

20 40 600 100 12080Time (ms)

(b)





















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




















[times1

times10

times11

times30

31times4

1]


Time (ms)90 100 110

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

120 130 140 150 160

Am

plitu

de (m

v)

Time (ms)110

0

ndash005

ndash01

ndash015

ndash02

120 130 140 150 160 170A

mpl

itude

(mv)

Time (ms)90

025

02

015

01

005

0

ndash005

ndash01

ndash015

ndash02

ndash025

100 110 120 130 140 150 160

Am

plitu

de (m

v)



100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


100 200 300 400 500 600 7000ndash01

001020304050607

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(a)

100 200 300 400 500 600 7000

ndash01

ndash02

0

01

02

03

04

05

Time (ms)

Am

plitu

de (m

v)

Qonset

Signal

Tend

(b)

100 200 300 400 500 600 7000ndash02

0

01

02

03

04

05

ndash01

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(c)

Am

plitu

de (m

v)

0

ndash1

05

ndash05

1000 200 300 400 500 600 700Time (ms)

Qonset

Signal

Tend

(d)

100 200 300 400 500 600 7000Time (ms)

Am

plitu

de (m

v)


002040608

Qonset

Signal

Tend

(e)

1000 200 300 400 500 600 700Time (ms)

ndash04ndash02

040608

002

1

Am

plitu

de (m

v)

Qonset

Signal

Tend

(f )


0

030201

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(g)

100 200 300 400 500 600 7000

08

ndash02

06

0

1

02

04

12

Am

plitu

de (m

v)

Time (ms)

Qonset

Signal

Tend

(h)





































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




































5 Conclusions


Data Availability




Acknowledgments


References





















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia






























RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


ECGMultilead QT IntervalEstimationUsingSupport...

Documents

Transcript of ECGMultilead QT IntervalEstimationUsingSupport...