research article a reliable method for rhythm analysis

Research ArticleA Reliable Method for Rhythm Analysis duringCardiopulmonary Resuscitation

U Ayala1 U Irusta1 J Ruiz1 T Eftestoslashl2 J Kramer-Johansen3

F Alonso-Atienza4 E Alonso1 and D Gonzaacutelez-Otero1

1 Communications Engineering Department University of the Basque Country UPVEHU Alameda Urquijo SN48013 Bilbao Spain

2Department of Electrical Engineering and Computer Science Faculty of Science and TechnologyUniversity of Stavanger 4036 Stavanger Norway

3Norwegian Centre for Prehospital Emergency Care (NAKOS) Oslo University Hospital and University of Oslo 0424 Oslo Norway4Department of Signal Theory and Communications University Rey Juan Carlos Camino del Molino SN 28943 Madrid Spain

Correspondence should be addressed to U Irusta unaiirustaehues

Received 10 February 2014 Revised 26 March 2014 Accepted 28 March 2014 Published 7 May 2014

Academic Editor Yongqin Li

Copyright copy 2014 U Ayala et alThis is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Interruptions in cardiopulmonary resuscitation (CPR) compromise defibrillation success However CPR must be interrupted toanalyze the rhythm because although currentmethods for rhythm analysis during CPR have high sensitivity for shockable rhythmsthe specificity for nonshockable rhythms is still too lowThis paper introduces a new approach to rhythm analysis during CPR thatcombines two strategies a state-of-the-art CPR artifact suppression filter and a shock advice algorithm (SAA) designed to optimallyclassify the filtered signal Emphasis is on designing an algorithmwith high specificityThe SAA includes a detector for low electricalactivity rhythms to increase the specificity and a shockno-shock decision algorithm based on a support vector machine classifierusing slope and frequency features For this study 1185 shockable and 6482 nonshockable 9-s segments corrupted by CPR artifactswere obtained from 247 patients suffering out-of-hospital cardiac arrest The segments were split into a training and a test set Forthe test set the sensitivity and specificity for rhythm analysis during CPR were 910 and 966 respectively This new approachshows an important increase in specificity without compromising the sensitivity when compared to previous studies

1 Introduction

Out-of-hospital cardiac arrest (OHCA) is a leading causeof mortality in the industrialized world with an estimatedannual incidence between 28 and 55 cases per 100000person-years [1] Early cardiopulmonary resuscitation (CPR)and early defibrillation are the key interventions for survivalafter cardiac arrest [2] Defibrillation may be administeredby an automated external defibrillator (AED) which incor-porates a shock advice algorithm (SAA) that analyzes theECG to detect shockable rhythms Current CPR guidelinesemphasize the importance of high quality CPR with minimalinterruptions in chest compressions (CCs) [3] However CPRmust be interrupted for a reliable rhythm analysis becauseCCs produce artifacts in the ECG These interruptionsadversely affect the probability of defibrillation success and

subsequent survival [4] Currently CPR is interrupted every2 minutes for rhythm reassessment on an artifact-free ECG

Although different approaches to rhythm analysis duringCPRhave been explored for instance algorithms that directlydiagnose the ECG corrupt with CPR artifacts [5 6] filteringthe CPR artifact has been a major approach (see [7] fora comprehensive review) The time-varying characteristicsof the CPR artifact and its spectral overlap with bothshockable and nonshockable cardiac arrest rhythmsmandatethe use of adaptive filters [8] which use reference signalsto model the CPR artifact Over the years many solutionshave been proposed including Wiener filters [9] Match-ing Pursuit Algorithms [10] Recursive Least Squares [11]least mean squares (LMS) [12] or Kalman filters [13 14]Adaptive solutions using exclusively the ECG have also beenexplored [15 16] but the results were poorer To evaluate

Hindawi Publishing CorporationBioMed Research InternationalVolume 2014 Article ID 872470 11 pageshttpdxdoiorg1011552014872470

2 BioMed Research International

the performance of these methods researchers first filteredthe CPR artifact and then analyzed the rhythm using aSAA to obtain the sensitivity and specificity of the methodthat is the proportion of correctly diagnosed shockable andnonshockable rhythms respectively However the SAAs usedwere originally designed to analyze artifact-free ECG insteadof the ECG after filtering

Currently rhythm analysis during CPR is not possible[17] Most methods have sensitivity above 90 the mini-mum value recommended by the American Heart Associ-ation (AHA) for SAA on artifact-free ECG [18] Howeverspecificity rarely exceeds 85 well below the 95 valuerecommended by the AHA A low specificity would result ina large number of false shock diagnoses during CPR whichwould unnecessarily increase the number of interruptionsin CPR Overall the main cause of the low specificity isfiltering residuals in nonshockable rhythms These residualsfrequently resemble a disorganized rhythm [10 12] and areoftenmisdiagnosed as shockable by SAAs designed to analyzeartifact-free ECG This problem is more prominent whenthe electrical activity of the underlying heart rhythm islow particularly for asystole (ASY) [14 16] because filteringresiduals may have amplitudes comparable or larger thanthose of the underlying ECG

In this study we explore the possibility of combiningadaptive filtering techniques with a SAA designed to opti-mally classify the rhythm after filteringThe aim is to improvethe accuracy of current approaches and in particular toovercome the low specificity When compared to previousstudies our results showed an increased specificity withoutcompromising the sensitivity for a comprehensive dataset ofOHCA rhythms

2 Materials and Methods

21 Data Collection The data for this study were extractedfrom a large prospective study of OHCA conducted between2002 and 2004 in three European sites [21 22] CPR wasdelivered by trained ambulance personnel in adherence to the2000 resuscitation guidelines Episodes were recorded usingmodified Laerdal Heartstart 4000 defibrillators (4000SP) andan external CPR assist pad to acquire additional referencesignals All signals were acquiredwith a 500Hz sampling rateThe initial rhythmand all subsequent changes in rhythmwereannotated by consensus of an experienced anesthesiologistand a biomedical engineer both specialized in resuscitation[21 22] Rhythm annotations comprised five types (see [21]for further details) VF and fast ventricular tachycardia(VT) in the shockable category and ASY pulseless electricalactivity (PEA) and pulse-generating rhythm (PR) in thenonshockable category Intervals with chest compressionswere annotated using the compression depth (CD) obtainedfrom the CPR assist pad

For this study specific records containing the ECG andCD signals were automatically extracted from the originalepisodes First rhythm transitions were identified using theoriginal annotations and then for each interval withoutrhythm transitions at most one record was extracted to avoid

bias due to data selection Records were extracted if thefollowing criteria were met duration of more than 20-songoing CCs and the same rhythm annotation before andafter CCs Following the AHA statement the records weregrouped into a shockable and a nonshockable category Theamplitude thresholds adopted for coarse VF and ASY arethose accepted in the literature on SAAs [6 18]The followingcriteria and rhythm definitions were checked in the cleanintervals before and after CCs

Shockable RhythmsThis category includes fast VT with rateabove 150 beats per minute (bpm) and coarse VF Coarse VFwas defined asVFwith peak-to-peak amplitude above 200 120583Vand a fibrillation frequency above 2Hz

Nonshockable Rhythms These rhythms were further dividedinto the following

(i) organized rhythms (ORG) all nonshockable rhythmsexcept ASY (PEA and PR)

(ii) asystole (ASY) rhythms with peak-to-peak ampli-tudes below 100 120583V for at least 2-s

All signals were resampled to 119891119904= 250Hz a sampling rate

similar to that used by commercial AEDs Inwhat follows thesample index and time variables are related by 119905 = 119899 sdot119879

119904(119879119904=

1119891119904) The ECG was band limited to 05ndash30Hz (order 10

Butterworth filter) a typical ECG monitor bandwidth usedin AEDs [5 6] which removes base line wander and highfrequency noise

Following standard practice in SAA design the rhythmanalysis method was designed to analyze three consecutive3 s windows so it gives a diagnosis every 9 s [23 24] A 3 swindow is sufficient to characterize the rhythm in terms ofrate stability and morphology and to make a shock (Sh)or no-shock (NSh) decision [23] SAA algorithms combineseveral consecutive diagnoses to avoid errors due to rhythmtransitions and to avoid shock diagnoses for short burstsof nonsustained VT Therefore each record was dividedinto nonoverlapping 9 s segments The 9 s segments wererandomly split into two separate sets one to train thealgorithm and an independent set to test the algorithm asrequired by the AHA statement In addition we made surethat the patients on both sets were different (AHA statement)and that the distribution of rhythm types was similar in bothsets

22 Rhythm Analysis Method The block diagram of therhythm analysis method is shown in Figure 1 First a CPRartifact suppression filter estimates the underlying rhythmthat is the filtered ECG signal 119904filt Then a SAA diagnosesevery 3 s window of the filtered signalThe SAA is designed tooptimally classify the filtered signal and is further composedof two sequential subalgorithms (1) a detector of rhythmswith low electrical activity (LEA) that is nonshockablerhythmswithout distinguishableQRS complexes such as ASYor idioventricular rhythms and (2) a ShNSh algorithm thatclassifies windows with electrical activity as shockable ornonshockable

BioMed Research International 3

++

CD Compdetector

Modelartifact

scor

scpr

LMS filter

sfiltminus

LEAdetector

ShNShalgorithm

Signal flowDecision flow

Sh

NSh NSh

SAA optimized to analyze sfilt

Figure 1 Block diagram of the new approach to rhythm analysisduring CPR in which an adaptive filter (LMS filter based on the CDsignal) is used in combination with a SAA designed to optimallyclassify the filtered signal 119904filt

11

minus11

0

s cor(m

V)

11

minus11

0

s cpr(m

V)

11

minus11

0

s filt(m

V)

minus6

minus3

0

Time (s)12960

CD (c

m)

t1 t2 middot middot middot tk

Figure 2 Filtering example of a 12-s segment In the first 3 s thereis no artifact and the underlying VF is visible The filter estimatesthe artifact 119904cpr using information derived from the CC marksindicated by vertical lines in CD (119905

119896instants)

23 Chest Compression Artifact Filter CPR artifacts weresuppressed using a state-of-the-art method based on an LMSfilter [12] In this method CC artifacts are modeled as aquasiperiodic interference with a time-varying fundamentalfrequency 119891

119900(119899) which is the instantaneous frequency of

the CCs This frequency is derived from the 119905119896instants the

CC marks shown in Figure 2 The LMS algorithm adaptivelyestimates the time-varying amplitudes 119888

119896(119899) and phases

120601119896(119899) of the first 5 harmonics of the artifact by fitting the

following model

119904cpr (119899) =5

sum

119896=1

119888119896(119899) cos (119896 sdot 2120587119891

119900(119899) sdot 119899 + 120601

119896(119899))

119891119900(119899) =

1

119905119896minus 119905119896minus1

for 119905119896minus1

le 119899119879119904lt 119905119896

(1)

In summary the LMS algorithm dynamically estimates theCPR artifact by adaptively estimating its harmonic contentFor this study we used the optimal values of the filterparameters as described in [12 19] As shown in Figure 1 thefiltered signal was obtained by subtracting the estimated CPRartifact from the corrupted ECG Figure 2 shows those signalsfor a 12-s segment with an underlying VF rhythm

24 Shock Advice Algorithm The SAA consists of a LEAdetector followed by the ShNSh algorithmTheLEAdetectoridentifies LEA windows as nonshockable the rest of thewindows are further processed by the ShNSh algorithm fora definitive diagnosis

241 LEA Rhythm Detector Some nonshockable rhythms(ASY bradyarrhythmias or idioventricular rhythms) maynot present QRS complexes in a 3 s analysis window Inthese cases filtering the CC artifact results in 119904filt withlow amplitudes and short intervals in which the electricalactivity is very low (see Figure 3(a)) To further improve LEAdetection 119904filt was high pass filtered with a 25Hz cut-offfrequency using an order 5 Butterworth filter which removedslow fluctuations of the ECG in LEA rhythms but preservedmost frequency components of VF as shown in Figure 3Theresulting signal 119904LEA was used to obtain the following twofeatures

(i) 119875LEA energy of 119904LEA in the 3 s window

119875LEA = sum119899

1199042

LEA (119899) (2)

(ii) 119871min minimum of the curve lengths of 119904LEA fornonoverlapping 05-s intervals which measures theminimum electrical activity in 05-s intervals Indiscrete form the curve length of the 119896th subintervalis [25]

119871119896=

(119896+1)1198911199042

sum

119899=1198961198911199042+1

radicΔ1199042LEA (119899) + 1198792

119904 (3)

where Δ119904LEA is the first difference of 119904LEA

LEA rhythms have smaller values of 119875LEA and 119871min thanshockable rhythms as shown in Figure 3 This block wasdesigned as a detector that is it gives a NSh diagnosis ifa LEA rhythm is detected otherwise the window is furtherprocessed by the ShNSh algorithm


03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

PLEA = 009 Lmin = 018

(a) ASY

PLEA = 274 Lmin = 135

03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

(b) VF

Figure 3 Examples of a LEA rhythm (a) and a VF (b) window after filtering the CC artifact 119904filt and preprocessed to suppress componentsunder 25Hz 119904LEA Vertical lines separate the 05 subintervals and the one with lowest activity (119871min) is shown in red

242 ShNSh Algorithm During resuscitation ORGrhythms with electrical activity may be very different interms of rate QRS width or QRS morphology Furthermoreeven after CPR artifact suppression rhythms may presentimportant filtering residuals that may resemble VF Fourfeatures derived from the frequency domain and slopeanalyses were defined For rhythms with electrical activitythese features emphasize the differences between nonshock-able (with QRS complexes) and shockable (without QRScomplexes) rhythms

243 SlopeAnalysis Features QRS complexeswere enhancedin 119904filt by computing the moving average of the square of itsfirst difference (its slope)

119889filt (119899) =1

119873

119873minus1

sum

119896=0

(119904filt(119899 minus 119896) minus 119904filt (119899 minus 119896 minus 1))2

(4)

where 119873 corresponds to the number of samples in a 100msinterval Then 119889filt was divided by its maximum value in theanalysis window to obtain 119889filt As shown in Figure 4 in ORGrhythms 119889filt is large only around QRS complexes and verysmall otherwise whereas in VF the values of 119889filt are moreevenly distributed and presentmany peaks Two featuresweredefined to measure these differences

(i) 119887119878 slope baseline a measure of how concentratedslope values are around small values (baseline) com-puted as the 10th percentile of 119889filt

(ii) 119899119875 number of peaks above a fixed threshold in 119889filt

Shockable rhythms will present larger values of 119887119878 and 119899119875 asshown in Figure 4

244 Frequency Domain Features For the frequency anal-ysis a Hamming window was applied to 119904filt and its zeropadded 1024-point FFT was computed The power spectral

density was estimated as the square of the magnitude of theFFT and normalized to total power of one to give 119875ss(119891)As shown in Figure 5 VF concentrates most of its poweraround the fibrillation frequency whereasORG rhythmsmayhave important power content at higher frequencies on theharmonics of the heart rate Two discrimination featureswere defined with limiting frequencies in line with thecharacteristics of human VF [26 27]

(i) 119875fib power proportion around the VF-fibrillationband (25ndash75Hz)

(ii) 119875ℎ power proportion in the high spectral bands

(above 12Hz)

Shockable rhythms have larger values of 119875fib but lower valuesof 119875ℎ(see Figure 5)

245 Support Vector Machine (SVM) Classifier The ShNShalgorithm classified windows using a SVM with a Gaussiankernel [28] First features were standardized to zero meanand unit variance using the data in the training set Thesex119894vectors of four normalized features were arranged as

(x1 1199101) (x

119899 119910119899) isin R4timesplusmn1 where 119910

119894= 1 for shockable

and 119910119894= minus1 for nonshockable windows After training the

discriminant function for a window with feature vector x is

119891 (x) =119873119904

sum

119894=1

120572119894119910119894exp (minus1205741003817100381710038171003817x minus x

119894

10038171003817100381710038172

) + 119887 (5)

where x119894are the support vectors 119873

119904is the number of

support vectors and120572119894and 119887 are coefficients estimated during

training Windows were classified as shockable for 119891(x) gt 0or nonshockable for 119891(x) le 0 Selecting an optimal SVMmodel for the classification problem involves selecting twoparameters 119862 and 120574 The width of the Gaussian kernel120574 determines the flexibility of the decision boundary [28]The soft margin parameter 119862 is used exclusively in theoptimization process and is a tradeoff between classification


04

minus04

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0139

dfil

t(nu)

nP = 19

(a)

14

minus14

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0014

dfil

t(nu)

nP = 7

(b)

Figure 4 Example of the slope analysis for VF (a) and an ORG (b) window During VF the slope 119889filt is irregular with many peaks whereasORG rhythms are regular with fewer peaks and concentrate most 119889filt values around the baseline

04

minus04

0

s filt(m

V)

Time (s)3210

0

Frequency (Hz)3020100

004

Pss(n u)

Pfib = 084 Ph = 001

(a) VF

1

minus1

0

s filt(m

V)

Time (s)3210

0

002

Pss( nu)


Pfib = 050 Ph = 012

(b) ORG

Figure 5 Example of the frequency domain analysis for VF (a) and an ORG (b) window VF concentrates most of its power around thefibrillation band (blue) ORG rhythms have a spectrum with many harmonics of the heart rate and thus larger 119875

ℎ(in green)

errors in training data and separating the rest of the trainingdata with maximummargin [28]

25 Data Analysis and Algorithm Optimization The rate anddepth characteristics of CPR in our data were analyzed foreach 9 s segment The distributions for rate and depth didnot pass the Kolmogorov-Smirnov test for normality and arereported as median and 5ndash95 percentiles

For each discrimination feature of the SAA statisticaldifferences in medians between the targeted classificationgroups of each subalgorithmweremeasured using theMann-Whitney119880 test The optimization process was carried out forthe 3 s windows of the training set in two sequential steps

(1) LEADetector ASY and shockable rhythmswere usedThe detection thresholds were determined through agreedy search on the two-dimensional feature space

to jointly maximize the number of detected ASYand minimize the number of shockable windowsincorrectly detected as nonshockable An additionalrestriction was imposed at maximum 5 of shock-able windows could be incorrectly classified

(2) ShNSh Algorithm Shockable and ORGwindows notdetected as NSh by the LEA detector were used tooptimize the SVM classifier To avoid overfitting theSVM to the training set 119862 and 120574 were selected using5-fold crossvalidation [29] to optimize the balancederror rate (BER)

BER = 1 minus 12(TPR + TNR) (6)

where the true positive rate (TPR) and the truenegative rate (TNR) are the capacity of the SVM


Table 1 Number of segments (patients in parenthesis) and characteristics of the CC rate and depth for the training and test datasets Valuesfor CC rate and depth are presented as median with 5ndash95 percentiles in parenthesis

Rhythm type Training Testing9-s seg Rate (cpm) Depth (mm) 9-s seg Rate (cpm) Depth (mm)

Shockable 563 (35) 116 (92ndash143) 38 (25ndash47) 622 (34) 113 (89ndash157) 35 (20ndash50)Nonshockable 3132 (110) 116 (88ndash155) 36 (20ndash51) 3350 (109) 116 (84ndash159) 35 (21ndash57)AS 1173 (66) 118 (92ndash164) 35 (18ndash52) 1309 (60) 117 (89ndash151) 34 (21ndash52)ORG 1959 (66) 114 (86ndash149) 37 (23ndash51) 2041 (76) 116 (79ndash164) 35 (21ndash59)Total 3695 (123) 116 (89ndash151) 36 (21ndash51) 3972 (124) 116 (86ndash159) 35 (21ndash58)

classifier to detect shockable and ORG windowsrespectively Weights were assigned to each class toresolve the unbalance in the number of instances perclass [28] The best SVMs using one two or threefeatures were compared to the optimal four-featureSVM using McNemarrsquos test

The performance of the algorithm was measured in thetest set in terms of sensitivity and specificity Since both 3 swindows and 9 s segments correspond to consecutive anal-yses within a record the sensitivities specificities and their90 low one-sided confidence intervals (CI) were adjustedfor clustering (longitudinal data) within each record usinga longitudinal logistic model fit by generalized estimatingequations (GEE) [30 31] The analysis was carried out in119877 using the geepack library [32] Finally the algorithmwas programmed in MATLAB R2013a (Mathworks Inc) forWindows and processing time performance tests were carriedout on a 29GHz Intel i7 with 4GB of RAM

3 Results

31 Database Description Our data comprise 7667 9 s seg-ments within 1396 records extracted from 247 OHCA patientepisodesThemedian number of 9 s segments per record was3 (1ndash19 range 1ndash44) Table 1 shows the number of 9 s segmentsand the rate and depth of CCs for those segments in thetraining and test sets The median CC rate and depth were116 (88ndash156) compressions per minute (cpm) and 36 (21ndash53)mm respectively

32 Shock Advice Algorithm

321 Training Figures 6(a) and 6(b) show the values of 119875LEAand119871min for theASY and shockable rhythmswhich presentedsignificant differences between the two groups (119875 lt 0001)The optimal detection thresholds of the LEA detector were

119875LEA lt 044 or 119871min lt 063 (7)

The LEA detector identified as NSh 721 of the ASY (truedetections) and 09 of the shockable (false detections)windows In addition 388 of the ORG windows werecorrectly identified as NSh these rhythms corresponded tovery low rate and low electrical activity intervals of ORGrhythms

Figures 6(c)ndash6(f) show the values of the features used inthe SVM classifier these values were statistically different for

theORGand shockable rhythms (119875 lt 0001)The SVMbasedon four features showed a significantly better performancewhen compared to the SVMs based on the best single pairor triplet of features (McNemarrsquos test 1205942 gt 10 119875 lt 0001 inall three cases)The optimal working point of the four-featureSVM was (119862 = 85 120574 = 01) which produced a BER = 0064TPR = 0927 and TNR = 0944 for the SVM classifierThe receiver operating characteristics analysis on the SVMfeatures resulted in the following area under the curve (AUC)values 0948 0928 0807 and 0733 for 119887119878 119899119875 119875fib and 119875ℎrespectively When combined in the SVM the resulting AUCwas 0971 which reveals the robustness of the classifier

322 Test The optimized SAA was used to classify the 3 swindows in the test set Table 2 shows a summary of theresults The overall sensitivity and specificity were 897(low one-sided 90 CI 855) and 951 (low 90 CI943) respectively The 9 s segments were diagnosed using amajority criterion on three consecutive window analyses thisincreased the overall sensitivity and specificity to 910 (low90CI 866) and 966 (low 90CI 959) respectively andAHA recommendations were met for all rhythm types (seeTable 2)

Figure 7 shows two examples (Figures 7(a) and 7(c)) ofcorrectly diagnosed segments and two examples (Figures 7(b)and 7(d)) of incorrectly diagnosed segments The examples(Figures 7(a) and 7(c)) show that the algorithm worksrobustly even in the presence of important filtering resid-uals However there were some instances of misdiagnosedsegments as shown in Figures 7(b) and 7(d) Errors weregenerally caused by spiky filtering residuals in shockablerhythms (Figure 7(b)) or large filtering residuals during ASY(Figure 7(d))

Processing time for the complete algorithm CPR sup-pression filter based on the LMS filter followed by the SAAwas on average 87ms per 3 s segment Processing time wasbroken down into 58ms for the LMS filter and 29ms forthe SAA For decisions taken by the LEA detector the SAArequired only 18ms and for windows in which the LEAdetector and the SVM were used it increased to 41ms Inthe worst case scenario processing time for the completealgorithm was under 10ms

4 Discussion

This study presents the first attempt to combine twoapproaches for rhythm analysis during CPR adaptive filters


0

4

8

12

ASY Sh

(a) 119875LEA

ASY Sh

0

1

2

3

(b) 119871min

000

006

012

018

ORG Sh

(c) 119887119878

ORG Sh

0

12

24

36

(d) 119899119875

ORG Sh

000

033

067

100

(e) 119875fib

ORG Sh

000

012

024

036

(f) 119875ℎ

Figure 6 Features of the SAA for the rhythm types used in each training stage of the algorithm For the LEA detector the figure comparesASY versus Sh (panels (a) and (b)) and for the SVM classifier ORG versus Sh (panels (c)ndash(f)) The boxes show the median and interquartileranges (IQR) and the whisker shows the last datum within the plusmn15IQR interval Significant differences were found for the median value ofthe features between the targeted groups (119875 lt 00001)

ShShSh

1

minus1

0

s filt(m

V)

Time (s)9630

(a) VF

NSh NShSh

Time (s)9630

1

minus1

0

s filt(m

V)

(b) VF

Time (s)9630

NSh NSh NSh

1

minus1

0

s filt(m

V)

(c) ORG

hShSN Sh

Time (s)9630

1

minus1

0

s filt(m

V)

(d) ASY

Figure 7 Examples of correctly and incorrectly classified 9 s segments Examples (a c) are correctly classified despite the presence of largefiltering residuals However in the VF of panel (b) spiky filtering artifacts cause the erroneous classification In the ASY of panel (d) filteringresiduals are large in the last two windows causing the shock diagnosis


Table 2 Final classification for the 3-s windows and 9-s segments of the test set compared to the AHA performance goals Sensitivitiesspecificities and low one-sided 90 CIs (in parenthesis) were obtained using GEE to adjust for clustering

Rhythm type 3-s window 9-s segment AHA goal [18]n SeSp n SeSp

Shockable 1866 897 (855) 622 910 (866) gt90 (for VF)Nonshockable 10050 951 (943) 3350 966 (959) gt95AS 3927 943 (931) 1309 965 (952) gt95ORG 6123 956 (946) 2041 967 (958) gt95

Table 3 Comparative assessment in terms of accuracy and the composition of the databases ( of ASY in nonshockable rhythm inparenthesis) between the method proposed in this study and previous methods tested on OHCA rhythms

Authors Method Accuracy Testing datasetsSe () Sp () Sh NSh

Eilevstjoslashnn et al [10] MC-RAMP 967 799 92 174 (30)Aramendi et al [19] LMS filter 954 863 87 285 (31)Tan et al [20] ART filter 921 905 114 4155 (NA)Li et al [5] Direct analysis 933 886 1256 964 (4)Krasteva et al [6] Direct analysis 901 861 172 721 (46)Proposed method Filtering + SAA 910 966 622 3350 (39)

to suppress the CPR artifact and an SAA optimized to analyzethe rhythm after filtering Our objective was to increase thespecificity because the low specificity of current methodshas restrained their implementation in current defibrillatorsOur results indicate that our new design approach mightcontribute to a substantial increase of the accuracy of rhythmanalysis methods during CPR with results that marginallymeet AHA performance goals

The design efforts were focused on obtaining a high speci-ficity during CPR to allow CCs to continue uninterrupteduntil the method gives a shock adviceThe positive predictivevalue (PPV) of the algorithm that is the confidence in ashock diagnosis must be kept high to avoid unnecessaryCPR interruptions if the underlying rhythm is nonshockableSince VF is the positive class the PPV depends on thesensitivityspecificity of the algorithm and on the prevalenceof VF 119875vf in the following way

PPV () = 100 times TPTP + FP

= 100 timesSe sdot 119875vf

Se sdot 119875vf + (1 minus Sp) sdot (1 minus 119875vf)

(8)

The exact prevalence of VF (reported for the initiallyobserved rhythm as stated in [33]) is unknown and variesamong OHCA studies with figures in the range of 23 to67 [34 35] For the original OHCA studies from which ourdatasets originated the prevalences of VF were 43 [21] and41 [22] within the previous range For the limits of the VFprevalence range the PPV of our algorithm is high in the889 to 982 range Furthermore since the PPV dependson the prevalences algorithms must be trained to optimizesensitivityspecificity with emphasis on a large specificity (aspecificity of 100 would result in a PPV of 100 regardlessof the prevalences)

To this date most methods for rhythm analysis duringCPR have focused on the accurate detection of shockablerhythms resulting in higher values for sensitivity than forspecificity Table 3 compares the accuracy of our method tothat of five well-known methods tested on OHCA data thatrepresent the twomost successful strategies for rhythm analy-sis during CPRThree of thosemethods are based on adaptivefilters [10 12 20] and the other two are algorithms designedto directly diagnose the corrupt ECG [5 6] Although thesensitivity of ourmethod is up to 4 points below that reportedby methods based on adaptive filters it is still above the valuerecommended by the AHA which ensures the detection of ahigh proportion of shockable rhythmsThe higher sensitivityof methods based on adaptive filters may be explained bythe fact that filtering residuals are frequently diagnosed asshockable by SAA designed to diagnose artifact-free ECG[14] In contrast the 966 specificity of our approach is animportant improvement with respect to previous approachesin which the specificity was below 91 We showed thatcombining the strong points of both approaches may resultin an increased accuracy

The characteristics of the OHCA data used in thesestudies may affect the sensitivityspecificity results and inparticular the characteristics of CPR the selection criteriafor VF and the proportion of ASY among nonshockablerhythms Rate and depth values of CPR in our data aresimilar to those reported in the original studies [21 22] andrepresent the wide range of CPR characteristics found in thefield In particular the CC rates are high (around 120 cpm)the spectral overlap with OHCA rhythms is therefore largeand suppressing the CPR artifact in our data should bechallenging [8] The CC depth was low even according tothe 2000 resuscitation guidelines and lower than the 5 cmrecommended in current guidelines [36] However no clear


correlation between depth and larger artifacts has beendemonstrated to date on human data Our database onlyincluded VF annotated as coarse as stated in the AHAstatement The three-phase model of cardiac arrest suggeststhat fine VF occurs when VF transitions from the electricphase into the circulatory or metabolic phases [37] Thereis no conclusive evidence that immediate defibrillation isthe optimal treatment in these latter phases of VF [38] sofrom a SAA design perspective it is a sound decision to onlyinclude coarseVFOn the other hand our database has a largeproportion of ASY among nonshockable rhythms (39)in agreement with the fact that ASY is the most frequentnonshockable OHCA rhythm [39]The high specificity of ourmethod for ASY is particularly important because ASY is themost difficult nonshockable rhythm to detect during CPR[14 16]

Our study shows that combining adaptive filtering withspecial SAAs that optimally diagnose the filtered ECG mayresult in an increased overall accuracy In addition thecomputational cost of the algorithm is low as shown by theprocessing time analysis The SAA algorithm computes atmost six ECG features and implementing our SVM in anAED requires only a few kilobytes of memory for the supportvectors and the computation of the discriminant function(see equation (5)) The LMS algorithm using 5 harmonicsinvolves only 10 coefficients [12] which substantially simpli-fies the filter In any case incorporating a CPR artifact filterto current AEDs is more complex than using algorithms thatdirectly analyze the corrupt ECG [5 6] Filtering techniquesbased on the CD signal require the use of external CPRquality devices [40 41] or modified defibrillation pads [4243] to record the acceleration signal Alternatively otherreference signals can be used such as the thoracic impedancerecorded through the defibrillation pads [19] CPR artifactfilters increase the complexity of the software and signalprocessing units of the AED and may even demand changesin its hardware to acquire reference signals

Finally several studies need to be completed beforeany method could be safely taken to the field First moreconclusive results require testing the algorithm on datarecorded by equipment different from those used for thisstudy and with CPR delivered according to the latest 2010CPR guidelines In addition retrospective studies based oncomplete resuscitation episodes should be conducted Inthis way the impact of using the method on CPR admin-istration could be evaluated This involves among otherthings a statistical evaluation of whether the method avoidsunnecessary CPR interruptions in nonshockable rhythmsand unnecessary CPR prolongations in shockable rhythms[36] The methodology for such an evaluation has recentlybeen developed [44]

5 Conclusions

This work introduces a new method for rhythm analysisduring CPR with a novel design approach aimed at obtaininga high specificity The method combines an adaptive LMS

filter to suppress the CPR artifact with a new shockno-shock classification method based on the analysis of thefiltered ECGThe method resulted in an increased specificityof 966 without compromising the sensitivity with overallperformance figures that met AHA requirements

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

Thiswork received financial support from SpanishMinisteriode Economıa y Competitividad (Projects TEC2012-31144 andTEC2012-31928) from the UPVEHU (unit UFI1116) andfrom the Basque government (Grants BFI-2010-174 BFI-2010-235 and BFI-2011-166)The authors would like to thankProfessor Rojo-Alvarez from the University Rey Juan Carlos(Madrid Spain) for his assistance with SVM classifiers andfor his thorough review of the paper

References

[1] J Berdowski R A Berg J G P Tijssen and R W KosterldquoGlobal incidences of out-of-hospital cardiac arrest and survivalrates systematic review of 67 prospective studiesrdquoResuscitationvol 81 no 11 pp 1479ndash1487 2010

[2] R O Cummins J P Ornato W H Thies et al ldquoImprovingsurvival from sudden cardiac arrest the ldquochain of survivalrdquoconceptrdquo Circulation vol 83 no 5 pp 1832ndash1847 1991

[3] C D Deakin J P Nolan J Soar et al ldquoEuropean ResuscitationCouncil Guidelines for resuscitation 2010 section 4 Adultadvanced life supportrdquo Resuscitation vol 81 no 10 pp 1305ndash1352 2010

[4] S Cheskes R H Schmicker J Christenson et al ldquoPerishockpause an independent predictor of survival from out-of-hospital shockable cardiac arrestrdquoCirculation vol 124 no 1 pp58ndash66 2011

[5] Y Li J Bisera F Geheb W Tang and M H Weil ldquoIdentifyingpotentially shockable rhythms without interrupting cardiopul-monary resuscitationrdquo Critical Care Medicine vol 36 no 1 pp198ndash203 2008

[6] V Krasteva I Jekova I Dotsinsky and J-P Didon ldquoShock advi-sory system for heart rhythm analysis during cardiopulmonaryresuscitation using a single ecg input of automated externaldefibrillatorsrdquo Annals of Biomedical Engineering vol 38 no 4pp 1326ndash1336 2010

[7] S Ruiz de Gauna U Irusta J Ruiz U Ayala E Aramendi andT Eftestoslashl ldquoRhythm analysis during cardiopulmonary resusci-tation past present and futurerdquoBioMedResearch Internationalvol 2014 Article ID 386010 13 pages 2014

[8] T Werther A Klotz M Granegger et al ldquoStrong corruptionof electrocardiograms caused by cardiopulmonary resuscita-tion reduces efficiency of two-channel methods for removingmotion artefacts in non-shockable rhythmsrdquo Resuscitation vol80 no 11 pp 1301ndash1307 2009


[9] S O Aase T Eftestoslashl J H Husoslashy K Sunde and P A SteenldquoCPR artifact removal from human ECG using optimal multi-channel filteringrdquo IEEETransactions on Biomedical Engineeringvol 47 no 11 pp 1440ndash1449 2000

[10] J Eilevstjoslashnn T Eftestoslashl S O Aase H Myklebust J H Husoslashyand P A Steen ldquoFeasibility of shock advice analysis duringCPR through removal of CPR artefacts from the human ECGrdquoResuscitation vol 61 no 2 pp 131ndash141 2004

[11] R D Berger J Palazzolo and H Halperin ldquoRhythm dis-crimination during uninterrupted CPR using motion artifactreduction systemrdquoResuscitation vol 75 no 1 pp 145ndash152 2007

[12] U Irusta J Ruiz S R de Gauna T Eftestoslashl and J Kramer-Johansen ldquoA least mean-Square filter for the estimation of thecardiopulmonary resuscitation artifact based on the frequencyof the compressionsrdquo IEEE Transactions on Biomedical Engi-neering vol 56 no 4 pp 1052ndash1062 2009

[13] K Rheinberger T Steinberger K Unterkofler M Baubin AKlotz and A Amann ldquoRemoval of CPR artifacts from theventricular fibrillation ECG by adaptive regression on laggedreference signalsrdquo IEEETransactions onBiomedical Engineeringvol 55 no 1 pp 130ndash137 2008

[14] J Ruiz U Irusta S Ruiz de Gauna and T Eftestoslashl ldquoCardiopul-monary resuscitation artefact suppression using a Kalman filterand the frequency of chest compressions as the reference signalrdquoResuscitation vol 81 no 9 pp 1087ndash1094 2010

[15] E Aramendi S R de Gauna U Irusta J Ruiz M F Arcochaand J M Ormaetxe ldquoDetection of ventricular fibrillationin the presence of cardiopulmonary resuscitation artefactsrdquoResuscitation vol 72 no 1 pp 115ndash123 2007

[16] S Ruiz deGauna J Ruiz U Irusta E Aramendi T Eftestoslashl andJ Kramer-Johansen ldquoA method to remove CPR artefacts fromhuman ECG using only the recorded ECGrdquo Resuscitation vol76 no 2 pp 271ndash278 2008

[17] Y Li and W Tang ldquoTechniques for artefact filtering from chestcompression corrupted ECG signals good but not enoughrdquoResuscitation vol 80 no 11 pp 1219ndash1220 2009

[18] R E Kerber L B Becker J D Bourland et al ldquoAutomaticexternal defibrillators for public access defibrillation recom-mendations for specifying and reporting arrhythmia analysisalgorithm performance incorporating new waveforms andenhancing safetyrdquoCirculation vol 95 no 6 pp 1677ndash1682 1997

[19] E Aramendi U Ayala U Irusta E Alonso T Eftestoslashl and JKramer-Johansen ldquoSuppression of the cardiopulmonary resus-citation artefacts using the instantaneous chest compressionrate extracted from the thoracic impedancerdquo Resuscitation vol83 no 6 pp 692ndash698 2012

[20] Q Tan G A Freeman F Geheb and J Bisera ldquoElectrocardi-ographic analysis during uninterrupted cardiopulmonaryresuscitationrdquo Critical Care Medicine vol 36 no 11 ppS409ndashS412 2008

[21] L Wik J Kramer-Johansen H Myklebust et al ldquoQuality ofcardiopulmonary resuscitation during out-of-hospital cardiacarrestrdquo Journal of the American Medical Association vol 293no 3 pp 299ndash304 2005

[22] J Kramer-Johansen H Myklebust L Wik et al ldquoQuality ofout-of-hospital cardiopulmonary resuscitation with real timeautomated feedback a prospective interventional studyrdquo Resus-citation vol 71 no 3 pp 283ndash292 2006

[23] D L Atkins W A Scott A D Blaufox et al ldquoSensitivityand specificity of an automated external defibrillator algorithmdesigned for pediatric patientsrdquo Resuscitation vol 76 no 2 pp168ndash174 2008

[24] A C Clifford ldquoComparative assessment of shockable ECGrhythm detection algorithms in automated external defibrilla-torsrdquo Resuscitation vol 32 no 3 pp 217ndash225 1996

[25] W Zong G B Moody and D Jiang ldquoA robust open-sourcealgorithm to detect onset and duration of QRS complexesrdquo inProceedings of the IEEE Computers in Cardiology pp 737ndash740September 2003

[26] C G Brown and R Ozwonczyk ldquoSignal analysis of the humanelectrocardiogram during ventricular fibrillation frequencyand amplitude parameters as predictors of successful counter-shockrdquoAnnals of EmergencyMedicine vol 27 no 2 pp 184ndash1881996

[27] K Minami H Nakajima and T Toyoshima ldquoReal-time discr-imination of ventricular tachyarrhythmia with Fourier-trans-form neural networkrdquo IEEE Transactions on Biomedical Engi-neering vol 46 no 2 pp 179ndash185 1999

[28] A Ben-Hur and J Weston ldquoA userrsquos guide to support vectormachinesrdquoDataMining Techniques for the Life Sciences vol 609pp 223ndash239 2010

[29] C-C Chang and C-J Lin ldquoLIBSVM a Library for supportvector machinesrdquo ACM Transactions on Intelligent Systems andTechnology vol 2 no 3 article 27 2011

[30] S L Zeger and K Y Liang ldquoLongitudinal data analysis fordiscrete and continuous outcomesrdquoBiometrics vol 42 no 1 pp121ndash130 1986

[31] M R Sternberg and A Hadgu ldquoA GEE approach to estimatingsensitivity and specificity and coverage properties of the con-fidence intervalsrdquo Statistics in Medicine vol 20 no 9-10 pp1529ndash1539 2001

[32] U Halekoh S Hoslashjsgaard and J Yan ldquoThe R package geepackfor generalized estimating equationsrdquo Journal of StatisticalSoftware vol 15 no 2 pp 1ndash11 2006

[33] I Jacobs V Nadkarni J Bahr et al ldquoCardiac arrest and cardio-pulmonary resuscitation outcome reports update and simpli-fication of the Utstein templates for resuscitation registriesA statement for healthcare professionals from a task force ofthe international liaison committee on resuscitation (AmericanHeart Association European Resuscitation Council AustralianResuscitation Council New Zealand Resuscitation CouncilHeart and Stroke Foundation of Canada InterAmerican HeartFoundation Resuscitation Council of Southern Africa)rdquo Resus-citation vol 63 no 3 pp 233ndash249 2004

[34] C AtwoodM S Eisenberg J Herlitz and T D Rea ldquoIncidenceof EMS-treated out-of-hospital cardiac arrest in Europerdquo Resus-citation vol 67 no 1 pp 75ndash80 2005

[35] G Nichol E Thomas C W Callaway et al ldquoRegional varia-tion in out-of-hospital cardiac arrest incidence and outcomerdquoJournal of the AmericanMedical Association vol 300 no 12 pp1423ndash1431 2008

[36] M R Sayre RWKosterM Botha et al ldquoPart 5 Adult basic lifesupport 2010 International Consensus on CardiopulmonaryResuscitation andEmergencyCardiovascularCare SciencewithTreatment Recommendationsrdquo Circulation vol 122 no 16supplement 2 pp S298ndashS324 2010

[37] M L Weisfeldt and L B Becker ldquoResuscitation after cardiacarrest a 3-phase time-sensitive modelrdquo Journal of the AmericanMedical Association vol 288 no 23 pp 3035ndash3038 2002

[38] J P Freese D B Jorgenson P Y Liu et al ldquoWaveform analysis-guided treatment versus a standard shock-first protocol forthe treatment of out-of-hospital cardiac arrest presenting inventricular fibrillation results of an international randomizedcontrolled trialrdquo Circulation vol 128 no 9 pp 995ndash1002 2013


[39] L A Cobb C E Fahrenbruch M Olsufka and M K CopassldquoChanging incidence of out-of-hospital ventricular fibrillation1980ndash2000rdquo Journal of the American Medical Association vol288 no 23 pp 3008ndash3013 2002

[40] C N Pozner A Almozlino J Elmer S Poole D McNa-mara and D Barash ldquoCardiopulmonary resuscitation feedbackimproves the quality of chest compression provided by hospitalhealth care professionalsrdquo American Journal of EmergencyMedicine vol 29 no 6 pp 618ndash625 2011

[41] G D Perkins R P Davies S Quinton et al ldquoThe effect ofreal-time CPR feedback and post event debriefing on patientand processes focused outcomes a cohort study trial protocolrdquoScandinavian Journal of Trauma Resuscitation and EmergencyMedicine vol 19 article 58 2011

[42] C F Babbs A E Kemeny W Quan and G Freeman ldquoA newparadigm for human resuscitation research using intelligentdevicesrdquo Resuscitation vol 77 no 3 pp 306ndash315 2008

[43] K G Monsieurs M de Regge K Vansteelandt et al ldquoExcessivechest compression rate is associated with insufficient compres-sion depth in prehospital cardiac arrestrdquo Resuscitation vol 83no 11 pp 1319ndash1323 2012

[44] J Ruiz U Ayala S Ruiz de Gauna et al ldquoDirect evaluationof the effect of filtering the chest compression artifacts on theuninterrupted cardiopulmonary resuscitation timerdquo AmericanJournal of Emergency Medicine vol 31 no 6 pp 910ndash915 2013

Submit your manuscripts athttpwwwhindawicom

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinsonrsquos Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom


the performance of these methods researchers first filteredthe CPR artifact and then analyzed the rhythm using aSAA to obtain the sensitivity and specificity of the methodthat is the proportion of correctly diagnosed shockable andnonshockable rhythms respectively However the SAAs usedwere originally designed to analyze artifact-free ECG insteadof the ECG after filtering

Currently rhythm analysis during CPR is not possible[17] Most methods have sensitivity above 90 the mini-mum value recommended by the American Heart Associ-ation (AHA) for SAA on artifact-free ECG [18] Howeverspecificity rarely exceeds 85 well below the 95 valuerecommended by the AHA A low specificity would result ina large number of false shock diagnoses during CPR whichwould unnecessarily increase the number of interruptionsin CPR Overall the main cause of the low specificity isfiltering residuals in nonshockable rhythms These residualsfrequently resemble a disorganized rhythm [10 12] and areoftenmisdiagnosed as shockable by SAAs designed to analyzeartifact-free ECG This problem is more prominent whenthe electrical activity of the underlying heart rhythm islow particularly for asystole (ASY) [14 16] because filteringresiduals may have amplitudes comparable or larger thanthose of the underlying ECG

In this study we explore the possibility of combiningadaptive filtering techniques with a SAA designed to opti-mally classify the rhythm after filteringThe aim is to improvethe accuracy of current approaches and in particular toovercome the low specificity When compared to previousstudies our results showed an increased specificity withoutcompromising the sensitivity for a comprehensive dataset ofOHCA rhythms

2 Materials and Methods

21 Data Collection The data for this study were extractedfrom a large prospective study of OHCA conducted between2002 and 2004 in three European sites [21 22] CPR wasdelivered by trained ambulance personnel in adherence to the2000 resuscitation guidelines Episodes were recorded usingmodified Laerdal Heartstart 4000 defibrillators (4000SP) andan external CPR assist pad to acquire additional referencesignals All signals were acquiredwith a 500Hz sampling rateThe initial rhythmand all subsequent changes in rhythmwereannotated by consensus of an experienced anesthesiologistand a biomedical engineer both specialized in resuscitation[21 22] Rhythm annotations comprised five types (see [21]for further details) VF and fast ventricular tachycardia(VT) in the shockable category and ASY pulseless electricalactivity (PEA) and pulse-generating rhythm (PR) in thenonshockable category Intervals with chest compressionswere annotated using the compression depth (CD) obtainedfrom the CPR assist pad

For this study specific records containing the ECG andCD signals were automatically extracted from the originalepisodes First rhythm transitions were identified using theoriginal annotations and then for each interval withoutrhythm transitions at most one record was extracted to avoid

bias due to data selection Records were extracted if thefollowing criteria were met duration of more than 20-songoing CCs and the same rhythm annotation before andafter CCs Following the AHA statement the records weregrouped into a shockable and a nonshockable category Theamplitude thresholds adopted for coarse VF and ASY arethose accepted in the literature on SAAs [6 18]The followingcriteria and rhythm definitions were checked in the cleanintervals before and after CCs

Shockable RhythmsThis category includes fast VT with rateabove 150 beats per minute (bpm) and coarse VF Coarse VFwas defined asVFwith peak-to-peak amplitude above 200 120583Vand a fibrillation frequency above 2Hz

Nonshockable Rhythms These rhythms were further dividedinto the following

(i) organized rhythms (ORG) all nonshockable rhythmsexcept ASY (PEA and PR)

(ii) asystole (ASY) rhythms with peak-to-peak ampli-tudes below 100 120583V for at least 2-s

All signals were resampled to 119891119904= 250Hz a sampling rate

similar to that used by commercial AEDs Inwhat follows thesample index and time variables are related by 119905 = 119899 sdot119879

119904(119879119904=

1119891119904) The ECG was band limited to 05ndash30Hz (order 10

Butterworth filter) a typical ECG monitor bandwidth usedin AEDs [5 6] which removes base line wander and highfrequency noise

Following standard practice in SAA design the rhythmanalysis method was designed to analyze three consecutive3 s windows so it gives a diagnosis every 9 s [23 24] A 3 swindow is sufficient to characterize the rhythm in terms ofrate stability and morphology and to make a shock (Sh)or no-shock (NSh) decision [23] SAA algorithms combineseveral consecutive diagnoses to avoid errors due to rhythmtransitions and to avoid shock diagnoses for short burstsof nonsustained VT Therefore each record was dividedinto nonoverlapping 9 s segments The 9 s segments wererandomly split into two separate sets one to train thealgorithm and an independent set to test the algorithm asrequired by the AHA statement In addition we made surethat the patients on both sets were different (AHA statement)and that the distribution of rhythm types was similar in bothsets

22 Rhythm Analysis Method The block diagram of therhythm analysis method is shown in Figure 1 First a CPRartifact suppression filter estimates the underlying rhythmthat is the filtered ECG signal 119904filt Then a SAA diagnosesevery 3 s window of the filtered signalThe SAA is designed tooptimally classify the filtered signal and is further composedof two sequential subalgorithms (1) a detector of rhythmswith low electrical activity (LEA) that is nonshockablerhythmswithout distinguishableQRS complexes such as ASYor idioventricular rhythms and (2) a ShNSh algorithm thatclassifies windows with electrical activity as shockable ornonshockable


++

CD Compdetector

Modelartifact

scor

scpr

LMS filter

sfiltminus

LEAdetector

ShNShalgorithm


Sh

NSh NSh



11

minus11

0

s cor(m

V)

11

minus11

0

s cpr(m

V)

11

minus11

0

s filt(m

V)

minus6

minus3

0

Time (s)12960

CD (c

m)



119896instants)





119896(119899) and phases


following model

119904cpr (119899) =5

sum

119896=1

119888119896(119899) cos (119896 sdot 2120587119891

119900(119899) sdot 119899 + 120601

119896(119899))

119891119900(119899) =

1

119905119896minus 119905119896minus1

for 119905119896minus1

le 119899119879119904lt 119905119896

(1)





119875LEA = sum119899

1199042

LEA (119899) (2)


119871119896=

(119896+1)1198911199042

sum

119899=1198961198911199042+1

radicΔ1199042LEA (119899) + 1198792

119904 (3)




03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

PLEA = 009 Lmin = 018

(a) ASY

PLEA = 274 Lmin = 135

03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

(b) VF




119889filt (119899) =1

119873

119873minus1

sum

119896=0


(4)









(above 12Hz)



(x1 1199101) (x





119891 (x) =119873119904

sum

119894=1


119894

10038171003817100381710038172

) + 119887 (5)






04

minus04

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0139

dfil

t(nu)

nP = 19

(a)

14

minus14

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0014

dfil

t(nu)

nP = 7

(b)


04

minus04

0

s filt(m

V)

Time (s)3210

0


004

Pss(n u)

Pfib = 084 Ph = 001

(a) VF

1

minus1

0

s filt(m

V)

Time (s)3210

0

002

Pss( nu)


Pfib = 050 Ph = 012

(b) ORG


ℎ(in green)















3 Results




119875LEA lt 044 or 119871min lt 063 (7)







4 Discussion



0

4

8

12

ASY Sh

(a) 119875LEA

ASY Sh

0

1

2

3

(b) 119871min

000

006

012

018

ORG Sh

(c) 119887119878

ORG Sh

0

12

24

36

(d) 119899119875

ORG Sh

000

033

067

100

(e) 119875fib

ORG Sh

000

012

024

036

(f) 119875ℎ


ShShSh

1

minus1

0

s filt(m

V)

Time (s)9630

(a) VF

NSh NShSh

Time (s)9630

1

minus1

0

s filt(m

V)

(b) VF

Time (s)9630

NSh NSh NSh

1

minus1

0

s filt(m

V)

(c) ORG

hShSN Sh

Time (s)9630

1

minus1

0

s filt(m

V)

(d) ASY














(8)








5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















++

CD Compdetector

Modelartifact

scor

scpr

LMS filter

sfiltminus

LEAdetector

ShNShalgorithm


Sh

NSh NSh



11

minus11

0

s cor(m

V)

11

minus11

0

s cpr(m

V)

11

minus11

0

s filt(m

V)

minus6

minus3

0

Time (s)12960

CD (c

m)



119896instants)





119896(119899) and phases


following model

119904cpr (119899) =5

sum

119896=1

119888119896(119899) cos (119896 sdot 2120587119891

119900(119899) sdot 119899 + 120601

119896(119899))

119891119900(119899) =

1

119905119896minus 119905119896minus1

for 119905119896minus1

le 119899119879119904lt 119905119896

(1)





119875LEA = sum119899

1199042

LEA (119899) (2)


119871119896=

(119896+1)1198911199042

sum

119899=1198961198911199042+1

radicΔ1199042LEA (119899) + 1198792

119904 (3)




03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

PLEA = 009 Lmin = 018

(a) ASY

PLEA = 274 Lmin = 135

03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

(b) VF




119889filt (119899) =1

119873

119873minus1

sum

119896=0


(4)









(above 12Hz)



(x1 1199101) (x





119891 (x) =119873119904

sum

119894=1


119894

10038171003817100381710038172

) + 119887 (5)






04

minus04

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0139

dfil

t(nu)

nP = 19

(a)

14

minus14

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0014

dfil

t(nu)

nP = 7

(b)


04

minus04

0

s filt(m

V)

Time (s)3210

0


004

Pss(n u)

Pfib = 084 Ph = 001

(a) VF

1

minus1

0

s filt(m

V)

Time (s)3210

0

002

Pss( nu)


Pfib = 050 Ph = 012

(b) ORG


ℎ(in green)















3 Results




119875LEA lt 044 or 119871min lt 063 (7)







4 Discussion



0

4

8

12

ASY Sh

(a) 119875LEA

ASY Sh

0

1

2

3

(b) 119871min

000

006

012

018

ORG Sh

(c) 119887119878

ORG Sh

0

12

24

36

(d) 119899119875

ORG Sh

000

033

067

100

(e) 119875fib

ORG Sh

000

012

024

036

(f) 119875ℎ


ShShSh

1

minus1

0

s filt(m

V)

Time (s)9630

(a) VF

NSh NShSh

Time (s)9630

1

minus1

0

s filt(m

V)

(b) VF

Time (s)9630

NSh NSh NSh

1

minus1

0

s filt(m

V)

(c) ORG

hShSN Sh

Time (s)9630

1

minus1

0

s filt(m

V)

(d) ASY














(8)








5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

PLEA = 009 Lmin = 018

(a) ASY

PLEA = 274 Lmin = 135

03

minus03

0

s filt(m

V)

Time (s)3210

03

minus03

0

Time (s)3210

s LEA

(mV)

(b) VF




119889filt (119899) =1

119873

119873minus1

sum

119896=0


(4)









(above 12Hz)



(x1 1199101) (x





119891 (x) =119873119904

sum

119894=1


119894

10038171003817100381710038172

) + 119887 (5)






04

minus04

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0139

dfil

t(nu)

nP = 19

(a)

14

minus14

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0014

dfil

t(nu)

nP = 7

(b)


04

minus04

0

s filt(m

V)

Time (s)3210

0


004

Pss(n u)

Pfib = 084 Ph = 001

(a) VF

1

minus1

0

s filt(m

V)

Time (s)3210

0

002

Pss( nu)


Pfib = 050 Ph = 012

(b) ORG


ℎ(in green)















3 Results




119875LEA lt 044 or 119871min lt 063 (7)







4 Discussion



0

4

8

12

ASY Sh

(a) 119875LEA

ASY Sh

0

1

2

3

(b) 119871min

000

006

012

018

ORG Sh

(c) 119887119878

ORG Sh

0

12

24

36

(d) 119899119875

ORG Sh

000

033

067

100

(e) 119875fib

ORG Sh

000

012

024

036

(f) 119875ℎ


ShShSh

1

minus1

0

s filt(m

V)

Time (s)9630

(a) VF

NSh NShSh

Time (s)9630

1

minus1

0

s filt(m

V)

(b) VF

Time (s)9630

NSh NSh NSh

1

minus1

0

s filt(m

V)

(c) ORG

hShSN Sh

Time (s)9630

1

minus1

0

s filt(m

V)

(d) ASY














(8)








5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















04

minus04

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0139

dfil

t(nu)

nP = 19

(a)

14

minus14

0

s filt(m

V)

Time (s)3210

0

1

05

Time (s)3210

bS

bS

= 0014

dfil

t(nu)

nP = 7

(b)


04

minus04

0

s filt(m

V)

Time (s)3210

0


004

Pss(n u)

Pfib = 084 Ph = 001

(a) VF

1

minus1

0

s filt(m

V)

Time (s)3210

0

002

Pss( nu)


Pfib = 050 Ph = 012

(b) ORG


ℎ(in green)















3 Results




119875LEA lt 044 or 119871min lt 063 (7)







4 Discussion



0

4

8

12

ASY Sh

(a) 119875LEA

ASY Sh

0

1

2

3

(b) 119871min

000

006

012

018

ORG Sh

(c) 119887119878

ORG Sh

0

12

24

36

(d) 119899119875

ORG Sh

000

033

067

100

(e) 119875fib

ORG Sh

000

012

024

036

(f) 119875ℎ


ShShSh

1

minus1

0

s filt(m

V)

Time (s)9630

(a) VF

NSh NShSh

Time (s)9630

1

minus1

0

s filt(m

V)

(b) VF

Time (s)9630

NSh NSh NSh

1

minus1

0

s filt(m

V)

(c) ORG

hShSN Sh

Time (s)9630

1

minus1

0

s filt(m

V)

(d) ASY














(8)








5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of






















3 Results




119875LEA lt 044 or 119871min lt 063 (7)







4 Discussion



0

4

8

12

ASY Sh

(a) 119875LEA

ASY Sh

0

1

2

3

(b) 119871min

000

006

012

018

ORG Sh

(c) 119887119878

ORG Sh

0

12

24

36

(d) 119899119875

ORG Sh

000

033

067

100

(e) 119875fib

ORG Sh

000

012

024

036

(f) 119875ℎ


ShShSh

1

minus1

0

s filt(m

V)

Time (s)9630

(a) VF

NSh NShSh

Time (s)9630

1

minus1

0

s filt(m

V)

(b) VF

Time (s)9630

NSh NSh NSh

1

minus1

0

s filt(m

V)

(c) ORG

hShSN Sh

Time (s)9630

1

minus1

0

s filt(m

V)

(d) ASY














(8)








5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















0

4

8

12

ASY Sh

(a) 119875LEA

ASY Sh

0

1

2

3

(b) 119871min

000

006

012

018

ORG Sh

(c) 119887119878

ORG Sh

0

12

24

36

(d) 119899119875

ORG Sh

000

033

067

100

(e) 119875fib

ORG Sh

000

012

024

036

(f) 119875ℎ


ShShSh

1

minus1

0

s filt(m

V)

Time (s)9630

(a) VF

NSh NShSh

Time (s)9630

1

minus1

0

s filt(m

V)

(b) VF

Time (s)9630

NSh NSh NSh

1

minus1

0

s filt(m

V)

(c) ORG

hShSN Sh

Time (s)9630

1

minus1

0

s filt(m

V)

(d) ASY














(8)








5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




























(8)








5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




















5 Conclusions





Acknowledgments


References




















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of



























































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















research article a reliable method for rhythm analysis

Documents