Non-invasive assessment of the atrial cycle length during atrial fibrillation in man: introducing,...

Ž .Cardiovascular Research 38 1998 69–81

Non-invasive assessment of the atrial cycle length during atrialfibrillation in man: introducing, validating and illustrating

a new ECG method

Magnus Holm a,), Steen Pehrson a, Max Ingemansson a, Leif Sornmo c, Rolf Johansson a,¨Lennart Sandhall b, Max Sunemark a, Birgit Smideberg a, Christian Olsson a, S. Bertil Olsson a

a Department of Cardiology, UniÕersity Hospital, S-221 85 Lund, Swedenb Department of Radiology, UniÕersity Hospital, Lund, Sweden

c Department of Applied Electronics, Lund UniÕersity, Lund, Sweden

Received 9 June 1997; accepted 10 November 1997

Abstract

Ž .Objectives: Atrial fibrillation AF in man has previously been shown to include a wide variety of atrial activity. Assessment of thecharacteristics of this arrhythmia with a commonly applicable tool may therefore be important in the choice and evaluation of differenttherapeutic strategies. As the AF cycle length has been shown to correlate locally with atrial refractoriness and globally with the degree ofatrial organization, with, in general, shorter cycle length during apparently random AF compared to more organized AF, we have

Ž .developed a new method for non-invasive assessment of the AF cycle length using the surface and the esophagus ESO ECG. Methodsand Results: From the frequency spectrum of the residual ECG, created by suppression of the QRST complexes, the dominant atrial

Ž .cycle length DACL was derived. By comparison with multiple intracardiac simultaneously acquired right and left AF cycle lengths inpatients with paroxysmal AF, we found that the DACL in lead V1, ranging from 130 to 185 ms, well represented a spatial average of theright AF cycle lengths, whereas the DACL in the ESO ECG, ranging from 140 to 185 ms, reflected both the right and the left AF cyclelength, where the influence from each structure depended on the atrial anatomy of the individual, as determined by MRI. In patients withchronic AF, the method was capable of following changes in the AF cycle length due to administration of D,L-sotalol and 5 min of ECGrecording was sufficient for the DACL to be reproducible. Conclusions: We conclude that this new non-invasive method, named

Ž .‘Frequency Analysis of Fibrillatory ECG’ FAF-ECG , is capable of assessing both the magnitude and the dynamics of the atrialfibrillation cycle length in man. q 1998 Elsevier Science B.V.

Keywords: Human; Atrium; Fibrillation; Cycle length; ECG; Frequency analysis; Non-invasive technique

1. Introduction

Ž .Atrial fibrillation AF is an arrhythmia that both in itsexperimental and in its clinical form exhibits a wide

w xvariety of atrial activity 1–8 . Atrial fibrillatory activityhas previously been explored by means of the invasivemapping technique, yielding knowledge on the number and

) Ž .Corresponding author. Fax: q46 46 157857; E-mail:[email protected]

propagation of activation waves, the atrial cycle length andw xthe atrial conduction velocity 3–5,8,9 . As some of these

studies have revealed that the global AF cycle lengthreflects the organization of the fibrillation and, further, thatthe local AF cycle length correlates to atrial refractoriness,access to the AF cycle length may be of importance for anunderstanding of the basic electrophysiological propertiesof this arrhythmia and for choosing and evaluating thera-

Time for primary review 37 days.

0008-6363r98r$19.00 q 1998 Elsevier Science B.V. All rights reserved.Ž .PII S0008-6363 97 00289-7

by guest on April 7, 2014

http://cardiovascres.oxfordjournals.org/D

ownloaded from

http://cardiovascres.oxfordjournals.org/


( )M. Holm et al.rCardioÕascular Research 38 1998 69–8170

w xpeutic interventions 4,10–13 . However, so far, only inva-sive techniques have been available for assessing the AFcycle length, a circumstance that has limited the use of thisvariable for practical and ethical reasons.

It has earlier been recognized that the surface ECGcontains information on the electrical and mechanical func-tion of the atria during fibrillation. In 1915, Hewlett andWilson distinguished between fine and coarse AF in man,based on the coarseness of the venous pulse tracings andthe oscillations of the baseline of the surface ECG during

w xdiastole 14 . More recent studies have addressed the corre-lation between the amplitude of the fibrillatory waves,called f-waves, of the ECG and the electrical events in the

w xatria and echocardiographic findings, respectively 15–19 .As detection and reliable time annotation of the f-wavesare difficult, Slocum and colleagues used the power spec-trum of the residual ECG, created by suppressing theventricular activity on the surface ECG, to discriminatebetween AF and other atrial rhythms and also followedchanges in the atrial fibrillatory rate due to interventionw x20,21 . Prompted by a belief that an estimate of the actualAF cycle length can be extracted from the f-waves in theECG, we have, using a similar technique as Slocum et al.,developed a new method named ‘Frequency Analysis of

Ž .Fibrillatory ECG’ FAF-ECG . As already preliminarilyreported, we have shown that it is indeed possible to assessboth the magnitude and the dynamics of the AF cycle

w xlength with this new spectral method of analysis 10 . Thepurpose of this paper is to describe the FAF-ECG methodin detail and the procedure of validating the method and toillustrate its applicability, advantages and drawbacks.

2. FAF-ECG method

This investigation conforms with the principles outlinedin the Declaration of Helsinki.

2.1. Data acquisition

ECGs were recorded, either from the body surface orŽ .the esophagus ESO , in patients in AF. The ECG was

digitized at a sampling rate of 1 kHz with an amplituderesolution of 0.6 mV using 16-bit ArD conversionŽequipment supplied by Siemens-Elema AB, Solna, Swe-

.den . The acquisition equipment was connected to anIBM-compatible personal computer and the data werestored on removable disc media.

2.2. Data analysis

Data analysis was performed on a lead-by-lead basisusing software developed by ourselves. To provide for

reliable QRS classification and subsequently QRST sup-pression, low-frequency components below 0.6 Hz wereattenuated using a linear-phase, high-pass filter. This filter-ing procedure was implemented by forwardrbackwardprocessing of the signal using a sixth-order Butterworthfilter. The classification of QRS morphology and the align-ment of beats were based on a cross-correlation procedureusing an automatically defined template beat. Time align-ment was done by shifting each new beat relative to thetemplate beat until the highest cross-correlation value wasfound.

Spectral estimation of the AF cycle length requires thatsuppression of QRST complexes is performed before theatrial activity can be detected. Suppression of the QRSTcomplexes was done by first creating a signal in which anaverage of the time-aligned, QRST complexes was posi-tioned at the location of each QRST complex. This signalwas then subtracted from the original ECG signal and theresidual signal was used for further spectral analysis. Mea-sures were taken such that the transitions between succes-

Žsive QRST averages were smooth interpolation was used.to fill in gaps between successive beats . It should be noted

that the QRST subtraction relies on the fact that atrialactivity is unsynchronized to ventricular activity.

As previous studies have shown that the atrial cyclew xlength during human AF rarely goes below 100 ms 4,8 ,

corresponding to an atrial rate of 10 Hz, we reduced thesampling rate before spectral analysis of the residual signalin order to speed up the analysis process. Reduction was

Ždone by resampling the residual signal after low-passfiltering with an eighth-order Chebyshev type I filter with

.cut-off frequency 80 Hz to 200 Hz. Thereafter, the Welchperiodogram was computed using 2048 points and a Han-ning window length of 512 samples with an intervaloverlap of 128 samples, giving a frequency resolutionbetter than 0.1 Hz and a peak distinction resolution ofapproximately 0.5 Hz. Various lengths of the window andvarious interval overlaps were tested, as described below,before the above values were chosen as a compromisebetween the ability to separate very closely located fre-quency peaks and low variance in the spectral estimate.

Two statistics were derived from the frequency spec-Ž .trum. 1 The distribution in the 3- to 12-Hz range, which

was named unimodal if one frequency component waspresent and multimodal if two or more frequency compo-nents were present. To be regarded as a frequency compo-nent, the peak magnitude had to be a local maximum. Onlyfrequency components with a peak magnitude of at least50% of the maximum magnitude in the 3–12 Hz intervalwere considered. If the maximum signal power in the 3–12

2 ŽHz range was below 5 mV rHz as a consequence of very.low amplitude of the f-waves , normally no single fre-

quency components could be distinguished, and the analy-Ž .sis was regarded as a failure. 2 The peak frequency of

each spectral component in the 3–12 Hz range was con-Ž .verted to a cycle length cycle lengths1rfrequency and



ownloaded from



( )M. Holm et al.rCardioÕascular Research 38 1998 69–81 71

Fig. 1. The different steps of the new FAF-ECG method. First, lowfrequency components below 0.6 Hz were attenuated using a linear-phase,sixth-order Butterworth high-pass filter, implemented by forwardrback-ward processing of the signal. Next, classification of QRS morphologyand the alignment of beats was performed based on a cross-correlation

Ž .procedure 1 . Thereafter, an average of the time-aligned QRST com-Ž .plexes was created and subtracted from the original ECG signal 2, 3 .

Finally, a frequency spectrum of the residual ECG signal was estimatedusing FFT, the dominant frequency components in the 3- to 12-Hz rangewere identified and the peak frequency of these was converted to a cycle

Ž . Ž .length and named dominant atrial cycle length DACL 4 . A moredetailed description of the method is given in the text.

Ž .named ‘dominant atrial cycle length’ DACL . It should benoted that in cases of multimodal frequency distributions,there was one DACL for each frequency component. TheDACLs were rounded off to the nearest multiple of 5 msŽ .for ease of interpretation . The different steps of themethod are illustrated in Fig. 1.

3. Validation of the FAF-ECG method

3.1. Material

A total of 23 patients, 5 females and 18 males with aŽ .mean age of 61"12 years range 42–77 years , were

included in the validation procedure. Patients 1–8 under-went an invasive electrophysiological study, where ECGand intracardiac electrograms were acquired simultane-ously for the purpose of comparing intracardiac AF cycle

lengths with non-invasively assessed AF cycle length us-ing the FAF-ECG method. The electrophysiological studywas performed as part of a clinical routine for identifica-tion of possible treatable causes of the AF. Patients 9–13underwent intervention, i.e. rapid intravenous injection ofD,L-sotalol, to evaluate the ability of the FAF-ECG methodto monitor changes in the AF cycle length, and patients14–23 underwent a 1-h ECG recording to determine thereproducibility of the FAF-ECG method. Intervention withD,L-sotalol was done as an attempt to cardiovert the pa-tients pharmacologically. Eleven of the patients had parox-

Ž . Ž .ysmal AF PAF and 12 had chronic AF CAF . Allantiarrhythmic drugs were withdrawn for at least 5 half-lives before the study. The clinical characteristics aresummarized in Table 1.

3.2. Methods

3.2.1. InÕasiÕe electrophysiological studyUnder fluoroscopic guidance, a 10-polar catheter with

Ž .5-mm spacing was placed in the coronary sinus CS withthe most proximal pole in the orifice of the CS and a

Ž .20-polar catheter Cordis Webster Deflectable Halo in theŽ .right atrium with the distal pole 1 in the low lateral right

Ž .atrium LLRA , poles 7 and 8 in the high lateral rightŽ .atrium HLRA and poles 13 and 14 in the high septal

Ž .right atrium HSRA . Furthermore, a 4-polar catheter wasŽ .placed in the right atrial appendage RAA and a 2-polar

Ž .electrode Medtronic 6992A, interpolar distance 20 mm inŽ . Ž .the esophagus ESO Fig. 2 . The ESO catheter was

Ž .placed with the distal pole ESO just above the CSdistŽ .catheter as seen in the right anterior oblique RAO 308

Ž .and the left anterior oblique LAO 308 fluoroscopic pro-Ž .jections. The pair of poles of the CS catheter CS thatprox

gave a bipolar electrogram with simultaneous local activa-tions compared to the bipolar esophageal electrograms waschosen for further analysis. In patients in AF, only thefluoroscopic criterion was used. Furthermore, a standard12-lead surface ECG was connected to the patient. Allcatheters and electrodes were motivated by the clinicalprocedure.

If the patient was in sinus rhythm, burst pacing atHLRA or CS was performed down to a cycle length of 120ms until AF was induced. Two 10-min recordings weremade with a 5-min interval during AF. The first recordingconsisted of the 12-lead surface ECG, one unipolar ESO

Ž .ECG ESO and two unipolar electrograms from CSdist prox

and HLRA or RAA, respectively. The second recordingconsisted of the 12-lead surface ECG and 5 unipolarelectrograms from LLRA, HLRA, HSRA, RAA and CS ,prox

respectively. All unipolar signals was referred to the Wil-son Central Terminal, i.e. the mean potential of the leftarm, the right arm and the left leg.

The FAF-ECG method was applied to all signals in thetwo recordings, i.e. not only to the ECG signals as required



ownloaded from




Table 1Patient characteristics

Type of investigation Patient Sex Age Heart disease Type of arrhythmia Duration of arrhythmiaŽ . Ž .MrF years

Invasive 1 M 56 None PAF 12 years2 M 48 None PAF 8 years3 M 50 None PAF 14 years4 F 56 None PAF 3 years5 M 42 None PAF 7 years6 M 47 None PAF 2 years7 M 62 None PAF 12 years8 M 45 None PAF 18 years

Intervention 9 M 71 HT PAF 1.5 years10 M 61 None PAF 1.5 years11 F 50 None CAF 1.5 years12 M 67 None PAF 1.5 months13 M 48 HT, HF, CMP CAF 2 years

1-h ECG 14 M 77 HT CAF )1 year15 F 74 HF CAF 3 months16 F 76 IHD PAF 3 months17 M 77 None CAF 1 year18 M 61 None CAF 2 years19 M 68 IHD CAF 5 months20 M 70 None CAF 13 months21 M 55 None CAF )1 year22 F 77 HT CAF 6 months23 M 67 None PAF 1.5 months

Fs female; Msmale; HTshypertension; HFsheart failure; IHDs ischemic heart disease; CMPscardiomyopathy; PAFsparoxysmal atrial fibrilla-tion; CAFschronic atrial fibrillation.

by the method, but also to the intracardiac electrograms toallow direct comparison between the DACLs in the periph-eral and the intracardiac signals. The validity of thiscomparison was demonstrated by comparing, in one 30-s

Ž .strip of electrogram RAA or HLRA from each patient,the DACL determined using the FAF-ECG method withthe median AF cycle length determined manually, usingthe intrinsic negative deflection in the unipolar electro-

grams as the times of local activation. The difference,Ž .0"2.5 ms mean"s.d. , ranging from y2.7 to 3.8 ms,

was not significant.Of the 12 surface ECG leads, the precordial lead, V1,

was chosen for the comparison between the peripheral andthe intracardiac DACLs as the f-waves in all patients weremost prominent in this lead and, anatomically, the V1electrode is considered to be closest to the right atrium.

Fig. 2. The position of the electrodes and the catheters used in patients undergoing an invasive electrophysiological study. In the left panel, a frontalŽ .fluoroscopic projection is seen in one of the patients patient 6 with the electrodes and catheters marked. The right panel shows a schematic drawing of the

Ž w x .atria with the position of the catheters and the esophagus electrode shown in more detail drawing reprinted from Cosio et al. 30 , with permission .



ownloaded from




The distance between the peripheral and the intracardiacrecording positions was determined in each patient fromaxial and sagittal oblique MR images obtained from aSiemens Magnetom Vision 1.5 Tesla with a circular polar-ized body array coil using a single slice, breath-holdT2-weighted turbo spin-echo pulse sequence with surface

ŽECG-triggering ETLs23, TRrTEs1465–2748 msr85ms, matrixs138–230=256–512, slice thicknesss7–8

.mm .

3.2.2. InterÕention with intraÕenous D,L-sotalolSimultaneous acquisition of the 12-lead standard sur-

face ECG was performed during 20 min. The patients wereallowed to rest in the supine position for 15 min beforerecording was started. Five minutes after the start of dataacquisition, 80 mg D,L-sotalol was injected into a periph-eral vein during 2 min. During the whole recording, non-invasive systolic blood pressure was manually measuredevery minute. The systolic blood pressure was estimated asthe pressure where consistent pulsations could be heard.The mean heart rate over 1 min was also calculated with a

1-min interval. The FAF-ECG method was applied to leadV1 and the DACLs were determined for consecutive inter-vals of 1 min each. The mean values of the blood pressure,the heart rate and the DACL were also determined for thefirst and the last 5 min of the recording.

3.2.3. Application to 1-h ECG recordingsOne hour of 12-lead surface ECG was obtained during

supine rest. Data acquisition was started after a 15-minresting period. The FAF-ECG method was applied and theDACLs in lead V1 were determined for consecutive inter-vals of 5, 10, 15, 20 and 30 min. The coefficient ofvariation of the DACLs was then calculated for eachduration of the consecutive intervals. We defined the DACLto be reproducible if its coefficient of variation in consecu-tive intervals was below 5%.

3.2.4. Validating the spectral estimateThe effect of various window lengths and interval over-

laps in the Welch periodogram method on the spectral

Table 2Ž . Ž .The dominant atrial cycle lengths DACLs of the peripheral and intracardiac signals in patients undergoing an invasive electrophysiological study a and

Ž .the distance between the peripheral and intracardiac recording positions according to MRI b

( ) Ž .a Patient DACLs ms

Recording 1 Recording 2

V1 ESO CS HLRA V1 LLRA HLRA HSRA RAA CS Mean RAdist prox prox

rRAA

1 180 185 185 NA 185 185 180 175 180 NA 1802 140 155 160 NA 140 135 135 140 135 155 137

175 200 1403 140 155 175 160 130 155 160 150 130 175 130

175 155 150 1544 165 175 195 175 160 160 160 160 170 185 1625 140 145 160 140 140 155 150 150 140 160 1496 160 170 180 160 160 155 160 170 170 180 1647 170 175 190 210 175 160 210 195 185 195 1878 145 140 175 150 145 150 150 150 150 175 150

Mean"s.d. 155"16 165"15 180"14 166"25 154"18 157"14 163"23 161"18 155"20 175"14 157"19

( ) Ž .b Distance between the recording positions cm

ESO-CS Min. ESO-LA Min. ESO-RA V1-CS V1-HLRA V1-LLRA V1-HSRA V1-RAA Min. V1-RAprox prox

1 2.5 2.5 2 15 9 8.5 10 8 52 3.5 2 3 11 9.5 8.5 9 9 4.53 NA NA NA NA NA NA NA NA NA4 2 2 2 12 8.5 8 8 8 55 4.5 4 3 14 9 9 10 10 46 2 2 1.5 12 8 8 11 8.5 67 2.5 3 1.5 13 9 8.5 10.5 9 58 2 2 1.5 14 8.5 8 11 10 5

Mean"s.d. 2.7"1.0 2.5"0.8 2.1"0.7 13"1.4 8.8"0.5 8.4"0.4 9.9"1.1 8.9"0.8 4.9"0.6

ESOsesophagus; CSscoronary sinus; distsdistal; prox sproximal; LLRA s low lateral right atrium; HLRA shigh lateral right atrium; HSRA shighŽseptal right atrium; RAA s right atrial appendage; Mean RA smean of the DACLs at LLRA, HLRA, HSRA and RAA; NA snot available due to

.claustrophobia .



ownloaded from






ownloaded from




estimate was evaluated. Three combinations of windowŽ . Ž .length and interval overlap, 1024, 256 , 512, 128 and

Ž . Ž .256, 64 , were applied in one peripheral signal lead V1Ž .and one intracardiac signal HLRA in patients 1–8. For

each combination and patient, the distribution in the 3- to12-Hz range and the DACLs was determined.

3.2.5. Statistical analysisCorrelation analysis and a t-test were used to test

whether there were any significant differences in the DACLwith respect to the patients’ age, duration and type of AF.Wilcoxon’s signed-rank tests and paired t-tests were usedto determine if there were any significant differences in theabsolute differences in DACL between the intracardiacsignals and the peripheral signals and if there were anysignificant changes in blood pressure, heart rate or DACLdue to intervention with sotalol. A P-value of less than0.01 was considered significant.

4. Results

4.1. General

A total of 31 recordings were successfully analyzedfrom the 23 patients. Another three patients were excludedfrom the material due to inability to induce more than

Ž .short less than 1-min duration paroxysms of AF duringŽ .the invasive electrophysiological study one patient and

failure to identify the DACL due to poor QRST suppres-sion caused by a marked variation in QRST morphologyand low amplitude of the f-waves in the surface ECG,

Ž .respectively, in the 1-h recordings two patients .In all patients, the DACL in V1 ranged between 120

Ž .and 195 ms during supine rest before any intervention . Inpatients undergoing an invasive study, the DACL in theintracardiac electrograms ranged between 130 and 195 msat the different intracardiac sites. No significant differencesin DACL were found with respect to the type of AF,duration of the arrhythmia or the patient’s age.

4.2. InÕasiÕe electrophysiological study

The DACLs for all signals and patients in the twosubsequent recordings are shown in Table 2a. A spatial

Ž .mean of the right DACLs Mean RA , i.e. the mean of theDACLs in LLRA, HLRA, HSRA and RAA electrogramsfrom recording 2, is shown in the last column. In two ofthe patients, certain signals had a multimodal frequencydistribution and, more than one DACL is therefore given.The absolute differences in DACL between V1 and CSprox

Žwas 24 ms and between ESO and CS 16 ms meandist prox.of all patients . The absolute differences between V1 and

CS was 24 ms and between V1 and Mean RA 5 ms, theprox

latter being significantly smaller than the former. It shouldŽ .be noted that in cases of multimodality patients 2 and 3 ,

we have compared the individual DACLs in each signalonly with the closest DACLs in the other signals. Thedistances between the peripheral and the intracardiacrecording positions, according to MRI, are given for eachpatient in Table 2b. The distance between the ESO record-

Ž .ing position and CS was 2.7 cm mean of all patients ,prox

which was much smaller than the distance between the V1electrode and the four right atrial positions, ranging be-tween 8.4 and 9.9 cm. From MRI, it was also obvious thatthe V1 electrode position is right in front of the right atrialfree wall and that the left atrium is almost entirely con-cealed by the right atrium from a V1 point of view. It isalso evident that the esophagus is equally close to the right

Ž .and the left atrium Table 2b .Ž .The DACL s and the frequency spectra for one of the

patients with an occasional multimodal distribution aregiven in Fig. 3.

4.3. InterÕention with intraÕenous D,L-sotalol

Comparing the mean values of the first 5 min with themean values of the last 5 min of the recordings, thesystolic blood pressure did not change significantly, theheart rate decreased significantly and the DACL in lead

Ž .V1 increased significantly. One of the patients patient 13had multimodally distributed spectra, and here similar

Ž .Fig. 3. Comparison between the dominant atrial cycle length DACL determined from the surface ECG lead V1 and the esophagus ECG and multipleŽ . Ž .simultaneously acquired intracardiac right and proximal coronary sinus CS atrial electrograms using the new FAF-ECG method patient 3 . The result

Ž . Ž . Ž .from Recording 1, including ECG lead V1, distal esophagus ECG ESO , proximal CS CS and high lateral right atrial HLRA electrograms, isdist prox

shown. The frequency distribution in ESO was multimodal with two DACLs, whereas the distribution was unimodal in the other signals. The DACL indist

CS was identical to one of the DACLs in ESO , whereas the second DACL in ESO was closer to the DACL in HLRA. The DACL in V1 was, inprox dist distŽ .this recording, 20 ms shorter than the DACL in HLRA. The results from Recording 2, including CS , HLRA, low lateral right atrial LLRA , highprox

Ž . Ž .septal right atrial HSRA , right atrial appendage RAA electrograms and ECG lead V1, in the same patient are shown. In this recording, two DACLsŽ .were seen in the V1 and RAA electrogram, one at 130 ms in both signals. The other DACL in V1 155 ms represented an approximate average of the

DACLs in the four right atrial electrograms, including the second DACL in RAA. It is interesting to note that while the frequency spectrum in V1 wasunimodal and quite wide in recording 1, the spectrum was multimodal in recording 2, illustrating the spontaneous variability seen in the AF cycle length in

Ž .humans discussed in more detail in the text . Another interesting observation is that in some patients, there is obviously a large spatial variability in theright AF cycle length and this variability appears to be reflected in ECG lead V1, i.e. this lead represents global right atrial activity.



ownloaded from




Fig. 4. An example of the results of intervention with intravenousŽ .D,L-sotalol patient 9 . The upper curve shows the non-invasively mea-

Ž .sured systolic blood pressure NIBP , the middle curve the heart rateŽ . Ž .HR and the lower curve the dominant atrial cycle length DACL asdetermined from ECG lead V1 using the FAF-ECG method. All statisticswere measured with 1 min resolution. At 6 min, 80 mg D,L-sotalol wasinjected in a peripheral vein during 2 min. The administration of D,L-

Ž .sotalol resulted in a decrease in systolic blood pressure 15–30 mmHg , aŽ .decrease in heart rate 30–35 bpm and an increase in the DACL in lead

Ž .V1 25 ms .

changes were seen in the two DACLs. The results fromone of the patients are illustrated in Fig. 4.

4.4. Application to 1-h ECG recordings

The results from applying the new method to consecu-tive intervals of ECG lead V1 with varying duration areshown in Table 3. In two of the 10 patients, the frequencydistribution was multimodal occasionally with two DA-CLs, exclusively seen in the shorter intervals. In these

cases, the mean of the two DACLs was used for theanalysis. In 9 of the 10 patients, 5 min of ECG recordingwas sufficient for the DACLs to be reproducible, i.e. thecoefficient of variation was below 5%. In the remainingpatient, the required duration could not be determined asthe coefficient of variation was slightly above 5%, regard-less of the duration of the ECG recording.

4.5. Validating the spectral estimate

A unimodal frequency distribution in the 3- to 12-Hzrange was found in all patients when using the two combi-

Ž . Ž .nations 256, 64 and 512, 128 of the window length andthe interval overlap, respectively, of the Welch peri-odogram. In 5 of 8 patients, the frequency distribution was

Ž .also unimodal when using the combination 1024, 256 ,whereas in the other 3 patients, a multimodal distributionwith two frequency components was found, both in thespectrum from lead V1 and in the HLRA electrogram. Inthe latter cases, the two frequency components were veryclose to each other and represented one cycle lengthslightly longer and one slightly shorter than the singledominant cycle length in the corresponding spectra calcu-lated using the other two combinations. In all but thesethree latter cases, the DACL was the same irrespective ofthe combination of window length and interval overlap

Ž .used. When using the combination 1024, 256 , the sidelobes of the dominant frequency component were visuallymore pronounced, i.e. the variance was larger, than in theother combinations, although this did not affect the mainfrequency peak in the 3- to 12-Hz range. An exampleshowing the effect of chosen window length and intervaloverlap on the spectral estimate is given in Fig. 5.

Table 3Ž .The coefficient of variation of the dominant atrial cycle lengths DACLs in lead V1 in consecutive intervals of different duration

Ž .Patient Frequency distribution Coefficient of variation % of the DACLs in consecutive intervals of:

5 min 10 min 15 min 20 min 30 min

14 UM 2.7 2.5 2 1.7 015 UM 2.1 1.8 1 1.1 016 UM 1.3 0.8 0.8 0.8 117 UM 1.6 0.8 1.4 0.8 118 UM 1.5 1.1 1.4 1.4 119 Occasionally MM 5.6 4.5 5.5 5.5 6.320 UM 0.9 0.6 0.7 0.7 0.821 Occasionally MM 3 3.5 2.9 2.1 0.922 UM 1.2 0.8 0 0 023 UM 0.9 0.6 0.3 0 0

Mean 2.1 1.7 1.6 1.4 1.1

DACLssdominant atrial cycle lengths; UMsunimodal; MMsmultimodal.



ownloaded from



()

M.H

olmet

al.rC

ardioÕascular

Research

381998

69–

8178

Fig. 5. An example showing the result from the validation of the spectral estimate. Three different spectra were calculated, each with a different combination of the window length and the interval overlapŽ . Ž .values in parentheses of the Welch periodogram method used to estimate the frequency spectra in the FAF-ECG method . As can be seen, by varying the method used to estimate the spectra, a quitedifferent appearance of the frequency distribution can be obtained. By using a window length of 1024 and an interval overlap of 256, a more distinct frequency peak is seen, but, on the other hand, the sidelobes of the dominant frequency component are visually more pronounced, i.e. the variance is larger than for the other combinations. One obviously has to compromise between small variance in thefrequency spectrum and good distinction between closely located frequency components when using the above method for frequency spectrum estimation.

by guest on April 7, 2014 http://cardiovascres.oxfordjournals.org/ Downloaded from




5. Discussion

We have shown that the atrial component in the ECGfrom patients in AF contains information on the magnitudeand the dynamics of the atrial cycle length. This informa-tion can be extracted using a new method called ‘FAF-ECG’ including time-averaging and spectral analysis tech-niques.

5.1. The DACL in lead V1

In unipolar leads, potentials originating close to theelectrode have a larger influence on the signal than poten-

w xtials originating far from the electrode 22,23 . The resultsfrom the MRI in this study show that the V1 electrode isfacing the free wall of the right atrium and that the leftatrium, from a V1 point of view, is concealed behind theright atrium. Therefore, the agreement in DACL betweenlead V1 and the right atrium is not at all surprising. In ourmaterial, the distance from the V1 electrode to the differ-ent parts of the right atrium ranged between approximately5 and 10 cm, whereas the distance to the proximal part ofthe CS was 13 cm, the latter circumstance explaining thepoor agreement between lead V1 and the CS. Furthermore,the circumstance that a multimodal distribution of the localAF cycle length, as seen in the right atrial appendage in

Ž .patient 2 Table 2a , was reflected in lead V1 indicates thatall parts of the right atrium, not only the closest ones,contribute to the V1 ECG signal. This contribution is, ofcourse, not entirely restricted to the right atrium as somestructures of the left atrium are at a similar distance fromthe V1 electrode as parts of the right atrium, but the resultsin the present study strongly suggest that the right atriumis responsible for the major contribution to the f-waves inlead V1.

5.2. The DACL in the esophagus ECG

While lead V1 may be used to assess the right AF cyclelength, the agreement between the ESO ECG and the CSelectrogram was less good, although it was slightly betterthan that between lead V1 and the CS electrogram. Itshould be considered in this context that the distance fromthe esophagus to both the proximal part of the CS and theclosest part of the right atrium and left atrium, respec-tively, is short. This implies that, the esophagus ECGbeing a unipolar signal, not only structures surrounding theCS, but also parts of the right and the left atrium con-tributed to the esophagus ECG. Thus, the esophagus ECGis most likely a mixture of the electrical activity in boththe left and the right atria as well as the interatrial septum,with interindividual variations in the contribution fromeach structure. This conclusion is in agreement with the

w xresults of Prystowsky et al. 24 . Using a somewhat differ-ent approach, comparing the esophageal atrial electrogramsduring orthodromic reciprocating tachycardia due to left-sided accessory pathways with electrograms from the CSand the AV-nodal area, they concluded that the distalesophageal electrode probably recorded posterior parasep-tal activity.

5.3. Monitoring the effect of interÕentions

The electrophysiological effects of administration ofsotalol in humans are well documented in the literature.Both the atrial and ventricular effective refractory periodare prolonged without any significant change in the atrial

w xor ventricular conduction velocity 25–28 . As the effec-tive refractory period sets the upper limit for the rate atwhich the atria can be activated during fibrillation, theatrial fibrillatory rate may be expected to decrease afteradministration of sotalol. In the present study, we havedocumented this effect, thus illustrating that by using thenew FAF-ECG method, changes in the fibrillatory rate canbe followed. As our increasing understanding of the elec-trophysiological mechanisms responsible for the initiationand perpetuation of AF in man forms the basis for newtherapeutic strategies, we believe that the need for non-in-vasive methods, capable of monitoring changes in thecharacteristics of this arrhythmia due to therapeutic inter-ventions, will increase dramatically.

5.4. Spontaneous Õariability in AF cycle length

We can conclude that 5 min of ECG recording issufficient for the DACL in lead V1 to be reproducible, i.e.that time window is enough to capture the majority of thetemporal dynamics of the AF cycle length. However, asalso mentioned below, an enhanced time resolution of thespectral estimate is needed to explore the spontaneousdynamic variations in atrial activation during chronic AFfurther.

5.5. The pathophysiological significance of DACL

The AF cycle length in humans has previously beenexpressed in terms of mean and standard deviation or

w xmedian and 10th–90th percentiles 4,8 . In cases of normaldistribution of the cycle length, the mean, the median andthe DACL are all identical. However, available data sug-

w xgest that this is not always the case 4,8 . These differentstatistics therefore represent different measures of the cy-cle length during AF. It has previously been shown that thelocal AF cycle length is an index of atrial refractorinessw x11–13 . As a drug-induced lengthening of the atrial refrac-

Ž .tory period intravenous D,L-sotalol resulted in an increase



ownloaded from




in the DACL in the present study, it is suggested, althoughnot proved, that a similar relationship exists between atrialrefractoriness and DACL during AF.

5.6. Limitations

As presented here, the FAF-ECG method has someimportant limitations:Ø Although spectral analysis can be performed on the

ECG signal without suppression of the QRST com-plexes, reliable and accurate identification of the atrialspectral component is dependent on such suppression.Here, suppression of the QRST complexes is done bysubtracting a time-averaged QRST complex from eachidentified QRST complex. As mentioned above, sup-pression of the ventricular component exclusively in theECG is dependent on the fact that the atrial and theventricular activity are completely unsynchronized.Thus, in cases of an atrial rhythm alternating betweenfibrillation and flutter, synchronization between atrialand ventricular activity will occasionally be present,implying that the atrial activity is also occasionallysuppressed.

Ø The method has limited ability to suppress QRST com-plexes that change morphology over time and, in addi-tion, ventricular extrasystoles are not suppressed at all.Temporal variability in the QRST complex morphologymay be due to changes in the electrical axes, as aconsequence of breathing, or due to variability in QT

w xtime 29 . Furthermore, the possible presence of so-called U-waves after the T-wave is not considered whencreating the time-averaged QRST complex and is there-fore not suppressed by the method. Both in cases of

Ž .large remaining QRST U complexes and in cases offrequent extrasystoles in the residual ECG, large fre-quency components are seen in the 0- to 3-Hz range

Ž .which may interfere with frequency component s cor-responding to the atrial activity in the 3- to 12-Hzrange.

Ø Although the method can accurately follow changes inthe AF cycle length, the technique of manually annotat-ing consecutive intervals of the data and calculating afrequency spectrum from each interval is a time-con-suming task that gives limited time-resolution. In someapplications, a true time–frequency representation maybe needed. We also believe that the method of estimat-ing the frequency spectrum itself can be further im-proved, providing an enhanced frequency resolution.

Ø As mentioned above, as only left atrial electrogramswere acquired from the CS, no firm conclusions can bedrawn regarding the interpretation of the DACL ob-tained from the esophagus ECG.

Ø Finally, the fact that we occasionally documented amultimodal frequency distribution in the 3- to 12-Hzrange, i.e. two or more components that, most likely,

were caused by the atrial activity, raises the issuewhether these were caused by spatial or temporaldifferences in the fibrillatory rate. Spatial differences inthe AF cycle length were seen in almost all patientsundergoing an invasive electrophysiological study, butonly in a few cases resulted in a multimodal frequencydistribution in the ECG signals. A temporal variabilityin the atrial activity during AF has earlier been docu-mented and was in the present study verified by the 1-hECG recordings. Whether the temporal variability is thecause of multimodality in the frequency distribution ofthe residual ECG signal or not cannot be determinedfrom the present data.

6. Summary

We have developed a new non-invasive method capableof assessing the dominant AF cycle length in man from astandard ECG recording using time-averaging and spectralanalysis techniques. By comparison with simultaneouslyacquired intracardiac data, we have shown that the domi-nant AF cycle length derived from the precordial lead V1well represents a spatial mean of the right intracardiac AFcycle length, whereas the dominant AF cycle length de-rived from the esophagus ECG is influenced by the right,left and septal intracardiac AF cycle lengths, where thedegree of influence from each structure depends on theanatomy of the individual. Furthermore, we have shownthat changes in the AF cycle length due to pharmacologi-cal intervention can be monitored with this new method.Finally, the spontaneous temporal variability in atrial cyclelength in patients with chronic AF requires a recordingtime of at least 5 min in order to obtain reproducibleresults.

Acknowledgements

This study was supported by grants from the SwedishHeart Lung Foundation. The authors wish to thank thestaff of the Electrophysiological Unit, Department of Car-diology, University Hospital, Lund, Sweden for their helpwith data acquisition

References

w x1 Wells JL, Karp RB, Kouchoukos NT, et al. Characterization of atrialfibrillation in man: studies following open heart surgery. PACE1978;1:426–438.

w x2 Allessie, M, Lammers, W, Smeets, J, Bonke, F, Hollen, J. Totalmapping of atrial excitation during acetylcholine-induced atrial flut-ter and fibrillation in the isolated canine heart. In: Kulbertus, HE,



ownloaded from




Olsson, SB, Schlepper, M, editors. Atrial fibrillation. Molndal,¨Sweden: Lindgren and Soner AB, 1982:44–61.¨

w x3 Cox JL, Canavan TE, Schuessler RB, et al. The surgical treatment ofatrial fibrillation. II. Intraoperative electrophysiologic mapping anddescription of the electrophysiologic basis of atrial flutter and atrialfibrillation. J Thorac Cardiovasc Surg 1991;101:406–426.

w x4 Konings K, Kirchhof C, Smeets J, et al. High-density mapping ofelectrically induced atrial fibrillation in humans. Circulation

Ž .1994;89 4 :1665–1680.w x5 Harada A, Sasaki K, Fukushima T, et al. Atrial activation during

chronic atrial fibrillation in patients with isolated mitral valve dis-ease. Ann Thorac Surg 1996;61:104–112.

w x6 Jais P, Haissaguerre M, Shah DC, et al. A focal source of atrialfibrillation treated by discrete radiofrequency ablation. Circulation1997;95:572–576.

w x7 Konings KTS, Smeets JLRM, Penn OC, Wellens HJJ, Allessie MA.Configuration of unipolar atrial electrograms during electricallyinduced atrial fibrillation in humans. Circulation 1997;95:1231–1241.

w x8 Holm M, Johansson R, Brandt J, Luhrs C, Olsson SB. Epicardial¨right atrial free wall mapping in chronic atrial fibrillation. Documen-tation of repetitive activation with a focal spread – a hithertounrecognised phenomenon in man. Eur Heart J 1997;18:290–310.

w x9 Kumagai K, Khrestian C, Waldo AL. Simultaneous multisite map-ping studies during induced atrial fibrillation in the sterile pericardi-tis model. Insight into the mechanism of its maintenance. Circulation1997;95:511–521.

w x10 Holm M, Olsson SB, Johansson R, et al. Non-invasive assessment ofthe atrial cycle length during chronic atrial fibrillation in man.

Ž . Ž .Circulation suppl 1996;94 8 :I–69.w x11 Lammers WJEP, Allessie MA, Rensma PL, Schalij MJ. The use of

fibrillation cycle length to determine spatial dispersion in electro-physiological properties and to characterize the underlying mecha-

Ž .nism of fibrillation. New Trends Arrhythmias 1986;II 1 :109–112.w x12 Capucci A, Biffi M, Boriani G, et al. Dynamic electrophysiological

behavior of human atria during paroxysmal atrial fibrillation. Circu-lation 1995;92:1193–1202.

w x13 Kim K-B, Rodefeld MD, Schuessler RB, Cox JL, Boineau JP.Relationship between local atrial fibrillation interval and refractoryperiod in the isolated canine atrium. Circulation 1996;94:2961–2967.

w x14 Hewlett AW, Wilson FN. Coarse auricular fibrillation in man. ArchIntern Med 1915;15:786–792.

w x15 Thurmann M, Janney JG. The diagnostic importance of fibrillatorywave size. Circulation 1962;25:991–994.

w x16 Peter RH, Morris JJ, McIntosch HD. Relationship of fibrillatory

waves and p waves in the electrocardiogram. Circulation1966;33:599–606.

w x17 Morganroth J, Horowitz LN, Josephson ME, Kastor JA. Relation-ship of fibrillatory wave amplitude to left atrial size and etiology of

Ž .heart disease. Am Heart J 1979;97 2 :184–186.w x18 Leier CV, Schaal SF. Biatrial electrograms during coarse atrial

Ž .fibrillation and flutter–fibrillation. Am Heart J 1980;99 3 :331–341.w x19 Li Y-H, Hwang J-J, Tseng Y-Z, Kuan P, Lien W-P. Clinical

significance of fibrillatory wave amplitude. A clue to left atrialappendage function in nonrheumatic atrial fibrillation. Chest1995;108:359–363.

w x20 Slocum J, Sahakian A, Swiryn S. Diagnosis of atrial fibrillation fromsurface electrograms based on computer-detected atrial activity. J

Ž .Electrocardiol 1992;25 1 :1–8.w x21 Slocum, JE, Ropella, KM. Correspondence between the frequency

domain characteristics of simultaneous surface and intra-atrialrecordings of atrial fibrillation. Comput Cardiol 1994;781–784.

w x22 Steinhaus BM. Estimating cardiac transmembrane activation andrecovery times from unipolar and bipolar extracellular electrograms:a simulation study. Biophys J 1988;53:424a.

w x23 Blanchard SM, Damiano RJ, Asano T, et al. The effects of distantcardiac electrical events on local activation in unipolar epicardial

Ž .electrograms. IEEE Trans Biomed Eng 1987;34 7 :539–546.w x24 Prystowsky EN, Pritchett ELC, Gallagher JJ. Origin of the atrial

electrogram recorded from the esophagus. CirculationŽ .1980;61 5 :1017–1023.

w x25 Gomoll AW, Bartek MJ. Comparative b-blocking activities andelectrophysiologic actions of racemic sotalol and its optical isomersin anesthetized dogs. Eur J Pharmacol 1986;132:123–135.

w x26 Rensma PL, Allessie MA, Lammers WJEP, Bonke FIM, Schalij MJ.Length of excitation wave and susceptibility to reentrant atrialarrhythmias in normal conscious dogs. Circ Res 1988;62:395–410.

w x27 Singh BN. Is class III antiarrhythmic activity important. CardiovascDrugs Ther 1990;4:597–602.

w x28 Karagueuzian, HS, Mandel, WJ. Antiarrhythmiac drugs: mode ofaction, pharmacokinetic properties, and therapeutic uses. In: Mandel,WJ, editor. Cardiac Arrhythmias. 3rd ed. Philadelphia: J.B. Lippin-cott, 1995:111.

w x29 Ehlert FA, Goldberger JJ, Rosenthal JE, Kadish AH. Relationbetween QT and RR intervals during exercise testing in atrial

Ž .fibrillation. Am J Cardiol 1992;70 3 :332–338.w x30 Cosio FG, Arribas F, Lopez-Gil M, Placios J. Atrial flutter mapping´

and ablation. I. Studying atrial flutter mechanisms by mapping andentrainment. PACE 1996;19:841–853.



ownloaded from



Non-invasive assessment of the atrial cycle length during atrial fibrillation in man: introducing,...

Documents

Transcript of Non-invasive assessment of the atrial cycle length during atrial fibrillation in man: introducing,...