IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 2, FEBRUARY 2014 273

Iterative Method to Detect Atrial Activations andMeasure Cycle Length From Electrograms During

Atrial FibrillationJason Ng∗, Vinod Sehgal, Justin K. Ng, David Gordon, and Jeffrey J. Goldberger

Abstract—Atrial fibrillation (AF) electrograms are character-ized by varying morphologies, amplitudes, and cycle lengths (CLs),presenting a challenge for automated detection of individual activa-tions and the activation rate. In this study, we evaluate an algorithmto detect activations and measure CLs from AF electrograms. Thisalgorithm iteratively adjusts the detection threshold level until themean CL converges with the median CL to detect all individualactivations. A total of 291 AF electrogram recordings from 13 pa-tients (11 male, 58 ± 10 years old) undergoing AF ablation wereobtained. Using manual markings by two independent reviewersas the standard, we compared the cycle length iteration algorithmwith a fixed threshold algorithm and dominant frequency (DF)for the estimation of CL. At segment lengths of 10 s, when com-paring the algorithm detected to the manually detected activation,the undersensing, oversensing, and total discrepancy rates were2.4%, 4.6%, and 7.0%, respectively, and with absolute differencesin mean and median CLs were 7.9 ± 9.6 ms and 5.6 ± 6.8 ms,respectively. These results outperformed DF and fixed threshold-based measurements. This robust method can be used for CL mea-surements in either real-time and offline settings and may be usefulin the mapping of AF.

Index Terms—Biomedical signal processing, cardiology, electro-cardiography, fibrillation.

I. INTRODUCTION

CONTACT atrial electrograms recorded during atrial fib-rillation (AF) are characterized by rapid deflections with

changing amplitudes, cycle lengths (CLs), and morphologies.Unlike other atrial tachyarrhythmias with regular activation pat-terns, AF has complex activation patterns that make elucidationof AF mechanisms difficult. Mapping of atrial CL or atrial acti-vation rates has been proposed as an alternative to mapping acti-vation sequences [1]–[5]. It has been hypothesized that sites withthe fastest activation rates represent the locations of the driversof AF (focal or reentrant) and could be possible ablation targets.

Manuscript received August 1, 2013; revised October 7, 2013 and October30, 2013; accepted October 31, 2013. Date of publication November 7, 2013;date of current version January 16, 2014. Asterisk indicates corresponding au-thor.

∗J. Ng is with the Feinberg School of Medicine and Northwestern University,Chicago, IL 60611 USA (e-mail: jsnng@northwestern.edu).

V. Sehgal is with the University of Illinois College of Medicine, Chicago, IL60612 USA (e-mail: vinod.sehgal.88@gmail.com).

J. K. Ng, D. Gordon, and J. J. Goldberger are with the Feinberg Schoolof Medicine and Northwestern University, Chicago, IL 60611 USA (e-mail:justin-ng-0@northwestern.edu; d-gordon@northwestern.edu; j-goldberger@northwestern.edu).

Digital Object Identifier 10.1109/TBME.2013.2290003

Evidence of this has been shown in both clinical [2], [5], [6] andexperimental studies [7], [8].

One of the main limitations of this activation rate mappingapproach is the technical difficulties involved in obtaining re-liable measurements. The complexity of AF electrograms canmake detection of deflections and the calculation of the CLsand activation rates difficult in both the time and frequency do-mains [9], [10]. Deflection-to-deflection intervals can be mea-sured manually by calipers but because of the variability of AFCLs, an average of several intervals are needed to characterizethe AF CL. However, this is an arduous task for the operatormaking the measurements. An alternative method is a manualsetting of an amplitude or slope threshold value that can beused to detect deflections. The limitation for manual thresh-olds is the subjectivity required to distinguish noise from atrialactivation. Additionally, automatic algorithms which detect de-flections based on fixed threshold levels are prone to oversensingor undersensing [11]. Dominant frequency (DF) analysis usesthe frequency that contains the most power to be the estimateof activation rate. DF analysis works well in the estimation ofactivation rates if the AF electrograms have a certain amount ofregularity, but the correlation is reduced with highly irregularwaveforms and complex morphologies [9], [10]. Thus, morerobust algorithms validated with rigorous testing are needed.

In this study, we evaluated a new algorithm which uses cyclelength iteration (CLI) to detect atrial complexes. The algorithmoperates on observations from previous work [9] that AF CLshave distributions where mean and median CLs are approxi-mately equal. Using manually marked electrograms as the stan-dard, we evaluated the accuracy of the CLI algorithm to detectactivations and compute AF CLs, and compared it with fixedthreshold algorithms and DF analysis.

II. METHODS

A. Electrogram Dataset

Electrogram recordings from 13 patients (11 male/2 female,58 ± 10 years old) undergoing AF ablation were obtained priorto radiofrequency ablation at Northwestern Memorial Hospi-tal. Four patients had paroxysmal AF and nine had persistentAF. Mapping and recording were performed using a Navi-Starcatheter (Biosense Webster, Inc., Diamond Bar, CA). Bipolarelectrograms were sequentially obtained from at least ten sitesin the right atrium and at least ten sites in the left atrium andstored on the Prucka CardioLab EP System (GE Healthcare,Waukesha, WI). The electrograms were sampled at a rate of

0018-9294 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

274 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 2, FEBRUARY 2014

977 Hz and filtered with a 30-Hz high-pass filter. The protocolwas approved by the Office for the Protection of Research Sub-jects’ Institutional Review Board of Northwestern University.

B. Manual Marking of Electrograms

All electrogram analyses were performed offline using toolsdeveloped in MATLAB (Mathworks, Natick, MA). Activa-tion complexes for each electogram recording were manuallymarked by two independent operators. The operators were in-structed to mark each activation at the point of the highest abso-lute amplitude. For electrograms with discrete but fractionatedactivation complexes with clear isoelectric periods before andafter the complex, the highest amplitude deflection was cho-sen. No markings were made within 50 ms of a previous mark.Simultaneous surface ECG recordings were available to distin-guish ventricular far-field activity from atrial activations whenperforming manual markings. Signals were excluded if the am-plitudes of the ventricular far-field complexes were greater thanthat of the atrial complexes. The manual reviewers were oth-erwise instructed not to mark ventricular complexes using thesurface ECG as a reference.

In addition to the two independent sets of markings, we alsoevaluated the CLI algorithm using the intersection of the two setsof manual marking. The intersection set was intended to providea set of activations that both reviewers were most confident in.The intersection was defined as the marks that are common toboth sets of manual markings within 50 ms. If the marked pointsagreed, then the time point of the first set was used.

C. CLI Algorithm

The electrograms are first preprocessed with similar stepsused by Botteron and Smith [12]: 1) 40-Hz high-pass filtering(second-order Butterworth); 2) rectification; and 3) 30-Hz low-pass filtering (second-order Butterworth). The iterative processfor detecting the activations is summarized in the flowchart ofFig. 1. The peak with the highest magnitude is the first detectedactivation time. Next, all peaks occurring within 50-ms blankingperiod before and after the detected beat are excluded. The nextlargest peak is then detected and added to the set. Then, theblanking period is applied again. The process of detecting thenext peak and applying the blanking period is repeated until themean calculated CL is less than 275 ms and one of the followingtwo conditions are met: 1) the mean CL is less than median CLplus 5 ms or 2) the magnitude of the current peak is 20% lessthan the magnitude of the previously detected peak.

The mean and median CL convergence criterion is based onprevious observations that AF CLs have distributions wheremean and median CLs are approximately equal [9]. Atrial ratesmeasured during human AF activity have been shown to be inthe 4–9 Hz range, which is equivalent to CLs between 250 and111 ms [13]. Therefore, we have selected a CL of 275 ms asupper bound for the activation detection in case the mean/mediancrossings occur prematurely. Fig. 2 shows an example of themean/median CL convergence when the 18th activation is addedto the set.

Fig. 1. Flowchart describing the steps of the CLI algorithm.

Fig. 2. Graphical illustration of the CLI method: (a) raw unfiltered electro-grams, (b) electrograms after rectification and low-pass filtering with peaksnumbered in order of the highest to lowest amplitude, (c) graph showing meanand median CLs calculation after adding peaks one by one. Mean and medianCLs converge after 18 peaks.

NG et al.: ITERATIVE METHOD TO DETECT ATRIAL ACTIVATIONS AND MEASURE CYCLE LENGTH 275

In order to detect potentially missed activations within longerintervals, a final postprocessing step involves finding activationintervals greater than 1.5 times the median CL. The largestpeak, if present, within the interval and not within 50 ms ofanother peak is included in the set of activations. This process isrepeated until there are no more intervals greater than 1.5 timesthe median CL with peaks between them.

D. Evaluation of the CLI Algorithm

The CLI algorithm was evaluated on contiguous AF seg-ments of each of the following segment lengths: 2, 4, 6, 8, 10,12, 14, and 16 s. Detected activation by the CLI algorithm thatis within 75 ms of a manually marked activation was consid-ered matched activations. As the manual marks were made onunfiltered signals and the tested algorithms were performed onfiltered electrograms, a 75 ms window was chosen to encompassthe width of a fractionated electrogram. Detected activations notwithin 75 ms of a manually marked activation that had not al-ready been classified as a matched activation were consideredoversensed activations. Manually marked activations that werenot within 75 ms of a detected activation were considered un-dersensed activations. Oversensed or undersensed activationsthat were either the first or last activation in the segment werenot counted in the oversensing and undersensing rates. Over-sensing and undersensing rates were calculated as percentagesof the total number of marked activations. Agreement in CLwas measured by calculating the absolute difference betweenthe mean/median CL of each segment between the manual andCLI algorithm.

E. Comparison With Other CL Estimation Methods

Agreement of the CLI algorithm with the manually markedactivations was compared against the following: 1) activationsdetected using an optimum fixed detection threshold; and 2) CLcalculated from DF.

Fixed threshold detection was performed using the filteredand rectified signal to detect the activations. With this approach,all peaks (taking into account the 50-ms blanking period) abovea giving threshold value were considered activations. A rangeof threshold values were tested to determine the optimal thresh-old value. Thresholds based on raw amplitude (in μVs) weretested in a 1–50 μV range with 1 μV increments. The am-plitudes of bipolar electrograms are in the microvolt range asthey are significantly attenuated following low-pass filtering.Thresholds based on the percentage of the maximum amplitudewere tested in a 0.5–25% range with 0.5% increments. Thresh-olds based on the percentage of one standard deviation weretested in a 2–100% range with 2% increments. For each typeand segment duration, the thresholds producing the minimumcombined oversensing and undersensing rates were consideredthe optimum thresholds.

DFs in this study were calculated as previously described [5].First, the electrogram segments were bandpass filtered withcutoff frequencies of 40 and 250 Hz using a second-ordered But-terworth filter. The filtered signals were then rectified and low-pass filtered at 20 Hz also using a second-ordered Butterworth

TABLE INUMBER OF SEGMENTS AND ACTIVATIONS ANALYZED FOR EACH SEGMENT

LENGTH STUDIED

filter. The power spectrum was then calculated using the FastFourier Transform. The DF was defined as the frequency withthe highest power in the 3–20 Hz band. The AF CL from DF(DF CL) was calculated as 1000/DF. As DF estimates activationrate without the detection of activations, only the absolute dif-ference between DF CL and manual CL was evaluated for eachsegment duration.

F. Statistics

Absolute differences in CL measurements and manuallymarked CLs of the multiple methods were compared using theWilcoxon signed rank test. P values less than 0.05 were consid-ered statistically significant.

III. RESULTS

A. Electrogram Characteristics

In total, 291 electrogram recordings were obtained. Twenty-seven electrograms where the atrial activations could not beeasily distinguished from the noise were excluded. The remain-ing set of electrograms were obtained from a total of 123 rightatrial sites and 141 left atrial sites with an average duration of27.5 ± 9.0 s. The average mean and median CLs determined bymanual marking were 159.9 ± 24.9 ms and 161.0 ± 24.9 ms,respectively. The average difference between the mean and me-dian CLs was 0.5 ± 5.7 ms and 82% of the recordings hadabsolute mean and median CL differences less than 5 ms whichsuggests that our assumptions of mean and median CL conver-gence used for the algorithm are valid for the large majority ofthe recordings. The average percent agreement between the twosets of manual markings was 92%. The number of contiguoussegments analyzed for each segment duration is summarized inTable I.

B. CLI Performance

The oversensing, undersensing, and total discrepancy rates ofthe CLI algorithm compared to the two sets of manual markingsand the intersection of the manual markings for the differentsegment durations are shown in Fig. 3(a)–(c). The undersens-ing rates were highest when shorter segment durations wereused. The oversensensing rate was more stable across the dif-ferent segment durations. Undersensing rates were higher forCLI when compared to the A markings than to the B markings.

276 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 2, FEBRUARY 2014

Fig. 3. Evaluation of the CLI algorithm against the manual markings frommarkers A and B and the intersection of the markings of A and B (A ∩ B) fordifferent segment lengths: panel (a) undersensing rate, (b) oversensing rates, (c)total discrepancy rates, (d) mean CL difference, and (e) median CL difference.

Oversensing rates were higher when compared to the B mark-ings than to the A markings. Undersensing rates were lowerand oversensing rates were higher when CLI was comparedto the intersection of A and B than when compared to eithermarkings A or B. At 10 s durations, when the total discrepancyrate seems to stabilize, the undersensing, oversensing, and totaldiscrepancy rates when CLI was compared to the A and B in-tersection markings were 2.4%, 4.6%, and 7.0%, respectively.Absolute differences in mean CL between the CLI algorithmand the manual markings are shown in Fig. 3(d). The abso-lute differences in median CL between the CLI and manualmarkings are shown in Fig. 3(e). Tracking with the undersens-ing rates, the absolute differences in CL are the highest withshorter segment durations. At 10 s durations, the absolute dif-ferences in mean and median CLs compared to the A and Bintersection markings were 7.9 ± 9.6 ms and 5.6 ± 6.8 ms,respectively.

For the 10 s segments, 25% of the recordings had CL stan-dard deviations less than 25 ms, 39% had standard deviationsbetween 25 and 40 ms, and 36% had standard deviations greaterthan 40 ms. The recordings with standard deviations less than25 ms had oversensing, undersensing, and total discrepancyrates of 0.3%, 1.2%, and 1.5%, respectively. Recordings withstandard deviations between than 25 and 40 ms had oversensing,undersensing, and total discrepancy rates of 2.8%, 2.2%, and5.1%, respectively. Recordings with standard deviations greaterthan 40 ms had oversensing, undersensing, and total discrepancyrates of 9.5%, 3.3%, and 12.8%, respectively.

2-s segments required an average of 4 ± 2 ms computationtime running on MATLAB with a computer equipped with a2-GHz Intel Core2Duo processor. 16-s segments required anaverage of 26 ± 5 ms computation time.

Fig. 4. Comparisons of (a) undersensing rates, (b) oversensing rates, and (c)total discrepancy rates with manual marking for the CLI algorithm and thefixed threshold algorithms based on: absolute amplitude (fixed-abs), percent-age of maximum amplitude (fixed-per), and percentage of standard deviation(fixed-sd).

C. CLI Versus Other CL Estimation Methods

The results comparing the undersensing, oversensing, andtotal discrepancy rates of the CLI algorithm with other activa-tion detection methods are shown in Fig. 4. For this analysis,we present only the data using the intersection markings as thestandard. Fixed threshold detections were evaluated using theoptimal thresholds that minimized the total discrepancy rates.The optimum absolute threshold values ranged from 8 to 9 μVdepending on segment duration. The optimum threshold as apercentage of the maximum peak amplitude ranged from 5.5%to 9.5%. The optimum threshold as a percentage of one stan-dard deviation ranged from 12% to 14%. Detection with theoptimum raw threshold values had the highest discrepancy ratesfor all segment durations. The CLI algorithm had much loweroversensing rates than any of the fixed threshold methods andlower undersensing rates for segment lengths of 6 s or greater.Overall CLI had the lowest total discrepancy rates for all thesegment lengths.

The absolute difference in mean and median CLs of the au-tomatic algorithms (including DF derived CL) with the man-ually marked intersection activations is shown in Fig. 5. CLIhad significantly lower absolute differences in both mean andmedian CLs than the other algorithms for all segment lengths.For the 10 s segment length, the CLI absolute difference inmean CL (7.5 ± 9.6 ms) was significantly less than that of DF

NG et al.: ITERATIVE METHOD TO DETECT ATRIAL ACTIVATIONS AND MEASURE CYCLE LENGTH 277

Fig. 5. Comparisons of (a) mean CL difference and (b) median CL differ-ence with manual marking for the CLI algorithm, DF derived CL, and thefixed threshold algorithms based on: absolute amplitude (fixed-abs), percent-age of maximum amplitude (fixed-per), and percentage of standard deviation(fixed-sd).

(13.3 ± 15.7 ms, p < 0.0001), fixed absolute threshold (28.2± 49.2 ms, p < 0.0001), fixed percentage threshold (13.9 ±19.2 ms, p < 0.0001), and fixed standard deviation threshold(11.5 ± 15.1 ms, p < 0.0001). The CLI absolute difference inmedian CL (5.4 ± 7.2 ms) was also significantly less than thatof DF (12.2 ± 14.8 ms, p < 0.0001), fixed absolute threshold(16.0 ± 33.5 ms, p < 0.0001), fixed percentage threshold (8.4± 13.9 ms, p < 0.0001), and fixed standard deviation threshold(7.1 ± 10.1 ms, p < 0.0091).

IV. DISCUSSION

The proposed CLI algorithm was shown to accurately de-tect atrial activations from bipolar electrograms recorded dur-ing AF and provide CL estimations that more closely matchedmanually marked activations than fixed threshold or DF-basedmethods. The criteria for mean and median CL convergenceof this algorithm provide detection criteria amidst the poten-tially very complex activation and morphology patterns of AFelectrograms. Although this study evaluated the algorithm onelectrograms in an offline setting, this automated algorithm isefficient enough to be used for real-time applications.

A variety of methods have been previously proposed to de-tect activations from electrograms in the setting of AF. Holmet al. [14] used a process which incorporates a cubic splineinterpolation of the activation wave and chooses the point thatdivides the interpolated wave to two equal parts as the acti-vation time. However, these methods require manual thresholdselection. Faes et al. [15] introduced an adaptive activation de-tection method which sets a threshold based on the amplitudeof the last ten detected peaks with exponentially decreasingweights and a blanking period of 55 ms. This method was com-pared with manual activation markings with differences rang-ing from 2.6 to 20 ms depending on the complexity of the AF.

Sensitivity and specificity of the measures were not reported.Lee et al. [16] evaluated a similar adaptive threshold algorithmfrom 20 000 electrograms recorded during a canine model ofAF. They showed that CLs obtained by the algorithm had a cor-relation of 0.96 when compared with those obtained by manualmarking. Although we have not directly compared the CLI withthese other methods, we postulate that the CLI algorithm wouldhave performance advantages for electrograms with highly vari-able electrogram amplitudes and longer durations, while theother adaptive threshold methods may have advantages withmore variable CLs and shorter durations.

We have attempted to use manually marked electrograms asthe basis for the evaluation of the CLI detection algorithm andcomparison with DF and the fixed threshold algorithms. Themanual marking of AF electrograms is also not a straightforwardprocess, as it requires sometimes subjective determinations ofwhat is considered an activation. Electrograms during AF canhave changing amplitudes and morphologies and can have veryfractionated morphologies that make activation markings verydifficult. Thus, we used the intersection set of markings fromtwo independent reviewers that in theory would include the onlyactivations with the highest confidence. The CLI algorithm hadthe lowest percentage of combined undersensed and oversensedactivations when the intersection set of manual markings wasused as the reference.

Although the sensitivity and specificity of detecting atrial acti-vations are important characteristics of the detection algorithm,the more critical evaluation criteria are the ability to robustlyestimate CL. When manually marking AF electrograms, a hu-man is able to see the recording in its entirety. The human eyecan process signals to identify where activations are expectedto occur based on the basic CL and thereby assign an activationto a low amplitude electrogram that may not be marked or de-tected by more rigid criteria. The iteration process of the CLIalgorithm to detect activation until the mean and median CLsconverge more closely mimics the human detection process,whereas fixed threshold methods do not use the contextual in-formation when detecting deflections. Thus, the CLI algorithmprovides a very robust measurement of AF CL.

There are some limitations to the CLI technique. First, thereis a requirement for clinical stability during the recording seg-ment in order for the algorithm to take advantage of mean andmedian CL convergence. Therefore, errors may occur if attempt-ing to use the technique during phases of rapid change in AFCL that may occur with drug or autonomic interventions. Sec-ond, the CLI algorithm works significantly better with longersegment lengths, although the CL measurements at the shortlengths were still better than those obtain with DF or fixedthresholds. Segment lengths of 10 s or more appear to provideoptimal activation detection and CL estimation. We show, how-ever, that for a subset of recording with highly irregular CLs,discrepancy rates compared to manual marking were consider-ably higher than those with less variable CLs. It is expected thatfar-field ventricular complexes would remain an issue with theCLI algorithm, as they do with CLs estimated with DF and fixedthreshold techniques. However, there are a variety of techniquesthat may be able to alleviate the effect of ventricular complexes.

278 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 2, FEBRUARY 2014

[17] Finally, it is not known whether the results of the CLI al-gorithm will provide more meaningful insights about AF mech-anisms. This will need to be explored in future studies.

V. CONCLUSION

Mapping AF CLs for the understanding of AF pathophysiol-ogy requires a certain amount of precision of the measurementsto detect the subtle differences that exist with regions, interven-tions, and between subjects. Obtaining precise measurements iscomplicated by the complex nature of AF electrograms. The CLIoffers some improvement in this regard over other commonlyused techniques and can be used in either offline or real-timesettings.

REFERENCES

[1] M. Haıssaguerre, P. Sanders, M. Hocini, L. F. Hsu, D. C. Shah, C. Scavee,Y. Takahashi, M. Rotter, J. L. Pasquie, S. Garrigue, J. Clementy, andP. Jaıs, “Changes in atrial fibrillation cycle length and inducibility duringcatheter ablation and their relation to outcome,” Circulation, vol. 109,pp. 3007–3013, 2004.

[2] P. Sanders, O. Berenfeld, M. Hocini, P. Jaıs, R. Vaidyanathan, L. F. Hsu,S. Garrigue, Y. Takahashi, M. Rotter, F. Sacher, C. Scavee, R. Ploutz-Snyder, J. Jalife, and M. Haıssaguerre, “Spectral analysis identifies sitesof high-frequency activity maintaining atrial fibrillation in humans,” Cir-culation, vol. 112, pp. 789–797, 2005.

[3] J. Sahadevan, K. Ryu, L. Peltz, C. M. Khrestian, R. W. Stewart,A. H. Markowitz, and A. L. Waldo, “Epicardial mapping of chronic atrialfibrillation in patients: Preliminary observations,” Circulation, vol. 110,pp. 3293–3299, 2004.

[4] S. R. Dibs, J. Ng, R. Arora, R. S. Passman, A. H. Kadish, andJ. J. Goldberger, “Spatiotemporal characterization of atrial activation inpersistent human atrial fibrillation: Multisite electrogram analysis andsurface electrocardiographic correlations—A pilot study,” Heart Rhythm,vol. 5, pp. 686–693, 2008.

[5] S. Lazar, S. Dixit, F. E. Marchlinski, D. J. Callans, and E. P. Gerstenfeld,“Presence of left-to-right atrial frequency gradient in paroxysmal but notpersistent atrial fibrillation in humans,” Circulation, vol. 110, pp. 3181–3186, 2004.

[6] A. Michelucci, P. Bartolini, G. Calcagnini, F. Censi, A. Colella, S. Morelli,L. Padeletti, P. Pieragnoli, and V. Barbaro, “Mapping the organization ofatrial fibrillation with basket catheters. Part II: Regional patterns in chronicpatients,” Pacing Clin. Electrophysiol., vol. 24, pp. 1089–1096, 2001.

[7] D. Filgueiras-Rama, N. F. Price, R. P. Martins, M. Yamazaki, U. M. Avula,K. Kaur, J. Kalifa, S. R. Ennis, E. Hwang, V. Devabhaktuni, J. Jalife,and O. Berenfeld, “Long-term frequency gradients during persistent atrialfibrillation in sheep are associated with stable sources in the left atrium,”Circulation Arrhythm Electrophysiol., pp. 5:1160–1167, 2012.

[8] R. Mandapati, A. Skanes, J. Chen, O. Berenfeld, and J. J. , “Stable mi-croreentrant sources as a mechanism of atrial fibrillation in the isolatedsheep heart,” Circulation, vol. 101, pp. 194–199, 2000.

[9] J. Ng, A. H. Kadish, and J. J. Goldberger, “Effect of electrogram charac-teristics on the relationship of dominant frequency to atrial activation ratein atrial fibrillation,” Heart Rhythm, vol. 3, pp. 1295–1305, 2006.

[10] J. Ng, A. H. Kadish, and J. J. Goldberger, “Technical considerations fordominant frequency analysis,” J. Cardiovascular Electrophysiol., vol. 18,pp. 757–764, 2007.

[11] O. Berenfeld, S. Ennis, E. Hwang, B. Hooven, K. Grzeda, S. Mironov,M. Yamazaki, J. Kalifa, and J. Jalife, “Time- and frequency-domain anal-yses of atrial fibrillation activation rate: The optical mapping reference,”Heart Rhythm, vol. 8, pp. 1758–1765, 2011.

[12] G. W. Botteron and J. M. Smith, “A technique for measurement of theextent of spatial organization of atrial activation during atrial fibrillationin the intact human heart,” IEEE Trans. Biomed. Eng., vol. 42, no. 6,pp. 579–586, Jun. 1995.

[13] J. Slocum, A. Sahakian, and S. Swiryn, “Computer discrimination ofatrial fibrillation and regular atrial rhythms from intra-atrial electrograms,”Pacing Clin. Electrophysiol., vol. 11, pp. 610–621, 1988.

[14] M. Holm, R. Johansson, S. B. Olsson, J. Brandt, and C. Luhrs, “A newmethod for analysis of atrial activation during chronic atrial fibrillation inman,” IEEE Trans. Biomed. Eng., vol. 43, no. 2, pp. 198–210, Feb. 1996.

[15] L. Faes, G. Nollo, R. Antolini, F. Gaita, and F. Ravelli, “A method for quan-tifying atrial fibrillation organization based on wave-morphology similar-ity,” IEEE Trans. Biomed. Eng., vol. 49, no. 12, pp. 1504–1513, Dec.2002.

[16] S. Lee, K. Ryu, A. L. Waldo, C. M. Khrestian, D. M. Durand, andJ. Sahadevan, “An algorithm to measure beat-to-beat cycle lengths forassessment of atrial electrogram rate and regularity during atrial fibrilla-tion,” J. Cardiovascular Electrophysiol., vol. 24, pp. 199–206, 2013.

[17] J. J. Rieta and F. Hornero, “Comparative study of methods for ventricularactivity cancellation in atrial electrograms of atrial fibrillation,” Physiol.Meas., vol. 28, pp. 925–936, 2007.

Authors’ photographs and biographies not available at the time of publication.