Interpreting Activation Mapping of Atrial Fibrillation: A Hybrid

Computational/Physiological Study

FRANCISCO SAHLI COSTABAL , JUNAID A. B. ZAMAN, ELLEN KUHL , and SANJIV M. NARAYAN

Stanford University, Stanford, CA, USA

(Received 13 September 2017; accepted 23 November 2017; published online 6 December 2017)

Associate Editor Estefanıa Pena oversaw the review of this article.

Abstract—Atrial fibrillation is the most common rhythmdisorder of the heart associated with a rapid and irregularbeating of the upper chambers. Activation mapping remainsthe gold standard to diagnose and interpret atrial fibrillation.However, fibrillatory activation maps are highly sensitive tofar-field effects, and often disagree with other opticalmapping modalities. Here we show that computationalmodeling can identify spurious non-local components ofatrial fibrillation electrograms and improve activation map-ping. We motivate our approach with a cohort of patientswith potential drivers of persistent atrial fibrillation. In acomputational study using a monodomain Maleckar model,we demonstrate that in organized rhythms, electrogramssuccessfully track local activation, whereas in atrial fibrilla-tion, electrograms are sensitive to spiral wave distance andnumber, spiral tip trajectories, and effects of fibrosis. In aclinical study, we analyzed n = 15 patients with persistentatrial fibrillation that was terminated by limited ablation. Infive cases, traditional activation maps revealed a spiral waveat sites of termination; in ten cases, electrogram timings wereambiguous and activation maps showed incomplete reentry.By adjusting electrogram timing through computationalmodeling, we found rotational activation, which was unde-tectable with conventional methods. Our results demonstratethat computational modeling can identify non-local deflec-tions to improve activation mapping and explain how andwhere ablation can terminate persistent atrial fibrillation.Our hybrid computational/physiological approach has thepotential to optimize map-guided ablation and improveablation therapy in atrial fibrillation.

Keywords—Electrophysiology, Simulation, Electrogram,

Atrial fibrillation, Rotors, Spiral waves.

INTRODUCTION

Atrial fibrillation is an abnormal, rapid and irreg-ular rhythm of the atria, the upper chambers of the

heart. To restore normal heart rate and rhythm,medications are typically the first treatment of choice.An alternative to medication is atrial fibrillationablation, a procedure that selectively scars or destroysspecific regions of the atria to disrupt rhythm disor-ders.4 Despite technical advances, mechanistic debatein atrial fibrillation continues to hinder major break-throughs.47 A major confounding factor is that theobserved fibrillation mechanisms—and with them thetargets of ablation—are highly sensitive to the selectedmapping approach. Optical mapping studies mostlyreveal rotational drivers in animal models32 andhuman atria.16 This work could explain clinical reportswhere local interventions terminate persistent atrialfibrillation.17,27 Conversely, traditional mapping tech-niques based on contour lines of electrogram onsetstypically displays disordered activation patterns.1,22

Adding to the confusion, the same data reveal orga-nized regions in human atrial fibrillation when ana-lyzed using methods that avoid electrogram onsets.37,50

Moreover, far field noise, generated from distantelectrical signals in the heart is captured by localelectrograms, complicating the precise determinationof activation times.

Here we hypothesized that far-field and otherundefined components28 of atrial fibrillation electro-grams alter contour maps. However, these factors canbe identified and filtered using computational tech-niques of wave propagation, tissue physiology, andelectrode size. This hypothesis was motivated by evi-dence that electrode spacing, orientation, and size alterelectrogram shape and impact mapping,7 as do wave-front direction,43 wavefront number, signal type,26 andtissue factors including fibrosis.10

We used a physiologically motived, computationalapproach to map electrograms in atrial fibrillation in acohort of patients with clinical endpoint in whomAddress correspondence to Ellen Kuhl, Stanford University,

Stanford, CA, USA. Electronic mail: ekuhl@stanford.edu

Annals of Biomedical Engineering, Vol. 46, No. 2, February 2018 (� 2017) pp. 257–269

https://doi.org/10.1007/s10439-017-1969-3

0090-6964/18/0200-0257/0 � 2017 Biomedical Engineering Society

257

limited ablation in a defined zone terminated persistentatrial fibrillation prior to pulmonary vein isolation.This provides a mechanistically relevant mapping tar-get, since termination of persistent atrial fibrillation bylocalized intervention17 is uncommon by randomablation.47 Our ultimate goal is to develop physiolog-ically-tailored rules to map atrial fibrillation based onelectrogram activation times.

METHODS

Clinical Study

We studied n = 15 patients with persistent atrialfibrillation undergoing ablation for routine indica-tions.4 In these cases ablation at sites identified aspotential fibrillation drivers terminated fibrillation tosinus rhythm or atrial tachycardia prior to pulmonaryvein isolation. Persistent atrial fibrillation was definedas atrial fibrillation for> 7 days without self-termi-nation (n = 13); longstanding persistent atrial fibrilla-tion was defined as continuous for> 12 months(n = 2). Patients were recruited from StanfordUniversity Hospital, Palo Alto, CA and University ofCalifornia, San Diego, CA. The study was approvedby both local IRBs.

Electrophysiological Study

Each patient in this series exhibited termination ofpersistent atrial fibrillation by ablating within a definedregion, as a reference for computational analyses. Pa-tients on anti-arrhythmic medications had these agentswithheld for five half-lives (in patients previously onamiodarone: median 200 days, range 90–1980 days).Heparin was infused to maintain an activated clottingtime> 300 s. Endocardial baskets (64 poles; FIRMap,Topera, Inc.) were advanced transvenously and posi-tioned in right then left atrium for atrial fibrillationmapping, verified by fluoroscopy and three-dimen-sional imaging (NavX, St. Jude Medical, St. Paul, MN;Carto, Biosense-Webster, Diamond Bar, CA). Map-ping used a phase-based approach,27 which has beenreported to identify fibrillation drivers in areas of 2–3 cm2 where ablation can eliminate atrial fibrilla-tion.15,25,41,44 Radiofrequency ablation applied withinthese regions terminated fibrillation to sinus rhythm oratrial tachycardia.

Data Acquisition

Electrograms in atrial fibrillation were recorded for1 min from 64-channel baskets (Bard LabSystem Pro,or GE Prucka Cardiolab) at 1000 Hz sampling rate,

filtered at 0.05–500 Hz. The mapping epoch usedprospectively to guide ablation was exported foranalysis using tools developed specifically for thisstudy. Analysis comprised creating contour maps ofencompassing sites where ablation later terminatedatrial fibrillation. We created maps using publishedmethods for traditional mapping. Activation timeswere assigned using criteria of minimum dV/dt, i.e.,maximum negative slope, within 50 ms starting fromthe local absolute maximum of unipolar signals.20 Weapplied a 100 ms blanking period to avoid analysiswithin repolarization.21 Figure 1 shows a map in apatient with persistent atrial fibrillation prior to ter-mination by localized ablation. Electrogram voltagesare shown in black, with continuous dV/dt in green.The activation time of each electrogram is indicated bythe local dV/dt minimum (vertical red lines), andindicates a rotational circuit where ablation terminatedatrial fibrillation. Maps were analyzed to examine sitesof termination. In this case, the signals are complex inthe sense that there are multiple candidates to markactivation times in a window of 100 ms. This com-plexity can be caused by far field noise and/or theintricacy of the AF electrical waves, which obscuresassignment of activation times. Depending on whichdeflection is selected for mapping, the apparentmechanism at the site of termination varies from afocal or partial reentry circuit to complete reentry. Wethus turned to realistic computer modeling to explainthe genesis of various electrogram deflections in atrialfibrillation. This tool is used to develop modified rulesto assign deflections that reflect local activation formapping.

Computational Model

To simulate spiral waves, we adopted the classicalmonodomain equations.9,12,14,18 This model is a par-ticular case of the bidomain model, where the ortho-tropic conductivity ratios between the extra- and intra-cellular domains are equal. We implemented an explicitfinite difference scheme, parallelized with OpenMP.For all simulations we used a time step size of 0.005 msand a mesh resolution of 0.1 mm. The conductivityvalue was set 0.13 mm2/ms. We employed the Mal-eckar model,24 a recent update of the Nygren model30

to represent the membrane kinetics of human atrialcells. To represent the action potential in the electri-cally remodeled atria, we modified the conductancesaccording to changes observed in myocytes subjectedto AF8,31,40,45,46: 70% reduction in L-type calciumcurrent, 50% reduction in transient outward current,50% reduction in ultra-rapid delayed potassium cur-rent and 100% increase in inward rectifier potassium

SAHLI COSTABAL et al.258

current. We advanced all models in time using an ex-plicit time integration.42 We initialized the cellularmodel by pacing for 1000 cycles at cycle length of500 ms to establish steady state conditions. To initiatea spiral wave, we used a S1–S2 stimulus protocol: Weapplied an S1 stimulus in the left boundary of thedomain with 5 ms pulse duration, followed by an S2stimulus in the lower left quadrant coupled at 215 msfor the baseline model and 155 ms for atrial fibrilla-tion. We allowed 1000 ms for the wave to stabilizebefore analyzing subsequent waves of activation andspirals.

Simulation of Electrograms

In contrast to prior modeling approaches based ontheoretical point electrodes, we considered electrodesof varying finite size. We hypothesized that the elec-trode size is a major parameter to interpret clinicaldata. Human atrial fibrillation has been studied usingelectrodes with a diameter of 0.3 mm and spacing from2.25 to 3.50 mm to approximate the scale of spiralwave trajectories and detect of far-field signals.1,2,11 Ingeneral, the potential /e of an electrode with coordi-nates xe = [xe, ye, ze] can be computed as a volumeintegral of the current sources Im and a transfer func-tion Z,

/e xeð Þ ¼Z

B

ImðxÞZ xe; xð ÞdB with x ¼ ½x; y; z�:

For the case of finite-size electrodes, we idealized theelectrode as a plate with constant potential. Then, theanalytical solution of the transfer function is19

Z xe; xð Þ ¼ 1

2predearcsin

� deffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðr� de=2Þ2 þ z2e

qþ

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðrþ de=2Þ2 þ z2e

q0B@

1CA;

where r ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðx� xeÞ2 þ ðy� yeÞ2

q; de is the electrode

diameter and re is the conductivity of the mediumsurrounding the tissue, which we assume to be equal toone. For comparison, we also computed electrodesusing the common point electrode transfer function6,39

Z xe; xð Þ ¼ 1

4preffiffiffiffiffiffiffiffiffiffiffiffiffiffir2 þ z2e

p :

Supplementary Fig. 1 shows transfer functions fordifferent electrode diameters. Transfer functions forplate electrodes are a plateau within the electrodediameter, meaning that all currents generated belowthe electrode are weighed equally. Point electrodes,

FIGURE 1. Mapped mechanisms at the site of atrial fibrillation termination by ablation depend on which electrogram deflection isselected. (a) Four atrial fibrillation cycles are shown with annotations of voltages (black trace) in the selected cycle based onminimum negative dV/dt (green trace). D3 has three similar deflections. (b) Localized ablation at this site terminated atrial fibril-lation. Mechanisms at the site of termination vary for selection of (i) first or (ii) second deflections, which reveal a focal impulse orpartial rotational circuits, (iii) third deflection, which reveals a complete rotation.

Interpreting Activation Mapping of Atrial Fibrillation 259

however, weigh currents below the center of the elec-trode with a nearly infinite value, capturing only localactivation. To avoid divergence, we assumed ze equalto 0.001 mm. We created an interactive software suitein Python to calculate electrograms. For interactiveanalysis, electrodes can be constructed at any positionwithin the simulated domain, and the local actionpotential and corresponding electrogram are immedi-ately calculated and displayed. This allows us to effi-ciently scrutinize electrograms in the vicinity of anyfeature including rotor tips.

Simulation of Fibrosis

We simulated fibrosis by growing an arbitrary re-gion from the center of the patch using a breadth firstsearch algorithm. Briefly, this algorithm first starts in aroot cell and explores the neighboring cells first, thenrecursively explores the neighbors of all explored cells,until a maximum number of iterations has beenreached. Without any modifications, this algorithmwould create a squared region. To create an amor-phously shaped region, each cell was explored ran-domly. Then, a fraction of the points within this regionwere randomly assigned as fibrotic, using a uniformdistribution. We included the effect of fibrotic tissue atthe cellular level using two different approaches: first,in all elements in the fibrotic region, we set the con-ductivity and the action potential source term equal tozero.48 In this case, fibrotic tissue acts as isolatingbarriers. Second, we considered patchy fibrosis inter-spersed with viable tissue, by simulating fibrosis aschanges in action potential. We further remodeled theatrial fibrillation version of Maleckar model to repli-cate experimentally observed changes during fibrosis.33

We reduced the inward rectifier potassium current by50%, reduced the L-type calcium current by 50%, re-duced the sodium current by 50%, and decreased theconductivity in the fibrotic cells by 50%.52

Quantification of Electrogram Characteristics

To systematically characterize electrograms in thevicinity of spiral waves, we computed signals every1 mm in an area that included the entire trajectory ofthe spiral with a margin of 1 mm, using electrodediameters of 0, 1, 3 and 5 mm. For fibrotic cases, wedefined this area as the patch with 1 mm of margin.For these signals, we computed activation timesautomatically using minimum dV/dt.36 For more thanone deflection in a window of 100 ms, we chose thedeflection with the minimum dV/dt to generate uniqueactivation maps. We labeled these cases as multipledeflections and stored them for further analysis. We

also defined the local activation time using the maxi-mum dV/dt of closest cell’s action potential. We cal-culated the differences between activation timescomputed by electrograms and local activation timesand related them to distance to the spiral tip at the timeof activation and magnitude of dV/dt.

Statistical Analysis

We used the Welch’s t test for independent samplesto determine if there were significant differencesbetween single and double deflection electrogrammarkings. We used a single group t-test to evaluatedifferences between electrodes sizes, by subtracting theerrors for the same the activation. The null hypothesisis that the mean of the difference in errors is zero. Forsample groups of size smaller than 20, we performedthe Shapiro–Wilk test for normality before we used thet-test. We used Bonferroni correction to account forthe multiple comparisons that we performed (n = 30).We then set the level of significance to 0.05/30 = 0.0016. We calculated the Pearson correlationcoefficient to estimate a linear dependency betweenvariables with its corresponding p-value to test thesignificance of the correlation. All statistics were per-formed in Python using the Scipy stats package.

RESULTS

Clinical Results and Mapping

Table 1 shows the characteristics of all subjects inthe clinical study. In each patient, ablation at onelocalized source region alone terminated atrial fibril-lation. Figure 1 indicates electrograms at the region oftermination in a 54 year old man with persistent atrialfibrillation. In panel a, some channels show simpleelectrograms while others show multiple componentsand low amplitude. In panel b, ablation in the regionproducing these electrograms terminated atrial fibril-lation to sinus rhythm. In panel c, the precise mecha-nism reported at the site of termination depends clearlyon which electrogram component is marked for themap at site D3. Local activation is traditionally as-signed at maximum—dV/dt. Marking (i) the first or (ii)second deflections at D3 made this site early consistentwith a focal site or breakthrough or a map of partialreentry. Marking (iii) the third deflection at D3 pushedthis site late, producing a map of complete rotation atthe site of atrial fibrillation termination. Overall, five of15 cases demonstrated rotational activity, which mayexplain atrial fibrillation termination at these sites.However, ten of 15 maps using activation timings re-

SAHLI COSTABAL et al.260

vealed only partial rotations or disordered activation,which does not readily explain atrial fibrillation ter-mination. We applied modeling to study whetherambiguities in assigning electrogram onset times couldbe improved by understanding the genesis of multiplesignal components. Our goal was to study whetherdeterministic selection of local over non-local deflec-tions may enable the development of novel rules foraccurate mapping.

Simulated Tip Trajectories and Effects on Electrograms

Electrograms were sensitive to spatial stability of thearrhythmia circuit. In organized atrial flutter, non-fibrillatory reentry in a fixed trajectory producedelectrograms with regular shape throughout the do-main, Supplementary Fig. 2. This was true for allelectrode diameters, 0–5 mm. As expected, activationmapping of electrograms in atrial flutter robustlyidentifies similar mechanisms regardless of recordingparameters. Conversely, spatially precessing waves andspiral tips in atrial fibrillation caused varying electro-gram morphology and deflections, which could dra-matically alter mapping. Figure 2 depicts a spiral wavein atrial fibrillation with tip precession in a ‘‘flower’’trajectory51 (Fig. 2a, panels 1–6), with resulting irreg-ular electrograms (Fig. 2b). Double deflections weredetected (Fig. 2c), which often did not reflect localactivation (phase 0, upstroke). Additional deflectionsvaried with the size of the recording electrode,including far-field activation from nearby wave frontsand other effects.

Electrogram Sensitivity with Respect to ElectrodeDiameter

Atrial fibrillation electrogram components reflectedthe capture of variable and complex propagationwaves by variably sized electrodes. In Fig. 2a, an

advancing wave front of activation (warm colors, pa-nel a1) inscribes a deflection which is small on thetheoretical electrode (0 mm) but larger on 1–5 mmelectrodes and distinct from local activation (time 1 vs.action potential upstroke in panel c). The seconddeflection (time 2, panels b, c) represents activationfrom the tight turn of the wave tip. This is likely notpart of repolarization because despite its comparableamplitude in the isopotential map (time 2, panel a), thewave is propagating forward. At times 3–4, panel a, thewave tip and wave tail produce simple activation withnon-complex electrograms at the electrode site thatreflect local activation (panel c). The fifth deflection(time 5 in panel b) represents the nearby passing wave(time 5, panel a and time 5, panel c) and is difficult toseparate in larger electrodes from the next true acti-vation at time 6 (panel c). We conclude that multipleelectrogram deflections in atrial fibrillation may rep-resent low amplitude depolarization and far field ef-fects, which are particularly confusing for clinicallysized electrodes and may contribute to the complexsignals observed in human atrial fibrillation in Fig. 1.

Electrogram Sensitivity with Respect to Fibrosis andElectrode Size

Structural remodeling, simulated as fibrosis, furtherincreased spiral wave precession and electrogramcomplexity (Fig. 3; Supplementary Fig. 3). Figure 3indicates exaggeration of timing errors in atrial fibril-lation electrograms with fibrosis of 50%, despite rela-tively simple spiral wave activation, which is similar toFig. 2. Figure 3a portrays spiral wave snapshots(panels a1–3) through the fibrotic region (panel d). Inpanel c, action potentials were impacted by the inter-action between fibrotic and non-fibrotic regions withaltered conductivity. Action potentials no longer pre-sent clear upstrokes at sites around which the spiralwave rotated. Electrograms from all electrode sizeswere complex and irregular in time (panel b). At time1, the 3 and 5 mm electrodes capture the nearbypassing wave front, but the 0 and 1 mm electrodescapture a deflection that does not represent the spiralwave at time 2.

Supplementary Fig. 3 shows a parallel analysis inthe presence of 15% fibrosis (panel d). Permeation ofthe spiral wave through these regions exaggerated theirtrajectory and produced electrograms with multipledeflections. Compared to point electrodes (Fig. 3b),larger electrodes recorded electrograms that no longertime-aligned with local activation (action potentialupstroke; Figs. 3b and 3c). The inaccuracy in electro-gram timing worsened with electrode size, with thetime difference between deflections being 31 ms for a

TABLE 1. Patient demographics.

Patient characteristics Mean ± SD

Age (years) 62.0 ± 10.0

Gender (% male) 93

LA diameter (mm) 48.5 ± 8.4

AF history (years) 6.06 ± 4.23

LVEF (%) 53.9 ± 11.0

Hypertension (%) 87

BMI (kg/m2) 30.4 ± 5.3

CHADSVASc score 2.3 ± 1.4

Site of termination (%LA/RA) 86.7/13.3

Mode of termination (%SR/AT) 67/33

Interpreting Activation Mapping of Atrial Fibrillation 261

3 mm electrode and 38 ms for a 5 mm electrode whencompared to the point electrode (time point 2). Smallerelectrodes produce lower errors for this time point(< 9 ms), but still produced discrepancies due to ac-tion potential heterogeneity in fibrotic regions. Thus,electrograms are less accurate for local activation inatrial fibrillation with increasing fibrosis, even for rel-atively simple underlying wave dynamics.

Electrogram Accuracy and Clinical RecordingParameters

Figure 4 shows the accuracy of atrial fibrillationelectrogram times for timing local activity based on thenumber of electrode deflections, dV/dt, distance ofelectrode to organized spiral wave core, electrode sizeand the impact of atrial fibrosis. The error between

FIGURE 2. Precessing spiral wave trajectory causes variable atrial fibrillation electrograms; with similar local and far-fieldcomponents on clinically sized electrodes. (a) Isopotential snapshots of the spiral wave. Insets 1–6 show zoomed views with blacklines representing spiral tip trajectory, at times referenced to electrograms. (b) Electrograms for electrodes of diameters marked byconcentric circles in panel (a). Recorded activity depends on electrode size independent of spacing. (c) Transmembrane actionpotentials. Time 1 indicates non-local activity from a passing wave, especially on larger electrodes. At time 2, low amplitude signalsreflect tight turn of the spiral tip. At times 3 and 4, electrograms represent activation. Time 5 represents non-local activation fromthe nearby passing wave, which is difficult to separate from the next true activation at time 6. Vertical black lines mark onset(minimum dV/dt) in each electrogram.

SAHLI COSTABAL et al.262

electrogram timing (dV/dt criteria) and action potentialupstroke increased with electrode size, from0.48 ± 3.92 ms for a point electrode to10.41 ± 13.11 ms for a 5 mm electrode (p< 0.001between all electrodes). When comparing electrogramsby morphology, we observe that signals with doubledeflections had, on average, higher error than singledeflection signals for 3 and 5 mm electrodes(p< 0.001, Fig. 4ai). Signal characteristics also variedwith distance of the electrode to the spiral core, andwith electrode size. Multiple component electrogramswere recorded, on average, closer to the spiral tip thansingle deflections (2.66 ± 1.20 vs. 4.88 ± 2.07 mm,p< 0.001, Fig. 4aii), and were more prevalent in larger

electrodes. Electrograms with two deflections showedsmaller average dV/dt magnitude (Fig. 4aiii), whichdepended on electrode size, with the reduction indeflection magnitude ranging from 62% for the pointelectrode to 38% for the 5 mm electrode (p< 0.001).These trends were exaggerated by increasing fibrosis to15% (Fig. 4b) and 50% (Fig. 4c). Timing errorsbetween electrograms and local action potentialsincreased with electrode size, ranging from0.61 ± 4.64 ms for the point electrode in the 15%fibrosis case to 12.11 ± 15.14 ms for the 5 mm elec-trode in the 50% remodeled fibrosis case when con-sidering all deflections (p< 0.001, when comparingelectrode sizes for each case). Complex electrograms

FIGURE 3. Atrial fibrillation electrograms poorly represents timing of local activity in the presence of substantial fibroticremodeling and for larger electrodes. (a) Spiral wave propagates through fibrotic tissue, with electrodes of varying diametersmarked by concentric circles. Panels 1–3 are isopotential snapshots at marked times; panel (d) indicates fibrotic region. (b) Atrialfibrillation electrograms generated by different electrodes sizes, with black lines marking minimum dV/dt. Large electrodes (3,5 mm) record a non-local deflection of the wave front that is passing nearby at time 1, whereas small electrodes (0, 1 mm) record adeflection that is not a complete activation at time 2 [panel (a), time 2, light blue]. (c) Action potential at the electrode site iscomplex due to the interaction of fibrotic and remodeled tissue.

Interpreting Activation Mapping of Atrial Fibrillation 263

were also more prevalent closer to the spiral tip(p< 0.001, Figs. 5b and 5cii) with lower dV/dt mag-nitude (p< 0.001, Figs. 4b and 4ciii).

Electrogram Deflections that Represent LocalActivation

Figure 5 summarizes analyses to infer the distancefrom the electrode to a spiral tip core, using timespacing between deflections. These two variablesshowed a moderate linear correlation (r = 0.50–0.71,p< 0.001) with 95% prediction intervals shown for

different electrodes sizes in gray. We systematicallystudied the impact of using the second electrogramdeflection to indicate local activation rather than tra-ditional minimum dV/dt. In 63% of cases for eachelectrode size of incorrect markings, the first deflectionhad a greater negative dV/dt than other deflections andwas selected as activation time. Figure 2b illustratesthis effect at times 1–2 for diameters> 1 mm. The firstdeflection in these incorrect markings had a lower dV/dt magnitude than in cases of correct markings, inwhich the first deflection with highest negative dV/dtminimized the error with respect to local activation

A

B

C

FIGURE 4. Errors in atrial fibrillation timing, number of electrogram deflections, and electrogram upstroke velocity varying withelectrode size. For all simulations [panels (a)–(c)], electrograms with two deflections present increased error with respect to thelocal activation times, were further from the spiral tip and decreased dV/dt of the deflection. Significant differences are marked with(*).

SAHLI COSTABAL et al.264

(p< 0.001 for electrodes of 3 and 5 mm). We deter-mined a dV/dt threshold to most accurately identifyelectrogram timings, i.e., that minimized the sum offalse negatives and false positives in the set of simu-lated electrograms. Activation times where dV/dt ex-ceeded threshold were considered accurate, while thosewith dV/dt below threshold were replaced by timings atthe second deflection. We tested this methodology in anew set of electrograms with an offset of 0.5 mm inboth directions with respect to the original electro-grams. Although this new set of data was obtainedfrom the same simulation, it can be used to assess thevalidity of the method without losing the characteristicof the AF model. This new approach (Fig. 6) reducedthe average timing error for local activation by 33% inelectrodes of 3 and 5 mm (p< 0.001), using thresholdsof 2546.2 and 1468.4 mV/ms, respectively, comparableto the error of single-deflection electrograms (Fig. 4ai).

Adjusting Electrogram Timing

Figure 7 provides another example of how compu-tational–physiological rules for marking electrogramsfrom this study modified reported mechanisms in thisseries. This may resolve the ambiguity highlighted inFig. 1 where different timings alter interpretation ofatrial fibrillation mechanisms at the termination site.In this 67 year old man with persistent atrial fibrilla-tion, annotating the first electrogram deflection pro-duces a map of multiple foci at sites C7 and BC5.Applying our modified rules to unipolar electrogramswith low amplitude and patient specific dV/dt thresh-old of 0.7 mV/ms produced a map consistent with acomplete rotational circuit at site CD56. Prospectiveablation in the case at site CD56, identified by a sep-arate method that did not require assignment of elec-

trogram onset (phase mapping), terminated atrialfibrillation to sinus rhythm.

DISCUSSION

This study combines clinical mapping with compu-tational modeling to improve mapping of human car-diac fibrillation. We established in silico simulations toidentify physiological contributions to local and far-field atrial fibrillation electrograms. Our analysis re-vealed components from wave curvature and perme-ation through fibrosis, which were indistinguishablefrom local depolarization. We developed novel elec-trogram shape criteria to attenuate far-field compo-

FIGURE 5. Estimation of distance to spiral tip based on time difference between electrogram deflections in atrial fibrillation. Alinear trend (p<0.001, n> 83 for all electrodes) was observed for each electrode size, with a positive correlation between the timeinterval between deflections and distance to the spiral tip at the time of the maximum negative dV/dt. 95% Prediction intervals areshown in gray and have a range of 3 mm for any given time difference.

FIGURE 6. Improved prediction of activation times in low dV/dt electrograms. By always selecting the second deflection aslocal activation time in cases of low dV/dt magnitude, we re-duced the error to that seen in single deflection electrogramsusing clinically sized electrodes (independent t-test: p< 0.001,3 mm electrode: n 5 123, 5 mm electrode: n 5 149).

Interpreting Activation Mapping of Atrial Fibrillation 265

nents. In a series of clinical cases, we demonstratedhow traditional rules may produce complex maps ofatrial fibrillation at sites where ablation terminatedpersistent atrial fibrillation. At the same time, ournovel combined computational approach revealedorganized patterns including rotational activity inagreement with phase mapping at sites where ablationterminated atrial fibrillation. Future studies can buildupon this approach to improve mapping of complexarrhythmias and ultimately guide therapy.

Electrograms Accuracy for Mapping Simple vs.Complex Arrhythmias

This study is among the first to address whyreported mechanisms for atrial fibrillation are highlysensitive to the underlying mapping technique. Inorganized, non-fibrillatory arrhythmias, where elec-trogram shape is regular across the mapped domain,this sensitivity is less pronounced which explains why,for example, atrial flutter circuits are revealed consis-tently by all electrode sizes and approaches with min-imal differences. In atrial fibrillation, however,reported mechanisms vary dramatically with mappingmethod: optical mapping consistently show localizeddrivers in human23 and other models of atrial fibrilla-tion, where ablation can terminate fibrillation and ex-plain clinical data.13,28 Conversely, traditionalmapping shows mostly disorganized waves.22,24 Tra-

ditional mapping is more straightforward when repo-larization data are available to filter far-fieldcomponents, either with monophasic action poten-tials26 or via optical mapping.23 However, there arecurrently no mapping criteria to separate local fromfar-field activity in atrial fibrillation.

Identifying Local and Non-Local Activity inElectrograms

Here we established methods to separate far-fieldfrom local electrograms in atrial fibrillation, withoutthe need for a reference channel of repolarizationdata29 which is rarely available. First, we identified abias in traditional mapping towards detecting earliercomponents than true local activation, particularlywhen using larger clinical electrodes. This may explainnumerous reports of early ‘foci’ or ‘breakthroughs’.Simply selecting the second deflection—if electrogramson the first deflection had dV/dt below our determinedthreshold-improved accuracy of local activation time.Second, while atrial fibrillation electrograms frommeandering rotor trajectories or fibrotic substrates arecomplex, double deflections may occur near a spiralwave tip with low amplitude and produce a large errorin estimating local activation time. Third, our simula-tions suggest that it may be possible to infer the dis-tance of a specific electrogram to a rotor core byconsidering the time interval between deflections, with

FIGURE 7. Applying a second deflection criterion changes the detection of atrial fibrillation mechanisms. (a) In this 67 years oldman marking the channels by maximum negative dV/dt (red lines), gives a map with a focal pattern of atrial fibrillation (b). However,in those channels with low dV/dt (C4–C6), applying the proposed rule of taking the second plausible negative dV/dt reveals arotational circuit (c), where earliest activation (red) meets latest activation (blue). Ablation at this site terminated atrial fibrillation.

SAHLI COSTABAL et al.266

narrower spacing indicating a closer proximity. Theseresults may assist in accurately marking electrogramsfor mapping, and also for finding an ablation target.

Next Clinical Steps and Limitations

Rules proposed by our hybrid computa-tional/physiological approach for improving atrialfibrillation mapping should be tested clinically. Thiscould be done prospectively in patients undergoingablation, or in blinded fashion retrospectively inpatients where localized ablation terminates persistentatrial fibrillation, similar to this study.

This study has several limitations: first, we onlysimulated two-dimensional surfaces to represent theatria and we used a simplified fibrotic model. In futurestudies, we plan to create patient specific geometriesand incorporate personalized fibrotic lesions frommagnetic resonance images.3,5 This will allow us tocharacterize secondary effects of atrial fibrillation, e.g.,mitral annular dynamics34,35 and left ventricularfunction.3 Second, we only simulated one spiral wave,as opposed to a more chaotic response observed clin-ically. However, it has been shown that many ad-vanced electrophysiology models, such as the one usedin this study, tend to be stable under atrial fibrilla-tion.51 Moreover, we have shown that even for a singlesource during a single cycle of fibrillation, there isconsiderable error, often up to 50 ms when estimatinglocal activity with electrograms. This has implicationson mechanisms and their stability for each successivecycle and importantly will impact therapy delivery.Third, we have disregarded the mechanical contractionof the tissue.13,49 Tissue deformation could induceadditional sources of fibrillation38 and disease pro-gression also alters the electro-mechanical propertiesof the tissue.23 Fourth, clinically, our cohort was notprospectively studied; this is the focus of our currentefforts.

CONCLUSION

In conclusion, we have identified physiologicalcontributions to local and far-field fibrillatory elec-trograms in atrial fibrillation using computationalsimulations. Our simulations define novel rules to mapatrial fibrillation. The major feature of these rules is toattenuate far-field contributions. Our pilot clinicalstudy revealed that traditional rules were incapable ofidentifying sites where ablation terminated persistentatrial fibrillation. Our novel rules reliably identifiedrotational activation patterns, which were consistentwith sites of successful ablation, and were concordantwith phase mapping. Our study can serve as a scientific

foundation for novel mapping strategies to guideablation therapy.

ELECTRONIC SUPPLEMENTARY MATERIAL

The online version of this article (https://doi.org/10.1007/s10439-017-1969-3) contains supplementary mate-rial, which is available to authorized users.

ACKNOWLEDGMENTS

Francisco Sahli Costabal is supported by a StanfordCardiovascular Institute Seed Grant, by the StanfordSchool of Engineering Fellowship, by the Becas Chile-Fulbright Fellowship, and by the National Institute ofHealth Grant U01 HL119578. Junaid Zaman is sup-ported by a Fulbright British Heart FoundationScholarship. Ellen Kuhl is supported by the NationalInstitute of Health Grant U01 HL119578. SanjivNarayan is supported by the National Institute ofHealth Grants R01 HL83359 and K24 HL8103800.

CONFLICT OF INTEREST

Sanjiv Narayan is co-author of intellectual propertyowned by the University of California Regents andlicensed to Topera, Inc. He held equity in Topera andreceived honoraria from Medtronic and St. JudeMedical. The other authors declare no conflict ofinterest.

REFERENCES

1Allessie, M. A., N. M. de Groot, R. P. Houben, U.Schotten, E. Boersma, J. L. Smeets, and H. J. Crijns.Electropathological substrate of long-standing persistentatrial fibrillation in patients with structural heart disease:longitudinal dissociation. Circ. Arrhythm. Electrophysiol.3:606–615, 2010.2Anter, E., C. M. Tschabrunn, and M. E. Josephson. High-resolution mapping of scar-related atrial arrhythmias usingsmaller electrodes with closer interelectrode spacing. Circ.Arrhythm. Electrophysiol. 8:537–545, 2015.3Baillargeon, B., N. Rebelo, D. D. Fox, R. L. Taylor, andE. Kuhl. The Living Heart Project: a robust and integrativesimulator for human heart function. Eur. J. Mech. A48:38–47, 2014.4Calkins, H., G. Hindricks, R. Cappato, Y. H. Kim, E. B.Saad, L. Aguinaga, J. G. Akar, V. Badhwar, J. Brugada, J.Camm, P. S. Chen, S. A. Chen, M. K. Chung, J. C. Niel-sen, A. B. Curtis, D. W. Davies, J. D. Day, A. d’Avila, R.Natasja de Groot, L. Di Biase, M. Duytschaever, J. R.Edgerton, K. A. Ellenbogen, P. T. Ellinor, S. Ernst, G.Fenelon, E. P. Gerstenfeld, D. E. Haines, M. Haissaguerre,R. H. Helm, E. Hylek, W. M. Jackman, J. Jalife, J. M.

Interpreting Activation Mapping of Atrial Fibrillation 267

Kalman, J. Kautzner, H. Kottkamp, K. H. Kuck, K.Kumagai, R. Lee, T. Lewalter, B. D. Lindsay, L. Macle,M. Mansour, F. E. Marchlinski, G. F. Michaud, H.Nakagawa, A. Natale, S. Nattel, K. Okumura, D. Packer,E. Pokushalov, M. R. Reynolds, P. Sanders, M. Scana-vacca, R. Scanavacca, R. Schilling, C. Tondo, H. M. Tsao,A. Verma, D. J. Wilber, and T. Yamane. 2017 HRS/EHRA/ECAS/APHRS/SOLAECE Expert ConsensusStatement on Catheter and Surgical Ablation of AtrialFibrillation. Washington, DC: Heart Rhythm Society,2017.5Chabiniok, R., V. Wang, M. Hadjicharalambous, L. As-ner, J. Lee, M. Sermesant, E. Kuhl, A. Young, P. Moireau,M. Nash, D. Chapelle, and D. A. Nordsletten. Multi-physics and multiscale modeling, data-model fusion andintegration of organ physiology in the clinic: ventricularcardiac mechanics. Interface Focus 6:20150083, 2016.6Chouvarda, I., N. Maglaveras, C. Pappas, F. J. L. VanCapelle, and J. DeBakker. Estimation of distance betweena unipolar recording electrode and a myocardial bundlebased on signal characteristics. Ann. Biomed. Eng. 32:1336–1347, 2004.7Correa de Sa, D. D., N. Thompson, J. Stinnett-Donnelly,P. Znojkiewicz, N. Habel, J. G. Muller, J. H. Bates, J. S.Buzas, and P. S. Spector. Electrogram fractionation: therelationship between spatiotemporal variation of tissueexcitation and electrode spatial resolution. Circ. Arrhythm.Electrophysiol. 4:909–916, 2011.8Courtemanche, M., R. J. Ramirez, and S. Nattel. Ionictargets for drug therapy and atrial fibrillation-inducedelectrical remodeling: insights from a mathematical model.Cardiovasc. Res. 42:477–489, 1999.9Dickopf, T., D. Krause, R. Krause, and M. Potse. Designand analysis of a lightweight parallel adaptive scheme forthe solution of the monodomain equation. SIAM J. Sci.Comput. 36:C163–C189, 2014.

10Gharaviri, A., M. Potse, S. Verheule, R. Krause, A.Auricchio, and U. Schotten. Epicardical fibrosis explainsincreased transmural conduction in a computer model ofatrial fibrillation. Comput. Cardiol. 43:237–240, 2016.

11Ghoraani, B., R. Dalvi, S. Gizurarson, M. Das, A. Ha, A.Suszko, S. Krishnan, and V. S. Chauhan. Localized rota-tional activation in the left atrium during human atrialfibrillation: relationship to complex fractionated atrialelectrograms and low-voltage zones. Heart Rhythm10:1830–1838, 2013.

12Goktepe, S., and E. Kuhl. Computational modeling ofelectrophysiology: a novel finite element approach. Int. J.Numer. Methods Eng. 79:156–178, 2009.

13Goktepe, S., and E. Kuhl. Electromechanics of the heart: aunified approach to the strongly coupled excitation–con-traction problem. Comput. Mech. 45:227–243, 2010.

14Goktepe, S., J. Wong, and E. Kuhl. Atrial and ventricularfibrillation: computational simulation of spiral waves incardiac tissue. Arch. Appl. Mech. 80:569–580, 2010.

15Haissaguerre, M., M. Hocini, A. Denis, A. J. Shah, Y.Komatsu, S. Yamashita, M. Daly, S. Amraoui, S. Zeller-hoff, M. Q. Picat, A. Quotb, L. Jesel, H. Lim, S. Ploux, P.Bordachar, G. Attuel, V. Meillet, P. Ritter, N. Derval, F.Sacher, O. Bernus, H. Cochet, P. Jais, and R. Dubois.Driver domains in persistent atrial fibrillation. Circulation130:530–538, 2014.

16Hansen, B. J., J. Zhao, T. A. Csepe, B. T. Moore, N. Li, L.A. Jayne, A. Kalyanasundaram, P. Lim, A. Bratasz, K. A.Powell, O. P. Simonetti, R. S. Higgins, A. Kilic, P. J.

Mohler, P. M. Janssen, R. Weiss, J. D. Hummel, and V. V.Fedorov. Atrial fibrillation driven by micro-anatomicintramural re-entry revealed by simultaneous sub-epicar-dial and sub-endocardial optical mapping in explantedhuman hearts. Eur. Heart J. 36:2390–2401, 2015.

17Herweg, B., M. Kowalski, and J. S. Steinberg. Terminationof persistent atrial fibrillation resistant to cardioversion bya single radiofrequency application. Pacing Clin. Electro-physiol. 26:1420–1423, 2003.

18Hurtado, D. E., and D. Henao. Gradient flows and vari-ational principles for cardiac electrophysiology: towardefficient and robust numerical simulations of the electricalactivity of the heart. Comput. Methods Appl. Mech. Eng.273:238–254, 2014.

19Jacquemet, V., and C. S. Henriquez. Genesis of complexfractionated atrial electrograms in zones of slow conduc-tion: a computer model of microfibrosis. Heart Rhythm6:803–810, 2009.

20Konings, K., J. Smeets, O. Penn, H. Wellens, and M. Al-lessie. Configuration of unipolar atrial electrograms duringelectrically induced atrial fibrillation in humans. Circulation95:1231–1241, 1997.

21Lau, D. H., B. Maesen, S. Zeemering, P. Kuklik, A.Hunnik, T. A. Lankveld, E. Bidar, S. Verheule, J. Nijs, J.Maessen, H. Crijns, P. Sanders, and U. Schotten. Indices ofbipolar complex fractionated atrial electrograms correlatepoorly with each other and atrial fibrillation substratecomplexity. Heart Rhythm 12:1415–1423, 2015.

22Lee, S., J. Sahadevan, C. M. Khrestian, I. Cakulev, A.Markowitz, and A. L. Waldo. Simultaneous bi-atrial highdensity epicardial mapping of persistent and long-standingpersistent atrial fibrillation in patients: new insights into themechanism of its maintenance. Circulation 132:2108–2117,2015.

23Lee, L. C., J. Sundnes, M. Geriet, J. F. Wenk, and S. T.Wall. An integrated electromechanical-growth heart modelfor simulating cardiac therapies. Biomech. Model. Me-chanobiol. 15:791–803, 2016.

24Maleckar, M. M., J. L. Greenstein, W. R. Giles, and N. A.Trayanova. K+ current changes account for the ratedependence of the action potential in the human atrialmyocyte. Am. J. Physiol. Heart Circ. Physiol. 297:H1398–H1410, 2009.

25Miller, J. M., R. C. Kowal, V. Swarup, J. P. Daubert, E. G.Daoud, J. D. Day, K. A. Ellenbogen, J. D. Hummel, T.Baykaner, D. E. Krummen, S. M. Narayan, V. Y. Reddy,K. Shivkumar, J. S. Steinberg, and K. R. Wheelan. Initialindependent outcomes from focal impulse and rotor mod-ulation ablation for atrial fibrillation: multicenter FIRMregistry. J. Cardiovasc. Electrophysiol. 25:921–929, 2014.

26Narayan, S. M., D. E. Krummen, A. M. Kahn, P. L.Karasik, and M. R. Franz. Evaluating fluctuations inhuman atrial fibrillatory cycle length using monophasicaction potentials. Pacing Clin. Electrophysiol. 29:1209–1218, 2006.

27Narayan, S. M., D. E. Krummen, K. Shivkumar, P.Clopton, W.-J. Rappel, and J. Miller. Treatment of atrialfibrillation by the ablation of localized sources: the con-ventional ablation for atrial fibrillation with or withoutfocal impulse and rotor modulation: CONFIRM trial. J.Am. Coll. Cardiol. 60:628–636, 2012.

28Narayan, S. M., M. Wright, N. Derval, A. Jadidi, A.Forclaz, I. Nault, S. Miyazaki, F. Sacher, P. Bordachar, J.Clementy, P. Jais, M. Haissaguerre, and M. Hocini. Clas-sifying fractionated electrograms in human atrial fibrilla-

SAHLI COSTABAL et al.268

tion using monophasic action potentials and activationmapping: evidence for localized drivers, rate accelerationand non-local signal etiologies. Heart Rhythm 8:244–253,2011.

29Narayan, S. M., J. A. Zaman, T. Baykaner, and M. R.Franz. Atrial fibrillation: can electrograms be interpretedwithout repolarization information? Heart Rhythm 13:962–963, 2016.

30Nygren, A., C. Fiset, L. Firek, J. W. Clark, D. S. Lindblad,R. B. Clark, and W. R. Giles. Mathematical model of anadult human atrial cell: the role of K+ currents in repo-larization. Circ. Res. 82:63–81, 1998.

31Pandit, S. V., O. Berenfeld, J. M. Anumonwo, R. M.Zaritski, J. Kneller, S. Nattel, and J. Jalife. Ionic determi-nants of functional reentry in a 2-D model of human atrialcells during simulated chronic atrial fibrillation. Biophys. J.88:3806–3821, 2005.

32Pandit, S. V., and J. Jalife. Rotors and the dynamics ofcardiac fibrillation. Circ. Res. 112:849–862, 2013.

33Pedrotty, D. M., R. Y. Klinger, R. D. Kirkton, and N.Bursac. Cardiac fibroblast paracrine factors alter impulseconduction and ion channel expression of neonatal ratcardiomyocytes. Cardiovasc. Res. 83:688–697, 2009.

34Rausch, M. K., W. Bothe, J. P. Kvitting, J. C. Swanson, N.B. Ingels, D. C. Miller, and E. Kuhl. Characterization ofmitral valve annular dynamics in the beating heart. Ann.Biomed. Eng. 39:1690–1702, 2011.

35Rausch, M. K., F. A. Tibayan, N. B. Ingels, D. C. Miller,and E. Kuhl. Mechanics of the mitral annulus in chronicischemic cardiomyopathy. Ann. Biomed. Eng. 41:2171–2180, 2013.

36Roney, C. H., C. D. Cantwell, N. A. Qureshi, R. A.Chowdhury, E. Dupont, P. B. Lim, E. J. Vigmond, J. H.Tweedy, F. S. Ng, and N. S. Peters. Rotor tracking usingphase of electrograms recorded during atrial fibrillation.Ann. Biomed. Eng. 45:910–923, 2017.

37Sahadevan, J., K. Ryu, L. Peltz, C. M. Khrestian, R. W.Stewart, A. H. Markowitz, and A. L. Waldo. Epicardialmapping of chronic atrial fibrillation in patients: prelimi-nary observations. Circulation 110:3293–3299, 2004.

38Sahli Costabal, F., F. A. Concha, D. E. Hurtado, and E.Kuhl. The importance of mechano-electrical feedback andinertia in cardiac electromechanics. Comput. Methods Appl.Mech. Eng. 320:352–368, 2017.

39Sahli Costabal, F., D. E. Hurtado, and E. Kuhl. Generat-ing Purkinje networks in the human heart. J. Biomech.49:2455–2465, 2016.

40Sanchez, C., A. Corrias, A. Bueno-Orovio, M. Davies, J.Swinton, I. Jacobson, P. Laguna, E. Pueyo, and B. Ro-driguez. The Na+/K+ pump is an important modulator ofrefractoriness and rotor dynamics in human atrial tissue.AJP Heart Circ. Physiol. 302:H1146–H1159, 2012.

41Sommer, P., S. Kircher, S. Rolf, S. John, A. Arya, B. Di-nov, S. Richter, A. Bollmann, and G. Hindricks. Successfulrepeat catheter ablation of recurrent longstanding persis-tent atrial fibrillation with rotor elimination as the proce-

dural endpoint: a case series. J. Cardiovasc. Electrophysiol.27:274–280, 2016.

42Spiteri, R. J., and R. C. Dean. Stiffness analysis of cardiacelectrophysiological models. Ann. Biomed. Eng. 38:3592–3604, 2010.

43Thompson, N. C., J. Stinnett-Donnelly, N. Habel, B.Benson, J. H. Bates, B. E. Sobel, and P. S. Spector. Im-proved spatial resolution and electrogram wave directionindependence with the use of an orthogonal electrodeconfiguration. J. Clin. Monit. Comput. 28:157–163, 2014.

44Tomassoni, G., S. Duggal, M. Muir, L. Hutchins, K.Turner, A. M. McLoney, and A. Hesselson. Long-termfollow-up of FIRM-guided ablation of atrial fibrillation: asingle-center experience. J. Innov. Card. Rhythm Manag.6:2145–2151, 2015.

45Van Wagoner, D. R., A. L. Pond, M. Lamorgese, S. S.Rossie, P. M. McCarthy, and J. M. Nerbonne. Atrial L-type Ca2+ currents and human atrial fibrillation. Circ. Res.85:428–436, 1999.

46Van Wagoner, D. R., A. L. Pond, P. M. McCarthy, J. S.Trimmer, and J. M. Nerbonne. Outward potassium currentdensities and Kv1.5 expression are reduced in chronichuman atrial fibrillation. Circ. Res. 80:772–781, 1997.

47Verma, A., C. Y. Jiang, T. R. Betts, J. Chen, I. Deisen-hofer, R. Mantovan, L. Macle, C. A. Morillo, W. Ha-verkamp, R. Weerasooriya, J. P. Albenque, S. Nardi, E.Menardi, P. Novak, and P. Sanders. Approaches to ca-theter ablation for persistent atrial fibrillation. N. Engl. J.Med. 372:1812–1822, 2015.

48Vigmond, E., A. Pashaei, S. Amraoui, H. Cochet, and M.Hassaguerre. Percolation as a mechanism to explain atrialfractionated electrograms and reentry in a fibrosis modelbased on imaging data. Heart Rhythm 13:1536–1543,2016.

49Vigueras, G., I. Roy, A. Cookson, J. Lee, N. Smith, and D.Nordsletten. Toward GPGPU accelerated human elec-tromechanical cardiac simulations. Int. J. Numer. MethodsBiomed. Eng. 30:117–134, 2014.

50Walters, T. E., G. Lee, G. Morris, S. Spence, M. Larobina,V. Atkinson, P. Antippa, J. Goldblatt, A. Royse, M.O’Keefe, P. Sanders, J. B. Morton, P. M. Kistler, and J. M.K. Kalman. Temporal stability of rotors and atrial acti-vation patterns in persistent human atrial fibrillation: ahigh density epicardial mapping study of prolongedrecordings. J. Am. Coll. Cardiol. Clin. Electrophysiol. 1:18–25, 2015.

51Wilhelms, M., H. Hettmann, M. M. Maleckar, J. T.Koivumaki, O. Dossel, and G. Seemann. Benchmarkingelectrophysiological models of human atrial myocytes.Front. Physiol. 3:487, 2012.

52Zahid, S., H. Cochet, P. M. Boyle, E. L. Schwarz, K. N.Whyte, E. J. Vigmond, R. Dubois, M. Hocini, M. Hais-saguerre, P. Jais, and N. A. Trayanova. Patient-derivedmodels link re-entrant driver localization in atrial fibrilla-tion to fibrosis spatial pattern. Cardiovasc. Res. 110:443–454, 2016.

Interpreting Activation Mapping of Atrial Fibrillation 269