TECHNICAL ISSUES
Europace (2013) 15, 447–452
doi:10.1093/europace/eus361
Validation of a novel algorithm for quantification
of the percentage of signal fractionation
in atrial fibrillation
Charlotte L. Haley1,2, Lorne J. Gula 1,2, Rodrigo Miranda 1,2, Kevin A. Michael 1,2,
Adrian M. Baranchuk 1,2, Christopher S. Simpson 1,2, Hoshiar Abdollah 1,2,
Adam J. West1,2, Selim G. Akl 1,2, and Damian P. Redfearn 1,2*
1
Division of Cardiology (Heart Rhythm Service), Mathematics and Statistics Department, Faculty of Medicine, Kingston General Hospital, School of Computing, Queen’s University,
Kingston, ON K7L 2V7, Canada; and 2Universidad de la Frontera, Temuco, Chile
Received 3 September 2012; accepted after revision 5 October 2012; online publish-ahead-of-print 15 November 2012
Aims
Catheter ablation for paroxysmal atrial fibrillation (AF) is rapidly becoming a standard practice. There is literature to
support that catheter ablation of persistent AF requires additional ‘substrate modification’. In clinical practice, operators rely on automated fractionation maps created by three-dimensional anatomic mapping systems to rapidly assess
complex ‘fractionated’ signals (CFAE). These systems use differing algorithms to automate the process. The agreement between operators and contemporary algorithms has not been examined. We sought to assess the agreement
between operators and a novel method of quantification calculating percentage fractionation (PF).
.....................................................................................................................................................................................
Methods
Expert opinion on 80 atrial electrogram 4 s signals of varying levels of activity were gathered and pooled for comand results
parison. Twelve independent experts visually quantified the signal fractionation and offered a threshold level for ablation. We developed an algorithm to find sites with high continuous electrical activity, or high PF. Correlation
between experts and PF was 0.78 [P , 0.01, 95% confidence interval (CI) (0.68–0.86)]. Receiver operating characteristics curve sensitivity and specificity for PF were 0.7727 and 0.8103 at the optimal cut-off point of 58.45 PF with
area under curve 0.89 CI (0.80–0.99).
.....................................................................................................................................................................................
Conclusion
The PF statistic represents a more robust and intuitive measure to represent fractionated atrial activity; importantly it
demonstrates excellent agreement with expert users and presents a new standard for algorithm assessment. Use of a
PF statistic should be considered in automated mapping systems.
----------------------------------------------------------------------------------------------------------------------------------------------------------Keywords
Fractionation † Complex fractionated atrial electrograms † Atrial fibrillation † Ablation
Introduction
Catheter ablation for persistent atrial fibrillation (AF) following a
‘substrate modification’ approach with or without pulmonary
vein isolation (PVI) or left linear ablation has been shown to be effective for slowing or termination of arrhythmia with low recurrence rates.1,2 Nademanee et al. reported high success rates for
maintaining post-procedure sinus rhythm in both paroxysmal and
persistent AF patients by targeting complex-fractionated atrial
electrograms (CFAEs). Complex-fractionated atrial electrogram
ablation has since been met with success in subsequent studies,
though there are inconsistent published reports as to the benefits
of CFAE ablation in addition to PVI.1,3,4
The most widely accepted definition of CFAE is given by the criteria of Nademanee et al., that is:
(1) fractionated electrograms composed of two or more deflections and/or a perturbation of the baseline with continuous deflection of a prolonged activation complex; and
(2) atrial electrograms with a very short cycle length (≤120 ms).1
In practice, these guidelines are limited in application because (i)
deflections in the electrogram may be obscured by noise artefact,
* Corresponding author. Tel: +1 613 549 6666; fax: +1 613 548 1387, Email: [email protected]
Published on behalf of the European Society of Cardiology. All rights reserved. & The Author 2012. For permissions please email: [email protected].
448
What’s new?
† A new metric, percentage fractionation (PF), to quantify
fractionation for substrate-based complex fractionated
atrial electrogram ablation is introduced.
† Based on clinical results which show that ablation at
sites with high continuous electrical activity is associated
with positive clinical outcome, PF reflects the percentage
(0–100%) of electrogram signal that is continuously
fractionated.
† Designed to be robust to spurious artefacts and noise, PF
performs favourably in comparison with a contemporary algorithm using the consensus opinion of a panel of experienced electrophysiologists as the gold standard.
† Because PF is designed with the criteria of the expert in
mind, and it agrees well with the opinion of the expert
user, it presents an intuitive tool for use during substrate ablation, and an important indicator for target areas showing a
high degree of continuous fractionation.
catheter movement, and/or improper catheter contact with the
endocardial surface (ii) operator time is often limited to a subjective visual assessment of each electrogram. In practice, operator decision criteria for substrate ablation involve (i) evaluation of
electrogram morphology—is fractionation as prescribed by the
Nademanee criteria present continuously over the entire course
of the signal, or does it appear in discrete bursts? (ii) Knowledge
of anatomic location and instrumentation settings, and (iii) correction for catheter movement, spurious artefacts, and noise.
Electroanatomic mapping systems such as EnSite NavX (St Jude
Medical, MN, USA) and CARTO (Biosense Webster, Diamond
Bar, CA, USA)5,6 can rapidly generate high-density fractionation
maps to provide a global overview of left atrial (LA) electrical activity. These systems guide ablation by way of automated electrogram assessment which mitigates subjective identification of
fractionated sites, and thus may reduce procedure times and risk
for the patient. The EnSite NavX system represents fractionation
on a colour-coded display of the average interval between deflections in the electrogram (CFE-mean). Electrograms are considered
CFAE when CFE-mean ≤120 ms, corresponding to a modified
‘Nademanee criterion’. Verma et al. 7 have confirmed that
CFE-mean guided substrate ablation is effective in slowing and
regularizing atrial fibrillation cycle length.
In a recent study, Takahashi et al.8 found that ablation of persistent AF at sites with electrograms displaying high amounts of ‘continuous electrical activity’ correlate with slowing and/or
termination of the arrhythmia, while the ‘local electrogram cycle
length’ did not compare significantly. An independent report has
reproduced this former result.9 While this study was done by
manual assessment, we propose an automated analogue to calculate the ‘percentage of continuous activity’ or percentage of fractionation (PF) on a scale of 0– 100%, expected to correspond
better with the opinion of the expert user. A panel of experts
from different centres was surveyed and responses for PF and
CFE-mean were compared against consensus expert opinion.
C.L. Haley et al.
We hypothesized that PF, since it rates continuous electrogram activity higher than discrete electrograms and is robust to noise artefact, would correlate well with expert opinion, as (i) substrate
ablation of electrograms with ‘continuous’ high-frequency activity
results in positive clinical outcome (as opposed to fractionated
electrograms separated by an isoelectric segment8 – 10) and (ii)
experienced cardiologists are able to assess CFAE while rejecting
spurious electrograms and noise artefact. Use of an electroanatomic mapping system equipped with the PF rating system would
shorten procedure times because the algorithm applies the same
decision criteria (a –c) as the expert user.
Methods
Study population
Eighteen consecutive patients (6 persistent, 12 paroxysmal, 2 female)
undergoing AF ablation for the first time were asked to participate
in the study. Antiarrhythmic drugs were discontinued for at least five
half-lives before the mapping and ablation procedure; except for amiodarone which was continued in three patients with persistent AF (see
Table 1). In paroxysmal patients, AF was induced by burst pacing as
part of an ongoing study,11 while for the purposes of this study only
the initial map was employed and AF allowed to stabilize for not
less than 5 min before the acquisition of fractionation points. All
patients provided written informed consent. Study methods were
approved by the Queen’s University Health Sciences and Affiliated
Teaching Hospitals Research Ethics Board, and the study complies
with the Declaration of Helsinki.
Using EnSite NavX (St Jude Medical Inc., St Paul, MN, USA) software
and previously acquired cardiac computerized tomography (CT), left
atrial electroanatomic maps were generated. An irrigated, 3.5 mm
tip, ablation catheter (St Jude Medical Inc.) with 2 mm interelectrode
spacing was used to collect signals during AF. Transseptal puncture
was performed and a left atrial geometry was created using a spiral
catheter with reference to the prior CT. On completion of the geometry, data were acquired using a point-by-point approach to ensure a
dense and even distribution of the LA endocardium. For the purposes
of this study, right atrial and coronary sinus data were not collected.
Data acquisition
With the ablation catheter in stable position, 4 s data segments (n ¼
265.6 + 116.4 per patient—paroxysmal and persistent AF cases combined, or 204.5 + 98.8 for paroxysmal and 341.9 + 91.7 for persistent
patients.) were collected at 1200 Hz and analogue filtered from 0.5–
250 Hz by the EnSite NavX preprocessing system, with 60Hz powerline interference removal.12
Table 1 Baseline characteristics of patients
Number of patients
Age
18
57.8 + 11.1
AF history (years)
6.4 + 3.0
LA diameter (mm)
Mean CL AF in LA
43.2 + 5.8
190 + 89
Persistent
6/18
Validation of a novel algorithm for quantification of the percentage of signal fractionation in AF
449
Figure 1 Results of the preprocessing steps on a fractionated electrogram. The top graph is the original signal, below is the preprocessed
output with signal deflections annotated. The total time the signal is fractionated is highlighted in red; divide this time by the total signal
time to get the percentage fractionation. The signal represents percentage fractionation of 65%, which (by later receiver operating characteristics analysis) categorizes it as complex-fractionated atrial electrogram. Significant baseline noise is also clearly present.
Expert annotation
The data from the patient cohort were examined and each sample was
inspected in a manner previously described.11 A total of 3481 samples
were available for inspection by the PF algorithm (see Algorithms) and
ranked according to PF statistic. A random number generator was then
employed to choose indices following an approximately normal distribution. Eighty 4 s samples were obtained and 20 repeated to evaluate
intraobserver consistency. A panel of 12 experts in catheter ablation of
AF from six Canadian centres were asked to evaluate the electrogram
samples blinded to case details and outcome. We created a questionnaire displaying each atrial electrogram, each accompanied by a continuous visual analogue scale from 0 to 100 ‘degrees fractionated’. In
order to assess whether PF or CFE-mean algorithms were good predictors of experts’ opinion on highly fractionated signal suitable for ablation, receiver operating characteristics (ROC) curves were
calculated. Experts agreed (unanimously) that they would ablate at
electrogram sites where scores exceeded 70 degrees fractionated.
Algorithms
A novel algorithm designed to assess the amount of continuous
electrical activity, or PF, scored each signal from 0 to 100%, where
0% is nowhere fractionated and 100% is continuously fractionated,
was developed on previously stored data and generated signals. The
algorithm implemented modified ‘Nademanee criteria’ to determine
the percentage of time the signal was fractionated, that is, the percentage of time deflections above baseline occurred ,120 ms apart
(Figure 1).13
Deflections were detected in the electrograms by way of a five-step
preprocessing algorithm consisting of (i) Baseline wander removal: using
an expansion in band-limited (discrete prolate spheroiodal sequences)
tapers. (ii) Signal rectification: using the Hilbert transform method.14 (iii)
Filtering: lowpass filtering from 0 to 100 Hz using an order 4 infinite
impulse response filter design.15 (iv) Deflection detection: instead of
using a threshold, the data were divided into segments of 30 ms
width to avoid over detection. Maximum voltage for each segment
was identified and in order to eliminate detections occurring at the
boundary, the data were re-segmented with an offset of 15 ms, and
the maximal voltages were reassessed. Peaks common to both segments were used for the analysis. (v) Percentage fractionation was calculated by summation of all peak-to-peak lengths (cycle lengths in
other literature) ,120 ms in length and divided by the total signal
length to give a percentage (Figure 1).
In addition, each signal was analysed online using the automated algorithm contained in the NavX System. The algorithm employed,
‘CFE-mean’, which delivers the average time between deflections in
the electrogram and is dependent upon three user-defined parameters: the deflection must (i) exceed an adaptive peak-to-peak sensitivity (typically 0.03– 0.05 mV), (ii) possess a down stroke and
exceed a (iii) refractory period (a blanking period to avoid over detection nominally set at 30 ms).4,12 Parameters were set following guidelines consistent with previous studies.4,7
Statistical analysis
Data are presented as mean + standard deviation unless stated otherwise. Interobserver variability was assessed quantitatively using
Pearson correlation and visually using ranked boxplots. Pearson or
Spearman correlation, when appropriate, was used to assess agreement between expert and algorithm, and ROC curves were created
to find optimal cut-off values for CFAE. Data were analysed using
MATLAB (MathWorks, Natick, MA, USA) and R (R Development
Core team, Vienna, Austria).
Results
Expert characteristics
Of 20 Canadian experts approached from six centres, 12 completed the assessment; the respondents had 5.5 + 2.8 years
450
C.L. Haley et al.
Figure 2 Boxplot of the 12 annotations given for each of the 80 signals, sorted in order of increasing median. The blue line at 70 degrees
fractionated represents the cut-off at which the experts agreed to ablate.
experience as primary operators with 5.4 + 1.9 ablation procedures per month, of which 26 + 11% were for persistent AF.
Intraobserver consistency
On the 20 repeated electrograms, the 12 experts displayed a high
level of reproducibility (Pearson correlation between the first and
second sighting of each signal was 0.85 + 0.067, P , 0.01).
Interobserver agreement
Median annotation vs. signal dispersion (interquartile range) for
each of the 80 non-repeated signals was shown using side-by-side
boxplots (Figure 2), ordered in increasing median degree of fractionation. The boxplots showed that near the extremes of
degree of fractionation the agreement tends to be better
(shorter box length) while the moderately fractionated signals
tend to demonstrate less agreement.
There was a significant positive correlation between PF and the
median value of degree of fractionation as assessed by the experts.
Pearson correlation for PF against consensus expert opinion was
0.78 [P , 0.01, 95% confidence interval (CI) (0.68–0.86)]. Receiver operating characteristics curve sensitivity and specificity
for PF were 0.77 and 0.81 at the optimal cut-off point of 58.45
PF with ROC curve area under curve (AUC) ¼ 0.89, CI (0.80–
0.99), with respect to the operator-defined cut-off of 70 degrees
fractionated.
There was no significant correlation between the CFE-mean
value and the degree of fractionation assessed by the experts.
Spearman correlation (with CFE ranked backwards so that positive
correlation indicates agreement) was 0.27 [P ¼ 0.016, 95% CI
(0.045 –0.49)]. Receiver operating characteristics curve area
(AUC) was not significant [P ¼ 0.602, AUC ¼ 0.602, 95% CI
(0.456 –0.747)]. Figure 3 shows CFE-mean and PF electroanatomic
maps for comparison. Note that the two metrics work differently:
CFE-mean indicates increasing fractionation by shorter cycle
length, while PF indicates increasing fractionation by higher
percentage. The colours have been reversed in the right-side
(CFE-mean) diagram for direct comparison and scaled to centre
near the optimal cut-off determined in the ROC analysis. It is
clear that the two metrics prioritize different sites for therapy.
Discussion
Substrate ablation for persistent AF is accepted by most as necessary to achieve an optimal outcome after catheter ablation.
However, the nature and targeting of CFAE remains debated
with poor understanding of the role of CFAE as functional or
truly anatomic substrate. The studies suggesting CFAE is important
for maintenance of AF used operator definitions and interpretation.8,9,13 To date, no study has rigorously compared an automated assessment algorithm with operator interpretation. We
examined the most consistently reported important feature of
CFAE—the percentage of time a signal is continuously fractionated—and designed an automated real-time assessment tool to
quantify this feature. In order to validate this algorithm, we compared it with consensus expert operator-defined values.
Alternative approaches to quantification of atrial electrograms
in both the time and frequency domain are being pursued.16,17
For example, dominant frequency (the highest peak in a power
spectrum estimate of the signal) has been implemented in new
releases of electroanatomic systems, despite conflicting clinical
reports8,17 – 19 which may be due to differences in implementation.
Dominant frequency assumes that a single frequency, or ‘cycle
length’, is adequate to describe the driving force of the fibrillation,
while rejecting secondary contributions to signal power (even if
they are significant), and without concern for non-stationarities
(changes in the average or variance of the signal over time). Rigorous assessment of automated algorithms is important when studying the physiopathology of AF and assumptions made must reflect
clinician understanding of the process.
Validation of a novel algorithm for quantification of the percentage of signal fractionation in AF
451
Figure 3 View of a posterior left atrium displaying electrogram data analysed using the two algorithms, PF map: Percentage fractionation (A)
and CFE-mean: complex-fractionated electrogram mean map (B). The colour scale is used to identify regions of fractionated electrograms; red
colour represents the areas highly fractionated in both maps. In (A) the scale is percent fractionation in (B) the scale is CFE-mean value over 4 s.
The percentage fractionation algorithm appears to represent more discrete regions of fractionation, and the two algorithms do not always
agree.
The main finding of this study was that the percentage of continuous activity in an atrial electrogram, as assessed by an automated algorithm, agrees with consensus expert identification of
CFAE. The PF correlated with experts’ opinion with a sensitivity
of 77% and specificity of 81%. Furthermore, with expert opinion
as the gold standard, PF prioritized CFAE over non-CFAE in 89%
of the signals under study. CFE-mean, which evaluates the mean
time between electrogram deflections did not compare significantly with expert opinion, as hypothesized, (CFAE-mean output
ignores variability in peak-to-peak length.) Notably, studies using
CFE-mean-guided ablation have reported significant positive clinical outcome.7
Algorithmic determination of PF differs from ‘CFE-mean’ in that
the PF statistic itself is robust (not-sensitive to outlying data points)
to anomalously long or short peak-to-peak distances, whereas an
assessment of the mean is unstable in this respect. Percentage fractionation reflects the overall percentage of continuous electrical
activity, whereas CFAE-mean reflects only the average behaviour
of the cycle lengths over the recording (though it includes standard
deviations of these cycle lengths). An algorithm for substrate ablation that is (i) based on the same decision criteria that are used by
operators and (ii) has been associated with positive clinical
outcome is required for implementation in intuitive software
systems that guide ablation. Percentage fractionation not only satisfies these two criteria, but is the only metric for CFAE that does.
Subsequent studies are now required to assess the efficiency of
PF-guided ablation with respect to clinical outcome.
Limitations
There is no gold standard for quantification of fractionation,
though we have used experienced expert opinion as a proxy.
There was a modest degree of interobserver variability as the
signals were derived from real cases and thus contained a practical
amount of noise artefact. Some experts remarked that this had
affected their judgement. Also, signals were presented to experts
context-free, i.e. there was no information given about the location
of the catheter during signal collection.
Conclusion
Catheter ablation for paroxysmal AFs is rapidly becoming standard
practice. There is concern that catheter ablation of persistent AF
requires additional ‘substrate modification’. Clinical studies have
confirmed that signals with high continuous fractionation represent
sites with positive clinical outcome after substrate ablation. The PF
statistic is designed to determine the percentage of the signal that
represents fractionated atrial activity. Importantly, PF demonstrates
excellent agreement with expert users and, as such, presents a new
standard for algorithm assessment. Use of a PF statistic should be
considered in future prospective studies of automated mapping
systems.
Acknowledgements
The authors are indebted to the electrophysiologists who completed the survey: Dr Allan Skanes, Dr Peter Leong-Sit, Dr Iqwal
Magnat, Dr Ratika Parkash, Dr George Veenhuyzen, Dr John
Sapp, and Dr Macle Laurent. Dr Damian Redfearn, Dr Rodrigo
Miranda, Dr Kevin Michael, and Dr Adrian Baranchuk also completed the survey.
Conflict of interest: none declared.
References
1. Nademanee K, Schwab MC, Kosar EM, Karwecki M, Moran MD, Visessok N et al.
Clinical outcomes of catheter substrate ablation for high-risk patients with atrial
fibrillation. J Am Coll Cardiol 2008;51:843 –9.
2. Haissaguerre M, Sanders P, Hocini M, Hsu LF, Shah DC, Scavee C et al. Changes in
atrial fibrillation cycle length and inducibility during catheter ablation and their relation to outcome. Circulation 2004;109:3007 – 13.
3. Oral H, Chugh A, Good E, Wimmer A, Dey S, Gadeela N et al. Radiofrequency
catheter ablation of chronic atrial fibrillation guided by complex electrograms. Circulation 2007;115:2606 –12.
452
4. Verma A, Mantovan R, Macle L, De Martino G, Chen J, Morillo CA et al. Substrate
and Trigger Ablation for Reduction of Atrial Fibrillation (STAR AF): a randomized,
multicentre, international trial. Eur Heart J 2010;31:1344 –56.
5. Ng J, Borodyanskiy AI, Chang ET, Villuendas R, Dibs S, Kadish AH et al. Measuring
the complexity of atrial fibrillation electrograms. J Cardiovasc Electrophysiol 2010;
21:649–55.
6. Wu J, Estner H, Luik A, Ücer E, Reents T, Pflaumer A et al. Automatic 3D mapping
of complex fractionated atrial electrograms (CFAE) in patients with paroxysmal
and persistent atrial fibrillation. J Cardiovasc Electrophysiol 2008;19:897 –903.
7. Verma A, Novak P, Macle L, Whaley B, Beardsall M, Wulffhart Z et al. A prospective, multicenter evlauation of ablating complex fractionated electrograms (CFEs)
during atrial fibrillation (AF) identified by an automated mapping algorithm: acute
effects on AF and efficacy as an adjuvant strategy. Heart Rhythm 2008;5:198 –205.
8. Takahashi Y, O’Neill MD, Hocini M, Dubois R, Matsuo S, Knecht S et al. Characterization of electrograms associated with termination of chronic atrial fibrillation
by catheter ablation. J Am Coll Cardiol 2008;51:1003 –10.
9. Hunter RJ, Diab I, Tayebjee M, Richmond L, Sporton S, Earley MJ et al. Characterization of Fractionated Atrial Electrograms Critical for Maintenance of Atrial Fibrillation: a randomized, controlled trial of Ablation Strategies (the CFAE AF trial).
Circ Arrhythm Electrophysiol 2011;4:622–9.
10. Haissaguerre M, Sanders P, Hocini M, Takahashi Y, Rotter M, Sacher F et al. Catheter ablation of long-lasting persistent atrial fibrillation: critical structures for termination. J Cardiovasc Electrophysiol 2005;16:1125 –37.
11. Redfearn DP, Simpson CS, Abdollah H, Baranchuk AM. Temporo-spatial stability
of complex fractionated atrial electrograms in two distinct and separate episodes
of paroxysmal atrial fibrillation. Europace 2009;11:1440 –4.
C.L. Haley et al.
12. Aizer A, Holmes DS, Garlitski AC, Bernstein NE, Smyth-Melsky JM, Ferrick AM
et al. Standardization and validation of an automated algorithm to identify
fractionation as a guide for atrial fibrillation ablation. Heart Rhythm 2008;5:
1134 –41.
13. Nademanee K, McKenzie J, Kosar E, Schwab M, Sunsaneewitayakul B, Vasavakul T
et al. A new approach for catheter ablation of atrial fibrillation: mapping of the
electrophysiologic substrate. J Am Coll Cardiol 2004;43:2044 –53.
14. Sörnmo L, Laguna P. Bioelectrical Signal Processing in Cardiac and Neurological Applications. Burlington, MA: Elsevier Academic Press; 2005.
15. Durrani T, Chapman R. Optimal all-pole filter design based on discrete prolate
spheroidal sequences. IEEE Trans Acoust, Speech, Signal Process 1984;32:716 –21.
16. Yoshida K, Ulfarsson M, Tada H, Chugh S, Good E, Kuhne M et al. Complex electrograms within the coronary sinus: time- and frequency-domain characteristics,
effects of antral pulmonary vein isolation, and relationship to clinical outcome
in patients with paroxysmal and persistent atrial fibrillation. J Cardiovasc Electrophysiol 2008;19:1017 –23.
17. Chang SH, Ulfarsson M, Chugh A, Yoshida K, Jongnarangsin K, Crawford T et al.
Time- and frequency-domain characteristics of atrial electrograms during sinus
rhythm and atrial fibrillation. J Cardiovasc Electrophysiol 2011;22:851 – 7.
18. Grzeda KR, Noujaim SF, Berenfeld O, Jalife J. Complex fractionated atrial electrograms: properties of time-domain versus frequency domain methods. Heart
Rhythm 2009;6:1475 –82.
19. Lo LW, Tai CT, Lin YJ, Chang SL, Udyavar AR, Hu YF et al. Predicting factors for
atrial fibrillation acute termination during catheter ablation procedures: implications for catheter ablation strategy and long-term outcome. Heart Rhythm 2009;
6:311 – 8.