Technical Considerations on Phase Mapping for Identification of Atrial Reentrant Activity in Direct- and Inverse-Computed Electrograms
Background: Phase mapping has become a broadly used technique to identify atrial reentrant circuits for ablative therapy guidance. This work studies the phase mapping process and how the signal nature and its filtering affect the reentrant pattern characterization in electrogram (EGM), body surface potential mapping, and electrocardiographic imaging signals.
Methods and Results: EGM, body surface potential mapping, and electrocardiographic imaging phase maps were obtained from 17 simulations of atrial fibrillation, atrial flutter, and focal atrial tachycardia. Reentrant activity was identified by singularity point recognition in raw signals and in signals after narrow band-pass filtering at the highest dominant frequency (HDF). Reentrant activity was dominantly present in the EGM recordings only for atrial fibrillation and some atrial flutter propagations patterns, and HDF filtering allowed increasing the reentrant activity detection from 60% to 70% of time in atrial fibrillation in unipolar recordings and from 0% to 62% in bipolar. In body surface potential mapping maps, HDF filtering increased from 10% to 90% the sensitivity, although provoked a residual false reentrant activity ≈30% of time. In electrocardiographic imaging, HDF filtering allowed to increase ≤100% the time with detected rotors, although provoked the apparition of false rotors during 100% of time. Nevertheless, raw electrocardiographic imaging phase maps presented reentrant activity just in atrial fibrillation recordings accounting for ≈80% of time.
Conclusions: Rotor identification is accurate and sensitive and does not require additional signal processing in measured or noninvasively computed unipolar EGMs. Bipolar EGMs and body surface potential mapping do require HDF filtering to detect rotors at the expense of a decreased specificity.
See Editorial by Aronis et al
WHAT IS KNOWN
Reentrant electric activity plays a decisive role in the perpetuation of atrial tachyarrhythmias, and phase mapping is becoming a reference technique for reentrant pattern identification in direct (electrogram, body surface potential mapping) or inverse (electrocardiographic imaging) mapping techniques.
Phase singularity detection may be dependent of the signal preprocessing, as well as to the phase quantification surrounding these reentrant patterns.
WHAT THE STUDY ADDS
Identification of reentrant patterns by phase mapping is accurate in presence of smooth signals, such as inverse-computed unipolar electrograms, and does not require additional signal processing for rotor identification.
Signals with complex morphology, such as bipolar electrograms or body surface potential mapping signals, require prefiltering to simplify the phase maps and to achieve reasonable sensitivity in the rotor detection.
Reentrant propagation of electric activity plays a decisive role in the perpetuation of atrial tachyarrhythmias. Although atrial flutter (AFL) is caused by a macroreentrant circuit around anatomic or functional obstacles that can be terminated ablating the critical isthmus,1 the nature of the waves’ mechanisms that maintain atrial fibrillation (AF) is still controversial.2 Fibrillatory electrograms analyzed using rules developed for organized rhythms, such as activation mapping,2 are unreliable because of inconsistencies in the estimated activation times in relation to the presence of far-field intrachamber crosstalk, noise, or multicomponent electrograms (EGMs).3 On the contrary, the value of sequential mapping during AF is limited because of the dynamic changes in activation sequence during AF. Despite these limitations, there is substantial experimental and clinical evidence, based on activation phase and frequency analyses, demonstrating that AF is maintained by functional reentries, or rotors, and that localized ablation of the atrial regions harboring such rotors can terminate AF episodes.4,5
Recent progress in ablative therapies for AF has been paired with increased understanding of the wave mechanisms responsible for AF as a direct consequence of the development of novel mapping systems to characterize spatiotemporal patterns of AF electric activity. These mapping systems include experimental optical systems based on the use of potentiometric dyes6 or clinical electric recording systems using multipolar catheters.5 Phase analysis of optical mapping signals has become the most reliable method to identify reentrant patterns because pivoting activity naturally renders singularity points (SPs) in the phase maps that can be clearly identified.7 However, optical mapping dyes are toxic in the clinical setting, and the extrapolation from experimental optical mapping to clinical electrode-based mapping lacks validation and raises the possibility of false association of phase SPs with AF reentries because nonreentrant electric activity may also cause the appearance of SPs under certain circumstances. On the contrary, multielectrode mapping catheters have difficulties to consistently map global biatrial activation with uniform accuracy. Moreover, activation time mapping that has been used to map AF5, can be ambiguous because of multicomponent EGMs and result in inconsistencies in the estimated activation times and wave descriptions.2 To overcome the ambiguity in marking the activation time, the phase analysis is considering the whole cardiac activation cycle indiscriminately.6 However, phase mapping techniques, as other techniques as well, make use of temporal signal filtering for improving the interpretation of propagation and reentrant pattern identification,8,9 adding another possible uncertainty into wave propagation studies in AF.
More recently, noninvasive systems have emerged as a panoramic mapping approach for simultaneous body surface recordings of biatrial activation during AF. Body surface potential mapping (BSPM)8,10 uses tens of electrodes for the analysis of surface ECG signals, whereas the electrocardiographic imaging (ECGI)11 computationally reconstructs the epicardial electric activity from the BSPM recordings. However, the accuracy of those technologies on determination of the driving role of observed rotors in human AF has not been established. Therefore, in this study, we use mathematical models of different atrial reentrant arrhythmias to provide a robust characterization of invasive and noninvasive mapping approach for localization of reentrant activation patterns in AF. The objective of the present study is to analyze the phase mapping processes in EGM, BSPM, and ECGI, including the effects of signals’ nature, temporal filtering, and an automatic detection of SPs in phase maps, to outline the validity and potential clinical use of those AF mapping approaches.
Materials and Methods
Atrial Mathematical Models
A realistic 3D model of the atrial anatomy composed by 284 578 nodes and 1 353 783 tetrahedrons (673.4±130.3 µm between nodes) was used to simulate the atrial electric activity.12 A gradient on the electrophysiological properties of the atrial myocardium, specifically on Ik,ACH, IK1, INa, and ICaL, was introduced into the atrial cell formulation13 to obtain propagation patterns maintained by rotors. Fibrotic tissue was modeled by disconnecting a percentage of nodes between 20% and 60% and scar tissue by disconnecting 100% of nodes in the scar region. The system of differential equations was solved by using Runge–Kutta integration based on a graphic processors unit (NVIDIA Tesla C2075 6G).14
An ensemble of 17 different arrhythmic electric patterns was simulated, divided into 4 groups according to the nature of their activation patterns. Group I was composed by one AF pattern driven by multiple rotational sources and 4 AF patterns driven by a single rotor at varying locations of the left atrium: pulmonary veins, posterior wall of the left atrium, and right atrial appendage. Group II was composed by 4 AFL patterns: a typical AFL, a clockwise atypical AFL, an inferior vena cava (IVC) atypical AFL, and an atypical AFL turning around the pulmonary veins because of the existence of inactive scar tissue in the posterior wall of the left atrium. Group III was composed by 4 focal atrial tachycardia (AT) models repeatedly stimulated at different locations of the atria (IVC, left superior pulmonary vein, right inferior pulmonary vein, and right atrial appendage). Group IV was composed by the same AT simulations of group III in which scar regions were added to create more complex propagation patterns.
Electric Signals Generation
For each simulation, a uniform mesh of 2048 unipolar EGMs was calculated surrounding the epicardial surface (1-mm distance) under the assumption of a homogenous, unbounded, and quasi-static conducting medium by summing up all effective dipole contributions over the entire model.15 Computed electrograms were stored for processing at a sampling frequency of 500 Hz. Bipolar electrograms were obtained as the potential difference between each node and the nearest neighbor.
The BSPM potentials on the torso model were calculated by solving the forward problem with the boundary element method16 in a mesh formed by 771 nodes and 1538 triangular patches (Figure 1). White Gaussian noise was added to the BSPM signals with a signal-to-noise ratio of 60 dB, and all signals were then referenced to the Wilson central terminal.
Inverse-computed EGMs (ECGI signals) were obtained by solving the inverse problem with zero-order Tikhonov regularization method and election of the regularization parameter based on the L curve.16,17
To evaluate the performance of an automatic rotor identification technique, epicardial EGMs of atrial scenarios with real reentrant activity were randomly reassigned to different nodes.18 These random epicardial EGM maps were processed as described above to obtain the BSPM and then the corresponding ECGI signals for the rotor analysis.
Baseline EGM, BSPM, and ECGI signals were estimated by decimation to 12.5 Hz and filtering with a Butterworth 10th-order low-pass filter with a cutoff frequency of 2 Hz. Signals were interpolated to 500 Hz and subtracted from the original signals. EGM, BSPM, and ECGI signals were then low-pass filtered with a 10th-order Butterworth filter with a cutoff frequency of 30 Hz. Processing procedures here were similar to clinical procedures elsewhere.10
For dominant frequency (DF) analysis, EGM, BSPM, and ECGI signals were baseline-removed as previously reported10 and were then low-pass filtered with a 10th-order Butterworth filter with a cutoff frequency of 10 Hz. Power spectral density of all signals was computed using Welch periodogram (65 536 point Fast Fourier Transform and 80% overlap) to determine the local DF with a spectral resolution of 0.01 Hz.10
We also tested the effect of narrow band-pass filtering of EGM, BSPM, and ECGI centered at the highest DF (HDF) found on the atrial surface by using a cascade of high-pass elliptic filters with a cutoff frequency equal to HDF −1 Hz and a low-pass elliptic filter with a cutoff frequency equal to HDF +1 Hz.8
Reentrant Activity Identification
Reentrant wave localization was performed by identification of SPs in the phase signal map obtained with the Hilbert transform.19 As shown in Figure 2, the phase transformation assigns a phase value between –π and π for each sample of the signal, and thus, each phase corresponds to a given state of the action potential (π for resting, π/2 for depolarization, 0 for plateau, and −π for repolarization). A phase map snapshot, therefore, allows inferring the propagation patterns and specifically the center of a pivoting rotor appears as a point in which phase is not defined (hence the term SP) surrounded by phases ranging monotonically from −π to π.
To identify SPs, phase values were evaluated along 3 different circles surrounding each tested point with increasing radii. Six to 12 points per circle were used for the phase analysis in which the EGM, BSPM, and ECGI signals were interpolated by a weighted average of the neighboring nodes, being d−2 the weight for each node and d the distance between nodes.
A tested point was assigned to be an SP only when the phases of at least 2 of these 3 circles were gradually increasing or decreasing for a total of 2π8 and if the mean phase error with respect to a straight line was lower than a threshold: 0.4 radians for EGM, 0.2 radians for BSPM, and no threshold for ECGI.
Testing with circles of various radii (Figure I in the Data Supplement), radii of 0.5, 1, and 1.5 cm were found to maximize sensitivity of SP identification in AF models, for both raw and HDF filtered signals and, therefore, selected for the study.
An SP reflects the instantaneous condition of phase reentry. Thus, a pivoting excitation pattern was considered to constitute a propagating wave when maintaining a sequential connection between its SPs across time. The distance between SPs at consecutive time instants should be <1 (EGM and ECGI) or 5 cm (BSPM) to be considered related and maintain a continuity of the wave rotation. In Figure II in the Data Supplement, we show the effect of this spatial threshold on true/spurious rotor identification. Finally, only long-lasting SP describing waves that complete at least 2 rotations were considered as rotors, and other SPs were discarded.
Sensitivity and Specificity Calculation
The different filtering strategies were evaluated in their ability to identify stable reentrant patterns (>2 rotations) in our models of AF as functional reentries (rotors) and in AFL patterns as anatomic reentries. Different criteria for considering a detected SP as true or false were applied for the atrial EGM and ECGI maps versus the BSPM. In EGM and ECGI maps, the sensitivity and specificity measures were based on SP location criterion because of the implication in the ablation guidance of the SP location, whereas in the BSPM, the location of the detected SP has no direct implication and, therefore, sensitivity and specificity measures considered only a presence or absence criterion.
Accordingly, when the EGM and ECGI maps were analyzed on the atrial wall, excluding valves and veins, only AF rotors detected <1.5 cm from the actual rotor core were considered as true-location positives (named as true rotors in Figures 4, 5, and 7), whereas AF rotors detected >1.5 cm from the actual rotor core were considered as false-location positives. We chose 1.5 cm as a threshold distance based on the rotor precession distance in our database, which was below this value (Figure III in the Data Supplement).
In our AFL simulations, a reentry was present around the tricuspid valve, left pulmonary veins, or IVC and its counter rotating wave was in the IVC or pulmonary vein orifices or at the septum. Therefore, in our AFL simulations, the electric reentrant pattern should generate phase singularities only inside the orifices or the septal areas. Because the EGM and ECGI time-series signals in the sensitivity and specificity analysis were not calculated at the orifices and the septal areas, all SPs detected during AFL are necessarily considered as false-location positives.
When the electric activity was analyzed on the torso surface (BSPM), the sensitivity and specificity measures were based on an SP presence criterion. In this case, the reentrant electric patterns generated by AF and AFL simulations can generate a rotor anywhere across the surface,8 so only their presence or absence was considered and not their location as in EGM or ECGI maps. Therefore, all surface reentries detected during AF and AFL patterns were considered as true-presence positives.
Additionally, reference sensitivity and specificity analyses of SP detections were as follows: (1) all SP detections (>2 rotations) during random distributions of the EGMs were considered as false positives and (2) all SP detections during AT and AT+scar rhythms which were simulated to be maintained by periodic focal stimulation were considered false positives.
All measures of continuous variables are reported as average±SD and displayed as bars with a height equal to the average and whiskers length equal to the SD. Statistical significance of differences between normally distributed continuous variables was estimated using the student t test. Linear fitting for phase measurements was performed by using the least squares method; R-square was calculated as the coefficient of determination, and phase errors were calculated as the square difference between phase measurements and their linear best-fitting. A P<0.05 was considered to be significant.
Restrictions in Rotor Identification
We found that we were able to identify more SPs in random EGM activity than in rotor-driven AF models (Figure 3D), and phase transitions around SPs that arise from nonrotating activity were less gradual than those arising from rotational activity. In Figure 3A and 3B, the phase transitions in 3 concentrical circles around detected SPs are shown for a rotational and a wavebreak pattern from an AF simulation. In this example, deviation from a linearly gradual change transition was largest in the outermost circle (1.5-cm radius) for the wavebreak pattern because phase was not monotonically increasing. Overall, this deviation was larger in the random patterns than in rotor-driven AF (1.00±0.04 versus 0.47±0.20 radius; P<0.01). To reject spurious SP detections, a linearity threshold (0.4 radius) was applied to SP detections, resulting in a reduction in the amount of detected SPs, as it can be observed in Figure 3D.
Transient SPs can also be found in our phase maps that arise from U turns around scars from an AT+scar simulation instead of from actual functional rotations. In Figure 4, one of such examples is depicted. Overall, if a duration of 0.5 turns is required to SPs to be considered as rotors, all false detections in random propagation patterns are rejected (Figure 4D), whereas most true rotation patterns are detected (66.5±47.2% of time for AF models). However, several false positives are detected (6.5±14.1% of time for AF patterns, 32.9±24.5% for AFL patterns, or 57.9±43.6% for AT+scars). Because SPs that do not arise from an actual rotation transiently disappear from the phase maps without completing a rotating cycle (Figure 4C), imposing a duration threshold of 2 turns reduces considerably the amount of false detections (to 0% for AF, 0% for AFL, and 15.9±28.8% for AT+scar) while keeping almost unaltered the detection of true rotors (60.0±54.7%). Figure IV in the Data Supplement shows rotor detection sensitivity when considering 0.25 to 4 rotations, where the incidence of spurious rotors detected in patterns other than AF decreases with the number of required rotations.
The reported detection ratio for AF models can be increased by preprocessing the EGMs before performing the phase transformation. Hilbert transform is particularly well suited for smooth or sinusoidal signals, and, therefore, a band-pass filter, centered at the activation rate, allows increasing the detection ratio (from 60.0±54.7% to 70.9±39.9%) for AF models while the false-positive rate detection in AF models is only 2.6±5.1% (Figure 5). This band-pass filtering was required for detecting rotors by using EGMs with multiple deflections, as found in bipolar EGMs. In bipolar EGMs, rotors were detected with the same detection rate than in unipolar EGMs but only after band-pass filtering (Figure 5C, 5D, and 5F).
However, the increased sensitivity for AF rotors detection after band-pass filtering takes place at the expense of increasing the detection ratio in AFL models, with ≤47.9±55.3% of time with detected rotors. Figure V in the Data Supplement shows an example of a stable macro-reentry around the IVC. Here, the upward propagation in the right atrium is followed by propagation through the Bachman bundle and subsequent downward depolarization of the posterior wall of the left atrium. This pattern was not reflected into a stable SP in the EGMs but got smoothened and stabilized after HDF filtering and a SP appeared. Therefore, HDF filtering may increase the false-positive detections that arise from actual rotating patterns—but not rotors—in the tail of the propagating wavefront.
Reentrant Activity in BSPM and ECGI
We have previously proposed to apply HDF filtering to BSPM during AF to increase the sensitivity of rotor detection8 but were unable to quantify the specificity of the method and whether it could be applied for computation of the ECGI maps. As shown in surface BSPMs for different mathematical models with and without HDF filtering (Figure 6), stable rotors can be observed after HDF filtering but not on the raw signals. However, HDF filtering also stabilized the patterns generated by random EGMs. Overall, HDF filtering allowed an increased detection of rotors in AF patterns, from 10.8±18.2% to 92.9±11.9% (Figure 7A) and in AFL, from 10.8±18.2% to 92.9±11.9%. However, it also resulted in false detections in complex AT patterns, from 0% to 15.9±31.8% and even in random AF patterns, from 0% to 32.4±28.4%.
When solving the inverse problem of electrocardiography for AF patterns, rotors can be accurately detected even without applying HDF filtering, as it is depicted in Figure 6. Overall, true rotors during AF could be detected during 72.5±42.0% of time in AF patterns, with only 4.7±10.7% of time with false detections for AF, 13.2±18.0% for random EGMs, and 25.0±50.0% for AFL and no false detections in the other situations (Figure 7B). HDF filtering applied after inverse problem solution, increased the detection of true rotors during AF ≤80.0±44.7%, but also increased the amount of false detections in all models: that is, 99.2±1.8% for random AF EGMs, 85.7±5.2% for random AFL EGMs, or 81.9±3.1% for complex ATs.
In this in silico study, we have found that rotor identification based on phase singularities detection is accurate and sensitive and does not require additional signal processing in smooth signals, such as unipolar EGMs, either measured or computed noninvasively. Bipolar EGMs and surface BSPM do require HDF filtering to detect rotors as phase singularities at the expense of a decreased specificity. HDF filtering is not recommended in the solution of the inverse problem of electrocardiography because of an increased susceptibility to detect artefactual phase singularities (Table).
Phase Mapping of Human AF
The mechanisms of AF are still unclear because the available mapping techniques yield diverse maps ranging from organized sources to highly disordered waves.2,5,7,11,18,20 Although phase analysis of signals has provided experimental evidence that localized reentrant sources or rotors drive AF,7,9 it has shown conflicting results when applied to endocardial signals or body surface electrocardiographic recordings in patients. On one hand, phased-analyzed multipolar endocardial recordings showed stable and long-lasting rotors, whereas short-lasting rotors that tend to recur to the same anatomic location were the hallmark of inverse-computed body surface maps.5,11 On the other hand, AF activation patterns reported using various noninvasive systems (ie, BSPM and ECGI) using different signal processing methods seem to be simpler than epicardial maps recorded in other studies, which do not report stable rotors.2,8,11 To clarify the effect of the filtering and validate phase processing on intracardiac AF activity and body surface recordings, we reproduced the mapping processing in computer simulations.
Rotors and Phase Singularities
The phase transform has been widely used for the identification of electric patterns in transmembrane potentials.6,19 However, the sole detection of a phase singularity does not imply the presence of an underlying rotor because singularities may arise from wavebreaks or fibrillatory conduction.2,9 Nevertheless, SPs arising from wavebreaks are more unstable and do not consistently present monotonical increases in phase. In this context, other authors have already proposed to search for phase singularities in 2 concentric rings around the SP18 and impose a restriction of a temporal span of at least 1 turn to increase specificity. In the same direction, in the present study, we found that application of time and space restrictions to detected SPs allows increasing the specificity in the detection of rotors. In particular, we propose the requirement of a good fit to a monotonical increase of phase in the 3 concentric rings. The use of 3 rings increases sensitivity as compared with a single ring because rotors occupying a small region are detected by the inner circles, whereas rotors with a large precession are detected by the outer rings. At the same time, the use of 3 rings reduces the chance for randomly distributed phases to be considered as SPs.
Phase Transformation and Signal Morphology
We have shown that the equivalence between propagation patterns and phase maps depends on signal morphology. Although the Hilbert transform results in an unambiguous phase assignment for signals with simple morphologies, for complex morphologies there is no relationship between the assigned phase values and the phase in the action potential of the tissue. Hilbert transform is mathematically defined for properly identifying the instantaneous phase value of a sinusoidal wave, assigning the whole range values from −π to π to the interval between signal peaks. However, the Hilbert transformation of complex signals with several deflections assigns the whole range of phase values from −π to π between 2 consecutive deflections, and, thus, this assignment does not convey any useful information for pattern identification. We have shown that phase singularities can be detected after the phase transformation of unipolar, noise-free EGMs. However, raw EGMs with multiple deflections, such as bipolar EGMs, are not suitable for SP detection and require a preprocessing step before applying Hilbert transform.
HDF Filtering and BSPM Phase Mapping
We have previously proposed the use of a narrow band-pass filter before the computation of the phase transformation to stabilize phase singularities in BSPM recordings.8 We showed that HDF filtering allows selecting the contribution of areas that activate at the HDF, whereas reduces the contribution on the body surface from regions that activate at a slower rate and are not harboring rotors.8 In the present work, we investigated the effect of HDF filtering on propagation patterns not maintained by rotors to quantify the proportion of artefactual detections introduced by our signal processing. According to our results, narrow band-pass filtering does induce false detections that can be as high as 30% in randomly distributed EGMs from AF models. For this reason, isolated SPs on BSPM maps obtained after HDF filtering, even if they last for longer than 2 turns, should be interpreted with care because they are not an unequivocal demonstration of the presence of a rotational activity. However, we have shown that a high incidence of long-lasting SPs is indicative for rotational activity because rotors were >2-fold detected during underlying rotational patterns than for nonrotational ones.
HDF filtering of BSPM results in a particularly high incidence of detected rotational patterns in AFL models. This was to be expected because raw BSPM data already show rotational patterns that get stabilized by the HDF filtering. This resemblance between AF and AFL patterns can be explained by the fact that electric potential recordings contain far-field components, and as such, the electric sources at the vicinity of the anatomic obstacle may generate rotational electric fields elsewhere even without an actual functional reentry source. Thus, the BSPM detection of SPs does not allow, in principle, discriminating between rotational patterns around an obstacle and functional rotors. However, here we studied the sensitivity and specificity of the BSPM to discriminate between rotational and nonrotational patterns, which is feasible and clinically relevant. Our simulations show that stable rotational patterns on the BSPM phase maps should be considered as indicative of either AFL or AF, and activation frequency should allow discriminating between these 2 rhythms.
HDF Filtering and EGM Phase Mapping
A aggressive band-pass filtering strategy has been proposed for detecting rotational patterns in multipolar catheter baskets,5,18 similar to our HDF filtering.8 Consistently, we have shown here that HDF filtering applied to EGMs increases the detection rate of rotors during AF at the expense of few false detections (Table). In addition, the smoothing effect of the HDF filtering seems to be necessary when the EGMs present multiple deflections so that the phase assignment by the Hilbert transform is related to a phase in the action potential.
However, HDF filtering of EGMs results in some artefactual detections that should be taken into consideration. In particular, when the underlying pattern presents a coincidental rotation and not a mother rotor, there is an increased chance of detecting a rotor because of the smoothing effect of the HDF filtering. These coincidental rotational patterns were especially relevant in our AFL model population in which either the activation tail or anatomic obstacles give rise to non-AF–driving rotations. Although these coincidental rotational patterns may not fulfill the eligibility criteria for rotors because there is no single rotational center where all phases between −π and π converge, phase homogenization that results after HDF filtering may make these patterns as qualified for rotor detection. This effect has been also seen in both ECGI, EGM, and BSPM phase maps.
HDF Filtering and ECGI Phase Mapping
Narrow band-pass filtering has also been used following inverse problem solution in mapping rotors during AF.9,11 The filtering has been shown to stabilize SPs; however, we demonstrate in this study that aggressive filtering strategies applied to the inverse-computed electrograms may also cause artefactual rotors. This comes as no surprise if we consider the ECGI virtual EGMs to depend on the BSPM recordings, which themselves are showing a limited sensitivity and specificity for SP and rotors detection. It is of notice although that the HDF filtering increases the detection of ECGI rotors generated by random EGMs more than for the BSPM (Figure 7), probably because of the additional smoothing by the inverse solution relative to the forward solution.
The present work is based on the use of mathematical models instead of patient data because current technology does not allow determining whether detected rotors are artefactual or they are in fact AF drivers. Mathematical models, instead, allow defining specific activation patterns in which the presence of mother rotors is known a priori and, thus, enable accurate classification. However, our mathematical AF models may be too simplistic and may not fully represent the whole spectra of AF patients.
Different thresholds for detection of reentrant activity had to be established, such as phase linearity or the radii of the circles for the phase assessment. The threshold election allowed increasing the specificity at the expense of decreasing the sensitivity of the reentrant activity detection and vice versa. These thresholds were chosen to achieve a balance between specificity and sensitivity according to our database. It should be further explored whether the proposed thresholds should be adapted to other scenarios.
Finally, we used the random distribution of the EGMs to generate propagation patterns with no stable reentrant patterns. Nevertheless, some of them could still retain some reentrant-like activity because of the casual alignment of the EGMs, although in this article, all reentries detected in random patterns have been classified as false positives.
The results of the present study may have several clinical implications that should be taken into consideration during phase analysis of AF signals. First, time and space restrictions should be applied to avoid false rotors detections. To this purpose, we suggest to only consider true rotors those rotational patterns lasting >2 turns. Secondly, differentiation between AFL and AF for correct classification of rotational patterns on the BSPM phase maps should be based on activation frequency. Thirdly, selection of signals prepocessing will depend on the recording type and method. Unipolar EGMs, either recorded from the endocardium or those computed noninvasively (ECGI), do not require additional signal processing.5,11,16 In contrast, endocardial bipolar EGMs and surface BSPM require HDF filtering to be able to detect rotors.8,18 However, care must be taken to exclude falsely detected rotors because of the methodology. Finally, aggressive filtering strategies should be avoided during ECGI because of an increased susceptibility to stabilization and detection of false rotors (Table).
Phase transformation and SP identification are robust methods to identify reentrant activity in the atrium. Smooth signals, such as inverse-computed unipolar EGMs, do not require additional signal processing for rotor identification. Rotor identification in signals with complex morphology, such us bipolar EGMs or BSPM signals, requires HDF filtering to simplify the phase maps at the expense of a decreased specificity.
Sources of Funding
This study was supported, in part, by Universitat Politècnica de València through its research initiative program; Generalitat Valenciana Grants (ACIF/2013/021); the Instituto de Salud Carlos III (Ministry of Economy and Competitiveness, Spain: PI13-01882, PI13-00903, PI14/00857, PI16/01123, TEC2013-46067-R, DTS16/0160, and IJCI-2014–22178); Spanish Society of Cardiology (Grant for Clinical Research in Cardiology 2015); Spanish Ministry of Science and Innovation (Red RIC RD12.0042.0001); and the National Heart, Lung, and Blood Institute (P01-HL039707, P01-HL087226, and Q1 R01-HL118304) and cofounded by FEDER.
Dr Atienza served on the advisory board of Medtronic and Livanova. Dr Berenfeld received research support from Medtronic and St. Jude Medical. He is a cofounder and scientific officer of Rhythm Solutions, Inc, research and development director for SAS Volta Medical, and consultant at Acutus Medical. None of the companies disclosed here financed the research described in this article. The other authors report no conflicts.
The Data Supplement is available at http://circep.ahajournals.org/lookup/suppl/doi:10.1161/CIRCEP.117.005008/-/DC1.
Circ Arrhythm Electrophysiol is available at http://circep.ahajournals.org.
- Received January 2, 2017.
- Accepted July 10, 2017.
- © 2017 American Heart Association, Inc.
- Garcia-Cosio F,
- Pastor-Fuentes A,
- Nunez-Angulo A
- Narayan SM,
- Krummen DE,
- Clopton P,
- Shivkumar K,
- Miller JM
- Rodrigo M,
- Guillem MS,
- Climent AM,
- Pedrón-Torrecilla J,
- Liberos A,
- Millet J,
- Fernández-Avilés F,
- Atienza F,
- Berenfeld O
- Vijayakumar R,
- Vasireddi SK,
- Cuculich PS,
- Faddis MN,
- Rudy Y
- Guillem MS,
- Climent AM,
- Millet J,
- Arenal Á,
- Fernández-Avilés F,
- Jalife J,
- Atienza F,
- Berenfeld O
- Haissaguerre M,
- Hocini M,
- Denis A,
- Shah AJ,
- Komatsu Y,
- Yamashita S,
- Daly M,
- Amraoui S,
- Zellerhoff S,
- Picat MQ,
- Quotb A,
- Jesel L,
- Lim H,
- Ploux S,
- Bordachar P,
- Attuel G,
- Meillet V,
- Ritter P,
- Derval N,
- Sacher F,
- Bernus O,
- Cochet H,
- Jais P,
- Dubois R
- Garcia-Molla VM,
- Liberos A,
- Vidal A,
- Guillem MS,
- Millet J,
- Gonzalez A,
- Martinez-Zaldivar FJ,
- Climent AM
- Rodrigo M,
- Climent AM,
- Liberos A,
- Calvo D,
- Fernández-Avilés F,
- Berenfeld O,
- Atienza F,
- Guillem MS
- Pedrón-Torrecilla J,
- Rodrigo M,
- Climent AM,
- Liberos A,
- Pérez-David E,
- Bermejo J,
- Arenal Á,
- Millet J,
- Fernández-Avilés F,
- Berenfeld O,
- Atienza F,
- Guillem MS
- Tikhonov AN
- Kuklik P,
- Zeemering S,
- van Hunnik A,
- Maesen B,
- Pison L,
- Lau D,
- Maessen J,
- Podziemski P,
- Meyer C,
- Schaffer B,
- Crijns H,
- Willems S,
- Schotten U
- Alhusseini M,
- Vidmar D,
- Meckler GL,
- Kowalewski CA,
- Shenasa F,
- Wang PJ,
- Narayan SM,
- Rappel WJ