Evaluation of RETRO‐Mapping for electrophysiological features including direction of plane activity during atrial fibrillation using multipolar catheters in humans

A quantifiable, automated standard of analyzing heart rhythm has long eluded cardiologists due, in part, to the limitations in technology and the ability to analyze large electrogram datasets. In this proof‐of‐concept study, we propose new measures to quantify plane activity in atrial fibrillation (AF) using our Representation of Electrical Tracking of Origin (RETRO)‐Mapping software.

Conclusion: RETRO-Mapping can measure electrophysiological features of activation activity and this proof-of-concept study suggests that this can be extended to the detection of plane activity in three types of AF. Wavefront direction may have a role in future work for predicting plane activity. For this study, we focused more on the ability of the algorithm to detect plane activity and less the differences between the types of AF. Future work should be in validating these results with a larger data set and comparing with other types of activation such as rotational, collision, and focal. Ultimately, this work can be implemented in real-time for prediction of wavefronts during ablation procedures. Epicardial mapping studies of a few seconds of AF using local activation times (LAT) appeared to support the theory. However, in clinical practice, episodes of AF can be prevented by ablating focal triggers and it has been postulated that persistent AF could also be maintained by focal drivers. This has challenged the multiple wavelet theory, but methods for investigating AF activation remain limited. 1 Lee et al. 1 studied activation patterns including preferential conduction patterns; however, they acquired epicardial mapping data during open-heart surgery and required the use of 128 electrodes. LAT mapping is laborious and usually restricted to short time windows and small areas. Global simultaneous mapping of the atria, both invasively and noninvasively, lack resolution and are difficult to apply to mechanistic studies.
Liu et al. 2 used noncontact charge density mapping to investigate the effects of pulmonary vein isolation (PVI), but this technique was limited by the need for manual annotation and offline retrospective analysis. Honarbakhsh et al. 3

created the Stochastic Trajectory
Analysis of Ranked signals (STAR) system that used unipolar electrograms to map focal drivers of AF and determine early sites of activation. The Focal Impulse and Rotor Mapping (FIRM) system, a global mapping tool based on unipolar electrograms from basket catheters, 4 successfully identified focal and rotational wavefronts. 5 The radically different, but more extensive approach of OPTIMA (Optimal Target Identification via Modeling of Arrhythmogenesis) created models from late-gadolinium enhanced MRIs and electrograms taken before procedure to predict AF ablation sites. 6 For this method, extensive computational power and time was required. 6 The current gold standard of mapping of AF is manual annotation of activation time. We previously developed RETRO-Mapping (RETRO-Representation of Electrical Tracking of Origin) as an automated technique to map longer time segments of AF. This visual tool gives real-time, two dimensional (2D) representation of wavefront propagation across the left atrium and overcomes the limitations in isochronal mapping of our previous Ripple Map. 7,8 Our previous study validated the method for identifying different types of wavefronts (uniform, focal, collision, and rotational) 9 against manual LAT mapping. The study also manually validated the method to demonstrate evidence of spatio-temporal stability in AF activation over several minutes.
In this proof-of-concept study, we focused on plane activation patterns seen in AF and investigated whether RETRO-Mapping could describe key characteristics such as conduction velocity (CV), cycle length (CL) and wavefront direction. We also investigated preferential directions of wavefront propagation across the left atrium.

| METHODS
RETRO-Mapping was used to detect and categorize activation wavefronts automatically. Plane wavefronts are defined by two criteria (1) having two endpoints, both at the edge of the 2D area of analysis, (2) traversing the area of analysis in an overall linear direction. We sought to quantify this linear behavior. We defined activation edge as the edge of a wavefront at a particular frame as it propagates across the left atrium, as seen in Although plane activity is the focus of this study, we also detected a total of 942 collision, 58 focal, and 7607 rotational activation edges.

| Data acquisition
The patients included in this study were undergoing clinically indicated ablation procedures under general anesthetic in accordance with local guidelines. 10 Before the procedure, all patients provided informed, written consent and their data were anonymised. This protocol has been approved by the research ethics committee (13/LO1169 and 14/LO1367).

| Algorithm development
We focused on the readings from the lower posterior wall where potentials are often recorded after pulmonary vein isolation 11 and where the catheter generates high quality data. All algorithms presented in this study were developed using the MATLAB software (MathWorks, Inc.).
We calculated the distributions of CV, activation edge direction, and CL across all wavefronts. CV and CL were calculated for each wavefront at the 5 bipolar electrogram positions with the best contact, based on a filtering algorithm. We defined a plane wavefront as one where all activation edges are within the same 90°. For each activation recorded across the lower posterior wall, we compared the plane activation edge direction between the current and preceding activation edges for different time intervals to ascertain consistency over time.

| Filtering/preprocessing
We developed an algorithm to determine which electrodes had good contact. This involved optimization of two criteria (1) maximization of peak-to-peak bipolar electrogram voltage and (2) geometric consistency (ensuring the 5-points were in contact). F I G U R E 2 19 bipolar electrogram positions are shown after projection onto the lower posterior wall. RETRO-Mapping calculates the positions, via a triangulation algorithm, at the midpoints between two consecutive electrodes on the 20-pole catheter. Highlighted in gray are the 5 bipolar electrograms with the best contact with myocardium based on a filtering algorithm combining highest peak-to-peak voltage and catheter geometry. In this example, the bipolar electrograms at positions 15, 16, 17, 18, and 8 were chosen, but this changes between cases. The two algorithms for geometric interpolation for bipolar calculation and best contact were repeated for each case. TOA TOA  TOA TOA   min egm  TOA TOA  TOA TOA   2  =  (  +  2  :  +  2   -+  2 :

| CL
Activation at a given point in the left atrium is given as a binary signal.
We defined CL as the time between two consecutive rising binary edges. A rising edge represents a point where inactive, repolarised tissue becomes depolarized (i.e., the signal "rises" from 0 to 1). This is recorded as relative ToA whose calculation we have previously presented. 9 ToA i = timing of activation i for given bipolar electrogram x ToA j = timing j of activation for given bipolar electrogram x where j is subsequent to i

| CV
Our CV algorithm utilized MATLAB's in-built Euclidean distance transform, which was performed on a binary image. 12 For each binary 0 pixel, it calculated the Pythagorean distance to the closest binary 1 pixel.
We calculated CV over the region of newly activated tissue at the given time. We calculated the Euclidean distance transform twice: (1) from the previous activation wavefront and (2)   We sought a means to quantify plane activity as wavefronts whose activation edge directions maintain the same direction (±45°) as they traverse the left atrium. For this reason, we calculated the difference between activation edge directions for subsequent time frames in RETRO-Mapping.
We also investigated whether the direction of the current plane wavefront was a predictor for the direction of the subsequent wavefront if that wavefront were another plane wavefront. We thus calculated the difference in wavefront directions for subsequent wavefronts to see how many were in the same direction (with a margin of ±45°). We compared the number of wavefronts whose subsequent wavefronts were in the same direction with the number of wavefronts whose subsequent wavefronts were in a different direction. This ratio was given as a percentage as calculated in Equation 5. We compared this for plane wavefronts that were 1, 2, 5, and 10 wavefronts ahead in time.

| Histogram analysis
The CV histogram in Figure 3

| Wavefront direction consistency
A lower difference in wavefront directions suggests a higher degree of wavefront direction consistency. As seen in Figure 5, Our histogram distribution suggests that our CV and CL algorithms return reasonable values compared to the literature. 15,16 This indicates that RETRO-Mapping can potentially measure key features such as CL and CV during AF. This should be formally validated in a study comparing the behaviors of different types of wavefronts such as focal, collision, and rotational activities. 9 In the future, obtaining these features, such as CV, in real-time during an ablation could allow physicians to update patient-specific models of AF to help guide their strategy. Theoretically, our method could F I G U R E 3 Thirty-nine thousand eight hundred sixty-seven plane activation edges were analyzed. We calculated the conduction velocity and cycle length at the five bipolar electrogram positions with the best contact and plotted the results on the above histogram diagrams. We also calculated the directions in degrees for each activation edge. All angles are represented in the above distribution suggesting a wide range of direction of propagation at the low posterior wall.

F I G U R E 4
The difference in direction between activation edges when time between edges increases is shown. Seven thousand, seven hundred forty-eight activation edges are from patients with paroxysmal AF. Eleven thousand nine hundred six are from those with persistent AF being treated with amiodarone. Fourteen thousand nine hundred fifty-nine are from patients with persistent AF but were not receiving amiodarone as treatment. The directions of the plane wavefronts were calculated as the angle in degrees. We calculated the difference between wavefront directions at different intervals of time from each other. The above graph displays the median and standard deviations for the wavefronts. The in direction remained below 45°for all frames indicating that our algorithm can successfully measure the presence of plane activity. We found that patients with persistent AF had a tendency for the next wavefront to have a more similar direction, in comparison to those in paroxysmal AF or persistent AF on amiodarone. However, in persistent AF, the CL was shorter, and this could restrict the range of new wavefront directions because of refractoriness in the tissue ( Figure 5).
When the preceding CL ( Figure 4) was accounted for, there was a tendency for paroxysmal AF patients to have a more variable wavefront direction. This merits further investigation and might be because healthier tissue supports more rapidly changing wavefront directions at any given CL: the wavefront direction variability may not there be a measure of 'organization' in the usual sense.
The results suggest wavefront direction for all three types of AF are similar for the subsequent wavefront, but that this trend quickly reduces over time. It is likely most reliable in the persistent without amiodarone T A B L E 1 Median difference in activation edge direction (degree) for the three groups of AF with increasing time between wavefronts (seen in Figure 4). T A B L E 2 Standard deviation difference in activation edge direction (degree) for the three groups of AF with increasing time between wavefronts (seen in Figure 4).
The difference in direction between wavefronts when time between edges increases is shown. W + 1 indicates a comparison between the current plane wavefront and the 1st subsequent wavefront if that wavefront is plane. This was repeated for the 2nd, 5th, and 10th subsequent wavefronts if that wavefront were plane. Calculated are the percentage of wavefronts where the difference in direction for the two wavefronts is between −45 to 45°( the criteria for wavefronts being in the same direction). This represents how far into the future the current plane wavefront direction predicts future plane wavefront directions. For example 48.8% of the plane wavefronts in persistent with amiodarone cases were in the same direction as the subsequent plane wavefront. However, 24.3% of the plane wavefronts in persistent with amiodarone cases were in the same direction as 10 plane wavefronts in the future. Data analysis is currently performed offline, and real-time analysis during ablation procedures is ideal and possible. The design of the algorithms was centered around their eventual real-time implementation thus their clinical translation will require few modifications. Moreover, the running of the algorithms is within the capability of any normal computer found within hospital clinics.
A key advantage of our technique is that it provides the CV and direction of activation edges and wavefronts. To our knowledge, none of the existing software packages used with electroanatomic mapping systems perform this analysis. Our technique uses the acquired data to describe the wavefronts, and future versions of the system could potentially provide this information in real-time.
Our results are encouraging in the use of activation edge direction to detect plane activity and to a lesser degree in the use of wavefront direction to predict future plane activity. It also gives hope for our RETRO-Mapping software to be used as a tool to calculate CV and CL at the time of ablation procedures. Future work should be undertaken to validate our algorithms and compare with other types of wavefront activity aside from plane. Our ultimate hope is that the direction and frequency of uniform plane wavefronts could help determine whether AF activation is more consistent with focal driver or multiple wavelet theories.

| LIMITATIONS
This is a proof-of-concept study aimed at evaluating whether RETRO-Mapping can quantify electrophysiological features and ultimately quantify plane activity. We are currently revising RETRO-Mapping for automatic classification of wavefront and measurement of disorganization during AF.
Most importantly, our algorithm assumes consistency in orientation of the multipolar catheter. Our RETRO-Mapping algorithm could account for this computationally, but it would be better if this were ensured at the time of procedure. We also designed the algorithm to account for lack of good contact of the catheter, although this should be ensured during data collection.
Due to the narrow field-of-view of high-density mapping systems, one of the limitations faced was distinguishing between rotational wavefronts and plane wavefronts, if remote from the central core. Computationally, we describe plane wavefronts as having endpoints of the activation edge at the edge of the field of view and rotational wavefronts as having a "loose" endpoint within the field. Additionally, it was challenging to detect poor contact when using the AFocus system. A future algorithm could provide an indication of poor contact to help guide the placement of electrodes.
Our data set comprises 30 s data segments and requires induction of AF. This was not possible in all patients and the samples were not included in this paper's analysis. Originally, 20 patients were collected for this study, but those for whom AF induction was not possible or catheter contact with the atrium was suboptimal, were excluded. However, the focus of the study was around variability between activation edges (for which we detected 34 613) thus we could draw initial, proof-of-concept conclusions.