RETRO‐mapping: A novel algorithm automating wavefront categorization using activation mapping during persistent atrial fibrillation demonstrates a reduction in wavefront collisions following pulmonary vein isolation

RETRO‐mapping was developed to automate activation mapping of atrial fibrillation (AF). We used the algorithm to study the effect of pulmonary vein isolation (PVI) on the frequency of focal, planar, and colliding wavefronts in persistent AF. An AFocusII catheter was placed on the left atrial endocardium to record 3 s of AF at six sites pre and post‐PVI in patients undergoing wide circumferential PVI for persistent AF. RETRO‐mapping analyzed each segment in 2 ms time windows for evidence of focal, planar, and colliding waveforms and the automated categorizations manually validated. Ten patients were recruited. A total of 360 s of data in 120 segments of 3 s from 60 left atrial locations were analyzed. RETRO‐map was highly effective at identifying focal waves and collisions during AF. PVI significantly reduced collision frequency but not focal and planar activation frequency. However, there was a significant reduction in the dispersion of activation directions. Larger studies may help determine factors associated with successful clinical outcome.


| INTRODUCTION
The underlying mechanism perpetuating electrical activation in atrial fibrillation (AF) remains uncertain.2][3][4][5][6] These theories have been tested by different mapping techniques but a consensus for the mechanism remains elusive.In paroxysmal AF, atrial ectopy originating in the pulmonary veins (PV) act as focal drivers initiating AF. 6 Electrically isolating these focal drivers by circumferential PV ablation has been an effective therapy in at least 60% of patients. 7The procedure has similar efficacy in persistent AF, although demonstrating the existence of similar focal drivers has been harder to prove. 8Recent mapping techniques have enabled invasive and noninvasive global simultaneous mapping of the atrium, but the benefit derived from ablation guided by these technologies continues to be debated. 5,9,10ctivation mapping has been the basis of determining mechanisms in other arrhythmias, but is difficult to apply in persistent AF.
We developed RETRO-mapping as a technique to create activation maps of AF and validated it against short segments of persistent AF using manually assigned local activation times. 11,12However, the volume of data generated was overwhelming and impractical for determining the underlying mechanism in real-time during clinical cases.We further developed RETRO-mapping to screen for activation patterns in short time windows, categorizing wavefronts into focal, planar, or collision.We manually validated RETRO-maps automated categorization of wavefronts and used it to study the impact of pulmonary vein isolation (PVI) on activation patterns in persistent AF.

| METHODS
Patients with symptomatic persistent AF (aged 18−85 years) undergoing clinically indicated catheter ablation procedures were prospectively recruited.Exclusion criteria included ejection fraction <25%, the presence of intracardiac thrombi, bleeding diatheses, and coagulopathy.All procedures were performed under general anesthesia.All patients were in AF on the day of their procedure.Anticoagulation regimens were not interrupted and all antiarrhythmic drugs except for amiodarone were discontinued at least five half-lives before the procedure.
The Ensite Precision 3D cardiac mapping system (Abbott) was used for all cases.Left atrial and pulmonary venous geometry was collected with a roving 20-pole spiral double loop catheter (Inquiry AFocus II; St Jude Medical; 4 mm interelectrode spacing, 20 mm fixed loop diameter).The AFocus II catheter design provides consistent electrode distribution which minimizes unmapped areas under the catheter recording surface.The following settings for mean-to-peak voltage were applied to the atrial bipolar electrograms during the data collection process: EGM width <10 ms, to avoid detecting farfield electrograms; EGM refractory time 50 ms; peak-to-peak sensitivity 0.08 mV, to avoid sensing noise; interpolation 10 mm and interior projection 5 mm, as surrogate markers to exclude inadequate electrode contact.
The AFocusII catheter was held in a stable position on the left atrial endocardium and when distinct electrograms were observed at each pole concurrently, a recording of 30 s of AF made.Multiple 30 s data segments were collected from the roof and the anterior, posterior, and inferior walls of the left atrium, where it was possible to achieve sufficient contact with the endocardium to record distinct electrograms on every pole concurrently.The position of each recording location was tagged as a "shadow" on the Precision left atrium geometry.After completing data collection, PVI was performed according to standard protocols.Following PVI, the AFocusII catheter was returned to the same locations and catheter orientation, and a further 30 s data segment was collected from each location.

Ethical approval for this study was provided by the Local
Research Ethics Committee for Imperial College Healthcare NHS Trust (14/LO1367), and all patients provided written informed consent before the procedure.

| Data analysis
All data were exported for offline processing with RETRO-mapping.
The algorithm was developed with the objective of identifying wavefronts that occur during AF using a windowed cross-correlation of the AFocus signals. 11Wavefronts are displayed via a graphic user interface (GUI) on an activation map (displaying an outline of the catheter face with activations superimposed) and automatically categorized as wavefront types with reference to the time point of occurrence.The user is able to confirm the automatic categorization of activation with the patterns observed in the GUI and the intracardiac electrograms (Figure 1).

| Validation of RETRO-mapping
RETRO-mapping automatically categorizes waves into focal, planar, or collisions.The algorithm categorizes activation every 2 ms.To validate RETRO-mapping, we examined whether the automatic categorization agreed with the wave observed in the activation map.A focal wave was defined as an activation where the wavefront forms a continuous loop that propagates outwards toward the periphery.A planar wave was defined as an activation showing directional uniformity moving across the 2D representation of the catheter from one side to the other.A collision was defined as the occurrence of two or more uniform waves colliding within the surface of the catheter.

| Effect of PVI
To determine the effect of PVI on activation patterns, activation in the first 3 s of each 30 s recording at each location were reviewed using RETRO-map and the number of planar waves, focal waves, and collisions occurring pre-PVI counted.The same was done post-PVI and compared by a paired t-test.
To understand the impact of PVI on direction of wavefronts, we developed a further level of analysis called an orbital plot. 12This provides a summary of activation direction and uniformity over the recording duration at a given location by assigning an average activation vector (which indicates the overall direction of activation) and its magnitude (which indicates the uniformity of activation).If all activation directions are the same then the average magnitude will be 1 (e.g., a uniform planar wavefront) but if the directions are not all aligned then the magnitude of the resultant is smaller.Polar coordinates were calculated using this method for every timewindow and each direction of activation assigned to 10°bins.Over a longer recording period (30 s) an orbital plot summarizes the wavefront direction by displaying a histogram of wavefront direction, weighted according to the magnitude of the wavefront uniformity.This provides a pictorial representation of the uniformity and direction of wavefronts in the mapped area.
RETRO-map was used to create orbital plots for each patient (using 30 s data segments from six locations pre & post-PVI).The orbital plots were reviewed subjectively and objectively (using an R50 analysis) to identify how the spread of wavefront activation, and therefore organization, changed after PVI.No data selection or preprocessing was required.R50 is the range in degrees in which 50% of the activation is contained. 12Data segments with a high degree of organization in a particular direction would be expected to show a smaller range of activation; a low R50.With completely random activation patterns, one would expect to have a theoretical R50 maximum of 180°.To calculate the R50; the data points from the orbital plot were extracted from Matlab and added to Excel where they were organized in descending order; angles were categorized in order per 10°bins (36 bins); the cumulative sum and the total sum of all the data points were calculated and finally it was established at which angle (from the 36 bins) the cumulative sum of the data points is greater or equal than 50% of the total sum of data points.

| Statistical analysis
All statistical analysis was conducted using GraphPad Prism 9 Version 9.3.1 (350) for Mac OS.Normality testing was conducted using the Shapiro−Wilk test and by visually inspecting histograms and q−q plots.Since all data sets were parametric, results are expressed as mean ± standard deviation to two decimal places.Paired t-tests were used to compare differences between activation patterns in each patient before and after PVI.p-Values ≤ .05were considered statistically significant.

| RESULTS
Ten patients were recruited (demographics Table 1).The mean age was 62.1 ± 9.0 years and 8 (80.0%) patients male.The mean duration of continuous persistent AF was 27.3 ± 16.5 months and mean LA diameter 45.0 ± 4.7 mm.A total of 360 s of data (120 segments of 3 s) from 60 left atrial locations were analyzed.
F I G U R E 1 Data was exported for analysis using RETRO-mapping.The first panel shows a 600 ms window with up to three activations at each electrode.The blue time line marks the wavefront displayed in the second panel.Here, a planar wave is seen in three timepoints as it progresses across the face of the catheter.An arrow demonstrates the direction of the wave.The algorithm categorizes every 2 ms time window into an activation pattern.The table that is generated (shown in the final panel) gives each 2 ms window and its categorization.In this case, the activation pattern appeared planar in each of the 2 ms time windows over a 20 ms period.This is consistent with the EGMs in panel 1 and the activation pattern visualized in panel 2.

| Validation of RETRO-mapping
Overall, 70 focal waves, 4474 planar waves, and 947 collisions were identified and manually validated.RETRO-map was highly effective at identifying focal waves and collisions (100% correlation with manual validation within the 2 ms time windows).For planar waves, the RETROmap planar categorization had to occur in 12 consecutive 2 ms windows (≥24 ms duration) to achieve an 80.3% correlation with manual validation.
The overall specificity in identifying planar waves was 37.7%, and hence, algorithm-detected planar waves were not used for further analysis.All algorithm categorized planar waves were manually validated, and only manually validated planar waves were used for subsequent analysis.

| Impact of PVI on activation patterns
Focal waves: Focal wavefronts were visualized throughout the left atrium and occurred repeatedly in certain locations.Figure 2 demonstrates recurrent focal activation in a single location before and after PVI.There was no reduction in mean rate of focal activation post-PVI (0.22 focals/s/patient pre-PVI, 0.17 focals/s/patient post-PVI, [p = .259]).In 7 patients, the number of focal activations decreased and in 2 patients there was an increase.
Manually validated planar waves: Recurrent planar activation in a single location was seen pre and post-PVI (Figure 3).There was no change in total number of manually validated planar waves post-PVI (2.41 planars/s/patient pre-PVI, 2.53 planars/s/patient post-PVI [p = .649]).

| Collisions
Wavefront collisions occurred frequently.The panels in Figure 5 illustrate the impact of PVI on each patient for each of the activation categories.

| Orbital plots
Orbital plots enabled visual representation of the impact of a reduction in collisions after PVI on the spread of activation directions.
For example, in Patient 5, a significant reduction in collisions was observed post-PVI (from 7 to 1 collision/s) at a specific location, where the activation direction became more confined to a range of ~90°(between 120°and 210°).Conversely, in Patient 3, an increase in collisions was observed post-PVI (from 3 to 7.33 collisions/s) and the activation direction remained broadly dispersed (Figure 6, Panel A).R50 was calculated before and after PVI for all locations mapped (n = 120).There was a significant decrease in R50 following PVI   Analyzing the manually validated wavefronts, we found a statistically significant reduction in collisions post-PVI, but no change in planar wavefronts using the current definition.There was a nonsignificant reduction in focal activation.We also used orbital plots to review the dispersion of all activation directions using the R50 and this showed a reduction in the spread of activation directions, which implies increased organization, post-PVI.
Interpreting the findings of this study is difficult as the mechanisms driving persistent AF remain controversial.The observation that AF is "more organized" after PVI has been widely F I G U R E 4 Collisions identified by the automated categorization algorithm at a site before PVI were validated by reviewing the activation wavefronts.Each panel shows the isochronal activation at four timepoints to illustrate the activation pattern and the corresponding electrogram with the time bar below.Separate wavefronts have different colors to help differentiate (Panels A−C).A 3D geometry with the catheter location is also shown.Following PVI there were no collisions identified and planar wavefront were seen at the same location (Panels D−F).PVI, pulmonary vein isolation.described using surrogate markers such as complex fractionated atrial electrograms (CFAEs) to indirectly assess the effect of PVI on mechanisms sustaining AF. [13][14][15] It was reported that PVI caused a decrease in the degree of fractionation and CFAEs suggesting more organized activation.However, the underlying activation patterns at CFAE sites is not certain as studies have also suggested that these may be AF drivers. 16,17Our findings indicate that the organizing effect of PVI is associated with a reduction in colliding wavefronts | 565 within the left atrium.By using activation mapping before and after PVI, we were able to avoid surrogate markers and demonstrate the organizing effect of PVI by directly assessing changes in the number of wave collisions.Based on the assumption that PVI removes PV drivers in persistent AF, fewer wavefronts would be expected to emanate from the PVs and this may lead to fewer collisions seen in the atrial body.
There was no significant change in the number of focal activations in the atrial body after PVI.The areas of repetitive focal activation identified could either represent an anatomical region with epicardial to endocardial activation or a focal driver site.Focal waves have been proposed as potential drivers in persistent AF 18,19 but it was beyond the scope of this study to determine if any of these focal activation sites could have been drivers.There is some evidence from the AcQMap system that ablating focal activation sites can terminate AF, but case numbers have been small. 14The CardioInsight system has also been used to target non-PV drivers with some promising case examples. 9In many of these studies, focal drivers were targeted in addition to PVI so determining for certain whether the ablated sites were drivers is difficult to conclude.In the GANGLIA-AF study focal ablation was performed at autonomic sites in the atrial body without PVI and this was able to prevent AF in about 50% of patients. 20It is not known if a similar mechanism would apply in persistent AF.We would need either larger series to determine if focal activation numbers determined clinical outcome after PVI or real-time RETRO-mapping to ablate focal activation sites and map the impact on activation patterns.
The RETRO-mapping algorithm requires further refining to identify planar wavefronts.It had a high relative specificity of 80% in recognizing planar waves of ≥24 ms duration, but the data volume would still make this challenging for real-time mapping.The error rate comparing automatic with manual categorization of planar wavefronts was primarily due to a planar wavefront needing to traverse the whole of the mapped area uniformly to be categorized as planar by manual validation.This occurred over a longer time period than the algorithm was created for (as opposed to focal & collisions which occur during a short time window) so many "planar" activations of shorter durations did not meet this criteria.Approximately 20% of the error rate was due to an area of conduction slowing and block which was also uniform and so detected by the algorithm but which we did not consider planar activation.PVI appeared to have no effect on the number of manually validated planar waves.We did not study other aspects of organization such as planar waves repeatedly following the same direction.
The strengths of our study include the extensive volume of AF data analyzed, spanning over 5 min of persistent AF and more than 5000 patterns of activation being visualized and reviewed.This allowed us to effectively validate RETRO-mapping and gain a better understanding of the mechanisms underlying AF.The use of RETROmapping enabled us to map activation patterns and study the effect of PVI on AF activation.

| Limitations
Patient numbers were small and data was analyzed retrospectively.However, it did show feasibility of being able to use RETRO-mapping to characterize AF.The ability to identify focal activation accurately may help guide ablation strategies when RETRO-mapping is implemented in real-time.Our results are reported as activations detected over time of data recording.However, the reported activations may be influenced by data quality, as not all recorded data time was derived from good-quality electrograms.Data quality could also be influenced by areas of scar which, at present, cannot be differentiated from areas of poor contact by the algorithm.In  F I G U R E 6 Panel (A) shows example of orbital plots before and after PVI in two patients.In the upper panel the patient had a reduction in collisions and this was associated with a clear reduction in the direction of activations seen after PVI.In the lower panel, the patient had an increase in the number of collisions with visual evidence of activation becoming more organized even though there were minor changes detected by the R50.Panel (B) shows graph of the reduction in R50 seen in all patients after PVI.PVI, pulmonary vein isolation.

4 |
DISCUSSIONRETRO-mapping is a novel algorithm for automatically categorizing wavefronts during persistent AF into focal, planar, and collisions.It was manually validated and found to be highly effective at determining focal activation and collisions.However, it needs further improvements to meaningfully categorize planar wavefronts.T A B L E 1 Demographic characteristics of patients undergoing clinically indicated PVI (n = 10) based on information collected before study start.F I G U R E 2 Focal activation identified by the automated categorization algorithm at the same site before and after PVI was validated by reviewing the activation wavefronts.Each panel shows the isochronal activation at four timepoints to illustrate the activation pattern and the corresponding electrogram with the time bar below.A 3D geometry with the catheter location is also shown.In this example there is evidence of focal activation from the same site both before (Panels A and B) and after PVI (Panels C and D).PVI, pulmonary vein isolation.

F I G U R E 3
Planar activation identified by the automated categorization algorithm at the same site before and after PVI was validated by reviewing the activation wavefronts.Each panel shows the isochronal activation at four timepoints to illustrate the activation pattern and the corresponding electrogram with the time bar below.A 3D geometry with the catheter location is also shown.In this example there is evidence of planar activation at the same site both before (Panels A−C) and after PVI (Panels D−F).PVI, pulmonary vein isolation.
addition, we were only able to analyze activation categorized by the algorithm.Within segments not categorized (discarded as no F I G U R E 5 Scatter plots demonstrating the total number of focal waves (Panel A), manually-validated planar waves (Panel B), and collisions (Panel C) for each patient before and after PVI.The total number of each wave type pre and post-PVI for each patient (n = 10) is depicted as a black square.The pre-PVI values for each patient are adjoined with the corresponding post-PVI values for the same patient to facilitate comparison.PVI, pulmonary vein isolation.activation, noise or poor catheter contact), there may have been activation patterns.The algorithm in the current study is limited by the catheter electrode configuration.In animal models of AF, it may be possible to increase spatial resolution with more electrodes over a larger area to better determine the underlying mechanisms of AF.Clinical studies considering ablation using RETRO-mapping would need more extensive and detailed mapping to avoid missing focal activation sites, unlike the current study objective to validate the automated categorization technique.

5 |
CONCLUSION RETRO-mapping is effective at automatically identifying focal activation and collisions during AF.Further refinement is required for automatic categorization of planar wavefronts.In patients undergoing PVI for persistent AF, RETRO-mapping demonstrated a reduction in the number of colliding wavefronts which may be the mechanism leading to more organized activation post-PVI.