Double Potential Gradient Analysis for Critical Isthmus Detection in Scar Related Atrial Tachycardia

Abstract

Background:

The presence of double potentials (DPs)—two discrete electrogram deflections separated by an isoelectric segment—is a validated indicator during ablation of scar-mediated atrial tachycardia (AT). A corridor of parallel DPs is often used as a marker of a line of conduction block, whereby DP duration decreases as one approaches the conducting gap. In this context, it is the spatial gradient of DP duration that is used for isthmus localization. Here, we calculate the DP duration gradient of atrial electroanatomic maps and assess how it compares to DP duration as a quantitative metric in isthmus identification.

Objective:

Assessment the utility of DP duration gradient for identifying AT isthmuses

Methods:

Patients undergoing ablation of scar-mediated AT were included if a complete high-density map was available. Isthmus points were identified by an electrophysiologist prior to analysis. As point clouds from the electroanatomic maps are irregular, calculations of DP gradients were interpolated to a regular grid via inverse distance weighting. Gradients were then back-interpolated to the original point clouds for analysis.

Results:

57 EAMs among 52 patients were available for analysis (13 female, median age 60.5). Pooled isthmus points had higher DP gradient values than non-isthmus points (Mann-Whitney, p<0.001). On logistic regression, the DP gradient was more predictive of isthmus location versus DP duration (AUC=0.667, 95% CI: 0.660 – 0.673 vs AUC=0.621, 95% CI: 0.616 – 0.627, DeLong test p<0.001.)

Conclusion:

DP gradient is a more accurate predictor of isthmus points than DP duration, and may be useful in identifying its location.

Introduction:

Catheter ablation is a commonly used therapy in the treatment of atrial and ventricular arrhythmias. The advent of high density electroanatomic mapping has allowed for improved characterization of complex scar-related arrhythmias¹, enabling a more accurate study of the electroanatomic properties of myocardium. The identification of adequate sites for ablation, however, still requires the visual analysis of electrogram characteristics, activation mapping and entrainment data. The presence of double potentials (DPs) is one well-validated electrogram feature used to guide the targeting of ablation sites.

Defined as two discrete electrogram deflections separated by an isoelectric segment, DPs have been shown to arise from the presence of functional or anatomic block with sequential activation on either side of the block. In principle, this potential splitting occurs due to electrotonic/far-field influence of tissues on either side of a fixed or functional barrier, as demonstrated in animal models of atrial flutter where double potentials were recorded at lines of block created by crush injury². A corridor of parallel DPs is often used as a marker of a line of block and can be seen in either complete or incomplete block². This characteristic has been validated as a guide in the identification of lines of block in both atria^3–6 and ventricles⁷, and to confirm block created with ablation⁸. The presence of DPs has accordingly been incorporated into modern mapping software as part of an algorithm that displays wavefront activation as a calculated best-fit solution⁹.

Beyond being a binary indicator of conduction block, the duration between deflections in an electrogram with DPs provides information regarding the proximity to a critical area of breakthrough³. Lines of complete conduction block typically have longer DP durations than those with incomplete block due to the greater delay in propagating to the other side^3,10. Likewise, when DPs are created by antidromic activation on either side of an incomplete line of block, DP duration decreases as one approaches the conducting gap¹¹.

Advancements in data collection density and analysis have provided the electrophysiologist with increasingly detailed electrical activation during ablation procedures, and DP identification can be performed automatically¹². The identification of the critical isthmus of a tachycardia, however, is still done largely through qualitative visual analysis. In this study we explored the possibility of using the calculated gradient of DP durations, rather than DP duration alone, for isthmus detection.

Methods

Patient Selection

Patients at least 18 years old undergoing atrial mapping and ablation of scar-mediated atrial tachycardia 2/2021 and 4/2025 were included. Ablations were performed at two sites. The study was approved the Institutional Review Boards at Columbia University Irving Medical Center and New York Hospital Queens. Scar-related atrial tachycardia was defined by the presence of non-centrifugal reentrant activation in AT¹³. For inclusion in the study, the mechanism of the tachycardia had to be adequately established in the opinion of the operating physician and investigators, and at least one isthmus identified with sufficient mapping density for structural characterization. DP duration and DP gradient values were not available to investigators during the labeling of isthmus points. Data from a subset of the patients included in this study were used in a previous study that assessed the ability of a trained neural network to predict AT isthmus locations¹⁴.

Mapping and Ablation

All procedures were performed under general anesthesia or moderate sedation by DS, HY, AB, GR, JH, or JK. Mapping was performed using the CARTO 3 EAMS and a multi-spline catheter (PENTARAY; 20-poles with 3 mm spacing or OCTARAY; 48-poles with 2 or 3 mm spacing, Biosense Webster, Diamond Bar CA) or the Ensite Precision system and a grid-shaped 16-pole catheter with 4 mm spacing (HD Grid; Abbott Cardiovascular, Plymouth MN). Approximately halfway through the study period, the Ensite X EAMS replaced Ensite Precision. Ablation was performed using either the SmartTouch SF (Biosense) or TactiCath (Abbott Cardiovascular) irrigated radiofrequency catheters. Ablation caused the termination of all arrhythmias studied.

Isthmus Identification and Double Potential Calculation

Coordinates and electrograms at time of acquisition were exported retrospectively from the electroanatomic mapping system. Complete data sets, including coordinate location, activation timing, voltage, and electrogram (EGMs) at the time of acquisition for each point sampled were exported from the mapping system for each case. A single cycle length window was selected for each EGM, and the signal was processed using an instantaneous energy operator¹⁵. Processed electrogram complexes that represented activity above or below baseline were counted as activity complexes in the manner described by Oesterlein et al¹².

A scar-related isthmus was defined as an area of abnormal bipolar (or omnipolar where available) voltage (< 1.5 mV) and conduction (<0.5 m/s) bounded on either side by functional or fixed conduction block, anatomic obstacle, or a scar (defined as < 0.10 mV)^9,16,17. Isthmus identity and mechanism were confirmed in off-line analysis by consensus opinion of two investigators (DS, EC).

DP duration was defined as the duration between complexes for signals with two or more complexes. A value of 0 was recorded for signals with a single complex. When signals demonstrated multiple complexes, the longest duration between deflections was recorded. Complexes that were divided by the cycle length window were combined and considered as a single complex. Calculations and feature extraction were performed using custom designed software (Python 3.9).

Double Potential Interpolation

To determine the spatial relationship between DP duration and isthmus points, we sought to determine whether the characteristic of the DP duration gradient over space was related to isthmus position. As point clouds from the electroanatomic maps are spatially irregular, the calculation of gradients first required interpolation of the point cloud onto a regular grid. A commonly implemented interpolation algorithm, inverse distance weighting (IDW) was utilized¹⁸. IDW requires the selection of multiple parameters, including the resolution of the interpolated grid, a power parameter (which determines how the local points impact the value of a grid point by determining how much a given impact is reduced with distance), and the number of local points that are considered when performing the interpolation calculation for each grid point¹⁹.

The optimal grid resolution was calculated using a K-nearest neighbor (KNN) plot for each EAM and assessing the local density of the EAMs. The mean distance between the 5 nearest local neighbors per EAM varied from approximately 0.6–1.5mm; thus to preserve as much information as possible while maintaining efficient memory utilization, a grid spacing of 0.5mm was selected. For each grid point, the DP duration value of the 5 nearest points was used in weighing, and weighing was based on the inverse geodesic distance with a power of 1.

Double Potential Gradient Calculation

To gain information regarding the spatial associations of DP duration in relationship to isthmus location, two metrics with respect to changes in DP duration in the 3D interpolated grid were calculated.

Gradient: The gradient (grad) of the DP duration scalar field was computed to obtain the direction of the maximum rate of increase of the field and the magnitude of that increase. Where f is the scalar DP duration field, grad f = ∇f.

Laplacian: The Laplacian of a scalar function is the divergence of its gradient. It reflects how f compares to its neighborhood average (local convexity/concavity). Positive values indicate locally concave-up structure (often near a local minimum), and negative values indicate concave-down structure (often near a local maximum). It is calculated as Laplacian of f = ∇²f = ∇ ⋅ ∇ f = fxx+fyy+fzz

The values derived above were then interpolated back onto the original point cloud to confirm fidelity of the workflow and for assessment of their ability to predict isthmus identity. The overall workflow is shared in Figure 1.

Figure 1:

Methods schema for the calculation done in this study. Irregular point cloud with DP durations was available from the original EAMs. These DP durations were interpolated onto a regular grid. This grid was used to calculate DP derivatives, which were then back-interpolated onto the original point cloud for analysis.

Statistical Analysis and Visualization

Continuous data is summarized as median (interquartile interval). Categorical data is summarized as percentage. Comparisons of isthmus and non-isthmus point summary statistics were done using the Mann-Whitney U-test on pooled data. Receiver operating characteristic (ROC) curves were plotted for each parameter tested and the effectiveness was assessed as the area under the curve (AUC). The combined AUC of multiple assessed features was determined via logistic regression. Differences between AUCs was assessed using the DeLong test. All parameter calculations were performed using custom software (Python 3.9). As individual mapping points within the same electroanatomic map are spatially correlated, this pooling strategy may overestimate statistical significance. Hierarchical or mixed-effects modeling was not performed in the primary analysis, though we assessed a per-map AUC (Supplemental 1) to demonstrate stability of effect size across EAMs. Statistical analysis was performed using GraphPad Prism Version 10.1.0 for Mac (GraphPad Software, Boston, Massachusetts USA). Point clouds were plotted with MATLAB Version R2023b, Natick, Massachusetts.

Results

Baseline Characteristics

57 EAMs among 52 patients were available for analysis (25.0% female, median age 69.1 (60.5, 73.5) The baseline characteristics are provided in Table 1.

Table 1.

Baseline characteristics of patients

Characteristic	Count (%) / Median (IQR)
Age (yrs)	69.1 (60.5, 73.5)
Sex
Male	39 (75%)
Female	13 (25%)
Race
White	33 (63.5%)
Black	2 (3.9%)
Asian	2 (3.9%)
Other	9 (17.3%)
Unspecified	6 (11.5%)
Body Mass Index	27.9 (24.7, 30.3)
Comorbidities
Hypertension	20 (38.5%)
Hyperlipidemia	44 (84.6%)
Congestive heart failure	41 (78.9%)
Coronary artery disease	25 (48.1%)
Diabetes mellitus	5 (9.6%)
Stroke	8 (15.4%)
Rate and Rhythm Control
Beta blocker	50 (96.2%)
Amiodarone	8 (15.4%)
Flecainide	3 (5.8%)
Sotalol	2 (3.9%)
Dronedarone	1 (1.9%)
Mexiletine	1 (1.9)

Summary Statistics

In pooled data, isthmus points had statistically higher DP duration than non-isthmus points (p<0.0001, Figure 2a). This effect persisted when only considering points with non-zero DP durations (p=0.035) (Figure 2b).

Figure 2:

a) DP durations of isthmus and non-isthmus points. b) DP durations of isthmus and non-isthmus points excluding points that have only 1 potential (measured as a DP of 0 in 2a).

Positional and Gradient analysis

Visually, isthmus points appear to occur roughly at transitions from high to low DP durations (Figure 3a). After interpolation of DP duration and calculation of the gradient and Laplacian, the interpolated grid of DP durations was interpolated back to the initial point cloud. Locations with elevated DP durations were preserved, indicating proper interpolation (Figure 3b). DP derivatives (gradient and Laplacian) are shown in Figure 4. Logistic regression was performed using each parameter independently to identify which were predictive of isthmus points. Non-interpolated DP duration, as was available in the initial data, was statistically significant, though weakly predictive of isthmus location (AUC of 0.621, 95% CI 0.616 – 0.627), Figure 5a. Gradient, as calculated above, was significantly predictive of the isthmus (p<0.001, AUC=0.667, CI: 0.660 – 0.673), with a significant improvement in isthmus discrimination relative to DP duration (p<0.001)(Figure 5b). The Laplacian was not a statistically significant predictor (Figure 5c). A logistic regression model that accounted for DP duration and DP gradient was statistically improved over models that assessed either only DP gradient, or only DP duration (P<0.0001) but with only a small absolute increase in AUC (AUC=0.670, 95% CI 0.664 – 0.676). A model that considered all 3 parameters (DP duration + DP gradient + DP Laplacian) performed similarly (AUC = 0.672, 95% CI: 0.666 – 0.678) and was not statistically different from the model considering DP duration and DP gradient only (p=0.233).

Figure 3:

A) Plotted original EAMs with DP durations. B) After interpolating all DP durations onto a regular grid and calculating DP derivatives, the interpolated DP durations were also back-interpolated onto the original point cloud to demonstrate fidelity of the interpolation process. As shown, regions of elevated DP duration are preserved in these two plots. Isthmus points overlayed as red circles.

Figure 4:

EAMs with DP derivatives back-interpolated onto their original point locations, isthmus points overlayed as red circles. A) DP gradient B) DP Laplacian

Figure 5:

ROC curves demonstrationg isthmus discrimination of the a) original DP durations, b) DP Gradient, c) DP Laplacian, d) model including DP duration and DP Gradient, e) model including DP duration, DP gradient, and DP Laplacian

Discussion

Double potentials (DPs), defined as two discrete deflections in an electrogram separated by an isoelectric segment, are a key metric used in the identification of the critical isthmus of a tachycardia. We show herein that the gradient of DP duration is predictive of isthmus location. We further demonstrate that using DP gradient as a quantitative predictor outperforms the use of simple DP duration, which itself is only weakly predictive of isthmus location.

The results demonstrated in this study are in line with prior research on DPs. Indeed, DP durations are expected to decrease, rather than increase, as one approaches a conducting gap/isthmus¹¹, and regions at many lines of block adjacent to the isthmus are expected to have non-zero double potentials. While DP gradient is often used to identify isthmus location, this is typically done implicitly and qualitatively, which can introduce operator bias in the isthmus identification process. Our method provides a framework for this feature to be operationalized quantitatively, reducing variability. Further, we explored the possibility that more complex derivatives of DP duration are predictive of isthmus location. While DP gradient is superior to DP duration, other DP duration derivatives such as Laplacian appear to be less predictive. This was assessed as one may theorize that if DP duration decreases approaching the critical isthmus, the isthmus itself may be a DP duration local minimum, which could be detected by higher order derivatives. This was not empirically shown in our assessment, however, suggesting that the simple calculation of gradient alone is sufficient. Logistic regression models that account for multiple of these features were only marginally improved relative to DP gradient alone. Altogether, our work is in line with other recent research that attempts to quantitatively identify the critical isthmus using electrogram-derived features, reducing reliance on qualitative visual assessment^20,21. Ultimately, clinical tools that automatically calculate and synthesize multiple such electrogram features would likely be most useful and accurate in assisting the electrophysiologist in intra-procedural isthmus detection.

There are several potential limitations of this study. The retrospective nature of the data analysis and the single-center design may limit the generalizability of the results. Future studies involving prospective data collection and multicenter trials are warranted to validate our findings and assess their applicability in diverse patient populations and clinical settings. Prospective trials would be needed to assess if the modest improvement in isthmus discrimination noted is clinically significant. Other structural features such as myocardial thickness at the isthmus site were not available for analysis and thus their impact could not be assessed. The study also relies on the validity of the ground truth isthmus point selection, though this validity is improved with the use of multiple established features by experienced clinicians in their point selection.

Additionally, the relationship between DP duration and isthmus proximity may be more complex in cases of extreme conduction delay, where prolonged or multi-component double potentials may be present at or near the critical isthmus. In such cases, the gradient may still provide directional information, but the interpretation of absolute DP duration values requires clinical context. Future studies should evaluate the performance of DP gradient analysis in arrhythmias with varying degrees of conduction delay. Our study included data from two mapping platforms (CARTO 3 and Ensite Precision/X), providing preliminary evidence of cross-platform applicability. However, differences in electrode configuration, mapping density, and signal annotation algorithms may influence DP gradient calculations, and validation across additional platforms and annotation methods is warranted. The applicability of DP gradient analysis to ventricular tachycardia and other chamber anatomies, where differences in wall thickness and scar architecture may alter the spatial distribution of double potentials, remains to be determined. Further, our selected parameters for KNN and interpolation were selected on the basis of standard best practices, but parameter optimization should be explored in future studies. A further limitation is that individual mapping points were pooled across all electroanatomic maps and treated as independent observations. Given the spatial and electrophysiologic correlation among points within the same map, the reported p-values should be interpreted with this consideration. That AUC was similar across EAMs suggests that our result was not driven by a set of outlier cases (Supplemental 1), but future analyses incorporating hierarchical or mixed-effects models may be warranted.

In conclusion, this study is the first to quantitatively demonstrate the strong relationship between DP gradient and isthmus location in atrial tachycardia. By using DP gradient quantitatively, we hope to reduce variability in interpretation of this metric and its ability to identify isthmus locations. As mapping throughput and data density progressively increases, the quantification of metrics such as DP gradient may also allow for utilization in increasingly complex models being developed for isthmus identification¹⁴ and may increase the precision and accuracy of isthmus ablation in atrial tachycardia.

Abbreviations:

DP

double potential

EAM

electroanatomic map

AT

Atrial Tachycardia

EGM

electrogram

IDW

inverse distance weighting

KNN

K-nearest neighbor

ROC

Receiver operating characteristic (ROC)

AUC

Area Under Curve