According to an estimate in 2014, 2.7–6.1 million people in the United States are afflicted by atrial fibrillation (AF). AF is the most common type of supraventricular arrhythmia in the United States and is associated with an increased risk of stroke (Go et al., 2001; Miyasaka et al., 2006). AF is characterized by the chaotic electrical activity in the atria (Munger et al., 2014). Furthermore, the prevalence of AF has been projected to increase to nearly 12.1 million people by 2050 (Miyasaka et al., 2006; Colilla et al., 2013). These reasons make AF a significant public health concern and underpin the need to develop more efficient treatment techniques.

Compared with antiarrhythmic drug therapy, catheter ablation improves AF symptoms and quality of life and reduces AF recurrence (Wazni et al., 2005; Jaïs et al., 2008; Wilber et al., 2010). Pulmonary vein (PV) isolation, the most commonly used catheter ablation approach, alone can be used as a strategy for catheter ablation in patients with paroxysmal AF (Sawhney et al., 2009; Ouyang et al., 2010; Chao et al., 2012). However, in the case of persistent and long-standing AF, electrical initiation of AF may arise outside the PV regions, which play an essential role in arrhythmogenesis and maintenance of AF (Elayi et al., 2010). Hence, it is important to identify AF ablation targets outside the PV regions for the improved success of catheter-based ablation therapy.

Several approaches, such as local activation maps, phase maps, and dominant frequency (DF) maps (Pandit and Jalife, 2013), are commonly used to identify ablation target sites. In our laboratory, we developed several novel techniques to identify active sites of AF, especially outside the PV regions, both in the frequency-domain [multiscale frequency (MSF) (Arunachalam et al., 2016b)] and time-domain [multiscale entropy (MSE) (Arunachalam et al., 2018) and kurtosis (Kt) (Arunachalam et al., 2016a)]. All novel techniques have been validated using optical mapping studies of ex vivo rabbit hearts and have been established to accurately identify the active sites or arrhythmias, such as pivot point of rotors in experiments where rotors have been visually observed (Annoni et al., 2018). These techniques were shown to be accurate under some clinical limitations, such as reduced signal time duration and decreased spatial resolution. It was also demonstrated that the novel techniques could differentiate between the intracardiac electrograms (iEGMs) recorded from patients with successful and unsuccessful AF ablation (Ravikumar et al., 2021). However, the use of individual approaches fails to robustly identify electrically active spatial sites for ablation (Ravikumar et al., 2021), and therefore, efforts on developing combinatory approaches to identify the active AF sites are needed.

In this study, we aimed to develop a similarity score by combining the various iEGM analysis approaches (DF, MSF, MSE, and Kt) based on an earth mover’s distance (EMD) method. We further demonstrated that this similarity score can identify active spatial sites of AF in patients with unsuccessful AF termination, while no active sites of AF were present in patients with successful AF termination.

First, we aimed to demonstrate whether individual techniques (MSF, MSE, DF, and Kt) can discriminate between patients with successful and unsuccessful AF termination when applied to BiEGMs. In Figure 1, the box plots of the MSF, MSE, DF, and Kt values are shown for all patients with successful (m = 4) and unsuccessful (m = 4) AF termination. Notably, all the methods show statistical significance (p < 0.05) between the two groups of patients, suggesting the successful discrimination based on all BiEGMs.

To further demonstrate the performance of the BiEGM analysis approaches in identifying active sites of AF, the 3D visualization of individual approaches was performed using the custom-made VIEgram software. Figures 2, 3 show the representative examples of 3D maps of DF, MSF, MSE, and Kt obtained via VIEgram software for patients with unsuccessful and successful AF termination, respectively. The atria are visualized to show the left superior PV (LSPV) and the left inferior PV (LIPV). The corresponding example of a BiEGM signal from a single electrode (#) is also shown along with the individual DF, MSF, MSE, and Kt values from this recording. Notably, the visual inspection of spatial similarity between all four approaches does not suggest any correlation between the techniques both in a patient with unsuccessful and successful AF termination. Therefore, individual approaches alone cannot correctly identify the active sites of AF that should only be present in patients with unsuccessful AF termination.

To improve the performance of the various BiEGM analysis techniques for the identification of active AF sites, a similarity score was developed based on EMD to quantify the correlation between various pairs of the techniques at each spatial site. In Figures 4, 5, the representative examples of the EMD values calculated for different pairs of approaches at different spatial sites are shown for patients with unsuccessful and successful AF termination, respectively. The gray shaded area indicates the regions with EMD < 0.2, where all pairs are strongly correlated, and therefore, the similarity score is high. As indicated in Figures 4, 5, different techniques have a different degree of correlation with each other, and the degree of correlation is also different depending on the spatial site. We hypothesized that the spatial sites in which at least 5 pairs of the techniques are correlated, i.e., with a high similarity score of 3, might be associated with active AF sites.

The correlation data from Figures 4, 5 are summarized in Table 1 (patient P1 with unsuccessful AF termination) and Table 2 (patient P8 with successful AF termination), respectively, where the EMD values below (orange) and above (gray) the threshold (EMD = 0.2) are shown for all the spatial sites. Notably, in patient P1 with unsuccessful AF termination, there are three spatial sites (# 3, 8, and 14, red color) in which the similarity score is high (=3). Such a high similarity score indicates that at least 5 pairs of different approaches are correlated in these sites, suggesting the presence of active AF drivers, which most probably were not properly targeted during the AF ablation procedure. However, the absence of such red sites with a high similarity score in a patient with successful AF termination (Table 2) was noted, indicating that all active sites were appropriately targeted during the AF ablation procedure. Similar tables for all remaining patients are present in Supplementary Tables 3–8.

In Figures 6, 7, the similarity scores for different spatial sites are shown for all patients with unsuccessful (m = 4) and successful (m = 4) AF termination, respectively. Notably, the spatial sites with high similarity scores (=3, red) are only identified in patients with unsuccessful AF termination, suggesting that these sites can be the active AF sites and potential AF ablation targets. The sites were distributed among the different regions of the heart, namely, LA appendage, lateral wall, roof wall, posterior wall, inferior wall, septal wall, LSPV, and LIPV.

Some recent studies indicated that a PV isolation alone could be a strategy for catheter ablation in patients with paroxysmal AF, while recurrence is not fully prevented (Sawhney et al., 2009; Ouyang et al., 2010; Chao et al., 2012). However, other studies suggested the need for mapping-specific ablation therapy in patients with persistent AF when active AF sites are outside of the PV regions (Barbhayia et al., 2015; Seitz et al., 2017). For instance, it has been shown that EGM dispersion occurs in the vicinity of AF drivers, and this dispersion is present both near and outside the PV regions. Therefore, it is important to develop new techniques for identifying AF substrates outside the vicinity of the PV regions and potentially improving the success rates of the catheter ablation procedure. Several studies demonstrated the successful identification of AF driver by the phase mapping of simultaneous recordings using a basket catheter (Narayan et al., 2012) and a noninvasive array of body surface electrodes (Haissaguerre et al., 2013). It was also demonstrated in the study by Roney et al. (2017) that spatial resolution is essential in identifying the AF reentrant activity. But the reported success rates using these guided strategies for patients with persistent AF remained inconclusive and ranged from 19 to 77% (Narayan et al., 2014; Miller et al., 2017; Mohanty et al., 2018).

Recently, our laboratory developed and validated the novel iEGM analysis techniques, such as MSF (Arunachalam et al., 2016b), MSE (Arunachalam et al., 2018), and Kt (Arunachalam et al., 2016a), to identify the sites with abnormal AF activity that can be a potential target for AF ablation. Furthermore, we demonstrated the feasibility of these techniques in identifying the active sites of arrhythmia under clinical limitations, with acceptable levels of specificities and sensitivities using numerically simulated iEGMs (Ravikumar et al., 2021). Multiple other studies have analyzed other signal processing-based ablation techniques (Lin et al., 2010; Okumura et al., 2011; Kumagai et al., 2013; Atienza et al., 2014) built on the DF- and PVI-based strategies. This study shows that the performance of individual techniques alone is insufficient in identifying the active AF spatial sites (Figures 1, 2). To overcome this, we developed a similarity score that can help identify the active AF spatial sites by a combination of multiple methods.

We developed a similarity score that can bring together various signal processing approaches and successfully identify active AF sites based on the correlation between all the approaches. In this study, we demonstrated the performance of the similarity score based on four approaches, namely, MSF, MSE, DF, and Kt, to identify abnormal electrical sites in patients with unsuccessful AF termination. However, the similarity score might be extended and optimized by incorporating other iEGM analysis approaches and/or by combining a different number of approaches.

In this study, we have used a custom 3D visualization software previously developed (VIEgram) to perform a visual inspection of clinically recorded iEGMs and implement the previously developed novel iEGM analysis techniques (DF, MSF, MSE, and Kt). The recorded sites were distributed in these anatomical regions of the atria, LA appendage, lateral wall, roof wall, posterior wall, inferior wall, septal wall, LSPV, and LIPV. It was observed that the use of these individual approaches fails to robustly identify spatial sites for ablation; the 3D representations of individual methods among the same patients are visually not similar in the distribution of the metrics, thus increasing the difficulty and time consumed to interpret the different approaches at the same time, especially in patients with unsuccessful AF termination. To overcome this, in this study, we have developed a similarity score that combines all the individual approaches to provide the identification of the active AF site.

In this retrospective study, we investigated the performance of individual BiEGM analysis approaches (DF, MSF, Kt, and MSE) and newly developed similarity scores to identify potential, abnormal, electrically active sites in patients with previously unsuccessful AF termination. The major findings of this study are as follows: (1) individual approaches can discriminate between patients with successful and unsuccessful AF termination but fail to robustly identify spatial sites with active AF drivers, (2) a novel EMD-based similarity score was developed and validated to identify the active AF sites in patients with unsuccessful AF termination, and (3) there was no single common region in the atria associated with active AF sites in patients with unsuccessful AF termination, thus indicating the need for patient-specific mapping and ablation therapy.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by the University of Minnesota’s Institutional Review Board (IRB: STUDY00003128). The patients provided written informed consent for ablation and data collection for research purposes.

EGT, HR, and VR contributed to conceptualization. EGT and VR contributed to methodology. VR, ST, and XK contributed to software and scripts. VR contributed to validation and writing – original draft preparation. VR and ST contributed to formal analysis. VR, ST, XK, HR, and EGT contributed to writing – review and editing. ET contributed to supervision, project administration, and funding acquisition. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.