Author + information
- Received August 25, 2016
- Revision received April 6, 2017
- Accepted April 11, 2017
- Published online December 4, 2017.
- David Vidmar, MSca,
- David E. Krummen, MDb,c,
- Justin Hayase, MDb,c,
- Sanjiv M. Narayan, MD, PhDd,
- Gordon Ho, MDb,c and
- Wouter-Jan Rappel, PhDa,∗ ()
- aDepartment of Physics, University of California San Diego, San Diego, California
- bDepartment of Medicine, University of California San Diego, San Diego, California
- cVeterans Affairs San Diego Healthcare System, San Diego, California
- dDepartment of Medicine, Stanford University, Palo Alto, California
- ↵∗Address for correspondence:
Dr. Wouter-Jan Rappel, Department of Physics, University of California San Diego, 9500 Gilman Drive, La Jolla, California 92093.
Objectives The objective of this study was to evaluate the spatiotemporal organization and progression of human ventricular fibrillation (VF) in the left ventricle (LV) and the right ventricle (RV).
Background Studies suggest that localized sources contribute to VF maintenance, but the evolution of VF episodes has not been quantified.
Methods Synchrony between electrograms recorded from 25 patients with induced VF is computed and used to define the Asynchronous Index (ASI), indicating regions which are out of step with surrounding tissue. Computer simulations show that ASI can identify the location of VF-maintaining sources, where greater values of the maximum ASI (ASImax) correlate with more stable sources.
Results Automated synchrony analysis shows elevated values of ASI in a majority of self-terminating episodes (LV: 8/9; RV: 7/8) and sustained episodes (LV: 11/11; RV: 12/12). The locations of ASImax in sustained episodes colocalize with rotor cores when rotational activity is simultaneously present in phase maps (LV: 8/8; RV: 5/7; p < 0.05). The distribution of the ASImax differentiates self-terminating from sustained episodes (mean ASImax = 0.60 ± 0.14 and 0.70 ± 0.16, respectively; p = 0.01). For sustained episodes, the LV exhibits an increase in ASImax with time.
Conclusions Quantitative analysis identifies localized asynchronous regions that correlate with sources in VF, with sustained episodes evolving to exhibit more stable activation in the LV. This successive increase in stability indicates a stabilizing agent may be responsible for perpetuating fibrillation in a “migrate-and-capture” mechanism in the LV.
Ventricular fibrillation (VF) is a common, life-threatening arrhythmia responsible for many of the nearly 300,000 cases of sudden cardiac death in the United States each year (1). Due to its inherent complexity, our understanding of the sustaining mechanisms of VF remains incomplete. Notably, survival to hospital discharge after VF is largely unchanged over the past 30 years (2).
Animal models of ischemia-related VF have shown that it is primarily maintained by localized drivers, found in nonischemic tissue, with unorganized activation into border zone and ischemic tissue (3). Separately, spiral wave reentry (rotors), focal sources, and disorganized activity have been described in human VF (4–6), but the relative importance of each was unclear. Recent work in our laboratory mapped human VF and demonstrated that spiral waves, identified visually from computed phase maps, were present in the majority of patients (7). Furthermore, this study found that sustained episodes requiring defibrillation to terminate were characterized by greater spiral wave stability (7), and that ablation of VF sources increased the VF threshold in an animal model and patient with clinical VF (8).
Potential criticism of this prior work is that visual identification of phase movies may be operator dependent, have a moderate learning curve, and does not always capture all spiral wave activity (9). We endeavored to address this criticism by developing and validating automated techniques that do not require subjective interpretation to examine VF mechanisms. We hypothesized that markers of activation synchrony could identify areas that activate independently from neighboring tissue, and thus autonomously identify drivers of VF. These automated techniques can then be used to quantify the evolution of identified sources in time and permit additional insights into VF mechanisms.
In silico studies
Patient enrollment, electrophysiology study, and electrogram analysis
We enrolled 35 consecutive patients presenting for ventricular arrhythmia ablation at the University of California San Diego and VA San Diego Healthcare System. VF was inducible in 25 patients (Table 1); only the first episode induced in each patient per chamber was used in this analysis to ensure independent measurements. In total, we analyze 9 left ventricular (LV) and 8 right ventricular (RV) self-terminating episodes and 11 LV and 12 RV sustained episodes. Our electrophysiology study and electrogram analysis has been published elsewhere and is detailed in the Online Appendix. In short, electrogram recordings were obtained from basket catheters (64-electrode, Constellation, Boston Scientific, Natick, Massachusetts). VF was induced by rapid pacing and defibrillated after 11 ± 3 s.
Continuous variables are expressed as mean ± standard deviation; error bars are represented with the standard error of the mean. The Student t test was used to compare continuous variables, the Fisher exact test was used to compare nominal variables. The 2-sample Kolmogorov Smirnov test was used to test the hypothesis that datasets were drawn from the same distribution. Statistics were calculated using SPSS version 19 (IBM, Somers, New York) and MATLAB (Mathworks Inc., Natick, Massachusetts).
Our methodology of inferring and analyzing phase synchrony of activation time information has recently been described as detailed in the Online Appendix (11). In short, we use activation times marked directly from electrograms recorded during VF to compute the time-dependent phase of each region of tissue. This phase is then used to compute an Asynchronous Index (ASI), which quantifies how dynamically “out of step” a given electrode is with the surrounding tissue. It is high when an electrode: 1) exhibits dynamics which differ significantly from the global trend of the entire domain; and 2) is immediately surrounded by synchronous tissue, whereas it is low when the electrode’s dynamics are similar to its surroundings. Further, the value of ASI at the location of a source encodes information about its spatiotemporal stability, with a larger ASI implying greater stability.
Our computational models demonstrate that electrical recordings near VF spiral wave cores can become asynchronous with neighboring tissue due to spiral wave meander, or precession. Figure 1A shows a meandering spiral tip trajectory in black, along with electrodes placed within (i) and outside the spiral tip trajectory (j). As the spiral wave reaches point 1 and travels to point 2, both electrodes will activate as the depolarization front passes over them. As the rotor revolves around point 2 and travels to point 3, the activation will only pass over electrode i. From point 3 to point 4, the activation front will again pass over both electrodes, and the cycle restarts. Counting up the number of activations per cycle, we see that electrode positioned inside of the spiral tip trajectory will be activated 3 times. In contrast, the electrode located outside of the trajectory will only activate 2 times. Thus, this 3:2 activation pattern will cause asynchrony between electrodes inside the rotor trajectory and those outside, resulting in an elevated value of ASI.
Figure 1B shows the trajectory of a single simulated spiral wave (white line) that encompasses only a single electrode. This trajectory is plotted on top of the corresponding ASI map, computed using an 8 × 8 virtual electrode grid, in which the ASI value is color coded with red/blue representing large/small values. As expected, the ASI value is elevated at the electrode inside the tip trajectory (0.86), but low everywhere else (0.035 ± 0.002). Changing the length scale of the spiral wave trajectory or, equivalently, decreasing the interelectrode distance, results in more electrodes that lie within the trajectory and a larger region of elevated ASI. This pattern is shown in Figure 1C where the tip trajectory, again shown in white, now crosses several electrodes, leading to a larger asynchronous domain.
The weighting factor λ in the definition of ASI, which computes the synchrony of surrounding tissue, increases its specificity for spiral wave cores by decreasing ASI values in areas of extended asynchrony due, instead, to spiral wave breakup. Figure 2A shows a snapshot of a localized spiral wave in the presence of spiral wave breakup. Here, the dashed line indicates the boundary of the stable domain and the location of the spiral tip is marked by a dot. We constructed local and global distributions of synchronization numbers using an 8 × 8 virtual electrode grid and computed the ASI map (Figure 2B), which shows that the electrode within the spiral tip trajectory has an elevated ASI value, whereas electrodes outside that region exhibit low ASI values. This result can be understood by examining the distribution of synchronization numbers for electrode 1, γlocal,1 and located within the tip trajectory, as shown on the right, together with the global distribution of synchronization numbers, γglobal, also shown on the right. Whereas γglobal is almost uniform, γlocal,1 is sharply peaked at low synchronization numbers. Thus, these 2 distributions are markedly different, resulting in an elevated ASI value of 0.64. In contrast, the distribution of synchronization numbers for electrode 2, γlocal,2 located in the organized region is almost identical to the global distribution, leading to a small ASI value of 0.21. Electrodes located in the breakup region also record a small value of ASI (0.20 ± 0.05) because they are surrounded mostly by disorganization, leading to a small weighting factor λ.
The ASI values are low in the presence of multiwavelet re-entry. Figure 2C shows a snapshot of a simulation of multiwavelet VF, during which VF is not driven by a source but self-perpetuates by a continuous breakup of spiral waves. Importantly, because the entire domain is driven by multiwavelet reentry, γlocal will be similar at each location and, thus, similar to γglobal. This is shown on the bottom right, where we plot the distribution of synchronization numbers at location 3 as well as the global distribution of synchronization numbers. Because each electrode displays asynchronous behavior, the prefactor λ is also small for each electrode, resulting in all electrodes recording low ASI values of 0.05 ± 0.02 (Figure 2D).
Elevated values of ASI are common, particularly in sustained human VF
Based on our simulation results we define an ASI of >0.5 to be elevated and examine the occurrence of ≥1 electrodes in the LV and RV to have an elevated ASI. For self-terminating cases we find 8 of 9 LV episodes and 7 of 8 RV episodes recording an elevated ASI. For sustained cases we find 11 of 11 LV episodes and 12 of 12 RV episodes recording an elevated ASI. An example ASI map of sustained VF in a 56-year-old patient, along with γglobal and γlocal for the electrode with the largest value of ASI and a corresponding isochronal map, are shown in Figure 3. Raw electrograms at the site of elevated ASI, as well as a neighboring site of low ASI, are also provided for comparison. The electrodes with elevated ASI indicate a tissue region that is asynchronous compared with the global level of synchrony. The isochronal map reveals that this area of asynchrony corresponds to rotational activity as determined from activation times. Furthermore, this patient was ablated at a region coincident with this site of elevated ASI and VF was subsequently found to be noninducible.
Locations of ASImax correlate with rotor core sites
We compared the locations of ASImax in sustained episodes to rotor core locations when strong rotational activity was simultaneously present in phase maps (Online Appendix). This revealed that ASImax corresponds with visually determined spiral wave locations in 8 of 8 cases in the LV and 5 of 7 cases in the RV (p < 0.05).
Distributions of ASImax differentiate sustained from self-terminating episodes
We compared the distributions of ASImax between self-terminating and sustained episodes over different 3-s windows (Online Appendix). Because most self-terminating episodes were around 3 s in duration, we calculated ASImax across the first 3 s of self-terminating episodes. To show that this distribution changes over time in sustained VF, we compare this with the distribution of ASImax across the last 3-s interval for which data were available in all sustained episodes. The resulting distributions are shown in Figure 4A, with the sustained episodes recording a higher mean ASImax (0.70 ± 0.16) than self-terminating episodes (0.60 ± 0.14). These distributions are statistically distinct (p = 0.01) with the sustained one clearly shifted toward higher values of ASImax, implying that sustained VF evolves to exhibit more stable activation dynamics than those exhibited during self-terminating VF.
ASImax progressively increases in the LV during sustained VF
The present analysis is able to map the evolution of the spatiotemporal organization during VF by computing ASI maps at different time points during the episode. We computed ASImax across 3-s intervals, averaged over all patients, for both the LV and RV throughout all sustained episodes. This is reported every 500 ms, up to the end of the shortest sustained episode, in Figure 4. As is shown in Figure 4B, the average ASImax for the RV remains roughly constant (red line), whereas the average ASImax for the LV first shows a slight decrease in ASImax values, followed by an increase in ASImax (blue line). This increase in ASImax is further examined in Figure 4C, where we plot the distributions of ASImax in the LV and RV during the 3-s time interval centered at 4 s after VF initiation and in Figure 4D, where we plot these distributions centered at 7 s after VF initiation. The RV distribution around 4 s is indistinct from the RV distribution around 7 s, and the LV distribution shifts markedly toward higher values of ASImax as the episode progresses. Indeed, the LV distribution around 4 s is distinct from the LV distribution around 7 s (p = 0.01).
Regions of elevated ASI can migrate
Using our synchrony analysis, we can examine the temporal evolution of the region of elevated ASI. An example of this is presented in Figure 5, where we show ASI and isochronal maps of a sustained VF episode in a 72-year-old man with ischemic cardiomyopathy, prior bypass surgery, and multiple percutaneous coronary interventions. The left panel shows VF at 4 s after initiation and the right panel shows VF at 6 s after initiation. The isochronal map of this patient (Figure 5, left side) shows clear rotational activity with a location that is migrating from the midbasal, posterolateral LV to the midapical posterolateral LV. The ASI map of this patient reveals regions of elevated ASI that are consistent with the rotational activity in the isochronal maps. This patient was described in an earlier report (8) and had, before ablation, repetitive implantable cardiac defibrillator shocks due to VF refractory to antiarrhythmic drugs. Targeted ablation was carried out at the site identified on the right side of Figure 5 (ASImax in the LV) and at 3 other sites, including one that colocalized with our computed ASImax in the RV. After ablation, VF was unable to be re-induced with standard pacing protocol. Furthermore, the patient has had no sustained VF events and zero implantable cardiac defibrillator therapies with no antiarrhythmic medications in the 3.3 years since the rotor site ablation procedure, suggesting that the ablated regions of tissue served a substantial role in allowing or perpetuating fibrillation.
Clinical VF results correspond with migrate-and-capture simulation
These results suggest a migrate-and-capture mechanism in which spiral waves that are formed after VF initiation may migrate and become preferentially located in specific parts of the left ventricle. To investigate this progression of VF episodes, we simulated a meandering spiral wave subject to drift as shown in the snapshots of Figure 5, right side. Here, the membrane potential is color coded and the spiral tip trajectory is shown in white. A square heterogeneity (Figure 5, right side), composed of tissue with 30% of cells randomly assigned as nonconducting, is introduced. As a result of the drift, the spiral wave migrates across the computational domain. Hence, the corresponding ASI map only shows weakly elevated ASI values at the tip region due to the spatiotemporal transiency of the source. Once the spiral wave has reached the heterogeneity, however, it becomes temporarily “captured” due to slowed conduction and experiences a relatively stationary meander pattern, resulting in an ASI map that shows a region with a highly elevated ASI, consistent with our clinical results.
This study shows that ASImax, a novel marker of VF source activity, can provide unique insight into the dynamics and progression of VF episodes. Our results demonstrate that localized sources are seen in nearly all cases of VF, with ASImax sites correlating with rotors identified visually. Tracking the stability of these sources with time suggests a migrate-and-capture mechanism for VF perpetuation in the LV.
Insights into VF sources
Our analysis is motivated by computational and experimental studies, which show spiral tips can exhibit complex and meandering trajectories (12–14). Consequently, one can distinguish at least 2 different dynamical regions within the heart: one that corresponds with the spiral tip path with complex and asynchronous activation and one that corresponds with synchronous tissue, which is activated by the arms of the spiral wave (Figure 1). Using computational simulations, we have shown that, as a result, we can determine the location of a meandering spiral wave using ASI, a measure of the spatiotemporal stability of a VF source (Figure 2).
The analysis we have presented here requires several factors to be in place for our methodology to be successful. First, the spiral wave tip needs to be meandering and its trajectory needs to cover at least 1 electrode. Thus, tip trajectories with a length scale much smaller than the spatial resolution or reentry with a nonmeandering tip trajectory will not result in an elevated ASI. For example, anatomic reentry, which is consistently anchored at tissue heterogeneities and likely to be associated with ventricular tachycardia, is not expected to be discernible on ASI maps. Second, the tissue harboring the tip trajectory needs to have different dynamics than the surrounding tissue. Thus, the coherent domain, that is, the arms, of the spiral wave reentry, needs to span at least the neighboring electrodes. Third, the global synchrony distribution needs to be sufficiently different from local distributions associated with sources. Therefore, as mean synchrony across electrodes becomes lower, it becomes harder to distinguish ASImax due to sources from inherent fluctuations. Finally, the reentry needs to be present for a minimum period of time, resulting in a sufficient number of activations, chosen here as 6 to 7 VF activations per electrode. Synchronization numbers are challenging to compute for a much smaller time window and much longer time windows will not capture migrating tip trajectories (Figure 5).
Advantages of ASI mapping
Compared with other strategies, our methodology has distinct features. Unlike phase or isochronal mapping analysis, ASI maps can be calculated in an automated fashion and determining the location of ASImax does not require a manual interpretation of movies. This objective localization has mechanistic implications, for determining stability or relationship to other physiological areas, but also clinical implications for ablation or other therapy. The automated index affords not only objectivity in determining VF dynamics, but also allows us to analyze many different patients and episodes, and summarize them in a simple and robust statistic. This ability to quantify and compare dynamical trends across VF episodes is unachievable through visual inspection of phase maps alone. Also, and unlike techniques that use information based on local electrode cycle length, including dominant frequency analysis (15,16), ASI mapping not only takes into account local activation sequences, but also takes into account global measures as a form of normalization. Thus, our methodology enables the determination of regions that are out of step compared with the global level of synchrony and allows us to distinguish a spatially stable spiral tip from multiwavelet reentry, potentially providing important insights into the mechanisms that sustain human VF.
Earlier studies have revealed evidence for rotors in human VF (6,17,18), and ablation at these tissue areas may suppress subsequent episodes of VF (12). More recent work using endocardial basket electrodes demonstrated that rotational activity is present in the vast majority of patients after induction of VF and that the stability of the rotors is larger in patients with sustained VF episodes compared with patients with self-terminating ones (7). A critical element in that study was the manual interpretation of animated sequences of phase maps. In contrast, our technique is fully automated and does not require subjective analysis of data. Importantly, for the patients in this dataset, elevated ASI values were seen in nearly all cases of VF, further supporting the role of spiral waves in maintaining this arrhythmia.
Evolution of fibrillation in the left and right ventricles
Our analysis revealed that the average ASImax for all sustained VF episodes does not vary significantly in the RV, although it decreases first and then increases in the LV (Figure 4). This pattern is also evident from the distribution of ASImax: for the LV, this distribution becomes more peaked around high values as the episode progresses. In other words, regions of elevated ASI become more pronounced in sustained episodes, indicating rotational activity that becomes spatiotemporally more stable and/or controls more surrounding tissue. The increased stability is evident in Figure 5, which shows a migrate-and-capture scenario: the spiral wave migrates toward an area where it is trapped and remains “captured” before defibrillation. Self-terminating episodes, in contrast, remain at relatively low values of ASImax, suggesting the absence of a stabilizing mechanism for VF sources in these episodes. Whether this scenario is applicable to all VF episodes and what other mechanisms may cause the increased stability of the spiral waves remains unclear. Of note, however, reports have suggested that zones of ischemia and scars might play a prominent role in the maintenance of VF (5,19–21). Furthermore, the reduced thickness of the RV has been hypothesized to permit only stable rotors whereas the thicker LV tissue tends to exhibit diverse activation patterns (22), consistent with these results. This difference in activation complexity between the LV and the RV has also been observed experimentally in mapping studies of fibrillating pig ventricles (23).
Altogether, our data suggest the following general progression of VF: initiation of VF creates multiple spiral wave reentry patterns in both sustained and self-terminating cases. In sustained cases, the migration of these spiral waves to regions in the LV favorable for prolonged spiral wave activity results in the perpetuation of VF. In self-terminating cases, activation is unstable; therefore, these spiral waves migrate and inherently self-terminate, either through spiral–spiral collisions or through interactions with nonconducting boundaries. Further studies are required to verify this potential scenario for the dynamical progression of VF.
In contrast with phase mapping, synchrony analysis is able to reveal sources of VF without explicitly searching for rotational activity. We compared the locations of ASImax with rotor locations whenever strong rotational activity was present in the phase maps. This comparison revealed that ASImax coincides with rotor cores in the majority of such patients. In several cases, the ASImax occurred during a time interval where the phase maps did not show clear rotational activity. This could be due to inexact activation times, which would also obscure rotational activity as determined using phase maps. Additionally, large-scale spiral wave meander could make the visual identification of rotational activation patterns challenging, although it will make the determination of sources in our synchrony analysis more straightforward. Therefore, our analysis could potentially identify sources obscured in phase map analysis by focusing on synchrony rather than on identifying complete rotational circuits.
First, the spatial resolution of our ASI maps is constrained by the spatial resolution of our baskets (5 to 10 mm). Importantly, however, the required resolution to determine asynchronous regions is such that at least 1 electrode falls within the spiral tip trajectory while its neighbors are within the domain controlled by coherent spiral wave activation. This argument is similar to the one that can be made for accurate mapping of spiral wave reentry (24). Second, the methodology relies on the accurate marking of activation times from electrogram data. Because the methodology compares against global synchrony, however, ambiguous signals and mismarkings can be discarded without penalty to the overall analysis, as long as a sufficient total number of clear signals exist. Third, the mapping system only registers the endocardium and is not able to record intramural events or epicardial activation. The system, however, is able to determine endocardial spatiotemporal activity that has mechanistic relevance (8,21). Combining endocardial mapping with an epicardial registration modality would be an interesting future direction. Fourth, the ASI metric has not been validated beyond the scope of this study. Finally, VF is a dynamic condition in which electrophysiological properties such as dominant frequency (19), action potential duration, conduction velocity, and tissue ischemia continually progress (25). In this study, VF was terminated at 11 ± 3 s for patient safety, and thus the results of this study may only be applicable in early VF. Nonetheless, early VF is an important link in the chain of events leading to sustained VF, and is also the time period in which symptoms such as syncope first present. Thus, an improved understanding of early VF is critical to improve current and emerging VF therapies, including antiarrhythmic drugs, defibrillation, and ablation (8). Furthermore, our findings are complementary to seminal work performed in longer duration VF (26), in which alternative mechanisms may predominate.
COMPETENCY IN MEDICAL KNOWLEDGE: In patients with VF, sources in the form of rotational activation patterns can be identified based on the characteristics of electrical recordings at spiral wave cores. These data suggest that the formation of stable spiral waves in the LV is a critical step in the maintenance of VF.
TRANSLATIONAL OUTLOOK: Additional research is needed to determine whether ASI can be used in conjunction with phase mapping to prospectively guide VF ablation.
This work was supported in part by the American Heart Association 10 BGIA 3500045 (to Dr. Krummen), 16PRE30930015 (to Mr. Vidmar), and UCSD Clinical Translational Research Institute (to Drs. Krummen and Ho). Drs. Rappel and Narayan are co-authors of intellectual property owned by the University of California Regents and licensed to Abbott EP. Dr. Narayan has received honoraria from Medtronic and St. Jude Medical. Dr. Krummen has consulted for Abbott EP (terminated >14 months ago); and has received EP fellowship support from Biotronik, Biosense-Webster, Boston Scientific, Medtronic, and St. Jude. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
Jonathan Kalman, MD, served as Guest Editor for this paper.
All authors attest they are in compliance with human studies committees and animal welfare regulations of the authors’ institutions and Food and Drug Administration guidelines, including patient consent where appropriate. For more information, visit the JACC: Clinical Electrophysiology author instructions page.
- Abbreviations and Acronyms
- asynchronous index
- left ventricle/ventricular
- right ventricle/ventricular
- ventricular fibrillation
- Received August 25, 2016.
- Revision received April 6, 2017.
- Accepted April 11, 2017.
- 2017 American College of Cardiology Foundation
- Sasson C.,
- Rogers M.A.,
- Dahl J.,
- Kellermann A.L.
- Zaitsev A.V.,
- Guha P.K.,
- Sarmast F.,
- et al.
- Nair K.,
- Umapathy K.,
- Farid T.,
- et al.
- Krummen D.E.,
- Hayase J.,
- Morris D.J.,
- et al.
- Krummen D.E.,
- Hayase J.,
- Vampola S.P.,
- et al.
- Vidmar D.,
- Narayan S.M.,
- Rappel W.J.
- Nash M.P.,
- Mourad A.,
- Clayton R.H.,
- et al.
- Davidenko J.M.,
- Kent P.F.,
- Chialvo D.R.,
- Michaels D.C.,
- Jalife J.
- Bradley C.P.,
- Clayton R.H.,
- Nash M.P.,
- et al.
- Jeyaratnam J.,
- Umapathy K.,
- Masse S.,
- et al.
- Winfree A.
- Huizar J.F.,
- Warren M.D.,
- Shvedko A.G.,
- et al.