Catheter-based radio-frequency ablation is a promising method to treat atrial fibrillation (AFib), but leads to atrial flutter (AFlut) in 4.7–31 % of all cases . Diagnosis of AFlut caused by AFib ablation, however, is particularly difficult, due to substrate modification of the myocardium. Since signals can have very low amplitudes in areas of slow conduction, common methods to diagnose AFlut such as P wave evaluation are not usable.
Therefore, this paper presents an approach to replicate the clinically measured flutter circuit in a computer model using a reentry variant of the fast marching (FaMa) method. The simulation considers ablation scars and is used to estimate the global and local conduction velocity (CV) to reproduce the flutter circuit . The dynamic flutter simulation will support the physician during diagnosis of and treatment planning for AFlut, for example by giving information about the effect of ablation lines on the flutter circuit.
2 Material and methods
2.1 Clinical data
Clinical data from two patients were acquired using the 3D electroanatomical mapping system Velocity (St. Jude Medical). Both patients provided informed consent. Patient A (male, aged 54) was mapped using a 10-pole single-loop catheter during AFlut characterized by a basic cycle length (BCL) of 253 ms, a macro reentry around the pulmonary veins (PVs) and an isthmus in the roof line. Patient B (male, aged 56) was mapped with a 20-pole double-loop catheter during counterclockwise peri-mitral flutter with a BCL of 267 ms. Both tachycardias were stable during mapping. The measured local activation time (LAT) maps were exported containing one second of data for each collected intracardiac electrogram (EGM) at a sampling rate of 2034.5 Hz.
2.2 Fast marching simulation
Electrophysiological models are used to simulate the electrical excitation of the heart and can be divided in Eikonal models and biophysical cellular models coupled with excitation propagation. The FaMa method is an Eikonal model simulating the excitation wave on the cardiac surface by calculating the activation time of each surface node. Thereby, the wave propagation is approximated with the Eikonal equation. In contrast to biophysical cellular models, Eikonal models do not consider ion transport through the membrane. Accordingly, Eikonal models have fast computing times and are consequently easier to use under clinical time constraints.
This work is based on a C++ implementation of the FaMa simulation (FaMaS) .
2.3 Manual feature detection
Anatomical features of the flutter circuit were detected manually on the basis of an excitation movie and an LAT map. As an example, the manual feature detection of the clinical flutter circuit of Patient A was as follows: The LAT map (see Figure 1a) indicates that the excitation wave came from the anterior wall, narrowed and extended to the posterior wall after passing through an isthmus (green circle) on the roof (LAT: 0 ms). An energy-based excitation movie showed that the PVs were not or only partially excited. Since the LATs on the antero-septal side of the LA (see Figure 1b) were smaller than the LATs on the lateral side, the excitation wave obviously conducted a U-turn on the anterior wall.
As the flutter circuit is a reentry mechanism, the start of excitation was defined in the isthmus where least tissue was excited. In contrast to the real clinical case, the simulated flutter circuit had to be started once which required an initial refractory time for the tissue ‘behind’ the excitation wave.
Scars and initial refractory lines were defined on the LA geometry using specific points of the respective line and a MATLAB implementation of the Dijkstra algorithm. The surface geometry, trimesh structure and state information (scar, start of excitation, initial refractory, normal) were written to a Visualization Toolkit (VTK) file for processing with FaMaS.
2.4 Estimation of the conduction velocity
2.4.1 Global conduction velocity
FaMaS was used to estimate the global CV by approximating it iteratively using the clinically measured BCL as a reference.
The first step was to run the simulation with any guessed global CV for four cycles to reduce the effect of initial refractory times. The difference between the calculated LAT of the fourth and third cycle (for each activated nodei ) served as simulated BCL:
with N being the number of activated nodes of the mesh. If the deviation between calculated BCLFaMaS and clinically measured cycle length BCLclinical was below a threshold for acceptance (here: 1 ms), the estimation of the global CV was finished. Otherwise, the CV for the next iterationk was approximated as follows:
2.4.2 Local conduction velocity
Since the myocardium is not homogeneous due to ablation scars and slow conduction areas, the local CV was estimated in a next step. The CV was approximated iteratively by comparing the calculated time stamp to the clinically measured LAT for each node where an electrode was present (‘electrode node’).
A ‘zone’ contained the electrode nodes which were considered in the current iterationk. In case of CV adaption, the zone was divided in two ‘subzones’. If the CV of both sub-zones of the current iteration was estimated, the zone of the previous iteration (‘superzone’) was examined.
First, the global CV was estimated as described above, and all electrode nodes were ordered according to their clinical LAT. Besides, the surface was subdivided in Voronoi cells (‘electrode areas’) according to the electrodes such that each node of the surface belonged to its closest electrode.
The LAT was computed for each electrode node of the current zone using the estimated global CV. If the deviation between calculated and clinically measured LAT for an electrode node was greater than a threshold for acceptance, the CV of this electrode area was adapted as follows:
In this work, the local estimation was evaluated for two thresholds for acceptance (1 ms and 5 ms). To achieve stability, the CV was limited by a lower and upper threshold, and by an upper threshold for the difference between CVs in adjacent electrode areas. The parameters were varied according to Table 1. For the next iteration, the current zone including the electrode nodes was divided in two subzones. The procedure started at the beginning regarding the first subzone.
Variation of the threshold parameters.
|Minimal CV difference||200||500||800|
If the deviation between calculated and clinically measured LAT was smaller than the threshold for all electrode nodes of the current subzone, the approximation of the CV for the second subzone followed.
If the approximation of the CV was finished for both sub-zones, the next superzone was examined.
After the CV of the last zone was adapted, the estimation of the local CV was finished and a second global adjustment followed:
The procedure was evaluated with the help of simulated test cases. Several combinations of material classes with different CVs were defined on a spherical geometry ( 10 cm, surface mesh: 1517 nodes). 114 nodes served as electrode nodes and their time stamps as clinical LATs.
For processing with clinical data, the LAT maps were interpolated using Laplacian minimization.
3.1 Manual feature detection
During the manual feature detection, a scar was placed along the roof line and around the PVs (see Figure 2a) for Patient A. In order to realize the U-turn, a further scar was placed on the anterior wall since otherwise the excitation would propagate directly and straightforward along the mitral valve in an area of homogeneous tissue (see Figure 2b). In order to assign the propagation direction, the start of excitation defined in the isthmus on the roof, was surrounded by an area of initial refractory nodes and by scars (see Figure 2a).
3.2 Estimation of the conduction velocity
3.2.1 Global conduction velocity
In the test cases with known CV (varying in the range of typical atrial CVs
After estimating the global CV for clinical cases, the calculated BCL deviated less than 1 ms from the clinical BCL (0.39 % for Patient A, 0.37 % for Patient B related to the BCL). The approximation of the global CV was already finished after one iteration and the computation time was shorter than 1 s.
3.2.2 Local conduction velocity
For the simulated test cases with spherical geometry, the CV of the majority of the surface was approximated with a deviation of less than
Table 2 presents the results of the calculation time and the deviation between calculated and measured LAT for the clinical cases. Concerning the latter, the average and median deviation between the LAT of all electrodes were calculated after the final global adjustment.
Local estimation results for the parameter combination which resulted in smallest deviation (A: error per electrode: 1 ms, B: error per electrode: 5 ms, max. CV:
|Patient||Calc. time||Average deviation||Median deviation|
|A||6.8–8 s||45 ms (0.18 % BCL)||48 ms (0.19 % BCL)|
|B||17–20 s||25 ms (0.09 % BCL)||18 ms (0.07 % BCL)|
A further quality criterion was the resulting BCL after the final global adjustment which agreed with the clinically measured one for both patients. Figure 3 shows the estimated local CVs on the LA geometry.
4.1 Estimation of the conduction velocity
4.1.1 Global conduction velocity
Test cases with known CV proved that the global CV can be estimated accurately (deviation < 2 %).
Concerning clinical cases, the final BCL deviated less than 1 ms from the clinical BCL which equated to nearly 0.4 % for both patients. For typical BCLs of AFlut (200– 350 ms [4, 5]) the deviation was less than 0.5 %. Therefore, the deviation was tolerable for both example patients and other typical cases of AFlut. The calculation time (< 1 s) was suitable for clinical use.
4.1.2 Local conduction velocity
The estimation of the local CV delivered good results for all simulated test cases. Despite wrongly estimated CVs on a boundary, the estimated local CV can be used to detect slow conduction areas, which was the main purpose.
Concerning clinical cases, single LATs after local estimation deviated significantly from the clinical measurement (Patient A: ~20 % of BCL, Patient B: ~8 % of BCL). Overall, the local estimation replicated the excitation propagation adequately, which can be seen by comparing the clinical excitation movie to the visualized FaMaS at which distinctive propagation features, for instance U-turn and wave front collision, were well replicated. The calculation time (< 20 s) was suitable for clinical use. In the course of estimating the local CV, the clinical LAT map was interpolated due to outlier LATs which may result from fractionated EGMs or from moving the catheter during the recording period.
The simulated flutter circuit provided information about the excitation propagation subject to manually defined anatomical features. The average CV of the flutter circuit can be well described by estimating the global CV. Concerning the local estimation, the clinical measurement was not reproduced exactly as the resulting LATs showed. Possible reasons were the interpolation of the LAT map and limitation of the CV. A larger number of signals with higher quality is supposed to provide better results. Overall, the estimated local CVs approximated the propagation on the atrial surface. Since FaMaS was able to simulate clinical cases of AFlut and to consider predefined scars, the effect of arbitrarily chosen and manually annotated scars on the excitation propagation can be evaluated. In order to support the physician during diagnosis of and treatment planning for AFlut, the next step is an automatized detection of electrophysiological features such as scars, e.g. with the help of EGM signal analysis.
Conflict of interest: Authors state no conflict of interest. Material and Methods: Informed consent: Informed consent has been obtained from all individuals included in this study. Ethical approval: The research related to human use has been complied with all the relevant national regulations, institutional policies and in accordance the tenets of the Helsinki Declaration, and has been approved by the authors’ institutional review board or equivalent committee.
Castrejón-Castrejón S, Ortega M, Pérez-Silva A, et al. Organized atrial tachycardias after atrial fibrillation ablation. Cardiology Research and Practice 2011; 1–16
Chinchapatnam P, Rhode KS, Ginks M, et al. Model-based imaging of cardiac apparent conductivity and local conduction velocity for diagnosis and planning of therapy. IEEE Transactions on Medical Imaging 2008; 27: 1631–1642
E Poremba. Implementation of a fast simulation C++ framework for the computation of vulnerability to artial arrhythmias using the fast marching algorithm. Institute of Biomedical Engineering, Karlsruhe Institute of Technology (KIT), 2013
Itoh T, Kimura M, Sasaki S, et al. High correlation of estimated local conduction velocity with natural logarithm of bipolar electrogram amplitude in the reentry circuit of atrial flutter. Journal of Cardiovascular Electrophysiology 2013; 1–8
Jaïs P, Matsuo S, Knecht S, et al. A deductive mapping strategy for atrial tachycardia following atrial fibrillation ablation: importance of localized reentry. Journal of Cardiovascular Electrophysiology 2009; 20: 480–491