Noncontact Frequency Mapping of Atrial Fibrillation
Frequency mapping of atrial fibrillation (AF) has gained widespread interest since Nademanee and colleagues reported the ablation of sites harboring high frequency electrograms as an alternative to pulmonary vein isolation for treating AF. However, frequency analysis of fibrillatory signals has been investigated for some time. Ropella and colleagues first performed Fourier analysis of right atrial fibrillatory signals and found that there was typically a peak in the 4-9 Hz band, suggesting an underlying organization to AF. Botteron and Smith derived a space constant that described the cross-correlation of fibrillatory signals over distance, and used this metric to measure the effect of antiarrhythmic drugs on AF organization. Monsour and colleagues performed Fourier analysis of biatrial fibrillatory signal in a sheep model of AF. They found that the fastest frequencies recorded during AF occurred in the posterior left atrium and that a left-to-right atrial frequency gradient existed during AF. This important contribution suggested that the posterior left atrium/pulmonary vein (PV) region played a role in maintaining AF, a finding that contradicted the prevailing multiple wavelet hypothesis. We later confirmed that this left-to-right atrial dominant frequency (DF) gradient exists during human paroxysmal AF, but was attenuated during longer-lasting persistent AF. This suggested that while the PV/posterior left atrial region was important for maintaining paroxysmal AF, other regions outside the posterior left atrium may play a role in maintaining persistent AF.
The question is how to best identify these AF drivers that lie outside of the posterior left atrium. Nademanee visually identified and targeted sites containing low amplitude and high frequency (<120 ms) electrograms, so-called complex fractionated atrial electrograms (CFAEs). However, this required extensive ablation at multiple sites throughout the right and left atria and the targeted sites were subjectively chosen. Unfortunately, we still do not have a good understanding of what CFAEs represent. It is unlikely they all represent "drivers" of AF. Many may simply be recordings of overlapping wavefronts in areas of complex anisotropic myocardium such as the atrial septum, left atrial appendage/left pulmonary vein ridge, or the junction of the pulmonary veins and posterior left atrium. Others may represent transient collision of multiple wavefronts. Empiric ablation of all these sites may not be the most efficient strategy for treating AF. We do not know how stationary these CFAE sites are over time or whether transient CFAE sites represent a different mechanism than persistent CFAE sites.
Several investigators have sought to use commercial mapping systems to quantify more objectively these sites. Sanders and colleagues performed biatrial frequency mapping of AF using electroanatomic mapping. They found that patients with persistent AF often had DF sites located outside the pulmonary veins and that ablation sites that resulted in AF termination retrospectively correlated to sites of high dominant frequency. However, construction of DF maps using this approach required sequential point-by-point acquisition, which may take 30 minutes or longer per map to acquire. This approach assumes that these DF sites are stationary, that is, not changing significantly over time. A system that allowed simultaneous high-density acquisition of multiple atrial sites would be extremely useful for assessing CFAE stability, distribution, and the mapping resolution required to identify CFAE sites. Noncontact mapping holds the promise of rapid simultaneous acquisition of electrograms throughout a cardiac chamber from an intracavitary multielectrode array (MEA) using the inverse solution method. Despite a prior study validating the reproduction of these "virtual electrograms" in the atria, the system has not achieved widespread use. This is largely explained by several concerns including (1) potential thrombogenicity of the electrode array in the systemic circulation, (2) inaccuracy of virtual electrogram reproduction with increasing distance from the MEA, and (3) limitations of far-field sensing of ventricular activation obscuring atrial activation during the QRS complex.
However, the idea that noncontact mapping can be used for global frequency mapping is an attractive one. A recent paper by Schilling and colleagues did find that the correlation between contact and noncontact atrial electrograms recorded during AF was reasonable (0.7), but fell off significantly with distance. It is important to realize, however, that a perfect reproduction of the contact electrogram morphology may not be required if the frequency content of the signal at each site could be adequately reproduced. Could the noncontact approach be particularly well suited to global frequency mapping of AF?
In this issue of the Journal, Lin and colleagues examine the accuracy of the noncontact MEA for reproducing contact electrogram morphology and DF measurements during human AF. The authors studied 12 patients with predominantly paroxysmal AF. Recordings from the noncontact multielectrode array were made during AF from the right atrium in seven patients and the left atrium in five patients. Contact unipolar electrograms were recorded simultaneously with noncontact "virtual" unipolar electrograms for comparison. Seven-second segments were exported for offline quantitative analysis. The authors localized each contact recording point into one of six right or six left atrial regions, and then grouped the total of 159 sites in both atria for the 12 patients together for analysis. They then compared the 7-second segments for electrogram morphology using the correlation coefficient and frequency spectra using the magnitude squared coherence (MSC) between frequency spectra and the absolute difference in DF between contact and noncontact electrograms. A cutoff of the correlation coefficient >0.7, mean MSC >0.5, and DF difference <0.5 Hz was chosen as representing an "acceptable" comparison between contact and noncontact electrograms.
The mean correlation between contact and noncontact electrogram morphology was 0.75 ± 0.17 and was less accurate with increasing distance from the array (>38 mm), as reported previously. The authors also note that the morphologic correlation was less accurate in the regions of the pulmonary veins, superior vena cava, left atrial appendage, and mitral areas, all locations where the "line of sight" to the MEA might be impaired. One reason for the lower correlation between contact and noncontact signals was a "time-misalignment," also known as a phase shift, between the two signals. When correcting for this time misalignment, the overall correlation improved impressively to 0.85 ± 12. The independent predictors of a poor correlation included a greater time misalignment, a greater distance from the contact catheter to the array center, and a lower peak-to-peak atrial voltage.
When comparing the frequency spectra of the 7-second segments, the mean MSC was >0.5 at most sites and there was no difference between sites closer to or further away from the array center. The dominant frequencies measured for contact and noncontact electrograms were within 0.5 Hz for 94% of mapping sites. The figures show nice examples of sites where the contact and noncontact comparisons were good and other sites where the comparisons were poor.
Are we ready to accept noncontact "virtual" frequency maps as a surrogate for contact frequency maps? There are several limitations and concerns that limit wholehearted acceptance of the noncontact approach for frequency mapping at the present time. First, both the right and left atria were not sampled in each patient, and the density of sampling in each patient was limited. Only approximately 13 points were compared per patient and we have no understanding of how these points were chosen. Were points randomly distributed throughout the atria? Were points deep inside the appendage or pulmonary veins that would be expected to have poor correlation sampled? It is difficult to have complete confidence in the robustness of the comparisons without a higher resolution sampling of atrial points. Second, the exact areas where the MEA has limitations--the pulmonary veins, the left atrial appendage, the mitral valve, and low voltage areas--turn out to be the areas we care most about! These are all sites where CFAEs have been most commonly reported. Inability to interpret frequencies in areas of low voltage may be a significant limitation of noncontact mapping. Additional details of a particular voltage cutoff below which noncontact mapping may have lower accuracy would have been welcome. Other reports have also found difficulties interpreting areas of split potentials in sinus rhythm, again typically areas of anisotropic myocardium that are of prime interest to those interested in localizing CFAE sites. In addition, dilated human atria with long-standing AF will often have significant surface areas >38 mm from the array center. Perhaps concentrating the array on specific regions and performing more detailed mapping in these regions will obtain better results. Finally, the cutoffs chosen by the authors for identifying "adequate" comparisons may have been a bit too liberal. Since correlation is insensitive to electrogram amplitude, a correlation coefficient of 0.75 may be present between two electrogram with grossly different appearance. An MSC of >0.5 Hz was chosen as a reasonable similarity of the frequency spectra; however, it is not clear that this cutoff is adequate. In a prior paper, we found that a left-to-right atrial DF gradient >0.25 Hz in patients with persistent AF predicted long-term freedom from AF. This difference is beyond the 0.5 Hz resolution reported by the authors.
In summary, Lin and colleagues provide several new findings fundamental to the concept of noncontact global frequency mapping. First, virtual unipolar electrograms have a phase lag between simultaneously recorded contact unipolar electrograms and this should be corrected before comparisons are made between contact and noncontact electrogram morphology. Second, while electrogram morphology comparisons fall off with distance from the array center, frequency comparisons do not. This is a significant strength of the noncontact system. Third, the authors have reported that reasonable correlations exist between the frequency spectra calculated by contact and noncontact virtual unipolar electrograms. These findings, for the first time, provide validation for using the MEA array for frequency mapping, keeping in mind the frequency resolution of 0.5 Hz for DF mapping. If these findings are reproduced in more detail in future studies, frequency mapping of AF may lead to a resurgence of noncontact mapping. Hopefully, a better understanding of the mechanism of CFAEs will eventually lead to a better understanding of the mechanism of AF as well as more appropriate targets for AF ablation.
Editorial Comment
Frequency mapping of atrial fibrillation (AF) has gained widespread interest since Nademanee and colleagues reported the ablation of sites harboring high frequency electrograms as an alternative to pulmonary vein isolation for treating AF. However, frequency analysis of fibrillatory signals has been investigated for some time. Ropella and colleagues first performed Fourier analysis of right atrial fibrillatory signals and found that there was typically a peak in the 4-9 Hz band, suggesting an underlying organization to AF. Botteron and Smith derived a space constant that described the cross-correlation of fibrillatory signals over distance, and used this metric to measure the effect of antiarrhythmic drugs on AF organization. Monsour and colleagues performed Fourier analysis of biatrial fibrillatory signal in a sheep model of AF. They found that the fastest frequencies recorded during AF occurred in the posterior left atrium and that a left-to-right atrial frequency gradient existed during AF. This important contribution suggested that the posterior left atrium/pulmonary vein (PV) region played a role in maintaining AF, a finding that contradicted the prevailing multiple wavelet hypothesis. We later confirmed that this left-to-right atrial dominant frequency (DF) gradient exists during human paroxysmal AF, but was attenuated during longer-lasting persistent AF. This suggested that while the PV/posterior left atrial region was important for maintaining paroxysmal AF, other regions outside the posterior left atrium may play a role in maintaining persistent AF.
The question is how to best identify these AF drivers that lie outside of the posterior left atrium. Nademanee visually identified and targeted sites containing low amplitude and high frequency (<120 ms) electrograms, so-called complex fractionated atrial electrograms (CFAEs). However, this required extensive ablation at multiple sites throughout the right and left atria and the targeted sites were subjectively chosen. Unfortunately, we still do not have a good understanding of what CFAEs represent. It is unlikely they all represent "drivers" of AF. Many may simply be recordings of overlapping wavefronts in areas of complex anisotropic myocardium such as the atrial septum, left atrial appendage/left pulmonary vein ridge, or the junction of the pulmonary veins and posterior left atrium. Others may represent transient collision of multiple wavefronts. Empiric ablation of all these sites may not be the most efficient strategy for treating AF. We do not know how stationary these CFAE sites are over time or whether transient CFAE sites represent a different mechanism than persistent CFAE sites.
Several investigators have sought to use commercial mapping systems to quantify more objectively these sites. Sanders and colleagues performed biatrial frequency mapping of AF using electroanatomic mapping. They found that patients with persistent AF often had DF sites located outside the pulmonary veins and that ablation sites that resulted in AF termination retrospectively correlated to sites of high dominant frequency. However, construction of DF maps using this approach required sequential point-by-point acquisition, which may take 30 minutes or longer per map to acquire. This approach assumes that these DF sites are stationary, that is, not changing significantly over time. A system that allowed simultaneous high-density acquisition of multiple atrial sites would be extremely useful for assessing CFAE stability, distribution, and the mapping resolution required to identify CFAE sites. Noncontact mapping holds the promise of rapid simultaneous acquisition of electrograms throughout a cardiac chamber from an intracavitary multielectrode array (MEA) using the inverse solution method. Despite a prior study validating the reproduction of these "virtual electrograms" in the atria, the system has not achieved widespread use. This is largely explained by several concerns including (1) potential thrombogenicity of the electrode array in the systemic circulation, (2) inaccuracy of virtual electrogram reproduction with increasing distance from the MEA, and (3) limitations of far-field sensing of ventricular activation obscuring atrial activation during the QRS complex.
However, the idea that noncontact mapping can be used for global frequency mapping is an attractive one. A recent paper by Schilling and colleagues did find that the correlation between contact and noncontact atrial electrograms recorded during AF was reasonable (0.7), but fell off significantly with distance. It is important to realize, however, that a perfect reproduction of the contact electrogram morphology may not be required if the frequency content of the signal at each site could be adequately reproduced. Could the noncontact approach be particularly well suited to global frequency mapping of AF?
In this issue of the Journal, Lin and colleagues examine the accuracy of the noncontact MEA for reproducing contact electrogram morphology and DF measurements during human AF. The authors studied 12 patients with predominantly paroxysmal AF. Recordings from the noncontact multielectrode array were made during AF from the right atrium in seven patients and the left atrium in five patients. Contact unipolar electrograms were recorded simultaneously with noncontact "virtual" unipolar electrograms for comparison. Seven-second segments were exported for offline quantitative analysis. The authors localized each contact recording point into one of six right or six left atrial regions, and then grouped the total of 159 sites in both atria for the 12 patients together for analysis. They then compared the 7-second segments for electrogram morphology using the correlation coefficient and frequency spectra using the magnitude squared coherence (MSC) between frequency spectra and the absolute difference in DF between contact and noncontact electrograms. A cutoff of the correlation coefficient >0.7, mean MSC >0.5, and DF difference <0.5 Hz was chosen as representing an "acceptable" comparison between contact and noncontact electrograms.
The mean correlation between contact and noncontact electrogram morphology was 0.75 ± 0.17 and was less accurate with increasing distance from the array (>38 mm), as reported previously. The authors also note that the morphologic correlation was less accurate in the regions of the pulmonary veins, superior vena cava, left atrial appendage, and mitral areas, all locations where the "line of sight" to the MEA might be impaired. One reason for the lower correlation between contact and noncontact signals was a "time-misalignment," also known as a phase shift, between the two signals. When correcting for this time misalignment, the overall correlation improved impressively to 0.85 ± 12. The independent predictors of a poor correlation included a greater time misalignment, a greater distance from the contact catheter to the array center, and a lower peak-to-peak atrial voltage.
When comparing the frequency spectra of the 7-second segments, the mean MSC was >0.5 at most sites and there was no difference between sites closer to or further away from the array center. The dominant frequencies measured for contact and noncontact electrograms were within 0.5 Hz for 94% of mapping sites. The figures show nice examples of sites where the contact and noncontact comparisons were good and other sites where the comparisons were poor.
Are we ready to accept noncontact "virtual" frequency maps as a surrogate for contact frequency maps? There are several limitations and concerns that limit wholehearted acceptance of the noncontact approach for frequency mapping at the present time. First, both the right and left atria were not sampled in each patient, and the density of sampling in each patient was limited. Only approximately 13 points were compared per patient and we have no understanding of how these points were chosen. Were points randomly distributed throughout the atria? Were points deep inside the appendage or pulmonary veins that would be expected to have poor correlation sampled? It is difficult to have complete confidence in the robustness of the comparisons without a higher resolution sampling of atrial points. Second, the exact areas where the MEA has limitations--the pulmonary veins, the left atrial appendage, the mitral valve, and low voltage areas--turn out to be the areas we care most about! These are all sites where CFAEs have been most commonly reported. Inability to interpret frequencies in areas of low voltage may be a significant limitation of noncontact mapping. Additional details of a particular voltage cutoff below which noncontact mapping may have lower accuracy would have been welcome. Other reports have also found difficulties interpreting areas of split potentials in sinus rhythm, again typically areas of anisotropic myocardium that are of prime interest to those interested in localizing CFAE sites. In addition, dilated human atria with long-standing AF will often have significant surface areas >38 mm from the array center. Perhaps concentrating the array on specific regions and performing more detailed mapping in these regions will obtain better results. Finally, the cutoffs chosen by the authors for identifying "adequate" comparisons may have been a bit too liberal. Since correlation is insensitive to electrogram amplitude, a correlation coefficient of 0.75 may be present between two electrogram with grossly different appearance. An MSC of >0.5 Hz was chosen as a reasonable similarity of the frequency spectra; however, it is not clear that this cutoff is adequate. In a prior paper, we found that a left-to-right atrial DF gradient >0.25 Hz in patients with persistent AF predicted long-term freedom from AF. This difference is beyond the 0.5 Hz resolution reported by the authors.
In summary, Lin and colleagues provide several new findings fundamental to the concept of noncontact global frequency mapping. First, virtual unipolar electrograms have a phase lag between simultaneously recorded contact unipolar electrograms and this should be corrected before comparisons are made between contact and noncontact electrogram morphology. Second, while electrogram morphology comparisons fall off with distance from the array center, frequency comparisons do not. This is a significant strength of the noncontact system. Third, the authors have reported that reasonable correlations exist between the frequency spectra calculated by contact and noncontact virtual unipolar electrograms. These findings, for the first time, provide validation for using the MEA array for frequency mapping, keeping in mind the frequency resolution of 0.5 Hz for DF mapping. If these findings are reproduced in more detail in future studies, frequency mapping of AF may lead to a resurgence of noncontact mapping. Hopefully, a better understanding of the mechanism of CFAEs will eventually lead to a better understanding of the mechanism of AF as well as more appropriate targets for AF ablation.