Author + information
- Received October 22, 2002
- Revision received January 6, 2003
- Accepted February 20, 2003
- Published online June 4, 2003.
- Edward P. Gerstenfeld, MD*,* (, )
- Sanjay Dixit, MD*,
- David J. Callans, MD*,
- Yadavendra Rajawat, MD*,
- Robert Rho, MD* and
- Francis E. Marchlinski, MD*
- ↵*Reprint requests and correspondence:
Dr. Edward P. Gerstenfeld, University of Pennsylvania Medical Center, 9 Founders Pavilion, 3400 Spruce Street, Philadelphia, Pennsylvania 19104, USA.
Objectives The purpose of this study was to objectively quantify the similarity of 12-lead electrocardiogram (ECG) waveforms using two quantitative metrics, the correlation coefficient (CORR) and the mean absolute deviation (MAD).
Background Comparison of the 12-lead ECG morphology between ventricular tachycardia (VT) and a pace-map is frequently performed; however, there are no objective criteria for quantifying the similarity between two waveform morphologies.
Methods During ablation of right ventricular outflow tract (RVOT) VT, 12-lead ECG pace-maps were acquired from three superior septal sites, three superior free wall sites, and before each ablation attempt in 15 patients. The 12-lead ECG waveforms of the clinical tachycardia and pace-maps were compared using both MAD and CORR at each site.
Results The MAD scores were lower (i.e., more closely matched) for septal compared with free wall sites (15.9 ± 5.3% vs. 25.3 ± 10.2%; p < 0.001). Successful ablation sites had a significantly lower MAD score compared with unsuccessful sites (9.5 ± 2.8% vs. 13.3 ± 5.6%; p = 0.01), whereas there was only a trend toward a higher CORR for successful ablation sites (98.2 ± 1.2% vs. 96 ± 4.7%; p = 0.07). A MAD score ≤12% was 93% sensitive and 75% specific for identifying a successful ablation site. There was an inverse correlation between MAD score and distance from the site of VT origin (r = 0.63, p < 0.001).
Conclusions A MAD score >12% between RVOT VT and a pace-map at any site suggests sufficient dissimilarity to dissuade ablation at that site. The MAD score can be used to standardize 12-lead ECG waveform morphology comparisons among different laboratories, and may be useful for guiding ablation of VT.
Pace-mapping is a useful tool for catheter ablation of atrial and ventricular tachycardias (VT), particularly tachycardias of focal origin. When local capture from a paced stimulus results in a 12-lead electrocardiogram (ECG) waveform identical to that during tachycardia, the catheter is assumed to be near the origin of the tachycardia. Although there are some limitations to this technique (spatial resolution, stimulus strength) (1,2), many studies have demonstrated efficacy using pace-mapping to choose ablation target sites (3–7). In addition, pacing during re-entrant tachycardias is also useful to identify critical isthmus sites (8).
Although comparison of the 12-lead ECG morphology between a pace-map and clinical tachycardia is frequently performed, there are few objective criteria for quantifying the similarity between two 12-lead ECG waveform morphologies. Such comparisons are frequently completely subjective or semiquantitative, i.e., a “10/12 lead match”. Discrepancies in ablation results may result in part from subjective differences in the opinion of a pace-map “match” to the clinical tachycardia. Furthermore, criteria for comparing the similarity in 12-lead ECG waveforms from one laboratory to another or for describing such comparisons in the literature are lacking.
We used two waveform comparison metrics, the correlation coefficient (CORR) and the mean absolute deviation (MAD) (9), to objectively quantify the similarity of 12-lead ECG waveforms during VT and pace-mapping. Because right ventricular outflow tract (RVOT) VT ablation is guided chiefly by pace-mapping in our laboratory and others, we used RVOT VT as a model for testing these quantitative comparison measures.
Consecutive patients undergoing radiofrequency (RF) catheter ablation of RVOT VT at the University of Pennsylvania Hospital and Presbyterian Medical Center were included in the study. All patients signed informed consent forms before the procedure. A quadripolar catheter was placed at the right ventricular (RV) apex for arrhythmia induction, and a Navistar catheter (Biosense Webster Inc., Diamond Bar, California) was used for mapping and ablation. An electroanatomic map of the RVOT was acquired in sinus rhythm. Pacing at twice diastolic threshold, 2 ms pulse width, and 400 ms cycle length was performed from three superior septal RVOT sites (1 = posterior, 2 = mid, 3 = anterior) (4), three opposite superior free wall RVOT sites (1 = posterior, 2 = mid, 3 = anterior), and the RV apex (Fig. 1). Arrhythmia was induced with burst atrial or ventricular pacing. If no arrhythmia occurred, isoproterenol up to 10 μg/min was infused and pacing was repeated. If no sustained tachycardia was induced, mapping and ablation was guided by repetitive monomorphic single premature ventricular beats (10). In our laboratory, mapping and ablation of RVOT VT is guided mainly by pace-mapping, finding the best 12-lead pace-map match to the clinical tachycardia, with confirmation of early activation before ablation. Pacing at twice threshold output, 2 ms pulse width, at the tachycardia cycle length is routinely performed in the area of the suspected tachycardia origin and before each ablation attempt. When only isolated premature ventricular complexes (PVCs) were identified, pacing was performed at 400 ms cycle length. When a 12-lead pace-map match was identified, RF lesions were delivered for 60 s each, and temperature was limited to 55° centrigrade with a maximum power of 50 W. An unsuccessful ablation attempt is defined as the delivery of a RF lesion for >30 s at a suspected site of tachycardia origin, after which a tachycardia of the same morphology was still induced. A successful ablation site is defined as the site of RF application after which no tachycardia of the same morphology could be induced with burst pacing and the same amount of isoproterenol before the procedure.
All surface ECG signals were stored on the Cardiolab electrophysiologic recording system (GE Medical Systems, Houston, Texas), with a sampling frequency of 1,000 Hz/channel, frequency band of 0.05 to 100 Hz, and filtered with a 60-Hz notch filter. Signals were digitally stored, extracted to binary data files, and analyzed off-line on a personal computer after the procedure. Signals were viewed and annotated using freely available software (11), and all quantitative waveform analysis was performed using custom written software in the C programming language.
For waveform comparison, both the MAD and the standard CORR were used. Both are quantitative methods of waveform comparison; however, the MAD tends to be more sensitive to differences in waveform amplitude. The normalized MAD was calculated by removing the mean from each waveform and then dividing the absolute value of the difference between the two waveforms by the absolute value of the sum of the area under the curve of the two waveforms. where X and Y are each vectors of length ncontaining the two waveforms to be compared. This is a computationally simple formula that is analogous to what an electrophysiologist performs intuitively in the electrophysiology laboratory, mentally superimposing two 12-lead waveforms to find the combination with the smallest difference between them. The MAD score quantifies waveform similarity and ranges from 0%, for two identical waveforms to 100% for completely different waveforms (Fig. 2).
The CORR was calculated in the usual fashion: where X and Y are vectors of length nrepresenting the two waveforms to be compared. The CORR typically varies from −1 for completely opposite waveform to +1 for identical waveforms (Fig. 2).
Typically, these waveform comparison measures are used to compare only two individual waveforms. To compare multiple waveforms, such as those in a 12-lead ECG, these formulas need to be extended. This was accomplished for the MAD score by summing the numerator and denominator of Equation 1 for all 12 waveforms in the standard ECG.
In the same manner, the CORR was calculated for the 12-lead ECG by summing the numerator and denominator of Equation 2 for all 12 ECG waveforms. In this manner, a single overall number can be calculated for the comparison of two 12-lead ECG waveforms using either MAD or CORR. Comparisons for each lead can also be easily performed if one needs further information about individual lead comparisons.
Pace-maps for each of the six anatomic RVOT pacing sites, unsuccessful ablation attempts, and successful ablation attempts were examined. After examining all 12 leads, the user chose a single 12-lead ECG beat for each paced site. The VT beat was chosen by choosing any two points before and after the desired VT complex without the need to choose the exact onset or offset of the waveform. Sustained VT was used if available, otherwise repetitive monomorphic premature ventricular beats were used for comparison. To avoid operator bias, when there was variation in morphology from one beat to the next for either pace-maps or VT, the second beat of uniform morphology was chosen for comparison. This was done because we often observed that the initial beat of sustained VT may have a subtly different morphology from the remaining beats.
The chosen VT beat was then automatically compared with each of the pace-map beats using customized software. The two beats were aligned by automatically convolving (sliding) the pace-map beat past the VT beat to obtain the lowest overall 12-lead MAD or highest CORR, depending on the desired measure. The 12-lead MAD and correlation were then recorded for the best alignment. This automatic alignment eliminated any operator bias in waveform comparison. Waveforms were viewed by the user to ensure proper alignment (Fig. 3).
Each paced site and all ablation sites (successful and unsuccessful) were tagged using electroanatomic mapping (Fig. 1). The distance from the final successful ablation site to each pace-map site was measured. The distance between each paced site and the successful ablation site was correlated with the MAD for that comparison.
Student ttest was used for comparisons. A simple linear regression was used to compare distance and MAD; p < 0.05 was considered significant.
A total of 15 patients (age 44 ± 11 years, 9 women) were included in the study. All patients had normal left ventricular and RV size and function by transthoracic echocardiography. All underwent successful ablation of a RVOT VT. Patients had sustained RVOT VT (11 patients) or reproducible monomorphic PVCs (4 patients) typically originating from the RV septum (14 patients) or RV free wall (1 patient).
The MAD and correlation scores (mean ± SD) of the clinical VT with pace-maps from the three septal, three free wall, unsuccessful, and successful ablation sites for the 14 patients with septal sites of VT origin are shown in Figures 4A and 4B. Note that MAD scores are lower (i.e., more closely matched) for septal compared with free wall sites (15.9 ± 5.3% vs. 25.3 ± 10.2%; p < 0.001), and the lowest MAD occurred at septal site 2, the most common origin for RVOT VT in this group. Although there was overlap between unsuccessful and successful ablation sites, successful ablation sites had a significantly lower MAD score compared with unsuccessful ablation sites (9.5 ± 2.8% vs. 13.3 ± 5.6%; p = 0.01). The sensitivity, specificity, positive predictive value, and negative predictive value for various MAD scores are shown in Table 1. Using a MAD score cutoff of ≤12%, the sensitivity was 93%, specificity 75%, and negative predictive value 94% for a successful ablation site. A MAD cutoff of <15% had 100% sensitivity and 100% negative predictive value but was not specific.
The CORR was also higher (more closely matched) for septal compared with free wall sites (94.8 ± 3.8% vs. 86.5 ± 11.3%; p < 0.01). There was also a trend toward a significant difference between the mean CORR for unsuccessful and successful ablation sites; however, this difference did not achieve statistical significance (98.2 ± 1.2% vs. 96 ± 4.7%; p = 0.07) (Fig. 4B). A CORR cutoff of >96% was 93% sensitive, but only 26% specific for a successful ablation site.
Because there is often beat-to-beat variation in waveforms during sustained VT, we also compared the chosen VT beat with 10 successive monomorphic VT beats in each patient. The MAD score found a mean beat-to-beat comparison measurement of 4.7 ± 1.7%, and the CORR yielded a mean beat-to-beat comparison of 99.5 ± 0.5% (Figs. 4A and 4B). An example of VT and pace-map 12-lead ECG waveform morphologies and corresponding MAD scores is shown in Figure 5.
We performed a correlation of MAD scores between the clinical VT and each paced site and the distance of the clinical VT origin (successful ablation site) to each pace-map site using electroanatomic mapping. There was a significant correlation between MAD score and distance from the site of VT origin (r = 0.63, p < 0.001) (Fig. 6).
We have described a quantitative measure for comparing 12-lead ECG waveform morphologies between pace-maps and a clinical tachycardia using a single number ranging from 0 to 1. The 12-lead MAD score performed better than the CORR, likely related to better sensitivity to amplitude differences between waveforms. Scores were better for successful compared with unsuccessful ablation sites, suggesting that an automated objective interpretation may have some advantage to human interpretation. The MAD scores were directly related to the distance from the tachycardia site of origin.
Others have described quantitative multi-lead waveform measures for comparing body surface potential maps and the 12-lead ECG. Lux et al. (12)described three quantitative measures to compare two body surface potential maps: the CORR, percent error, and root-mean-square error. These authors acknowledged the limitations of the CORR, which was sensitive to differences in map contour but not amplitude, and used % error and root-mean-square error to describe amplitude differences between maps. They also quantitatively compared body surface potential maps during pacing from multiple left ventricular sites, and found that the CORR decreased with distance (2). Goyal et al. (13)used the CORR and root-mean-square error to compare individual lead morphology differences during pacing at different coupling intervals and cycle lengths. They found that the root-mean-square error was a better discriminator of individual lead waveform differences than the CORR. Throne et al. (14)used the bin area method for template matching of intracardiac electrograms. However, quantitative measures have not been developed that can be used to easily quantify the entire 12-lead ECG and have not been used clinically to assess catheter ablation.
We tested both the MAD and CORR because previous experience with pace-mapping has revealed the importance of matching waveform amplitude in addition to morphology in obtaining a successful ablation site. Although correlation is the much more commonly used statistic, the MAD is much more sensitive to differences in waveform amplitude. Therefore, it is not surprising that the MAD was a better discriminator of successful from unsuccessful ablation sites. The most common human error when turning on RF energy was not appreciating subtle amplitude or precordial lead transition differences between two ECG patterns (Fig. 5). It is important to note that such subtle differences in multiple leads can be reflected in a single quantitative number.
We found that during sustained RVOT VT, there were beat-to-beat differences in 12-lead ECG waveform morphology quantifiable as approximately 5% difference using the MAD score. Some degree of these beat-to-beat changes may be related to intermittent noise, baseline wander, or respiratory artifact. However, these beat-to-beat changes in QRS morphology during sustained VT are frequently observed, and their mechanism requires further investigation.
There are multiple clinical applications of the MAD score. Most obvious would be guiding RF ablation of RVOT VT in centers with less experienced operators. The MAD score is computationally simple and should be easy to incorporate into current electrophysiologic recording systems, allowing feedback to the physician of pace-map comparison before turning on the RF energy. A MAD score of ≤12% was 93% sensitive and 75% specific for a successful ablation site. It is not surprising that the MAD score is more sensitive than specific. Characteristics other than a 12-lead ECG match are necessary for a successful ablation, including catheter-tissue contact, catheter orientation, and tissue heating. Our data suggest that a MAD score >12%, and certainly >15% (100% negative predictive value) suggests sufficient dissimilarity between pace-map and clinical tachycardia to dissuade ablation at that site. The MAD scores ≤12% should be considered an excellent match, and ablation at these sites is warranted if catheter contact and stability are adequate.
Recently, anatomic methods of scar-based tachycardia ablation have been developed that create linear lesions along scar borders based on pace-map similarity to the clinical tachycardia to define the tachycardia exit site. Although studies describing this technique state that there was similarity between pace-maps and VT waveforms at these sites (15,16), there was no existing method for describing the degree of pace-map similarity. The MAD score would allow such a quantitative comparison.
Variation in pace-map waveform morphology with distance, current strength, and rate, as well as beat-to-beat waveform differences during sustained tachycardia described earlier are poorly understood. The MAD score would allow these differences to be analyzed in more detail.
Good signal quality is clearly important in performing these comparisons. Because this study was performed retrospectively, there was no attempt by the clinicians to obtain superior signal quality for the purposes of this study, thus these results should be applicable to any digital signal acquisition system of reasonable quality.
Some laboratories rely more on activation mapping than pace-mapping to ablate RVOT VT. The purpose of this study was not to compare the two but only to use pace-mapping as a model for developing a quantitative metric.
This study was a retrospective one, conducted to validate the MAD metric as a meaningful quantitative indicator of 12-lead ECG waveform morphology. Although all patients in this study underwent eventual successful RF ablation, the MAD score may eventually result in fewer delivered RF lesions and may help to quantitatively describe and guide ablations. Furthermore, the goal of this study was not to demonstrate superiority to conventional ablation techniques, but to validate a quantitative measure of multiple waveform comparison in the electrophysiology laboratory.
The comparison of two 12-lead ECG waveforms can be quantified using the MAD metric. This measure grades 12-lead ECG waveform similarity as a single number ranging from 0 (identical) to 100% (completely different). A MAD score >12% between RVOT VT and a pace-map at any site suggests sufficient dissimilarity to dissuade ablation at that site. The MAD score can be used to standardize 12-lead ECG waveform morphology comparisons among different laboratories, and may be useful for guiding ablation of RVOT VT.
☆ Dr. Gerstenfeld is supported by a Scientist Development Grant from the American Heart Association.
- correlation coefficient
- mean absolute deviation
- premature ventricular complexes
- right ventricular
- right ventricular outflow tract
- ventricular tachycardia
- Received October 22, 2002.
- Revision received January 6, 2003.
- Accepted February 20, 2003.
- American College of Cardiology Foundation
- Kadish A.H.,
- Childs K.,
- Schmaltz S.,
- Morady F.
- Green L.S.,
- Lux R.L.,
- Ershler P.R.,
- Freedman R.A.,
- Marcus F.I.,
- Gear K.
- Coggins D.L.,
- Lee R.J.,
- Sweeney J.,
- et al.
- Klein L.S.,
- Shih H.T.,
- Hackett F.K.,
- et al.
- Stevenson W.G.,
- Friedman P.L.,
- Sager P.T.,
- et al.
- ↵Kenney JF, Keeping ES. Mean absolute deviation. In: Mathematics of Statistics, Pt. 1. 3rd edition. Princeton, NJ: Van Nostrand, 1962:76–7
- Zhu D.W.,
- Maloney J.D.,
- Simmons T.W.,
- et al.
- Goldberger A.L.,
- Amaral L.A.N.,
- Glass L.,
- et al.
- Goyal R.,
- Harvey M.,
- Daoud E.,
- Brinkman K.,
- et al.
- Soejima K.,
- Suzuki M.,
- Maisel W.H.,
- et al.
- Marchlinski F.E.,
- Callans D.J.,
- Gottlieb C.D.,
- Zado E.