Author + information
- Frank Bogun, MD∗ ( and )
- Mohammed Saeed, MD, PhD
- ↵∗Address for correspondence:
Dr. Frank Bogun, Division of Cardiovascular Medicine, University of Michigan, Cardiovascular Center, SPC 5853, 1500 East Medical Center Drive, Ann Arbor, Michigan 48109-5853.
The 12-lead electrocardiogram can be helpful in localizing the exit site in scar-related ventricular tachycardia (VT). Pioneering work by Waxman and Josephson (1) and later Miller et al. (2) introduced the concept of dividing the endocardium into several regions (typically 10 to 12) and studied whether particular 12-lead electrocardiographic patterns during VT could be reliably mapped to specific regions. Several algorithms have been proposed using mapping schemes to guide electrophysiologists during catheter ablation procedures (1–3). In particular, in cases in which activation mapping may be limited by hemodynamic instability during VT, the use of pace mapping to identify VT exit sites has been demonstrated to be of critical importance (4). Yokokawa et al. (5) proposed a semiautomated algorithm based on a supervised machine-learning technique (support-vector machines) trained on a set of patients and then validated on a different set of patients with approximately 70% accuracy in localizing the VT sites of origin to 1 of 10 LV endocardial regions. Such advances in computing and machine learning, along with the availability of automatically acquired pace-mapping data, open exciting possibilities for developing real-time “intelligent” decision support systems to guide VT ablation procedures and potentially shorten procedure times and improve outcomes.
In this issue of JACC: Clinical Electrophysiology, Sapp et al. (6) report on the utility of using the 12-lead electrocardiogram to determine the exit site in scar-related VT by using a novel computerized algorithm. Conventional 12-lead electrocardiographic data (from 38 patients), along with 120-point body-surface potential mapping data (in 18 patients of the 38-patient cohort) acquired during endocardial pacing were pooled. This resulted in surface electrocardiographic data from a set of 1,012 known pacing sites, which were then labeled as originating from 1 of 16 left ventricular (LV) endocardial areas. Two complementary localization techniques are presented. The first technique involves the use of a template matching algorithm to map surface electrocardiographic data to 1 of the 16 LV segments and further to 1 of 238 triangles with a regression algorithm that specifies the x, y, and z coordinates of the candidate site in 3-dimensional space on the basis of a 3-dimensional generic model of the LV endocardium. Localization error was reported to be a mean of 12 mm using a bootstrapping method. In another 20 patients, pace mapping was performed from a “limited” number of sites within a region of interest (guided by the segment localization approach) to train a patient-specific regression model that allowed (on the basis of an “optimal” set of 3 surface electrocardiographic leads) improved localization of pace-mapping sites in the LV endocardium. The reported localization error using patient-specific mapping coefficients was reported to be as low as 5 mm in select patients but varied from 2.6 to 20.7 mm.
The segment localization approach appears to be similar to the work of Yokokawa et al. (5), who first demonstrated the feasibility of using a computational algorithm for localizing VT exit sites using pace mapping–generated QRS configurations in post-infarct patients as training data. Unfortunately, Sapp et al. (6) do not provide the individual error analysis for the segment localization approach and instead provide only summary statistics (mean error localization of 12 mm) and report 75.9% accuracy over all patients when the correct segment ranks among the top 3 of the candidate segments (Online Table 1 in Sapp et al. ). The size of this area of interest is not indicated but is likely in the 30- to 50-cm2 range (5). Unfortunately, the accuracy of the top template match is not given. Furthermore, the investigators do not describe the distribution of the pace-mapping sites and their relationships to the distribution of known scar, as this may significantly affect the configuration of the acquired 12-lead electrocardiograms and hence the accuracy of their template matching algorithm.
Sapp et al. (6) report that the “personalized” regression coefficients localize endocardial points from surface electrocardiographic data with surprising accuracy in some patients. No clear-cut validation data are provided for all patients. Figure 3, for example, demonstrates that from generic patient pooled data (from 38 patients), a template match with a correlation coefficient (CC) of 0.99 could be generated for an individual patient. It is unclear how one is to interpret such a high template CC. In practice, it is quite difficult to obtain a CC of 0.99 in a given patient when pace mapping is performed to identify VT exit sites (and pace-mapping configurations are compared with VT configurations in the same patient). There is a wide variance in the localization accuracy. For example, Patient #25 had a localization mean error of 20.7 mm. This is much higher than the reported localization error using the population-based template matching approach.
The investigators describe a work flow for how this method can be used for the practicing electrophysiologist: VT is induced, and the computerized algorithm will rank the best 3 LV endocardial segments matching best with the induced VT configuration. This is displayed as a CC. Then the operator proceeds to place the catheter in the proximity of these segments and performs pace mapping at as many sites as possible (minimum of 5 sites), and local regression is performed in real time to provide the candidate exit site. Eventually the operator will look at the actual matches of pace mapping rather than predictions of an algorithm that may or may not be correct.
The work of Sapp et al. (6) is ambitious in that they attempted to localize VT exit sites to actual continuous 3-dimensional points rather than only discrete segments of the LV endocardium, as others have done. The “inverse problem” of electrocardiography has been pursued by many others, and typically required high-density body-surface potential mapping electrocardiographic data, in addition to information regarding heart-torso geometry and cardiac morphology from advanced cardiac morphometric imaging analysis (computed tomography and/or magnetic resonance imaging) (7). It is surprising that a localization error as low as 5 mm using only 3 electrocardiographic leads can be achieved in the absence of such 3-dimensional geometric data. The investigators did not describe the body habitus, cardiac volumes, scar location and orientation of their patient cohort. It would be helpful to understand the reliability and robustness of computer algorithms in relation to individual patient characteristics and scar distribution patterns. The work presented by Sapp et al. (6) can be seen as a preliminary proof-of-concept study that certainly warrants further investigation.
However, several caveats need to be considered: First, it is important to emphasize that the localization error was based on predicting the site of pacing rather than true VT exit sites. Validation data detailing the localization accuracy of target sites for given VTs is reported for only 2 patients. The accuracy of the method used is reported as the localization error between the estimated 3-dimensional coordinate and the actual point where pacing was done. There were 20 patients used for computing the continuous coordinates, in which each patient had at least 10 pacing sites within a confined neighborhood. This clustering effect of pace mapping sites (1,012 points distributed over 238 segments) invariably leads to sparse sampling in some segments (some triangles may have 0 to 2 corresponding pace maps), with resulting under- and/or oversampled regions of the generic LV endocardial model. Furthermore, scar-related arrhythmias may likely lead to additional variability in spatial resolution of pace-mapping sites.
Second, it would be of more interest to evaluate how this method performed for localizing VT exit sites in all patients rather than calculating localization predictions of known endocardial mapping points. Not all VTs have exit sites in the endocardium. It is not clear how this algorithm performs for epicardial or intramural VTs. VTs originating from the septum with intramural or right ventricular exit sites may be incorrectly classified, because right ventricular data were not included in the training data.
In summary, the computer algorithm described by Sapp et al. (6) helps regionalize VT exit sites in patients with structural heart disease by using training data from prior patients. Accuracy can be improved if more pace mapping is performed in the region of interest. The value of this method for VT ablation procedures will need to be determined prospectively.
↵∗ Editorials published in JACC: Clinical Electrophysiology reflect the views of the authors and do not necessarily represent the views of JACC: Clinical Electrophysiology or the American College of Cardiology.
Both authors have reported that they have no relationships relevant to the contents of this paper to disclose.
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.
- 2017 American College of Cardiology Foundation
- Miller J.M.,
- Marchlinski F.E.,
- Buxton A.E.,
- Josephson M.E.
- Marchlinski F.E.,
- Callans D.J.,
- Gottlieb C.D.,
- Zado E.
- Sapp J.L.,
- Bar-Tal M.,
- Howes A.J.,
- et al.