Improving localization accuracy for non-invasive automated early left ventricular origin localization approach

Abstract

Background: We previously developed a non-invasive approach to localize the site of early left ventricular activation origin in real time using 12-lead ECG, and to project the predicted site onto a generic LV endocardial surface using the smallest angle between two vectors algorithm (SA). Objectives: To improve the localization accuracy of the non-invasive approach by utilizing the K-nearest neighbors algorithm (KNN) to reduce projection errors. Methods: Two datasets were used. Dataset #1 had 1012 LV endocardial pacing sites with known coordinates on the generic LV surface and corresponding ECGs, while dataset #2 included 25 clinically-identified VT exit sites and corresponding ECGs. The non-invasive approach used "population" regression coefficients to predict the target coordinates of a pacing site or VT exit site from the initial 120-m QRS integrals of the pacing site/VT ECG. The predicted site coordinates were then projected onto the generic LV surface using either the KNN or SA projection algorithm. Results: The non-invasive approach using the KNN had a significantly lower mean localization error than the SA in both dataset #1 (9.4 vs. 12.5 mm, p < 0.05) and dataset #2 (7.2 vs. 9.5 mm, p < 0.05). The bootstrap method with 1,000 trials confirmed that using KNN had significantly higher predictive accuracy than using the SA in the bootstrap assessment with the left-out sample (p < 0.05). Conclusion: The KNN significantly reduces the projection error and improves the localization accuracy of the non-invasive approach, which shows promise as a tool to identify the site of origin of ventricular arrhythmia in non-invasive clinical modalities.

Keywords: ECG; k-nearest neighbors (KNN) algorithm; pace-mapping; radiofrequency ablation; ventricular tachycardia.

FIGURE 1

The K-nearest neighbors (KNN) algorithm with Euclidean distance measurement was used to project a predicted pacing/VT-exit site marked by the red ball onto one of the 238-triangle centers of the generic LV endocardial mesh surface. The example illustrates that a predicted pacing/VT-exit site marked by the red ball was projected onto the No. 196 triangle center by finding the shortest distance from all of the 238 Euclidean distances calculated by the predicted pacing/VT-exit site and all of the 238-triangle centers of the generic LV endocardial mesh surface.

FIGURE 2

Left Panel, Comparison based on the test set (n = 202) for the error measured as geodesic distance using the two projection algorithms (KNN, SA). Mean values are shown numerically. Right Panel: Box plot of localization error for using the two projection algorithms (KNN, SA). Plots represent data for mean localization error in terms of geodesic distance on the generic LV-endocardial surface for the left-out sample (n = 1,012/e≃371). Boxes represent interquartile ranges; a line inside the box marks the median, “whiskers” above and below the box indicate range, ∗∗ represent outliers.

FIGURE 3

Left panel: the distributions for the non-invasive automated approach using the K-nearest neighbors (KNN) projection algorithm of the bootstrap left-out-sample errors, based on localization errors on the LV endocardial surface for 1,000 bootstrap trials. Right panel: the distributions for the non-invasive automated approach using the smallest angle between two vectors algorithm of the bootstrap left-out-sample errors, based on localization errors on the LV endocardial surface for 1,000 bootstrap trials. Error was measured as geodesic distance (approximated by the arc length) between the centroid of the projected triangle and the centroid of the pacing-site triangle on the 238-triangle generic LV endocardial mesh surface.

FIGURE 4

Box plot of localization error of the 25 VT exit sites for using the non-invasive automated approach based the two projection algorithms. Plots represent data for mean localization error in terms of Euclidean distance.

FIGURE 5

Localization of a ventricular tachycardia (VT) exit by the non-invasive automated localization using the two projection algorithms. (A), The recorded 12-lead ECG of an induced monomorphic VT during the procedure. The onset of one VT beat was automatically detected (Kemmelings et al., 1994); the user can edit the onset of the 120 m window (rectangle box) if correction is necessary. (B), Bull’s eye icon that indicates the estimated VT-exit locations using the non-invasive automated approach based on the smallest angle between two vectors (SA algorithm). The red ball indicates the VT reference site on the 238-triangle generic LV endocardial mesh surface, which was registered manually from an endocardial electroanatomic mapping map (panel D). Localization error of the VT exit site is 6.7 mm between the bull’s eye icon and the red ball. The large number within each segment is the correlation coefficient (%) for match by the 12-lead ECG VT pattern with population based 12-lead ECG templates; the small number identifies the segment. (C), Bull’s eye icon that indicates the estimated VT-exit locations using the non-invasive automated approach based on the K-nearest neighbors algorithm (KNN). The red ball registered manually from an endocardial electroanatomic mapping map (panel D) indicates the VT reference site on the 238-triangle generic LV endocardial mesh surface. Localization error of the VT exit site is 5.8 mm between the bull’s eye icon and the red ball. The large number within each segment is the correlation coefficient (%) for match by the 12-lead ECG VT pattern with population based 12-lead ECG templates; the small number identifies the segment. (D), an endocardial electroanatomic substrate map, with areas featuring bipolar signal amplitude ≥1.50 mV in purple, and the site of VT exit (identified by contact mapping) depicted by the yellow arrow, yellow starand gold ball.