ECG-based automated 3D localization of earliest right ventricular activation
L Ding, M Tang, S J ZhouAbstract
Background
The Automated Arrhythmia Origin Localization (AAOL) algorithm predicts the location of earliest ventricular activation in real time using standard ECG and patient-specific electroanatomic mapping (EAM) geometry. We previously extended AAOL to output 3D coordinates (AAOL-3D) in a multicenter cohort; however, right-ventricular (RV) evaluation was limited by a small sample size.
Objective
To validate AAOL-3D for localizing RV pacing sites (earliest activation locations) in a larger RV cohort and to compare its performance with AAOL-Surface.
Methods
RV pacing data from patients with premature ventricular contraction were analyzed. For each site, 120-ms QRS integrals (leads III, V2, V6) and 3D EAM coordinates were extracted. Within each patient, leave-one-site-out cross-validation was performed (test site; remaining sites for training). AAOL-Surface predicted a point on the EAM surface; AAOL-3D predicted full 3D coordinates. Localization error was the Euclidean distance between predicted and true coordinates. Paired tests compared methods overall and by RV subregion (RVOT vs non-RVOT).
Results
A total of 20 patients with 196 sites were enrolled (9.8±1.7 sites/patient). The mean age was 47.1 years and 63.6 % were female. The common comorbidities including hypertension (25.0%), diabetes (9.4%), and heart failure (6.3 %). Overall error was 7.44±2.88 mm for AAOL-3D vs 6.80±3.46 mm for AAOL-Surface (P < 0.001). RVOT (n=57): 7.83±2.92 vs 7.39±3.48 mm (P=0.133). Non-RVOT (n=139): 7.28±2.86 vs 6.55±3.43 mm (P < 0.001). Figure 1 shows statistical results and a pacing-site localization example.
Conclusion
Both methods achieved clinically acceptable accuracy (<10 mm). In RVOT, AAOL-3D was comparable to AAOL-Surface, supporting feasibility of direct 3D coordinate prediction in this anatomically complex, crescent-shaped region where surface projection can introduce geometric artifacts. Further evaluation and optimization across RV subregions are indicated.