Interpretable machine learning for pre-ablation localization of outflow tract ventricular arrhythmias

Accurate pre-ablation differentiation between left and right outflow tract ventricular arrhythmias (OTVAs; LVOT and RVOT) is essential for selecting appropriate vascular access and treatment strategies. Conventional ECG algorithms show inconsistent accuracy, limited validation, and do not fully leverage information from complete 12-lead ECG data. This study aimed to develop and evaluate a clinically interpretable machine learning (ML) model for pre-ablation localization in idiopathic OTVAs. We analyzed 689 patients who underwent successful catheter ablation between 2012 and 2019, forming a relatively large single-center idiopathic OTVA cohort. Class imbalance was addressed using SMOTE-based resampling. Feature selection was performed using the LASSO method and the Boruta algorithm. Six ML models were trained using 5-fold cross-validation. Performance was evaluated across discrimination, calibration, and clinical utility on an internal hold-out testing set and compared with the best available traditional algorithm, the “Weighted hybrid score”. Shapley Additive Explanations analysis was used to rank feature importance and construct a simplified model. The support vector machine (SVM) model showed the highest observed AUC among the six ML models in the internal hold-out testing set (AUC 0.83). Compared with the “Weighted hybrid score” (AUC 0.76), the full SVM showed a numerically higher AUC, but the difference was not statistically significant (ΔAUC 0.07, 95% CI − 0.01 to 0.15; P = 0.09). A simplified SVM model using ten routinely obtainable features (Ra_V3, Sd_V2, TZI, Sd_V3, QRS i40 _V3, Ra_III, LVESD, TIPN_II, V4Sa_sinus, and age) preserved similar discrimination (AUC 0.81). The simplified model reduced the predictor set from 24 to 10. The calibration slopes were 0.49 and 0.47 for the full and simplified SVM models, respectively, indicating imperfect probability calibration. This study developed clinically interpretable SVM models for pre-ablation differentiation between LVOT and RVOT arrhythmias. Although the models showed numerically higher AUCs than the conventional algorithm, statistical superiority was not demonstrated. External validation in larger multicenter cohorts is warranted.

Authors

Institutions

Publication Details

Journal
BMC Medical Informatics and Decision Making
Published
2026-11-19
DOI
https://doi.org/10.1186/s12911-026-03998-w
Primary Topic
Cardiac Arrhythmias and Treatments
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Interpretable machine learning for pre-ablation localization of outflow tract ventricular arrhythmias

Zhuoqiao He, Xuerui Tan, Pengxiang Ying, Dong Lin et al.
BMC Medical Informatics and Decision Making
Cardiac Arrhythmias and Treatments
article

Interpretable machine learning for pre-ablation localization of outflow tract ventricular arrhythmias

Zhuoqiao He, Xuerui Tan, Pengxiang Ying, Dong Lin, Ming Liu
article en

Abstract

Accurate pre-ablation differentiation between left and right outflow tract ventricular arrhythmias (OTVAs; LVOT and RVOT) is essential for selecting appropriate vascular access and treatment strategies. Conventional ECG algorithms show inconsistent accuracy, limited validation, and do not fully leverage information from complete 12-lead ECG data. This study aimed to develop and evaluate a clinically interpretable machine learning (ML) model for pre-ablation localization in idiopathic OTVAs. We analyzed 689 patients who underwent successful catheter ablation between 2012 and 2019, forming a relatively large single-center idiopathic OTVA cohort. Class imbalance was addressed using SMOTE-based resampling. Feature selection was performed using the LASSO method and the Boruta algorithm. Six ML models were trained using 5-fold cross-validation. Performance was evaluated across discrimination, calibration, and clinical utility on an internal hold-out testing set and compared with the best available traditional algorithm, the “Weighted hybrid score”. Shapley Additive Explanations analysis was used to rank feature importance and construct a simplified model. The support vector machine (SVM) model showed the highest observed AUC among the six ML models in the internal hold-out testing set (AUC 0.83). Compared with the “Weighted hybrid score” (AUC 0.76), the full SVM showed a numerically higher AUC, but the difference was not statistically significant (ΔAUC 0.07, 95% CI − 0.01 to 0.15; P = 0.09). A simplified SVM model using ten routinely obtainable features (Ra_V3, Sd_V2, TZI, Sd_V3, QRS i40 _V3, Ra_III, LVESD, TIPN_II, V4Sa_sinus, and age) preserved similar discrimination (AUC 0.81). The simplified model reduced the predictor set from 24 to 10. The calibration slopes were 0.49 and 0.47 for the full and simplified SVM models, respectively, indicating imperfect probability calibration. This study developed clinically interpretable SVM models for pre-ablation differentiation between LVOT and RVOT arrhythmias. Although the models showed numerically higher AUCs than the conventional algorithm, statistical superiority was not demonstrated. External validation in larger multicenter cohorts is warranted.

BMC Medical Informatics and Decision Making
Edith Cowan University (AU), Wuhan University (CN), First Affiliated Hospital of Shantou University Medical College (CN), Shantou University Medical College (CN), Wuhan Asia Heart Hospital (CN)
Cardiac Arrhythmias and Treatments
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon