Catheter Ablation -> Ventricular Arrhythmias -> Mapping & Imaging D-PO01 - Featured Poster Session (ID 11) Poster

D-PO01-185 - A Machine Learning Approach For Computer-guided Localization Of The Origin Of Ventricular Tachycardia Using 12-lead Electrocardiograms (ID 956)

Abstract

Background: Machine learning may help localize the site of origin of ventricular tachycardia (VT) using 12-lead electrocardiograms (ECGs). Population-based models, however, suffer from inter-subject anatomical variations within ECG data, while patient-specific models face the challenge of selecting an optimal set of pace-mapping sites to train the model.
Objective: This study presents and validates the first hybrid model that combines population and patient-specific machine learning for rapid “computer-guided pace-mapping”.
Methods: A population-based deep neural network was first trained on pace-mapping data from 38 patients to disentangle inter-subject variations and regionalize the site of VT origin. An on-line patient-specific model, initialized by the population-based prediction, then actively prompts in real time where to pace next and progressively improves the prediction towards the site of VT origin. The integrated model was tuned on one patient, and tested on a separate cohort of eight patients in localizing 1) 193 LV endocardial pacing sites, and 2) nine VTs with clinically determined exit sites.
Results: The presented model achieved a localization error of 4.8 ± 2.8 mm using only 4.9 ± 1.1 pacing sites. As shown in Fig. 1, this gave a higher accuracy with a significantly smaller number of training sites (p<0.0001, paired t-tests) in comparison to models without active guidance, even a “best-scenario” model trained using pacing sites intentionally selected from within a 25-mm radius of the target site (5.7 ± 3.2 mm using 7.3 ± 2.3 sites).
Conclusion: The hybrid machine-learning model has the potential to assist rapid pace-mapping of interventional targets in VT.

Collapse