Learning Domain Shift in Simulated and Clinical Data: Localizing the Origin of Ventricular Activation From 12-Lead Electrocardiograms

Abstract

Building a data-driven model to localize the origin of ventricular activation from 12-lead electrocardiograms (ECG) requires addressing the challenge of large anatomical and physiological variations across individuals. The alternative of a patient-specific model is, however, difficult to implement in clinical practice because training data must be obtained through invasive procedures.Here, we present a novel approach that overcomes this problem of the scarcity of clinical data by transferring the knowledge from a large set of patient-specific simulation data while utilizing domain adaptation to address the discrepancy between simulation and clinical data. The method that we have developed quantifies non-uniformly distributed simulation errors, which are then incorporated into the process of domain adaptation in the context of both classification and regression. This yields a quantitative model that, with the addition of 12-lead ECG data from each patient, provides progressively improved patient-specific localizations of the origin of ventricular activation. We evaluated the performance of the presented method in localizing 75 pacing sites on three in-vivo premature ventricular contraction (PVC) patients. We found that the presented model showed an improvement in localization accuracy relative to a model trained on clinical ECG data alone or a model trained on combined simulation and clinical data without considering domain shift. Further, we demonstrated the ability of the presented model to improve the real-time prediction of the origin of ventricular activation with each added clinical ECG data, progressively guiding the clinician towards the target site.

Reconstruction of three-dimensional biventricular activation based on the 12-lead electrocardiogram via patient-specific modelling

“…Our method is neither the first to require only a 12-lead ECG 9 , 10 nor the first to predict the activation sequence in the entire ventricular volume, 11 but it is the first to achieve both of these feats without compromising the physiological accuracy or strongly limiting the applicability of the approach. Our method combines the best of two worlds: ECG mapping 1 , 2 , 11–13 and patient-specific modelling.…”

“… 18 Personalization of cardiac models from 12-lead ECG has already been investigated by others. 10 , 14–16 Compared with those methods our approach is more general, being capable of identifying an optimal set of EAMs with no restriction on the number of sites, and very competitive in terms of time to solution, hence widening the spectrum of clinical applications.…”

A Spatial Adaptation of the Time Delay Neural Network for Solving ECGI Inverse Problem

“…The first layer is identical to the TDNN expressed by Eq.2. Then, we perform an element-wise multiplication of the first layer output by the first order adjacency matrix Adj (1) . This allows, for each point, to only keep the weights corresponding to its adjacent points and reduces the others to zero.…”

“…More details about these methods can be found in [3] . In recent years, some papers introduced a new vision of the inverse problem by using machine learning algorithms [2,5,8,9,1]. In this paper, we will introduce a new approach to solve the inverse problem using a neural network inspired from the work of Jiaqiu Wang et al [7] for travel time prediction.…”

A Novel Model Based on Spatial and Morphological Domains to Predict the Origin of Premature Ventricular Contraction

“…An alternative strategy is to build patient-specific prediction models. To our knowledge, some studies have used the image-based simulated ECG data to train a customized prediction model for each patient ( Potse et al, 2000 ; Yang et al, 2018 ) and the domain adaptation method has newly been applied to modify the prediction model with clinical data to account for the potential errors in the simulation data ( Alawad and Wang, 2019 ).…”