A Workflow for Probabilistic Calibration of Models of Left Atrial Electrophysiology

Abstract

Atrial fibrillation is an increasingly common condition. Computational models that describe left atrial electrophysiology have the potential to be used to guide interventions such as catheter ablation. Calibration of these models to faithfully represent left atrial structure and function in a particular patient is challenging because electrophysiology observations obtained in the clinical setting are typically sparse and noisy, and can be difficult to register to a mesh obtained from imaging.Probabilistic approaches show promise as a way to obtain personalised models while taking account of noise, sparseness, and uncertainty. We have developed a workflow in which parameter fields are represented as Gaussian processes, and the posterior distribution is inferred using MCMC. Our workflow has been tested using synthetic data, generated from simulations where the spatial variation in model parameters is known, and we have shown that both features and parameters can be recovered from simulated sparse measurements.

A Review of Personalised Cardiac Computational Modelling Using Electroanatomical Mapping Data

“…Workflows have been proposed to recover parameters from sparse EAM data of atria in which Gaussian processes (GPs) are used to represent parameter fields, and the posterior distribution of CV is inferred using Markov chain Monte Carlo. [ 50 , 106 , 124 ] The starting point of the workflow is a mesh representing the left atrium, and bipolar EGMs are recorded at different locations within the left atrium and at different pacing cycle lengths. From these observations, LATs are estimated at the electrode locations using a modified centre of mass method.…”

“…Having obtained a probabilistic interpolation of LAT, the third step is to calculate the inverse of the gradient in LAT to obtain an uncertain estimate of CV at each mesh vertex. The last step is to use these estimates to calibrate the EP model based on the workflow suggested by Coveney et al[ 124 ] For recovering parameters fields related to tissue conductivity and restitution properties and to calibrated EP models, an extension has been suggested using latent GPs by Coveney et al[ 106 ]…”