Scalable Construction of Anatomically Detailed Biatrial Models with Personalised Electrophysiology from Imaging or Electroanatomic Mapping Data

Caterina Vidal Horrach1, Laura Bevis1, Mahmoud Ehnesh2, Semhar Biniam Misghina1, Ovais Ahmed Jaffery1, Carlos Edgar Lopez Barrera1, Alexander М Zolotarev1, Fuyu Cheng1, Steven Niederer3, Edward Vigmond4, Caroline H Roney1
1Queen Mary University of London, 2Queen Mary University of London, School of Engineering and Materials Science, 3Imperial College London, 4LIRYC - University of Bordeaux

Abstract

Success rates for catheter ablation therapy in persistent atrial fibrillation (AF) remain low, in part due to the complex interplay of anatomical, structural and functional factors that contribute to arrhythmia initiation and maintenance. Clinically, it is difficult to isolate the individual contribution of each factor. Patient-specific biophysical models offer a physiology- and physics-constrained framework to simulate AF inducibility and predict responses to multiple treatment strategies for an individual. In parallel, large-scale in silico trials allow for assessment of treatment efficacy across virtual populations.

To translate these cardiac computational models from the research environment to the clinic, models must be constructed quickly, reproducibly and at scale. We previously introduced atrialmtk, an open-source pipeline for building anatomically detailed atrial models from imaging or electroanatomic mapping (EAM) data. This platform generates atrial meshes with regional labelling, fibre architecture, and transmural wall variation. It is user-friendly, cross-platform compatible, and scalable, as demonstrated by its application to a cohort of 1000 geometries for population-based in silico studies.

Anatomical atrial models may be further personalised to incorporate structural and functional information from imaging or EAM data; however, it is challenging to determine the optimal methodology for this personalisation. Here we utilise atrialmtk to build anatomical models from imaging and EAM data and then compare different methods of model calibration as follows. For LGE-MRI data personalisation, we compare different types of fibrotic remodelling including conduction slowing and replacement fibrosis (percolation). For EAM-based calibration, we compare conduction velocity calibration from local activation time maps collected at different pacing cycle lengths to personalisation approaches based on bipolar voltage or omnipolar voltage maps. Finally, we demonstrate methodologies for calibrating to AF cycle length maps.