Automated workflow to Integrate Electroanatomic Maps into Patient-Specific Bi-atrial Models For Personalized AF Treatment

Caterina Vidal Horrach1, Sam Coveney2, Ovais Ahmed Jaffery1, Mahmoud Ehnesh3, Steven Niederer4, Shohreh Honarbakhsh1, Sanjiv Narayan5, Caroline H Roney1
1Queen Mary University of London, 2University of Leeds, 3Queen Mary University of London, School of Engineering and Materials Science, 4Imperial College London, 5Stanford University

Abstract

Introduction: Personalized computational models can improve atrial fibrillation (AF) treatment but require invasive imaging or sinus rhythm electroanatomic mapping (EAM) data which is difficult to apply in real-time during a case. This study aims to create an automated workflow for building patient-specific models from AF EAM data, combining anatomical, structural, and functional details.

Methods: A pipeline integrating bi-atrial anatomical models, AF recordings, and calibrated simulations was used to study AF dynamics with the Atrial Modelling Toolkit. AF cycle length (CL) measured from bi-atrial basket recordings was mapped onto atrial anatomy using Gaussian Process Manifold Interpolation (GPMI) and used in model calibration. AF was simulated using baseline parameters.

Results: Basket electrogram recordings in the LA ranged from 64-256 (192 ± 78.38) and in the RA from 64-128 (115.2 ± 28.62). AFCL recordings for raw clinical data in the LA ranged from 111-220ms (173.30ms ± 13.73ms) and in the RA from 123-220ms (174.30ms ± 15.10ms) across all cases. AFCL recordings for simulated AF in the LA ranged from 71-290ms (198.13ms ± 8.91ms) and in the RA from 92.5-370ms (225.52ms ± 10.90ms) across all cases. AF simulations with baseline parameters showed complex and chaotic activation patterns, highlighting the effects of AF maintenance.

Conclusion: Personalized computer models can be created rapidly from this pipeline to create simulations that may provide insights into AF mechanisms and guide individual therapy.