Simulated activation sequences fitted to ECG improve model accuracy in porcine models of chronic infarct

Caroline Mendonca Costa1, Phillip Gemmel1, Aurel Neic2, Gernot Plank2, Mark O'Neill1, John Whitaker1, Martin Bishop1
1King's College London, 2Medical University of Graz


Background: Patient-specific models of cardiac anatomy and electro-physiology (EP) built from clinical data are increasingly being recognized as a valuable tool to predict arrhythmia risk and ablation targets in infarct pa-tients. Clinical images used to build patient-specific models are typically ac-quired at low out-of-plane spatial resolution (~8-12mm) and EP data for parameterization is often not available. We investigate the impact of image resolution and EP parameterization on simulated activation sequences.

Experimental data: High-resolution (1mm isotropic) LGE-MRI data and EP data (high density electro-anatomical mapping (EAM) and 12-lead ECG data) recorded during right-ventricular (RV) pacing were obtained from 7 pigs with chronic (6 weeks) infarcts.

Methods: Ventricular anatomy, scar and border zone (BZ) morphology were segmented from the high-resolution LGE-MRI. The segmentations were downsampled to out-of-plane resolutions of 4mm (medium) and 10mm (low). Computational models were created for each segmentation. The im-pact on scar morphology was evaluated using Dice Scores relative to the high resolution model. RV pacing activation was also simulated 1) using literature-based conduction velocities; and 2) with conduction velocities tuned to fit the total activation time to the QRS duration on recorded 12-lead ECG. Sim-ulated activation sequences were compared to local activation times (LAT) from EAM data.

Results: Dice scores for scar and BZ elements, and for scar transmurality were 0.9, 0.42, and 0.8 for medium, and 0.3, 0.47, and 0.4 for low resolution models. The mean RMS errors relative to EAM measured LAT for activation sequences with fitted conduction velocities were 24.6±6.5, 25.5±5.3, 23.3±6.4, for high, medium, and low resolution, respectively, and 37.9±14.6, 37.1±13.8, 30.6±7.2, for activation sequences with literature conduction ve-locities.

Conclusion: The lack of EP parameterization leads to errors in activation sequences, whereas the use of low resolution images has little impact. How-ever, low resolution affects scar morphology, which may affect simulated arrhythmia circuits.