Comparative influence of imaging and electrophysiology data on patient-specific atrial model calibration and prediction of ablation targets in atrial fibrillation

Background Patient-specific left atrial models are increasingly used to explore Atrial Fibrillation (AF) mechanisms and optimise ablation strategies, yet accurately representing individual electrophysiological properties remains challenging due to low-density AF mapping and complex AF dynamics Objectives To 1 compare atrial model outcomes calibrated using late gadolinium enhancement manetic resonance imaging (LGE-MRI), Peak-to-peak voltage electrograms (EGM) during AF and coronary sinus pacing (CSP), and conduction velocity CV from EGM, 2 evaluate the effect of EGM interpolation methods on model predictions, 3 assess the influence of calibration data on predicting patient-specific ablation targets and response to pulmonary vein isolation PVI ablation Methods Patient-specific left atrial models were constructed using LGE-MRI and EGM data from nine AF patients undergoing PVI ablation. Model calibration focused on conductivity changes without ionic modifications, incorporating LGE-MRI, peak-to-peak voltage bipolar/omnipolar EGM during AF and CSP, and CV of bipolar/omnipolar EGM. EGM values were assigned to each simulation mesh element using Radial Basis Function RBF and Gaussian Process Manifold Interpolation GPMI methods. Fibrosis in LGE-MRI was modelled as regions of conduction slowing, with velocities determined by image intensity ratio (IIR) at four thresholds Fig 1, 2D. Peak-to-peak voltage EGM data defined conduction slowing regions using low, median, and high voltage thresholds in both rhythms Fig 1, 2D. For CV EGM, each patient’s CV map was divided into ten regions, with conductivity values adjusted based on tissue patch simulation experiments. Phase singularity (PS) analysis identified potential driver sites in 15-second AF simulations, though PS may represent passive phenomena rather than true drivers. For all calibrated models, PVI was simulated as a non-conductive region surrounding the PVs Results AF was sustained in 77.78% of LGE-calibrated models, 100% with AF EGMs, and 94.45% with CSP EGM and CV-derived CSP EGM. Interpolation methods had no impact on ablation target prediction or AF sustainability in EGM-based models Fig 1: RBF, 2: GPMI. Among the 117 analysed models, pre-PVI ablation targets were consistently detected in all models (PS hotspots, mean ± SD: 7.10±3.35). Following PVI ablation, a significant reduction in PS hotspots was observed (mean ± SD: 3.94±3.35, P >0.05). Figures 1C,F and 2C,F demonstrate a reduction in PS hotspots outside PVI sites. This reduction may be influenced by the decrease in atrial area rather than the elimination of true PV triggers Conclusion This study developed a framework for calibrating patient-specific atrial models using imaging and electrophysiology data. Interpolation methods had no impact on ablation targets or AF sustainability. In silico, PVI reduces PS hotspots outside PVs, further investigation of different ablation strategies and long term clinical validation of PVI are needed