A Generalizable Pipeline for Integrating 3D Dixon Late Gadolinium Enhancement and Electroanatomic Mapping to Monitor Ablation-Induced Atrial Remodeling

Background: CMR enables non-invasive quantitative characterization of cardiac function, tissue and morphology in patients with arrhythmic disorders such as atrial fibrillation (AF). Integration of 3D CMR data such as 3D Dixon Late Gadolinium Enhancement (LGE) with complementary 3D modalities including electroanatomic mapping (EAM) or CT is challenging. Often it requires specialized proprietary pipelines limiting transfer to other disease entities and hindering broader adoption in (pre)clinical practice. In this work, we sought to develop and evaluate a versatile, generalizable pipeline for multimodal integration of 3D CMR data with complementary modalities, and to demonstrate its utility with the example of integrating left atrial (LA) EAM with isotropic 3D Dixon LGE to monitor ablation-induced remodeling.

Methods: The LA was segmented manually from 3D LGE images and blood-pool-normalized image intensities were projected onto a 3D surface reconstruction for image intensity ratio (IIR) analysis(1). Areas with an IIR >1.34 were considered fibrotic. Registration of both surfaces (LGE and corresponding EAM) was performed in two steps: 1. Global rigid registration using the iterative closest point algorithm and 2. Local as-rigid-as-possible surface deformation (2) using paired landmarks. Registration performance was evaluated for periablational regions and the posterior LA wall using mean absolute surface distance (MSD) and the 95% symmetric Hausdorff Distance (HD95).

To demonstrate its clinical utility, we applied the pipeline to quantify ablation-induced changes in LGE amount in the periablational region (within ±7.5mm surface distance from ablation sites (3)) and non-ablated LA region in sixteen AF-patients who underwent 1.5T CMR with high-resolution (1.3mm³) 3D LGE based on a GRE Dixon research sequence(4) and high resolution EAM (1086±558 mapping sites). CMR was performed immediately before pulmonary vein isolation and repeated 6 months after.

Results: The pipeline achieved a median [interquartile range] HD95 of 2.80mm [1.70–3.42] for the periablational area and 0.98mm [0.76–1.21] for the posterior wall. The MSD was 0.62mm [0.42–0.75] for the periablational area and 0.28mm [0.22–0.37] for the posterior wall.

Non-rigid surface deformation significantly improved initial rigid registration (Figure 2) for both periablational and posterior LA surface wall. Applying the pipeline to monitor ablation induced changes in LA-LGE, an increase in fibrotic tissue amount of 7.74% [2.08-10.07], p< 0.001 in the periablational area was observed, with no significant change in the remaining part of the LA (0.19% [-0.97-1.26], p=0.375) (Figure 3).

Conclusion: The pipeline enables robust integration of 3D electroanatomic mapping with 3D Dixon LGE reconstructions. While here its utility for quantifying ablation-induced remodeling in AF patients was demonstrated, the general framework is broadly applicable to other cardiac chambers, disease entities, and multimodal imaging datasets.

Figure 1: (A) Following manual segmentation of the left atrial (LA) wall on 3D Dixon Late Gadolinium Enhancement (LGE) images, normalized image intensities (IIR) were projected onto the surface reconstruction for quantification of fibrosis amount. Fibrosis (defined by an IIR>1.34) is denoted in red. Electroanatomic mapping (EAM) data and ablation sites (red dots) were imported from the EAM workstation into the pipeline. (B) EAM and LGE meshes were first aligned by rigid registration, resulting in a global alignment of the two surfaces with local gaps due to deformation of the LA between LGE and EAM measurements (purple arrows). (C) Non-rigid surface deformation resulted in an improvement in the local alignment of the two meshes effectively compensating for this inter-measurement deformation.
Figure 2: Symmetric 95% Hausdorff Distance (HD95) and Mean Surface Distance (MSD) for the rigid and non-rigid surface registration steps of the pipeline. The non-rigid surface deformation led to a significant reduction in HD95 and MSD in both the periablational area and the posterior left atrial wall when compared to global rigid registration alone (all p<0.001 indicated by ***). Boxplots represent median (solid line inside the box), interquartile range (IQR, box) and Q1 - 1.5*IQR or Q3 + 1.5*IQR (whiskers).
Figure 3: Clinical application of the pipeline: (A) Spatial correlation of ablation sites (dark red points) from electroanatomic mapping (EAM) with isotropic Late Gadolinium Enhancement (LGE) for regionalized quantitative assessment of ablation-induced fibrosis (periablational area displayed solid, remote left atrial (LA) wall displayed opaque, image intensity ratios >1.34 marked in red, image intensity ratios >1.2 marked in yellow). Note that for quantification of scar amount in the remote (non-ablated) LA wall, pulmonary veins, LA appendage and mitral valve were excluded from the quantification. (B) At 6 months post pulmonary vein isolation (PVI) an increase in fibrotic substrate amount was observed predominantly in the periablational area. No significant change (p=0.375 Wilcoxon signed-rank test) was observed in the remote LA wall. Boxplots represent median (solid line inside the box), interquartile range (IQR, box) and Q1 - 1.5*IQR or Q3 + 1.5*IQR (whiskers).