QF3-01 - Integration of electro-anatomical and cardiac MRI imaging data in patients with Implantable Cardioverter-Defibrillators undergoing ventricular tachycardia ablation.

Background: Late Gadolinium enhancement (LGE) cardiovascular MRI (CMR) identifies myocardial scar and may facilitate procedural planning in patients undergoing ventricular tachycardia (VT) ablation. Integration of electroanatomical maps (EAM) and CMR-derived scar was previously shown to potentially improve procedural outcomes (1,2). However, the majority of patients undergoing VT ablation have implantable cardioverter defibrillators (ICD) in situ, potentially limiting the accuracy of scar detection due to artefact. Previously, LGE-CMR and EAM datasets have been co-registered in ICD patients (3,4) by prospective fitting of LV models, however the optimal method for independent co-registration and the optimal LGE-CMR voxel intensity threshold to detect electrical myocardial scarring remains unclear. In this study, we evaluated the accuracy of LGE-CMR and EAM data integration using both left ventricular and extra-cardiac landmarks.

Methods: We retrospectively analyzed data from 10 patients with ICD who underwent LGE-CMR prior to undergoing VT ablation (8 subendocardial, 2 epicardial). A wideband motion corrected phase sensitive inversion recovery sequence for LGE imaging was deployed to minimize device artifact (5). Epicardial and endocardial borders were segmented on a short axis stack (4mm slice thickness, contiguous slices), generating a 3D pixel signal intensity map using custom software (ADAS-VT, Galgo Medical, Spain) (Figure 1). The LGE-CMR map was co-registered with the EAM generated from the electroanatomical navigation system (CARTO, Biosense Webster, CA), using LV and aortic landmarks as fiducial markers (Figure 2). For the EAM map, scar was defined as bipolar voltage (< 1.5mV) whereas for the LGE-CMR map an ROC curve was used in each case to derive the optimal threshold for defining scar (Figure 3).

Results: Segmentation and co-registration of EAM and CMR pixel intensity maps was feasible in all subjects. Comparison was performed on a point-by-point basis, with a median of 220 (99-279) points paired per patient. The median accuracy of LGE-CMR to identify EAM-defined scar was 74.2% (62.2%-83.3%) with a median sensitivity 74.0% (63.3%-80.0%), specificity 79.5% (70.6%-89.5%) and AUC 0.78 (0.61-0.87). The median pixel-intensity threshold for defining scar for LGE-CMR was 39.7% (35% - 41.4%) of the maximum pixel intensity.

Conclusion: Using both LV and extra-cardiac landmarks, co-registration between LGE-CMR and invasive EAM is feasible despite the image artifact caused by the ICD. The highest scar point matching between the two modalities was obtained when a case-specific threshold was used for the LGE-CMR pixel-intensity map. Limitations to accurate independent co-registration of CMR and EAM-derived scar maps remain, although incorporation of aortic anatomy is helpful.