Reproducible target transfer from electroanatomic mapping to radiotherapy planning systems for cardiac radioablation - cross-validation for the RAVENTA trial

Abstract

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – EU funding. Main funding source(s): RAVENTA is part of the EU-Horizon-2020 STOPSTORM project. Background Ventricular arrhythmias are a leading cause of morbidity and mortality in patients with structural heart disease. Non-invasive stereotactic single-fraction arrhythmia radioablation (STAR) with 25 Gy has been proposed as a bail-out strategy in patients where catheter-based ablation is either not feasible or has previously failed to control the arrhythmia. Defining the exact radiotherapy target, however, presents unique challenges, and manual transfer of locations from the electroanatomic map (EAM) into the cardiac CT results in a high degree of uncertainty. Different workflows have been developed to register EAM with the cardiac CT and allow for target definition in EAM space with semi-automated transfer of the target structure to CT coordinates. Validation of these approaches is presently lacking. Here we report the initial cross-validation results of two conceptually different software solutions for target transfer quality assurance in STAR. Methods Clinical case data of patients treated with STAR (EAM data, cardiac CT and screenshots of the map with marked target region) were collected retrospectively from centres participating in the RAdiosurgery for VENtricular TAchycardia (RAVENTA) trial. Two blinded investigators independently created a target volume in the cardiac CT based on the marked target region using either a 3D-3D registration approach (3D Slicer / EAMapReader) or a 2D-3D registration (CARDIO-RT), as described previously. The resulting target structures were analysed in a core lab for overlap and similarity by means of surface distance metrics and conformity. Results Three clinical data sets were analysed, one for each mapping system in widespread clinical use. Agreement between structures created with both approaches was good, with DICE coefficients of 0.88, 0.69 and 0.506 and median surface distances of 0 mm, 1.2 mm and 2.2 mm, respectively. The least agreement was seen in a case where no additional structures were mapped besides the LV, whereas additional anchor points such as the ascending aorta seem to reduce registration ambiguity. Conclusion Using two conceptually different approaches for the target transfer from EAM to cardiac CT we were able to show good agreement between both methods. Both techniques thus can be used for quality assurance and analysis in STAR. Further study in a larger sample of patient data sets is warranted to elucidate differences between transfer methods and predictors for improved registration.