Excellent interobserver agreement and steep learning curve for target volume delineation for stereotactic arrhythmia radioablation using a commercial software

AIMS: Stereotactic arrhythmia radioablation (STAR) has emerged as bail-out treatment for ventricular tachycardia (VT). Accurate, reproducible, and easy-to-use data transfer from electroanatomical mapping (EAM) systems to radiotherapy planning CT is desirable. We aim to evaluate interobserver variability, ease of use, and learning curve for EAM based target volume (CardTV-EPinv) creation and transfer using available software packages. METHODS AND RESULTS: In patients considered for STAR, CardTV-EPinv were created using ADAS and Slicer3D for workflow comparison. Four CardTV-EPinv (clinically targeted volume and three mock targets) were created by an experienced operator and a 2nd-year medical student, based on endocardial EAM tags indicating VT substrate location. CardTV-EPinv sizes, Hausdorff distances (HDs), and workflow duration were measured to assess interobserver variability and learning curve. Agreement between CardTV-EPinv was high using ADAS and Slicer3D workflows (HD 3.64 mm [2.7-4.5]). ADAS workflow was faster and more robust (ADAS 26 min [24-29] vs. Slicer3D 65 min [61-70], P < 0.001; system crashes: ADAS 0 vs. Slicer3D 7). In 20 patients (80% non-ischaemic cardiomyopathy, LVEF 35 ± 14%), 80 CardTV-EPinv were created using ADAS. CardTV-EPinv size was similar for both observers (11.8 mL [10.1-13.7] vs. 10.7 mL [9.6-11.8], P = 0.17), with high interobserver agreement (HD 1.68 mm [1.45-1.96]; 95th percentile HD < 4.8 mm [3.5-5.7]). Linear regression showed a steep learning curve for the student (P = 0.01). CONCLUSION: CardTV-EPinv creation showed excellent interobserver agreement and was faster and more robust using ADAS than 3D slicer. The steep learning curve appears clinically relevant given the limited use of STAR even in high-volume VT ablation centres.