ESTRO 2025 - Abstract Book

S2010

Clinical - Urology

ESTRO 2025

2767

Digital Poster Comparison of two auto-segmentation programmes for delineation of elective nodal volumes in prostate cancer patients Sinead Horan 1 , Ciaran Malone 2 , Christina Skourou 2 , Brian O'Neill 1 1 Radiation Oncology, St Luke's Radiation Oncology Network, Dublin, Ireland. 2 Medical physics, St Luke's Radiation Oncology Network, Dublin, Ireland Purpose/Objective: To evaluate and compare the performance and flexibility of two AI-based auto-segmentation systems, MVision and TheraPanacea, in contouring prostate cancer elective nodal regions. Our institutional practice follows the RTOG and NRG-consensus atlases, the latter extending the RTOG-elective nodal clinical target volume superiorly. Both MVision and TheraPanacea are based on the 2009 RTOG guidelines, initiating contouring at the distal common iliac vessels (L5/S1 interspace). We assessed both systems' ability to accommodate scenarios where Radiation Oncologists (RO) use different guidelines — RTOG and NRG — within an institution. Material/Methods: We completed a retrospective analysis of 34 patients treated with elective nodal irradiation, without gross nodal disease, manually delineated by a single experienced RO. First, we compared the institutional NRG volumes (RO delineated), cropped superiorly to align with RTOG landmarks, to both auto-segmentation systems. Volumes were evaluated using the Dice Similarity Coefficient (DSC), Surface-Dice Similarity Coefficient at 2mm (S-DSC 2mm), and 95% Hausdorff Distance (HD95). Second, we assessed each system's ability to produce NRG level contours. MVision allows for post-processing modifications to extend the nodal volumes superiorly to match the NRG atlas, whereas TheraPanacea does not. The auto-segmented contours (with MVision modified for NRG) were compared to the original institutional NRG contours using the same metrics. Results: For the RTOG level comparison, no statistically significant difference was found between MVision and TheraPanacea for the DSC (p=0.17) or HD95 (p=0.91). MVision performed significantly better for the S-DSC 2 mm (p=0.01), indicating fewer deviations greater than 2mm. Both auto-segmentation systems generated clinically acceptable RTOG-contours as effective starting points, likely yielding significant time savings in contouring elective nodes. For the NRG-level comparison, MVision showed higher DSC and S-DSC 2mm scores and a lower HD95 value (p<0.0001). While TheraPanacea was expected not to match the NRG levels due to its basis on the RTOG atlas with no post-processing extension possible, it agreed well within the RTOG boundaries. However, HD95 highlighted the manual extension required to align its contours with the superior boundary of the NRG consensus.