Clinical Equivalency of Alternate Head-And-Neck OAR Delineations
MNH Rashad1*, F Badry1, V Leandro Alves1, H Nourzadeh2, W Choi2, J Siebers1, (1) University of Virginia, Charlottesville, VA, (2) Thomas Jefferson University, Philadelphia, PA,
Presentations
PO-GePV-M-211 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: To assess the clinical equivalence of alternative head-and-neck OAR delineations via an alternative truth plan and evaluation assessment.
Methods: For 14 head-and-neck datasets, alternative dose distributions for 9-beam IMRT plans were created using the initial PTV delineations with 3 alternative OAR delineations: manual (MD) and two commercial AI-based auto-delineation algorithms (AI1, AI2). OAR dose-volume metrics and NTCPs were evaluated using the OAR used for planning, as well as with each alternative for each plan before and after probabilistic evaluations which account for +/-XX mm inherent systematic and random patient setup variations. Delineations are deemed clinically equivalent when the differences between plan OAR metrics and those evaluated with the alternative were within 2% of the prescription dose.
Results: For SpinalCord D01% 79% of AI delineations were equivalent to MD for static plan. In probabilistic evaluations 79% were equal for D_01_95% and 82% for D_01_50%. SpinalCord NTCP was equivalent in 100% of AI delineations for static plan and probabilistic evaluations with NTCP_95% and NTCP_mean. For Parotid_Left D50%, 67% of AI delineations are equivalent to MD for static plan and probabilistic evaluation with D_50_95% and D_50_50%. For NTCP, 83% are equivalent for static plan and 79% and 83% respectively for probabilistic evaluation with NTCP_95% and NTCP_mean. For Parotid_Right D50%, 40% are equivalent to MD for static plan and 44% and 40% respectively for probabilistic evaluations with D_50_95% and D_50_50%. For NTCP, 56% are equivalent to MD for static and 56% and 52% respectively for probabilistic evaluation with NTCP_95% and NTCP_mean.
Conclusion: The alternative-truth assessment found that the OAR delineation equivalence depends upon the OAR, patient, and the delineation algorithm. Incorporation of setup errors in the evaluation reduced the equivalence for 95ᵗʰ percentile while 50ᵗʰ percentile produces similar result to static plan except for Parotid_R. Dose-metric equivalence was less than the NTCP-based equivalence.
Keywords
Segmentation, Treatment Planning, CT
Taxonomy
Contact Email