Sensitivity of contour curation to the empirical threshold

Determine the sensitivity of the digital-twin contour-uncertainty estimate to the empirically selected curation threshold used to exclude library sessions with poor propagated-contour agreement, and establish whether the curation procedure remains valid when applied to another institution’s library.

Background

The framework removes library sessions that produce implausible deformations, using an empirically chosen threshold based on mean high-dose clinical target volume Dice agreement. This curation is intended to prevent a small number of failed sessions from dominating the consensus contour and uncertainty estimate.

The authors explicitly acknowledge that the threshold was not systematically varied and that the curation procedure would need to be repeated for another institution’s library. Consequently, the robustness of the reported forecast uncertainty to threshold selection and institutional dataset variation remains unresolved.

References

The curation threshold was chosen empirically and not swept, so the sensitivity of the estimate to a poorly chosen value is unknown, and the curation would have to be repeated on another institution’s library.

— A digital-twin framework for forecasting treatment-day imaging with contour uncertainty in adaptive proton radiotherapy  (2609.24725 - Wu et al., 21 Sep 2026) in Section 5, Discussion, paragraph beginning “Several limitations bound the work”