Reply to comment on impact of deep learning on CT-based organ-at-risk delineation for flank irradiation in paediatric renal tumours: A SIOP-RTSG radiotherapy committee

Le résumé fourni par la source

We appreciate the commentary from Saad et al., which offers an opportunity to clarify key methodological and clinical aspects of our study assessing the impact of deep learning-based CT auto-contouring for organ-at-risk delineation in paediatric flank irradiation for renal tumours. First, the annotation protocol was provided in the original Supplementary Materials, and additional delineation instructions followed established SIOP-RTSG standards. Second, as already mentioned in the manuscript discussion, we acknowledge the inherent bias associated with STAPLE consensus contours, and we addressed this by including an additional single-expert reference in our evaluation. Third, although dose analysis can provide valuable clinical insights, it is not essential at this stage, as geometric evaluation remains the main benchmark for validating auto-contouring performance. Fourth, while uncertainty quantification is a promising research direction, our study was designed to reflect current clinical practice, where uncertainty-aware segmentation has not yet been integrated into routine auto-segmentation systems. Finally, we recognize that our controlled workshop environment does not fully reflect real-world clinical workflows, a limitation already discussed in our original manuscript. We hope these clarifications foster a balanced understanding of our work and support ongoing efforts toward the safe and effective clinical adoption of AI-assisted contouring in paediatric radiotherapy.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.