Abstract
MRI-based gross tumor volume (GTV) delineation is central to postoperative flank irradiation planning for Pediatric renal tumor, yet heterogeneous imaging protocols across centers limit the robustness and clinical usability of automated segmentation models. This challenge is amplified in pediatric radiotherapy, where data are scarce and fragmented across institutions. We developed and evaluated a sequenceagnostic MRI deep learning approach for automated Wilms tumor GTV segmentation, using a national multi-sequence MRI cohort (T1, T1-Gd, and T2) for development and an external public cohort to test generalization. We show that single-sequence models generalize poorly across MRI contrasts, whereas multi-sequence training maintains in-domain accuracy while improving robustness across sequences. Fine-tuning of foundation models like DINO-UNet, MedDINOv3, and VoxTell did not outperform this multi-sequence baseline. RC-based augmentation improved generalization to unseen sequences for single-sequence models. When applied to the multi-sequence model, lightweight random-convolution augmentation further improved external generalization, increasing the median DSC from 81.2 [67.6–86.0] to 82.5 [72.4–87.6] (p = 0.004), with negligible in-domain cost. These results indicate that multi-sequence training combined with lightweight appearance augmentation is a practical strategy for robust Wilms tumor GTV segmentation across heterogeneous MRI protocols.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PedAItrics_019.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{DinMia_SequenceAgnostic_MICCAISAT2026,
author = { Ding, Mianyong AND Radu, Adrian-Marian AND van den Heuvel-Eibrink, Marry M. AND Janssens, Geert O. AND Maspero, Matteo},
title = { { Sequence-Agnostic MRI Segmentation of Pediatric Renal Tumors for Postoperative Flank Irradiation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17257},
month = {pending},
page = {pending}
}