Abstract
Accurate delineation of the gross tumor volume (GTV) using multi-modal imaging is essential for postoperative radiotherapy planning in pediatric patients. Unlike adult preoperative cases, for which most multi-modal segmentation methods are designed, pediatric postoperative GTV delineation presents distinct challenges. The diversity of tumor types and complexity of postoperative changes result in case-specific variations in the imaging modalities that provide relevant diagnostic information. Methods that assume agreement between modalities or normalize attention across modalities are not well suited to such scenarios. We propose Task-aware Modality Contribution Fusion (TMCF), a per-voxel, per-modality gating module that scores each modality as the product of a complementarity factor, which measures whether the modality contributes information beyond other modalities at a given location, and a task-relevance factor, which measures whether that information reflects a tumor-related abnormality. The resulting contribution maps are reused at multiple encoder feature scales through lightweight gates, propagating modality-aware modulation through the backbone rather than acting only at the input. TMCF was evaluated on 133 pediatric postoperative proton therapy cases covering four tumor types using 5-fold cross-validation. It attained the highest mean Dice similarity coefficient of 0.7483 among the compared state-of-the-art convolutional and transformer methods, with significantly higher recall than all baselines and a smaller case-level standard deviation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PedAItrics_008.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{HouZhu_TaskAware_MICCAISAT2026,
author = { Hou, Zhuo AND Liu, Xing’an AND Zhang, Xuanrong AND Shimizu, Shosei AND Shimizu, Akinobu},
title = { { Task-Aware Modality Contribution Fusion Framework for Pediatric Postoperative Gross Tumor Volume Segmentation } },
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}
}