Abstract
Pediatric brain tumors (PBTs) such as glioblastoma, diffuse midline glioma, and diffuse intrinsic pontine glioma remain highly aggressive, with survival rates below 20%. Accurate segmentation of tumor subregions, including the enhancing region (ET), non-enhancing components (NET), cystic component
(CC), and edema (ED), is essential for diagnosis and treatment but remains diffi
cult due to heterogeneous pediatric neuroanatomy and imaging limitations.
Therefore, we proposed a deep learning framework for automated segmentation of PBTs. Preprocessing includes skull stripping, intensity normalization, and data augmentation, followed by training 3D U-Net and Swin UNETR models using a suitable combination of loss functions for each tumor subregion (ET, NET, ED, and CC). Furthermore, the best-performing model is selected to predict the subregion. ET and NET are best predicted by Swin UNETR, and CC and ED are best predicted by 3D U-Net, which are combined into a single prediction mask in a fixed priority order of ET, NET, CC, and ED to resolve overlapping predictions within the subregion. Test-time augmentation is applied to all subregion models, followed by post-processing based on connected-component analysis, which reduces false positives and improves the overall dice score of the whole tumor. The proposed framework achieves a global dice similarity coefficient of 0.86 and 0.88 on the validation data for the whole tumor and tumor core, respectively.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_PEDs_013.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=V13ItDhzxK
BibTex
@InProceedings{SriSan_SubRegionSpecific_MICCAISAT2026,
author = { Srivastava, Sanskriti AND Kehara, Thulani AND Yadav, Harsh AND Singh, Anup},
title = { { Sub-Region-Specific Deep Learning Model for Pediatric Brain Tumor Segmentation on Multi-Parametric MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17253},
month = {pending},
page = {pending}
}
