Abstract
Pediatric brain tumors can change during and after treatment, which makes accurate tumor monitoring important for clinical follow-up and treatment planning. Automated segmentation of tumor subregions from multi-modal magnetic resonance imaging (MRI) can support the comparison of pre-treatment and post-treatment scans by providing consistent measurements of tumor location and volume. In this work, we present a three-dimensional segmentation framework called MedSwinNet for the BraTS-PED task. The model uses four MRI modalities, including T1-weighted contrast-enhanced, native T1-weighted, T2-FLAIR, and T2-weighted images. MedSwinNet combines MedNeXt-style convolutional blocks with Swin Transformer blocks to capture both local tumor details and wider spatial information. The decoder is guided by an auxiliary whole-tumor prediction through spatially adaptive normalization, while a classwise modality attention module allows each tumor subregion to use information from the most relevant MRI sequences. The proposed model directly predicts enhancing tumor, nonenhancing tumor, cystic component, and peritumoral edema. Training uses class-balanced patch sampling, spatial and intensity augmentation, deep supervision, and a combined loss containing Dice–cross-entropy, weighted focal loss, and auxiliary whole-tumor loss. During inference, sliding-window prediction, test-time augmentation, probability averaging, and connected-component filtering are applied. The proposed framework provides a structured approach for pediatric brain tumor segmentation and may support future longitudinal analysis of tumor changes before and after treatment.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_PEDs_015.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=6yEoBGlFcu
BibTex
@InProceedings{HosMD_Pediatric_MICCAISAT2026,
author = { Hossen, MD Fayaz Bin AND Evans, Michael L. AND Sadique, MD Shibly AND Farzana, Walia AND Rahman, Asfaqur AND Temtam, Ahmed AND Iftekharuddin, Khan M.},
title = { { Pediatric Brain Tumor Segmentation with MedSwinNet } },
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}
}