Abstract

Accurate pediatric brain tumor segmentation from multi-parametric magnetic resonance imaging is essential for quantitative diagnosis, treatment planning, and longitudinal follow-up. However, the BraTS 2026 Pediatric Brain Tumor Segmentation (PED) task remains challenging because pediatric tumors present heterogeneous appearance, variable anatomical locations, small lesion subregions, and strong class imbalance. In this work, we propose AFS-Net, an Adaptive Fusion Skip network built on nnU-Net v2 for reliable pediatric brain tumor segmentation. Rather than introducing a separate prior-guided model, we focus on improving the nnU-Net baseline through channel recalibration (SE-style attention), a widened encoder, specificity-driven regularization, and adaptive skip fusion. These design choices are motivated by prior BraTS-PED nnU-Net improvements and recent work on dynamic skip connections in U-like architectures. AFS-Net achieves DSC scores of 0.538, 0.913, 0.247, 0.000, 0.941, and 0.941 for ET, NET, CC, ED, TC, and WT, respectively, and NSD scores of 0.448, 0.638, 0.250, 0.000, 0.657, and 0.658. These results suggest that carefully designed nnU-Net enhancements remain a strong and reliable strategy for the BraTS 2026 PED task.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_PEDs_008.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=7GDDj93lxx

BibTex

@InProceedings{LiHao_AFSNet_MICCAISAT2026,
        author = { Li, Hao AND Peng, Yanjun},
        title = { { AFS-Net: Adaptive Fusion Skip Network for Pediatric Brain Tumor Segmentation } },
        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}
}


back to top