Abstract
Brain tumor segmentation remains challenging due to substantial tumor heterogeneity across pathological subtypes, with variations in lesion size, morphology, and spatial distribution limiting cross-tumor generalization. Spiking neural networks (SNNs) provide an alternative perspective for medical image analysis through sparse event-driven computation and temporal dynamic representation. In this work, based on our previously proposed SpikeFormer-UNet, we develop a pseudo-label enhanced training framework for the BraTS 2026 Generalizability Across Tumors (GoAT) challenge. The framework employs heterogeneous ensembles of CNN, Transformer, and SNN segmentation models to generate pseudo-labels from unlabeled MRI data, enabling SpikeFormer-UNet adaptation to diverse tumor distributions without architectural modification. During inference, a confidence-aware ET adaptive thresholding (CEAT) strategy combined with connected-component filtering (CCF) is introduced to refine segmentation results. On the BraTS 2026 GoAT validation set, the proposed framework achieves Dice scores of 0.8726, 0.8244, and 0.7764 for WT, TC, and ET, respectively. Analysis further reveals that SpikeFormer-UNet exhibits complementary prediction patterns to conventional architectures, including lower false positive (FP) rates and improved ET segmentation. By integrating these complementary representations into pseudo-label generation and ensemble inference, the proposed framework achieves robust brain tumor segmentation performance. Code is available at: https://github.com/jianan-zhao/Pseudo-Label-Enhanced-SpikeFormer-UNet.git.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_GoAT_009.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/BraTS_GoAT_009_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=JmEyoG6KSa
BibTex
@InProceedings{ZhaJia_PseudoLabel_MICCAISAT2026,
author = { Zhao, Jianan AND Ahmad, Parvez AND Huang, Ziyi AND Shim, Vickie AND Park, Thomas AND Kasabov, Nikola AND Wang, Alan},
title = { { Pseudo-Label Enhanced SpikeFormer-UNet for Generalizable 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}
}