AIstatLab describes consensus pseudo-label pipeline for BraTS-GoAT 2026 brain-tumor segmentation
A 3D ResidualEncoderUNet trained on screened consensus pseudo-labels reached mean DSC 0.834 on the challenge validation server, with the clearest component gains on ET regions.
In plain words
The team describes a neural network that segments brain tumors into shared regions used by the BraTS-GoAT challenge. To get more training data, they had five teacher models label 1,138 unlabeled scans, then kept only the 983 predictions that passed confidence and agreement checks. Their final model scored a mean DSC of 0.834 on the challenge's validation server.
What was announced
Writing on arXiv, the AIstatLab team reports its entry for BraTS-GoAT 2026 Task 3, which evaluates brain-tumor segmentation across heterogeneous entities using shared ET, TC, and WT regions. The system is a 3D ResidualEncoderUNet trained with quality-controlled consensus pseudo-labels, a training-only nested-region objective, and a four-expert bottleneck adapter. Five-fold ResEnc-M/L teachers produced predictions for 1,138 unlabeled cases; after fusion, hierarchy enforcement, and confidence-and-agreement screening, 983 cases were retained for training alongside 1,351 annotated cases.
The final five-fold ResEnc-M entry achieved mean DSC 0.834, mean NSD 0.510, and Acalc of 0.6720 on the validation server. The authors state that component effects were modest and region-dependent, with the clearest gains on ET. Source code is available in a public GitHub repository.
What it does not show
- The source is an abstract only, so full method details and complete results are not available in the supplied evidence.
- Scores come from the challenge validation server as reported by the authors; no independent confirmation is provided.
- The abstract reports component effects as modest and region-dependent without giving per-component numbers.
Sources
Source-support review completed 21 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 21 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.