ZeroHour
Hugging Face daily paperspublished ()ingested Tristan Kirscher, Vivian Metzger, Philippe Meyer

Assessing nnU-Net Generalization across Brain Tumor Populations in BraTS-GoAT 2026

infoAI researchimportance 25
AI summary · glm-5.3-flash

A 3D nnU-Net trained on 1,351 BraTS-GoAT 2026 cases shows mean Dice dropping 0.0747 from out-of-fold to pooled validation, exposing generalization gaps.

Researchers trained a conventional 3D nnU-Net on 1,351 labeled brain tumor cases with five-fold cross-validation and 1,000 epochs per fold, averaging folds with test-time mirroring for the final predictor. On pooled official validation, global Dice scores were 0.7805 for enhancing tumor, 0.8288 for tumor core, and 0.8854 for whole tumor. Under matched fold-0 inference, mean regional Dice fell from 0.9058 on source out-of-fold cases to 0.8310 on pooled validation, and failures correlated with smaller and more disconnected enhancing tumor components rather than volume alone.

  • 3D nnU-Net trained on 1,351 labeled BraTS-GoAT cases, five-fold CV
  • Global DSC: 0.7805 ET, 0.8288 TC, 0.8854 WT on pooled validation
  • Mean Dice dropped 0.0747 from OOF source cases to pooled validation
  • Test-time mirroring gave small single-fold gains, no ensemble benefit
  • Failures linked to smaller, more disconnected enhancing tumor components
ProductsnnU-Net
OrganizationsBraTS-GoAT
Full article135 words · extracted from huggingface.co · click to collapse

BraTS-GoAT evaluates tumor segmentation across heterogeneous populations. We trained a conventional 3D nnU-Net on 1,351 labeled cases using five-fold cross-validation and 1,000 epochs per fold. The final predictor averaged all folds and applied test-time mirroring. On pooled official validation, global DSC values were 0.7805, 0.8288, and 0.8854 for enhancing tumor (ET), tumor core (TC), and whole tumor (WT). Under matched fold-0 inference, mean regional Dice decreased from 0.9058 on source out-of-fold (OOF) cases to 0.8310 on pooled validation (difference--0.0747). Mirroring gave small single-fold gains but no clear ensemble benefit; a residual-encoder alternative reached 0.8282 mean Dice. In labeled OOF predictions, failure cases had substantially smaller reference ET volumes; after adjustment for ET and WT volume, lower Dice remained associated with more disconnected ET components and a smaller fraction of ET contained in the largest component.

Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.15524