Abstract

The growing shift toward volumetric response assessment in pediatric neuro-oncology requires accurate 3D segmentation. However, pediatric brain tumor segmentation on multiparametric MRI is complicated by limited annotated data, heterogeneous tumor appearance, and imbalance among enhancing tumor (ET), non-enhancing tumor (NET), cystic component (CC) and peritumoral edema (ED). We investigate a fixed three-dimensional Laplacian-of-Gaussian (LoG) input transform in vanilla and residual-encoder (ResEnc-M) nnU-Net configurations and evaluate whether adding class-weighted cross-entropy benefits the LoGaugmented models.We use patient-level five-fold cross-validation to compare two nnU-Net architectures under two loss settings and combine the selected standard nnU-Net branch, the selected LoG-augmented branch, and HFF-Net in an architecture-diverse ensemble. Cross-validation selected the class-weighted vanilla nnU-Net as the standard branch and the default-loss LoG-augmented ResEnc-M nnU-Net as the LoG branch. Class weighting did not consistently benefit the LoG-augmented models. On the official BraTS-PEDs 2026 challenge validation set, the final ensemble achieved global Dice scores of 0.246 (CC), 0.000 (ED), 0.494 (ET), 0.907 (NET), 0.937 (tumor core, TC), and 0.937 (whole tumor, WT), and global normalized surface distance scores of 0.246 (CC), 0.000 (ED), 0.408 (ET), 0.639 (NET), 0.661 (TC), and 0.662 (WT).

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_PEDs_005.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=lOVVxXaSDX

BibTex

@InProceedings{CheYon_LoGAugmented_MICCAISAT2026,
        author = { Chen, Yong AND Medrea, Sebastian AND Khalvati, Farzad},
        title = { { LoG-Augmented nnU-Net Ensemble 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