Abstract
Brain metastasis segmentation on magnetic resonance imaging (MRI) poses significant challenges, as lesions are often small, spatially heterogeneous, and exhibit substantial variability between pre- and post-treatment scans. Post-surgical cases introduce additional difficulty due to the presence of resection cavities, which are rare and visually variable. In this work, we presented a multi-network ensemble strategy for Task 1 of the BraTS-METS 2026 challenge. The proposed approach combined three complementary 3D segmentation models, MedNeXt, nnU-Net, and STU-Net, trained on multi-sequence MRI scans including T1-weighted, contrast-enhanced T1-weighted, T2-weighted, and FLAIR images. Model predictions were fused using class-specific weighted probability averaging for standard tumor-related labels and the resection-cavity class, across three stages. A first ensemble stage optimized the segmentation of non-enhancing tumor core, surrounding FLAIR hyperintensity, and enhancing tumor, while a second stage applied class-specific calibration for resection cavities. A final connected-component post-processing step was employed to reduce small false-positive predictions. On the validation leaderboard, the final ensemble improved the mean Dice score from 0.609 to 0.627 compared with the non-post-processed ensemble, with gains across enhancing tumor, tumor core, whole tumor, and resection cavity regions. These results suggest that class-specific ensemble fusion and conservative post-processing provide a practical strategy for robust brain metastasis segmentation in heterogeneous clinical MRI datasets. Our code is available at https://github.com/NicoloPecco/Segmentation_METs_BraTS2026.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_METS_002.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=aVQSLIIKX0
BibTex
@InProceedings{PecNic_AMultiNetwork_MICCAISAT2026,
author = { Pecco, Nicolo AND Veronese, Lorenzo},
title = { { A Multi-Network Ensemble Strategy for the BraTS 2026 Brain Metastasis Segmentation Challenge } },
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}
}
