Abstract
Automated brain-metastasis segmentation is complicated by lesion multiplicity, small volumes, heterogeneous MRI, postoperative anatomy, and overlapping evaluation regions. We evaluated a two-phase BraTS 2026 Task 1 framework using organizer-provided data. Excluding one mask with an uncorrected invalid label left 1,295 development cases; official validation comprised 179 unlabeled cases. Five-fold 2D and 3D full-resolution nnU-Net models were trained with mutually exclusive class labels and then overlapping whole-tumor, tumor-core, enhancing-tumor, and resection-cavity targets using identical images, plans, and partitions. On the 1,295-case development cohort, pooled five-fold out-of-fold RC50 mean DSC, NSD, and small-lesion F1 were 0.5605, 0.5788, and 0.3064; on the 179-case official validation cohort, the corresponding means were 0.6456, 0.6555, and 0.3153. Within region-based out-of-fold analysis, RC50 versus global 50/50 fusion changed mean DSC by +0.0050, mean NSD by −0.0010, and mean small-lesion F1 by +0.0323. Relative to the class-based equal-fusion system, observed mean differences were +0.0588 in DSC, +0.0451 in NSD, and +0.0659 in small-lesion F1. Because target formulation and fusion both changed, the official comparison does not isolate either component. The method used no external data, test-time augmentation, or component postprocessing and was packaged offline.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_METS_023.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=7E6bOlTO1g
BibTex
@InProceedings{MojAmi_RegionAligned_MICCAISAT2026,
author = { Mojtahedzadeh, Amirmahdi AND Akinmuleya, Oluwatobi Iyanuoluwa AND Daniel, Jeremiah Oluwatomi Itodo AND Adinavi, Maxwell},
title = { { Region-Aligned Training and Target-Specific 2D–3D nnU-Net Fusion for BraTS 2026 Brain Metastasis 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}
}