Abstract
Accurate volumetric assessment of brain metastases is crit-
ical for treatment planning and response evaluation, yet manual delin-
eation is time-consuming and subject to inter-observer variability, par-
ticularly for small and multifocal lesions. In this work, we address the
problem of automated brain metastasis detection and multi-class segmen-
tation, including post-treatment resection cavities, using the nnU-Net
framework. We integrate two MRI foundation encoders—Triad (convolu-
tional, used for encoder initialization) and BrainIAC (vision-transformer,
used as a frozen auxiliary feature extractor)—through the transfer inter-
face each architecture permits. Representation probing of both frozen en-
coders guided the design of class-specific fusion weights, and confidence-
aware connected-component filtering was applied as post-processing. Our
experiments on the BraTS-METS 2026 dataset demonstrate improved
segmentation across all four evaluation regions relative to a plain nnU-
Net baseline, reaching lesion-wise DSC of 0.703, 0.560, 0.727 and 0.686
for ET, RC, TC and WT. An ablation shows that the contributions are
region-dependent: RC case oversampling produced the largest resection-
cavity gain, with further increments from the BrainIAC branch and RC-
specific fusion, whereas confidence-aware filtering drives the gains in ET,
TC and WT.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_METS_020.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=nkj0nDxqp1
BibTex
@InProceedings{KimDoy_Segmentation_MICCAISAT2026,
author = { Kim, Doye AND Han, Yeon Gyu AND Lee, Dongheon},
title = { { Segmentation of Pre- and Post-treatment Brain Metastases Using an Ensemble of MRI Foundation Models } },
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}
}