Abstract
Brain metastases are the most common brain tumor. They
occur in 10-20% of adults with malignancies and are ten times more
common than primary brain tumors. Brain metastases typically appear
as multiple small and widely scattered lesions on MRI images, which
makes their accurate segmentation difficult. Manually delineating them
is laborious, time-consuming, inconsistent between readers, and prone
to missing small lesions. Accurate and robust segmentation is needed
for enabling efficient and reproducible clinical assessment. In this work,
we trained two distinct networks, a self-configuring nnU-Net v2 model
and MedNeXt-M. Both models were trained on the same training dataset
and were refined using post-processing steps involving lesion-size filtering
and hole filling. Evaluation by Synapse on validation data showed Med-
NeXt performed better on whole tumor (WT), tumor core (TC), and
enhancing tumor (ET) based on lesion-wise Dice, while nnU-Net was
substantially better on the resection cavity (RC) after post-processing,
with a lesion-wise Dice coefficient of 0.476 against 0.448. We ensembled
these two trained models to extract the best features of all 4 labels.
We also found that the post-processing threshold had a greater impact
on lesion-detection than on segmentation metrics, with a low lesion size
threshold improving small lesion detection substantially while only sac-
rificing a very low Dice coefficient. Further optimizations are required for
enhancing the accuracy of resection cavities. The codes are available at https://github.com/Rossette-46/BraTS-MET
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_METS_034.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/profile?id=~Anup_Singh2
BibTex
@InProceedings{RahAbd_Segmentation_MICCAISAT2026,
author = { Rahman, Abdul AND Raj, Abhishek AND Kassaw, Ewunate Assaye AND Singh, Anup},
title = { { Segmentation of Pre- and Post-treatment Brain Metastases Using Ensemble Model } },
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}
}
