Abstract
Brain metastases are the most common intracranial tumors, and segmenting them by hand is slow work: a single scan can carry a dozen lesions, many only a few millimeters across. We built a pipeline for the BraTS 2026 Metastases challenge using MONAI Auto3DSeg with a SegResNetDS backbone, trained from scratch on the 1,296 provided cases with no pretrained weights: five folds split by patient rather than by case, 245 epochs each on four NVIDIA H200 GPUs, averaged into one probability map and post-processed with per-region thresholds, cavity hysteresis and label-hierarchy enforcement. On the 179 validation cases this reaches lesion-wise instance F1 of 0.72 for tumor core, 0.71 for enhancing tumor, 0.69 for whole tumor, and 0.28 for resection cavity. A case-level decomposition shows the cavity figure to be precision-driven: our final two configurations detect the same 8 cavity instances in the same 23 cavity-bearing cases and differ only in false positives. We also report interventions that did not improve performance under the configurations tested, four of which reflect measured properties of the lesion-wise evaluation setting rather than of the model. Cavity detection remains unresolved: 15 of 23 cavity-bearing cases are missed under every threshold we swept.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_METS_036.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=QJtnjjhBwX
BibTex
@InProceedings{GhoMah_ADetectionAware_MICCAISAT2026,
author = { Ghoraba, Mahmoud AND Ghoraba, Mohamed AND Ghonim, Mohanad AND Ghonim, Mohamed},
title = { { A Detection-Aware SegResNetDS Ensemble and a Measured Null Space for Brain Metastasis Segmentation in BraTS 2026 } },
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}
}
