Abstract
We present a clinically informed, dataset-adaptive lightweight pipeline for brain metastasis sub-region segmentation within the nnU-Net v2 framework. Targeting the four BraTS 2026 METS sub-regions — enhancing tumour (ET), non-enhancing tumour core (NETC), surrounding non-enhancing FLAIR hyper-intensity (SNFH), and resection cavity (RC) — our approach addresses key bar-riers to clinical deployment through five complementary contributions: (i) RC-balanced dataset curation using stratified K-means clustering to counter severe post-treatment class imbalance; (ii) Connected Component Dice (CCDice) loss to explicitly supervise multi-lesion spatially-disconnected structures; (iii) Jaco-bian sensitivity-guided XTiny model compression reducing model size by 80% relative to the standard nnU-Net ResEncM baseline; (iv) an on-the-fly 3D clinical prior attention map derived from T1ce spatial gradients encoding grey-white mat-ter interface, vascular junction, and Rolandic region predilections; and (v) al-ways-active modality dropout enforcing robustness to missing MRI sequences. Independent ablation shows CCDice loss and RC-balanced curation as the high-est-impact modifications, with RC DSC gains of +29.1% and +21.9% respec-tively over the full-dataset nnU-Net baseline. The resulting model is compact, modality-robust, and anatomically informed, advancing the case for clinically deployable brain metastasis segmentation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_METS_001.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=%7EWillem_Petrus_Engelbrecht_Boonzaier1
BibTex
@InProceedings{BooWil_Clinically_MICCAISAT2026,
author = { Boonzaier, Willem P. E. AND Pretorius, Izak S. AND Iorumbur, Aondona Moses AND Anazodo, Udunna C.},
title = { { Clinically Informed, Dataset-Adaptive Lightweight Segmentation of Brain Metastases within the nnU-Net Framework } },
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}
}
