Abstract
Multiparametric MRI is the clinical standard for glioma assessment, yet complete four-sequence protocols are rarely guaranteed in Sub-Saharan African settings, where conventionally trained segmentation models degrade sharply when sequences are missing. We investigate the minimum set of MRI sequences required for reliable automated glioma segmentation under these constraints. Using the BraTS-Africa 2023 dataset and an nnU-Net v2 residual-en-coder backbone, we train a baseline and five probabilistic modality-dropout variants and evaluate all configurations across nine modality combinations. Independent per-modality dropout (p = 0.25) was the most robust. It reduced boundary error by ~45% relative to the baseline under full input and retained a mean Dice of 0.83 across all combinations, where the baseline collapsed. Guided by these results, we identify two sequences (T1Gd + T2-FLAIR) as a minimum viable protocol for automated segmentation. A dedicated reduced-modality model and the dropout model retained mean Dice of 0.901 and 0.894, respectively, on this two-sequence input, versus 0.560 for the baseline. A single dropout-trained model thus delivers near-specialist accuracy on a reduced imaging protocol while remaining resilient to missing sequences for glioma segmentation in resource-constrained clinics.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_013.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_013_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=3CafKTftzl
BibTex
@InProceedings{PreIza_Toward_MICCAISAT2026,
author = { Pretorius, Izak S. AND Zhanje, Thandiwe AND Anazodo, Udunna C. AND Iorumbur, Aondona Moses AND Raymond, Confidence AND Boonzaier, Willem P. E.},
title = { { Toward a minimum viable MRI protocol for glioma segmentation in resource-constrained settings } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17264},
month = {pending},
page = {pending}
}