Abstract
Glioma is the most common primary brain tumour, yet patients in Sub-Saharan Africa (SSA) face survival outcomes far worse than those in high-income settings, compounded by a severe shortage of neuro-oncology expertise and imaging infrastructure. While deep learning models trained on Global North datasets (BraTS 2021) achieve state-of-the-art performance, their transferability to African clinical acquisitions remains unquantified. In this study, we benchmark three representative architectures—3D U-Net, DynUNet, and Swin UNETR—trained on BraTS 2021 and evaluated zero-shot on BraTS-Africa (n = 146).
All three architectures achieve mean Dice of 81.2–83.1% in-domain, but experience a 13.8–15.0 percentage point (∆Dice) drop on BraTS-Africa (DynUNet zero-shot mean Dice: 69.0%, 95% CI: [67.8%, 70.2%]). Stratifying evaluation by pathology isolates disease shift from acquisition shift: for glioma-only subjects (n = 95), performance reaches 69.9% mean Dice (95% CI: [68.6%, 71.3%]), narrowing the domain gap to 10.3 pp, whereas non-glioma cases (n = 51) drop to 63.7% (95% CI: [61.5%, 65.9%]). We also introduce a convention aware label-harmonization transform resolving silent channel erasure errors. These results establish an unassisted lower bound for direct deployment, motivating future lightweight domain
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_040.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_040_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=hrzEXomUOm
BibTex
@InProceedings{OluAbr_Characterizing_MICCAISAT2026,
author = { Oluwasegun, Abraham Ajeolu AND Cheruto, Amy AND Kawesha, Mwansa AND Kiboi, Sammy Wanjohi AND Oyebisi, Oyetayo AND Otieno, Dishan AND Iorumbur, Aondona Moses AND Confidence, Raymond AND Anazodo, Udunna C. AND Djoumessi, Kerol},
title = { { Characterizing Domain Shift in Glioma Segmentation Between Global North and African MRI Datasets } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17265},
month = {pending},
page = {pending}
}