Abstract
Glioma mortality in Sub-Saharan Africa (SSA) has risen by approximately 25% over three decades, yet the deep learning models now transforming neuro-oncology are trained almost exclusively on North American and European data. We conduct the first paired, subregion-level fairness audit of two state-of-the-art glioma segmentation models: GLIM (BraTS 2023 global winner, n≈1,250 training cases) and SSAM (BraTS-SSA 2025 winner, n≈60 African cases), spanning global glioma, African glioma, and African non-glioma neoplasm cohorts. SSAM outperforms GLIM on every clinically critical subregion in African gliomas, necrotic core (d=+0.34), enhancing tumor (d=+0.23), and edema (d=+0.12), despite GLIM’s marginally higher whole-tumor Dice. A linear mixed-effects model reveals this apparent advantage is volume-confounded (β=−0.001, p=0.969 after adjustment). Both models fail equivalently on non-glioma neoplasms (p=0.570), exposing shared cross-pathology brittleness. That a regionally trained model (SSAM) achieves superior subregion fidelity on 20× fewer cases suggests that population-specific training is a more efficient path to equitable neuro-oncology AI than scale alone.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_028.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=V6A7KKrwr2
BibTex
@InProceedings{MuhAbb_Fairness_MICCAISAT2026,
author = { Muhammad, Abbas Rabiu AND Musah, Toufiq AND Zubair, Abdulrazaq A. AND Kalaiwo, Chinasa AND Ibrahim, Ummulkhairi},
title = { { Fairness and Cross-Regional Generalizability of State-of-the-Art Glioma Segmentation Models: Evaluation Across African and Global Cohorts } },
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}
}