Do Medical Foundation Models Generalize on the African Brain?

2026-07-30

AI Summary

This study addresses the lack of systematic evaluation of generalization capabilities of existing medical foundation models on brain MRI data from African populations, a gap that raises concerns about potential algorithmic bias. We present the first comprehensive assessment of both general-purpose and segmentation-specific foundation models—including BrainIAC, 3DINO, and MedSAM2—on African (Nigeria Dementia Dataset, BraTS-Africa) and non-African brain MRI datasets across dementia classification and brain tumor segmentation tasks, benchmarking against models trained from scratch. Results show modest gains in classification performance (maximum ROC-AUC of 0.86) but substantially improved segmentation accuracy over baselines (maximum Dice score of 0.86). Model performance was primarily driven by training data scale rather than geographic origin, with no evidence of inherent bias against African populations.

Abstract

Medical foundation models (FMs) are increasingly used for brain MRI analysis. However, their evaluation remains dominated by high-resource datasets, leaving generalization to African cohorts underexplored. We assess whether FMs generalize equally to African and non-African brain MRI data across two tasks: dementia classification using a Nigerian dataset and brain tumor segmentation using BraTS-Africa. We evaluate two generalist FMs (BrainIAC, 3DINO) and two segmentation-specific FMs (MedSAM2, Medical-SAM2) against a from-scratch baseline. For classification, FMs provide limited gains (highest ROC-AUC of 0.86 with BrainIAC), whereas for segmentation they consistently improve performance, reaching up to 0.86 Dice with MedSAM2. Performance differences between African and non-African cohorts are inconsistent and appear more related to dataset size than data origin. These results suggest that FMs do not exhibit an inherent bias against African cohorts, and highlight the limited availability and diversity of African neuroimaging datasets as the main barrier to robust evaluation and deployment.

Problem

• medical foundation models
• generalization
• African brain MRI
• health equity
• neuroimaging datasets

Innovation

• medical foundation models
• brain MRI
• generalization
• African cohorts
• neuroimaging datasets