Abstract
Large vision–language models (LVLMs) have demonstrated strong performance in chest X-ray analysis and automated radiology report generation, but often exhibit racial bias, leading to lower accuracy on data from low- and middle-income countries (LMICs). To address these challenges, we evaluated existing Med-VLMs on Naija-CXR, a newly curated dataset from a tertiary hospital in Nigeria, and on the MIMIC-CXR dataset from the United States. Existing MedVLMs show population bias. To mitigate poor performance due to population-domain shifts, we introduce NaijaCXR-VLM, a lightweight, multitask framework that combines supervised adversarial adaptation and parameter-efficient fine-tuning for pathology classification and report generation. Built on MedGemma-4B, our model employs gradient-reversal and supervised alignment modules to learn domain-invariant representations, thereby mitigating population bias. This method enhances cross-domain generalization and achieves 80.23% accuracy, 50.94 BLEU-1, 57.82 RadGraph, and 65.22% AUROC on Naija-CXR. NaijaCXR-VLM outperforms all baselines models on report generation performance and achieves highest F1-score, precision, and recall for pathology classification, highlighting the role of domain adaptation in mitigating population bias.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AFRICAI_003.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=TpEawNyxFL
BibTex
@InProceedings{MusAmi_NaijaCXRVLM_MICCAISAT2026,
author = { Musa, Aminu AND Prasad, Rajesh AND Hernandez, Monica},
title = { { NaijaCXR-VLM: A Multi-task Vision-Language Model for Chest X-ray Analysis in Low-resource 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}
}
