Abstract
In sub-Saharan Africa, fetal growth restriction drives high perinatal mortality, yet screening requires standard plane identification demanding scarce sonographer expertise. AI tools trained on high-resource data fail under domain shift: we show a European-trained baseline classifier drops from 95.8% F1 on European data to 57.0% F1 on African data, concentrated in the critical abdomen and femur planes. We present Afro-FetalNet, a domain-adaptive pipeline utilizing selective CLAHE, Leave-One-Country-Out (LOCO) transfer learning, and model-soup averaging. Evaluated under a strict held-out protocol on Malawian data, our model recovers 41.7 F1 points, achieving 98.7% macro F1 (95% Wilson score CI: [92.8%, 99.8%]) using only 153 labelled training images per fold (51 images per source country) while preserving source-domain accuracy (95.2% F1). Grad-CAM analysis demonstrates sharp anatomical focus on key diagnostic boundaries, establishing an interpretable pathway for low-resource deployment.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AFRICAI_014.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=VdRhk0V6fE
BibTex
@InProceedings{OluSam_AfroFetalNet_MICCAISAT2026,
author = { Oluwatobi, Samuel Olusola AND Mugume, Paul Okwija AND Muriira, Patricia AND Gael, Ishimwe Karekezi Guy AND Mukamakuza, Carine},
title = { { Afro-FetalNet: Fetal Plane Classification for Low-Resource African 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}
}