Abstract
Breast cancer remains a major public health concern in Africa, where
limited access to diagnostic resources delays early detection. Point-of-care
ultrasound offers a practical solution for community screening, but interpretation
requires expertise often unavailable in low-resource settings. We propose a
transfer learning framework for breast lesion detection using
community-acquired ultrasound images from the ABreast/PACE dataset in
Nigeria and Uganda, including 280 participants. The framework integrates an
EfficientNetV2 backbone with multimodal fusion of clinical metadata and an
attention mechanism to highlight diagnostically relevant regions. BI-RADS
scores were used solely as a reference standard for evaluation and were not
included as input features. Model performance was assessed using stratified
five-fold cross-validation, yielding average accuracy of 92.4 ± 0.1 percent,
sensitivity of 90.1 ± 0.2 percent, specificity of 93.7 ± 0.1 percent, F1-score of
0.91 ± 0.1, and area under the curve of 0.95 ± 0.1. Ablation confirmed the added
value of transfer learning and metadata fusion compared to training from scratch.
Grad-CAM++ visualizations supported interpretability by highlighting clinically
relevant lesion regions.
These findings highlight the potential of AI-enabled ultrasound to strengthen
breast cancer screening in resource-constrained settings, while underscoring the
need for larger datasets and external validation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_056.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_056_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=jXhgVriLPk
BibTex
@InProceedings{GboTem_Breast_MICCAISAT2026,
author = { Gbolahan, Temitayo Ismail A. AND Chabi, Wahabou K. Taba AND Hounton, Johannes AND Cocouvi, Alexandre AND Missihoun, Jonathan Suru AND Hovozounkou, Rodrigue AND Legbassi, Gildas Sèyigbénan AND Minaba, Sêgnimaké Tatiana Carine AND Musah, Toufiq AND Kalaiwo, Chinasa AND Anazodo, Udunna C. AND Raymond, Confidence AND Iorumbur, Aondona Moses AND Bankole, Nourou Dine Adeniran},
title = { { Breast Lesion Detection from Community-Acquired POCUS Images: A Framework Proposal Based on Transfer Learning for Improved Screening in Africa } },
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}
}