Abstract
Breast ultrasound holds great potential for cancer screening
in resource-limited settings, yet lesion characterization remains challenging due to speckle noise, acoustic shadowing, and limited specialist access. We propose an uncertainty-aware dual-task framework for joint lesion segmentation and malignancy classification on 2,773 images from three public datasets (5-fold cross-validation), comparing nine augmentation combinations across ResNet34, ResNet50, and EfficientNet-B3. We also present DUSpeckNet, a lightweight dual-task network trained from scratch. With ultrasound specific augmentation, ResNet50 achieved Dice= 0.88 ± 0.01 and EfficientNet-B3 achieved AUC= 0.94 ± 0.02, Accuracy= 0.89 ± 0.02. DUSpeckNet achieved Dice= 0.85±0.01, AUC= 0.92±0.01 at 51.8% fewer parameters with no pretraining. On the ABreast cohort (a small African handheld point-of-care dataset, n=20, all benign), transfer is assessed for segmentation only, as the benign-only composition prevents malignancydetection evaluation. EfficientNet-B3 and DUSpeckNet achieved segmentation Dice= 0.85±0.00 and 0.82±0.02 respectively. Classification uncertainty tracked error for EfficientNet-B3 (r = 0.34, p = 0.0005) but not for DUSpeckNet, identifying uncertainty calibration under pretraining-free transfer as a key future direction.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_026.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_026_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=aMddr4NUex
BibTex
@InProceedings{IorAon_DomainSpecific_MICCAISAT2026,
author = { Iorumbur, Aondona Moses AND Sanni, Henry Ananyi AND Raymond, Confidence},
title = { { Domain-Specific Augmentation and Uncertainty Quantification for Robust Breast Ultrasound Lesion Segmentation and Classification } },
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}
}