BADF-Net: a boundary-aware dynamic fusion network for robust multi-anatomical structure segmentation in abdominal ultrasound
Frontiers in Medicine · 20.11.2026 · 13
DOI: 10.3389/fmed.2026.1998681 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Medicine, online erschienen am 20.11.2026. Verfasst von Md. Rashed und 9 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Introduction
Automated segmentation of multiple anatomical structures in abdominal ultrasound could enhance clinical decision–making by providing consistent quantitative measurements for disease monitoring. However, ultrasound–specific artifacts including speckle noise, low contrast and weak boundaries challenge existing deep learning models. Materials and methods We propose BADF-Net, an architecture that integrates a Boundary-Aware Dynamic Attention (BADA) module with Boundary Confidence Estimation (BCE) at every stage of a squeeze-and-excitation (SE) recalibrated ResNet-34 encoder, a Cross-Scale Boundary Fusion (CSBF) module that adaptively combines multi-scale boundary cues into a shared guidance feature and a boundary-guided decoder with attention-gated skip fusion, to improve boundary localization and tissue discrimination.
Results
On a multi-anatomical kidney and spleen ultrasound dataset, BADF-Net achieved a Dice score of 0.8174, Pixel Accuracy of 0.9703 and Mean Pixel Accuracy of 0.8521, outperforming 11 baselines with the lowest HD95 (11.49), and showing the largest gains in low-contrast structures such as kidney cortex and medulla. Agreement with expert radiologist annotations was strongest for well-defined structures such as the spleen (0.9786). By reducing operator dependency and inter-observer variability, BADF-Net can serve as a reliable second-reader tool in high-volume clinical workflows, processing images in 0.016 seconds for real-time integration.
Discussion
Our approach demonstrates how dynamic boundary attention, confidence-weighted feature refinement and channel recalibration jointly address key ultrasound segmentation challenges with direct clinical applicability. The complete experiment code is publicly available at our GitHub repository: https://github.com/rashed200613/BADF-Net .
Zitieren
APA
Rashed, M., Jamil, M., Hossain, M. I., Kavak, A., Hassan, R., Ohidujjaman, Hosain, M. S., Çakır, Ö., Fotouhi, H., & Srivastava, G. (2026). BADF-Net: a boundary-aware dynamic fusion network for robust multi-anatomical structure segmentation in abdominal ultrasound. Frontiers in Medicine, 13. https://doi.org/10.3389/fmed.2026.1998681
Vancouver
Rashed M, Jamil M, Hossain MI, Kavak A, Hassan R, Ohidujjaman, et al. BADF-Net: a boundary-aware dynamic fusion network for robust multi-anatomical structure segmentation in abdominal ultrasound. Frontiers in Medicine. 2026;13. doi:10.3389/fmed.2026.1998681
BibTeX
@article{Rashed2026,
author = {Rashed, Md. and Jamil, Muhammad and Hossain, Md. Imran and Kavak, Adnan and Hassan, Riad and Ohidujjaman and Hosain, Md. Sarwar and Çakır, Özgür and Fotouhi, Hossein and Srivastava, Gautam},
title = {BADF-Net: a boundary-aware dynamic fusion network for robust multi-anatomical structure segmentation in abdominal ultrasound},
journal = {Frontiers in Medicine},
year = {2026},
volume = {13},
doi = {10.3389/fmed.2026.1998681}
}
EUVIMED