EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

SR–MobileNet–MSA: A Super‐Resolution and Multi‐Scale Attention Network for Dandy–Walker Malformation Diagnosis in Fetal Ultrasound

Amene Vatanparast, Mansoor Fateh, Hoda Mashayekhi

The Journal of Engineering · 2026-01 · 2026 (1)

DOI: 10.1049/tje2.70206 ↗

Kurz gefasst

Fachartikel in der Zeitschrift The Journal of Engineering, online erschienen am 06.07.2026. Verfasst von Amene Vatanparast und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The automated diagnosis of Dandy–Walker malformation (DWM) in fetal ultrasound images remains challenging due to the subtle and localized nature of the anomaly, as well as inherent imaging artefacts such as low contrast, speckle noise and loss of fine anatomical details during image resizing. Although deep learning (DL) has shown promise, existing approaches often exhibit non‐zero false negative (FN) rates, which pose a critical clinical risk. In this work, we propose a lightweight, task‐specific DL framework, termed Super‐Resolution MobileNet and Multi‐Scale Attention Network (SR–MobileNet–MSA), designed to address these challenges through the integration of complementary components tailored to ultrasound‐based DWM detection. The framework incorporates an efficient sub‐pixel convolutional super‐resolution (ESPCN) module at the input stage, enabling the recovery of fine anatomical details within an end‐to‐end learning pipeline. This is followed by a multi‐scale feature fusion mechanism that combines low‐level, mid‐level and high‐level representations to capture both subtle structural cues and broader contextual information. A squeeze‐and‐excitation (SE) attention block is further employed to adaptively emphasize diagnostically relevant feature channels. We curated a dataset of 894 fetal ultrasound images, comprising 446 normal and 448 DWM cases. Experimental results demonstrate that the proposed framework, with approximately 1.01 M parameters, achieves 99.34% accuracy and 100% recall on the test set, with no FNs observed in this evaluation. These results highlight the effectiveness of the proposed integration strategy in enhancing sensitivity to subtle abnormalities while maintaining computational efficiency. Additional robustness experiments under source shift and image degradation further suggest that the proposed framework maintains strong sensitivity under more challenging conditions. Code and data are available at https://github.com/amenevatanparast/SR_MobileNet_MSA .

Zitieren

APA

Vatanparast, A., Fateh, M., & Mashayekhi, H. (2026). SR–MobileNet–MSA: A Super‐Resolution and Multi‐Scale Attention Network for Dandy–Walker Malformation Diagnosis in Fetal Ultrasound. The Journal of Engineering, 2026(1). https://doi.org/10.1049/tje2.70206

Vancouver

Vatanparast A, Fateh M, Mashayekhi H. SR–MobileNet–MSA: A Super‐Resolution and Multi‐Scale Attention Network for Dandy–Walker Malformation Diagnosis in Fetal Ultrasound. The Journal of Engineering. 2026;2026(1). doi:10.1049/tje2.70206

BibTeX
@article{Vatanparast2026,
  author = {Vatanparast, Amene and Fateh, Mansoor and Mashayekhi, Hoda},
  title = {SR–MobileNet–MSA: A Super‐Resolution and Multi‐Scale Attention Network for Dandy–Walker Malformation Diagnosis in Fetal Ultrasound},
  journal = {The Journal of Engineering},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1049/tje2.70206}
}
Kostenlos starten →