Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging

Jiancheng Han; Heqing Wang; Yifan Feng; Qi Yang; Jingtan Li; Haojie Zhang; Yihua He; Jiang Liu; Toru Nakamura; Yang Cao; Naidi Sun; Kun Qian; Bin Hu; Xinru Gao; Yan Xia; Zongjie Weng; Björn W. Schuller; Yoshiharu Yamamoto

doi:10.1016/j.bspc.2026.109795

Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging

Jiancheng Han
, Heqing Wang
, Yifan Feng
, Qi Yang
, Jingtan Li
, Haojie Zhang
, Yihua He
, Jiang Liu
, Toru Nakamura
, Yang Cao
, Naidi Sun^*
, Kun Qian
, Bin Hu
, Xinru Gao
, Yan Xia
, Zongjie Weng
, Björn W. Schuller
, Yoshiharu Yamamoto

^*此作品的通讯作者

医学科学与工程学院

Capital Medical University
Beijing Key Laboratory of Maternal-Fetus Medicine in Fetal Heart Disease (BZ0308)
Beijing Institute of Technology
Waseda University
The University of Osaka
Institute of Science Tokyo
Northwest Women's and Children's Hospital
Peking University
Fujian Medical University
Imperial College London
Technical University of Munich
The University of Tokyo

科研成果: 期刊稿件 › 文章 › 同行评审

摘要

This study presents the first application of federated learning (FL) for prenatal detection of Interrupted Aortic Arch (IAA) using fetal ultrasound images. To address the challenges of data scarcity, privacy constraints, and inter-institutional variability, we develop a federated learning IAA detection method and systematically evaluate three representative strategies (FedAvg, FedProx, and FedBABU) across five clinical centres. Results show that FL improves model performance over local training in recall and F1-score in data-scarce centres. Among FL algorithms, FedAvg and FedProx consistently outperform FedBABU in stability and generalisation. Among the three CNN architectures compared — ResNet-50, EfficientNet-B3, and DenseNet-121 — DenseNet-121 demonstrates superior overall performance, particularly in non-independent and identically distributed (Non-IID) scenarios. Our framework demonstrates the feasibility of collaborative AI for rare disease detection without data sharing, laying the foundation for scalable, real-world prenatal screening of congenital heart defects.

源语言	英语
文章编号	109795
期刊	Biomedical Signal Processing and Control
卷	119
DOI	https://doi.org/10.1016/j.bspc.2026.109795
出版状态	已出版 - 15 6月 2026

访问文件

10.1016/j.bspc.2026.109795

其它文件与链接

链接到 Scopus 的出版物

引用此

Han, J., Wang, H., Feng, Y., Yang, Q., Li, J., Zhang, H., He, Y., Liu, J., Nakamura, T., Cao, Y., Sun, N., Qian, K., Hu, B., Gao, X., Xia, Y., Weng, Z., Schuller, B. W., & Yamamoto, Y. (2026). Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging. Biomedical Signal Processing and Control, 119, 文章 109795. https://doi.org/10.1016/j.bspc.2026.109795

@article{3a3aef46aa8645f2aa9655e2f0fbca0b,

title = "Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging",

abstract = "This study presents the first application of federated learning (FL) for prenatal detection of Interrupted Aortic Arch (IAA) using fetal ultrasound images. To address the challenges of data scarcity, privacy constraints, and inter-institutional variability, we develop a federated learning IAA detection method and systematically evaluate three representative strategies (FedAvg, FedProx, and FedBABU) across five clinical centres. Results show that FL improves model performance over local training in recall and F1-score in data-scarce centres. Among FL algorithms, FedAvg and FedProx consistently outperform FedBABU in stability and generalisation. Among the three CNN architectures compared — ResNet-50, EfficientNet-B3, and DenseNet-121 — DenseNet-121 demonstrates superior overall performance, particularly in non-independent and identically distributed (Non-IID) scenarios. Our framework demonstrates the feasibility of collaborative AI for rare disease detection without data sharing, laying the foundation for scalable, real-world prenatal screening of congenital heart defects.",

keywords = "Federated learning, Fetal ultrasound imaging, Interrupted aortic arch",

author = "Jiancheng Han and Heqing Wang and Yifan Feng and Qi Yang and Jingtan Li and Haojie Zhang and Yihua He and Jiang Liu and Toru Nakamura and Yang Cao and Naidi Sun and Kun Qian and Bin Hu and Xinru Gao and Yan Xia and Zongjie Weng and Schuller, \{Bj{\"o}rn W.\} and Yoshiharu Yamamoto",

note = "Publisher Copyright: {\textcopyright} 2026",

year = "2026",

month = jun,

day = "15",

doi = "10.1016/j.bspc.2026.109795",

language = "English",

volume = "119",

journal = "Biomedical Signal Processing and Control",

issn = "1746-8094",

publisher = "Elsevier Ltd.",

}

Han, J, Wang, H, Feng, Y, Yang, Q, Li, J, Zhang, H, He, Y, Liu, J, Nakamura, T, Cao, Y, Sun, N , Qian, K , Hu, B, Gao, X, Xia, Y, Weng, Z, Schuller, BW & Yamamoto, Y 2026, 'Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging', Biomedical Signal Processing and Control, 卷 119, 109795. https://doi.org/10.1016/j.bspc.2026.109795

TY - JOUR

T1 - Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging

AU - Han, Jiancheng

AU - Wang, Heqing

AU - Feng, Yifan

AU - Yang, Qi

AU - Li, Jingtan

AU - Zhang, Haojie

AU - He, Yihua

AU - Liu, Jiang

AU - Nakamura, Toru

AU - Cao, Yang

AU - Sun, Naidi

AU - Qian, Kun

AU - Hu, Bin

AU - Gao, Xinru

AU - Xia, Yan

AU - Weng, Zongjie

AU - Schuller, Björn W.

AU - Yamamoto, Yoshiharu

PY - 2026/6/15

Y1 - 2026/6/15

N2 - This study presents the first application of federated learning (FL) for prenatal detection of Interrupted Aortic Arch (IAA) using fetal ultrasound images. To address the challenges of data scarcity, privacy constraints, and inter-institutional variability, we develop a federated learning IAA detection method and systematically evaluate three representative strategies (FedAvg, FedProx, and FedBABU) across five clinical centres. Results show that FL improves model performance over local training in recall and F1-score in data-scarce centres. Among FL algorithms, FedAvg and FedProx consistently outperform FedBABU in stability and generalisation. Among the three CNN architectures compared — ResNet-50, EfficientNet-B3, and DenseNet-121 — DenseNet-121 demonstrates superior overall performance, particularly in non-independent and identically distributed (Non-IID) scenarios. Our framework demonstrates the feasibility of collaborative AI for rare disease detection without data sharing, laying the foundation for scalable, real-world prenatal screening of congenital heart defects.

AB - This study presents the first application of federated learning (FL) for prenatal detection of Interrupted Aortic Arch (IAA) using fetal ultrasound images. To address the challenges of data scarcity, privacy constraints, and inter-institutional variability, we develop a federated learning IAA detection method and systematically evaluate three representative strategies (FedAvg, FedProx, and FedBABU) across five clinical centres. Results show that FL improves model performance over local training in recall and F1-score in data-scarce centres. Among FL algorithms, FedAvg and FedProx consistently outperform FedBABU in stability and generalisation. Among the three CNN architectures compared — ResNet-50, EfficientNet-B3, and DenseNet-121 — DenseNet-121 demonstrates superior overall performance, particularly in non-independent and identically distributed (Non-IID) scenarios. Our framework demonstrates the feasibility of collaborative AI for rare disease detection without data sharing, laying the foundation for scalable, real-world prenatal screening of congenital heart defects.

KW - Federated learning

KW - Fetal ultrasound imaging

KW - Interrupted aortic arch

UR - https://www.scopus.com/pages/publications/105029520430

U2 - 10.1016/j.bspc.2026.109795

DO - 10.1016/j.bspc.2026.109795

M3 - Article

AN - SCOPUS:105029520430

SN - 1746-8094

VL - 119

JO - Biomedical Signal Processing and Control

JF - Biomedical Signal Processing and Control

M1 - 109795

ER -

Federated learning for prenatal detection of interrupted aortic arch using fetal ultrasound imaging

摘要

访问文件

其它文件与链接

学术指纹

引用此