A Structure-Aware Self-Supervised Framework for Transferable Fetal Ultrasound Screening
- 1Shanghai Medical College, Fudan University, Shanghai, China
Abstract
Accurate prenatal screening from ultrasound imaging is severely constrained by the extreme spatial heterogeneity of fetal structures and the data scarcity inherent to rare congenital anomalies. To overcome these computational bottlenecks, we present RadarMAE, a data-efficient self-supervised representation learning framework tailored for fetal ultrasound analysis. By employing a novel, structure-aware masking strategy, RadarMAE enables robust pre-training on highly limited clinical datasets. Benchmarking the framework on first-trimester screening for trisomy 21 (Down syndrome), our model achieved exceptional diagnostic performance on an internal cohort, with an area under the curve (AUC) of 0.99, 97% accuracy, 94% sensitivity, and 98% specificity. Crucially, through domain-adversarial adaptation, the framework demonstrates robust cross-institutional generalization under extreme source-target data imbalance, bounding performance degradation to[≤] 0.04 across all key metrics in external testing. Furthermore, by integrating spatial registration with explainable AI, we constructed the first population-level facial risk atlas for trisomy 21. This data-driven interpretability framework validates standard clinical markers while revealing the comparable diagnostic weight of underutilized features, such as mandibular dysmorphology. Ultimately, this approach provides a scalable, transparent, and high-performance solution for equitable prenatal diagnostics and phenotypic biomarker discovery.
Related
- Screening for rheumatoid arthritis-associated interstitial lung disease using lung ultrasound: a systematic review and quality appraisal of recent studies
- Predicting Cervical Cancer Screening Utilization Among Women of Reproductive Age in Ethiopia Using Supervised Machine Learning: A Nationally Representative 2024/25 EDHS Study
- An integrative meningioma surfaceome atlas and a transferable framework for CAR-T target nomination.
- ULTRASOUND OF QUADRICEPS FEMORIS AS A RULE-OUT TOOL FOR MALNUTRITION SCREENING IN HEART FAILURE CLINICS
- A Semi-Automated Interactive Annotation Framework for Intraoperative Neurosurgical Ultrasound: A Proof-of-Concept Study
BibTeX
@misc{zhu2026structureaware,
title = {{A Structure-Aware Self-Supervised Framework for Transferable Fetal Ultrasound Screening}},
author = {Y. Zhu and J. Wu and L. Gao and N. Zhang and L. Wu and H. Li and K. Chen and H. Luo and C. Sun and D. Lu and Q. Wu and H. Chen},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.21.26361576},
url = {https://www.medrxiv.org/content/10.64898/2026.09.21.26361576v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - A Structure-Aware Self-Supervised Framework for Transferable Fetal Ultrasound Screening
AU - Y. Zhu
AU - J. Wu
AU - L. Gao
AU - N. Zhang
AU - L. Wu
AU - H. Li
AU - K. Chen
AU - H. Luo
AU - C. Sun
AU - D. Lu
AU - Q. Wu
AU - H. Chen
PY - 2026
DA - 2026/09/21
PB - medRxiv
DO - 10.64898/2026.09.21.26361576
UR - https://www.medrxiv.org/content/10.64898/2026.09.21.26361576v1
AB - Accurate prenatal screening from ultrasound imaging is severely constrained by the extreme spatial heterogeneity of fetal structures and the data scarcity inherent to rare congenital anomalies. To overcome these computational bottlenecks, we present RadarMAE, a data-efficient self-supervised representation learning framework tailored for fetal ultrasound analysis. By employing a novel, structure-aware masking strategy, RadarMAE enables robust pre-training on highly limited clinical datasets. Benchmarking the framework on first-trimester screening for trisomy 21 (Down syndrome), our model achieved exceptional diagnostic performance on an internal cohort, with an area under the curve (AUC) of 0.99, 97% accuracy, 94% sensitivity, and 98% specificity. Crucially, through domain-adversarial adaptation, the framework demonstrates robust cross-institutional generalization under extreme source-target data imbalance, bounding performance degradation to[≤] 0.04 across all key metrics in external testing. Furthermore, by integrating spatial registration with explainable AI, we constructed the first population-level facial risk atlas for trisomy 21. This data-driven interpretability framework validates standard clinical markers while revealing the comparable diagnostic weight of underutilized features, such as mandibular dysmorphology. Ultimately, this approach provides a scalable, transparent, and high-performance solution for equitable prenatal diagnostics and phenotypic biomarker discovery.
ER -