Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
AI Achieves 96% Accuracy in Detecting Fetal Brain Abnormalities on Ultrasound
A two-stage deep learning system successfully detects fetal brain abnormalities in routine second-trimester ultrasound with 96% accuracy, offering potential to democratize prenatal screening globally.
Background
Detecting fetal brain abnormalities early is critical for prenatal planning, yet specialized expertise may be unavailable in many settings. Researchers developed an AI system using 319 ultrasound images from nine international fetal medicine centers to enable standardized screening between 19-24 weeks of gestation.
Key Findings
- Two-stage pipeline achieved 96% diagnostic accuracy (AUC 0.96) with 87% sensitivity and 91% specificity
- Object detection model identified six brain regions with 93% mean average precision
- Processing time: 50 milliseconds per image
- Negative predictive value of 99.99% at population screening prevalence
- Consistent performance between cross-validation and independent testing
Why It Matters
This screening-support tool could enhance detection regardless of operator experience and assist resource-limited settings with limited access to specialty-trained sonographers, while prioritizing complex cases for expert review.
Limitations
The study used standardized images from specialized centers, which may not reflect broader screening populations with variable image quality. Further validation in routine clinical settings would strengthen real-world applicability.
Original paper: Development of an Integrated Deep Learning Approach for Detecting Fetal Brain Abnormalities in Routine Second Trimester Ultrasound Scan: A Multicenter Study. — Radiology. Artificial intelligence. 10.1148/ryai.250737
