Related Experiment Video
Updated: Nov 5, 2025
07:34
Fetal Mouse Cardiovascular Imaging Using a High-frequency Ultrasound 30/45MHZ System
Published on: May 5, 2018
11.9K
An ensemble of neural networks provides expert-level prenatal detection of complex congenital heart disease
Rima Arnaout1,2,3,4,5, Lara Curran6,7, Yili Zhao8
1Division of Cardiology, Department of Medicine, University of California, San Francisco, San Francisco, CA, USA. rima.arnaout@ucsf.edu.
Nature Medicine
|May 15, 2021
Summary
This study developed an AI model to improve fetal congenital heart disease (CHD) detection using ultrasound images. The AI demonstrated high accuracy, offering a promising tool for early diagnosis of this common birth defect.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiology
Background:
- Congenital heart disease (CHD) is the most common birth defect, posing a significant diagnostic challenge.
- Current fetal screening ultrasound sensitivity for complex CHD is suboptimal, often as low as 30% despite guidelines.
- Accurate prenatal detection of CHD is crucial for timely intervention and improved outcomes.
Purpose of the Study:
- To develop and validate an AI-powered tool for enhancing the detection of fetal CHD.
- To improve the accuracy and consistency of identifying cardiac views and diagnosing complex CHD in fetuses.
- To assess the performance of AI models in calculating standard fetal cardiothoracic measurements.
Main Methods:
- Trained an ensemble of neural networks on over 100,000 fetal cardiac ultrasound images.
- Utilized segmentation models for calculating fetal cardiothoracic measurements.
- Validated the AI model on a large dataset of fetal surveys, including images of varying quality and from different institutions.
Main Results:
- Achieved an Area Under the Curve (AUC) of 0.99 in distinguishing normal from abnormal fetal hearts.
- Demonstrated high sensitivity (95%) and specificity (96%) for CHD detection.
- Model performance remained robust across diverse image qualities and hospital settings, with clinically relevant feature identification.
Conclusions:
- Ensemble learning models show significant potential to improve fetal CHD detection rates.
- The AI tool offers comparable sensitivity to clinicians and provides reliable cardiac measurements.
- This technology addresses a critical global diagnostic challenge in prenatal cardiology.
