Related Experiment Video
Updated: Sep 13, 2025
07:04
A Novel Use of Three-dimensional High-frequency Ultrasonography for Early Pregnancy Characterization in the Mouse
Published on: October 24, 2017
8.4K
Perinatal artificial intelligence in ultrasound (PAIR) study: predicting delivery timing
Neil Patel1, John O'Brien2, Robert Bunn3
1Department of Maternal Fetal Medicine, Ascension Sacred Heart, Pensacola, FL, USA.
Summary
Artificial intelligence (AI) can predict delivery timing from ultrasound images, aiding in preterm birth prediction. Continuous retraining of the AI model enhances its accuracy for predicting delivery dates and preterm births.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Healthcare
- Perinatology
Background:
- Accurate prediction of delivery timing is crucial for optimal maternal and neonatal care.
- Preterm birth remains a significant challenge, necessitating improved predictive tools.
Purpose of the Study:
- To evaluate an AI model's ability to predict days until delivery using only ultrasound images.
- To assess the impact of model retraining on improving prediction accuracy, especially for preterm births.
Main Methods:
- Developed and trained an AI model using de-identified ultrasound images from a large cohort.
- Utilized 79% of data for training and 21% for validation, with delivery outcomes blinded.
- Retrained the AI model multiple times with updated datasets and methodologies.
Main Results:
- The AI model demonstrated strong performance in predicting days to delivery (R² up to 0.92).
- Initial preterm birth prediction sensitivity was 39% and specificity 93%.
- Retraining improved overall prediction accuracy and preterm birth prediction capabilities.
Conclusions:
- AI can effectively predict delivery timing from ultrasound data alone.
- The AI shows potential for preterm birth prediction, though sensitivity is currently limited.
- Model retraining with advanced techniques can enhance predictive performance.
