Artificial Intelligence-Based Multimodal Ultrasound Model for Prediction of Spontaneous Preterm Birth: Development...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Obstetrics

Background:

• Spontaneous preterm birth (SPB) is a leading cause of neonatal mortality.
• Early prediction of SPB is crucial for timely intervention and improved outcomes.
• Current prediction methods have limitations in accuracy and accessibility.

Purpose of the Study:

• To develop and validate a multimodal artificial intelligence (AI) model for predicting spontaneous preterm birth (SPB).
• To integrate cervical 2D ultrasound and elastography images with ultrasound parameters for enhanced prediction accuracy.
• To assess the predictive performance of deep learning and machine learning models in identifying pregnancies at risk for SPB.

Main Methods:

• A prospective cohort study included 721 pregnant women screened at 20-24 weeks gestation.
• Deep learning models (DenseNet121) were trained on ultrasound images to generate scores.
• Multimodal AI models combined image-derived scores with ultrasound parameters for SPB prediction.

Main Results:

• The DenseNet121 deep learning model achieved high performance (AUC: 0.92/0.93) on ultrasound images.
• A neural network model integrating image scores and ultrasound parameters demonstrated strong predictive performance (AUC=0.85) in validation.
• Feature importance analysis highlighted the significant contributions of elastography and 2D ultrasound scores.

Conclusions:

• Multimodal AI models integrating ultrasound imaging and parameters show promising predictive capabilities for SPB in the second trimester.
• This AI-driven approach can aid in identifying high-risk pregnancies for timely intervention.
• Further validation is warranted to translate these findings into clinical practice for SPB prevention.