AI learning for pediatric right ventricular assessment: development and validation across multiple centers.

Area of Science:

• Pediatric Cardiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Right ventricular (RV) dysfunction is a complex issue in pediatric heart disease, affecting ~1% of children globally.
• Accurate RV assessment in children is challenging due to anatomical variability and irregular geometry.
• Existing methods for RV assessment in pediatric patients lack consistency and efficiency.

Purpose of the Study:

• To develop and validate a deep learning framework for automated RV functional assessment in pediatric patients.
• To enable real-time, expert-level quantification of pediatric ventricular function.
• To improve diagnostic consistency and support earlier interventions for pediatric heart disease.

Main Methods:

• Utilized a dataset of 24,984 echocardiograms from 3993 children across North America and Asia.
• Developed a video-based deep learning framework using a U²-Net architecture for automated ventricular segmentation and functional assessment.
• Evaluated the model's performance in frame-level segmentation, fractional area change (FAC) estimation, RV disease classification, and left ventricular ejection fraction (LV EF) prediction.

Main Results:

• Achieved high segmentation accuracy (Dice = 0.86 [A4C], 0.88 [PSAX]).
• Demonstrated strong RV disease classification performance (AUC = 0.95 U.S., 0.97 Asia).
• Outperformed previous methods in LV EF prediction across different cohorts.

Conclusions:

• The validated deep learning framework provides expert-level, real-time quantification of pediatric ventricular function.
• This automated approach enhances diagnostic consistency and reduces manual workload.
• The tool has the potential to support earlier intervention for children with heart disease, especially in resource-limited settings.