Automated interpretation of fetal cardiac function evaluation from the echocardiogram

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Fetal Cardiology

Background:

• Prenatal assessment of fetal cardiac function is vital for neonatal outcomes.
• Manual echocardiographic measurements are time-consuming and prone to subjectivity.
• Limitations in current methods necessitate more objective and efficient assessment tools.

Purpose of the Study:

• To develop and validate a fully automated artificial intelligence (AI) workflow for fetal cardiac function parameter estimation.
• To compare the AI workflow's performance against manual measurements and existing automated tools.
• To establish a dynamic Z-score model for standardized interpretation of fetal cardiac function.

Main Methods:

• Developed a deep learning model for real-time cardiac structure detection and segmentation.
• Integrated quality control and geometric calculation into the AI workflow.
• Validated the AI using large internal and external datasets of normal and abnormal fetal echocardiograms.

Main Results:

• Achieved high segmentation accuracy (Dice >92%, IoU >85%) across all datasets.
• Demonstrated superior reproducibility and accuracy compared to manual and Fetal Heart Quantification (Fetal HQ) methods.
• Established a dynamic Z-score model for gestational age-referenced analysis.

Conclusions:

• The fully automated AI workflow offers accurate, efficient, and reproducible fetal cardiac function quantification.
• This AI system has the potential for standardized clinical application in prenatal care.
• The findings support the use of AI for improving the objectivity and reliability of fetal echocardiography.