Accurate birth weight prediction from fetal biometry using the Gompertz model.

Area of Science:

• Maternal-fetal medicine
• Biometry
• Growth modeling

Background:

• Accurate fetal growth monitoring and birth weight estimation are crucial clinical practices.
• Current methods rely on ultrasound biometry but lack a simple, predictive growth-model-based formula.
• Existing birth weight estimation depends on late-term ultrasound data.

Purpose of the Study:

• To model fetal biometry growth using the Gompertz model.
• To develop a machine learning model for predicting birth weight based on Gompertz parameters.
• To assess the model's predictive accuracy compared to existing methods.

Main Methods:

• Utilized ultrasound biometry measurements from the "Seethapathy cohort" (774 pregnant women).
• Applied the Gompertz model, a constrained growth model, to analyze fetal biometry.
• Trained a machine learning model on inferred Gompertz parameters to predict birth weight (BW).

Main Results:

• The Gompertz model demonstrated a strong fit for fetal biometry growth.
• Two Gompertz parameters appeared universal, while a third captured individual fetal scale.
• The ML model predicted birth weight with an 8% error, outperforming late-term ultrasound methods, and achieved 8.4% error on an independent cohort.

Conclusions:

• The Gompertz model effectively fits fetal biometry growth and enables birth weight estimation without late-term ultrasounds.
• The model's single scale parameter ($$) explains most individual variation, suggesting its utility for future growth standards.
• This approach offers significant clinical value for early and accurate fetal growth assessment.