A new algorithm for improving fetal weight estimation from ultrasound data at term.

Area of Science:

• Medical Imaging
• Biostatistics
• Obstetrics

Background:

• Accurate fetal birth weight estimation is crucial for managing pregnancy and delivery.
• Traditional ultrasonographic measurements have limitations in predicting birth weight, particularly for macrosomic infants.

Purpose of the Study:

• To improve the accuracy of fetal birth weight estimation using traditional ultrasonographic measurements at term.
• To evaluate a novel regression method, isotonic regression, against existing algorithms for birth weight prediction.

Main Methods:

• Retrospective review of delivery records from two hospitals, identifying 223 cases of macrosomic infants and 212 controls.
• Utilized isotonic regression to develop a birth weight prediction function, ensuring monotonic increase with input variables.
• Compared the performance of the isotonic regression model with linear regression models.

Main Results:

• The developed biometric algorithms had a mean absolute error of 312-344 g at a 95% confidence interval.
• Isotonic regression significantly improved birth weight prediction accuracy compared to linear regression models.
• Suspicion of macrosomia was based on clinical factors and ultrasound estimations exceeding 4,000 g.

Conclusions:

• The isotonic regression method offers improved accuracy for fetal birth weight estimation.
• This method enhances the ultrasound detection of macrosomia, aiding clinical decision-making.