Potential of Artificial Intelligence to Estimate the Birthweight of Newborns Based on Ultrasound Data – Comparison between AI and Hadlock Formula
Year: 2026
Type: article
Abstract: Objective: When managing labor and delivery, an accurate estimation of fetal weight is essential. The aim of this study was to compare if a prediction model developed by Artificial intelligence (AI) can predict fetal birthweight or macrosomia with greater accuracy than the Hadlock formula, which is commonly used to estimate fetal weight at the time of ultrasound. Methods: 17691 ultrasonography exams from 2013-2022 were included in this study. The dataset was split. The first 70% of the data were used for training the models, the results were validated on 15% of the data and the remaining 15% were designated as the testing dataset. Several machine learning methods were examined. Three AI models were created: the regression model to predict the birthweight and two classification models to predict a birthweight of over 4000 g and over 4500 g respectively. Results: In the regression model the best prediction accuracy was attained with random forest regression (RMSE 315.071, SD 10.055). These results were better compared to the Hadlock formula (RMSE of 323.14). In the classification model with a threshold of 4000 g the best result was achieved with deep neural network (AUC 0.889, SD 0.008). Similarly in the second classification model (threshold 4500 g) the best results were observed with random forest classification (AUC 0.894, SD 0.054) and with deep neural network (AUC 0.910, SD 0.032). Conclusion: AI-supported models such as random forest and deep neural network facilitate accurate prediction of macrosomia and fetal weight at birth by considering a variety of parameters.
Source: Geburtshilfe und Frauenheilkunde
DOI: 10.1055/a-2915-6606
Authors: Natalia Anna Kaufmann, Peter Oppelt, Jens Meier, Teresa Eichinger, Sabine Enengl, Julia Lastinger, Elisabeth Reiter
Language: German
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7204080669