Identification of key fetometric markers for predicting neonatal birth weight based on ultrasound fetometry using classical and machine-learning algorithms

BACKGROUND: Accurate prediction of neonatal birth weight using ultrasound fetometry remains a relevant challenge in modern perinatology. Despite numerous studies in this field, a gap persists in defining the optimal combination of fetometric parameters required for precise estimation of fetal birth weight, which substantially affects timely diagnosis of intrauterine growth restriction. AIM: To determine the contribution of individual fetometric parameters to predicting neonatal birth weight and to develop an optimal fetal weight estimation model based on comprehensive analysis of ultrasound measurement data. METHODS: This retrospective study used data from 5161 full-term neonates and 8022 fetal ultrasound examinations. The dataset included neonates with complete clinical information for the main fetometric parameters available. The observation period covered the interval from the first ultrasound examination to delivery. The primary endpoint was neonatal birth weight measured at birth. To assess parameter contributions, descriptive and correlation analyses, multiple linear regression, quantile regression with coefficients reported to three decimal places, and machine-learning methods (Random Forest and XGBoost) were applied. Additionally, principal component analysis was performed to derive a general fetal growth factor, enabling integration of multiple fetometric measurements into a single composite indicator that more reliably reflects overall fetal size and is more closely associated with birth weight. RESULTS: Abdominal circumference showed the strongest correlation with birth weight (r = 0.820), whereas femur length and head circumference demonstrated correlations of r = 0.620 and r = 0.540, respectively. Multiple regression including these three parameters yielded R2 = 0.730. Quantile regression showed that the coefficient for abdominal circumference increased at the upper weight quantile (β = 23.500 at τ = 0.900) compared with the median (β = 18.900 at τ = 0.500). Machine-learning methods confirmed the dominant role of abdominal circumference in birth weight prediction (feature importance 50%–55%). Principal component analysis indicated that the first principal component, interpreted as a general size measure, explained 78% of the variance and was highly correlated with birth weight (r = 0.850). CONCLUSION: These findings indicate that abdominal circumference is the most informative fetometric parameter for predicting fetal birth weight, whereas femur length provides complementary information by reflecting skeletal growth. Head circumference contributes minimally. The combined use of classical statistical approaches and machine-learning algorithms supports prioritizing accurate measurement of abdominal circumference and femur length when estimating fetal weight.