Birth weight prediction using artificial intelligence-based placental assessment from macroscopic photo: a retrospective study

Background

This study aimed to predict newborn birth weight through multifactorial analysis of macroscopic placental images using artificial intelligence (AI).

Methods

We retrospectively reviewed the data of singleton pregnant women whose placentas were histopathologically examined at Asan Medical Center from January 2021 to December 2021. A total of 15 placental features were included in the machine learning using four regression analysis methods. Predictive performance matrics, including the mean absolute error (MAE), mean relative error (MRE), root mean square error (RMSE) and R² score, were calculated for each algorithm. The study population was divided into two groups according to placental pathological findings, which were subsequently compared.

Results

A total of 131 cases were included. For the machine learning analysis using all features, all 15 placental features were used. The second analysis using univariate predictive features was performed by excluding five features whose correlation values with birth weight were <0.5. Using AI, a predictive algorithm was developed, with a minimum MAE of 257.72 g, MRE of 0.15, RMSE of 338.42 g and a maximum R² score of 0.77. The group with non-pathologic placentas showed a higher overall predictive performance than that with pathological placentas. Subanalysis of the machine learning model, excluding the placental weight, showed similar trends.

Conclusions

The AI algorithm developed using machine learning for multifactorial analysis of the placenta can be used for neonatal birth weight prediction. This algorithm might assist in estimating fetal growth if it can potentially be adapted for prenatal ultrasonography by analyzing changes in placental measurements.