Automated prediction of the mode of birth delivery using geometric features of uterine contraction segments

Abstract

Pregnancy is a unique, complex process and it's hard to predict the mode of birth delivery due to the lack resources. This study aims to develop a clinical decision support system to predict birth delivery mode according to three categories: spontaneous vaginal delivery, induced vaginal delivery, and cesarean section. This methodology entails the automated extraction of uterine contraction segments from the electrohysterogram (EHG) signal based on the zero-crossing rate. Each segment is processed through a discrete Fourier transform to obtain the Fourier coefficients. Geometric features, including area, perimeter, circularity, variance, and bending energy were extracted from the boundary shape of these complex coefficients using the convex hull method. We found the women who experience spontaneous deliveries exhibit higher feature values and a lower circularity value compared to those who undergo cesarean sections or induced births. Based on these extracted parameters, the random forest (RF) model yielded promising results: reaching an accuracy above 90% in the classification between caesarean and spontaneous deliveries and spontaneous and induced vaginal deliveries and somewhat lower, around 70% between induced and cesarean. To conclude, the utilization of all proposed EHG parameters through machine learning can enhance obstetricians' ability to predict the mode of birth delivery.