Prediction of Delivery Mode from Fetal Heart Rate and Electronic Medical Records Using Machine Learning

Abstract

Cardiotocography (CTG) is the most common method for monitoring the fetus during the early stages of labor, making effective decisions based on CTG data is a challenge. Overinterpretation of CTG leads to unnecessary surgical intervention. This paper is conducted in two parts to identify the delivery mode: First, traditional fetal heart rate (FHR) and maternal electronic medical record (EMR) features are extracted from the available FHR and prenatal medical records respectively. Then, several prediction models are trained to classify the delivery mode for patients: vaginal or Cesarean Section (CS). A feature selection algorithm is employed on the features sets to remove unnecessary features to improve the performance of classifiers. The prediction models based on FHR and EMR features take into account accuracy, sensitivity, specificity, and area under the curve receiver operating characteristic (AUC) in the outcomes. K Nearest Neighbor (KNN) has the best accuracy (62.72%) and sensitivity (81.31%). We conclude that machine learning is proved with high value in predicting CS, and is a useful tool for reducing the false-positive rate and unnecessary operative interventions by employing FHR data and EMR information.