Predicting the Success of Oxytocin-Induced Labor Using TOCO Signals with Machine Learning Modeling

Oxytocin induction of labor, as an important pharmacological method of induction, is closely linked to uterine contraction activity. The aim of this study is to address the problem of limited information obtained on TOCO, the contraction pressure signal, and to assist obstetricians in making the correct decision to induce labor. The study introduces a robust method utilizing an database composed of TOCO records from 72 pregnant women at gestational ages between 37 and 42 weeks for feature set engineering design. A set of robust signal processing techniques were applied to the raw TOCO records to extract 11 features. The F-test feature selection algorithm was employed to remove features with low discriminative ability. Several simple yet powerful machine learning algorithms were trained using the feature set, and their performance was evaluated using real test data. The results was encouraging and the best results were obtained using SVM classifier with 0.8309 for Sensivity, 0.8115 for Specificity, 0.8299 for Accuracy, 0.7997 for F1 Score and 0.8212 for AUC. This study delved into deeper features within TOCO to characterize uterine contraction activity, and utilized machine learning methods to assist obstetricians in optimizing oxytocin prescriptions, offering new insights for research and application in related fields.