Prediction Of Labor Outcome Using Generative Artificial Intelligence

Abstract

Labor complications and emergency operative delivery are linked to sever fetal and maternal morbidity and mortality. It is crucial to predict the delivery mode before labor onset. We use Generative Artificial Intelligence to build a large balanced synthetic dataset having as seed a real dataset that contains 268 nulliparous patients. A Deep Learning model was trained on the synthetic set and applied on the real data to determine the type of birth weeks before the onset of labor. The University of Medicine and Pharmacy of Craiova approved an observational prospective cohort study, no. 96/28.09.2020, regarding data from 268 nulliparous patients. Weekly, we have evaluated sonographic and clinical low-risk nulliparous pregnant patients at term. We have used a Generative Adversarial Network to build a larger more balanced synthetic dataset, having as starting point the real data. We have further trained a Deep Learning model on the synthetic dataset. The trained model was applied on the real dataset to predict delivery mode weeks before the onset of labor. The Deep learning model trained on data generative by Artificial Intelligence was able to predict the delivery mode weeks before the onset of labor with an accuracy over 83%, when applied on real data. Other statistical measurements revealed a precision of 0.836, a recall of 0.836, and an F1-score of 0.837. Keywords: generative AI; deep learning; delivery mode; statistical assessement; transperineal ultrasound.

Ähnliche Arbeiten