Dynamic Model for Early Detection of Preterm Labor

Abstract

Preterm labor is a major challenge in maternal and neonatal health due to its strong association with high rates of newborn morbidity and mortality. Early detection is critical, yet conventional static methods often fail to identify risks accurately and promptly. This study proposes the development of a dynamic, machine learning-based preterm birth risk prediction model using the Long Short-Term Memory (LSTM) architecture combined with Bayesian Updating. The model is designed to process multivariate time-series data from various clinical sources such as electronic health records (EHR), electrohysterography (EHG), cardiotocography (CTG), and vital signs collected longitudinally during pregnancy. Leveraging LSTM’s ability to capture temporal dependencies and Bayesian mechanisms for probabilistic adaptation, the model provides weekly, real-time, and adaptive risk estimates. Predictions are visualized through interactive graphs with risk categorization (low, medium, high) to support fast and accurate clinical interpretation. Importantly, this study used only dummy data entirely simulated for 500 virtual pregnancies to evaluate model functionality. No real patient data were involved. Results demonstrate that the system dynamically adjusts risk predictions as new data becomes available. This research contributes to the advancement of AI-based clinical decision support systems in maternal health. Future work will involve integration with real clinical datasets and external validation in hospital settings to improve accuracy and ensure the system’s applicability in real-world obstetric care.