UTILIZING DEEP NEURAL NETWORKS FOR PREDICTING POSTPARTUM HEMORRHAGE RISK IN ATHLETE AND NON-ATHLETE PREGNANT WOMEN: A FACTOR ANALYSIS APPROACH

Abstract

Background: Postpartum hemorrhage (PPH) is a critical obstetric complication and a leading cause of maternal mortality. Its rapid onset can lead to severe consequences, such as hemorrhagic shock, particularly in high-risk groups like athletes, who may have unique physiological profiles. Understanding and predicting the risk of PPH in these individuals are crucial for implementing timely and effective interventions. Objective: This study leverages a deep neural network (DNN) to develop a risk prediction model for PPH that is particularly tailored to the needs of pregnant athletes, incorporating factors unique to this group. Methods: Utilizing the Deep Belief Network (DBN) framework, this research constructs a predictive model for PPH by inputting specific risk factors and outputting the corresponding risk levels. To enhance the model's accuracy and mitigate the issues associated with random initializations of network weights, an Improved Particle Swarm Optimization (IPSO) algorithm is employed. This approach optimizes the initial parameters of the DBN model, refining both the inertia weight and learning factors to enhance predictive performance. Results: By integrating the IPSO method for optimizing basic DBN network parameters, the developed Improved DBN (IDBN) algorithm effectively predicts the risk of PPH. This model is particularly adapted to predict PPH in pregnant athletes by considering specific athletic-related physiological and health indicators that may influence hemorrhage risks. Conclusion: The application of an IDBN algorithm in the risk assessment of PPH among pregnant athletes represents a significant advancement in prenatal care within sports medicine. This model offers a promising tool for healthcare providers to predict and manage the risks associated with PPH, ensuring better health outcomes for athletic mothers. Further research will focus on validating this model across broader athletic populations and integrating additional athlete-specific variables to enhance its applicability and accuracy.