A geometric nonlinear return mapping framework for automated term–preterm delivery classification: a comparative analysis of TOCO and EHG signals

OBJECTIVE: Preterm delivery remains the leading cause of neonatal mortality worldwide. Early detection and accurate prediction of women at risk are critical for improving maternal and neonatal outcomes. This study aimed to diagnose preterm delivery using uterine contraction signals, providing a non-invasive approach to assist in clinical decision-making. Method: Electrohysterography (EHG) and tocodynamometry (TOCO) signals were recorded from a dataset of 300 samples, including both term and preterm deliveries. Linear and nonlinear signal processing techniques were employed to extract a wide range of features across time, frequency, and nonlinear domains. Features derived from return mapping and other methods were utilized as inputs to a Multi-Layer Perceptron (MLP) classifier for distinguishing between term and preterm deliveries. Result: The proposed method achieved exceptional classification performance, with an accuracy of 95.18±1.21%, sensitivity of 93.84±1.76%, and specificity of 95.61±1.09%. These results demonstrate the method's robustness and reliability. Comparative analysis with recent research highlighted the superior performance of the proposed approach. Conclusion: The findings underscore the potential of a computer-aided diagnostic system for preterm delivery detection using EHG and TOCO signals. This method provides a non-invasive, efficient, and accurate tool that could significantly reduce unnecessary hospitalizations and mitigate risks for high-risk pregnancies. By enabling timely intervention, the proposed approach holds promise for improving maternal and neonatal healthcare outcomes. .