Research on Fetal Chromosome Abnormality Risk in NIPT Based on Multivariate Nonlinear Modeling

Abstract

This study addresses key challenges in fetal chromosomal abnormality risk assessment for non-invasive prenatal testing (NIPT) by establishing an integrated analytical framework combining multivariate nonlinear modeling and machine learning. The innovative contributions of this research are highlighted in several aspects: For male fetuses, a novel trivariate cubic polynomial regression model was developed to precisely quantify the complex nonlinear relationships among Y-chromosome concentration, gestational age, and maternal BMI, with a joint optimization algorithm determining personalized testing time windows—overcoming limitations of existing methods in dynamic prediction and timing selection. For female fetuses, an anomaly identification model integrating ADASYN oversampling and decision trees was constructed, effectively addressing data imbalance through multi-dimensional feature selection and significantly enhancing classification performance in the absence of Y-chromosome reference. Overall, the proposed gender-specific modeling strategy provides a more precise and personalized screening solution for clinical practice. Validation on a clinical dataset of 1,056 male and 605 female fetal cases demonstrates that this framework significantly improves risk prediction accuracy and offers an effective tool for optimizing prenatal screening strategies.