A Study on Optimal NIPT Timing Selection and Intelligent Fetal Abnormality Dectection Based on Machine Learning

Abstract

To address the challenges of uncertain testing time points and the difficulty in determining abnormalities due to individual variations in non-invasive prenatal testing (NIPT), this paper introduces an integrated machine learning framework that amalgamates prediction, optimization, and classification. The framework is designed to identify key factors influencing the quality of testing data and, based on these factors, to offer personalized optimal testing time points and highly accurate solutions for abnormality determination. Initially, the study utilizes an XG-Boost regression model to elucidate the complex nonlinear relationships among critical quality control indicators and factors such as gestational age at testing and maternal BMI. Building upon this, a novel approach that integrates weighted K-means clustering with nonlinear programming is employed to ascertain the optimal testing window for groups of pregnant women with varying physiological characteristics. Finally, to address the challenge of classifying rare abnormal samples, an adaptively weighted and optimized random forest model is developed. Experimental results indicate that this framework effectively handles the data nonlinearity and individual heterogeneity. The final classification model achieves a high recall rate of 85% and a precision of 97%, underscoring the advanced nature of the proposed method and its potential for clinical application.