NIPT time point selection and abnormality determination of female fetuses via KMeans and random forest

Abstract

This paper explores non-invasive prenatal testing (NIPT), emphasizing the optimization of detection timing and the assessment of chromosomal abnormalities, utilizing data predominantly from pregnant women with high body mass index (BMI). The study begins by analyzing the relationship between Y-chromosome concentration in male fetuses and influencing factors, such as gestational age and BMI, through data preprocessing, visualization, Spearman correlation, and multiple regression modeling, which demonstrates significant linear associations. For male fetuses, it further applies K-means clustering based on BMI to group subjects, determining optimal detection timing via the 80th percentile of the first gestational week per group, with validity confirmed by ANOVA and silhouette scores. The research then extends to incorporate additional factors like height, age, and weight for refined grouping, while highlighting how measurement errors can distort rates and risk assessments, potentially complicating clinical decisions. Regarding female fetuses, the investigation develops a random forest model that integrates multiple features, including Z-scores of chromosomes 13, 18, 21, and X, GC content, read ratio, and BMI, to enhance the accuracy of aneuploidy detection beyond the limitations of single-feature approaches. Finally, the paper discusses the strengths and weaknesses of the proposed models, offering recommendations for their practical application and generalization.