A Machine Learning Approach for Fetal Chromosome Abnormality Identification Based on Multi-Feature Fusion

作者：Yi Duo Wang, Rouyi Fan, Yaning Yin, Tianyuan Liu · 发表于：Journal of big data and computing. · 年份：2025 · DOI：10.62517/jbdc.202501412 · 研究领域：Prenatal Screening and Diagnostics、Fetal and Pediatric Neurological Disorders、Genomic variations and chromosomal abnormalities

One of the prenatal screening techniques employed by NIPT (Non-Invasive Prenatal Testing) is high-throughput sequencing of foetal cell-free DNA isolated from maternal peripheral blood. Due to its high sensitivity, potential for early diagnosis, and non-invasive nature, it has emerged as a crucial method for identifying chromosomal abnormalities in fetuses. The Y chromosome's concentration is a vital reference point for quality assessment and anomaly analysis, necessitating its presence for the identification of male foetal abnormalities. However, because the Y chromosome lacks a definitive marker, the only techniques available for identifying chromosomal defects in female embryos are multidimensional feature fusion analysis and the X chromosome. For the diagnosis of female foetal anomalies, this methodology creates serious gaps and difficulties in feature utilisation, model stability, and interpretability. This study utilizes regional NIPT data to develop a multi-feature fusion and machine learning-based approach for identifying chromosomal abnormalities in female fetuses. SMOTE was employed to address the class imbalance brought on by the lack of aberrant samples in the training dataset. A feature set comprising Z-scores, GC content, and read duration metrics was methodically created. The LightGBM model was used to identify foetal chromosome abnormalities in females. Experimental results demonstrate that LightGBM outperformed Random Forest, XGBoost, CatBoost, and logistic re...