Author(s)

Wendi Zhao, Junru Ren, Yulin Li

Corresponding Author

Wendi Zhao

Abstract

With the rapid development of non-invasive prenatal testing (NIPT) technology, how to reasonably select the detection time and accurately determine fetal abnormalities has become an important research topic in the field of prenatal diagnosis. Aiming at the problem of point-in-time optimization in NIPT detection, a multi-objective optimization model is proposed based on generalized average non-expansion operator (GAN) theory and machine learning methods. The results of model analysis showed that the gestational age was significantly positively correlated with the concentration of Y chromosome, while the BMI of pregnant women and age were significantly negatively correlated, and the model successfully revealed the key physiological indicators that affect the basis of NIPT detection. In the second step, in order to minimize the risk, a dynamic grouping and detection time optimization model based on BMI was constructed, and the model divided pregnant women into 5 BMI groups through GAN operator solving, and gave the corresponding optimal detection time point (11-12 weeks), so that the total risk value was reduced to 2851.0, and the balance between personalized detection and risk control was realized. In the third step, in order to integrate multi-dimensional physiological indicators for more refined grouping, an intelligent clustering model fusing random forest and GAN operator was proposed, and the model identified three groups of pregnant women with different characteristics, and verified that BMI (importance: 0.2536) and gestational age (importance: 0.1902) were the core basis for personalized grouping. The unified optimization framework proposed in this paper innovatively applies the GAN operator to the whole process of NIPT data analysis, and the model has the advantages of fast convergence, good stability and strong interpretation, which can provide scientific and accurate decision-making support for the selection of clinical NIPT detection time points.

Keywords

Generalized mean non-inflated operator; NIPT detection; Multiple nonlinear regression; Fetal abnormality determination; Target optimization