关键词: 无创产前检测, Y染色体浓度, 体质量指数, 多目标优化, 集成学习, 女胎异常

Abstract: BACKGROUND: Non-invasive prenatal testing has become an important tool for prenatal screening of fetal chromosomal abnormalities, and the choice of testing timing directly affects the reliability of the results and the window for clinical intervention. Currently, a uniform testing strategy starting at 12 weeks of gestation is widely adopted in clinical practice. However, studies have indicated that gestational age and body mass index significantly influence the proportion of fetal cell-free DNA, suggesting that the current strategy lacks individualized basis. Furthermore, the diagnosis of abnormalities in female fetuses faces greater challenges due to the absence of Y-chromosome markers. 
OBJECTIVE: To explore strategies for selecting the optimal testing timing for the non-invasive prenatal testing based on individual maternal characteristics, analyze key factors influencing Y-chromosome concentration, and construct an intelligent diagnostic model for chromosomal abnormalities in female fetuses, thereby providing a basis for the precise application of non-invasive prenatal testing. 
METHODS: A retrospective analysis was performed on multiple non-invasive prenatal testing data from 267 pregnant women carrying male fetuses and 147 carrying female fetuses. Lasso regression was used to screen key factors influencing Y chromosome concentration, and a multiple linear regression model was constructed for validation. K-means++ clustering was applied to group pregnant women carrying male fetuses based on the body mass index, and survival analysis was used to plot the probability curves for the Y chromosome concentration that reached the standard. A risk quantification function was constructed, and a single-objective optimization problem was solved to minimize potential risk, obtaining the optimal testing timing for each group. The error delay factor was analyzed based on quartiles to assess the impact of detection errors. Further incorporating height, body mass, and age, a multi-objective programming model was constructed with the goals of minimizing risk and maximizing Y-chromosome concentration. An improved NSGA-II algorithm was used to solve the model, and the impact of different grouping strategies on the Pareto front was compared. For female fetus data, using the Z-values of the X chromosome and chromosomes 13, 18, and 21, GC content, read counts, and body mass index as input features, a Soft Voting ensemble learning method was used to fuse Gaussian-linear mixed kernel support vector machine, random forest, and logistic regression to construct a diagnostic model for chromosomal abnormalities in female fetuses. 
RESULTS AND CONCLUSION: (1) Lasso regression identified four key features: Z-value of chromosome 18, gestational age, body mass index, and age. (2) Multiple linear regression showed that the Z-value of chromosome 18 and body mass index were significantly negatively correlated with Y-chromosome concentration; gestational age was significantly positively correlated; and age had no significant effect. (3) K-means++ clustering divided body mass index into five groups with cluster centers of 28.96, 30.64, 32.34, 34.1, and 36.43 kg/m². (4) Single-objective optimization yielded optimal testing timings concentrated between 10.08 and 10.44 weeks, with the [33.2, 35.3) kg/m² group having the lowest risk and the [20, 29.8) kg/m² group having the highest risk. (5) Error analysis showed that detection errors could lead to a delay of 0.67-2.02 weeks. (6) In multi-objective optimization, the equal-frequency binning strategy performed best, with recommended timings distributed between 11.98 and 17.52 weeks. (7) In the diagnostic model for female fetuses, the Soft Voting ensemble model performed excellently with an accuracy of 0.931 3 and an area under the curve value of 0.896, superior to classifiers. These results indicate that body mass index and the Z-value of chromosome 18 are key negative factors influencing Y-chromosome concentration, while gestational age is a positive factor. Individualized NIPT timing selection strategy based on body mass index grouping can effectively reduce potential risks, equal-frequency binning performs best in multi-objective optimization, and the ensemble learning model can accurately identify chromosomal abnormalities in female fetuses.

Key words: non-invasive prenatal testing, Y chromosome concentration, body mass index, multi-objective optimization, integrated learning, abnormality in female fetuses