Method and system for determining NIPT optimal time point of male fetus and judging abnormity of female fetus

Abstract

The invention provides a male fetus NIPT optimal time point determination and female fetus abnormity judgment method and system, and relates to the technical field of biomedicine. Preprocessing the NIPT detection data of the pregnant woman; extracting male fetus detection data in the pre-processed data, dividing the number of gestational weeks into a plurality of stages, respectively calculating the correlation between the Y chromosome concentration and the number of gestational weeks and between the Y chromosome concentration and the BMI of the pregnant woman in each stage, and constructing a relation model to obtain a correlation result between the Y chromosome concentration and the number of gestational weeks and between the Y chromosome concentration and the BMI of the pregnant woman; dividing the BMI of the male fetus and the pregnant woman into a plurality of groups, obtaining detection related indexes of each group under different gestational weeks through a survival analysis method by combining an association result and fusing multiple factors, constructing a multi-target planning model, and solving an optimal NIPT time point of each group of pregnant women; and extracting female fetus detection data in the preprocessed data, obtaining influence features through feature statistical analysis and significance screening, constructing a logistic regression model, and judging whether the female fetus is abnormal or not according to a model output result. The detection accuracy and the clinical auxiliary intervention timeliness are realized.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, and in particular to a method and system for determining the optimal timing of NIPT in male fetuses and for identifying abnormalities in female fetuses. Background Technology

[0002] As an auxiliary means of health assessment, it analyzes fetal cell-free DNA fragments in maternal plasma to help determine chromosomal abnormalities. Its detection effectiveness is closely related to the timing of the test and needs to be considered in conjunction with factors such as gestational age and maternal BMI.

[0003] In current practice, the use of uniform testing time points and simple BMI grouping can easily lead to problems such as insufficient Y chromosome concentration compliance rates in male fetuses and biases in the auxiliary determination of chromosomal abnormalities in female fetuses, thus affecting the reliability of auxiliary assessments.

[0004] Existing technologies often overlook the synergistic effects of multiple factors and interference from detection errors. BMI grouping lacks scientific basis, and the optimal timing solution does not take into account the achievement rate, intervention window, and risk control. Furthermore, the characteristics of auxiliary judgment of female fetal abnormalities are not specific enough, making it difficult to adapt to the personalized auxiliary testing needs of pregnant women with different physical conditions. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a method and system for determining the optimal timing of NIPT in male fetuses and for identifying abnormalities in female fetuses, which balances detection accuracy with timely clinical intervention, thereby reducing maternal and infant health risks.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for determining the optimal timing of NIPT in male fetuses and for assessing abnormalities in female fetuses, comprising: Preprocessing of NIPT test data from pregnant women; the NIPT test data from pregnant women includes gestational age, pregnant woman's BMI, and Y chromosome concentration; Male fetal detection data were extracted from the preprocessed data. The gestational age was divided into multiple stages. The Spearman correlation between Y chromosome concentration and gestational age and maternal BMI was calculated for each stage. A relationship model was constructed to obtain the association results between Y chromosome concentration and gestational age and maternal BMI. The BMI of pregnant women with male fetuses was divided into multiple groups. Based on the correlation results and by integrating multiple factors, the relevant indicators of each group at different gestational weeks were obtained by survival analysis. A multi-objective programming model was constructed to solve the optimal NIPT time point for each group of pregnant women. Female fetal detection data were extracted from the preprocessed data. Influencing features were obtained through feature statistical analysis and significance screening. A logistic regression model was constructed, and the model output results were used to determine whether the female fetus was abnormal.

[0007] Secondly, the present invention provides a system for determining the optimal timing of NIPT in male fetuses and for assessing abnormalities in female fetuses, comprising: The data processing module is used to preprocess the NIPT test data of pregnant women; the NIPT test data of pregnant women includes gestational age, pregnant woman's BMI and Y chromosome concentration; The male fetus data analysis module is used to extract male fetus detection data from the preprocessed data, divide the gestational age into multiple stages, calculate the Spearman correlation between Y chromosome concentration and gestational age and maternal BMI in each stage, and construct a relationship model to obtain the correlation results between Y chromosome concentration and gestational age and maternal BMI. The module for determining the timing of male fetuses is used to divide pregnant women with male fetuses into multiple groups based on their BMI. Combining the correlation results and integrating multiple factors, the module obtains relevant indicators for each group at different gestational weeks through survival analysis, constructs a multi-objective programming model, and solves the optimal NIPT timing for each group of pregnant women. The female fetus abnormality determination module is used to extract female fetus detection data from preprocessed data, obtain influencing features through feature statistical analysis and significance screening, construct a logistic regression model, and determine whether the female fetus is abnormal based on the model output results.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for determining the optimal timing of NIPT in male fetuses and for judging abnormalities in female fetuses as described in the first aspect.

[0009] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the method for determining the optimal timing of NIPT for male fetuses and for judging abnormalities in female fetuses as described in the first aspect.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention first preprocesses NIPT test data by imputing missing values ​​and identifying outliers to ensure data reliability and lay a high-quality foundation for subsequent analysis. For male fetuses, Spearman correlation is calculated in stages to accurately capture the differences in the association between gestational age and Y chromosome concentration. Then, a nonlinear relationship model is constructed using a generalized additive model to improve the accuracy of the association results. By combining BMI grouping and survival analysis to obtain detection indicators, a multi-objective programming model is constructed to solve for the optimal timing, ensuring both the target Y chromosome concentration rate and avoiding missing the intervention window. For female fetuses, feature statistical analysis and significance screening are used to focus on key influencing factors, and a logistic regression model is constructed to achieve accurate determination of chromosomal abnormalities in female fetuses. The overall solution takes into account the differences in testing between male and female fetuses, providing doctors with objective data support and decision-making reference.

[0011] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0013] Figure 1 This is a flowchart illustrating the main steps of a method for determining the optimal timing of NIPT in male fetuses and for identifying abnormalities in female fetuses, as provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the average Y chromosome concentration variation curve at different gestational weeks provided in an embodiment of the present invention; Figure 3 Figure 1 shows the correlation between gestational age and Y chromosome concentration at four stages provided in this embodiment of the invention; Figure 2(a) shows the correlation between gestational age 11-15 and Y chromosome concentration; Figure 3(b) shows the correlation between gestational age 16-18 and Y chromosome concentration; Figure 4(c) shows the correlation between gestational age 19-22 and Y chromosome concentration; Figure 5(d) shows the correlation between gestational age 23-25 ​​and Y chromosome concentration. Figure 4 Figure (a) is a schematic diagram of the correlation between pregnant women's BMI and Y chromosome concentration, provided in an embodiment of the present invention; Figure (b) is a schematic diagram of the correlation between pregnant women's BMI and Y chromosome concentration; Figure 5 A heatmap of the correlation coefficients of the remaining indicators provided in an embodiment of the present invention; Figure 6 Figure (a) shows the relationship between gestational age, maternal BMI, and Y chromosome concentration, as provided in this embodiment of the invention. Figure (b) shows the relationship between gestational age and Y chromosome concentration. Figure 7 This is a schematic diagram of the cumulative achievement rate curves for each BMI group provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time point for BMI group 1 in the first stage of the optimal NIPT timing solution provided in the embodiments of the present invention. Figure 9 This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time point for BMI group 2 in the first stage of the optimal NIPT timing solution provided in this embodiment of the invention. Figure 10This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time for BMI group 3 in the first stage of the optimal NIPT timing solution provided in the embodiments of the present invention. Figure 11 This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time for BMI group 4 in the first stage of the optimal NIPT timing solution provided in the embodiments of the present invention. Figure 12 This is a schematic diagram illustrating the optimal number of groups for the elbow method provided in an embodiment of the present invention; Figure 13 This is a schematic diagram of the cumulative achievement rate curves for each BMI group provided in an embodiment of the present invention; Figure 14 This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time point in the second stage of NIPT timing solution provided in this embodiment of the invention. Figure 15 This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time point for BMI group 2 in the second stage of NIPT timing solution provided in this embodiment of the invention. Figure 16 This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time point for BMI group 4 in the second stage of NIPT timing solution provided in this embodiment of the invention. Figure 17 This is a schematic diagram of the dynamic risk of NIPT compliance and the survival model curve of the optimal detection time point for BMI group 5 in the second stage of NIPT timing solution provided in this embodiment of the invention. Figure 18 This is a schematic diagram of the confusion matrix between normal and abnormal data provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Example 1 like Figure 1 As shown, this embodiment discloses a method for determining the optimal timing of NIPT in male fetuses and for judging abnormalities in female fetuses, including the following steps: S1: Preprocessing the NIPT test data of pregnant women; the NIPT test data of pregnant women includes gestational age, pregnant woman's BMI and Y chromosome concentration; S2: Extract male fetal detection data from the preprocessed data, divide the gestational age into multiple stages, calculate the Spearman correlation between Y chromosome concentration and gestational age and maternal BMI in each stage, and construct a relationship model to obtain the correlation results between Y chromosome concentration and gestational age and maternal BMI. S3: Divide the BMI of pregnant women with male fetuses into multiple groups, combine the aforementioned correlation results, and integrate multiple factors. Use survival analysis to obtain the relevant indicators of each group at different gestational weeks, construct a multi-objective programming model, and solve the optimal NIPT time point for each group of pregnant women. S4: Extract female fetal detection data from the preprocessed data, obtain influencing features through feature statistical analysis and significance screening, construct a logistic regression model, and determine whether the female fetus is abnormal based on the model output results.

[0016] Next, combined Figure 1 This embodiment provides a detailed description of a method for determining the optimal timing of NIPT in male fetuses and for identifying abnormalities in female fetuses.

[0017] First, in order to accurately determine the optimal timing for NIPT testing and scientifically assess fetal abnormalities, modeling analysis was conducted based on data characteristics and clinical needs, and the following assumptions were made: 1. Assume that the pregnant woman in the data is always in a healthy state.

[0018] Reason: It is believed that aneuploidy can only originate from the fetus, and the focus is on analyzing the health status of the fetus.

[0019] 2. It is assumed that within the same mother, the same fetus at the same gestational age can be considered as the same sample.

[0020] Reason: It is believed that the fetus's physical condition will gradually improve with the gestational age.

[0021] 3. Assume that the height and weight data of the pregnant woman given in the appendix (NIPT test data of pregnant women) are reliable and accurate.

[0022] Reason: To avoid considering too many factors that may affect the data and reduce contradictions between indicators.

[0023] 4. Assume that the same mother will not have multiple pregnancies within the statistical period.

[0024] Reason: It is believed that the changes in various indicators are not caused by multiple fetuses.

[0025] Non-invasive prenatal genetic testing (NIPT) currently uses a standardized protocol: testing is conducted between 10 and 15 weeks of gestation, with accuracy determined by a Y chromosome abnormality of ≥4% for male fetuses and a normal X chromosome for female fetuses; abnormalities are identified by the concentration of chromosomes 13, 21, and 18, with low risk in early detection (within 12 weeks) and high risk in mid-to-late stages (after 13 weeks). However, significant individual differences exist among pregnant women in terms of pregnancy status, age, and BMI, and a standardized protocol can easily affect the accuracy of the test.

[0026] Therefore, this embodiment focuses on male fetuses. First, it analyzes the correlation between Y chromosome concentration and indicators such as gestational age and BMI of pregnant women, screens key factors to construct a joint model, and determines the optimal timing for personalized NIPT testing for pregnant women with different individual characteristics, so as to improve the accuracy of testing and the timeliness of intervention.

[0027] In S1, firstly, NIPT test data of pregnant women are collected and preprocessed.

[0028] This embodiment selected 21 indicators as NIPT test data for pregnant women, including age, height, weight, gestational age, maternal BMI, number of original reads, proportion of reads aligned to the reference genome, proportion of duplicate reads, number of uniquely aligned reads, GC content, Z-value of chromosome 13, Z-value of chromosome 18, Z-value of chromosome 21, Z-value of X chromosome, Z-value of Y chromosome, Y chromosome concentration, X chromosome concentration, GC content of chromosome 13, GC content of chromosome 18, GC content of chromosome 21, and the proportion of reads filtered out. These data were used to test and analyze the correlation between fetal Y chromosome concentration and maternal gestational age and BMI.

[0029] The preprocessing methods can be selected by those skilled in the art according to the actual situation, including missing value handling, cross-validation, numerical transformation and box plot analysis, etc. The specific methods are not limited in this embodiment.

[0030] In S2, further, for male fetuses, the optimal timing for NIPT detection is determined by exploring the correlation between Y chromosome concentration and other indicators.

[0031] First, before analyzing the correlation between the concentration of the Y chromosome in male fetuses and various indicators, the distribution of the indicator data was verified by Shapiro-Wilk. Since most of them did not meet the normal distribution, Spearman correlation coefficient was used to determine their correlation with the concentration of the Y chromosome.

[0032] (1) Correlation between gestational age and Y chromosome concentration To investigate the impact of gestational age on Y chromosome concentration, this embodiment used 988 processed male fetal detection data as a sample. The gestational age data from 10 to 25 weeks were discretized (step size 1 week). Non-discrete data points were rounded to the corresponding nodes. The average Y chromosome concentration at each node was calculated, and a change curve was plotted. Figure 2 As shown. Since the attached sample lacks data for week 10, weeks 11-25 are divided into stages based on the curve trend.

[0033] The curve shows that Y chromosome concentration exhibits significant phased changes with gestational age: fluctuating and rising from weeks 11-15, dropping to a minimum at week 16-18 before rapidly rebounding, stabilizing from weeks 19-22, and initially rising then falling from weeks 23-25, but still higher than weeks 19-22, continuing an upward trend. Based on this, gestational age is divided into four phases, and the Spearman correlation coefficient is calculated using Python. Figure 3 As shown.

[0034] The results showed that the correlation coefficients at all four stages were <0.3, indicating a weak positive correlation. Gestational age had a limited effect on promoting Y chromosome concentration. The correlation coefficient was highest at 11-15 weeks, with the most significant promoting effect. The correlation decreased sequentially at 16-18 weeks and 19-22 weeks, with the weakest correlation at 19-22 weeks. The correlation rebounded somewhat at 23-25 ​​weeks.

[0035] In summary, the impact of Y chromosome concentration varies at different gestational weeks, possibly related to fetal development and maternal metabolic dynamics. This indicates that gestational age is not the sole criterion for determining the optimal timing for NIPT; a comprehensive assessment system combining key indicators such as maternal BMI and age is necessary to improve the accuracy of testing timing selection.

[0036] (2) Correlation between pregnant women's BMI and Y chromosome concentration The Spearman correlation coefficient between pregnant women's BMI and Y chromosome concentration was calculated using Python, and a scatter plot was generated to visually illustrate the correlation trend between the two. Specific results are shown below. Figure 4 As shown.

[0037] Depend on Figure 4 The Spearman correlation coefficient is -0.143. ( The significance level (a commonly used marker in statistics, representing extremely high statistical reliability) indicates a highly significant weak negative correlation between maternal BMI and Y chromosome concentration. That is, as maternal BMI increases, Y chromosome concentration tends to decrease. Furthermore, while the blue data points in the scatter plot are relatively dispersed, the overall distribution, along with the red fitted dashed line, shows a slight upward and downward tilt, visually demonstrating the negative association of a slight decrease in Y chromosome concentration with increasing BMI. At the same time, the high dispersion of the scatter plot also reflects that BMI has limited explanatory power for Y chromosome concentration, only capturing a weak overall trend; the actual degree to which BMI affects Y chromosome concentration varies significantly among individuals.

[0038] This relatively weak correlation indicates that BMI is not the primary determinant of Y chromosome concentration. Y chromosome concentration cannot be inferred solely from BMI. Multiple factors, including the pregnant woman's underlying medical conditions, diet, and biological characteristics, should be considered to further elucidate the regulatory mechanism of Y chromosome concentration.

[0039] (3) Correlation between other indicators and Y chromosome concentration As can be seen from the above, determining the optimal timing for NIPT requires focusing on the relationship between gestational age and Y chromosome concentration. Achieving a Y chromosome concentration of ≥4% is crucial for accurate detection of male fetuses, and this concentration varies with gestational age. Therefore, it is necessary to first clarify the relationship between the two to establish a baseline timeframe. Simultaneously, there is a significant weak negative correlation between maternal BMI and Y chromosome concentration, and the difficulty of achieving the target varies significantly among different BMI groups. Introducing BMI allows for group-specific analysis, avoiding the limitations of a uniform timing point.

[0040] However, relying solely on gestational age and BMI has limitations: both have limited explanatory power for Y chromosome concentration, and individual differences are also influenced by factors such as age, height, and sequencing quality indicators (e.g., GC content, read ratio). Therefore, it is necessary to further incorporate these key indicators and improve the accuracy and individual suitability of optimal timing determination through multi-factor collaborative modeling.

[0041] Using Python, the Spearman correlation coefficients were calculated among 19 indicators, including age, height, weight, number of original reads, proportion of reads aligned to a reference genome, proportion of duplicate reads, number of uniquely aligned reads, GC content, Z-scores and GC content of chromosomes 13, 18, and 21, and Z-scores and concentrations of chromosomes X and Y. Figure 5 As shown in the figure. The shades of color correspond to the magnitude of the correlation coefficient between the two indicators; the darker the red, the stronger the positive correlation, and the darker the blue, the stronger the negative correlation.

[0042] Depend on Figure 5 It can be seen that the correlations among the various indicators are significantly different. There is a highly significant positive correlation between X chromosome concentration and Y chromosome concentration, with a correlation coefficient reaching 0.466. This indicates that fluctuations in X chromosome concentration and changes in Y chromosome concentration show a more pronounced synchronous trend, with a relatively higher degree of linear correlation between the two. Furthermore, GC content-related indicators exhibit a moderate positive correlation with sequencing alignment indicators such as raw reads, the proportion of duplicate reads, and the number of uniquely aligned reads. However, their linear correlation with core detection indicators such as chromosome Z-score and cell-free DNA concentration is limited, suggesting that the direct linear impact of basic sequencing quality parameters on chromosome abnormality detection results is not significant.

[0043] There is a moderate positive correlation between basic physiological indicators such as age, height, and weight, and the absolute values ​​of the correlation coefficients with sequencing alignment indicators such as raw readings and the proportion of alignments on the reference genome, as well as core chromosome detection indicators such as Z-values ​​of each chromosome and cell-free DNA concentration, are mostly close to 0, indicating no significant linear association overall.

[0044] Clarifying the differences in the correlation between these indicators can provide a basis for selecting key indicators for the optimal timing of NIPT (prioritizing strongly correlated indicators such as X chromosome concentration and weakening the individual role of basic physiological indicators).

[0045] Furthermore, to clarify the relationship between gestational age, maternal BMI, and Y chromosome concentration, and thus select a suitable statistical model, this embodiment first performs a visual analysis of the relationship between gestational age, maternal BMI, and Y chromosome concentration. By overlaying scatter plots with trend fitting, the potential association characteristics between variables are visually explored. The results are as follows: Figure 6 As shown in the figure, preliminary observations based on the scatter plot indicate that there is no obvious linear correlation between gestational age and Y chromosome concentration. Furthermore, similar patterns appear in the scatter plot of pregnant women's BMI and Y chromosome concentration, both in the scatter plot and the fitted line.

[0046] It is known that Y chromosome concentration, as a core observed variable, may have a non-linear relationship with gestational age and maternal BMI. To flexibly fit the complex non-linear relationship between variables, this embodiment selects a generalized additive model (GAM) to construct a regression model to solve the specific relationship between Y chromosome concentration and gestational age and maternal BMI.

[0047] Before constructing the GAM model, the feature set containing gestational age, maternal BMI and its squared terms was centered by collinearity test. The variance inflation factor (VIF) and feature correlation matrix were used to evaluate the collinearity strength and correlation direction. It was found that the feature set had no serious risk of collinearity and could be fully included in the subsequent model construction, providing a reliable feature basis for the model's parameter estimation and nonlinear correlation fitting.

[0048] Accordingly, the specific construction of the Generalized Additive Model (GAM) is as follows: (1) Feature engineering and data preparation A nonlinear enhanced feature matrix is ​​constructed using relevant indicators such as gestational age and BMI of pregnant women from the preprocessed data, and this matrix is ​​used as the input basis for the GAM model to capture the complex relationship between gestational age, BMI and Y chromosome concentration.

[0049] (2) Enhancement of nonlinear characteristics To enhance the GAM model's ability to capture complex relationships, an enhanced feature matrix is ​​constructed based on the original features and derived features. Specifically, retain the original gestational age of the pregnant woman ( ) and pregnant women ( ), and construct derived features: the square of the gestational age of the pregnant woman ( The quadratic nonlinear trend of gestational age and Y chromosome concentration in pregnant women is captured by the squared term, as well as the pregnant woman's... square( The quadratic nonlinear relationship between BMI and Y chromosome concentration is captured by using the squared term. Nonlinear trends are captured by introducing derived features.

[0050] in, Indicates the first One sample, ; Indicates the first The gestational weeks of the pregnant women in the sample. The unit is week; Indicates the first BMI of pregnant women in a sample The unit is kg / m³ 2 .

[0051] Next, construct the enhanced feature matrix. :

[0052] in, For dimension The matrix represents the number of data samples (988) and the number of features (4). Represents the enhanced feature matrix The Middle Line number The elements of the column provide comprehensive feature information for the subsequent input of the GAM model. Indicates the first One enhanced feature, ; (3) Model form and smoothing function Because the target variable is the Actual Y chromosome concentration in each sample Since the variables are continuous and follow a normal distribution, the linear GAM model is chosen, with its link function being the identity function. This differs from traditional linear regression's "linear coefficients..." The "features" take different forms. The essence of the linear GAM model is "intercept term + linear combination of multiple nonlinear smoothing functions". The model form is more flexible and can fit complex nonlinear relationships.

[0053] Linear GAM for the first Predicted Y chromosome concentration for each sample The formula is:

[0054] in, Indicates the first Predicted Y chromosome concentration for each sample The model intercept term represents the baseline predicted Y chromosome concentration when all enhancement features are set to 0. Indicates the first One enhanced feature The smoothing function is used to characterize the nonlinear relationship between the feature and the target variable.

[0055] To ensure the flexibility and stability of the smoothing function, a natural cubic spline is used for construction. Its mathematical essence is to transform features The range of values ​​is divided into several intervals, and a cubic polynomial is used to fit each interval. The intervals are connected in a way that ensures that the function values ​​are continuous, the first derivative is continuous, and the second derivative is continuous. In order to avoid boundary oscillations, the second derivative is forced to be 0 at the boundaries of the feature values ​​to improve the generalization ability of the model.

[0056] For the One enhanced feature Smoothing function Its mathematical expression is:

[0057] in, , representing the number of splines. The more splines there are, the stronger the smoothing function's ability to fit the data, but the higher the risk of overfitting. Here, it is determined using 5-fold cross-validation. This is the optimal value. Indicates the first A natural cubic spline basis function is defined on the feature Piecewise cubic polynomials within the range of values ​​constitute the "basic units" of a smooth function, and all basis functions are linearly independent. Indicates the first In the nth characteristic smoothing function, the th The coefficients of the individual spline basis functions are the core parameters that the model needs to learn. The magnitude and sign of the coefficients determine the degree of contribution of the corresponding basis functions to the smoothing function.

[0058] When linear GAM is used for regression tasks, the following assumptions must be met: No. The actual Y chromosome concentration of each sample It consists of the model prediction component and random error, namely:

[0059] Assumptions for the random error term:

[0060] in, For the first The random error of a sample Let be the variance of the error term. This assumption implies that the error term follows a normal distribution with a mean of 0 and a constant variance, ensuring the rationality of subsequently using the least squares method to solve for the parameters.

[0061] The goal of model solving is to minimize the empirical loss and regularization term, balancing model fit and generalization ability. Loss function The formula is:

[0062] in, This is the experience loss, i.e., the sum of squared mean residuals. For regularization terms, For regularization strength, For smoothing function The second derivative. This part prevents overfitting by penalizing the "curvature" of the smoothing function.

[0063] (4) Regularization and solution The solution is obtained through penalized least squares optimization, and the specific steps are as follows: a. Spline basis function expansion: For the enhanced feature matrix... Each enhanced feature in ,use Natural cubic spline basis functions Expanding, we obtain the basis matrix for each feature. With the first Taking an enhanced feature as an example, the basis matrix The structure is as follows:

[0064] in, Let m represent the m-th natural cubic spline basis function.

[0065] b. Construction of joint basis matrix: constructing the basis matrix of all features With intercept term Concatenate the columns to obtain the joint basis matrix. :

[0066] c. Penalized Least Squares Optimization: Model Parameter Vector The solution formula is:

[0067] in, It contains one intercept term and The coefficients of the spline basis functions. joint basis matrix The transpose of the matrix, for The symmetric matrix reflects the correlation between basis functions. This represents the target variable vector. This represents a diagonal regularized matrix.

[0068] d. Post-processing of predicted values: Since the actual range of Y chromosome concentration is 0-0.5 (concentration cannot be negative, and the upper limit of experimental detection is 0.5), the initial predicted values ​​of the model need to be post-processed. By limiting the range, the final predicted value is obtained. Its formula is:

[0069] (5) Model evaluation To comprehensively evaluate the model's fit and generalization ability, four core evaluation metrics were used: coefficient of determination, mean squared error, root mean square error, and mean absolute error. All metrics were based on the processed predicted values. Compared with actual value Calculation, the smaller the index value or The closer the value is to 1, the better the model performance.

[0070] (6) Significance test Through overall significance testing and eigensignificance verification, it can be demonstrated that there is a correlation between gestational age, maternal BMI, and Y chromosome concentration, but it is not a simple linear relationship. It should be understood that the specific verification methods can be selected by those skilled in the art based on the actual situation, and are not limited here.

[0071] In S3, after obtaining the correlation results between Y chromosome concentration and gestational age and maternal BMI, this embodiment uses the quartile grouping method to classify the BMI of pregnant women with male fetuses in order to accurately analyze the impact of maternal BMI on NIPT testing. Specifically, the 25th, 50th, and 75th percentiles are used as the cutoff points, which can ensure that the sample size of each group is balanced and objectively reflect the hierarchical differences in BMI, avoiding subjective grouping bias.

[0072] Using 988 valid male fetal test data points selected after preprocessing as a sample, the pregnant women's BMI was divided into 4 groups using Python. The specific grouping results are shown in Table 1.

[0073] Table 1 BMI Grouping;

[0074] Before determining the optimal NIPT time point for each BMI group, this embodiment first uses Python to perform statistical analysis on the sample data. The specific results are as follows: Figure 7 As shown in the figure. The yellow dashed line represents the compliance rate constraint threshold of 75.0%, and the four colored curves correspond to the BMI groups in turn.

[0075] Depend on Figure 7 It was found that the cumulative achievement rate of all groups increased with gestational age, which is consistent with the biological principle that prolonged gestational age promotes the accumulation of fetal cell-free DNA, thereby improving the achievement rate of NIPT testing. However, there were significant differences among the BMI groups. The cumulative achievement rate curves of groups 1 and 3 increased rapidly and showed similar performance in the early stages. The achievement progress of group 2 was between that of groups 1, 3, and 4. The curve of group 4 increased the slowest, and the cumulative achievement rate of group 4 was significantly lower than that of other groups at each gestational age stage. This indicates that the higher the pregnant woman's BMI, the more difficult it is to achieve the target concentration of fetal Y chromosome. It is evident that high BMI pregnant women need to be given special management during pregnancy to improve NIPT achievement. Pregnant women with a BMI in the range of [20.7, 33.8) showed relatively better achievement performance, which can provide a reference for the selection of the time window for NIPT testing during pregnancy and the formulation of intervention strategies.

[0076] In NIPT, fetal Y chromosome concentration is a key indicator for determining fetal sex, and clinically, it is necessary to ensure that the concentration reaches or exceeds 4% to guarantee detection accuracy. Gestational age and maternal BMI are the core factors affecting Y chromosome concentration: Too early a gestational age may result in false negatives because the Y chromosome concentration has not reached the detection limit; too late a gestational age, while increasing the detection rate, may miss the window of opportunity for clinical intervention; and a high maternal BMI may dilute the Y chromosome concentration due to maternal adipose tissue, reducing the detection rate.

[0077] To avoid the aforementioned problems and balance detection accuracy, timely intervention, and risk control, this embodiment first constructs a multi-objective programming model for pregnant women in different BMI groups to solve for the optimal NIPT testing time point in the first stage. Before model construction, survival analysis is used to quantify the risk of Y chromosome concentration not reaching the target under different gestational weeks and BMI conditions, setting reasonable constraints and objective function weights for the model.

[0078] First, to address the issue that traditional BMI achievement rate analysis ignores the "when the target was achieved" time dimension, the Nelson-Aalen model survival analysis method was used to process the sample data of pregnant women in different BMI groups: The study calculates four core indicators: cumulative risk of not meeting the target, survival probability of "meeting the target", cumulative rate of meeting the target, and instantaneous risk rate of not meeting the target. It quantifies the detection risk and probability of meeting the target at different gestational weeks and provides input for the multi-objective programming model. Since the survival analysis results are discrete "gestational age-indicator" data, linear interpolation is used to convert them into continuous functions, and data consistency processing is performed to ensure that they meet the model requirements.

[0079] Furthermore, based on the continuous index function obtained from survival analysis, a multi-objective programming model is constructed to solve for the optimal NIPT timing for pregnant women in different BMI groups: The continuous variable "optimal gestational week of detection t (t∈[10,25] weeks, accuracy 0.1 weeks)" is used as the decision variable. Piecewise linear constraints are adopted to transform the nonlinear relationship between the cumulative achievement rate, gestational week of detection, risk adjustment time and the decision variable into linear constraints. With the goal of "maximizing the cumulative achievement rate, minimizing the gestational week of detection, and minimizing the risk adjustment time", a linear weighted objective function is constructed after index normalization, and the solution is obtained under the constraint of "cumulative achievement rate ≥ 75%".

[0080] The multi-objective programming model was solved using Python, and the specific results of the optimal NIPT timing for each group under this BMI grouping are shown in Table 2: Table 2 Optimal NIPT timing for four BMI groups;

[0081] Furthermore, four sets of core indicator curves for the BMI survival model were plotted, visually demonstrating the dynamic changes of the three core indicators—cumulative risk of not reaching the target, probability of survival reaching the target, and instantaneous risk of not reaching the target—with gestational age at different BMI levels. Specific results are as follows: Figures 8-11 As shown, green represents the survival probability of achieving the target, that is, the probability that "the Y chromosome concentration is still not up to standard"; red represents the cumulative risk rate of not achieving the target, that is, "the cumulative risk of not achieving the target from the beginning of pregnancy to the current gestational week"; blue represents the instantaneous risk rate of not achieving the target, that is, "the instantaneous probability of not achieving the target per unit time at the current gestational week"; and the purple dashed line marks the optimal NIPT time point calculated from the BMI of each group.

[0082] Depend on Figures 8-11 It can be seen that Group 1 had a clear advantage in achieving the target in the early stage, but the risk increased sharply after 16.7 weeks and then tended to stabilize; Group 2 performed the best in maintaining the target during pregnancy, with a gradual decline in the survival probability of achieving the target, the lowest cumulative risk of not achieving the target, and a later fluctuation in instantaneous risk; Group 3 showed a continuous unstable characteristic, with the survival probability of achieving the target continuously decreasing after 16.8 weeks, and the overall instantaneous risk increasing; Group 4 maintained a high probability of achieving the target for a long time, but the cumulative risk rate of not achieving the target continued to rise after 17.0 weeks, and the risk of a surge in the later stage was significant.

[0083] This result confirms that the impact of BMI on NIPT target achievement risk exhibits regional heterogeneity, and also verifies the reliability of the BMI grouping and optimal NIPT timing determined in this embodiment.

[0084] As one implementation method, the model performance was verified through multiple validations, and the results were all excellent. In the analysis of the impact of measurement error, after adding a 10% error to each BMI group, the compliance rate of the optimal detection time point for each group decreased slightly but remained stable overall, showing a gradient stratification that is negatively correlated with BMI level. This conforms to the rule that the higher the BMI, the greater the difficulty in achieving the target, indicating that the optimal NIPT time point has strong robustness to detection error.

[0085] In terms of model validation, ANOVA and F-tests showed that BMI and gestational age had extremely significant nonlinear effects on Y chromosome concentration variation (P < 0.001 for both), confirming the rationality of the four-group division of BMI. In the Bootstrap sampling test, the deviation between the original value and the sampling mean of the optimal detection gestational age in each group was minimal, and the frequency distribution was concentrated, indicating good robustness of the model results. In the sensitivity analysis, after ±10% perturbation of the core parameters, the fluctuation range of the optimal detection time point was within an acceptable range, demonstrating excellent anti-interference and adaptability of the model.

[0086] As can be seen, the model has undergone multiple validations and possesses good reliability, robustness, and stability, and can accurately recommend the optimal NIPT testing time for pregnant women in different BMI groups.

[0087] Furthermore, based on the optimal NIPT timing determined by the above BMI grouping study, a second-stage optimal NIPT timing solution is performed. This embodiment further integrates 12 key factors, including the pregnant woman's BMI, gestational age, height, weight, age, Y chromosome concentration, X chromosome concentration, Z-score of chromosome 13, Z-score of chromosome 18, Z-score of chromosome 21, pregnancy number code, and delivery number code. Taking into account the impact of detection error and the proportion of fetal Y chromosome concentration reaching the target, the optimal NIPT timing within each group is used as the target. The BMI of male fetus pregnant women is rationally grouped to minimize the potential risks to pregnant women. Finally, the optimal gestational age for detection is determined for different BMI groups, providing more comprehensive decision support for clinical NIPT testing.

[0088] First, before grouping by BMI, considering the multicollinearity among the 12 factors such as age, height, and weight, principal component analysis (PCA) was used for dimensionality reduction.

[0089] A feature matrix containing 12 relevant features was constructed. The feature data was standardized to eliminate differences in units and ensure that each feature contributed equally to the clustering. Then, the covariance matrix of the standardized features was calculated, and principal components were extracted through eigenvalue decomposition. Principal components with a cumulative explained variance of ≥95% were retained, which eliminated the effects of collinearity and avoided information loss. Finally, the principal component comprehensive score of each sample was obtained by weighted calculation.

[0090] Next, the number of groups was determined, and the elbow method was used to determine the optimal number of groups. Four grouping cases (k=2 to 5) were tested sequentially. Based on the principal components after PCA dimensionality reduction, an objective equation was constructed, aiming to minimize the maximum range of principal component scores within a group. The optimal grouping scheme was solved using integer programming. Combining the elbow method analysis results, such as... Figure 12 As shown, the objective function values ​​of the 5 groups are optimal, the differences within the groups are minimal, and the differences between the groups are significant. Therefore, the final number of groups is determined to be 5.

[0091] The integer programming model was solved using Python to obtain the specific grouping results of pregnant women's BMI, as shown in Table 3. The BMI was divided into five intervals: ≤29.1, 29.1-31.3, 31.3-32.6, 32.6-35.1, and >35.1. The sample size distribution of each group is reasonable, ensuring the reliability of subsequent analysis.

[0092] Table 3 BMI Grouping;

[0093] Furthermore, before determining the optimal NIPT timing point, statistical analysis is performed on the multi-factor sample data, such as... Figure 13 As shown, the cumulative achievement rate curves for each BMI group are presented. The results show that all five groups exhibit a continuous upward trend in cumulative achievement rate with increasing gestational age, consistent with the biological principle that fetal cell-free DNA accumulates more fully as gestational age increases. However, significant differences exist between groups: Group 4 (32.6-35.1) showed the fastest rate of increase and the highest final level, approaching 1.0 at 24 weeks; Groups 1, 2, and 3 had achievement rates in the middle range; Group 5 (>35.1) showed the slowest increase, remaining significantly lower than other groups at 24 weeks. This indicates that achieving the NIPT target is more difficult for pregnant women with high BMI, while pregnant women with a BMI in the 32.6-35.1 range showed relatively the best achievement performance.

[0094] Considering the impact of random errors such as experimental operation and instrument precision in Y chromosome concentration detection, a Monte Carlo simulation is used to quantify the effect of errors on the compliance rate. Assuming that random errors follow a log-normal distribution (closely reflecting the characteristics of actual detection error distribution), 100 perturbed concentration values ​​are generated for each sample, and a compliance label is defined (Y chromosome concentration ≥4% is considered compliant), providing multi-scenario data support for subsequent risk quantification.

[0095] For each BMI group, the Nelson-Aalen model is used to quantify the dynamic relationship between gestational age and target achievement rate. By calculating the mathematical expectation of indicators under 100 scenarios, the average cumulative target achievement rate, average instantaneous non-target risk rate, and average risk adjustment time are obtained, avoiding random interference from a single scenario and accurately characterizing the detection risk features of different gestational ages.

[0096] Using the optimal gestational age for detection in each group as the decision variable (range 10-25 weeks), a multi-objective programming model is constructed. Nonlinear indicators are transformed into linear constraints through piecewise linear constraints. The objective function balances maximizing the cumulative achievement rate, minimizing the gestational age for detection, and minimizing the risk adjustment time. After indicator normalization, a linear weighted function is constructed to fully consider the impact of detection errors, improving the adaptability of the solution. The optimal detection time is then determined under the constraint of the cumulative achievement rate.

[0097] The optimal NIPT timing for five BMI groups was determined using Python, as shown in Table 4. The optimal gestational age for testing in each group was concentrated between 16.6 and 17.3 weeks, with group 4 having the highest success rate (66.03%) and group 5 having the lowest (34.81%).

[0098] Table 4. Optimal NIPT timing;

[0099] Select groups 1, 2, 4, and 5 to plot the core indicator curves of the survival model, as follows: Figures 14-17 As shown, the dynamic changes of indicators such as the risk rate of not meeting the target and the survival probability of meeting the target with gestational age are intuitively displayed, which verifies the rationality of the optimal timing selection and is consistent with the core principle that BMI affects the timing of testing.

[0100] As one implementation method, the model performance was verified through BMI grouping rationality tests, stability tests, and sensitivity analyses, all of which showed excellent results. In the BMI grouping rationality test, ANOVA was performed on key features such as principal component composite score, BMI, and age. Except for age, all other features showed significant differences among different BMI groups, indicating that the grouping method effectively distinguished the variation among key features, demonstrating sufficient rationality. The stability test showed that although the mean optimal gestational age varied among the groups, the standard deviation was ≤0.11, and the fluctuation range did not exceed 0.3, indicating that all groups were considered stable with low variability in optimal gestational age within each group, making the results reliable. Sensitivity analysis was conducted for three types of parameter perturbation scenarios. The fluctuation range of the optimal time point for all groups was ≤0.8 weeks, with extremely low dispersion and maximum deviation, demonstrating strong anti-interference ability and excellent sensitivity performance.

[0101] In summary, the five BMI groups are reasonable, and the optimal gestational age for testing is concentrated between 16.6 and 17.3 weeks. The high BMI group is more difficult to reach the target and has higher risks. The model has undergone multiple tests and has good reliability, stability and anti-interference ability, and can accurately recommend the timing of NIPT testing for pregnant women with different BMIs.

[0102] In S4, the determination of abnormalities in female fetuses is based on the aneuploidy of chromosomes 21, 18, and 13 in the pregnant woman with a female fetus. The determination process for female fetal abnormalities is given by comprehensively considering factors such as the Z-value, GC content, number of reads and related ratios, BMI, etc. of the X chromosome and the above chromosomes.

[0103] 1. Data Processing Because the appendix contains a large number of data columns for female fetal detection, this section first performs data verification and preparation to eliminate noise interference and balance the category distribution. The specific steps are as follows: (1) Feature selection First, based on the analysis above and the information given in the title, this embodiment identifies 13 features: age, pregnant woman's BMI, gestational age, Z-value of chromosomes 13, 18, 21, and X, GC content of chromosomes 13, 18, and 21, proportion of alignments on the reference genome, proportion of duplicate reads, and proportion of reads that are filtered out.

[0104] (2) Data extraction and label definition a. Data extraction: Read the preprocessed data, parse the abnormal identifiers in the "Aneuploidy of Chromosomes" column, namely "T21", "T13" and "T18", extract the abnormal types according to the rules, retain only the chromosome abnormal identifiers in the format of "T + number", and exclude non-standard characters.

[0105] b. Binary label definition: Define the target variable The "Is it abnormal?" function simplifies the determination of aneuploidy, with the following rules:

[0106] This forms the dimension. A well-labeled initial dataset, The sample size is 13, which represents 13 features, and 4 represents 4 label columns: "chromosome aneuploidy", "abnormal chromosome", "abnormality", and "simplified abnormality type".

[0107] (3) Data feature analysis a. Statistical distribution analysis: Calculate the mean, standard deviation, quartiles, and other statistical measures for 13 characteristics.

[0108] b. Analysis of abnormal distribution patterns: Statistically analyze the proportion of abnormal samples to normal samples in T21, T18, T13, and X to clarify the category imbalance problem of "more normal samples and fewer abnormal samples", and provide a basis for subsequent balancing.

[0109] (4) Feature verification To eliminate redundant features that lack discriminative power, Welch's t-test was used to compare the "normal group". "and "abnormal group" The method of differentiating the mean values ​​of the abnormal groups does not require the assumption of homogeneity of variance between the two groups, compared to the traditional t-test, and is more suitable for heteroscedastic scenarios in medical data where "the abnormal group has greater fluctuations in characteristics".

[0110] a. Test statistic and degrees of freedom calculation: t-statistic, used to measure the degree of standardization of the difference between the means of two groups:

[0111] in, Indicates the first The characteristic mean of the group (k=1 is the abnormal group, k=2 is the normal group), Indicates the first The sample variance of the group Indicates the first The sample size of the group.

[0112] b. Welch degrees of freedom: To avoid the failure of test power due to unequal variances, the formula is:

[0113] (5) Class imbalance handling: double oversampling strategy The original data contains an imbalance between abnormal and normal samples. In medical data, the imbalance of "few abnormal samples and many normal samples" can cause the model to be biased towards the majority class and miss abnormal samples. Therefore, this embodiment uses a combination of SMOTE synthesis oversampling and Gaussian noise probability oversampling to deal with this problem. The specific steps are as follows.

[0114] a. Target Ratio Setting: The target ratio of normal samples to abnormal samples is set at 2:1. This ratio avoids both a majority-class-dominated model and the over-synthesis of abnormal samples that deviate from true biological characteristics. The formula for calculating the target number of abnormal samples is as follows:

[0115] in, Number of abnormal samples This represents the normal sample size. The target ratio.

[0116] b. SMOTE synthetic oversampling: For original anomalous samples... In its Neighbor Within the specified range, synthetic samples that "conform to the true feature distribution" are generated through linear interpolation, as shown in the following formula:

[0117] For new sample data, For nearest neighbor samples, It represents the random coefficients in the interval [0,1].

[0118] c. Gaussian noise probability oversampling: This applies to both SMOTE synthesized samples and the original samples. Gaussian noise was added to generate additional samples to simulate natural fluctuations in clinical settings.

[0119] in, This represents noise that follows a normal distribution. This represents the noise intensity parameter.

[0120] (6) Feature standardization Because different features have significantly different dimensions, directly inputting them into the model would lead to an optimization bias towards features with large dimensions. StandardScaler is used for standardization, as shown in the following formula:

[0121] Represents the original feature values. Representing features on the training set The mean, Representing features on the training set The standard deviation.

[0122] Since the logistic regression model can output the contribution coefficients of each feature to the anomaly detection, it has strong interpretability and suppresses overfitting through L2 regularization.

[0123] (1) Model objective function The model aims to minimize the cross-entropy loss and introduces L2 regularization to constrain the absolute value of the regression coefficients, preventing overfitting to noisy samples. The objective function is as follows:

[0124] in, For sample size, For the sample The binary classification labels (1=abnormal, 0=normal). sample The probability of an anomaly The regularization coefficient is . For hyperparameters, For the intercept term, This represents the squared L2 norm, used to constrain the absolute value of regression coefficients.

[0125] (2) Model optimization algorithm Coordinate descent is computationally efficient for high-dimensional features and can stably handle L2 regularization terms. Therefore, this embodiment uses coordinate descent to optimize the objective function. The core logic is to iteratively optimize a single regression coefficient while keeping other coefficients fixed, minimizing the loss function until convergence. The specific steps are as follows: Step 1: Initialize regression coefficients ; Step 2: For each coefficient With other coefficients fixed, solve for the loss function. The smallest Update value; Step 3: Repeat Step 2 until all coefficient updates are less than the convergence threshold. Step 4 outputs the final regression coefficients; the model passes... Output anomaly probability, set If the condition is met, it is considered abnormal; otherwise, it is considered normal.

[0126] By using Python to apply the coordinate descent method to fit and solve the logistic regression model, a confusion matrix between outliers and normal samples was obtained, visually displaying the data distribution of true negative examples, false positive examples, false negative examples, and true positive examples. Specifically... Figure 18 As shown.

[0127] Depend on Figure 18 As can be seen, True Negative Examples (TN) are the number of samples that actually belong to the normal category and are predicted as normal by the model. In this embodiment, the value is 63, representing the number of normal samples correctly identified by the model. False Positive Examples (FP) are the number of samples that actually belong to the normal category but are misclassified as abnormal by the model. The value is 31, corresponding to the number of "Type I errors" where normal samples are misdiagnosed as abnormal. False Negative Examples (FN) are the number of samples that actually belong to the abnormal category but are misclassified as normal by the model. There are 33 such samples, representing the "Type II errors" where abnormal samples are missed and are not diagnosed as normal. True Examples (TP) are the number of samples that actually belong to the abnormal category and are predicted as abnormal by the model. Here, the value is 111, which means that the number of abnormal samples correctly identified by the model is 111.

[0128] This embodiment uses 5-fold cross-validation to evaluate the model's generalization performance and stability, dividing the dataset into training and test sets in an 8:2 ratio. The cross-validation results show that the model's average accuracy is 0.6729 ± 0.0305, indicating that the model has relatively stable predictive ability across different data subsets. In the independent test set validation, the model's accuracy reaches 0.7059, further validating the model's effective identification performance for non-fixed factors.

[0129] In logistic regression models, the Wald test examines whether the regression coefficient corresponding to a certain independent variable is statistically significant (not zero), thereby inferring whether there is a statistically significant association between the independent and dependent variables. Specific results are shown in Table 5.

[0130] Table 5. Wald test results;

[0131] Table 5 shows that features such as GC content of chromosome 21, GC content of chromosome 13, age, Z-score of chromosome 21, Z-score of chromosome X, and the proportion of filtered reads had P < 0.001, indicating extremely significant impact on the model. Features such as Z-score of chromosome 18, maternal BMI, and gestational age had P values ​​between 0.001 and 0.05, indicating significant impact on the model. Features such as GC content of chromosome 18 had P ≥ 0.05, indicating no statistical significance. Overall, most core features among chromosome-specific and sequencing quality indicators passed the Wald test, validating the rationality of feature selection and further demonstrating the good interpretability and predictive performance of the model.

[0132] The model's performance in identifying "normal" and "abnormal" categories was further evaluated based on precision, recall, and F1 score. The specific evaluation results are shown in Table 6.

[0133] Table 6. Model evaluation results;

[0134] Table 6 shows that for the normal class, the model precision is 0.6071, indicating that only about 60.71% of the samples predicted as normal actually belong to the normal category; the recall is 0.7234, meaning that only about 72.34% of the actual normal samples were correctly identified; and the F1 score is 0.6602, reflecting that the balance between precision and recall for the normal class is at a moderate level. For the abnormal class, the model precision reaches 0.7937, indicating that about 79.37% of the samples predicted as abnormal are actually abnormal, resulting in a relatively low false positive rate; the recall is 0.6944, meaning that only about 69.44% of the actual abnormal samples were correctly identified, still indicating a certain risk of missed diagnoses; and the F1 score is 0.7407, showing a better balance between precision and recall than the normal class. Overall, the model performs better in terms of precision for the abnormal class, but there is still room for improvement in both the false positive rate for the normal class and the false negative rate for the abnormal class.

[0135] This specific embodiment addresses the shortcomings of existing NIPT technologies, such as insufficient accuracy due to uniform testing time points and simple BMI grouping, and the lack of a systematic method for identifying female fetal abnormalities. In data processing, rigorous cleaning preserves valid data, and accurate identification of biological variants avoids accidental deletion, laying a reliable foundation for subsequent analysis. Regarding optimization of male fetal detection, for the first time, the correlation between Y chromosome concentration and gestational age is analyzed in four stages according to the trend of gestational age. A generalized additive model is used to capture the non-linear relationship between gestational age, BMI, and Y chromosome concentration. Combined with Nelson-Aalen survival analysis and multi-objective programming, the optimal time point is determined for each BMI group. Monte Carlo simulation and Bootstrap sampling are used to verify the model's stability and anti-interference capabilities. For identifying female fetal abnormalities, multi-dimensional features are integrated, and double-oversampled balanced data is used to construct a logistic regression model with L2 regularization, achieving accurate identification and providing a personalized and scientific auxiliary solution for clinical NIPT testing.

[0136] Example 2 This embodiment provides a system for determining the optimal timing of NIPT in male fetuses and for assessing abnormalities in female fetuses, including: The data processing module is used to preprocess the NIPT test data of pregnant women; the NIPT test data of pregnant women includes gestational age, pregnant woman's BMI and Y chromosome concentration; The male fetus data analysis module is used to extract male fetus detection data from the preprocessed data, divide the gestational age into multiple stages, calculate the Spearman correlation between Y chromosome concentration and gestational age and maternal BMI in each stage, and construct a relationship model to obtain the correlation results between Y chromosome concentration and gestational age and maternal BMI. The module for determining the timing of male fetuses is used to divide pregnant women with male fetuses into multiple groups based on their BMI. Combining the correlation results and integrating multiple factors, the module obtains relevant indicators for each group at different gestational weeks through survival analysis, constructs a multi-objective programming model, and solves the optimal NIPT timing for each group of pregnant women. The female fetus abnormality determination module is used to extract female fetus detection data from preprocessed data, obtain influencing features through feature statistical analysis and significance screening, construct a logistic regression model, and determine whether the female fetus is abnormal based on the model output results.

[0137] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in Embodiment 1 above.

[0138] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in Embodiment 1 above.

[0139] The steps or modules involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any method of the present invention. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining the optimal timing of NIPT in male fetuses and for assessing abnormalities in female fetuses, characterized in that, include: Preprocessing of NIPT test data from pregnant women; The NIPT test data for pregnant women includes gestational age, maternal BMI, and Y chromosome concentration; Male fetal detection data were extracted from the preprocessed data. The gestational age was divided into multiple stages. The Spearman correlation between Y chromosome concentration and gestational age and maternal BMI was calculated for each stage. A relationship model was constructed to obtain the association results between Y chromosome concentration and gestational age and maternal BMI. The BMI of pregnant women with male fetuses was divided into multiple groups. Based on the correlation results and by integrating multiple factors, the relevant indicators of each group at different gestational weeks were obtained by survival analysis. A multi-objective programming model was constructed to solve the optimal NIPT time point for each group of pregnant women. Female fetal detection data were extracted from the preprocessed data. Influencing features were obtained through feature statistical analysis and significance screening. A logistic regression model was constructed, and the model output results were used to determine whether the female fetus was abnormal.

2. The method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in claim 1, characterized in that, The NIPT test data for pregnant women also includes: pregnant woman's age, height, weight, number of original reads, proportion of reads aligned to the reference genome, proportion of duplicate reads, number of uniquely aligned reads, GC content, Z-value of chromosome 13, Z-value of chromosome 18, Z-value of chromosome 21, Z-value of X chromosome, Z-value of Y chromosome, X chromosome concentration, GC content of chromosome 13, GC content of chromosome 18, GC content of chromosome 21, and proportion of reads that were filtered out.

3. The method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in claim 1, characterized in that, The preprocessing includes missing value handling, cross-validation, numerical transformation, and box plot analysis.

4. The method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in claim 1, characterized in that, The division of gestational age into multiple stages is as follows: The preprocessed male fetal test data was obtained, and the original gestational week data from 10 to 25 weeks was extracted and discretized with a discretization step size of 1 week. The Y chromosome concentration value at each discretized point is used as the average of all data included at that point, and a curve showing the average change of Y chromosome concentration at each gestational week is plotted based on this. Based on the differences in concentration trends within each interval, gestational weeks are divided into four stages: 11 to 15 weeks, 16 to 18 weeks, 19 to 22 weeks, and 23 to 25 weeks.

5. The method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in claim 1, characterized in that, The constructed relationship model yielded correlation results between Y chromosome concentration and gestational age and maternal BMI, specifically including: A generalized additive model was used to construct a model of the relationship between Y chromosome concentration and gestational age and maternal BMI; The model input includes raw features of gestational age and maternal BMI, as well as their squared-term derived features, to capture non-linear trends; The nonlinear relationship between variables was fitted by a natural cubic spline smoothing function, and the model was solved and regularized using penalized least squares method. Finally, the correlation results of Y chromosome concentration with gestational age and BMI were obtained.

6. The method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in claim 1, characterized in that, The process involves dividing pregnant women with male fetuses into multiple groups based on their BMI, combining the correlation results, and integrating multiple factors. Survival analysis is used to obtain relevant indicators for each group at different gestational weeks. A multi-objective programming model is then constructed to solve for the optimal NIPT timing point for each group of pregnant women. Specifically, the optimal NIPT timing point solution is achieved in two stages: the first stage only considers the BMI indicator, while the second stage considers the integration of multiple factors. The solution for the optimal NIPT timing in the first stage is as follows: the BMI of pregnant women with male fetuses is divided into four groups according to the quartiles, and the Nelson-Aalen survival analysis method is used to quantify the cumulative target achievement rate, instantaneous non-target risk rate and risk adjustment time of each group at different gestational weeks. Based on the aforementioned indicators, a multi-objective programming model is constructed. With the cumulative achievement rate not lower than a set threshold as a constraint, the model maximizes the cumulative achievement rate, minimizes the gestational age at detection and risk adjustment time, and solves for the optimal NIPT timing for each group.

7. The method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in claim 6, characterized in that, The second stage involves finding the optimal NIPT timing point, specifically as follows: Integrating 12 key factors, including pregnant women's BMI, gestational age, height, weight, age, Y chromosome concentration, X chromosome concentration, Z-value of chromosome 13, Z-value of chromosome 18, Z-value of chromosome 21, pregnancy number code, and delivery number code, we first eliminated collinearity by dimensionality reduction through principal component analysis, and then determined the number of BMI groups to be 5 based on the elbow method. Monte Carlo simulation was used to introduce detection errors, generate multi-scenario disturbance data, and calculate the average cumulative compliance rate and risk indicators of each group. A multi-objective programming model was reconstructed. Under the constraint of cumulative achievement rate, the optimal gestational age for testing was solved for each group with the objectives of maximizing the cumulative achievement rate and minimizing the gestational age for testing and the risk adjustment time.

8. A system for determining the optimal timing of NIPT in male fetuses and for assessing abnormalities in female fetuses, characterized in that, include: The data processing module is used to preprocess NIPT test data from pregnant women; The NIPT test data for pregnant women includes gestational age, maternal BMI, and Y chromosome concentration; The male fetus data analysis module is used to extract male fetus detection data from the preprocessed data, divide the gestational age into multiple stages, calculate the Spearman correlation between Y chromosome concentration and gestational age and maternal BMI in each stage, and construct a relationship model to obtain the correlation results between Y chromosome concentration and gestational age and maternal BMI. The module for determining the timing of male fetuses is used to divide pregnant women with male fetuses into multiple groups based on their BMI. Combining the correlation results and integrating multiple factors, the module obtains relevant indicators for each group at different gestational weeks through survival analysis, constructs a multi-objective programming model, and solves the optimal NIPT timing for each group of pregnant women. The female fetus abnormality determination module is used to extract female fetus detection data from preprocessed data, obtain influencing features through feature statistical analysis and significance screening, construct a logistic regression model, and determine whether the female fetus is abnormal based on the model output results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method for determining the optimal timing of NIPT in male fetuses and judging abnormalities in female fetuses as described in any one of claims 1-7.

Images