A method for assessing the risk of pregnancy outcome of pregnant women based on multi-factor feature analysis

Abstract

This invention relates to the field of medical data analysis, and more particularly to a method for assessing the risk of pregnancy outcomes in pregnant women based on multifactor feature analysis. The method includes: acquiring individual measured time-series datasets and baseline reference data for physiological expectations; obtaining a gestational age-specific pathological surge penalty factor by weighting the ratio of angiogenic factors to anti-angiogenic factors; obtaining a cross-system pathological cascade time coupling factor by temporally tracing and coupling abnormal deviations in vascular system indicators and blood system indicators; obtaining a fused feature vector by correcting and fusing the initial linear regression slope features; and obtaining the pregnancy outcome risk assessment result by performing machine learning prediction on the fused feature vector. This method solves the problem that existing prediction methods based on linear regression slope extraction and feature splicing cannot effectively identify gestational age-specific early high-risk signals and cross-system short-term pathological cascade signals.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, and in particular to a method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis. Background Technology

[0002] In the management of pregnant women, early risk identification for adverse pregnancy outcomes such as severe preeclampsia is crucial for ensuring maternal and infant safety and improving the timeliness of clinical intervention. Current clinical practice typically involves risk assessment based on longitudinal monitoring data from multiple prenatal checkups during pregnancy. This includes continuously collecting physiological indicators closely related to placental dysfunction and target organ damage, such as the ratio of angiogenic factors to anti-angiogenic factors and platelet counts, and analyzing the trends of pathological changes during pregnancy in conjunction with gestational age information. Because this prenatal data often exhibits dynamic changes with gestational age, variable sampling frequency, and parallel evolution of multiple indicators, current techniques typically first extract features from the variable-length time-series data obtained from multiple prenatal checkups, and then input the extracted fixed-dimensional features into a machine learning model for classification and prediction to assist in assessing the risk of adverse pregnancy outcomes.

[0003] In existing technologies, a common approach is to use a moving time window or all prenatal checkup points, employing least squares to perform linear regression fitting on measured data of various physiological indicators changing with gestational age, and extracting the corresponding linear regression slopes as time-series dynamic features. Subsequently, the slope features obtained from different indicators are concatenated into vectors and input into machine learning models such as extreme gradient boosting trees for risk classification. This type of method can, to some extent, compress longitudinal prenatal checkup data into a fixed-length input that is easy for models to process. It has advantages such as relatively direct implementation and convenient engineering deployment, and therefore has a certain application basis in the field of pregnancy complication risk prediction.

[0004] However, the clinicopathological progression of adverse pregnancy outcomes, especially conditions like preeclampsia, is not a simple linear process but exhibits significant gestational age specificity and cross-system cascade evolution characteristics. On one hand, the same magnitude of increase or decrease in an indicator at different gestational weeks carries different clinical risks. For example, a vascular-related indicator might be considered a normal physiological fluctuation near delivery in late pregnancy, while a similar change occurring prematurely in mid-pregnancy often indicates a higher pathological risk. On the other hand, there is usually a sequential and time-separation difference between upstream vascular system abnormalities and downstream blood system damage. If both types of abnormalities occur consecutively within a short period, it is more likely to indicate a rapid deterioration of the condition. Existing techniques based on linear regression slope extraction and direct splicing of multiple features rely solely on the absolute changes in the measured data for fitting. They fail to incorporate the expected baseline of normal physiological function at different gestational weeks into the feature calculation process and fail to express the temporal causal relationships and cascading urgency between different systems during the multivariate fusion stage. This makes it difficult to effectively distinguish between normal physiological fluctuations and pathological surges, and also makes it difficult to capture the high-risk signal of short-term resonant deterioration across multiple systems. Therefore, how to construct a method for assessing the risk of pregnancy outcomes in pregnant women that can take into account both differences in physiological background at gestational age and the temporal linkage of cross-system pathological factors has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the present invention aims to propose a method for assessing the risk of pregnancy outcomes in pregnant women based on multifactor feature analysis, in order to solve the problem that existing prediction methods based on linear regression slope extraction and feature splicing cannot effectively identify gestational age-specific early high-risk signals and cross-system short-term pathological cascade signals.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] A method for assessing the risk of pregnancy outcomes in pregnant women based on multivariate characteristic analysis, the method comprising:

[0008] Step S1: Obtain individual measured time series datasets and physiological expected baseline reference data by acquiring and preprocessing multimodal longitudinal sequence data and physiological baseline data;

[0009] Step S2: Obtain the gestational age-specific pathological surge penalty factor by calculating the ratio of angiogenic factors to anti-angiogenic factors relative to the physiological baseline offset and gestational age risk weighting.

[0010] Step S3: Obtain the cross-system pathological cascade time coupling factor by performing time-series tracing coupling calculation on the abnormal offsets of vascular system indicators and blood system indicators;

[0011] Step S4: Obtain the fused feature vector by correcting the initial linear regression slope features with a gestational age-specific pathological surge penalty factor and combining it with a cross-system pathological cascade time coupling factor;

[0012] Step S5: Obtain pregnancy outcome risk assessment results by performing machine learning prediction on the fused feature vector.

[0013] Furthermore, the acquisition and preprocessing of multimodal longitudinal sequence data and physiological baseline data to obtain individual measured time series datasets and physiological expectation baseline reference data includes:

[0014] The process involves acquiring basic profile information and multiple prenatal checkup records of the pregnant woman to be assessed within the risk assessment time window. The basic profile information includes at least the pregnant woman's identification information, the total number of prenatal checkups, and the chronological order of each checkup. For each checkup number, the specific gestational age, measured angiogenesis factor to anti-angiogenesis factor ratio, and measured platelet count data for that checkup are obtained. These data are then linked and stored chronologically to obtain multimodal longitudinal sequence data for the pregnant woman to be assessed. The multimodal longitudinal sequence data undergoes format standardization, outlier removal, and missing value interpolation to complete the data, resulting in an individual measured time series dataset.

[0015] Based on a pre-constructed retrospective clinical database of healthy pregnant women, reference data on the ratio of angiogenic factors to anti-angiogenic factors and platelet counts at different gestational weeks were extracted from healthy pregnant women without records of pregnancy complications or adverse pregnancy outcomes. The reference data on the ratio of angiogenic factors to anti-angiogenic factors and platelet counts were then smoothed to construct a physiological expectation baseline reference library that varies with gestational week. Based on the specific gestational week values ​​corresponding to each prenatal checkup in the individual measured time-series dataset, the physiological expectation baseline values ​​for the ratio of angiogenic factors to anti-angiogenic factors and platelet counts at the corresponding gestational week were matched and extracted from the physiological expectation baseline reference library to obtain physiological expectation baseline reference data.

[0016] Furthermore, the method of obtaining gestational age-specific pathological surge penalty factors by weighting the ratio of angiogenic factors to anti-angiogenic factors against physiological baseline deviation and gestational age risk includes:

[0017] By performing relative offset processing based on the physiological expectation baseline on the ratio data of angiogenic factors and anti-angiogenic factors, pathological excess change characterization data are obtained.

[0018] By applying gestational age risk weighting and nonlinear enhancement processing to the pathological excess change characterization data, a gestational age-specific pathological surge penalty factor was obtained.

[0019] Furthermore, the process of obtaining pathological excess change characterization data by performing relative offset processing based on the physiological expectation baseline on the ratio data of angiogenic factors and anti-angiogenic factors includes:

[0020] For the pregnant woman to be evaluated, the first and second prenatal examination numbers of any two adjacent prenatal examinations within the risk assessment time window are extracted from the individual measured time series dataset. The measured data of the previous specific gestational week value and the ratio of the previous angiogenic factor to anti-angiogenic factor corresponding to the previous prenatal examination number are extracted from the individual measured time series dataset. The measured data of the next specific gestational week value and the ratio of the next angiogenic factor to anti-angiogenic factor corresponding to the next specific gestational week value are extracted from the physiological expected baseline reference data.

[0021] The difference between the measured ratio of the next angiogenic factor to anti-angiogenic factor and the measured ratio of the previous angiogenic factor to anti-angiogenic factor is taken as the measured difference. The difference between the value of the next specific gestational week and the value of the previous specific gestational week is taken as the gestational week interval data. The result of dividing the measured difference by the gestational week interval data is used as the measured change rate assessment between adjacent prenatal examinations.

[0022] The difference between the physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor in the latter and the physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor in the former is taken as the baseline difference. The baseline difference is divided by the calculation result of the gestational week interval data as the baseline change rate assessment between adjacent prenatal checkups.

[0023] The difference between the measured rate of change assessment and the baseline rate of change assessment is used as the relative deviation difference assessment. When the relative deviation difference assessment is less than or equal to a constant 0, the pathological excess change characterization data between adjacent prenatal examinations is set to a constant 0; when the relative deviation difference assessment is greater than a constant 0, the relative deviation difference assessment is used as the pathological excess change characterization data between adjacent prenatal examinations.

[0024] Furthermore, by performing gestational age-specific pathological surge penalty factors on the pathological excess change characterization data through gestational age risk weighting and nonlinear enhancement processing, the following factors are obtained:

[0025] For the pregnant woman to be evaluated, the second prenatal examination number in any two adjacent prenatal examinations within the risk assessment time window is extracted from the individual measured time series dataset. The specific gestational week value corresponding to the second prenatal examination number and the measured data of the ratio of angiogenic factor to anti-angiogenic factor are extracted from the individual measured time series dataset. The physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor corresponding to the specific gestational week value is extracted from the physiological expected baseline reference data. At the same time, the pathological excess change characterization data corresponding to the two adjacent prenatal examinations are extracted.

[0026] The difference between the measured ratio of the next angiogenic factor to anti-angiogenic factor and the physiologically expected baseline value of the ratio of the next angiogenic factor to anti-angiogenic factor is used as the relative deviation difference assessment. The relative deviation difference assessment divided by the calculated physiologically expected baseline value of the ratio of the next angiogenic factor to anti-angiogenic factor is used as the relative deviation rate assessment. The relative deviation rate assessment is then subjected to an exponential mapping with the natural constant as the base to obtain the nonlinear enhancement weights corresponding to the two adjacent prenatal examinations.

[0027] The difference between the preset full-term gestational age constant and the value of the next specific gestational age is used as the early onset risk assessment value. The early onset risk assessment value is divided by the preset full-term gestational age constant to obtain the gestational age risk weighting coefficient for the corresponding two adjacent prenatal checkups.

[0028] The pathological surge penalty value is calculated by multiplying the pathological excess change characterization data, the nonlinear enhancement weight, and the gestational age risk weighting coefficient. The pathological surge penalty values ​​corresponding to all two adjacent prenatal examinations within the risk assessment time window are summed to obtain the gestational age-specific pathological surge penalty factor.

[0029] Furthermore, the step of obtaining the cross-system pathological cascade time coupling factor by performing time-series tracing coupling calculations on abnormal offsets of vascular system indicators and abnormal offsets of blood system indicators includes:

[0030] By performing relative physiological baseline decline offset calculation on blood system indicator data, downstream abnormal response characterization data are obtained;

[0031] By performing relative physiological baseline offset tracing calculation on vascular system indicator data, upstream abnormal trigger characterization data are obtained;

[0032] By performing time-series decay coupling processing on downstream abnormal response characterization data and upstream abnormal trigger characterization data, the cross-system pathological cascade time coupling factor is obtained.

[0033] Furthermore, the process of calculating the relative physiological baseline decline offset of blood system indicator data to obtain downstream abnormal response characterization data includes:

[0034] For any prenatal check-up number of a pregnant woman to be evaluated within the risk assessment time window, extract the specific gestational age value and platelet count measurement data corresponding to the prenatal check-up number from the individual measured time series dataset, and extract the physiological expected baseline value of platelet count corresponding to the specific gestational age value from the physiological expected baseline reference data.

[0035] The difference between the expected baseline value of platelet count and the measured platelet count data is used as the blood system decline deviation difference assessment, and the blood system decline deviation difference assessment divided by the calculated result of the expected baseline value of platelet count is used as the blood system relative decline deviation assessment.

[0036] When the relative downward offset assessment of the blood system is less than or equal to a constant 0, the downstream abnormal response characterization data of the corresponding prenatal examination number is set to a constant 0; when the relative downward offset assessment of the blood system is greater than a constant 0, the relative downward offset assessment of the blood system is used as the downstream abnormal response characterization data of the corresponding prenatal examination number.

[0037] Furthermore, the process of performing relative physiological baseline shift calculation on vascular system indicator data to obtain upstream abnormal trigger characterization data includes:

[0038] For any prenatal examination number within the risk assessment time window of the pregnant woman to be assessed, the specific gestational age value and the measured data of the ratio of angiogenic factor to anti-angiogenic factor corresponding to the target traceability node are extracted from the individual measured time series dataset. The physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor corresponding to the specific gestational age value is extracted from the physiological expected baseline reference data.

[0039] The difference between the measured data of the ratio of angiogenic factors to anti-angiogenic factors and the physiologically expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors is used as the vascular system upward deviation difference assessment. The vascular system upward deviation difference assessment is divided by the calculated result of the physiologically expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors as the vascular system relative upward deviation assessment.

[0040] When the relative upward offset assessment of the vascular system is less than or equal to a constant 0, the upstream anomaly triggering characterization data of the corresponding target tracing node is set to a constant 0; when the relative upward offset assessment of the vascular system is greater than a constant 0, the relative upward offset assessment of the vascular system is used as the upstream anomaly triggering characterization data of the corresponding target tracing node.

[0041] Furthermore, the step of obtaining the cross-system pathological cascade time coupling factor by performing time-series decay coupling processing on downstream abnormal response characterization data and upstream abnormal trigger characterization data includes:

[0042] For any prenatal checkup number of a pregnant woman to be evaluated within the risk assessment time window, the specific gestational week value corresponding to the downstream response node is extracted from the individual measured time series dataset, and the downstream abnormal response characterization data corresponding to the downstream response node is extracted.

[0043] For any downstream response node, among all the prenatal examination serial numbers that are the same as the downstream response node, select one of them as the upstream traceability node, extract the specific gestational age value corresponding to the upstream traceability node from the individual measured time series dataset, and extract the upstream abnormal trigger characterization data corresponding to the upstream traceability node.

[0044] The difference between the specific gestational age value corresponding to the downstream response node and the specific gestational age value corresponding to the upstream trace node is used as the cross-system time interval data. The result of adding the cross-system time interval data to the constant 1 is used as the time decay denominator. The result of dividing the constant 1 by the time decay denominator is used as the time decay weight between the corresponding downstream response node and the upstream trace node.

[0045] The calculation result of multiplying the downstream abnormal response characterization data, the upstream abnormal trigger characterization data, and the time-series decay weight is used as the cascade coupling evaluation value between the corresponding downstream response node and the upstream tracing node. All cascade coupling evaluation values ​​corresponding to any downstream response node are summed to obtain the cascade response value corresponding to that downstream response node. The cascade response values ​​corresponding to all downstream response nodes within the risk assessment time window are summed to obtain the cross-system pathological cascade time coupling factor.

[0046] Furthermore, the step of modifying the initial linear regression slope features with a gestational age-specific pathological surge penalty factor and then fusing them with a cross-system pathological cascade time coupling factor to construct a fused feature vector includes:

[0047] Based on the specific gestational age values ​​and the measured ratio of angiogenic factors to anti-angiogenic factors corresponding to each prenatal checkup in the individual measured time series dataset, linear regression fitting was performed to obtain the initial linear regression slope feature corresponding to the ratio of angiogenic factors to anti-angiogenic factors; based on the specific gestational age values ​​and the measured platelet count data corresponding to each prenatal checkup in the individual measured time series dataset, linear regression fitting was performed to obtain the initial linear regression slope feature corresponding to the platelet count.

[0048] The result of multiplying the gestational age-specific pathological surge penalty factor by the preset first scaling factor and adding it to the constant 1 is used as the slope correction coefficient. The result of multiplying the initial linear regression slope feature corresponding to the ratio of angiogenic factor to anti-angiogenic factor by the slope correction coefficient is used as the corrected slope feature of the ratio of angiogenic factor to anti-angiogenic factor.

[0049] The result of multiplying the cross-system pathological cascade time coupling factor by a preset second scaling factor is used as the cross-system coupling feature value;

[0050] The modified angiogenesis factor to anti-angiogenesis factor ratio slope feature, the initial linear regression slope feature corresponding to platelet count, and cross-system coupling feature value are concatenated to obtain a fused feature vector.

[0051] Compared with the prior art, the present invention has the following advantages:

[0052] This invention describes a method for assessing pregnancy outcome risk in pregnant women based on multifactorial feature analysis. It utilizes longitudinal monitoring data from multiple prenatal checkups during pregnancy, introducing a baseline of normal physiological expectations matched to specific gestational weeks. Abnormal changes in key indicators such as the ratio of angiogenic factors to anti-angiogenic factors are corrected for based on gestational age, moving beyond a mechanical extraction of absolute slope or numerical changes in risk characteristics. This allows for accurate differentiation between normal physiological fluctuations in the mid-to-late stages of pregnancy and pathological surges occurring prematurely in the second trimester. By non-linearly enhancing the degree of relative baseline deviation and assigning differentiated weights based on the clinical pattern that earlier onset of disease carries higher risk, it can more sensitively identify latent high-risk individuals whose changes are not yet statistically significant but already exhibit clear pathological indications. This improves the clinical sensitivity and early warning capability of adverse pregnancy outcome risk assessment, enabling physicians to intervene before the onset of severe complications such as severe preeclampsia. Meanwhile, this invention does not simply treat various physiological indicators as independent input variables. Instead, it utilizes the sequential relationship and temporal proximity between vascular system abnormalities and blood system damage to establish a coupled expression mechanism for cross-system pathological responses. This enables the model to identify the true critical state reflected by the rapid and continuous deterioration of multiple organs within a short period. This technique effectively compensates for the shortcomings of traditional feature splicing methods in reflecting the urgency of pathological cascades. The resulting risk assessment not only reflects the danger level of individual indicator abnormalities but also further reveals the severe progression trend of the disease from upstream abnormalities to downstream damage, thereby improving the specificity and stability of risk assessment and its clinical decision-making value for high-risk maternal and infant events. Attached Figure Description

[0053] 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 an undue limitation of the invention. In the drawings:

[0054] Figure 1 This is a flowchart illustrating a method for assessing the risk of pregnancy outcomes in pregnant women based on multifactor feature analysis, as described in an embodiment of the present invention. Detailed Implementation

[0055] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0056] See Figure 1 This is a flowchart of a method for assessing the risk of pregnancy outcomes in pregnant women based on multi-factor feature analysis, as provided in Embodiment 1 of the present invention. Figure 1 As shown, a method for assessing the risk of pregnancy outcomes in pregnant women based on multivariate characteristic analysis may include:

[0057] Step S1 involves acquiring and preprocessing multimodal longitudinal sequence data and physiological baseline data to obtain individual measured time series datasets and physiological expected baseline reference data.

[0058] First, obtain the basic information of the pregnant woman to be assessed within the risk assessment time window, as well as her multiple prenatal check-up records. The basic information should include at least the identification information of the pregnant woman to be assessed, the total number of prenatal check-ups, and the sequence number of each prenatal check-up arranged in chronological order.

[0059] For each prenatal checkup sequence number, the specific gestational age value, measured data of the ratio of angiogenic factors to anti-angiogenic factors, and measured platelet count corresponding to that prenatal checkup are obtained. The specific gestational age value, measured data of the ratio of angiogenic factors to anti-angiogenic factors, and measured platelet count corresponding to each prenatal checkup are associated and stored in chronological order to obtain multimodal longitudinal sequence data of the pregnant woman to be evaluated.

[0060] After obtaining the multimodal longitudinal sequence data of the pregnant women to be evaluated, the data is further processed by standardizing the format, removing outliers, and interpolating missing values ​​to obtain individual measured time series datasets.

[0061] After obtaining the individual measured time series dataset of the pregnant women to be evaluated, we continued to extract reference data on the ratio of angiogenic factors to anti-angiogenic factors and platelet counts of healthy pregnant women without records of pregnancy complications or adverse pregnancy outcomes at different gestational weeks based on a pre-constructed retrospective clinical database of healthy pregnant women. We then performed smooth fitting on the reference data on the ratio of angiogenic factors to anti-angiogenic factors and platelet counts to construct a reference library of physiological expectations that change with gestational weeks.

[0062] Based on the specific gestational week values ​​corresponding to each prenatal checkup in the individual measured time series dataset, the physiological expected baseline values ​​of the ratio of angiogenic factors to anti-angiogenic factors and the physiological expected baseline values ​​of platelet count at the corresponding gestational week are matched and extracted from the physiological expected baseline reference library to obtain the corresponding physiological expected baseline reference data.

[0063] This completes the acquisition and preprocessing of multimodal longitudinal sequence data and physiological baseline data to obtain individual measured time series datasets and expected physiological baseline reference data.

[0064] Step S2: Obtain the gestational age-specific pathological surge penalty factor by calculating the ratio of angiogenic factors to anti-angiogenic factors relative to the physiological baseline offset and gestational age risk weighting.

[0065] When extracting the time-series characteristics of the key indicator, the ratio of angiogenic factors to anti-angiogenic factors, existing techniques mainly rely on linear regression algorithms based on least squares to calculate the overall slope within a time window. However, this ratio is not a static constant in clinical pathological progression; healthy pregnant women also experience a physiological, slow increase in this indicator during the mid-to-late stages of pregnancy. This leads to the least squares method focusing only on the linear fit and absolute slope of the measured data points in geometric space, detached from the physiological context of the gestational age in which the data is situated. The same slope value, if occurring at 38 weeks of gestation, is usually considered a normal pre-labor physiological change, but if it occurs at 24 weeks of gestation, when it has previously been at a low and stable level, it suggests a risk of early-onset preeclampsia. Because traditional linear regression slope algorithms are time-translation invariant, they cannot identify the different pathological significance of the same slope at different gestational ages. To address the issue of high-risk feature masking caused by the algorithm's deviation from the dynamic physiological baseline, it's insufficient to simply extract the slope of absolute data. Instead, it's necessary to introduce the dynamic physiological baseline curve of this indicator in healthy individuals as a reference standard, transforming the calculation object from an absolute measured rate of change to an excess pathological surge rate relative to the physiological baseline. Simultaneously, considering the objective clinical pattern that earlier disease onset leads to worse prognosis, a gestational age attenuation penalty mechanism needs to be introduced during the temporal feature extraction stage. This reconstructs a purely geometric slope into a pathological burden metric carrying clinical hazard weights.

[0066] In summary, this invention first obtains pathological excess change characterization data by performing relative offset processing on the ratio of angiogenic factors to anti-angiogenic factors based on the physiological expected baseline. Specifically, for the prenatal examination serial number and the next prenatal examination serial number of any two adjacent prenatal examinations within the risk assessment time window for the pregnant woman to be evaluated, the specific gestational week value, the measured ratio of angiogenic factors to anti-angiogenic factors corresponding to the previous prenatal examination serial number, and the measured ratio of angiogenic factors to anti-angiogenic factors corresponding to the next prenatal examination serial number are extracted from the individual measured time series dataset. Furthermore, the physiological expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors corresponding to the previous specific gestational week value and the physiological expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors corresponding to the next specific gestational week value are extracted from the physiological expected baseline reference data. The difference between the measured ratio of a subsequent angiogenic factor to anti-angiogenic factor (AGF) and the measured ratio of a previous AGF to anti-angiogenic factor (anti-angiogenic factor) is used as the measured difference. The difference between the measured ratio of a subsequent specific gestational week and the measured ratio of a previous specific gestational week is used as the gestational week interval data. The result of dividing the measured difference by the gestational week interval data is used as the measured rate of change assessment between adjacent prenatal examinations. The difference between the expected physiological baseline value of the AGF to anti-angiogenic factor ratio of a subsequent and the expected physiological baseline value of the AGF to anti-angiogenic factor ratio of a previous and the measured baseline difference is used as the baseline difference. The result of dividing the baseline difference by the gestational week interval data is used as the baseline rate of change assessment between adjacent prenatal examinations. The difference between the measured rate of change assessment and the baseline rate of change assessment is used as the relative deviation difference assessment. When the relative deviation difference assessment is less than or equal to a constant 0, the pathological excess change characterization data between the corresponding adjacent prenatal examinations is set to a constant 0; when the relative deviation difference assessment is greater than a constant 0, the relative deviation difference assessment is used as the pathological excess change characterization data between the corresponding adjacent prenatal examinations.

[0067] After obtaining the pathological excess change characterization data, further gestational age risk weighting and nonlinear enhancement processing are applied to the pathological excess change characterization data to obtain gestational age-specific pathological surge penalty factors. Specifically, for the pregnant woman to be evaluated, for any two adjacent prenatal examinations within the risk assessment time window, the specific gestational age value corresponding to the next prenatal examination number and the measured data of the ratio of the next angiogenic factor to anti-angiogenic factor are extracted from the individual measured time series dataset. The physiological expected baseline value of the ratio of the next angiogenic factor to anti-angiogenic factor corresponding to the specific gestational age value is extracted from the physiological expected baseline reference data. Simultaneously, the pathological excess change characterization data corresponding to the two adjacent prenatal examinations are extracted. The difference between the measured data of the ratio of the next angiogenic factor to anti-angiogenic factor and the physiological expected baseline value of the ratio of the next angiogenic factor to anti-angiogenic factor is used as the relative deviation difference assessment. The relative deviation difference assessment divided by the calculated result of the physiological expected baseline value of the ratio of the next angiogenic factor to anti-angiogenic factor is used as the relative deviation rate assessment. The relative deviation rate assessment is subjected to an exponential mapping with a natural constant as the base to obtain the nonlinear enhancement weights for two consecutive prenatal checkups. The difference between the preset full-term gestational age constant and the value of the next specific gestational age is used as the early onset risk assessment value. The result of dividing the early onset risk assessment value by the preset full-term gestational age constant is used as the gestational age risk weighting coefficient for two consecutive prenatal checkups. In this embodiment of the invention, the full-term gestational age is set to 40. The result of multiplying the pathological excess change characterization data, the nonlinear enhancement weights, and the gestational age risk weighting coefficient is used as the pathological surge penalty value for two consecutive prenatal checkups. The pathological surge penalty values ​​corresponding to all two consecutive prenatal checkups within the risk assessment time window are summed to obtain the gestational age-specific pathological surge penalty factor.

[0068] In one implementation, assume the first The total number of prenatal checkups for each pregnant woman within the risk assessment time window was: ;No. The pregnant woman in the first The specific gestational week at the time of the last prenatal checkup was: ;No. The pregnant woman in the first The first prenatal checkup at gestational week The measured data for the ratio of angiogenic factor to anti-angiogenic factor was as follows: ;No. The pregnant woman in the first The first prenatal checkup at gestational week The measured data for the ratio of angiogenic factor to anti-angiogenic factor was as follows: ; gestational age The physiological expected baseline value for the ratio of angiogenic factors to anti-angiogenic factors is [value missing]. ; gestational age The physiological expected baseline value for the ratio of angiogenic factors to anti-angiogenic factors is [value missing]. If the full-term gestational age is 40 weeks, then the first... The formula for calculating the gestational age-specific pathological surge penalty factor for each pregnant woman is as follows:

[0069]

[0070] in, Indicates the first A surge in age-specific pathological factors in pregnant women; Indicates the first The pregnant woman in the first The specific gestational week at the time of the first prenatal checkup; Indicates the first The total number of prenatal checkups for each pregnant woman within the risk assessment time window; Indicates the first The pregnant woman in the first The first prenatal checkup at gestational week The measured data of the ratio of angiogenic factors to anti-angiogenic factors; Indicates the first The pregnant woman in the first The first prenatal checkup at gestational week The measured data of the ratio of angiogenic factors to anti-angiogenic factors; Indicates gestational age The physiological expectation baseline value of the ratio of angiogenic factors to anti-angiogenic factors; Indicates gestational age The physiological expectation baseline value of the ratio of angiogenic factors to anti-angiogenic factors; Represents an exponential function with the natural constant as the base; This represents the maximum value function.

[0071] It should be noted that existing time-series feature extraction algorithms, when processing longitudinal data during pregnancy, can confuse physiological increases with pathological surges by simply calculating the absolute slope of the entire dataset. To address this algorithmic shortcoming, the gestational age-specific pathological surge penalty factor of this invention features a structured design that matches real-world clinical scenarios. First, the data difference part on the left side of the formula departs from the conventional direct difference method. Instead, it differs the rate of change of the measured data with the rate of change of the normal physiological baseline for the same period, and uses this result as the input parameter for the maximum value function. The purpose of this design is to achieve directional filtering during algorithm execution: if the increase in the pregnant woman's indicators does not exceed the baseline change caused by normal physiological development at that gestational week, the difference result will be less than or equal to zero, and the maximum value function will output zero; only when the upward slope of the measured data deviates abnormally from the normal physiological trajectory will this abnormal change be quantified and accumulated. This design solves the problem that traditional linear slope extraction processes cannot effectively remove background noise during the normal physiological period. Secondly, for early-onset abnormalities, the absolute values ​​of indicators in early pregnancy are often small. Directly using absolute differences can lead to these hidden features being easily masked by larger numerical features in late pregnancy during subsequent model fusion. Therefore, this invention constructs an exponential amplifier based on relative deviation rate in the intermediate stage of the formula. By calculating the relative error between the measured value and the physiological baseline and performing exponential operations, the degree of deviation of the measured indicator from the physiological baseline can be mapped to the feature value with non-linear weights, improving the algorithm's sensitivity to minor but pathologically oriented numerical mutations. Finally, since traditional mathematical models lack consideration of the clinical significance of the time dimension, this invention introduces a temporal proportion based on the full-term gestational age constant as a decay product term at the end of the formula. By introducing this constant structure, for the same relative excess pathological surge rate, the smaller the gestational age at occurrence, the larger the penalty multiplier factor. This design transforms the objective understanding in the medical field that early-onset diseases have poor prognoses into an adaptive adjustment mechanism at the algorithm's underlying level, ensuring that the extracted temporal features can objectively reflect the urgency of real clinical intervention corresponding to different stages of disease, effectively improving the machine learning model's ability to identify the risk of abnormal pregnancy outcomes.

[0072] Thus, the process of obtaining gestational age-specific pathological surge penalty factors by weighting the ratio of angiogenic factors to anti-angiogenic factors against physiological baseline deviation and gestational age risk has been completed.

[0073] Step S3: Obtain the cross-system pathological cascade time coupling factor by performing time-series tracing coupling calculation on the abnormal offsets of vascular system indicators and blood system indicators.

[0074] After completing step S2 above, which extracts hidden surge features from single-dimensional physiological indicators, the multi-factor risk assessment model still faces the problem of temporal fragmentation in cross-system pathological cascades when performing multi-dimensional feature fusion. Preeclampsia and other adverse pregnancy outcomes are essentially multi-organ system syndromes. Imbalances in upstream placental angiogenesis, characterized by an abnormal ratio of angiogenic factors to anti-angiogenic factors, can trigger target organ damage in the downstream blood system later in the pathological progression (a typical objective indicator is a sharp drop in platelet count). Existing machine learning algorithms, when processing multivariate time series fusion, typically input simple vector concatenation of independently extracted temporal or statistical features from each dimension. This concatenation method treats different physiological indicators as absolutely independent feature dimensions, losing the temporal phase difference and causal delay linkage information between organ damage in different physiological systems. In fact, if a pregnant woman's vascular endothelial injury index is abnormally high, and within a very short time window there is a sharp drop in blood system indicators relative to their baseline, this cross-system short-term deterioration resonance is an extremely high-risk signal that the condition will immediately endanger the lives of both mother and baby. If the two events occur at different times, they may be independent, sporadic fluctuations. To address the problem that existing algorithm feature splicing mechanisms are detached from the spatiotemporal causal laws of disease development, it is necessary to introduce the calculation of the sequence and urgency of pathological reactions during the multivariate feature fusion stage, construct cross-system temporally coupled feature factors, and transform cross-system causal correlation information into readable fusion weights for the model.

[0075] In summary, this invention first obtains downstream abnormal response characterization data by calculating the relative physiological baseline decline offset of blood system indicator data. Specifically, for any prenatal checkup sequence number within the risk assessment time window for the pregnant woman to be evaluated, the specific gestational age value and platelet count measurement data corresponding to that prenatal checkup sequence number are extracted from the individual measured time series dataset, and the physiological expected baseline value of platelet count corresponding to that specific gestational age value is extracted from the physiological expected baseline reference data. The difference between the physiological expected baseline value of platelet count and the measured platelet count data is used as the blood system decline offset difference assessment, and the result of dividing the blood system decline offset difference assessment by the physiological expected baseline value of platelet count is used as the blood system relative decline offset assessment. When the blood system relative decline offset assessment is less than or equal to a constant 0, the downstream abnormal response characterization data for the corresponding prenatal checkup sequence number is set to a constant 0; when the blood system relative decline offset assessment is greater than a constant 0, the blood system relative decline offset assessment is used as the downstream abnormal response characterization data for the corresponding prenatal checkup sequence number.

[0076] After obtaining the downstream abnormal response characterization data, further processing is performed on the vascular system indicator data to calculate the relative physiological baseline rise offset, thereby obtaining the upstream abnormal trigger characterization data. Specifically, for any prenatal examination number within the risk assessment time window of the pregnant woman to be evaluated, the specific gestational age value and the measured data of the ratio of angiogenic factors to anti-angiogenic factors corresponding to the target traceability node are extracted from the individual measured time series dataset. The physiological expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors corresponding to the specific gestational age value is extracted from the physiological expected baseline reference data. The difference between the measured data of the ratio of angiogenic factors to anti-angiogenic factors and the physiological expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors is used as the vascular system rise offset difference assessment. The result of dividing the vascular system rise offset difference assessment by the calculated physiological expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors is used as the vascular system relative rise offset assessment. When the relative upward offset assessment of the vascular system is less than or equal to a constant 0, the upstream anomaly triggering characterization data of the corresponding target tracing node is set to a constant 0; when the relative upward offset assessment of the vascular system is greater than a constant 0, the relative upward offset assessment of the vascular system is used as the upstream anomaly triggering characterization data of the corresponding target tracing node.

[0077] After obtaining the upstream anomaly trigger characterization data, the downstream anomaly response characterization data and the upstream anomaly trigger characterization data are coupled using time-series decay coupling processing to obtain the cross-system pathological cascade time coupling factor. Specifically, for any prenatal examination number within the risk assessment time window of the pregnant woman to be evaluated, the specific gestational age value corresponding to the downstream response node is extracted from the individual measured time series dataset, and the downstream anomaly response characterization data corresponding to the downstream response node is also extracted. For any downstream response node, from all prenatal examination numbers preceding and identical to the downstream response node, one is selected as the upstream tracing node, and the specific gestational age value corresponding to the upstream tracing node is extracted from the individual measured time series dataset, and the upstream anomaly trigger characterization data corresponding to the upstream tracing node is also extracted. The difference between the specific gestational age value corresponding to the downstream response node and the specific gestational age value corresponding to the upstream tracing node is used as the cross-system time-series interval data. The result of adding the cross-system time-series interval data to a constant 1 is used as the time-series decay denominator, and the result of dividing the constant 1 by the time-series decay denominator is used as the time-series decay weight between the corresponding downstream response node and the upstream tracing node. The calculation result of multiplying the downstream abnormal response characterization data, the upstream abnormal trigger characterization data, and the time-series decay weight is used as the cascade coupling evaluation value between the corresponding downstream response node and the upstream tracing node. All cascade coupling evaluation values ​​corresponding to any downstream response node are summed to obtain the cascade response value corresponding to that downstream response node. The cascade response values ​​corresponding to all downstream response nodes within the risk assessment time window are summed to obtain the cross-system pathological cascade time coupling factor.

[0078] In one implementation, it is assumed that the downstream target organ damage assessment node is the first The serial number of the prenatal checkup was The first upstream pathological trigger tracing node The serial number of the prenatal checkup was ;No. The pregnant woman in the first The gestational week count at the last prenatal checkup was: ;No. The pregnant woman in the first The actual platelet count data at the last prenatal checkup was: ; gestational age The physiological expected baseline value of platelet count at that time is ;No. The pregnant woman in the first The first prenatal checkup at gestational week The measured data of the ratio of angiogenic factors to anti-angiogenic factors at that time were: ; gestational age The physiological expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors at that time is Then the first The formula for calculating the cross-system pathological cascade time coupling factor for a pregnant woman is as follows:

[0079]

[0080] in, Indicates the first Time coupling factors of cross-systemic pathological cascades in a pregnant woman; Indicates gestational age The expected baseline value of platelet count at that time; Indicates the first The pregnant woman in the first Actual platelet count data from the first prenatal checkup; Indicates the first The pregnant woman in the first The first prenatal checkup at gestational week Measured data on the ratio of angiogenic factors to anti-angiogenic factors at the time of angiogenesis; Indicates gestational age The physiological expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors at that time; The first upstream pathology trigger tracing node Prenatal checkup serial number; The first node representing the downstream target organ damage assessment node Prenatal checkup serial number; Indicates the first The total number of prenatal checkups for each pregnant woman within the risk assessment time window; This represents the maximum value function.

[0081] It should be noted that the cross-system pathological cascade time coupling factor of this invention is based on a cross-system pathological time-series coupling design. First, the formula uses a double maximum function to control the direction of cross-index pathological evolution. The outer side of the formula calculates the decline deviation rate of downstream platelet indicators (using the baseline minus the measured value to ensure the direction of deterioration is positive), while the inner side calculates the rise deviation rate of upstream vascular indicators. Only when both upstream and downstream indicators deviate abnormally from the normal physiological baseline in the direction of deterioration is the system recognized as a valid cumulative term, actively filtering out characteristic noise caused by measurement errors or unidirectional random fluctuations. Second, traditional correlation coefficient calculations can only assess the trend consistency of two sets of variables over the entire time axis, failing to capture sudden, short-term causal linkages. Therefore, the formula of this invention incorporates a layer of historical prenatal checkup nodes. The backtracking summation mechanism is implemented, and a time-inverse decay term is introduced. When in the When a prenatal checkup detects damage and deterioration in downstream target organs, the algorithm automatically traces the abnormal initiating source in the upstream vascular system back to the pregnant woman's historical gestational age. The smaller the time interval (i.e., gestational age difference) between a surge in upstream vascular indicators and a sharp drop in downstream platelets, the smaller the denominator of the attenuation term, and the closer the weight obtained by the cross-system product is to its maximum value. Conversely, if the onset times are far apart, the temporal coupling weight exhibits an inversely proportional and sharp attenuation. This mechanism transforms the phenomenon of short-term, continuous damage and resonance across organs into a high-dimensional feature penalty factor in the feature fusion stage of the machine learning model through mathematical mapping at the algorithm's underlying level. This design fills the technical gap in traditional linear feature concatenation when processing multimodal causal evolution sequences, enabling the final predictive model to not only assess the risk level of a single indicator but also to provide accurate early warnings of impending severe illness based on the urgency of the multi-system deterioration sequence.

[0082] Thus, the cross-system pathological cascade time coupling factor was obtained by performing time-series tracing coupling calculations on the abnormal offsets of vascular system indicators and blood system indicators.

[0083] Step S4: The initial linear regression slope features are corrected by applying a gestational age-specific pathological surge penalty factor and combined with a cross-system pathological cascade time coupling factor to construct a fused feature vector.

[0084] After obtaining the independently calculated gestational age-specific pathological surge penalty factor and the cross-system pathological cascade temporal coupling factor, these optimization factors need to be applied to the existing extreme gradient boosting tree model input feature construction framework. Current techniques extract the linear regression slopes of various measured data sequences using the least squares method and then concatenate the vectors. To embed the penalty mechanism and temporal coupling mechanism of the optimization factors into the model and avoid disrupting the original model's understanding of the basic data trends, the gestational age-specific pathological surge penalty factor is used as a product weight to correct the initial linear regression slope of the original angiogenesis factor to anti-angiogenesis factor ratio, thus imbuing the slope feature with clinical hazard perception. Simultaneously, the cross-system pathological cascade temporal coupling factor is treated as a newly generated higher-order causal feature dimension and directly incorporated into the final feature concatenation set along with the corrected features and the initial linear regression slope of platelet count, to compensate for the missing cross-system collaborative information in the original concatenation model.

[0085] Specifically, firstly, based on the measured gestational age values ​​and the ratio of angiogenic factors to anti-angiogenic factors corresponding to each prenatal checkup in the individual measured time-series dataset, linear regression fitting was performed to obtain the initial linear regression slope characteristic corresponding to the ratio of angiogenic factors to anti-angiogenic factors. Then, based on the measured gestational age values ​​and platelet count data corresponding to each prenatal checkup in the individual measured time-series dataset, linear regression fitting was performed to obtain the initial linear regression slope characteristic corresponding to the platelet count.

[0086] After obtaining the initial linear regression slope features corresponding to the ratio of angiogenic factors to anti-angiogenic factors and the initial linear regression slope features corresponding to platelet count, the result of multiplying the gestational age-specific pathological surge penalty factor by a preset first scaling factor and adding it to a constant 1 is used as the slope correction coefficient. In this embodiment, the first scaling factor is set to 0.5. The result of multiplying the initial linear regression slope feature corresponding to the ratio of angiogenic factors to anti-angiogenic factors by the slope correction coefficient is used as the corrected slope feature of the ratio of angiogenic factors to anti-angiogenic factors. The result of multiplying the cross-system pathological cascade time coupling factor by a preset second scaling factor is used as the cross-system coupling feature value. In this embodiment, the second scaling factor is set to 0.2. The corrected slope feature of the ratio of angiogenic factors to anti-angiogenic factors, the initial linear regression slope feature corresponding to platelet count, and the cross-system coupling feature value are then concatenated to obtain a fused feature vector.

[0087] Thus, the process of constructing a fused feature vector by correcting the initial linear regression slope features with a gestational age-specific pathological surge penalty factor and combining it with a cross-system pathological cascade time coupling factor was completed.

[0088] Step S5: Obtain pregnancy outcome risk assessment results by performing machine learning prediction on the fused feature vector.

[0089] After completing the aforementioned feature reconstruction and fusion, a fused feature vector has been obtained that can simultaneously characterize both the gestational age-specific pathological surge of a single key indicator and the short-term cascade deterioration of multiple systems. Compared with the feature sets in existing technologies that consist only of the initial linear regression slopes of each individual indicator, the fused feature vector constructed in this invention not only retains the basic information of the overall trend of longitudinal prenatal examination data, but also further introduces a gestational age-specific pathological surge penalty factor that amplifies early-onset occult deterioration signals, and a cross-system pathological cascade temporal coupling factor that reflects the temporal urgency between vascular system abnormalities and blood system damage. Therefore, this fused feature vector forms a highly discriminative input feature set that better reflects the true pathological evolution of pregnancy complications, providing a more reliable data foundation for subsequent machine learning models to identify risks.

[0090] Specifically, the fused feature vector corresponding to the pregnant woman to be evaluated is input into a pre-trained machine learning prediction model, preferably an extreme gradient boosting tree model. The extreme gradient boosting tree model constructs a training set based on historical pregnant woman samples. These samples include at least the fused feature vectors corresponding to each historical pregnant woman and the actual pregnancy outcome label for each historical pregnant woman. The actual pregnancy outcome label characterizes whether the corresponding historical pregnant woman experienced the target adverse pregnancy outcome. By using the training set for supervised training of the extreme gradient boosting tree model, the model learns the mapping relationship between different fused feature vectors and actual pregnancy outcomes, thereby establishing a classification prediction model for risk assessment.

[0091] During the model prediction phase, the extreme gradient boosting tree model receives the input fused feature vector and then uses multiple internal decision trees to perform layer-by-layer node partitioning and gain evaluation on each dimension of the fused feature vector, identifying feature combinations highly correlated with adverse pregnancy outcomes. Since the modified angiogenic factor to anti-angiogenic factor ratio slope can reflect the clinical risk of abnormal surge signals at different gestational weeks, and the cross-system coupling feature value can reflect the urgency of the cascade deterioration between upstream vascular abnormalities and downstream blood system damage, the model can utilize the tree node splitting process to perform a more refined nonlinear distinction between occult high-risk samples and ordinary fluctuating samples, thereby improving its ability to identify individuals at high risk of adverse pregnancy outcomes such as severe preeclampsia.

[0092] Preferably, the classification score output by the extreme gradient boosting tree model is normalized and mapped to obtain the risk probability value of the pregnant woman to be evaluated for the target adverse pregnancy outcome. Further, this risk probability value can be compared with a preset risk judgment threshold: when the risk probability value is greater than or equal to the preset risk judgment threshold, the pregnant woman to be evaluated is determined to be in a high-risk group for pregnancy outcome, and a high-risk warning result is output; when the risk probability value is less than the preset risk judgment threshold, the pregnant woman to be evaluated is determined to be in a low-risk group for pregnancy outcome, and a routine follow-up result is output. As a preferred embodiment, the preset risk judgment threshold can be set to 0.5; in application scenarios with higher sensitivity requirements for identifying high-risk pregnant women, the preset risk judgment threshold can be further lowered to 0.4 to improve the early warning capability for high-risk individuals.

[0093] Ultimately, the pregnancy outcome risk assessment results output in this step include at least the probability value of adverse pregnancy outcomes for the pregnant woman being assessed and the corresponding risk level determination. If necessary, the risk probability value, risk level determination, and corresponding prenatal checkup time information can be further output to a clinical monitoring terminal or obstetric auxiliary decision-making system to provide quantitative evidence for physicians to determine the timing of intensive monitoring, hospitalization intervention, or termination of pregnancy. Through the above methods, this invention achieves a complete mapping from fused feature vectors to clinical risk results, enabling machine learning models to more accurately identify the true high-risk states corresponding to multifactorial abnormal evolution during pregnancy.

[0094] This completes the process of obtaining pregnancy outcome risk assessment results through machine learning prediction of fused feature vectors.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis, characterized in that, The method includes: Step S1: Obtain individual measured time series datasets and physiological expected baseline reference data by acquiring and preprocessing multimodal longitudinal sequence data and physiological baseline data; Step S2: Obtain the gestational age-specific pathological surge penalty factor by calculating the ratio of angiogenic factors to anti-angiogenic factors relative to the physiological baseline and gestational age risk weighting. Step S3: Obtain the cross-system pathological cascade time coupling factor by performing time-series tracing coupling calculation on the abnormal offsets of vascular system indicators and blood system indicators; Step S4: Obtain the fused feature vector by correcting the initial linear regression slope features with a gestational age-specific pathological surge penalty factor and combining it with a cross-system pathological cascade time coupling factor; Step S5: Obtain pregnancy outcome risk assessment results by performing machine learning prediction on the fused feature vector.

2. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 1, characterized in that, The process of acquiring and preprocessing multimodal longitudinal sequence data and physiological baseline data to obtain individual measured time series datasets and expected physiological baseline reference data includes: The process involves acquiring basic profile information and multiple prenatal checkup records of the pregnant woman to be assessed within the risk assessment time window. The basic profile information includes at least the pregnant woman's identification information, the total number of prenatal checkups, and the chronological order of each checkup. For each checkup number, the specific gestational age, measured angiogenesis factor to anti-angiogenesis factor ratio, and measured platelet count data for that checkup are obtained. These data are then linked and stored chronologically to obtain multimodal longitudinal sequence data for the pregnant woman to be assessed. The multimodal longitudinal sequence data undergoes format standardization, outlier removal, and missing value interpolation to complete the data, resulting in an individual measured time series dataset. Based on a pre-constructed retrospective clinical database of healthy pregnant women, reference data on the ratio of angiogenic factors to anti-angiogenic factors and platelet counts at different gestational weeks were extracted from healthy pregnant women without records of pregnancy complications or adverse pregnancy outcomes. The reference data on the ratio of angiogenic factors to anti-angiogenic factors and platelet counts were then smoothed to construct a physiological expectation baseline reference library that varies with gestational week. Based on the specific gestational week values ​​corresponding to each prenatal checkup in the individual measured time-series dataset, the physiological expectation baseline values ​​for the ratio of angiogenic factors to anti-angiogenic factors and platelet counts at the corresponding gestational week were matched and extracted from the physiological expectation baseline reference library to obtain physiological expectation baseline reference data.

3. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 1, characterized in that, The method involves calculating the ratio of angiogenic factors to anti-angiogenic factors based on relative physiological baseline deviation and gestational age risk weighting to obtain gestational age-specific pathological surge penalty factors, including: By performing relative offset processing based on the physiological expectation baseline on the ratio data of angiogenic factors and anti-angiogenic factors, pathological excess change characterization data are obtained. By applying gestational age risk weighting and nonlinear enhancement processing to the pathological excess change characterization data, a gestational age-specific pathological surge penalty factor was obtained.

4. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 3, characterized in that, The process involves performing relative offset processing on the ratio data of angiogenic factors and anti-angiogenic factors based on the physiological expectation baseline to obtain pathological excess change characterization data, including: For the pregnant woman to be evaluated, the first and second prenatal examination numbers of any two adjacent prenatal examinations within the risk assessment time window are extracted from the individual measured time series dataset. The measured data of the previous specific gestational week value and the ratio of the previous angiogenic factor to anti-angiogenic factor corresponding to the previous prenatal examination number are extracted from the individual measured time series dataset. The measured data of the next specific gestational week value and the ratio of the next angiogenic factor to anti-angiogenic factor corresponding to the next specific gestational week value are extracted from the physiological expected baseline reference data. The difference between the measured ratio of the next angiogenic factor to anti-angiogenic factor and the measured ratio of the previous angiogenic factor to anti-angiogenic factor is taken as the measured difference. The difference between the value of the next specific gestational week and the value of the previous specific gestational week is taken as the gestational week interval data. The result of dividing the measured difference by the gestational week interval data is used as the measured change rate assessment between adjacent prenatal examinations. The difference between the physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor in the latter and the physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor in the former is taken as the baseline difference. The baseline difference is divided by the calculation result of the gestational week interval data as the baseline change rate assessment between adjacent prenatal checkups. The difference between the measured rate of change assessment and the baseline rate of change assessment is used as the relative deviation difference assessment. When the relative deviation difference assessment is less than or equal to a constant 0, the pathological excess change characterization data between adjacent prenatal examinations is set to a constant 0; when the relative deviation difference assessment is greater than a constant 0, the relative deviation difference assessment is used as the pathological excess change characterization data between adjacent prenatal examinations.

5. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 3, characterized in that, The process involves applying gestational age-specific pathological surge penalty factors to the pathological excess change characterization data through gestational age risk weighting and nonlinear enhancement processing, including: For the pregnant woman to be evaluated, the second prenatal examination number in any two adjacent prenatal examinations within the risk assessment time window is extracted from the individual measured time series dataset. The specific gestational week value corresponding to the second prenatal examination number and the measured data of the ratio of angiogenic factor to anti-angiogenic factor are extracted from the individual measured time series dataset. The physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor corresponding to the specific gestational week value is extracted from the physiological expected baseline reference data. At the same time, the pathological excess change characterization data corresponding to the two adjacent prenatal examinations are extracted. The difference between the measured ratio of the next angiogenic factor to anti-angiogenic factor and the physiologically expected baseline value of the ratio of the next angiogenic factor to anti-angiogenic factor is used as the relative deviation difference assessment. The relative deviation difference assessment divided by the calculated physiologically expected baseline value of the ratio of the next angiogenic factor to anti-angiogenic factor is used as the relative deviation rate assessment. The relative deviation rate assessment is then subjected to an exponential mapping with the natural constant as the base to obtain the nonlinear enhancement weights corresponding to the two adjacent prenatal examinations. The difference between the preset full-term gestational age constant and the value of the next specific gestational age is used as the early onset risk assessment value. The early onset risk assessment value is divided by the preset full-term gestational age constant to obtain the gestational age risk weighting coefficient for the corresponding two adjacent prenatal checkups. The pathological surge penalty value is calculated by multiplying the pathological excess change characterization data, the nonlinear enhancement weight, and the gestational age risk weighting coefficient. The pathological surge penalty values ​​corresponding to all two adjacent prenatal examinations within the risk assessment time window are summed to obtain the gestational age-specific pathological surge penalty factor.

6. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 1, characterized in that, The method of obtaining cross-system pathological cascade time coupling factors by performing time-series tracing coupling calculations on abnormal deviations of vascular system indicators and abnormal deviations of blood system indicators includes: By performing relative physiological baseline decline offset calculation on blood system indicator data, downstream abnormal response characterization data are obtained; By performing relative physiological baseline offset tracing calculation on vascular system indicator data, upstream abnormal trigger characterization data are obtained; By performing time-series decay coupling processing on downstream abnormal response characterization data and upstream abnormal trigger characterization data, the cross-system pathological cascade time coupling factor is obtained.

7. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 6, characterized in that, The process involves calculating the relative physiological baseline shift of blood system indicator data to obtain downstream abnormal response characterization data, including: For any prenatal check-up number of a pregnant woman to be evaluated within the risk assessment time window, extract the specific gestational age value and platelet count measurement data corresponding to the prenatal check-up number from the individual measured time series dataset, and extract the physiological expected baseline value of platelet count corresponding to the specific gestational age value from the physiological expected baseline reference data. The difference between the expected baseline value of platelet count and the measured platelet count data is used as the blood system decline deviation difference assessment, and the blood system decline deviation difference assessment divided by the calculated result of the expected baseline value of platelet count is used as the blood system relative decline deviation assessment. When the relative downward offset assessment of the blood system is less than or equal to a constant 0, the downstream abnormal response characterization data of the corresponding prenatal examination number is set to a constant 0; when the relative downward offset assessment of the blood system is greater than a constant 0, the relative downward offset assessment of the blood system is used as the downstream abnormal response characterization data of the corresponding prenatal examination number.

8. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 6, characterized in that, The process involves performing relative physiological baseline shift calculations on vascular system indicator data to obtain upstream anomaly trigger characterization data, including: For any prenatal examination number within the risk assessment time window of the pregnant woman to be assessed, the specific gestational age value and the measured data of the ratio of angiogenic factor to anti-angiogenic factor corresponding to the target traceability node are extracted from the individual measured time series dataset. The physiological expected baseline value of the ratio of angiogenic factor to anti-angiogenic factor corresponding to the specific gestational age value is extracted from the physiological expected baseline reference data. The difference between the measured data of the ratio of angiogenic factors to anti-angiogenic factors and the physiologically expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors is used as the vascular system upward deviation difference assessment. The vascular system upward deviation difference assessment is divided by the calculated result of the physiologically expected baseline value of the ratio of angiogenic factors to anti-angiogenic factors as the vascular system relative upward deviation assessment. When the relative upward offset assessment of the vascular system is less than or equal to a constant 0, the upstream anomaly triggering characterization data of the corresponding target tracing node is set to a constant 0; when the relative upward offset assessment of the vascular system is greater than a constant 0, the relative upward offset assessment of the vascular system is used as the upstream anomaly triggering characterization data of the corresponding target tracing node.

9. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 6, characterized in that, The process of obtaining a cross-system pathological cascade time coupling factor by performing time-series decay coupling processing on downstream abnormal response characterization data and upstream abnormal trigger characterization data includes: For any prenatal checkup number of a pregnant woman to be evaluated within the risk assessment time window, the specific gestational week value corresponding to the downstream response node is extracted from the individual measured time series dataset, and the downstream abnormal response characterization data corresponding to the downstream response node is extracted. For any downstream response node, among all the prenatal examination serial numbers that are the same as the downstream response node, select one of them as the upstream traceability node, extract the specific gestational age value corresponding to the upstream traceability node from the individual measured time series dataset, and extract the upstream abnormal trigger characterization data corresponding to the upstream traceability node. The difference between the specific gestational age value corresponding to the downstream response node and the specific gestational age value corresponding to the upstream trace node is used as the cross-system time interval data. The result of adding the cross-system time interval data to the constant 1 is used as the time decay denominator. The result of dividing the constant 1 by the time decay denominator is used as the time decay weight between the corresponding downstream response node and the upstream trace node. The calculation result of multiplying the downstream abnormal response characterization data, the upstream abnormal trigger characterization data, and the time-series decay weight is used as the cascade coupling evaluation value between the corresponding downstream response node and the upstream tracing node. All cascade coupling evaluation values ​​corresponding to any downstream response node are summed to obtain the cascade response value corresponding to that downstream response node. The cascade response values ​​corresponding to all downstream response nodes within the risk assessment time window are summed to obtain the cross-system pathological cascade time coupling factor.

10. The method for assessing the risk of pregnancy outcomes in pregnant women based on multifactorial feature analysis according to claim 1, characterized in that, The process involves correcting the initial linear regression slope features with a gestational age-specific pathological surge penalty factor and then fusing them with a cross-system pathological cascade time coupling factor to construct a fused feature vector, including: Based on the specific gestational age values ​​and the measured ratio of angiogenic factors to anti-angiogenic factors corresponding to each prenatal checkup in the individual measured time series dataset, linear regression fitting was performed to obtain the initial linear regression slope feature corresponding to the ratio of angiogenic factors to anti-angiogenic factors; based on the specific gestational age values ​​and the measured platelet count data corresponding to each prenatal checkup in the individual measured time series dataset, linear regression fitting was performed to obtain the initial linear regression slope feature corresponding to the platelet count. The result of multiplying the gestational age-specific pathological surge penalty factor by the preset first scaling factor and adding it to the constant 1 is used as the slope correction coefficient. The result of multiplying the initial linear regression slope feature corresponding to the ratio of angiogenic factor to anti-angiogenic factor by the slope correction coefficient is used as the corrected slope feature of the ratio of angiogenic factor to anti-angiogenic factor. The result of multiplying the cross-system pathological cascade time coupling factor by a preset second scaling factor is used as the cross-system coupling feature value; The modified angiogenesis factor to anti-angiogenesis factor ratio slope feature, the initial linear regression slope feature corresponding to platelet count, and cross-system coupling feature value are concatenated to obtain a fused feature vector.

Images