A method for dynamically predicting pre-eclampsia preterm birth risk based on pregnancy monitoring indicators

Abstract

The present application relates to the field of medical data analysis, and more particularly to a method for dynamically predicting the risk of pre-eclampsia premature birth based on monitoring indicators during pregnancy, which comprises: obtaining an individual measured time series set and a basic time weighted average feature; obtaining a recent transient decompensation momentum factor by performing recent deterioration dynamics analysis on monitoring data near the current evaluation time; obtaining a steady-state fracture and barrier breakdown factor by jointly evaluating the evolution trend and fluctuation potential of the historical monitoring sequence; obtaining a reconstructed fusion feature vector by fusing and reconstructing the basic time weighted average feature and the steady-state fracture and barrier breakdown factor; and obtaining a dynamic prediction result of the risk of premature birth by inputting the reconstructed fusion feature vector into a Cox proportional hazards model for risk prediction, thereby solving the problem that the existing prediction method based on time weighted average and linear feature fusion cannot effectively identify recent decompensation deterioration signals and maternal-fetal state deviation risks.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, and in particular to a method for dynamic prediction of the risk of preeclampsia and premature birth based on pregnancy monitoring indicators. Background Technology

[0002] In the clinical practice of dynamically predicting the risk of preeclampsia and premature birth, in order to assess the dynamic risk of the disease in pregnant women as pregnancy progresses and to assist in determining the timing of termination of pregnancy, it is usually necessary to continuously collect physiological laboratory indicators of pregnant women and fetal Doppler ultrasound indicators. Because the number of monitoring sessions and the follow-up intervals vary among pregnant women in actual follow-up, current techniques typically use a time-weighted averaging algorithm to process the irregular longitudinal time series, extracting a uniform average exposure feature from each monitoring data point, and then combining this with a Cox proportional hazards model for linear feature weighting to generate a nomogram score for dynamically predicting the risk of preeclampsia.

[0003] However, the pathological progression of preeclampsia exhibits a clear nonlinear decompensation characteristic. During the compensatory phase, relevant indicators change relatively slowly, while in the critical stage approaching preterm intervention, key indicators often show an accelerated upward trend. Existing techniques, based on time-weighted average feature extraction, perform mean smoothing on the full-cycle fluctuations, making it easy for the rapidly deteriorating critical decompensation acceleration signals in the near term to be diluted and masked by relatively stable historical data in the long term, resulting in a lag in model warnings.

[0004] Furthermore, in clinical practice, to prolong gestation, pregnant women are often given antihypertensive drugs and other treatments, which may cause a false sense of stability in maternal apparent indicators. However, worsening placental perfusion can still lead to a rapid increase in fetal hypoxia-related indicators. Existing Cox prediction models, when performing multivariate fusion, mainly rely on the linear addition of feature scores. This can easily cause the low-risk contribution corresponding to the improvement of maternal indicators to cancel out the high-risk contribution corresponding to the deterioration of fetal indicators at the underlying level. This masks the danger signals of discrepancies between maternal appearance and the actual fetal condition, reducing the accuracy of dynamic prediction of preterm birth risk. Therefore, improving the ability to identify recent decompensated deterioration signals in preeclampsia and the risk of discrepancies between maternal and fetal status has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the present invention aims to propose a dynamic prediction method for the risk of preeclampsia and premature birth based on pregnancy monitoring indicators, in order to solve the problem that existing prediction methods based on the fusion of time-weighted average and linear features cannot effectively identify recent decompensated deterioration signals and the risk of deviation between maternal and fetal status.

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

[0007] A method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators, the method comprising:

[0008] Step S1: By acquiring and processing the dynamic monitoring time series data and clinical baseline parameters of pregnant women with preeclampsia, the individual measured time series set and baseline time-weighted average characteristics are obtained;

[0009] Step S2: Obtain the recent transient decompensation momentum factor by performing a recent deterioration dynamics analysis on the monitoring data close to the current assessment time;

[0010] Step S3: Obtain steady-state fracture and barrier breakdown factors by jointly evaluating the evolution trend and fluctuation potential energy of historical monitoring sequences;

[0011] Step S4: Obtain the reconstructed fused feature vector by fusing and reconstructing the basic time-weighted average features with the steady-state fracture and barrier breakdown factors;

[0012] Step S5: By inputting the reconstructed fusion feature vector into the Cox proportional hazards model for risk prediction, dynamic prediction results of preterm birth risk are obtained.

[0013] Furthermore, by acquiring and processing the dynamic monitoring time-series data and clinical baseline parameters of pregnant women with preeclampsia, individual measured time-series sets and baseline time-weighted average characteristics are obtained, including:

[0014] For pregnant women with preeclampsia awaiting evaluation, multiple routine clinical follow-up records and biochemical test records from the time of diagnosis or registration of preeclampsia to the current evaluation time are retrieved from the electronic medical record system of the medical institution. The total number of monitoring times up to the current evaluation time is obtained, and the monitoring timestamps are obtained in chronological order, where the monitoring timestamps are represented by gestational weeks. For each monitoring timestamp, the actual serum uric acid data at the corresponding monitoring timestamp is extracted, and the monitoring timestamps are correlated with the actual serum uric acid data corresponding to each monitoring timestamp to obtain the original dynamic monitoring time series data of the pregnant women with preeclampsia.

[0015] The original dynamic monitoring time series data were processed by data anonymization, format cleaning and missing value interpolation and completion, and the processed monitoring timestamps and corresponding serum uric acid measured data were used as individual measured time series sets.

[0016] The clinical risk upper limit threshold corresponding to serum uric acid is retrieved from the electronic medical record system of medical institutions. Based on the time interval between two adjacent monitoring timestamps in the individual measured time series set and the corresponding measured serum uric acid data, a time-weighted average is calculated to obtain the basic time-weighted average feature.

[0017] Furthermore, the step of obtaining the recent transient decompensation momentum factor by performing recent deterioration dynamics analysis on monitoring data close to the current assessment time includes:

[0018] By performing time-series localization and difference extraction on continuous monitoring data close to the current assessment time, recent change slope data is obtained;

[0019] By combining recent change slope data with clinical risk upper limit threshold, the recent transient decompensated momentum factor is obtained.

[0020] Furthermore, the step of obtaining recent change slope data by performing time-series localization and difference extraction processing on continuous monitoring data close to the current evaluation time includes:

[0021] For any pregnant woman with preeclampsia whose condition is to be evaluated, extract the last monitoring timestamp closest to the current evaluation time and the penultimate monitoring timestamp adjacent to the last monitoring timestamp from the individual measured time series set, and extract the measured serum uric acid data corresponding to the last monitoring timestamp and the measured serum uric acid data corresponding to the penultimate monitoring timestamp respectively.

[0022] The difference between the measured serum uric acid data corresponding to the last monitoring timestamp and the measured serum uric acid data corresponding to the penultimate monitoring timestamp is taken as the recent indicator change, and the difference between the last monitoring timestamp and the penultimate monitoring timestamp is taken as the recent monitoring time interval.

[0023] The recent index change is used as the numerator, the recent monitoring time interval is used as the denominator, and the corresponding fraction is used as the recent change slope data.

[0024] Furthermore, the method of obtaining the recent transient decompensated momentum factor by jointly gating the recent slope data with the upper limit threshold of clinical risk includes:

[0025] For any pregnant woman with preeclampsia whose condition is to be evaluated, when the recent change slope data is less than or equal to constant 0, the unidirectional deterioration slope assessment corresponding to the recent change slope data is set to constant 0; when the recent change slope data is greater than constant 0, the recent change slope data is used as the corresponding unidirectional deterioration slope assessment.

[0026] The serum uric acid measured data corresponding to the last monitoring timestamp of the preeclamptic pregnant woman were extracted, and the upper limit threshold of clinical risk corresponding to serum uric acid was extracted from the clinical baseline parameters. The difference between the measured serum uric acid data corresponding to the last monitoring timestamp and the upper limit threshold of clinical risk was used as the numerator, and the upper limit threshold of clinical risk was used as the denominator. The corresponding fraction was used as the risk threshold deviation assessment.

[0027] The result of multiplying the danger threshold deviation assessment by the preset slope activation sensitivity constant is used as the exponential gating input value. The negative of the exponential gating input value is subjected to exponential mapping with the natural constant as the base. The constant 1 is added to the corresponding exponential mapping result, and the result of dividing the constant 1 by the sum is used as the threshold gating coefficient.

[0028] The result of multiplying the unidirectional deterioration slope assessment by the threshold gating coefficient is used as the recent transient decompensated momentum factor.

[0029] Furthermore, the method of jointly evaluating the evolution trend and fluctuation potential energy of historical monitoring sequences to obtain steady-state fracture and barrier breakdown factors includes:

[0030] By extracting upward fluctuations and performing time decay processing on historical monitoring sequence data, historical upward fluctuation potential energy data can be obtained.

[0031] By performing first-to-last evolution trend analysis on historical monitoring sequence data, historical overall evolution gradient data can be obtained.

[0032] By jointly enhancing recent transient decompensated momentum factors, historical upward wave potential energy data, and historical overall evolution gradient data, steady-state fracture and barrier breakdown factors are obtained.

[0033] Furthermore, the step of extracting and time-decaying historical monitoring sequence data to obtain historical upward fluctuation potential energy data includes:

[0034] For any pregnant woman with preeclampsia in any state to be evaluated, the penultimate monitoring timestamp and all historical monitoring timestamps before the penultimate monitoring timestamp are extracted from the individual measured time series set, and the serum uric acid measured data corresponding to each historical monitoring timestamp are extracted.

[0035] For any adjacent historical monitoring timestamp, the difference between the measured serum uric acid data corresponding to the next historical monitoring timestamp and the measured serum uric acid data corresponding to the previous historical monitoring timestamp is taken as the historical indicator change, the difference between the next historical monitoring timestamp and the previous historical monitoring timestamp is taken as the historical monitoring time interval, the historical indicator change is taken as the numerator, the historical monitoring time interval is taken as the denominator, and the corresponding fraction is taken as the historical change slope assessment.

[0036] When the historical change slope assessment is less than or equal to a constant 0, the upward volatility assessment corresponding to the historical change slope assessment is set to a constant 0; when the historical change slope assessment is greater than a constant 0, the historical change slope assessment is used as the corresponding upward volatility assessment.

[0037] For any subsequent historical monitoring timestamp, the constant 1 is divided by the calculated result of the difference between the current evaluation time and the subsequent historical monitoring timestamp, and then added to the constant 1. The corresponding sum is then subjected to a natural logarithm operation to obtain the time decay weight corresponding to the subsequent historical monitoring timestamp.

[0038] The result of multiplying the upward fluctuation assessment by the time decay weight is used as the weighted upward fluctuation assessment for the corresponding historical interval. The weighted upward fluctuation assessments corresponding to all historical intervals are squared and summed. The sum is divided by the number of historical intervals and then square rooted to obtain the historical upward fluctuation potential energy data.

[0039] Furthermore, the step of obtaining historical overall evolution gradient data by performing first-to-last evolution trend analysis on historical monitoring sequence data includes:

[0040] For any pregnant woman with preeclampsia in any state to be evaluated, the first monitoring timestamp, the second to last monitoring timestamp, the serum uric acid measured data corresponding to the first monitoring timestamp, and the serum uric acid measured data corresponding to the second to last monitoring timestamp are extracted from the individual measured time series set;

[0041] The difference between the measured serum uric acid data corresponding to the penultimate monitoring timestamp and the measured serum uric acid data corresponding to the first monitoring timestamp is taken as the historical overall indicator change, and the difference between the penultimate monitoring timestamp and the first monitoring timestamp is taken as the historical overall monitoring time span.

[0042] The change in the overall historical index is used as the numerator, the monitoring time span of the overall historical index is used as the denominator, and the corresponding fraction is used as the historical overall evolution gradient data.

[0043] Furthermore, the method of obtaining steady-state fracture and barrier breakdown factors by jointly enhancing recent transient decompensated momentum factors, historical upward wave potential energy data, and historical overall evolution gradient data includes:

[0044] For any pregnant woman with preeclampsia in any state to be evaluated, the result of multiplying the recent transient decompensated momentum factor by the historical overall evolutionary gradient data is used as the trend inner product assessment. The result of multiplying the absolute value of the recent transient decompensated momentum factor by the absolute value of the historical overall evolutionary gradient data is added to a preset small positive constant to obtain the trend normalized denominator assessment.

[0045] The trend inner product assessment is used as the numerator, the trend normalized denominator assessment is used as the denominator, and the corresponding fraction is used as the trend direction similarity assessment. The difference between the constant 1 and the trend direction similarity assessment is used as the trend deflection intensity assessment, and the trend deflection intensity assessment is mapped by the hyperbolic tangent function to obtain the trend deflection enhancement coefficient.

[0046] The square of the recent transient decompensated momentum factor is used as the momentum energy level assessment. The calculation result of adding the square of the historical upward wave potential energy data to the preset physiological wave stability constant is used as the historical barrier base assessment. The momentum energy level assessment is used as the numerator, the historical barrier base assessment is used as the denominator, and the corresponding fraction is used as the barrier penetration strength assessment. The barrier penetration strength assessment is then subjected to an exponential mapping with the natural constant as the base to obtain the barrier penetration enhancement coefficient.

[0047] The result of multiplying the recent transient decompensated momentum factor, trend deflection enhancement coefficient, and barrier penetration enhancement coefficient is used as the steady-state fracture and barrier breakdown factors.

[0048] Furthermore, the process of fusing and reconstructing the basic time-weighted average features with the steady-state fracture and barrier breakdown factors to obtain the reconstructed fused feature vector includes:

[0049] For any pregnant woman with preeclampsia in any state to be evaluated, the result of multiplying the steady-state fracture and barrier breakdown factors by the preset scaling factor is used as a dynamic amplification evaluation.

[0050] The result of adding the constant 1 to the dynamic amplification evaluation is used as the feature reconstruction weight;

[0051] The result of multiplying the base time-weighted average feature by the feature reconstruction weight is used as the reconstruction fusion feature;

[0052] Extract the conventional cross-sectional clinical features corresponding to the preeclamptic pregnant woman, and combine the reconstructed fusion features with the conventional cross-sectional clinical features to obtain the reconstructed fusion feature vector.

[0053] Compared with the prior art, the present invention has the following advantages:

[0054] This invention presents a dynamic prediction method for the risk of preeclampsia and premature birth based on pregnancy monitoring indicators. While retaining the fundamental advantage of existing time-weighted average features in characterizing long-term pathological burden, it further introduces a targeted identification mechanism for recent rapid deterioration trends. This mechanism can more sensitively capture abnormal signals indicating a shift from a stable to a decompensated state in key indicators as the clinical risk window approaches. Compared to traditional methods that rely solely on the average exposure over the entire pregnancy cycle for risk modeling, this invention avoids the dilution and masking of recent rapid deterioration information by early stable data. This allows the model to identify critical deterioration stages more promptly and more closely reflect the actual pathological progression, improving the early warning sensitivity for short-term preeclampsia risk changes in pregnant women with preeclampsia and providing obstetricians with more reliable quantitative evidence for timely intervention. Furthermore, this invention does not simply linearly superimpose monitoring indicators but jointly characterizes recent deterioration momentum, historical evolution trends, and individual past adaptive capacity to fluctuations. This gives the risk characteristics an adaptive expression capacity to individual differences in compensatory capacity and the intensity of disease mutations. This technique can more accurately distinguish between long-term, slow fluctuations and truly clinically significant steady-state disruption events, reducing misjudgments caused by superficial stability or general fluctuations, and enhancing the model's specificity in identifying high-risk individuals. In practical applications, this dynamic feature representation method, which considers both long-term baseline load and recent acute changes, helps improve the accuracy and stability of preterm birth risk prediction results, and enhances the application value of dynamic nomograms in continuous clinical follow-up, determination of the timing of pregnancy termination, and stratified management of high-risk patients. Attached Figure Description

[0055] 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:

[0056] Figure 1 This is a flowchart illustrating a method for dynamically predicting the risk of preeclampsia and premature birth based on pregnancy monitoring indicators, as described in an embodiment of the present invention. Detailed Implementation

[0057] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0058] See Figure 1 This is a flowchart of a method for dynamically predicting the risk of preeclampsia and premature birth based on pregnancy monitoring indicators, as provided in Embodiment 1 of the present invention. Figure 1 As shown, a method for dynamically predicting the risk of preeclampsia and premature birth based on pregnancy monitoring indicators may include:

[0059] Step S1 involves acquiring and processing the dynamic monitoring time-series data and clinical baseline parameters of pregnant women with preeclampsia to obtain individual measured time series sets and baseline time-weighted average characteristics.

[0060] First, for pregnant women with preeclampsia awaiting evaluation, multiple routine clinical follow-up records and biochemical test records from the time of diagnosis or registration of preeclampsia to the current evaluation time are retrieved from the medical institution's electronic medical record system. The total number of monitoring sessions up to the current evaluation time is obtained, and the monitoring timestamps are obtained in chronological order, with each monitoring timestamp represented by gestational weeks. For each monitoring timestamp, the corresponding serum uric acid measurement data is extracted, and the monitoring timestamps are correlated with the corresponding serum uric acid measurement data to obtain the original dynamic monitoring time-series data of the pregnant women with preeclampsia.

[0061] The original dynamic monitoring time series data were processed by data anonymization, format cleaning and missing value interpolation, and the processed monitoring timestamps and corresponding serum uric acid measured data were used as individual measured time series sets.

[0062] The clinical risk upper limit threshold corresponding to serum uric acid is retrieved from the electronic medical record system of medical institutions. Based on the time interval between two adjacent monitoring timestamps in the individual measured time series set and the corresponding measured serum uric acid data, a time-weighted average is calculated to obtain the basic time-weighted average feature.

[0063] It should be noted that, in the embodiments of the present invention, a weighted average is calculated by multiplying the mean of two adjacent monitoring data by the time interval, summing the sums, and then dividing by the total observation time.

[0064] Thus far, the acquisition and processing of dynamic monitoring time-series data and clinical baseline parameters of pregnant women with preeclampsia have been completed, resulting in individual measured time series sets and baseline time-weighted average characteristics.

[0065] Step S2: By performing a recent deterioration dynamics analysis on the monitoring data close to the current assessment time, the recent transient decompensation momentum factor is obtained.

[0066] To overcome the drawback of traditional time-weighted averaging algorithms that dilute recent sharp deterioration with historical stable data during feature extraction, it is necessary to remove the interference from long-term historical data and extract the dynamic state of local time segments close to the current prediction time separately. When determining whether the disease has entered the decompensated stage, the key is to capture the speed and degree of deviation of the indicator from the clinical control safety line. Therefore, this step extracts a small time window closest to the current assessment time. Using the ratio of the displacement increment of the value approaching the danger boundary within this cross-section to the time taken, and coupling it with a nonlinear activation function based on the clinical warning line, a transient scalar reflecting the unidirectional deterioration rate at the current moment is constructed, thus preliminarily quantifying the recent risk mutation signal.

[0067] First, by performing time-series localization and difference extraction on continuous monitoring data close to the current assessment time, recent change slope data is obtained. Specifically, for any pregnant woman with preeclampsia to be assessed, the last monitoring timestamp and the penultimate monitoring timestamp closest to the current assessment time are extracted from the individual measured time series set. The serum uric acid measured data corresponding to the last monitoring timestamp and the penultimate monitoring timestamp are then extracted respectively. The difference between the serum uric acid measured data corresponding to the last monitoring timestamp and the penultimate monitoring timestamp is taken as the recent indicator change, and the difference between the last monitoring timestamp and the penultimate monitoring timestamp is taken as the recent monitoring time interval. The recent indicator change is used as the numerator, the recent monitoring time interval is used as the denominator, and the resulting fraction is taken as the recent change slope data.

[0068] After obtaining the recent change slope data of the pregnant women to be evaluated, the recent transient decompensated momentum factor is obtained by jointly gating the recent change slope data with the upper limit threshold of clinical risk. Specifically, for any pregnant woman with preeclampsia to be evaluated, when the recent change slope data is less than or equal to a constant 0, the unidirectional deterioration slope assessment corresponding to the recent change slope data is set to a constant 0; when the recent change slope data is greater than a constant 0, the recent change slope data is used as the corresponding unidirectional deterioration slope assessment.

[0069] The measured serum uric acid data corresponding to the last monitoring time stamp of the preeclamptic pregnant woman is extracted, and the upper limit of clinical risk threshold corresponding to serum uric acid is extracted from the clinical baseline parameters. The difference between the measured serum uric acid data corresponding to the last monitoring time stamp and the upper limit of clinical risk threshold is used as the numerator, and the upper limit of clinical risk threshold is used as the denominator. The resulting fraction is used as the risk threshold deviation assessment. The result of multiplying the risk threshold deviation assessment by a preset slope activation sensitivity constant is used as the exponential gating input value. In this embodiment, the slope activation sensitivity constant is set to 10. The negative of the exponential gating input value is exponentially mapped with the natural constant as the base. The constant 1 is added to the corresponding exponential mapping result, and the result of dividing the constant 1 by the sum is used as the threshold gating coefficient. The result of multiplying the unidirectional deterioration slope assessment by the threshold gating coefficient is used as the recent transient decompensation momentum factor.

[0070] In one embodiment, it is assumed that the last monitoring timestamp and the penultimate monitoring timestamp are respectively and The actual serum uric acid data corresponding to the last monitoring time stamp and the penultimate monitoring time stamp are as follows: and The upper limit of clinical risk for serum uric acid is: The slope activation sensitivity constant is Then the first The formula for calculating the recent transient decompensated momentum factor of a pregnant woman to be evaluated is as follows:

[0071]

[0072] in, Indicates the first Recent transient decompensated momentum factor of a pregnant woman to be evaluated; and These represent the measured serum uric acid data corresponding to the last monitoring timestamp and the penultimate monitoring timestamp, respectively. and These represent the timestamps of the last monitoring session and the penultimate monitoring session, respectively. This indicates the upper limit of clinical risk corresponding to serum uric acid; Represents an exponential function with the natural constant as the base; This represents the maximum value function.

[0073] It should be noted that existing feature extraction algorithms, due to their global averaging characteristics, are slow to identify sudden deterioration of the condition. The recent transient decompensated momentum factor constructed in this invention abandons the full-cycle summation structure. First, the left side of the formula only extracts the unit slope between the two most recent points and nests a maximum value function. This design forms a one-way filtering mechanism: as long as the recent indicator does not show a numerical increase or deterioration (i.e., the slope is less than or equal to zero), the transient momentum calculation result immediately returns to zero, eliminating false positive interference caused by the natural decline or stable fluctuation of the indicator. Second, a single slope value has different clinical significance at different absolute background levels. Therefore, a risk threshold-based factor is introduced on the right side of the formula. Anchored Sigmoid soft threshold gating structure. If the latest metric... Currently at a safe low level, the absolute value of the exponential part of the sigmoid term is large and negative, causing the output of the entire term to be close to 0. At this point, even if there are large slope fluctuations recently, they will be judged as changes within the safe physiological redundancy range and weakened. However, once the current indicator approaches or crosses the warning line... This term will rapidly undergo a nonlinear leap and approach 1, thus fully releasing the high-slope feature calculated on the left side. This differential probe, combined with a high-gated nonlinear composite structure, maps the basic geometric slope to a clinically relevant high-level decompensation transient risk value, enabling targeted quantitative extraction of hidden, rapidly deteriorating signals.

[0074] Thus, the recent transient decompensation momentum factor was obtained by conducting a recent deterioration dynamic analysis on monitoring data close to the current assessment time.

[0075] Step S3: By jointly evaluating the evolution trend and fluctuation potential energy of historical monitoring sequences, the steady-state fracture and barrier breakdown factors are obtained.

[0076] After initially extracting the recent transient decompensated momentum factor in step S2, directly inputting it into the prediction model still cannot fully reflect the overall picture of system decompensation. In pathological progression, the pregnant woman's body possesses regulatory resilience, i.e., a compensatory barrier. Determining whether the steady state has substantially collapsed requires not only assessing the changes in current values ​​but also measuring the degree of deviation of the current evolutionary trajectory from the historical dominant trajectory, and whether this abrupt change exceeds the upper limit of historically accumulated fluctuation adaptation. If a pregnant woman recently experiences a high-momentum, recent transient decompensated momentum factor, and this change deviates from her long-term stable trajectory while instantly breaking through her historically established upper limit of fluctuation, the risk of disease deterioration indicated by this steady-state tension rupture is extremely high. To introduce this multi-dimensional spatiotemporal rupture tension into the model, it is necessary to abandon the traditional static mean and variance comparisons and construct a higher-order phase space deflection and asymmetric potential energy penetration function based on the recent transient decompensated momentum factor to uncover the deep deterioration characteristics at the moment of collapse of the body's compensatory system.

[0077] In summary, this invention first extracts and time-decays historical monitoring sequence data to obtain historical upward fluctuation potential energy data. Specifically, for any pregnant woman with preeclampsia to be evaluated, the penultimate monitoring timestamp and all historical monitoring timestamps before the penultimate monitoring timestamp are extracted from the individual measured time series set, and the serum uric acid measured data corresponding to each historical monitoring timestamp are extracted.

[0078] For any adjacent historical monitoring timestamp, the difference between the measured serum uric acid data corresponding to the later historical monitoring timestamp and the measured serum uric acid data corresponding to the earlier historical monitoring timestamp is taken as the historical indicator change. The difference between the later and earlier historical monitoring timestamps is taken as the historical monitoring time interval. The historical indicator change is taken as the numerator, and the historical monitoring time interval is taken as the denominator. The resulting fraction is used as the historical change slope assessment. When the historical change slope assessment is less than or equal to a constant 0, the upward fluctuation assessment corresponding to the historical change slope assessment is set to a constant 0; when the historical change slope assessment is greater than a constant 0, the historical change slope assessment is used as the corresponding upward fluctuation assessment.

[0079] For any subsequent historical monitoring timestamp, the constant 1 is divided by the calculated result of the difference between the current assessment time and the subsequent historical monitoring timestamp, and then added back to the constant 1. The sum is then subjected to a natural logarithm operation to obtain the time decay weight corresponding to the subsequent historical monitoring timestamp. The result of multiplying the upward fluctuation assessment by the time decay weight is used as the weighted upward fluctuation assessment for the corresponding historical interval. The weighted upward fluctuation assessments corresponding to all historical intervals are squared and summed. The sum is then divided by the number of historical intervals and then square-rooted to obtain the historical upward fluctuation potential energy data.

[0080] After obtaining historical upward fluctuation potential energy data, the historical overall evolution gradient data is obtained by performing first-to-last evolution trend analysis on the historical monitoring sequence data. Specifically, for any pregnant woman with preeclampsia to be evaluated, the first monitoring time stamp, the penultimate monitoring time stamp, the serum uric acid measured data corresponding to the first monitoring time stamp, and the serum uric acid measured data corresponding to the penultimate monitoring time stamp are extracted from the individual measured time series set. The difference between the serum uric acid measured data corresponding to the penultimate monitoring time stamp and the serum uric acid measured data corresponding to the first monitoring time stamp is taken as the historical overall indicator change, and the difference between the penultimate monitoring time stamp and the first monitoring time stamp is taken as the historical overall monitoring time span. The historical overall indicator change is taken as the numerator, the historical overall monitoring time span is taken as the denominator, and the corresponding fraction is taken as the historical overall evolution gradient data.

[0081] After obtaining the historical overall evolution gradient data, the steady-state fracture and barrier breakdown factors are obtained by jointly enhancing the recent transient decompensated momentum factor, historical upward wave potential energy data, and historical overall evolution gradient data. Specifically, for any preeclampsia pregnant woman in any state to be evaluated, the result of multiplying the recent transient decompensated momentum factor by the historical overall evolution gradient data is used as the trend inner product assessment. The result of multiplying the absolute value of the recent transient decompensated momentum factor by the absolute value of the historical overall evolution gradient data is added to a preset small positive constant to obtain the trend normalized denominator assessment. The trend inner product assessment is used as the numerator, and the trend normalized denominator assessment is used as the denominator. The resulting fraction is used as the trend direction similarity assessment. The difference between the constant 1 and the trend direction similarity assessment is used as the trend deflection intensity assessment. The trend deflection intensity assessment is then mapped using a hyperbolic tangent function to obtain the trend deflection enhancement coefficient.

[0082] The square of the recent transient uncompensated momentum factor is used as the momentum energy level assessment. The result of adding the square of the historical upward wave potential energy data to a preset physiological wave stability constant is used as the historical barrier basis assessment. In this embodiment of the invention, the physiological wave stability constant is set to 400. The momentum energy level assessment is used as the numerator, the historical barrier basis assessment is used as the denominator, and the corresponding fraction is used as the barrier penetration strength assessment. The barrier penetration strength assessment is then subjected to an exponential mapping with the natural constant as the base to obtain the barrier penetration enhancement coefficient.

[0083] The result of multiplying the recent transient decompensated momentum factor, trend deflection enhancement coefficient, and barrier penetration enhancement coefficient is used as the steady-state fracture and barrier breakdown factors.

[0084] In one implementation, assume the first The actual serum uric acid data corresponding to the timestamp of this monitoring is The timestamp of the current evaluation moment is: The total number of monitoring sessions up to the current assessment time is: The expression for calculating the historical wave potential energy data is as follows:

[0085]

[0086] in, This represents historical data on upward fluctuations in potential energy. This indicates the total number of monitoring sessions up to the current assessment time. Indicates the first Serum uric acid measured data corresponding to the timestamp of each monitoring session; Indicates the first Serum uric acid measured data corresponding to the timestamp of each monitoring session; Indicates the first The timestamp of the monitoring session; Indicates the first The timestamp of the monitoring session; The timestamp indicates the current evaluation time; Represents the logarithmic function with the natural constant as the base; This represents the maximum value function.

[0087] Furthermore, assuming the historical overall evolution gradient data is... The physiological fluctuation stability constant is ; a small positive constant is Then the first The calculation formulas for the steady-state fracture and barrier breakdown factor of a pregnant woman to be evaluated are as follows:

[0088]

[0089] in, Indicates the first Steady-state fracture and barrier breakdown factor of a pregnant woman to be evaluated; Indicates the first Recent transient decompensated momentum factor of a pregnant woman to be evaluated; This represents the overall historical evolution gradient data; This represents a small positive constant, which is set in the embodiments of the present invention. ; This represents historical data on upward fluctuations in potential energy. This represents the hyperbolic tangent function.

[0090] It should be noted that existing multivariate combined models often rely on static Z-score normalization, which makes it difficult to detect sudden reversals in the evolution direction of time series and nonlinear kinetic energy overflow. The steady-state fracture and barrier breakdown factor of this invention adopts a dual structure of trajectory deflection and kinetic energy breakdown.

[0091] First, the middle section of the formula Utilizing recent momentum within the structure gradient with historical benchmark The inner product and modulus ratio are calculated to evaluate the cosine similarity of the angle between the recent deterioration vector and the historical trend vector. Subtracting this similarity from 1 means that if the historical sequence has been stable or declining, and the recent trend suddenly reverses and surges upward (the angle approaches 180 degrees, and the cosine value is negative), the structure will output a large positive penalty (close to 2), which is then smoothly converged to the activation multiplier by the hyperbolic tangent function; conversely, if the historical trend is in a slow upward phase, and the recent continued rise is an inertial continuation, the structure will output a lower value. This nonlinear angle design aims to capture the clinically dangerous phenomenon of disease trajectory reversal.

[0092] Secondly, the natural exponent term at the end of the formula The denominator is used to compensate for the kinetic energy penetration coefficient. This invention independently constructs the denominator. It is a root mean square of historical unidirectional upward kinetic energy rectified by a maximum function and weighted with a time logarithmic decay. This means that only historically occurring, and more recent, deteriorating fluctuations constitute an effective compensatory adaptive barrier for the organism. The exponential term uses the square of the recent transient momentum (…). Divide by the square of the historical compensatory potential energy ( This creates a penetration ratio. When the recent deterioration kinetic energy significantly exceeds the pregnant woman's historically accumulated adaptation threshold, the exponential function will produce a sharp numerical amplification. This phase-space kinetic energy structure forcibly amplifies the critical signal when the body's compensatory limit is breached, enhancing the dynamic prediction model's ability to determine the timing of emergency preterm labor intervention.

[0093] Thus, the steady-state fracture and barrier breakdown factors were obtained by jointly evaluating the evolution trend and fluctuation potential energy of historical monitoring sequences.

[0094] Step S4: Obtain the reconstructed fused feature vector by fusing and reconstructing the basic time-weighted average features with the steady-state fracture and barrier breakdown factors.

[0095] After obtaining the independently calculated steady-state fracture and barrier breakdown factors, these optimized factors need to be applied to the input framework of the existing dynamic prediction model. Existing prediction models typically directly input the average exposure features extracted from the time-weighted average (TWA) into the Cox proportional hazards model. To smoothly embed the critical signal of organismal compensatory collapse into the model and avoid compromising the original model's ability to assess the long-term pathological burden trend, this invention does not discard the original TWA features. Instead, it directly applies the steady-state fracture and barrier breakdown factors as non-linear, dynamic multiplicative amplification weights to the initial time-weighted average features. This operation endows the static features, which originally only reflected long-term average exposure levels, with the ability to perceive recent decompensation mutations, thereby generating a reconstructed fusion feature that combines historical background burden and recent acute deterioration attributes.

[0096] Specifically, for any preeclampsia pregnant woman in any state to be evaluated, the result of multiplying the steady-state fracture and barrier breakdown factors by a preset scaling factor is used as the dynamic amplification assessment. In this embodiment of the invention, the scaling factor is set to 0.5. The result of adding a constant 1 to the dynamic amplification assessment is used as the feature reconstruction weight. The result of multiplying the baseline time-weighted average feature by the feature reconstruction weight is used as the reconstruction fusion feature. The conventional cross-sectional clinical features corresponding to the preeclampsia pregnant woman are extracted, and the reconstruction fusion feature is combined with the conventional cross-sectional clinical features to obtain the reconstruction fusion feature vector.

[0097] It should be noted that by superimposing the steady-state fracture and barrier breakdown factor as penalty incentive amplification terms onto the base time-weighted average feature, if the pregnant woman's condition is in a relatively stable compensatory phase (no recent severe deterioration or deterioration that has not penetrated the historical barrier), the calculated results of the steady-state fracture and barrier breakdown factor approach 0. At this time, the reconstructed fusion feature degenerates back to the original base time-weighted average feature, ensuring the stability of the model's prediction results during the safe phase and avoiding false positives and overtreatment. However, once the pregnant woman's body experiences a rapid steady-state tension fracture in the near future, the values ​​of the steady-state fracture and barrier breakdown factor increase rapidly and nonlinearly, causing the reconstructed fusion feature to amplify beyond its original mean. This structure corrects the defect of the average value algorithm in existing technologies that masks recent acute deterioration signals from the source of algorithm input, enabling the prediction model to obtain a directional identification dimension for critical preterm birth high-risk states. In the actual computer programming and clinical system deployment phase, to prevent high-risk extreme value samples from triggering floating-point arithmetic overflows in the underlying computer and to ensure the linear convergence of the subsequent Cox proportional hazards model, the system applies conventional program safety throttling operations to the upper limit of the exponential function calculation and the final generated reconstructed fusion features when performing the above feature extraction. In this embodiment, the upper limit threshold for the output of the reconstructed fusion features is set to 2000. The overflow prevention truncation mechanism is an engineering protection measure and does not change the core nonlinear mathematical evolution logic of this invention.

[0098] Thus, the reconstruction of the fused feature vector is completed by fusing the basic time-weighted average features with the steady-state fracture and barrier breakdown factors.

[0099] Step S5 involves inputting the reconstructed fusion feature vector into the Cox proportional hazards model for risk prediction, thereby obtaining dynamic prediction results for preterm birth risk.

[0100] After completing the aforementioned feature extraction and fusion reconstruction, a reconstructed fusion feature vector has been obtained that can simultaneously characterize the long-term pathological burden level and the intensity of recent decompensated deterioration in preeclampsia pregnant women. Compared with the traditional method that only uses baseline time-weighted average features as model input, the reconstructed fusion feature vector constructed in this invention not only retains the overall exposure level of key monitoring indicators of pregnant women throughout the entire observation period, but also further embeds enhanced information on the dynamic state of recent deterioration, the degree of deviation of historical evolutionary trajectory, and the breakthrough of individual historical fluctuation adaptation capacity. Therefore, it can more completely reflect the true disease progression status of preeclampsia pregnant women at the current stage. In order to further transform the above features into risk assessment results that can directly serve clinical decision-making, this step inputs the reconstructed fusion feature vector into a pre-trained Cox proportional hazards model to perform dynamic prediction of preterm birth risk for a future preset time window.

[0101] Specifically, for any pregnant woman with preeclampsia to be assessed, the reconstructed fusion feature vector corresponding to that woman is first extracted, and then input into a Cox proportional hazards model pre-trained based on historical clinical cohort samples. The historical clinical cohort samples include dynamic monitoring time-series data, clinical baseline parameters, follow-up outcome records, and corresponding preterm birth outcome markers generated during continuous follow-up of multiple pregnant women with preeclampsia during pregnancy. The preterm birth outcome markers at least include whether an iatrogenic preterm birth event due to disease deterioration occurred within a preset risk prediction time window. By using the historical clinical cohort samples to train the parameters of the Cox proportional hazards model, the partial regression coefficients and baseline risk functions corresponding to each input feature are obtained, thereby establishing a mapping relationship between the reconstructed fusion feature vector and the risk of preterm birth within the target time window.

[0102] During the model prediction phase, the reconstructed and fused feature vector of the pregnant woman to be evaluated is input into the Cox proportional hazards model. The model calculates the risk contribution of each feature based on the partial regression coefficients of the internal fit, and, combined with the baseline risk function, outputs the probability that the pregnant woman will experience a worsening of preeclampsia requiring iatrogenic preterm labor within a preset time window at the current evaluation time. In this embodiment, the preset time window is preferably set to the next 7 days. The higher the risk probability output by the model, the greater the likelihood that the pregnant woman to be evaluated will enter the high-risk termination of pregnancy decision range in the short term.

[0103] To improve the clinical interpretability and ease of use of the prediction results, after obtaining the risk probability, the risk probability can be further mapped to an individualized dynamic risk score, and a corresponding survival probability curve or dynamic nomogram can be generated simultaneously to display the results. Based on the dynamic risk score, combined with the pregnant woman's current gestational age, fetal intrauterine status, changes in maternal organ function, and clinical intervention conditions, a judgment can be made on whether to strengthen monitoring, adjust treatment strategies, or terminate the pregnancy early. Because the reconstructed fusion feature vector used in this invention can enhance the ability to identify the recent acute exacerbation stage of preeclampsia, under the same Cox proportional hazards model framework, this invention can expose critical high-risk signals that are difficult to reflect by traditional time-weighted average features earlier, thereby improving the accuracy, timeliness, and clinical guidance value of the dynamic prediction results of preterm birth risk.

[0104] Thus, the risk prediction results for preterm birth risk have been obtained by inputting the reconstructed and fused feature vectors into the Cox proportional hazards model.

[0105] 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 dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators, characterized in that, The method includes: Step S1: By acquiring and processing the dynamic monitoring time series data and clinical baseline parameters of pregnant women with preeclampsia, the individual measured time series set and baseline time-weighted average characteristics are obtained; Step S2: Obtain the recent transient decompensation momentum factor by performing a recent deterioration dynamics analysis on the monitoring data close to the current assessment time; Step S3: Obtain steady-state fracture and barrier breakdown factors by jointly evaluating the evolution trend and fluctuation potential energy of historical monitoring sequences; Step S4: Obtain the reconstructed fused feature vector by fusing and reconstructing the basic time-weighted average features with the steady-state fracture and barrier breakdown factors; Step S5: By inputting the reconstructed fusion feature vector into the Cox proportional hazards model for risk prediction, dynamic prediction results of preterm birth risk are obtained.

2. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 1, characterized in that, The process involves acquiring and processing time-series data from dynamic monitoring of pregnant women with preeclampsia and clinical baseline parameters to obtain individual measured time-series sets and baseline time-weighted average characteristics, including: For pregnant women with preeclampsia awaiting evaluation, multiple routine clinical follow-up records and biochemical test records from the time of diagnosis or registration of preeclampsia to the current evaluation time are retrieved from the electronic medical record system of the medical institution. The total number of monitoring times up to the current evaluation time is obtained, and the monitoring timestamps are obtained in chronological order, where the monitoring timestamps are represented by gestational weeks. For each monitoring timestamp, the actual serum uric acid data at the corresponding monitoring timestamp is extracted, and the monitoring timestamps are correlated with the actual serum uric acid data corresponding to each monitoring timestamp to obtain the original dynamic monitoring time series data of the pregnant women with preeclampsia. The original dynamic monitoring time series data were processed by data anonymization, format cleaning and missing value interpolation and completion, and the processed monitoring timestamps and corresponding serum uric acid measured data were used as individual measured time series sets. The clinical risk upper limit threshold corresponding to serum uric acid is retrieved from the electronic medical record system of medical institutions. Based on the time interval between two adjacent monitoring timestamps in the individual measured time series set and the corresponding measured serum uric acid data, a time-weighted average is calculated to obtain the basic time-weighted average feature.

3. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 1, characterized in that, The method involves performing recent deterioration dynamics analysis on monitoring data close to the current assessment time to obtain the recent transient decompensated momentum factor, including: By performing time-series localization and difference extraction on continuous monitoring data close to the current assessment time, recent change slope data is obtained; By combining recent slope data with clinical risk upper limit thresholds, recent transient decompensated momentum factors are obtained.

4. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 3, characterized in that, The process of obtaining recent change slope data by performing time-series localization and difference extraction on continuous monitoring data close to the current evaluation time includes: For any pregnant woman with preeclampsia whose condition is to be evaluated, extract the last monitoring timestamp closest to the current evaluation time and the penultimate monitoring timestamp adjacent to the last monitoring timestamp from the individual measured time series set, and extract the measured serum uric acid data corresponding to the last monitoring timestamp and the measured serum uric acid data corresponding to the penultimate monitoring timestamp respectively. The difference between the measured serum uric acid data corresponding to the last monitoring timestamp and the measured serum uric acid data corresponding to the penultimate monitoring timestamp is taken as the recent indicator change, and the difference between the last monitoring timestamp and the penultimate monitoring timestamp is taken as the recent monitoring time interval. The recent index change is used as the numerator, the recent monitoring time interval is used as the denominator, and the corresponding fraction is used as the recent change slope data.

5. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 3, characterized in that, The method involves jointly gating recent slope data with a clinical risk upper limit threshold to obtain the recent transient decompensated momentum factor, including: For any pregnant woman with preeclampsia whose condition is to be evaluated, when the recent change slope data is less than or equal to constant 0, the unidirectional deterioration slope assessment corresponding to the recent change slope data is set to constant 0; when the recent change slope data is greater than constant 0, the recent change slope data is used as the corresponding unidirectional deterioration slope assessment. The serum uric acid measured data corresponding to the last monitoring timestamp of the preeclamptic pregnant woman were extracted, and the upper limit threshold of clinical risk corresponding to serum uric acid was extracted from the clinical baseline parameters. The difference between the measured serum uric acid data corresponding to the last monitoring timestamp and the upper limit threshold of clinical risk was used as the numerator, and the upper limit threshold of clinical risk was used as the denominator. The corresponding fraction was used as the risk threshold deviation assessment. The result of multiplying the danger threshold deviation assessment by the preset slope activation sensitivity constant is used as the exponential gating input value. The negative of the exponential gating input value is subjected to exponential mapping with the natural constant as the base. The constant 1 is added to the corresponding exponential mapping result, and the result of dividing the constant 1 by the sum is used as the threshold gating coefficient. The result of multiplying the unidirectional deterioration slope assessment by the threshold gating coefficient is used as the recent transient decompensated momentum factor.

6. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 1, characterized in that, The method of obtaining steady-state fracture and barrier breakdown factors by jointly evaluating the evolution trend and fluctuation potential energy of historical monitoring sequences includes: By extracting upward fluctuations and performing time decay processing on historical monitoring sequence data, historical upward fluctuation potential energy data can be obtained. By performing first-to-last evolution trend analysis on historical monitoring sequence data, historical overall evolution gradient data can be obtained. By jointly enhancing recent transient decompensated momentum factors, historical upward wave potential energy data, and historical overall evolution gradient data, steady-state fracture and barrier breakdown factors are obtained.

7. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 6, characterized in that, The process of extracting upward fluctuations and performing time decay processing on historical monitoring sequence data to obtain historical upward fluctuation potential energy data includes: For any pregnant woman with preeclampsia in any state to be evaluated, the penultimate monitoring timestamp and all historical monitoring timestamps before the penultimate monitoring timestamp are extracted from the individual measured time series set, and the serum uric acid measured data corresponding to each historical monitoring timestamp are extracted. For any adjacent historical monitoring timestamp, the difference between the measured serum uric acid data corresponding to the next historical monitoring timestamp and the measured serum uric acid data corresponding to the previous historical monitoring timestamp is taken as the historical indicator change, the difference between the next historical monitoring timestamp and the previous historical monitoring timestamp is taken as the historical monitoring time interval, the historical indicator change is taken as the numerator, the historical monitoring time interval is taken as the denominator, and the corresponding fraction is taken as the historical change slope assessment. When the historical change slope assessment is less than or equal to a constant 0, the upward volatility assessment corresponding to the historical change slope assessment is set to a constant 0; when the historical change slope assessment is greater than a constant 0, the historical change slope assessment is used as the corresponding upward volatility assessment. For any subsequent historical monitoring timestamp, the constant 1 is divided by the calculated result of the difference between the current evaluation time and the subsequent historical monitoring timestamp, and then added to the constant 1. The corresponding sum is then subjected to a natural logarithm operation to obtain the time decay weight corresponding to the subsequent historical monitoring timestamp. The result of multiplying the upward fluctuation assessment by the time decay weight is used as the weighted upward fluctuation assessment for the corresponding historical interval. The weighted upward fluctuation assessments corresponding to all historical intervals are squared and summed. The sum is divided by the number of historical intervals and then square rooted to obtain the historical upward fluctuation potential energy data.

8. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 6, characterized in that, The process of analyzing the first and last evolutionary trends of historical monitoring sequence data to obtain historical overall evolutionary gradient data includes: For any pregnant woman with preeclampsia in any state to be evaluated, the first monitoring timestamp, the second to last monitoring timestamp, the serum uric acid measured data corresponding to the first monitoring timestamp, and the serum uric acid measured data corresponding to the second to last monitoring timestamp are extracted from the individual measured time series set; The difference between the measured serum uric acid data corresponding to the penultimate monitoring timestamp and the measured serum uric acid data corresponding to the first monitoring timestamp is taken as the historical overall indicator change, and the difference between the penultimate monitoring timestamp and the first monitoring timestamp is taken as the historical overall monitoring time span. The change in the overall historical index is used as the numerator, the monitoring time span of the overall historical index is used as the denominator, and the corresponding fraction is used as the historical overall evolution gradient data.

9. A method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 6, characterized in that, The method involves jointly enhancing recent transient decompensated momentum factors, historical upward wave potential energy data, and historical overall evolution gradient data to obtain steady-state fracture and barrier breakdown factors, including: For any pregnant woman with preeclampsia in any state to be evaluated, the result of multiplying the recent transient decompensated momentum factor by the historical overall evolutionary gradient data is used as the trend inner product assessment. The result of multiplying the absolute value of the recent transient decompensated momentum factor by the absolute value of the historical overall evolutionary gradient data is added to a preset small positive constant to obtain the trend normalized denominator assessment. The trend inner product assessment is used as the numerator, the trend normalized denominator assessment is used as the denominator, and the corresponding fraction is used as the trend direction similarity assessment. The difference between the constant 1 and the trend direction similarity assessment is used as the trend deflection intensity assessment, and the trend deflection intensity assessment is mapped by the hyperbolic tangent function to obtain the trend deflection enhancement coefficient. The square of the recent transient decompensated momentum factor is used as the momentum energy level assessment. The calculation result of adding the square of the historical upward wave potential energy data to the preset physiological wave stability constant is used as the historical barrier base assessment. The momentum energy level assessment is used as the numerator, the historical barrier base assessment is used as the denominator, and the corresponding fraction is used as the barrier penetration strength assessment. The barrier penetration strength assessment is then subjected to an exponential mapping with the natural constant as the base to obtain the barrier penetration enhancement coefficient. The result of multiplying the recent transient decompensated momentum factor, trend deflection enhancement coefficient, and barrier penetration enhancement coefficient is used as the steady-state fracture and barrier breakdown factor.

10. The method for dynamic prediction of preeclampsia and preterm birth risk based on pregnancy monitoring indicators according to claim 1, characterized in that, The process involves fusing and reconstructing the basic time-weighted average features with the steady-state fracture and barrier breakdown factors to obtain a reconstructed fused feature vector, including: For any pregnant woman with preeclampsia in any state to be evaluated, the result of multiplying the steady-state fracture and barrier breakdown factors by the preset scaling factor is used as a dynamic amplification evaluation. The result of adding the constant 1 to the dynamic amplification evaluation is used as the feature reconstruction weight; The result of multiplying the base time-weighted average feature by the feature reconstruction weight is used as the reconstruction fusion feature; Extract the conventional cross-sectional clinical features corresponding to the preeclamptic pregnant woman, and combine the reconstructed fusion features with the conventional cross-sectional clinical features to obtain the reconstructed fusion feature vector.

Images