Intelligent prediction method for gestational diabetes risk based on multimodal data fusion

Abstract

This invention discloses an intelligent prediction method for gestational diabetes mellitus risk based on multimodal data fusion, specifically relating to the field of data processing technology. The method includes S1: setting a prediction window; S2: acquiring risk detection data; S3: constructing a multimodal time series matrix; S4: risk analysis; S5: constructing a pregnancy risk change curve; and S6: visualization. This invention detects risks using physiological and biochemical risk indicators and constructs a multimodal time series matrix based on the risk detection data. This ensures accurate data collection and forms a matrix structure of time periods and multiple indicators. It then correlates pregnancy risk with the detection time period, constructing a pregnancy risk change curve that visually demonstrates the dynamic evolution of risk throughout pregnancy. Furthermore, inflection point detection allows for precise location of the risk occurrence time, facilitating focused detection of risk occurrence times and laying a structured foundation for subsequent risk trend and fluctuation amplitude analysis.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method for intelligent prediction of gestational diabetes risk based on multimodal data fusion. Background Technology

[0002] Gestational diabetes mellitus (GDM) refers to diabetes mellitus first diagnosed or occurring during pregnancy. This disease poses significant risks to both mother and fetus. For example, pregnant women may experience complications such as abnormal embryonic development or death, gestational hypertension, infections, and polyhydramnios. Fetuses may experience macrosomia, fetal growth restriction, miscarriage and premature birth, fetal distress, and intrauterine fetal death. Furthermore, patients with GDM have an increased risk of developing type 2 diabetes postpartum, and a higher incidence of long-term cardiovascular diseases.

[0003] Existing methods for predicting the risk of gestational diabetes mellitus (GDM) are mainly based on static risk factors such as age, BMI, and family history. They construct empirical or semi-quantitative assessment systems to determine the probability of GDM occurrence. However, these methods still have some shortcomings in practical use. First, existing risk factor assessment and prediction methods are mainly based on static risk factors such as age, BMI, and family history, resulting in low prediction accuracy. They cannot conduct dynamic risk assessments and fail to fully consider individual differences. In this case, multi-data assessment is needed. However, existing methods for addressing low prediction accuracy are mostly limited to monitoring single biochemical indicators, mainly through monitoring fasting blood glucose and postprandial blood glucose. The monitoring results are easily affected by various factors, lack assessment of the overall system status, and the warning time is relatively late, making it difficult to achieve early intervention.

[0004] Second, current methods for detecting gestational diabetes mainly rely on one-time data from before or in early pregnancy, which cannot track dynamic changes in indicators during pregnancy. Pregnancy is a process of dramatic physiological changes, especially in the second and third trimesters when placental hormone secretion increases, leading to increased insulin resistance. In this case, dynamic risk monitoring during pregnancy is needed, but current methods cannot capture the risk inflection points in this dynamic process. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an intelligent prediction method for the risk of gestational diabetes based on multimodal data fusion. By performing physiological risk analysis and biochemical risk analysis separately during the detection of gestational diabetes, and thereby calculating the pregnancy risk index, the method maximizes the combination of data detection and risk prediction. After calculating the pregnancy risk index, a pregnancy risk change curve is constructed to determine the time of risk occurrence during pregnancy, thus effectively solving the problems raised in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] S1: Set the prediction window, which includes the early pregnancy window, the mid-pregnancy window, and the late pregnancy window, and label them as T1, T2, and T3 respectively. The prediction window time is set according to the medical diagnosis and treatment standards for diabetes.

[0008] S2: Risk Detection Data Acquisition: Extract the current pregnancy detection time from the prediction window. The pregnancy detection time is in units of gestational week and is divided and marked according to gestational week as 1, 2, ..., j, ..., m, where j represents the j-th pregnancy detection time. Risk detection data corresponding to each pregnancy detection time is collected. The risk detection data includes physiological risk indicators and biochemical risk indicators, and then the pregnancy detection time period corresponding to the risk detection data is obtained.

[0009] S3: Multimodal time series matrix construction: Based on the pregnancy detection time period corresponding to the risk detection data, the time series corresponding to the risk detection data is generated in a preset order, and then a multimodal time series matrix is ​​constructed based on the time series;

[0010] S4: Risk Analysis: Perform physiological risk analysis and biochemical risk analysis on the multimodal time series matrix corresponding to each pregnancy detection period, and analyze the pregnancy risk index for each pregnancy detection period accordingly.

[0011] S5: Constructing a pregnancy risk change curve: Correlation analysis is performed on the pregnancy risk index corresponding to each pregnancy detection period and the pregnancy detection period to construct a pregnancy risk change curve, thereby determining the time of risk occurrence;

[0012] S6: Visualization: The risk occurrence time is sent to the user terminal according to the preset summary method, which includes report summary, picture summary and chart summary. Managers can use the summary content to focus on the risk occurrence time of gestational diabetes. The summary content includes multimodal time series matrix, pregnancy risk status and pregnancy risk change curve.

[0013] The technical effects and advantages of this invention are as follows:

[0014] 1. This invention uses wearable devices and prenatal testing terminals to detect physiological and biochemical risk indicators of gestational diabetes mellitus, and constructs a multimodal time series matrix based on the risk detection data. This allows for multi-dimensional data collection, which helps to provide a two-way assessment of the risk during pregnancy and provides comprehensive and reliable data support for determining the timing of risk occurrence, avoiding the one-sidedness of a single indicator. On the other hand, marking the pregnancy detection time in gestational week sequence to form a time series can maximize the accuracy of data collection and lay a structured foundation for subsequent risk change trends and fluctuations.

[0015] 2. This invention strictly aligns physiological and biochemical risk indicators on a timeline by using pregnancy testing time periods, solving the problem of inconsistent testing frequencies across different modalities. This forms a matrix structure of time periods and multiple indicators, thereby enabling correlation analysis between pregnancy risk and testing time periods. From this, a pregnancy risk change curve is constructed, visually demonstrating the dynamic evolution of risk throughout pregnancy. Furthermore, inflection point detection can accurately pinpoint the time of risk occurrence, facilitating focused detection of the risk's occurrence time to minimize the risk of gestational diabetes and providing doctors with a strong reference for diabetes diagnosis. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0017] Figure 2 This is a flowchart of the risk analysis for this invention.

[0018] Figure 3 The flowchart illustrates the steps for determining the timing of risk occurrence in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] As attached Figure 1-3 The intelligent prediction method for gestational diabetes risk based on multimodal data fusion shown includes a wearable device, a prenatal examination terminal, a user terminal, and a control center. The wearable device and the prenatal examination terminal are connected to the user terminal via Bluetooth.

[0021] In a more specific application of the present invention, the wearable device is used to monitor vital signs, specifically weight, height, resting heart rate, and electrocardiogram (ECG) signals. The wearable device can be a smart bracelet, which is combined with a heart rate monitoring device for monitoring. For example, the weight and height of a pregnant woman are continuously recorded by the smart bracelet, and the heart rate monitoring device can be an ECG monitor for continuous monitoring of resting heart rate and ECG signals.

[0022] The prenatal checkup terminal is used to collect information on the insulin levels of pregnant women. Specifically, it can be an RFID reader, an insulin detector, and a prenatal checkup database. The insulin level is monitored by tracking the pregnant woman's fasting blood glucose and insulin concentration. The monitoring results are stored in the prenatal checkup database, and the RFID reader automatically identifies the pregnant woman's insulin level.

[0023] The control center is used to analyze data and control relevant parameters based on the data from the aforementioned equipment during the gestational diabetes detection process.

[0024] The specific embodiments of the present invention include the following steps:

[0025] S1: Set prediction windows, which include early pregnancy window, mid-pregnancy window and late pregnancy window, and label them as T1, T2 and T3 respectively. The prediction window time is set according to the medical diagnosis and treatment standards for diabetes. For example, T1 is 6-12 weeks of pregnancy, T2 is 24-28 weeks of pregnancy and T3 is 36-40 weeks of pregnancy.

[0026] It should be explained that the purpose of setting the prediction window is to scientifically define the key time range for data collection based on the significant physiological changes during pregnancy. This ensures the timeliness, relevance, and standardization of multimodal data. The early pregnancy window is the time of data collection at the first prenatal checkup, used to assess pre-gestational diabetes risk factors. The mid-pregnancy window is the second trimester, when the placenta secretes anti-insulin hormones at their peak, leading to a significant increase in insulin resistance. Therefore, this stage is the high-incidence period for gestational diabetes. The late pregnancy window is due to the increasing risk of gestational diabetes as the fetus gradually forms, requiring intensive testing before delivery. Physiological indicators during pregnancy change dynamically with gestational week. Testing in non-critical windows may have limited value due to the gradual changes, while focusing on key and sensitive windows can highlight the sudden changes in gestational diabetes-related indicators.

[0027] S2: Risk Detection Data Acquisition: Extract the current pregnancy detection time from the prediction window. The pregnancy detection time is in units of gestational week and is divided and marked according to gestational week as 1, 2, ..., j, ..., m, where j represents the j-th pregnancy detection time. Risk detection data corresponding to each pregnancy detection time is collected. The risk detection data includes physiological risk indicators and biochemical risk indicators, and then the pregnancy detection time period corresponding to the risk detection data is obtained.

[0028] In this embodiment, it should be specifically explained that the specific process for obtaining physiological risk indicators includes:

[0029] ECG signals are extracted using wearable devices, and the peak value of the R wave is identified based on the ECG signals, thereby calculating the interval between adjacent R waves.

[0030] The root mean square of the R-wave is calculated based on the difference between adjacent R-wave intervals, and then the standard deviation of all R-wave intervals is calculated using the root mean square of the R-wave.

[0031] Weight and height are extracted, and the body mass index (BMI) is obtained through the BMI calculation formula. Then, the weight gain rate is obtained by comparing the BMI of adjacent pregnancy detection times with the time interval between adjacent pregnancy detection times.

[0032] The standard deviation of R-wave intervals and the rate of weight gain were integrated as a physiological risk indicator.

[0033] It should be explained that the standard deviation of the R-wave interval and the rate of weight gain are core physiological indicators reflecting the maternal autonomic nervous system function and metabolic load. The standard deviation of the R-wave interval is the standard deviation of the heartbeat cycle. The larger the standard deviation of the R-wave interval, the more balanced the activity of the autonomic nervous system, the stronger the cardiac regulation ability, and the more stable insulin secretion and glucose metabolism. The smaller the standard deviation of the R-wave interval, the more autonomic nervous system is disordered, insulin resistance is aggravated, leading to an increase in fasting blood glucose and increasing the risk of gestational diabetes. The rate of weight gain is the change in body mass index between adjacent pregnancy monitoring times divided by the time interval. When the rate is too fast, fat cells proliferate excessively, directly inhibiting the insulin signaling pathway, thereby increasing the risk of disease.

[0034] In this embodiment, it should be specifically explained that the process of obtaining biochemical risk indicators includes:

[0035] The pregnant woman's insulin status is identified through the prenatal examination terminal, and her fasting blood glucose and insulin concentration are extracted from it;

[0036] Abnormalities in pregnant women's fasting blood glucose levels are identified and classified according to diagnostic criteria:

[0037] ,

[0038] in, The results of abnormal conditions in pregnant women at various pregnancy testing times were marked. This indicates the pregnant woman's fasting blood glucose levels at various pregnancy testing times. and These are the standard values ​​for fasting blood glucose in each abnormal state, specifically obtained according to the medical diagnosis and treatment standards for diabetes.

[0039] The abnormal status markers for each pregnancy test time are summed. If the sum is not equal to 0, it is determined that there is an abnormality in the fasting blood glucose corresponding to each pregnancy test time; otherwise, it is determined that the fasting blood glucose corresponding to each pregnancy test time is normal.

[0040] Extract the fasting blood glucose corresponding to each pregnancy test time with abnormalities, subtract the fasting blood glucose corresponding to adjacent pregnancy test times and take the absolute value, and compare it with the maximum value of the two to obtain the blood glucose fluctuation coefficient.

[0041] The insulin concentrations were compared, and the insulin concentrations corresponding to adjacent pregnancy detection times were subtracted from each other. The insulin secretion rate was then obtained by comparing the results with the time intervals between adjacent pregnancy detection times.

[0042] Integrating blood glucose variability and insulin secretion rate as a biochemical risk indicator.

[0043] It needs to be explained that the blood glucose variability coefficient and insulin secretion rate are used as biochemical risk indicators because the blood glucose variability coefficient is the amplitude of blood glucose fluctuation with gestational age, while the insulin secretion rate is the dynamic compensation of insulin secretion. The higher the blood glucose variability coefficient, the higher the insulin secretion rate, and the lower the short-term regulatory ability of insulin on blood glucose. This indicates that cells increase secretion to compensate for insulin resistance, and thus the biochemical risk is higher. At the same time, by processing the fasting blood glucose and insulin concentration corresponding to each pregnancy test time with adjacent pregnancy test times, it can reflect the changes in biochemical risk indicators during pregnancy, rather than missing cases of large blood glucose fluctuations but not exceeding the standard due to single time point status testing.

[0044] It should be further explained that, since the detection frequencies of physiological risk indicators and biochemical risk indicators are different, by extracting the current pregnancy detection time from the prediction window, data from different modalities can be forcibly aligned to the same time range, avoiding fusion errors caused by time misalignment. At the same time, after obtaining physiological risk indicators and biochemical risk indicators, new detection times are generated by comparing different pregnancy detection times, which are marked as 1, 2, ..., j-1, ..., m-1, where j-1 represents the j-1th pregnancy detection time period.

[0045] S3: Multimodal time series matrix construction: Based on the pregnancy detection time period corresponding to the risk detection data, the time series corresponding to the risk detection data is generated in a preset order, and then a multimodal time series matrix is ​​constructed based on the time series.

[0046] In this embodiment, it should be specifically noted that the time series corresponding to the risk detection data includes physiological risk time series and biochemical risk time series. The physiological risk time series arranges the physiological risk indicators in a preset order, specifically as follows: Each time period corresponds to physiological risk characteristics including the standard deviation of the R-wave interval and the rate of weight gain. Similarly, a biochemical risk time series is generated, specifically represented as follows: Each time period corresponds to biochemical risk characteristics including glycemic variability and insulin secretion rate. Furthermore, different modalities are aligned using pregnancy testing time periods, with missing testing time periods filled using time-series interpolation. A multimodal time series matrix is ​​constructed based on physiological and biochemical risk time series, specifically represented as follows:

[0047] ,

[0048] in, This represents a multimodal time series matrix, where each row corresponds to a pregnancy detection time period and each column corresponds to a modal feature. This represents the physiological risk characteristics corresponding to the (m-1)th pregnancy detection time period in the physiological risk time series. This represents the biochemical risk characteristics corresponding to the (m-1)th pregnancy testing time period in the biochemical risk time series.

[0049] It should be explained that time series interpolation is a data processing method used to solve the problem of missing data. Time series interpolation can improve the accuracy of data analysis, thereby improving prediction results. A specific time series interpolation method can be linear interpolation. For example, by traversing the time series, identifying consecutive missing time periods, calculating the interpolation according to the linear interpolation formula, and then filling in the missing positions in turn.

[0050] It should be further explained that the time series corresponding to the risk detection data are generated in a preset order, where the preset order is the chronological order of the pregnancy detection time periods corresponding to the risk detection data.

[0051] S4: Risk Analysis: Physiological risk analysis and biochemical risk analysis were performed on the multimodal time series matrix corresponding to each pregnancy detection period, and the pregnancy risk index of each pregnancy detection period was analyzed accordingly.

[0052] In this embodiment, the physiological risk analysis is specifically described as follows: physiological risk features are extracted based on the multimodal time series matrix corresponding to each pregnancy detection period, including the standard deviation of the R-wave interval and the rate of weight gain.

[0053] By comparing the standard deviation of the R-wave interval with the rate of weight gain, the risk coefficient of physiological abnormalities corresponding to each pregnancy monitoring period is calculated, as follows:

[0054] ,

[0055] in, This indicates the risk coefficient for physiological abnormalities corresponding to each pregnancy testing time. This represents the standard deviation of the R-wave interval. This represents the standard deviation of the ideal R-wave interval. This indicates the rate of weight gain.

[0056] It should be explained that the risk of physiological abnormalities is greatly affected by heart rate variability, as well as the degree of obesity in pregnant women. Heart rate variability is described using the standard deviation of the R-wave interval. If the standard deviation of the R-wave interval is close to the ideal standard deviation of the R-wave interval, it means that the heart rate variability is relatively large during this pregnancy testing period. The degree of obesity is described using the rate of weight gain. If the rate of weight gain is larger, it means that the pregnant woman's weight changes more during this pregnancy testing period, and the higher the risk coefficient of physiological abnormalities corresponding to this pregnancy testing period.

[0057] It should be further explained that the biochemical risk analysis is as follows: biochemical risk features, including blood glucose fluctuation coefficient and insulin secretion rate, are extracted based on the multimodal time series matrix corresponding to each pregnancy detection period.

[0058] By comparing and calculating the glycemic variability coefficient and insulin secretion rate, the risk coefficient of biochemical abnormalities corresponding to each pregnancy testing period is obtained, as specifically expressed as follows:

[0059] ,

[0060] in, This indicates the risk coefficient for biochemical abnormalities corresponding to each pregnancy testing period. , These represent the glycemic variability coefficient and insulin secretion rate, respectively, in the biochemical risk characteristics. , These represent the corresponding standard blood glucose variability coefficient and insulin secretion rate, respectively, which are obtained from the medical diagnostic and treatment standards for diabetes.

[0061] It should be explained that by analyzing the risk of biochemical abnormalities using the blood glucose fluctuation coefficient and insulin secretion rate, the slower the insulin secretion rate, the worse the blood glucose regulation ability during the pregnancy testing period. At this time, blood glucose fluctuations are larger, and the larger the blood glucose fluctuation coefficient, the greater the risk coefficient of biochemical abnormalities corresponding to the pregnancy testing period.

[0062] It should be further explained that the pregnancy risk index is based on the weighted sum of the risk coefficients for physiological abnormalities and biochemical abnormalities corresponding to each pregnancy testing period, as specifically expressed as follows:

[0063] ,

[0064] in, This indicates the pregnancy risk index corresponding to each pregnancy testing period. and These represent the physiological and biochemical abnormality risk coefficients for the (m-1)th pregnancy testing period, respectively. and These represent the weighting coefficients corresponding to the physiological abnormality risk coefficient and the biochemical abnormality risk coefficient, respectively. The weighting coefficients corresponding to the physiological abnormality risk coefficient and the biochemical abnormality risk coefficient are set through the prediction window corresponding to each pregnancy testing time period. For example, if the prediction window corresponding to the pregnancy testing time period is T1, then the weighting coefficients corresponding to the physiological abnormality risk coefficient and the biochemical abnormality risk coefficient corresponding to the pregnancy testing time period are 0.7 and 0.3, respectively.

[0065] S5: Constructing a pregnancy risk change curve: Correlation analysis is performed between the pregnancy risk index corresponding to each pregnancy testing time period and the pregnancy testing time period to construct a pregnancy risk change curve, thereby determining the time of risk occurrence.

[0066] In this embodiment, the specific method for constructing the pregnancy risk change curve is as follows: a two-dimensional coordinate system is constructed with the pregnancy detection time as the horizontal axis and the pregnancy risk index as the vertical axis. Several points are marked in the constructed two-dimensional coordinate system for the pregnancy detection time and pregnancy risk index of each detection within each pregnancy detection time period to form a pregnancy risk change curve within the pregnancy detection time period, thereby determining the time of risk occurrence. Specifically, the inflection point is marked on the pregnancy risk change curve within the pregnancy detection time period.

[0067] Based on the labeled inflection points, the pregnancy risk change curve is divided into an ascending segment, a descending segment, and a flat segment. The percentage and magnitude of the ascending segment are then statistically analyzed. The ascending segments of the pregnancy risk change curve are sequentially numbered from front to back according to their position.

[0068] The probability of risk occurring in each upward segment is calculated based on the proportion and magnitude of the upward segment. Specifically, the proportion and magnitude of the upward segment are correlated using a probability calculation formula, as follows:

[0069] ,

[0070] in, This indicates the probability of risk occurring in each upward segment. and These represent the percentage of the upward segment and the magnitude of the upward segment, respectively. , as well as These represent the risk fitting parameters, which are estimated using the maximum likelihood method.

[0071] By comparing the probability of risk occurrence in each rising segment of the pregnancy risk change curve within the pregnancy testing period, the pregnancy testing period corresponding to the highest probability of risk occurrence is selected as the risk occurrence time.

[0072] It should be further explained that the statistics for the percentage and magnitude of upward segments are as follows:

[0073] The pregnancy testing time in the rising segment was extracted from the pregnancy risk change curve and the percentage of the rising segment was calculated by comparing it with the total pregnancy testing time.

[0074] Obtain the slope of each rising segment in the pregnancy risk change curve, and calculate the mean to obtain the amplitude of the rising segment.

[0075] It is important to understand that the percentage and magnitude of the rising segment in the pregnancy risk change curve during the pregnancy monitoring period are used as the basis for the average risk change probability. This is because these indicators comprehensively consider the changing trends at different stages of the pregnancy monitoring period, rather than focusing on just one point in time. By considering the percentage and magnitude of the rising segment, the overall dynamic changes in pregnancy risk can be assessed more comprehensively, thereby more accurately predicting the risk of gestational diabetes.

[0076] S6: Visualization: The risk occurrence time is sent to the user terminal according to the preset summary method, which includes report summary, picture summary and chart summary. Managers can use the summary content to focus on the risk occurrence time of gestational diabetes. The summary content includes multimodal time series matrix, pregnancy risk status and pregnancy risk change curve.

[0077] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0078] In conclusion, 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 intelligent prediction of gestational diabetes risk based on multimodal data fusion, characterized in that, include: S1: Set the prediction windows, which include the early pregnancy window, the mid-pregnancy window, and the late pregnancy window, and label them as T1, T2, and T3 respectively; S2: Risk detection data acquisition: Extract the current pregnancy detection time from the prediction window and label them as 1, 2, ..., j, ..., m, where j represents the j-th pregnancy detection time. Collect the risk detection data corresponding to each pregnancy detection time. The risk detection data includes physiological risk indicators and biochemical risk indicators, and then obtain the pregnancy detection time period corresponding to the risk detection data. S3: Multimodal time series matrix construction: Based on the pregnancy detection time period corresponding to the risk detection data, the time series corresponding to the risk detection data is generated in a preset order, and then a multimodal time series matrix is ​​constructed based on the time series; S4: Risk Analysis: Perform physiological risk analysis and biochemical risk analysis on the multimodal time series matrix corresponding to each pregnancy detection period, and analyze the pregnancy risk index for each pregnancy detection period accordingly. S5: Constructing a pregnancy risk change curve: Correlation analysis is performed on the pregnancy risk index corresponding to each pregnancy detection period and the pregnancy detection period to construct a pregnancy risk change curve, thereby determining the time of risk occurrence; S6: Visualization: The time of risk occurrence is sent to the user terminal according to the preset summary method. Managers can use the summary content to focus on the detection of gestational diabetes based on the time of risk occurrence. The physiological risk analysis is as follows: physiological risk features are extracted based on the multimodal time series matrix corresponding to each pregnancy detection period, including the standard deviation of the R-wave interval and the rate of weight gain; By comparing the standard deviation of the R-wave interval with the rate of weight gain, the risk coefficient of physiological abnormalities corresponding to each pregnancy detection period was obtained. The specific biochemical risk analysis is as follows: biochemical risk features are extracted based on the multimodal time series matrix corresponding to each pregnancy detection period, including blood glucose fluctuation coefficient and insulin secretion rate; By comparing and calculating the glycemic variability coefficient and insulin secretion rate, the risk coefficient of biochemical abnormalities corresponding to each pregnancy testing period is obtained, as specifically expressed as follows: ， in, This indicates the risk coefficient for biochemical abnormalities corresponding to each pregnancy testing period. , These represent the glycemic variability coefficient and insulin secretion rate, respectively, in the biochemical risk characteristics. , These represent the corresponding standard blood glucose variability coefficient and insulin secretion rate, respectively; The pregnancy risk index is based on a weighted sum of the risk coefficients for physiological and biochemical abnormalities corresponding to each pregnancy testing period, specifically expressed as follows: ， in, This indicates the pregnancy risk index corresponding to each pregnancy testing period. and These represent the physiological and biochemical abnormality risk coefficients for the (m-1)th pregnancy testing period, respectively. and These represent the weighting coefficients corresponding to the physiological abnormality risk coefficient and the biochemical abnormality risk coefficient, respectively. The specific method for constructing the pregnancy risk change curve is as follows: a two-dimensional coordinate system is constructed with the pregnancy detection time as the horizontal axis and the pregnancy risk index as the vertical axis. Several points are marked in the constructed two-dimensional coordinate system for the pregnancy detection time and pregnancy risk index of each detection within each pregnancy detection time period to form a pregnancy risk change curve within the pregnancy detection time period, thereby determining the time of risk occurrence. The specific time of risk occurrence is determined as follows: the inflection point is marked on the pregnancy risk change curve during the pregnancy testing period; Based on the labeled inflection points, the pregnancy risk change curve is divided into an ascending segment, a descending segment, and a flat segment. The percentage and magnitude of the ascending segment are then statistically analyzed. The ascending segments of the pregnancy risk change curve are sequentially numbered from front to back according to their position. The probability of risk occurrence for each upward segment is calculated based on the proportion and magnitude of the upward segment. Specifically, the proportion and magnitude of the upward segment are correlated through a probability calculation formula. By comparing the probability of risk occurrence in each rising segment of the pregnancy risk change curve within the pregnancy testing period, the pregnancy testing period corresponding to the highest probability of risk occurrence is selected as the risk occurrence time.

2. The intelligent prediction method for gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that: The specific process for collecting the physiological risk indicators includes: ECG signals are extracted using wearable devices, and the peak value of the R wave is identified based on the ECG signals, thereby calculating the interval between adjacent R waves. The root mean square of the R-wave is calculated based on the difference between adjacent R-wave intervals, and then the standard deviation of all R-wave intervals is calculated using the root mean square of the R-wave. The weight and height corresponding to the pregnancy detection time are extracted, and the body mass index is obtained through the body mass index calculation formula. Then, the weight gain rate is obtained by comparing the body mass index of adjacent pregnancy detection times with the time interval between adjacent pregnancy detection times. The standard deviation of R-wave intervals and the rate of weight gain were integrated as a physiological risk indicator.

3. The intelligent prediction method for gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that: The specific process for collecting the biochemical risk indicators includes: The pregnant woman's insulin status is identified through the prenatal examination terminal, and her fasting blood glucose and insulin concentration are extracted from it; Abnormalities in pregnant women's fasting blood glucose levels are identified and classified according to diagnostic criteria: ， in, The results of abnormal conditions in pregnant women at various pregnancy testing times were marked. This indicates the pregnant woman's fasting blood glucose levels at various pregnancy testing times. and These are the standard values ​​for fasting blood glucose in each abnormal state, specifically obtained according to the medical diagnosis and treatment standards for diabetes. The abnormal status markers for each pregnancy test time are summed. If the sum is not equal to 0, it is determined that there is an abnormality in the fasting blood glucose corresponding to each pregnancy test time; otherwise, it is determined that the fasting blood glucose corresponding to each pregnancy test time is normal. Extract the fasting blood glucose corresponding to each pregnancy test time with abnormalities, subtract the fasting blood glucose corresponding to adjacent pregnancy test times and take the absolute value, and compare it with the maximum value of the two to obtain the blood glucose fluctuation coefficient. The insulin concentrations were compared, and the insulin concentrations corresponding to adjacent pregnancy detection times were subtracted from each other. The insulin secretion rate was then obtained by comparing the results with the time intervals between adjacent pregnancy detection times. Integrating blood glucose variability and insulin secretion rate as a biochemical risk indicator.

4. The intelligent prediction method for gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that: The preset order refers to the chronological order of the risk detection data corresponding to the pregnancy detection time periods. The time series corresponding to the risk detection data includes physiological risk time series and biochemical risk time series. The physiological risk time series arranges the physiological risk indicators in a preset order, specifically as follows: Each time period corresponds to physiological risk characteristics including the standard deviation of the R-wave interval and the rate of weight gain. Similarly, a biochemical risk time series is generated, specifically represented as follows: Each time period corresponds to biochemical risk characteristics including glycemic variability and insulin secretion rate. Different modalities are aligned using pregnancy testing time periods, with missing testing time periods filled using time-series interpolation. A multimodal time series matrix is ​​constructed based on physiological and biochemical risk time series, specifically represented as follows: ， in, This represents a multimodal time series matrix, where each row corresponds to a pregnancy detection time period and each column corresponds to a modal feature. This represents the physiological risk characteristics corresponding to the (m-1)th pregnancy detection time period in the physiological risk time series. This represents the biochemical risk characteristics corresponding to the (m-1)th pregnancy testing time period in the biochemical risk time series.

Images