Gestational diabetes risk early warning method based on multi-modal data fusion

Abstract

The present application relates to the technical field of digital medical treatment, in particular to a pregnancy-induced diabetes risk early warning method based on multi-modal data fusion. Multi-modal data such as blood glucose time series data, exercise amount, body weight and placenta thickness of pregnant women during pregnancy are acquired, the peak value and back-off characteristics of blood glucose time series are analyzed, the exercise amount is combined, the blood glucose back-off trend degree is quantified, and the dynamic metabolic control capability is accurately reflected; further, on the basis of placenta thickness data, the body weight change condition, the exercise amount and the blood glucose back-off trend degree are combined to determine the placenta thickness estimation value time series data, and the physiological correlation characteristics between various data are improved; finally, the blood glucose back-off trend degree mean value and the placenta thickness sustained increasing trend characteristics are fused to reveal the correlation characteristics of blood glucose and abnormal placenta development, a neural network model is trained to capture multi-factor nonlinear relationships, and early and accurate risk prediction of pregnancy-induced diabetes (GDM) is realized.

Description

Technical Field

[0001] This invention relates to the field of digital medical technology, specifically to a method for early warning of gestational diabetes risk based on multimodal data fusion. Background Technology

[0002] Gestational diabetes mellitus (GDM) refers to abnormal blood sugar levels that occur during pregnancy, typically in women who did not previously have diabetes. Physiological changes during pregnancy, such as increased glucose requirements and heightened insulin resistance, can disrupt a pregnant woman's blood sugar regulation system, leading to elevated blood sugar. For the pregnant woman, GDM can cause infections, abnormally high amniotic fluid levels, and other problems; for the fetus, excessively high blood sugar levels can lead to excessive birth weight and complications such as hypoglycemia after birth. Therefore, early detection and management of GDM are crucial.

[0003] Existing methods for early warning of gestational diabetes typically rely on a single type of data (such as blood glucose or weight) to determine the risk level of gestational diabetes by comparing it to preset thresholds. This method fails to capture the temporal drift of blood glucose and weight caused by dietary fluctuations and lack of exercise during pregnancy, and lacks correlation analysis between ultrasound images (such as placental thickening) and physiological indicators (such as blood glucose fluctuations). Relying solely on a single data point not only has limited accuracy but also fails to comprehensively reflect the health status of pregnant women, lacking dynamic monitoring and comprehensive assessment of their health. Summary of the Invention

[0004] To address the issue of relying on a single type of data (such as blood glucose or weight) to determine the risk level of gestational diabetes by comparing it with preset thresholds, this method fails to capture the temporal drift patterns of blood glucose-weight caused by dietary fluctuations and lack of exercise during pregnancy. Furthermore, it lacks correlation analysis between ultrasound images (such as placental thickening) and physiological indicators (such as blood glucose fluctuations). Relying solely on single data points not only has limited accuracy but also fails to comprehensively reflect the health status of pregnant women, lacking the technical problem of dynamic monitoring and comprehensive assessment of their health status. The purpose of this invention is to provide a gestational diabetes risk warning method based on multimodal data fusion. The specific technical solution adopted is as follows:

[0005] Collect sample data for each pregnant woman during her pregnancy. The sample data includes placental thickness data, daily exercise volume, weight, and blood glucose time series data at each examination within a preset time period.

[0006] In each sample data, the numerical characteristics and changes of blood glucose values ​​in the daily blood glucose time series data are analyzed to determine the daily blood glucose decline performance. Within a preset time period, the blood glucose decline performance corresponding to each sample data is combined with the amount of exercise to analyze the numerical change trend and obtain the daily blood glucose decline trend.

[0007] In each sample data, based on placental thickness data, the characteristics of weight change are analyzed within a preset time period, and combined with the blood glucose decline trend, weight value and exercise volume to determine the time series data of placental thickness estimation.

[0008] The study analyzed the sustained increase in placental thickness estimates in the sample data over a preset time period, and combined this with the trend of blood glucose decline to train a neural network. The trained neural network was then used to predict the risk of gestational diabetes in the sample data.

[0009] Furthermore, the method for obtaining the blood glucose decline performance includes:

[0010] In each sample data, the daily blood glucose time series data is segmented based on time to obtain blood glucose data segments;

[0011] In each blood glucose data segment, the ratio of the maximum blood glucose value to the first blood glucose value is used as the peak anomaly coefficient for each blood glucose data segment;

[0012] In each blood glucose data segment, the changes in blood glucose values ​​are analyzed and combined with numerical characteristics to determine the ideal degree of blood glucose decline in each blood glucose data segment;

[0013] In each sample data, the peak abnormality coefficient of the blood glucose data segment is normalized and used as the blood glucose decline weight. The blood glucose decline ideality is weighted and fused based on the blood glucose decline weight, and the weighted result is normalized and used as the daily blood glucose decline performance of each sample data.

[0014] Furthermore, the method for obtaining the ideal degree of blood glucose reduction includes:

[0015] In each blood glucose data segment, the data segment following the time sequence of the maximum blood glucose value is taken as the blood glucose fallback segment. In the blood glucose fallback segment, the blood glucose values ​​that are lower than the preset blood glucose fallback threshold are taken as the target values. The ratio of the number of all target values ​​to the total number of blood glucose values ​​in the blood glucose fallback segment is taken as the blood glucose fallback ideal factor.

[0016] The ratio of the last blood glucose value in the time sequence of the blood glucose decline segment to the preset blood glucose decline threshold is normalized and negatively correlated. The product of this value and the blood glucose decline ideal factor is used as the blood glucose decline ideality of each blood glucose data segment.

[0017] Furthermore, the method for obtaining the blood glucose decline trend includes:

[0018] In each sample data, the blood glucose decline performance of all days within the preset time period is fitted according to time sequence, and the number of days corresponding to the maximum value point in the fitted curve is taken as the number of days to be evaluated.

[0019] Analyze the variation characteristics of exercise volume between the number of days to be evaluated and adjacent days, and determine the exercise influence coefficient corresponding to each number of days to be evaluated;

[0020] The product of the exercise impact coefficient and the blood glucose decline performance for each day to be evaluated is used as the relative blood glucose decline performance for each day to be evaluated.

[0021] Replace the blood glucose decline performance with the relative performance of blood glucose decline corresponding to the number of days to be evaluated, so as to obtain the time series data of blood glucose decline performance for each sample data within the preset time period.

[0022] In the time series data of blood glucose decline performance corresponding to each sample data, the normalized value of the difference between the corresponding value of each day and the corresponding value of the last day is used as the daily blood glucose decline trend.

[0023] Furthermore, the method for obtaining the motion influence coefficient includes:

[0024] The average of the exercise volume of two consecutive days in the time series of each day to be evaluated is used as the mean comparison value;

[0025] The normalized value of the difference between the amount of exercise and the mean for each day to be evaluated is used as the exercise influence coefficient for each day to be evaluated.

[0026] Furthermore, the method for obtaining the time-series data of the placental thickness estimate includes:

[0027] In each sample data, the weight of all days within the preset time period is fitted in time sequence, and the number of days corresponding to the maximum point in the fitted curve is taken as the number of days to be predicted.

[0028] Before each estimated number of days, the number of days corresponding to the placental thickness data closest to the time of each estimated number of days is used as the reference number of days;

[0029] The normalized value of the ratio between the weight of each day to be estimated and the weight of the corresponding reference day is used as the weight gain factor for each day to be estimated.

[0030] The sum of the exercise volume of all days between each day to be predicted and the corresponding reference day is normalized and used as the exercise performance value of each day to be predicted.

[0031] The weight gain factor for each day to be estimated is used to weight and fuse the blood glucose decline trend and exercise performance value to obtain the predictive placental growth coefficient for each day to be estimated.

[0032] The difference between the placental thickness data of the reference days for each number of days to be estimated and the placental thickness data closest to the time after each number of days to be estimated is used as the placental thickness fluctuation value for each number of days to be estimated.

[0033] The product of the placental thickness fluctuation value for each number of days to be estimated and the estimated placental growth coefficient, and the sum of the placental thickness data for the reference number of days, are used as the estimated placental thickness value for each number of days to be estimated.

[0034] Based on the estimated placental thickness for all days to be estimated and the placental thickness data at all examination times, a curve was fitted over time to obtain the time series data of the estimated placental thickness.

[0035] Furthermore, the method for obtaining the estimated placental growth factor includes:

[0036] The product of the weight gain factor and the blood glucose decline trend for each day to be estimated is used as the first placental growth factor. The product of the weight gain factor after negative correlation mapping for each day to be estimated and the exercise performance value for each day to be estimated is used as the second placental growth factor.

[0037] The sum of the first placental growth factor and the second placental growth factor corresponding to each number of days to be predicted is used as the predictive placental growth coefficient for each number of days to be predicted.

[0038] Furthermore, the method for obtaining the trained neural network includes:

[0039] Analyze the sustained increase in placental thickness estimates for each sample data within a preset time period to determine the degree of sustained increase in placental thickness for each sample data.

[0040] The normalized value of the product of the mean of the blood glucose decline trend of each sample data for all days within the preset time period and the product of the continuous increase trend of placental thickness of each sample data is used as the blood glucose-placental linkage performance of each sample data.

[0041] The risk level of gestational diabetes for each sample data is determined. The expression of blood glucose-placental linkage changes in the sample data is used as the input of the neural network, and the risk level of gestational diabetes is used as the output of the neural network. The neural network is then trained to obtain a well-trained neural network.

[0042] Furthermore, the method for obtaining the degree of continuous increase in placental thickness includes:

[0043] In the time series data of placental thickness estimates corresponding to each sample data, for any data value, the ratio of the data value to the adjacent previous data value is taken as the trend value of the data value.

[0044] The normalized value of the difference between the trend value of each data value and the preset change threshold is used as the change growth factor for each data value.

[0045] In the time series data of placental thickness estimates for each sample data, the placental thickness estimates are weighted and fused using the data value change growth factor, and the normalized value of the weighted result is used as the degree of continuous growth trend of placental thickness for each sample data.

[0046] Furthermore, the neural network may be a CNN.

[0047] The present invention has the following beneficial effects:

[0048] This invention acquires sample data from each pregnant woman during her pregnancy, including various data types such as time-series blood glucose data, exercise volume, weight, and placental thickness data. It overcomes the limitations of traditional methods that rely on a single indicator (such as blood glucose or weight alone), comprehensively covering multi-dimensional information on metabolism, behavior, and imaging, thus improving the completeness of risk assessment. Analyzing the numerical characteristics and trends of time-series blood glucose data, combined with exercise data, quantifies the "blood glucose decline trend," accurately reflecting the pregnant woman's dynamic blood glucose control ability during pregnancy and helping to analyze whether she has a good blood glucose decline status. Because placental secretion of insulin-antagonistic hormones can lead to decreased insulin sensitivity in surrounding tissues, potentially causing insulin resistance and increasing the risk of gestational diabetes, and because changes in maternal weight usually reflect fetal growth and the pregnant woman's nutritional status, it can be used to indirectly characterize placental growth characteristics. Therefore, further, based on placental thickness data, it combines the numerical characteristics of weight changes with blood glucose decline trend, weight, and exercise volume to determine the estimated placental thickness time-series data, making it more consistent with the individual pregnant woman's metabolic state and improving the physiological relevance of indicators such as placental thickness. Finally, by integrating the "blood glucose decline trend" with the "continuous increase trend of placental thickness," the potential pathological link between hyperglycemia and abnormal placental development was directly revealed, providing a deep biological basis for risk prediction. The neural network model was trained to capture the nonlinear relationship of multiple factors, significantly improving the accuracy of early prevention and risk prediction. Attached Figure Description

[0049] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating a method for early warning of gestational diabetes risk based on multimodal data fusion, as provided in one embodiment of the present invention.

[0051] Figure 2 A flowchart illustrating a method for obtaining blood glucose decline performance according to an embodiment of the present invention;

[0052] Figure 3 This is a flowchart illustrating a method for obtaining time-series data of placental thickness estimation, as provided in an embodiment of the present invention. Detailed Implementation

[0053] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a gestational diabetes risk warning method based on multimodal data fusion proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0055] The following description, in conjunction with the accompanying drawings, details a specific scheme for a gestational diabetes risk warning method based on multimodal data fusion provided by the present invention.

[0056] Please see Figure 1 The diagram illustrates a flowchart of a method for early warning of gestational diabetes risk based on multimodal data fusion, according to an embodiment of the present invention. The method includes the following steps:

[0057] Step S1: Obtain sample data for each pregnant woman during her pregnancy. The sample data includes placental thickness data, daily exercise volume, weight, and blood glucose time series data during each examination within a preset time period.

[0058] Gestational diabetes not only affects the health of pregnant women but also increases the risk of birth defects, macrosomia, and premature birth in the fetus. In particular, pregnant women in the second trimester (13-27 weeks of gestation) experience a significant increase in insulin-antagonistic hormones secreted by the placenta (such as human placental lactogen, progesterone, estrogen, and cortisol), leading to decreased insulin sensitivity in peripheral tissues and potentially causing insulin resistance, thus increasing the risk of gestational diabetes. At the same time, the second trimester is known as the period of worsening insulin resistance (IR) and the golden period for intervention. Therefore, pregnant women in the second trimester should be closely monitored to analyze their potential risk of gestational diabetes.

[0059] Therefore, in this embodiment of the present invention, sample data is obtained for each pregnant woman in the second trimester. The sample data includes placental thickness data obtained from each ultrasound examination within a preset time period, daily exercise volume (specifically, step count in this embodiment of the present invention, recorded by wearable devices such as smart bracelets, sports watches, etc.) within the preset time period, daily weight (weighed by an electronic scale, with three weighings per day and the average value taken), and blood glucose time series data (relying on rtCGM technology to continuously obtain blood glucose values ​​for two hours after each meal each day, and then stitching them together to obtain daily blood glucose time series data).

[0060] Thus, sample data of multiple pregnant women during pregnancy can be obtained. It should be noted that in this embodiment of the present invention, the preset time period is set to 13-27 weeks of pregnancy, and the number of sample data is set to at least 1000. The specific value can be adjusted according to the implementation scenario and is not limited here.

[0061] In the embodiments of the present invention, the collection and acquisition of personal information data are authorized by the relevant users, and the process does not violate relevant laws and regulations, nor does it violate public order and good morals.

[0062] Step S2: In each sample data, analyze the numerical characteristics and changes of blood glucose values ​​in the daily blood glucose time series data to determine the daily blood glucose decline performance; within a preset time period, integrate the blood glucose decline performance corresponding to each sample data with the amount of exercise, analyze the numerical change trend, and obtain the daily blood glucose decline trend.

[0063] For healthy pregnant women, blood sugar levels rise after meals but usually return to normal within two hours. However, for pregnant women with gestational diabetes, the surge in placental hormone HPL during mid-pregnancy can block insulin receptor phosphorylation. When these women consume the same amount of carbohydrates, their insulin demand increases but its efficiency decreases, causing glucose to remain in the blood for a longer period. This can prevent blood sugar from returning to normal, resulting in a persistent hyperglycemic state. Therefore, to capture this potential pathological blood sugar fluctuation pattern in pregnant women, we analyze the numerical characteristics and changes in blood sugar values ​​in daily time-series data for each sample to determine the daily blood sugar regression performance and evaluate the daily blood sugar fluctuation characteristics of the pregnant women in the corresponding sample data.

[0064] Preferably, in one embodiment of the present invention, the method for obtaining the blood glucose decline performance includes:

[0065] Please see Figure 2 The diagram illustrates a method flowchart for obtaining the performance of blood glucose decline in one embodiment of the present invention. The method includes the following steps:

[0066] Step S201: In each sample data, the daily blood glucose time series data is segmented based on time to obtain blood glucose data segments.

[0067] In step S1, since the daily blood glucose time series data is obtained by splicing blood glucose values ​​obtained continuously for two hours after each meal, the time of each meal can be used as the starting point and two hours as the length, so as to segment the required blood glucose data segment from the blood glucose time series data.

[0068] Step S202: In each blood glucose data segment, determine the peak abnormality coefficient of each blood glucose data segment based on the numerical characteristics of the blood glucose value.

[0069] Postprandial blood glucose levels in pregnant women are determined by factors such as the glycemic index of food and the compensatory capacity of their own insulin. Insulin sensitivity also varies at different stages of the day. Even if the same pregnant woman eats the same food, her blood glucose levels will differ at different stages. Therefore, the baseline blood glucose value and the peak blood glucose value of each blood glucose data segment can be compared and analyzed to assess abnormal blood glucose rise in each segment. In each blood glucose data segment, the ratio of the maximum blood glucose value (representing the peak blood glucose value) to the first blood glucose value (representing the baseline blood glucose value) is used as the peak abnormality coefficient for each blood glucose data segment. The larger the peak abnormality coefficient, the greater the increase in blood glucose after eating, and the more likely there is an abnormal increase in blood glucose.

[0070] Step S203: In each blood glucose data segment, analyze the changes in blood glucose values ​​and combine them with numerical characteristics to determine the ideal degree of blood glucose decline for each blood glucose data segment.

[0071] Gestational diabetes not only causes abnormally high blood sugar levels, but also delays the return of blood sugar to normal levels. In other words, in cases of potentially pathological blood sugar fluctuations, after blood sugar reaches its peak, there will be a delay in the subsequent decline, meaning that it does not return to below the pregnant woman's blood sugar threshold within two hours after a meal.

[0072] Therefore, within each blood glucose data segment, the maximum blood glucose value is used as a dividing line, and the data segment following it is designated as the blood glucose decline segment. Within the blood glucose decline segment, blood glucose values ​​below a preset decline threshold are taken as target values ​​(i.e., blood glucose values ​​where the decline meets expectations). The ratio of all target values ​​to the total number of blood glucose values ​​in the decline segment is used as the ideal blood glucose decline factor. The larger the ideal blood glucose decline factor, the more consistent the fluctuations in blood glucose values ​​within that data segment are with normal conditions.

[0073] The last blood glucose value in a blood glucose decline segment often reflects the final state of the decline. A smaller value indicates a more ideal decline. Therefore, the ratio of the last blood glucose value in the time sequence of the decline segment to a preset blood glucose decline threshold is normalized and negatively correlated. This normalized value is then multiplied by a blood glucose decline ideality factor. This product is used as the blood glucose decline ideality for each blood glucose data segment. A higher blood glucose decline ideality indicates a more normal fluctuation pattern in blood glucose values ​​within that segment, and a closer approximation to the ideal state. The normalization and negative correlation mapping can be performed using the following formula: ,in, Let x represent the normalization function, and let x represent the independent variable.

[0074] It should be noted that the preset blood glucose drop threshold in this embodiment of the invention is 7.8 mmol / L.

[0075] Step S204: In each sample data, combine the peak abnormality coefficient of the daily blood glucose data segment with the ideal blood glucose decline to obtain the daily blood glucose decline performance of each sample data.

[0076] For pregnant women, a higher peak abnormality coefficient (PAC) in their blood glucose data segments (used as a marker signal) indicates greater influence from food and insulin metabolism. Therefore, greater attention should be paid to the ideal degree of blood glucose decline (focusing on the possibility of a delay in the blood glucose decline process). Thus, the two aforementioned indicators can be combined to evaluate the daily blood glucose decline performance of each sample. Within each sample, the PAC of the blood glucose data segments is normalized (the normalization here uses...). The blood glucose decline weight is used as a function to measure the ideal blood glucose decline. A larger weight indicates a greater need to focus on the ideal blood glucose decline. Therefore, a weighted fusion of the ideal blood glucose decline is performed based on the weight. Specifically, in each sample data point, the weight of the blood glucose decline for each daily blood glucose data segment is multiplied by the ideal blood glucose decline, and the sum of these products is normalized. This normalized sum is then used as the daily blood glucose decline performance for each sample data point. Based on the aforementioned analysis, a higher blood glucose decline performance indicates more stable blood glucose levels and a closer approximation to a normal state. Normalization is a technique well-known to those skilled in the art. The normalization function can be linear normalization, standard normalization, or, for example, maximum / minimum value normalization. Specific normalization methods are not limited here.

[0077] Gestational diabetes mellitus (GDM) is not merely a manifestation of blood glucose regulation imbalance, but a multi-factorial (including diet, hormones, and exercise) and multi-stage evolutionary process. These factors work together to cause abnormal blood glucose decline and may also lead to complications such as placental thickening, gradually affecting the health of the pregnant woman and the development of the fetus. Dietary structure has a significant impact on blood glucose fluctuations in pregnant women. High-sugar, high-carbohydrate diets cause a rapid rise in postprandial blood glucose, and hormonal changes lead to insulin resistance. Furthermore, insufficient physical strength or concerns about the impact of exercise on the fetus may result in insufficient exercise, further exacerbating insulin resistance and making blood glucose control difficult, especially during the postprandial blood glucose decline process. Therefore, it is possible to further combine the blood glucose decline performance of each sample data with exercise volume to analyze the trend of numerical changes, thereby determining the daily blood glucose decline trend, which can be used to describe the potentially persistent delayed trend of blood glucose decline.

[0078] Preferably, in one embodiment of the present invention, the method for obtaining the blood glucose decline trend includes:

[0079] In each sample data, the blood glucose decline performance of all days within the preset time period is fitted according to time sequence (the least squares method can be used, and the well-known techniques will not be elaborated). The development of metabolic abnormalities during pregnancy is often a dynamic process. Blood glucose control ability may gradually become abnormal from normal, and local peaks often correspond to the critical point where metabolic capacity begins to show extreme changes. Therefore, the number of days corresponding to the maximum value point in the fitted curve is taken as the number of days to be evaluated.

[0080] Exercise is a behavioral factor regulating blood glucose. Sudden changes in daily exercise volume can affect the degree of blood glucose decline. Therefore, this study analyzes the variation characteristics of exercise volume between the assessed days and adjacent days to determine the exercise influence coefficient for each assessed day. The average exercise volume of the two nearest adjacent days for each assessed day is used as the mean comparison value. The difference between the exercise volume for each assessed day and the mean comparison value is calculated. A positive and larger difference indicates higher exercise volume on the assessed days, which will reduce insulin resistance to some extent and improve blood glucose control, thus having a positive feedback effect on the degree of blood glucose decline. Conversely, a negative and smaller difference indicates insufficient exercise, which may further aggravate insulin resistance, leading to difficulty in blood glucose control, thus having a negative feedback effect on the degree of blood glucose decline. This difference is then normalized to serve as the exercise influence coefficient for each assessed day. The normalization process can be performed using... function.

[0081] Based on the foregoing analysis, the larger the exercise influence coefficient, the more positive feedback it has on the performance of blood glucose decline. Therefore, the product of the exercise influence coefficient and the blood glucose decline performance for each day to be evaluated can be used as the relative performance of blood glucose decline for each day to be evaluated. Then, the relative performance of blood glucose decline for each day to be evaluated replaces the blood glucose decline performance, and curve fitting is performed again to obtain the time series data of blood glucose decline performance for each sample data within the preset time period. At this time, the blood glucose decline performance incorporates the exercise data, so it can reflect the true metabolic situation to a certain extent, making the subsequent trend analysis more accurate.

[0082] Finally, since the last value in the time series data of blood glucose decline performance represents the "final metabolic state," the difference between the daily value and the last day's value is calculated for each sample's blood glucose decline performance time series data. This difference reflects the magnitude of change between the daily blood glucose decline state and the last day; the larger the value, the more obvious the decline trend. Therefore, the normalized value of this difference is used as the daily blood glucose decline trend degree. Based on the aforementioned analysis, a larger blood glucose decline trend degree indicates that the blood glucose data for that day is more likely to have a greater decline trend. The normalization process here can be performed using... function.

[0083] Step S3: In each sample data, based on the placental thickness data, analyze the numerical change characteristics of body weight within a preset time period, and combine it with the blood glucose decline trend, body weight and exercise volume to determine the time series data of the estimated placental thickness.

[0084] For pregnant women, ultrasound examinations are performed less frequently, especially in the second trimester, so the amount of data on placental thickness may be limited. To compensate for this lack of data, other methods can be used to infer placental thickness or to estimate it indirectly.

[0085] Changes in a pregnant woman's weight usually reflect fetal growth and the mother's nutritional status. Weight gain may indirectly reflect placental growth, as placental size and fetal growth are related. However, weight changes are usually influenced by a variety of factors and may not be entirely linearly related to placental thickness. Therefore, more data is needed to model the changes.

[0086] In this embodiment of the invention, for each sample data, based on its placental thickness data, the numerical change characteristics of body weight are analyzed within a preset time period, and combined with the blood glucose decline trend, body weight value and exercise volume, so as to indirectly analyze the placental thickness and determine the time series data of the estimated placental thickness value.

[0087] Preferably, in one embodiment of the present invention, the method for obtaining time-series data of placental thickness estimation includes:

[0088] Please see Figure 3 The diagram illustrates a method flowchart for obtaining time-series data of placental thickness estimation values ​​according to an embodiment of the present invention. The method includes the following steps:

[0089] Step S301: In each sample data, analyze the numerical characteristics of weight and filter out the number of days to be predicted.

[0090] Pregnant women's weight gain during pregnancy is not linear and may have phases of acceleration or deceleration. Therefore, in each sample data, the weight of all days within the preset time period is fitted in time sequence (this can be based on the least squares method, which is a well-known technique and will not be elaborated here). The maximum points (such as local peaks) in the fitted curve often correspond to the "critical days" of weight change. These days may be accompanied by significant changes in placental thickness (such as compensatory thickening of the placenta due to accelerated fetal growth). Therefore, the days corresponding to the maximum points in the fitted curve are used as the number of days to be inferred.

[0091] Step S302: Determine the reference number of days corresponding to each number of days to be predicted.

[0092] Before each estimated number of days, the number of days corresponding to the placental thickness data closest to each estimated number of days is taken as the reference number of days. In this way, the placental thickness of each estimated number of days can be estimated in subsequent processes by using the data between each estimated number of days and the reference number of days.

[0093] Step S303: Compare the weight difference characteristics between each number of days to be predicted and the corresponding reference number of days to determine the weight gain factor for each number of days to be predicted.

[0094] Weight gain is a combined result of fetal growth, placental development, and increased maternal body fluids. Therefore, the normalized ratio between the weight of each day to be estimated and the weight of the corresponding reference day is used as the weight gain factor for each day to be estimated. The larger the weight gain factor, the more significant the weight gain. Normalization is a technique well-known to those skilled in the art, and the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0095] Step S304: In terms of time sequence, integrate the amount of motion over all days between each day to be predicted and the corresponding reference day to determine the motion performance value of each day to be predicted.

[0096] The greater the weight gain, the larger the weight gain factor, the more attention should be paid to the decline in blood sugar levels between the reference number of days and the estimated number of days when assessing placental thickness for the estimated number of days (corresponding to weight gain status); conversely, the smaller the weight gain, the more attention should be paid to exercise performance between the reference number of days and the estimated number of days when assessing placental thickness for the estimated number of days (corresponding to weight loss status).

[0097] Therefore, in this sub-step, it is necessary to obtain the exercise performance data: the sum of the exercise volume of all days between each day to be predicted and the corresponding reference day is normalized and used as the exercise performance value for each day to be predicted. The larger the exercise performance value, the more sufficient the exercise volume. Normalization is a technique well known to those skilled in the art. The normalization function can be linear normalization or standard normalization, etc., and the specific normalization method is not limited here.

[0098] Step S305: Use the weight gain factor of each day to be predicted to weight the blood glucose decline trend and the exercise performance value to weighted fuse, thereby obtaining the predictive placental growth coefficient for each day to be predicted.

[0099] Based on the logic in step S304, a weight gain factor can be used to weightedly fuse the blood glucose decline trend and exercise performance value: When the weight gain factor is larger, the blood glucose decline trend has higher reference value; therefore, the product of the weight gain factor and the blood glucose decline trend for each day to be estimated is used as the first placental growth factor. When the weight gain factor is smaller, the exercise performance value has higher reference value; therefore, the product of the negatively correlated weight gain factor for each day to be estimated and the exercise performance value for each day to be estimated is used as the second placental growth factor. Since the weight gain factor is a normalized value, the negative correlation mapping here can be performed using a formula... , where x represents the independent variable.

[0100] Finally, the sum of the first placental growth factor and the second placental growth factor corresponding to each number of days to be estimated is used as the predicted placental growth coefficient for each number of days to be estimated. The predicted placental growth coefficient integrates multiple data, thus it can more accurately reflect the possible changes in placental thickness. Moreover, the larger the value, the greater the possibility of placental thickness increase at the number of days to be estimated.

[0101] Step S306: Based on placental thickness data and the predictive placental growth coefficient for each number of days to be estimated, determine the estimated placental thickness for each number of days to be estimated, thereby obtaining time series data of placental thickness estimates.

[0102] The difference between the placental thickness data of the reference days for each day to be estimated and the placental thickness data closest to the time after each day to be estimated is used as the placental thickness fluctuation value for each day to be estimated. The placental thickness fluctuation value reflects the natural change in placental thickness between two ultrasound examinations before and after the day to be estimated.

[0103] Then, the placental thickness fluctuation value is amplified or reduced by the predicted placental growth factor: the product of the placental thickness fluctuation value for each day to be estimated and the predicted placental growth factor is calculated, and this product is used as the adjustment amount. The larger the adjustment amount, the greater the increase in placental thickness for the day to be estimated. The placental thickness data of the reference days corresponding to the day to be estimated is used as the baseline value. The baseline value is added to the adjustment amount, and the sum is used as the estimated placental thickness value for each day to be estimated.

[0104] Finally, based on the estimated placental thickness for all days to be estimated and the placental thickness data at all examination times, curve fitting is performed on the time series (curve fitting is based on the least squares method, which is a well-known technique, and the specific process will not be elaborated). Thus, the time series data of the placental thickness estimate can be obtained (the estimated placental thickness for each day within a preset time period can be obtained based on the time series data of the placental thickness estimate).

[0105] Step S4: Analyze the continuous increase of the estimated placental thickness in the sample data within the preset time period, and integrate the blood glucose decline trend to train the neural network. The trained neural network is then used to predict the risk of gestational diabetes in the sample data to be tested.

[0106] The delayed decline in blood glucose levels during pregnancy reflects insulin resistance. When blood glucose remains high for an extended period after meals, it not only negatively impacts the mother's health but may also promote excessive placental growth through placental glucose-insulin exchange. Simultaneously, placental thickening may lead to increased insulin and glucose delivery to the fetus, potentially causing excessive fetal weight gain and further exacerbating the mother's insulin resistance, creating a vicious cycle. This cycle not only affects the rate of blood glucose decline but may also worsen gestational diabetes. Therefore, in this embodiment of the invention, the increasing trend of estimated placental thickness within a preset time period can be analyzed and integrated with the blood glucose decline trend. This allows for the determination of a more comprehensive feature index to train the neural network, resulting in a better and more accurate predictive ability, enabling effective risk prediction of the tested sample data.

[0107] Preferably, in one embodiment of the present invention, the method for obtaining a trained neural network includes:

[0108] First, we analyze the sustained increase in placental thickness estimates for each sample data point within a preset time period to determine the sustained increasing trend of placental thickness for each sample data point: In the time series data of placental thickness estimates for each sample data point, for any given data value, the ratio of that data value to its preceding adjacent data value is used as the trend value of that data value. A trend value greater than 1 and the larger it is, indicates a significant increasing trend in placental thickness; conversely, a trend value less than 1 and the smaller it is, indicates a significant decreasing trend in placental thickness. Since the first data value has no preceding adjacent data value, the trend value of the first data value is set to be the same as the trend value of the second data value.

[0109] Therefore, the difference between the trend value of each data point and the preset change threshold (set to 1) is calculated. Based on the previous analysis, a positive and larger difference indicates a stronger growth trend, while a negative and smaller difference indicates a stronger decline trend. Therefore, the normalized value of this difference is used as the growth factor for each data point. The normalization process here can be performed using... function.

[0110] Then, in the time series data of placental thickness estimates for each sample data, the growth factor of the data value change is used as the growth weight. The placental thickness estimates are weighted and fused using the growth factor of the data value change, and the normalized value of the weighted result is used as the placental thickness continuous growth trend degree for each sample data. The larger the placental thickness continuous growth trend degree, the more obvious the growth trend of the placental thickness. Normalization is a technique well known to those skilled in the art. The choice of normalization function can be linear normalization or standard normalization, etc., and the specific normalization method is not limited here.

[0111] The greater the sustained downward trend of postprandial blood glucose in pregnant women, the greater the increasing trend of placental thickness. This indicates that the two indicators maintain similar correlation and change characteristics under the influence of factors such as diet. Therefore, the normalized value of the product of the mean of the downward trend of blood glucose and the sustained increasing trend of placental thickness for each sample data over the preset period is used as the performance of the correlation between blood glucose and placental changes for each sample data.

[0112] Finally, a professional doctor can conduct a manual assessment to determine the gestational diabetes risk level (e.g., level 1, level 2, level 3) for each sample data. The blood glucose-placental linkage variation performance of the sample data is used as the input of a neural network (CNN is used in this embodiment of the invention), and the gestational diabetes risk level is used as the output of the neural network, thereby training the neural network to obtain a trained neural network.

[0113] The training process is a well-known technique. In short, after assessing the risk level of all sample data, the training set and validation set are divided in a 7:3 ratio. The neural network is trained using the training set samples, with the cross-entropy function as the loss function. Gradient descent is used to train until the loss function converges. The robustness of the training results is then verified using the validation set, thus obtaining the trained neural network.

[0114] After obtaining the trained neural network, the risk of gestational diabetes can be predicted from the test sample data. This involves calculating the blood glucose-placental linkage variation of the test sample data, and then obtaining the potential risk level of the test sample based on the trained neural network.

[0115] In summary, this invention acquires sample data from each pregnant woman during her pregnancy, including various data types such as time-series blood glucose data, exercise volume, weight, and placental thickness data. It overcomes the limitations of traditional methods that rely on a single indicator (such as blood glucose or weight alone), comprehensively covering multi-dimensional information on metabolism, behavior, and imaging, thus improving the completeness of risk assessment. Analyzing the numerical characteristics and trends of time-series blood glucose data, combined with exercise data, quantifies the "blood glucose decline trend," accurately reflecting the pregnant woman's dynamic blood glucose control ability during pregnancy and helping to analyze whether she has a good blood glucose decline status. Because placental secretion of insulin-antagonistic hormones can lead to decreased insulin sensitivity in surrounding tissues, potentially causing insulin resistance and increasing the risk of gestational diabetes, and because changes in maternal weight usually reflect fetal growth and the pregnant woman's nutritional status, it can be used to indirectly characterize placental growth characteristics. Therefore, further, based on placental thickness data, combining the numerical characteristics of weight changes with blood glucose decline trend, weight, and exercise volume, time-series data for estimated placental thickness is determined, making it more consistent with the individual pregnant woman's metabolic state and improving the physiological relevance of indicators such as placental thickness. Finally, by integrating the "blood glucose decline trend" with the "continuous increase trend of placental thickness," the potential pathological link between hyperglycemia and abnormal placental development was directly revealed, providing a deep biological basis for risk prediction. The neural network model was trained to capture the nonlinear relationship of multiple factors, significantly improving the accuracy of early prevention and risk prediction.

[0116] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0117] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for early warning of gestational diabetes mellitus risk based on multimodal data fusion, characterized in that, The method includes: Collect sample data for each pregnant woman during her pregnancy. The sample data includes placental thickness data, daily exercise volume, weight, and blood glucose time series data at each examination within a preset time period. In each sample data, the numerical characteristics and changes of blood glucose values ​​in the daily blood glucose time series data are analyzed to determine the daily blood glucose decline performance. Within a preset time period, the blood glucose decline performance corresponding to each sample data is combined with the amount of exercise to analyze the numerical change trend and obtain the daily blood glucose decline trend. In each sample data, based on placental thickness data, the characteristics of weight change are analyzed within a preset time period, and combined with the blood glucose decline trend, weight value and exercise volume to determine the time series data of placental thickness estimation. The analysis showed a sustained increase in the estimated placental thickness of the sample data over a preset period, and combined this with the trend of blood glucose decline to train the neural network. The trained neural network was then used to predict the risk of gestational diabetes in the sample data. The method for obtaining the blood glucose decline trend includes: in each sample data, fitting the blood glucose decline performance of all days within a preset time period according to time sequence, and taking the days corresponding to the maximum point in the fitted curve as the days to be evaluated; analyzing the change characteristics of exercise volume between the days to be evaluated and adjacent days to determine the exercise influence coefficient corresponding to each day to be evaluated; taking the product of the exercise influence coefficient corresponding to each day to be evaluated and the blood glucose decline performance as the relative blood glucose decline performance corresponding to each day to be evaluated; replacing the blood glucose decline performance with the relative blood glucose decline performance corresponding to the days to be evaluated, thereby obtaining the time series data of blood glucose decline performance corresponding to each sample data within the preset time period; in the time series data of blood glucose decline performance corresponding to each sample data, taking the normalized value of the difference between the corresponding value of each day and the corresponding value of the last day as the daily blood glucose decline trend; The method for obtaining time-series data for placental thickness estimation includes: in each sample data set, fitting the weight of all days within a preset time period according to time sequence, and using the days corresponding to the maximum points in the fitted curve as the days to be estimated; before the time sequence of each day to be estimated, using the days corresponding to the placental thickness data closest to the time of each day to be estimated as reference days; normalizing the ratio between the weight of each day to be estimated and the weight of the corresponding reference days as the weight gain factor for each day to be estimated; normalizing the sum of the exercise volume of all days between each day to be estimated and the corresponding reference days as the exercise performance value for each day to be estimated; and using each day to be estimated... The weight gain factor is weighted and fused with the blood glucose decline trend and exercise performance values ​​to obtain the predicted placental growth coefficient for each number of days to be estimated. The difference between the placental thickness data of the reference days for each number of days to be estimated and the placental thickness data of the nearest time after each number of days to be estimated is used as the placental thickness fluctuation value for each number of days to be estimated. The product of the placental thickness fluctuation value and the predicted placental growth coefficient for each number of days to be estimated, and the sum of the placental thickness data of the reference days, is used as the estimated placental thickness value for each number of days to be estimated. Based on the estimated placental thickness values ​​for all numbers of days to be estimated and the placental thickness data at all examination times, a curve is fitted over time to obtain the time series data of the estimated placental thickness.

2. The method for early warning of gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that, Methods for obtaining the performance of blood glucose decline include: In each sample data, the daily blood glucose time series data is segmented based on time to obtain blood glucose data segments; In each blood glucose data segment, the ratio of the maximum blood glucose value to the first blood glucose value is used as the peak anomaly coefficient for each blood glucose data segment; In each blood glucose data segment, the changes in blood glucose values ​​are analyzed and combined with numerical characteristics to determine the ideal degree of blood glucose decline in each blood glucose data segment; In each sample data, the peak abnormality coefficient of the blood glucose data segment is normalized and used as the blood glucose decline weight. The blood glucose decline ideality is weighted and fused based on the blood glucose decline weight, and the weighted result is normalized and used as the daily blood glucose decline performance of each sample data.

3. The method for early warning of gestational diabetes risk based on multimodal data fusion according to claim 2, characterized in that, Methods for determining the ideal degree of blood sugar reduction include: In each blood glucose data segment, the data segment following the time sequence of the maximum blood glucose value is taken as the blood glucose fallback segment. In the blood glucose fallback segment, the blood glucose values ​​that are lower than the preset blood glucose fallback threshold are taken as the target values. The ratio of the number of all target values ​​to the total number of blood glucose values ​​in the blood glucose fallback segment is taken as the blood glucose fallback ideal factor. The ratio of the last blood glucose value in the time sequence of the blood glucose decline segment to the preset blood glucose decline threshold is normalized and negatively correlated. The product of this value and the blood glucose decline ideal factor is used as the blood glucose decline ideality of each blood glucose data segment.

4. The method for early warning of gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that, Methods for obtaining the motion influence coefficient include: The average of the exercise volume of two consecutive days in the time series of each day to be evaluated is used as the mean comparison value; The normalized value of the difference between the amount of exercise and the mean for each day to be evaluated is used as the exercise influence coefficient for each day to be evaluated.

5. The method for early warning of gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that, Methods for obtaining the predictive placental growth factor include: The product of the weight gain factor and the blood glucose decline trend for each day to be estimated is used as the first placental growth factor. The product of the weight gain factor after negative correlation mapping for each day to be estimated and the exercise performance value for each day to be estimated is used as the second placental growth factor. The sum of the first placental growth factor and the second placental growth factor corresponding to each number of days to be predicted is used as the predictive placental growth coefficient for each number of days to be predicted.

6. The method for early warning of gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that, Methods for obtaining a pre-trained neural network include: Analyze the sustained increase in placental thickness estimates for each sample data within a preset time period to determine the degree of sustained increase in placental thickness for each sample data. The normalized value of the product of the mean of the blood glucose decline trend of each sample data for all days within the preset time period and the product of the continuous increase trend of placental thickness of each sample data is used as the blood glucose-placental linkage performance of each sample data. The risk level of gestational diabetes for each sample data is determined. The expression of blood glucose-placental linkage changes in the sample data is used as the input of the neural network, and the risk level of gestational diabetes is used as the output of the neural network. The neural network is then trained to obtain a well-trained neural network.

7. The method for early warning of gestational diabetes risk based on multimodal data fusion according to claim 6, characterized in that, Methods for obtaining the degree of continuous increase in placental thickness include: In the time series data of placental thickness estimates corresponding to each sample data, for any data value, the ratio of the data value to the adjacent previous data value is taken as the trend value of the data value. The normalized value of the difference between the trend value of each data value and the preset change threshold is used as the change growth factor for each data value. In the time series data of placental thickness estimates for each sample data, the placental thickness estimates are weighted and fused using the data value change growth factor, and the normalized value of the weighted result is used as the degree of continuous growth trend of placental thickness for each sample data.

8. The method for early warning of gestational diabetes risk based on multimodal data fusion according to claim 1, characterized in that, A neural network such as CNN can be used.

Images