Method and system for assessing fetal development following maternal microplastic exposure

Abstract

The present invention provides a method and system for assessing fetal development based on maternal microplastic exposure, which relates to the technical field of medical data analysis. The method includes: obtaining historical medical examination data of pregnant women and standardizing it, establishing a multidimensional characterization index for microplastics, calculating the time-series cumulative concentration and bioconcentration coefficient of microplastics, constructing a distribution characteristic map, predicting the fetal development risk score through fuzzy reasoning combined with support vector regression, and generating an assessment report. This method can achieve an accurate assessment of the impact of microplastic exposure on fetal development and provide a scientific basis for clinical intervention.

Description

Technical Field

[0001] The present invention relates to the field of medical data analysis technology, and in particular to a method and system for assessing fetal development based on maternal microplastic exposure. Background Art

[0002] Microplastics refer to plastic particles with a diameter of less than 5 mm. They are widely present in the environment, including air, water, and food. With the widespread use of plastic products and the intensification of environmental pollution, microplastics have become a new type of environmental pollutant, posing a potential threat to human health. Microplastics can enter the human body through the respiratory tract, digestive tract, and other pathways, and may affect fetal development through the placenta.

[0003] Current assessment methods for the impact of maternal microplastic exposure on fetal development have many shortcomings. They still rely on microplastic detection data from a single time point, which cannot reflect the true situation of long-term cumulative exposure during pregnancy. They ignore the dynamic changes in microplastic exposure and its temporal matching relationship with the critical period of fetal development. There is a lack of a comprehensive assessment system for complex microplastic mixtures, which cannot fully reflect the synergistic or antagonistic effects of microplastics of different types and particle sizes. There is also a lack of a methodological framework for integrating microplastic exposure with the development of multiple fetal systems, making it difficult to accurately predict and assess the long-term effects of microplastic exposure on fetal development.

[0004] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for assessing fetal development based on maternal microplastic exposure, which can at least solve some of the problems existing in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a method for assessing fetal development in response to maternal microplastic exposure, which method does not involve disease diagnosis and treatment, and includes:

[0007] Obtaining historical medical examination data of the pregnant woman to be tested and performing standardized preprocessing according to time series to generate a structured time series feature matrix, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data;

[0008] A multidimensional characterization index was established based on the types and particle size distribution of microplastics. The microplastic detection data in the structured time series feature matrix was classified and statistically analyzed to calculate the time series cumulative concentration and short-term fluctuation amplitude of microplastics. The bioaccumulation coefficient of microplastics was calculated and the comprehensive exposure index was determined by combining pre-collected maternal blood and urine indicators. Based on tissue and organ-specific analysis methods, a distribution characteristic map of microplastics was constructed. A scoring matrix was established by calculating the penetration rate and residual time of microplastics in various tissues and organs, and a developmental impact score was calculated.

[0009] The comprehensive exposure index and developmental impact score are input into the fuzzy reasoning module. The fuzzy rule base is constructed and the fuzzy risk value is calculated based on the spatiotemporal characteristics of the critical period of embryonic development. The fuzzy risk value is subjected to time-frequency analysis, and the characteristic coefficients of different critical periods of development are extracted and combined with support vector regression to predict the fetal development risk score.

[0010] In an optional embodiment,

[0011] Obtain the historical medical examination data of the pregnant woman to be tested and perform standardized preprocessing according to the time series to generate a structured time series feature matrix including:

[0012] Obtaining historical medical examination data of the pregnant woman to be tested, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data;

[0013] Sequence alignment of the historical medical examination data is performed according to a unified time standard, the original sampling interval of the periodic examination data is maintained, the temporary examination data collected irregularly is aligned to the most recent examination time point based on the principle of proximity, and the missing test data is supplemented by linear interpolation;

[0014] The aligned historical medical examination data are preprocessed with standardization, the numerical features are normalized to their maximum and minimum values, the categorical features are converted to one-hot encoding, and the processed data are integrated according to the time dimension, feature dimension, and sample dimension to generate a structured time series feature matrix.

[0015] In an optional embodiment,

[0016] A multidimensional characterization index is established based on the types and particle size distribution of microplastics. The microplastic detection data in the structured time series feature matrix are classified and statistically analyzed to calculate the time series cumulative concentration and short-term fluctuation amplitude of microplastics. Combined with the pre-collected maternal blood and urine indicators, the bioaccumulation coefficient of microplastics is calculated and the comprehensive exposure index is determined, including:

[0017] By analyzing the types and particle size distribution of microplastics, we can obtain particle size distribution parameters, chemical composition parameters, and morphological parameters, and construct a multidimensional characterization index for microplastics.

[0018] Obtain microplastic detection data in the structured time series feature matrix and perform classification statistics, construct a time decay weight matrix, and calculate the corresponding time series cumulative concentration and short-term fluctuation amplitude of microplastics;

[0019] Collect maternal blood and urine indicators, and correct the ratio of maternal blood and urine indicators using a correction coefficient matrix to obtain a bioconcentration coefficient;

[0020] The multidimensional characterization indicators, time-series cumulative concentration, short-term fluctuation amplitude and bioconcentration coefficient are combined to form a joint feature vector and added to the pre-set Bayesian deep belief network. The probabilistic dependency relationship between the dimensions of the joint feature vector is established through variational inference.

[0021] Based on the probabilistic dependency relationship, the Fisher information matrix is ​​constructed to evaluate the importance of features and assign weight coefficients. The expectation maximization algorithm is executed to optimize the parameters of the Bayesian deep belief network. The risk probability density function is generated based on the updated Bayesian deep belief network. The expected value and variance of the risk score are calculated. The comprehensive exposure index and its confidence interval are determined by combining the confidence coefficient obtained by uncertainty estimation.

[0022] In an optional embodiment,

[0023] Based on the probability dependency, the Fisher information matrix is ​​constructed to evaluate the feature importance and assign weight coefficients. The expectation maximization algorithm is executed to optimize the parameters of the Bayesian deep belief network. The risk probability density function is generated based on the updated Bayesian deep belief network, including:

[0024] Calculating the second-order partial derivative of the log-likelihood function according to the probability dependency relationship, constructing a Fisher information matrix, and performing weighted summation on each element of the Fisher information matrix and the corresponding dependency strength weight to obtain a feature importance score;

[0025] Inputting the feature importance score into an exponential function adjusted by a temperature parameter for normalization to obtain an initial weight vector, and iteratively updating the initial weight vector based on a learning rate parameter to obtain a weight coefficient;

[0026] Constructing a posterior distribution using the weight coefficients and substituting the posterior distribution into an expectation-maximization algorithm, performing an expectation step by calculating the posterior expectation, performing a maximization step in combination with a regularization term, and iteratively optimizing the parameters of the Bayesian deep belief network;

[0027] The hidden layer output of the optimized Bayesian deep belief network is combined with the pre-acquired observation features to construct a conditional probability model. Based on the conditional probability model, the joint distribution of the hidden layer output and the observation features is calculated and the joint distribution is marginalized and integrated to obtain the risk probability density function.

[0028] In an optional embodiment,

[0029] Based on tissue and organ-specific analysis methods, a distribution characteristic map of microplastics was constructed. A scoring matrix was established by calculating the penetration rate and residual time of microplastics in various tissues and organs. The developmental impact scores were calculated as follows:

[0030] Based on the tissue-organ specific analysis method, the specific response characteristics of different tissues and organs to microplastics are obtained, the organ importance weights are determined according to the specific response characteristics to construct an organ weight coefficient matrix, and the spatial distribution vector is constructed based on the organ weight coefficient matrix and the pre-collected organ microplastic concentration, organ volume and organ tissue density;

[0031] The time-varying penetration coefficient of microplastics in tissues and organs is calculated based on the specific response characteristics of the tissues and organs. The time-varying penetration coefficient is the product of the ratio of the organ microplastic concentration to the exposure concentration at the last detection and the exponential function of the organ decay time. The time-varying penetration coefficient is integrated to obtain the cumulative penetration index;

[0032] Based on the specific metabolic mechanism of tissues and organs for microplastics, the concentration-time curve of microplastics in tissues and organs is obtained and fitted through a double exponential, the rapid clearance ratio, rapid clearance rate and slow clearance rate of microplastics by tissues and organs are obtained, and the double exponential function is substituted to construct a residual time distribution function, and the product of the residual time distribution function and time is infinitely integrated to obtain the average residual time;

[0033] The normalized inner product of the cumulative penetration index, the average residual time, and the spatial distribution vector with the organ weight coefficient matrix is ​​multiplied by the corresponding weight coefficient to obtain a scoring matrix, the scoring matrix is ​​normalized to obtain a normalized impact score, and the normalized impact score and the preset developmental stage sensitivity coefficient are double-summed to obtain a developmental impact degree score.

[0034] In an optional embodiment,

[0035] The comprehensive exposure index and developmental impact score are input into the fuzzy reasoning module. The fuzzy rule base is constructed and the fuzzy risk value is calculated based on the spatiotemporal characteristics of the critical period of embryonic development. The fuzzy risk value is subjected to time-frequency analysis, and the characteristic coefficients of different critical periods of development are extracted and used in combination with support vector regression to predict the fetal development risk score.

[0036] The comprehensive exposure index and the developmental impact score are added as input variables to a pre-set fuzzy reasoning module, and Gaussian membership functions are established for fuzzy processing. The Gaussian membership function includes five linguistic variables: low, medium-low, medium, medium-high, and high. The central value and standard deviation of the Gaussian membership function are determined based on expert experience and historical data to obtain the membership values ​​of the comprehensive exposure index and the developmental impact score;

[0037] Based on the membership value, a fuzzy rule base is constructed in combination with the spatiotemporal characteristics corresponding to the critical period of embryonic development. The fuzzy rules in the fuzzy rule base include an input fuzzy set of the comprehensive exposure index, an input fuzzy set of the developmental impact score, and an output fuzzy set. The fuzzy rules are fuzzy inferenced using the Mandani reasoning method, and are clarified using the center of gravity method to obtain a fuzzy risk value representing the risk status of fetal development. The critical period of embryonic development includes the organ formation period, the functional differentiation period, and the system maturation period.

[0038] Performing a continuous wavelet transform on the fuzzy risk value, calculating the wavelet energy spectrum at different scales through scale parameters and translation parameters, and extracting characteristic coefficients corresponding to each critical period of embryonic development based on the wavelet energy spectrum, wherein the characteristic coefficients include the energy value, variance value and entropy value of the wavelet energy spectrum;

[0039] The characteristic coefficients are combined to construct a characteristic vector, a support vector regression algorithm is used to establish a risk prediction model, a radial basis kernel function is selected and the penalty factor, kernel function parameters and fault tolerance parameters of the radial basis kernel function are optimized through cross-validation to obtain an optimized risk prediction model, the characteristic vector is input into the optimized risk prediction model to obtain a fetal development risk score.

[0040] A second aspect of an embodiment of the present invention provides a fetal development assessment system for maternal microplastic exposure, which does not involve the diagnosis and treatment of diseases and includes:

[0041] The first unit is used to obtain historical medical examination data of the pregnant woman to be tested and perform standardized preprocessing according to the time series to generate a structured time series feature matrix, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data;

[0042] The second unit is used to establish multidimensional characterization indicators based on the types and particle size distribution of microplastics, classify and count the microplastic detection data in the structured time series feature matrix, calculate the time series cumulative concentration and short-term fluctuation amplitude of microplastics, combine pre-collected maternal blood and urine indicators to calculate the bioaccumulation coefficient of microplastics and determine the comprehensive exposure index, construct a distribution characteristic map of microplastics based on tissue and organ specific analysis methods, establish a scoring matrix by calculating the penetration rate and residual time of microplastics in various tissues and organs, and calculate the developmental impact score;

[0043] The third unit is used to input the comprehensive exposure index and the developmental impact score into the fuzzy reasoning module, build a fuzzy rule base and calculate the fuzzy risk value based on the spatiotemporal characteristics of the critical period of embryonic development, perform time-frequency analysis on the fuzzy risk value, extract the characteristic coefficients of different critical development periods, and use support vector regression to predict the fetal development risk score.

[0044] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0045] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0046] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0047] In the present invention, by obtaining historical medical examination data of pregnant women and performing standardized preprocessing, a structured time series feature matrix is ​​established, and combined with the multidimensional characterization indicators and bioconcentration coefficients of microplastics, a microplastic distribution feature map is constructed, which realizes the accurate quantification and evaluation of the degree of microplastic exposure, improves the comprehensiveness and accuracy of the evaluation, and constructs a scoring matrix based on the tissue and organ specific analysis method to calculate the developmental impact score. In combination with fuzzy reasoning technology and the spatiotemporal characteristics of the critical period of embryonic development, a dynamic assessment of fetal developmental risk is realized, which can capture the specific risk changes in different developmental stages and improve the scientific nature and pertinence of risk assessment. The fuzzy risk value is processed by time-frequency analysis and support vector regression method, and the characteristic coefficients of different critical developmental periods are extracted to generate an evaluation report containing abnormal feature analysis and intervention recommendations, which provides a scientific basis for clinical intervention and prevention and helps to reduce the adverse effects of microplastic exposure on fetal development. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a method for assessing fetal development following maternal microplastic exposure according to an embodiment of the present invention;

[0049] Figure 2 This is a comparison diagram of the risk probability density function of the fetal development assessment method for maternal microplastic exposure according to an embodiment of the present invention;

[0050] Figure 3 A heat map of the average residual time of microplastics in tissues and organs for a method for assessing fetal development following maternal microplastic exposure according to an embodiment of the present invention;

[0051] Figure 4 This is a comparison chart of the risk assessment performance of different developmental key periods of the fetal development assessment method for maternal microplastic exposure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0053] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0054] Figure 1 This is a flow chart of a method for assessing fetal development in maternal microplastic exposure according to an embodiment of the present invention. This method does not involve the diagnosis and treatment of diseases, such as Figure 1 As shown, the method includes:

[0055] Obtaining historical medical examination data of the pregnant woman to be tested and performing standardized preprocessing according to time series to generate a structured time series feature matrix, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data;

[0056] A multidimensional characterization index was established based on the types and particle size distribution of microplastics. The microplastic detection data in the structured time series feature matrix was classified and statistically analyzed to calculate the time series cumulative concentration and short-term fluctuation amplitude of microplastics. The bioaccumulation coefficient of microplastics was calculated and the comprehensive exposure index was determined by combining pre-collected maternal blood and urine indicators. Based on tissue and organ-specific analysis methods, a distribution characteristic map of microplastics was constructed. A scoring matrix was established by calculating the penetration rate and residual time of microplastics in various tissues and organs, and a developmental impact score was calculated.

[0057] The comprehensive exposure index and developmental impact score are input into the fuzzy reasoning module. The fuzzy rule base is constructed and the fuzzy risk value is calculated based on the spatiotemporal characteristics of the critical period of embryonic development. The fuzzy risk value is subjected to time-frequency analysis, and the characteristic coefficients of different critical periods of development are extracted and combined with support vector regression to predict the fetal development risk score.

[0058] This technical solution is essentially a prediction system based on data analysis and risk assessment. It collects and processes existing medical examination data, applies mathematical models and computer algorithms to conduct data mining and statistical analysis, and ultimately outputs risk assessment results. This technical solution does not directly perform any diagnostic or treatment operations on the human body, nor does it involve directly determining or eliminating the cause of the disease. Instead, it predicts possible risks and provides reference suggestions through analysis of historical data. This technical solution processes existing examination data such as microplastic detection data, fetal development indicators, and maternal health status, and uses mathematical calculation methods such as standardization, feature extraction, and fuzzy reasoning to construct a risk assessment model. It is information processing at the data level and will not cause any direct intervention or therapeutic effect on living organisms. Therefore, this technical solution is a technical solution for data analysis and risk warning, rather than a medical diagnosis and treatment method.

[0059] In an optional embodiment,

[0060] Obtain the historical medical examination data of the pregnant woman to be tested and perform standardized preprocessing according to the time series to generate a structured time series feature matrix including:

[0061] Obtaining historical medical examination data of the pregnant woman to be tested, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data;

[0062] Sequence alignment of the historical medical examination data is performed according to a unified time standard, the original sampling interval of the periodic examination data is maintained, the temporary examination data collected irregularly is aligned to the most recent examination time point based on the principle of proximity, and the missing test data is supplemented by linear interpolation;

[0063] The aligned historical medical examination data are preprocessed with standardization, the numerical features are normalized to their maximum and minimum values, the categorical features are converted to one-hot encoding, and the processed data are integrated according to the time dimension, feature dimension, and sample dimension to generate a structured time series feature matrix.

[0064] Historical medical examination data of the pregnant women being tested was extracted from the medical institution's electronic health record system. Microplastic detection data included microplastic concentrations in blood (in ng / mL), urine (in ng / mL), and amniotic fluid (in ng / mL). Fetal developmental indicators included fetal length (in mm), fetal weight (in g), fetal head circumference (in mm), fetal abdominal circumference (in mm), fetal heart rate (in beats / minute), and fetal movement frequency (in beats / hour). Maternal health data included blood pressure (in mmHg, divided into systolic and diastolic pressures), blood glucose level (in mmol / L), weight (in kg), hemoglobin content (in g / L), and thyroid function indicators (in mIU / L). Data was automatically retrieved from the hospital's information system via an API or extracted from paper examination reports using scanning and optical character recognition technology and stored in a local database.

[0065] In the time series alignment step, historical medical examination data from different sources and time points are aligned according to a unified time standard. Regular examination data for pregnant women are identified, typically collected at fixed gestational weeks (such as 12, 16, 20, 24, 28, 32, 36, and 38). For each pregnant woman, a timeline indexed by gestational week is created, starting at 4 weeks of pregnancy and continuing each week until delivery. The original sampling interval of periodic examination data is maintained, that is, regular prenatal examination data is kept at the original gestational week of collection. For ad hoc examination data conducted at unusual gestational weeks, they are aligned to the nearest examination time point using the principle of proximity. For example, if a pregnant woman had an ad hoc examination at 18 weeks and 3 days, these data are aligned to the 18th week time point. Missing examination data at certain time points are supplemented using linear interpolation. Specifically, if a pregnant woman's blood microplastic concentration is 5.2 ng / mL and 7.8 ng / mL at weeks 20 and 28, respectively, but this data is missing at week 24, the estimated value at week 24 is calculated by linear interpolation to be 6.5 ng / mL (calculated as 5.2 + (7.8 - 5.2) × (24 - 20) / (28 - 20) = 6.5). For long-term missing data, a maximum interpolation interval is set (e.g., 8 weeks). Missing data beyond this interval are not interpolated but marked as missing.

[0066] During the normalization preprocessing step, the aligned historical medical examination data was normalized to eliminate dimensional differences between different indicators. For numerical features (such as blood microplastic concentration, fetal weight, and maternal blood pressure), a minimum-maximum normalization method was used to map all values ​​to the interval [0,1]. For example, assuming that the minimum value of blood microplastic concentration is 0.5 ng / mL and the maximum value is 12.3 ng / mL across all historical data, then for a sample with a concentration of 5.2 ng / mL, the normalized value is (5.2 - 0.5) / (12.3 - 0.5) = 0.398. For categorical features (such as fetal position and amniotic fluid status), a one-hot encoding conversion method was used. For example, fetal position can be "cephalic," "breech," or "transverse." These are converted into three-dimensional vectors, with cephalic position represented as [1, 0, 0], breech as [0, 1, 0], and transverse as [0, 0, 1]. After standardization, all processed data are integrated according to the time, feature, and sample dimensions to generate a structured time series feature matrix. The dimensions of this matrix are [number of samples × number of time points × number of features]. For example, if data from a pregnant woman is tracked from week 12 to week 40 (a total of 29 time points), and each time point has 30 features (including microplastic detection data, fetal development indicators, and maternal health data), the dimensions of the structured time series feature matrix for this pregnant woman are [1 × 29 × 30].

[0067] In this embodiment, a unified time standard is used for sequence alignment to solve the problem that medical data collected at different frequencies are difficult to directly compare. Missing data are supplemented by linear interpolation method, which effectively improves the integrity of the data and provides a reliable basis for subsequent analysis. The maximum and minimum values ​​of numerical features are normalized to eliminate the dimensional differences between different indicators. The categorical features are converted into one-hot encoding to achieve quantitative expression of qualitative data, which significantly improves the comparability between different types of medical data. The processed data are integrated according to the time dimension, feature dimension and sample dimension to generate a structured time series feature matrix, so that microplastic detection data, fetal development indicator data and maternal health status data form a unified data representation in the time series, providing a multi-dimensional perspective for the subsequent analysis of the relationship between microplastic exposure and fetal development.

[0068] In an optional embodiment,

[0069] A multidimensional characterization index is established based on the types and particle size distribution of microplastics. The microplastic detection data in the structured time series feature matrix are classified and statistically analyzed to calculate the time series cumulative concentration and short-term fluctuation amplitude of microplastics. Combined with the pre-collected maternal blood and urine indicators, the bioaccumulation coefficient of microplastics is calculated and the comprehensive exposure index is determined, including:

[0070] By analyzing the types and particle size distribution of microplastics, we can obtain particle size distribution parameters, chemical composition parameters, and morphological parameters, and construct a multidimensional characterization index for microplastics.

[0071] Obtain microplastic detection data in the structured time series feature matrix and perform classification statistics, construct a time decay weight matrix, and calculate the corresponding time series cumulative concentration and short-term fluctuation amplitude of microplastics;

[0072] Collect maternal blood and urine indicators, and correct the ratio of maternal blood and urine indicators using a correction coefficient matrix to obtain a bioconcentration coefficient;

[0073] The multidimensional characterization indicators, time-series cumulative concentration, short-term fluctuation amplitude and bioconcentration coefficient are combined to form a joint feature vector and added to the pre-set Bayesian deep belief network. The probabilistic dependency relationship between the dimensions of the joint feature vector is established through variational inference.

[0074] Based on the probabilistic dependency relationship, the Fisher information matrix is ​​constructed to evaluate the importance of features and assign weight coefficients. The expectation maximization algorithm is executed to optimize the parameters of the Bayesian deep belief network. The risk probability density function is generated based on the updated Bayesian deep belief network. The expected value and variance of the risk score are calculated. The comprehensive exposure index and its confidence interval are determined by combining the confidence coefficient obtained by uncertainty estimation.

[0075] A multidimensional characterization index was established based on the types and size distribution of microplastics. Raman and infrared spectroscopy were used to determine the primary chemical composition of microplastics in environmental samples. For the 100 aquatic samples collected, the system identified five main types of microplastics: polyethylene (PE), polypropylene (PP), polystyrene (PS), polyethylene terephthalate (PET), and polyvinyl chloride (PVC), assigning them values ​​from 1 to 5. Microplastic particle size was measured using a laser particle size analyzer, and the particle size range was divided into four intervals: less than 20 μm, 20-50 μm, 50-100 μm, and greater than 100 μm, designated as size intervals 1 to 4. Morphological analysis used a digital microscope to observe the microplastic morphology and categorize it into fragments, fibers, films, foams, and spheres, corresponding to morphological intervals 1 to 5. These three sets of parameters constitute the multidimensional characterization index for microplastics. For example, a fibrous microplastic made of PE with a particle size of 35 μm could be represented as [1, 2, 2].

[0076] When classifying and statistically analyzing the microplastic detection data in the structured time-series feature matrix, data collected over 30 consecutive days were classified according to the aforementioned multidimensional characterization indicators, forming a time-series matrix T. To reflect the accumulation and metabolic effects of microplastics in the body, a time-attenuation weight matrix W was constructed, in which the weight of the most recent day was 1.0, and the weight decreased by 0.05 each day until the weight of the 20th day was 0.05. The weight of earlier data was 0.05. By multiplying the detection concentration matrix with the attenuation weight matrix, the time-series cumulative concentration of PE microplastics was calculated to be 0.83 ng / mL, PP to 0.62 ng / mL, PS to 0.45 ng / mL, PET to 0.37 ng / mL, and PVC to 0.29 ng / mL. The short-term fluctuation amplitude was obtained by calculating the standard deviation of the detection data over the past 7 days: 0.15 ng / mL for PE, 0.12 ng / mL for PP, 0.09 ng / mL for PS, 0.08 ng / mL for PET, and 0.06 ng / mL for PVC.

[0077] Blood and urine samples were collected from 50 participants for index measurement. Blood indicators include red blood cell count, white blood cell count, hemoglobin, four liver function tests and three kidney function tests. Urine indicators include urine protein, urine sugar, urobilinogen and uric acid. Based on existing studies, the system established a correction coefficient matrix C, which takes into account the impact of factors such as age and weight on the measured values. For example, for a 35-year-old participant, the concentration of PE microplastics in the blood was 0.75 ng / mL and in the urine was 0.12 ng / mL. After the system applied a correction coefficient of 1.2, the bioconcentration factor of PE microplastics was calculated to be 7.5. Similarly, the bioconcentration factors of PP, PS, PET and PVC were calculated to be 6.8, 5.9 and 5.2, respectively.

[0078] The above multidimensional characterization indicators, time-series cumulative concentration, short-term fluctuation amplitude, and bioconcentration coefficient are combined to form a 25-dimensional joint feature vector V, which is then input into a pre-configured Bayesian deep belief network. This network consists of an input layer, three hidden layers, and an output layer, with 32, 16, and 8 hidden layer nodes, respectively. Using variational inference methods, probabilistic dependencies between the dimensions of the joint feature vector are established. For example, the correlation between the cumulative concentration of PE microplastics and their bioconcentration coefficient is found to be 0.78, while the correlation with their short-term fluctuation amplitude is 0.65.

[0079] Based on the above probabilistic dependencies, a Fisher information matrix I was constructed with a size of 25×25, with diagonal elements representing the importance of each feature. Analysis showed that the indicator importances for PE and PP microplastics were 0.25 and 0.22, respectively, significantly higher than for other microplastic types. The characteristic importance for microplastics with a particle size less than 20 μm was 0.30, higher than for other particle size ranges. Weighting coefficients were assigned: microplastic type was weighted at 0.35, particle size distribution was weighted at 0.25, cumulative concentration over time was weighted at 0.20, short-term fluctuation amplitude was weighted at 0.10, and bioconcentration factor was weighted at 0.10.

[0080] High risk. Monte Carlo integration calculated the expected value of the risk score to be E = 6.7, with a variance of Var = 1.2. Further calculation of the uncertainty estimate at a 95% confidence level yielded a confidence coefficient of α = 1.96. The overall exposure index was determined to be 6.7, with a 95% confidence interval of [4.5, 8.9], indicating that the participant faced a moderate to high risk of microplastic exposure.

[0081] In this embodiment, by analyzing the particle size distribution parameters, chemical composition parameters and morphological parameters of microplastics, a multidimensional characterization index was constructed, and differentiated characterization of different types of microplastics was achieved, overcoming the limitation of traditional methods that only focus on concentration and ignore the differences in their physical and chemical properties. The time-attenuated weight matrix was introduced to achieve accurate calculation of the temporal cumulative effects and short-term fluctuations of microplastic exposure, solving the technical problem that traditional methods cannot effectively characterize the differences between long-term low-dose and short-term high-dose exposure. The ratio of maternal blood and urine indicators was corrected by the correction coefficient matrix to obtain a more accurate bioconcentration coefficient, which effectively solved the exposure assessment bias caused by the differences in the accumulation of different microplastics in the organism. The Bayesian deep belief network and variational inference technology were used to establish a probabilistic dependency relationship between the dimensions of the joint eigenvector, breaking through the limitation that traditional linear models cannot capture complex nonlinear relationships, and improving the model's adaptability to complex microplastic exposure scenarios.

[0082] In an optional embodiment,

[0083] Based on the probability dependency, the Fisher information matrix is ​​constructed to evaluate the feature importance and assign weight coefficients. The expectation maximization algorithm is executed to optimize the parameters of the Bayesian deep belief network. The risk probability density function is generated based on the updated Bayesian deep belief network, including:

[0084] Calculating the second-order partial derivative of the log-likelihood function according to the probability dependency relationship, constructing a Fisher information matrix, and performing weighted summation on each element of the Fisher information matrix and the corresponding dependency strength weight to obtain a feature importance score;

[0085] Inputting the feature importance score into an exponential function adjusted by a temperature parameter for normalization to obtain an initial weight vector, and iteratively updating the initial weight vector based on a learning rate parameter to obtain a weight coefficient;

[0086] Constructing a posterior distribution using the weight coefficients and substituting the posterior distribution into an expectation-maximization algorithm, performing an expectation step by calculating the posterior expectation, performing a maximization step in combination with a regularization term, and iteratively optimizing the parameters of the Bayesian deep belief network;

[0087] The hidden layer output of the optimized Bayesian deep belief network is combined with the pre-acquired observation features to construct a conditional probability model. Based on the conditional probability model, the joint distribution of the hidden layer output and the observation features is calculated and the joint distribution is marginalized and integrated to obtain the risk probability density function.

[0088] Based on the probabilistic dependencies between the dimensions of the obtained joint feature vector, the second-order partial derivative of the log-likelihood function was calculated. For a sample of 500 pregnant women, the second-order partial derivative of the log-likelihood function for the two features of polyethylene microplastics (average particle size of 0.8 μm) and their cumulative concentration in maternal blood (average of 0.015 mg / L) was calculated to be 0.376. Similarly, for polystyrene microplastics (average particle size of 0.5 μm) and their bioconcentration coefficient (BCF) (average of 2.8), the second-order partial derivative was calculated to be 0.591. This calculation was performed for all 36 pairs of feature combinations, forming a 6×6 Fisher information matrix.

[0089] Next, each element of the Fisher information matrix was weighted with its corresponding dependency strength weight. The dependency strength weight for small-size (less than 1 micron) microplastics was set to 1.5, the dependency strength weight for medium-size (1-5 microns) microplastics was set to 1.2, and the dependency strength weight for large-size (greater than 5 microns) microplastics was set to 0.8. These weights were multiplied element-wise by the Fisher information matrix and summed across each row to obtain the importance scores for the six features. The proportion of small-size polystyrene microplastics had the highest score, at 2.865, followed by the bioconcentration coefficient of polyethylene microplastics, at 2.327. The short-term fluctuation amplitude of total microplastics received a score of 2.102.

[0090] After obtaining the feature importance score, it is input into the exponential function regulated by the temperature parameter for processing. In the present embodiment, the temperature parameter is set to 0.75. The importance score of each feature is divided by the temperature parameter and the index value is calculated. For example, the small particle size ratio of polystyrene microplastics is calculated as exp(2.865 / 0.75)=44.12. All index values ​​are normalized to ensure that the sum is 1 to obtain an initial weight vector. After normalization, the initial weight of the small particle size ratio of polystyrene microplastics is 0.387, the initial weight of the bioconcentration coefficient of polyethylene microplastics is 0.252, and the initial weight of the short-term fluctuation amplitude of total microplastics is 0.193.

[0091] The initial weight vector is iteratively updated based on the learning rate parameter. The initial value of the learning rate is set to 0.05 and decays at a rate of 0.99 after each iteration. In the first round of iteration, the direction and amplitude of the weight adjustment are calculated based on the difference between the risk value predicted by the current weight and the actual fetal development indicators. For example, the weight adjustment value of the small particle size ratio of polystyrene microplastics is -0.023, and the updated weight becomes 0.364. After 50 rounds of iteration, the weight coefficient tends to stabilize. The weight of the small particle size ratio of polystyrene microplastics is 0.342, the bioconcentration coefficient weight of polyethylene microplastics is 0.278, and the short-term fluctuation amplitude weight of total microplastics is 0.215.

[0092] The posterior distribution is constructed using the optimized weight coefficients. For example, a three-layer Bayesian deep belief network (DBN) has six nodes in the input layer corresponding to six features, 12 nodes in the first hidden layer, and eight nodes in the second hidden layer. The posterior distribution is fed into the expectation-maximization algorithm to begin network parameter optimization. In the expectation step, the current network parameters are fixed and the posterior expected values ​​of the hidden layer nodes are calculated. For example, in the first iteration, the posterior expected value of the third node in the first hidden layer is 0.723.

[0093] After the expectation calculation is completed, the expected values ​​of the hidden layer nodes obtained in the previous step are fixed, and the network parameters are updated to maximize the log-likelihood function. A parameter regularization term with a regularization coefficient of 0.01 is introduced to constrain the parameter size. After the first round of parameter updates, the weight connecting the second node in the input layer and the fourth node in the first hidden layer is updated from the initial value of 0.35 to 0.382. These two steps are repeated alternately for a total of 100 rounds, and the network parameters converge to obtain the optimized Bayesian deep belief network.

[0094] The hidden layer output information was extracted from the optimized network. For example, a test sample contained 78% small-size polystyrene microplastics, a polyethylene microplastic bioaccumulation factor of 3.2, and a short-term fluctuation of 0.27 mg / L in total microplastics. These characteristics were input into the optimized network, resulting in the following output values ​​for the eight nodes in the second hidden layer: [0.825, 0.423, 0.651, 0.278, 0.592, 0.137, 0.502, 0.764].

[0095] These hidden layer outputs were combined with the original observational features to construct a conditional probability model. This model uses a mixed Gaussian distribution with three Gaussian components, corresponding to low, medium, and high risk states. The model parameters were determined through maximum likelihood estimation, yielding means of 0.25, 0.55, and 0.85, standard deviations of 0.08, 0.12, and 0.10, and mixing weights of 0.35, 0.40, and 0.25, respectively.

[0096] Based on the conditional probability model, the joint distribution of the hidden layer output and the observed features is calculated. Substituting the features and hidden layer output of the above test sample into the model, the probabilities of the three risk states are calculated: low risk 0.15, medium risk 0.30, and high risk 0.55. By marginalizing and integrating the joint distribution, the risk probability density function is obtained. This function has a probability density of 0.32 near a risk value of 0.2, a probability density of 0.58 near a risk value of 0.5, and a probability density of 1.25 near a risk value of 0.8. This function intuitively shows the risk distribution of the sample, with the probability density of the high-risk state being significantly higher than that of the other states.

[0097] In this embodiment, information theory tools are introduced to evaluate feature importance, breaking through the limitations of linear analysis methods. A feature weight distribution mechanism adjusted by temperature parameters is adopted to enhance the model's adaptability. The expectation maximization algorithm and regularization technology are combined to optimize network parameters, improving the model's stability and generalization ability. The risk probability density function is constructed through a conditional probability model and marginalized integration to achieve a complete probabilistic expression of risk assessment.

[0098] Existing technologies mainly rely on simple linear models or traditional machine learning methods to assess microplastic exposure risks. These methods cannot effectively capture the complex nonlinear dependencies between features, resulting in important features being ignored or minor features being overemphasized. Gradient descent-based methods are often used, which have difficulty handling multimodal distribution problems and are prone to falling into local optimal solutions, affecting model stability and generalization capabilities. The output results lack interpretability and cannot provide a transparent and reliable basis for risk quantification for medical decision-making.

[0099] This embodiment constructs the Fisher information matrix by calculating the second-order partial derivative of the log-likelihood function, which can fully capture the nonlinear dependencies between features, overcome the limitation of traditional linear correlation analysis that cannot identify complex feature interactions, and achieve accurate quantification of feature importance. It uses an exponential function adjusted by temperature parameters for normalization, and dynamically adjusts the weight distribution according to the information contribution of the feature, solving the problem of insufficient adaptability of the traditional fixed weight method and enhancing the model's sensitivity to different microplastic exposure patterns. By combining the expectation maximization algorithm with the regularization technology, it ensures convergence and avoids overfitting when performing parameter optimization. Compared with the traditional gradient descent method, it greatly improves the stability and generalization ability of the model under heterogeneous data conditions.

[0100] Figure 2 This is a comparison chart of the risk probability density functions for fetal development assessment methods for maternal microplastic exposure, based on an embodiment of the present invention. It shows a comparison of the risk probability density functions calculated using three different methods. The horizontal axis represents the risk value (ranging from 0 to 1), and the vertical axis represents the probability density.

[0101] This technical solution (represented by the solid line) constructs a conditional probability model through the combination of the hidden layer output of the Bayesian deep belief network and the observation features. The calculated risk distribution has a clear three-peak characteristic, with peak values ​​of probability density of 0.32, 0.58 and 1.25 formed at risk values ​​of 0.2, 0.5 and 0.8, respectively. Among them, the peak in the high-risk area (near 0.8) is the most significant. Although the naive Bayes method (represented by the dotted line) also presents a multi-peak distribution, the peak in the high-risk area is only 0.87, which cannot effectively distinguish between medium and high-risk states; the linear regression model (represented by the dotted line) is completely unable to capture the multi-peak characteristics of the risk distribution, and presents a unimodal distribution that is approximately normal, with the weakest risk identification ability.

[0102] This technical solution constructs a mixed Gaussian distribution through three Gaussian components (with means of 0.25, 0.55 and 0.85, standard deviations of 0.08, 0.12 and 0.10, and mixing weights of 0.35, 0.40 and 0.25, respectively). It can not only accurately characterize the multimodal characteristics of risk, but also provide doctors with an intuitive risk distribution map by calculating the probability density function obtained by marginalized integral. In particular, its sensitive identification ability for high-risk states (probability density 1.25) is significantly higher than other methods, providing a reliable basis for clinical intervention decisions.

[0103] In an optional embodiment,

[0104] Based on tissue and organ-specific analysis methods, a distribution characteristic map of microplastics was constructed. A scoring matrix was established by calculating the penetration rate and residual time of microplastics in various tissues and organs. The developmental impact scores were calculated as follows:

[0105] Based on the tissue-organ specific analysis method, the specific response characteristics of different tissues and organs to microplastics are obtained, the organ importance weights are determined according to the specific response characteristics to construct an organ weight coefficient matrix, and the spatial distribution vector is constructed based on the organ weight coefficient matrix and the pre-collected organ microplastic concentration, organ volume and organ tissue density;

[0106] The time-varying penetration coefficient of microplastics in tissues and organs is calculated based on the specific response characteristics of the tissues and organs. The time-varying penetration coefficient is the product of the ratio of the organ microplastic concentration to the exposure concentration at the last detection and the exponential function of the organ decay time. The time-varying penetration coefficient is integrated to obtain the cumulative penetration index;

[0107] Based on the specific metabolic mechanism of tissues and organs for microplastics, the concentration-time curve of microplastics in tissues and organs is obtained and fitted through a double exponential, the rapid clearance ratio, rapid clearance rate and slow clearance rate of microplastics by tissues and organs are obtained, and the double exponential function is substituted to construct a residual time distribution function, and the product of the residual time distribution function and time is infinitely integrated to obtain the average residual time;

[0108] The normalized inner product of the cumulative penetration index, the average residual time, and the spatial distribution vector with the organ weight coefficient matrix is ​​multiplied by the corresponding weight coefficient to obtain a scoring matrix, the scoring matrix is ​​normalized to obtain a normalized impact score, and the normalized impact score and the preset developmental stage sensitivity coefficient are double-summed to obtain a developmental impact degree score.

[0109] When obtaining tissue and organ-specific response characteristics, at least five typical tissues and organs were selected for microplastic response analysis, including liver, kidney, brain tissue, heart, and gonads. High-sensitivity mass spectrometry technology was used to detect changes in microplastic concentrations in various tissues and organs, and combined with tissue pathological analysis, the sensitivity and response intensity of each tissue and organ to microplastics were determined. For example, in one set of experiments, the response coefficient of the liver to polystyrene microspheres was 0.82, the kidney was 0.65, the brain tissue was 0.91, the heart was 0.43, and the gonads was 0.88. Based on these response characteristic data, an organ weight coefficient matrix can be established, in which the weights of the liver are 0.25, the kidney is 0.20, the brain tissue is 0.30, the heart is 0.10, and the gonads are 0.15.

[0110] The construction of the spatial distribution vector requires the combination of organ microplastic concentration, organ volume and tissue density data. For example, the concentration of polyethylene microplastics in the liver was detected to be 125 ng / g and the organ volume was 1500 cm 3 , tissue density is 1.05g / cm 3The concentration of polyethylene microplastics in the kidney was 85 ng / g, and the organ volume was 280 cm 3 , tissue density is 1.03 g / cm 3 The concentration of polyethylene microplastics in brain tissue was 62 ng / g, and the organ volume was 1350 cm 3 , tissue density is 1.04 g / cm 3 The concentration of polyethylene microplastics in the heart was 45 ng / g, and the organ volume was 320 cm 3 , tissue density is 1.06 g / cm 3 The concentration of polyethylene microplastics in the gonads was 103 ng / g, and the organ volume was 65 cm 3 , tissue density is 1.04 g / cm 3 By multiplying the microplastic concentration by the organ volume and tissue density, the total amount of microplastics in each organ was obtained, and combined with the organ weight coefficient matrix to construct a complete spatial distribution vector.

[0111] The calculation of the time-varying penetration coefficient is based on continuous monitoring data. In the experiment, model organisms were treated with microplastic solutions at different exposure concentrations (e.g., 100 ng / L, 500 ng / L, and 1000 ng / L). The concentrations of microplastics in various tissues and organs were measured at different time points (e.g., 24 hours, 72 hours, 168 hours, and 336 hours). For example, at an exposure concentration of 100 ng / L, the concentration in the liver was 15 ng / g at 24 hours, 38 ng / g at 72 hours, 76 ng / g at 168 hours, and 105 ng / g at 336 hours. To calculate the time-varying penetration coefficient, the ratio of the 72-hour detection value to the 24-hour exposure concentration (38 / 100 = 0.38) was multiplied by the liver's decay time exponential function for this type of microplastic (approximately 0.85), resulting in a time-varying penetration coefficient of 0.323 at that time point. By integrating the time-varying penetration coefficients at different time points, the cumulative penetration index of the liver to this type of microplastics was found to be 0.62.

[0112] The persistence of microplastics in tissues and organs was determined through metabolic clearance experiments. Organisms pre-exposed to microplastics were transferred to a clean environment, and the concentration of microplastics in various tissues and organs was regularly monitored. For example, in brain tissue, starting from an initial concentration of 62 ng / g, the concentration decreased to 45 ng / g on day 1, 32 ng / g on day 3, 25 ng / g on day 7, 19 ng / g on day 14, and 12 ng / g on day 30. Fitting this concentration-time curve with a biexponential function revealed a rapid clearance ratio of 0.35, a rapid clearance rate of 0.28 / day, and a slow clearance rate of 0.05 / day for these microplastics. Substituting the parameters into the biexponential function to construct a retention time distribution function, and integrating the product of this function and time to infinity, the average retention time of microplastics in brain tissue was 13.6 days.

[0113] The scoring matrix was constructed by comprehensively considering the cumulative penetration index, mean residual time, and spatial distribution vector. In practical applications, the cumulative penetration indexes calculated for five tissues and organs were 0.62 (liver), 0.51 (kidney), 0.72 (brain), 0.38 (heart), and 0.66 (gonads). The mean residual times were 8.5 days (liver), 6.2 days (kidney), 13.6 days (brain), 5.8 days (heart), and 10.3 days (gonads). This data was combined with the spatial distribution vector and the organ weight coefficient matrix to generate the initial scoring matrix. This matrix was normalized, resulting in normalized impact scores of 0.75 (liver), 0.58 (kidney), 0.93 (brain), 0.36 (heart), and 0.82 (gonads).

[0114] The calculation of the developmental impact score requires integration of the sensitivity coefficients for each developmental stage. The sensitivity of various tissues and organs to microplastics varies significantly across different stages of fetal development. For example, during organogenesis (gestational weeks 3-8), the sensitivity coefficients for the neural tube and heart primordium are 1.8 and 1.6, respectively; during functional differentiation (gestational weeks 9-20), the sensitivity coefficients for the liver and kidney are 1.5 and 1.3, respectively; and during systemic maturation (gestational weeks 21-40), the sensitivity coefficients for the cerebral cortex and lung are 1.4 and 1.2, respectively. Multiplying the normalized impact score with the sensitivity coefficient for the corresponding fetal developmental stage and summing the results yields impact scores for each developmental stage: 4.35 for organogenesis, 3.72 for functional differentiation, and 3.25 for systemic maturation. Quantitative results indicate that the most significant impact of microplastics on fetal development occurs during the early organogenesis period, followed by functional differentiation and systemic maturation. This is highly consistent with the developmental sensitivity characteristics of key fetal periods.

[0115] In this embodiment, the specific response characteristic analysis and organ weight coefficient matrix construction are used to accurately characterize the spatial distribution of microplastics. The dynamic penetration ability of microplastics is quantified by introducing the time-varying penetration coefficient and cumulative penetration index. The residence time of microplastics in tissues is accurately calculated by double exponential fitting and residual time distribution function analysis. By constructing a multidimensional scoring matrix and combining the developmental stage sensitivity coefficient, the risk differences at different stages of fetal development are accurately quantified.

[0116] Existing technologies typically use single-organ models or holistic exposure models, which fail to reflect the differentiated responses of different organs to microplastics. They ignore the dynamic distribution and temporal accumulation effects of microplastics in organisms, focus only on instantaneous concentrations, and lack quantitative assessment of the penetration ability and residual characteristics of microplastics, resulting in a lack of biological basis for risk assessment results.

[0117] This embodiment successfully captures the temporal evolution of microplastic penetration by considering the exponential function of organ decay time, making the assessment results more consistent with the actual characteristics of biological processes. By distinguishing between rapid clearance and slow clearance processes, the accuracy of residual time estimation is greatly improved, providing a reliable basis for long-term risk assessment. By introducing the developmental stage sensitivity coefficient, the risk assessment results are more consistent with the laws of developmental biology.

[0118] Figure 3 The average residual time of microplastics in tissues and organs of the fetal development assessment method for maternal microplastic exposure is thermally analyzed in an embodiment of the present invention, and the average residual time prediction results of four common microplastics (polyethylene, polystyrene, polyvinyl chloride and polypropylene) in five key tissues and organs are intuitively displayed.

[0119] This technical solution obtains high-precision average residual time data by fitting the concentration-time curve with a biexponential function and calculating the infinite integral. The shade of the color block in the heat map represents the length of the residual time, and the value is accurate to 0.1 day. As can be seen from the data, the average residual time of all types of microplastics in brain tissue is the longest. Polystyrene remains in brain tissue for an average of 15.8 days, which is more consistent with the experimental observation value of 16.3 days than the 11.2 days predicted by the existing single exponential decay model. The gonads are second, with polyethylene remaining in the gonads for an average of 10.3 days, while the traditional tissue distribution model only predicts 7.5 days.

[0120] This technical solution also revealed a correlation between microplastic type and retention time. The retention time of polyvinyl chloride in the liver (9.8 days) was significantly longer than that of polypropylene (6.7 days), a difference not accurately reflected in traditional physicochemical property prediction models. By comprehensively considering both rapid and slow clearance rates, this solution achieved a prediction accuracy of 92.7%, a significant improvement over the existing technologies of 78.3% (single exponential model) and 65.9% (tissue partitioning model), providing a more reliable data foundation for assessing the long-term health risks of microplastics.

[0121] In an optional embodiment,

[0122] The comprehensive exposure index and developmental impact score are input into the fuzzy reasoning module. The fuzzy rule base is constructed and the fuzzy risk value is calculated based on the spatiotemporal characteristics of the critical period of embryonic development. The fuzzy risk value is subjected to time-frequency analysis, and the characteristic coefficients of different critical periods of development are extracted and used in combination with support vector regression to predict the fetal development risk score.

[0123] The comprehensive exposure index and the developmental impact score are added as input variables to a pre-set fuzzy reasoning module, and Gaussian membership functions are established for fuzzy processing. The Gaussian membership function includes five linguistic variables: low, medium-low, medium, medium-high, and high. The central value and standard deviation of the Gaussian membership function are determined based on expert experience and historical data to obtain the membership values ​​of the comprehensive exposure index and the developmental impact score;

[0124] Based on the membership value, a fuzzy rule base is constructed in combination with the spatiotemporal characteristics corresponding to the critical period of embryonic development. The fuzzy rules in the fuzzy rule base include an input fuzzy set of the comprehensive exposure index, an input fuzzy set of the developmental impact score, and an output fuzzy set. The fuzzy rules are fuzzy inferenced using the Mandani reasoning method, and are clarified using the center of gravity method to obtain a fuzzy risk value representing the risk status of fetal development. The critical period of embryonic development includes the organ formation period, the functional differentiation period, and the system maturation period.

[0125] Performing a continuous wavelet transform on the fuzzy risk value, calculating the wavelet energy spectrum at different scales through scale parameters and translation parameters, and extracting characteristic coefficients corresponding to each critical period of embryonic development based on the wavelet energy spectrum, wherein the characteristic coefficients include the energy value, variance value and entropy value of the wavelet energy spectrum;

[0126] The characteristic coefficients are combined to construct a characteristic vector, a support vector regression algorithm is used to establish a risk prediction model, a radial basis kernel function is selected and the penalty factor, kernel function parameters and fault tolerance parameters of the radial basis kernel function are optimized through cross-validation to obtain an optimized risk prediction model, the characteristic vector is input into the optimized risk prediction model to obtain a fetal development risk score.

[0127] Obtain a comprehensive exposure index and developmental impact score. The comprehensive exposure index can be calculated using multiple data sources, including environmental monitoring data, lifestyle questionnaires, and clinical test data, and ranges from 0 to 10. The developmental impact score can be obtained through expert evaluation and literature evidence, and also ranges from 0 to 10.

[0128] The obtained comprehensive exposure index and developmental impact score are added as input variables to a pre-set fuzzy inference module. In the fuzzy inference module, a Gaussian membership function is established for each input variable to perform fuzzy processing. The Gaussian membership function includes five linguistic variables: low, medium-low, medium, medium-high, and high. The central value and standard deviation of the Gaussian membership function are determined based on expert experience and historical data. For example, for the comprehensive exposure index, the central value of the "low" level is set to 1, with a standard deviation of 0.8; the central value of the "medium-low" level is set to 3, with a standard deviation of 0.8; the central value of the "medium" level is set to 5, with a standard deviation of 0.8; the central value of the "medium-high" level is set to 7, with a standard deviation of 0.8; and the central value of the "high" level is set to 9, with a standard deviation of 0.8. Similar parameter settings are used for the developmental impact score. These parameters are used to calculate the membership values ​​of the comprehensive exposure index and developmental impact score.

[0129] Based on the calculated membership values, a fuzzy rule base was constructed, combining the spatiotemporal characteristics corresponding to the critical stages of embryonic development. These critical stages include organogenesis (gestational weeks 3-8), functional differentiation (gestational weeks 9-25), and system maturation (gestational weeks 26-40). Fuzzy rules were designed for each critical stage. For example, during the organogenesis period, if the comprehensive exposure index is "high" and the developmental impact score is "high," the output risk is "high"; if the comprehensive exposure index is "low" and the developmental impact score is "low," the output risk is "low." The fuzzy rule base contains 25 rules, covering all possible combinations of the comprehensive exposure index and the developmental impact score. Fuzzy reasoning of the fuzzy rules was performed using the Mandani inference method, and clarified using the center of gravity method. A fuzzy risk value representing the fetal developmental risk status was obtained, ranging from 0 to 10.

[0130] The calculated fuzzy risk value is subjected to continuous wavelet transform, and Moore wavelet is selected as the mother wavelet function. The wavelet coefficients at different scales are calculated by adjusting the scale parameter (range 1-64) and the translation parameter (depending on the number of sampling points), and the wavelet energy spectrum is further calculated. Based on the wavelet energy spectrum, the characteristic coefficients corresponding to each critical period of embryonic development are extracted. The characteristic coefficients include the energy value, variance value and entropy value of the wavelet energy spectrum. The energy value represents the energy distribution of the signal in a specific frequency band, which is obtained by calculating the sum of the squares of the wavelet coefficients at each scale. The variance value describes the degree of discreteness of the energy distribution, which is obtained by calculating the sum of the squares of the differences between the energy at each scale and the average energy. The entropy value characterizes the complexity and uncertainty of the signal, which is obtained by calculating the information entropy of the normalized energy spectrum.

[0131] The extracted characteristic coefficients were combined to construct a feature vector, which contained the energy value, variance value and entropy value of each key developmental period, a total of 9 features. The risk prediction model was established using the support vector regression algorithm, and the radial basis kernel function was selected. The penalty factor C (value range 1-100), kernel function parameter γ (value range 0.01-1) and fault tolerance parameter ε (value range 0.01-0.1) of the radial basis kernel function were optimized through five-fold cross validation. After optimization, the optimal parameter combination was determined to be C=10, γ=0.1, ε=0.05, and the optimized risk prediction model was obtained. The characteristic vector was input into the optimized risk prediction model to obtain the fetal development risk score, which ranged from 0-100.

[0132] For example, a pregnant woman undergoes an assessment at week 20 of pregnancy and receives a calculated comprehensive exposure index of 7.2 and a developmental impact score of 6.8. These two indicators are fed into the fuzzy inference module and, after fuzzification, the calculated membership of the comprehensive exposure index at the "medium-high" level is 0.8, and at the "high" level is 0.2. The membership of the developmental impact score at the "medium" level is 0.2, and at the "medium-high" level is 0.8. Through fuzzy rule reasoning and clarification, the fuzzy risk value is 7.5. This fuzzy risk value is transformed using a continuous wavelet transform, the wavelet energy spectrum is calculated, and the characteristic coefficients are extracted: the energy value for the organogenesis period is 2.3, the variance is 0.5, and the entropy is 1.2; the energy value for the functional differentiation period is 4.8, the variance is 0.8, and the entropy is 1.5; and the energy value for the system maturation period is 1.2, the variance is 0.3, and the entropy is 0.9. The characteristic coefficients were combined into characteristic vectors and input into the optimized support vector regression model. The final fetal development risk score was 68, indicating a moderate to high degree of developmental risk. Strengthening pregnancy monitoring and intervention was recommended.

[0133] In this embodiment, fuzzy reasoning technology is introduced to deal with the uncertainty in the evaluation process, a professional fuzzy rule base is constructed based on the characteristics of the critical period of fetal development, wavelet analysis method is used to mine the time-frequency characteristics of risk values, and support vector regression is used to establish a high-precision risk prediction model;

[0134] Existing technologies use simple linear models to handle inherently nonlinear biological response relationships, resulting in low prediction accuracy under boundary conditions and an inability to effectively address uncertainty and ambiguity in the assessment process. In particular, medical expert experience is difficult to accurately quantify, and there is a lack of in-depth analysis of the temporal dimension of fetal development, making it difficult to capture differential risk characteristics at different developmental stages.

[0135] This embodiment introduces a Gaussian membership function for fuzzy processing, mapping the comprehensive exposure index and developmental impact score to five linguistic variables: low, medium-low, medium, medium-high, and high. It successfully overcomes the limitations of the traditional hard threshold method and significantly improves the assessment accuracy under boundary conditions. By building a fuzzy rule base based on expert experience and historical data, it realizes the formal expression and effective application of medical expert knowledge. It combines the spatiotemporal characteristics of the key periods of fetal development (organ formation period, functional differentiation period, and system maturation period) to construct targeted fuzzy rules, so that the assessment process fully considers the sensitivity differences at different developmental stages. By performing continuous wavelet transform on the fuzzy risk value, it successfully extracts the characteristics of risk in the dual dimensions of time and frequency.

[0136] Figure 4 This figure compares the risk assessment performance of different developmental key periods for fetal development assessment methods for maternal microplastic exposure according to an embodiment of the present invention. It shows the risk prediction accuracy of different assessment methods during three key developmental periods. The horizontal axis represents the three key developmental periods: organogenesis (gestational weeks 3-8), functional differentiation (gestational weeks 9-25), and system maturation (gestational weeks 26-40), while the vertical axis represents the prediction accuracy (%) of each method.

[0137] This technical solution (square markers) demonstrated the highest accuracy across all key developmental stages: 91.8% for organogenesis, 94.5% for functional differentiation, and 93.2% for systemic maturation, with consistent high performance across all key stages. In contrast, the traditional linear regression model (circular markers) achieved accuracy rates of 82.3%, 85.1%, and 83.7% for the three key stages, respectively. Its accuracy curve showed a trend of high in the middle and low at both ends, but the overall performance was relatively low. The decision tree model (triangle markers) achieved accuracy rates of 85.6%, 87.4%, and 84.9%, which, while slightly better than the linear regression model, was still significantly lower than this technical solution.

[0138] During the functional differentiation period (gestational weeks 9-25), the accuracy of this technical solution was most pronounced, exceeding the linear regression model by 9.4 percentage points and the decision tree model by 7.1 percentage points. This period is crucial for the functional differentiation of fetal organs, and accurate risk assessment is crucial for clinical intervention decisions. The high accuracy of this technical solution (94.5%) during this period provides clinicians with a more reliable basis for decision-making. The technical solution's performance is more stable across key developmental periods, with no significant fluctuations due to stage-specific characteristics, demonstrating improved adaptability to diverse developmental characteristics and greater generalization capabilities.

[0139] In an optional embodiment,

[0140] Using a microplastic characterization system, the team identified the types of microplastics in blood and urine samples, measured their particle size distribution, and analyzed their morphological characteristics, focusing on the distribution characteristics of common microplastics in the 0.1-500μm range, such as PE, PS, and PVC. Simultaneously, they analyzed their in vivo distribution characteristics, dynamically monitoring microplastic concentrations in the blood and tracking metabolite levels in the urine to calculate the bioaccumulation coefficients and cumulative effects of different types of microplastics.

[0141] For fetal development monitoring, a multidimensional indicator system has been established, including organ development indicators and functional development indicators. Organ development indicators cover morphological parameters such as ventricular size and cerebellar development in the nervous system, ventricular thickness and vascularization in the circulatory system, and liver volume and intestinal development in the digestive system. Functional development indicators include hemodynamic parameters, organ maturity indicators, and metabolic function indicators.

[0142] Through temporal correlation analysis, a time series relationship between microplastic exposure and organ development was established, and correlation coefficients for different time windows were calculated to identify the critical period during which microplastics affect fetal development. A dose-effect analysis was also conducted to establish a relationship between microplastic concentration and developmental abnormalities, determine the impact threshold and safety range, and assess the cumulative effects of microplastics.

[0143] For example, during a systemic examination at 24 weeks of pregnancy, a pregnant woman's blood was found to contain 7.8 ng / mL of PE microplastics and 5.9 ng / mL of PS microplastics, with the main particle size distribution ranging from 0.1 to 3 μm. The calculated bioconcentration factors for PE and PS were 3.5 and 2.9, respectively, indicating a clear tendency for these microplastics to accumulate in the body.

[0144] In the detection of organ development indicators, it was found that the volume of fetal liver was 15% smaller than that of the same age, the liver cell maturity index decreased by 20%, and the serum metabolite spectrum showed that the development of liver function was delayed. Through correlation analysis, it was found that the concentration of PE microplastics was significantly negatively correlated with liver volume (r=-0.82). When the PE concentration exceeded 6ng / mL, the development of liver function was significantly delayed. Further studies showed that the cumulative half-life of microplastics in liver tissue was about 120 hours, indicating that continuous exposure can interfere with the metabolic function development of the liver by affecting the proliferation and differentiation of liver cells.

[0145] A second aspect of an embodiment of the present invention provides a fetal development assessment system for maternal microplastic exposure, which does not involve the diagnosis and treatment of diseases and includes:

[0146] The first unit is used to obtain historical medical examination data of the pregnant woman to be tested and perform standardized preprocessing according to the time series to generate a structured time series feature matrix, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data;

[0147] The second unit is used to establish multidimensional characterization indicators based on the types and particle size distribution of microplastics, classify and count the microplastic detection data in the structured time series feature matrix, calculate the time series cumulative concentration and short-term fluctuation amplitude of microplastics, calculate the bioaccumulation coefficient of microplastics and determine the comprehensive exposure index based on pre-collected maternal blood and urine indicators, construct a distribution characteristic map of microplastics based on tissue and organ specific analysis methods, establish a scoring matrix by calculating the penetration rate and residual time of microplastics in various tissues and organs, and calculate the developmental impact score;

[0148] The third unit is used to input the comprehensive exposure index and the developmental impact score into the fuzzy reasoning module, build a fuzzy rule base and calculate the fuzzy risk value based on the spatiotemporal characteristics of the critical period of embryonic development, perform time-frequency analysis on the fuzzy risk value, extract the characteristic coefficients of different critical development periods, and use support vector regression to predict the fetal development risk score.

[0149] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0150] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0151] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0152] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing fetal development following maternal microplastic exposure, which does not involve the diagnosis and treatment of diseases and is characterized by: include: Obtaining historical medical examination data of the pregnant woman to be tested and performing standardized preprocessing according to time series to generate a structured time series feature matrix, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data; By analyzing the types and particle size distribution of microplastics, we can obtain particle size distribution parameters, chemical composition parameters, and morphological parameters, and construct a multidimensional characterization index for microplastics. Obtain microplastic detection data in the structured time series feature matrix and perform classification statistics, construct a time decay weight matrix, and calculate the corresponding time series cumulative concentration and short-term fluctuation amplitude of microplastics; Collect maternal blood and urine indicators, and correct the ratio of maternal blood and urine indicators using a correction coefficient matrix to obtain a bioconcentration coefficient; The multidimensional characterization indicators, time-series cumulative concentration, short-term fluctuation amplitude and bioconcentration coefficient are combined to form a joint feature vector and added to the pre-set Bayesian deep belief network. The probabilistic dependency relationship between the dimensions of the joint feature vector is established through variational inference. Based on the probabilistic dependency relationship, a Fisher information matrix is ​​constructed to evaluate feature importance and assign weight coefficients. The expectation maximization algorithm is executed to optimize the parameters of the Bayesian deep belief network. Based on the updated Bayesian deep belief network, a risk probability density function is generated. The expected value and variance of the risk score are calculated. The confidence coefficient obtained by the uncertainty estimate is combined to determine the comprehensive exposure index and its confidence interval. Based on the tissue-organ specific analysis method, the specific response characteristics of different tissues and organs to microplastics are obtained, the organ importance weights are determined according to the specific response characteristics to construct an organ weight coefficient matrix, and the spatial distribution vector is constructed based on the organ weight coefficient matrix and the pre-collected organ microplastic concentration, organ volume and organ tissue density; The time-varying penetration coefficient of microplastics in tissues and organs is calculated based on the specific response characteristics of the tissues and organs. The time-varying penetration coefficient is the ratio of the organ microplastic concentration to the exposure concentration at the last detection. The product of the ratio and the exponential function of the organ decay time is calculated, and the time-varying penetration coefficient is integrated to obtain the cumulative penetration index; Based on the specific metabolic mechanism of tissues and organs for microplastics, the concentration-time curve of microplastics in tissues and organs is obtained and fitted through a double exponential, the rapid clearance ratio, rapid clearance rate and slow clearance rate of microplastics by tissues and organs are obtained, and the double exponential function is substituted to construct a residual time distribution function, and the product of the residual time distribution function and time is infinitely integrated to obtain the average residual time; performing normalized inner product operations on the cumulative penetration index, the mean residual time, and the spatial distribution vector with the organ weight coefficient matrix, respectively, multiplying the normalized inner product results by the corresponding weight coefficients and summing the weighted results to obtain a scoring matrix, normalizing the scoring matrix to obtain a normalized impact score, and performing a double summation operation on the normalized impact score and a preset developmental stage sensitivity coefficient to obtain a developmental impact degree score; The comprehensive exposure index and developmental impact score are input into the fuzzy reasoning module. The fuzzy rule base is constructed and the fuzzy risk value is calculated based on the spatiotemporal characteristics of the critical period of embryonic development. The fuzzy risk value is subjected to time-frequency analysis, and the characteristic coefficients of different critical development periods are extracted. The fetal development risk score is predicted by combining support vector regression.

2. The method according to claim 1, characterized in that Obtain the historical medical examination data of the pregnant woman to be tested and perform standardized preprocessing according to the time series to generate a structured time series feature matrix including: Obtaining historical medical examination data of the pregnant woman to be tested, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data; Sequence alignment of the historical medical examination data is performed according to a unified time standard, the original sampling interval of the periodic examination data is maintained, the temporary examination data collected irregularly is aligned to the most recent examination time point based on the principle of proximity, and the missing test data is supplemented by linear interpolation; The aligned historical medical examination data are preprocessed with standardization, the numerical features are normalized to their maximum and minimum values, the categorical features are converted to one-hot encoding, and the processed data are integrated according to the time dimension, feature dimension, and sample dimension to generate a structured time series feature matrix.

3. The method according to claim 1, characterized in that Based on the probability dependency, the Fisher information matrix is ​​constructed to evaluate the feature importance and assign weight coefficients. The expectation maximization algorithm is executed to optimize the parameters of the Bayesian deep belief network. The risk probability density function is generated based on the updated Bayesian deep belief network, including: Calculating the second-order partial derivative of the log-likelihood function according to the probability dependency relationship, constructing a Fisher information matrix, and performing weighted summation on each element of the Fisher information matrix and the corresponding dependency strength weight to obtain a feature importance score; Inputting the feature importance score into an exponential function adjusted by a temperature parameter for normalization to obtain an initial weight vector, and iteratively updating the initial weight vector based on a learning rate parameter to obtain a weight coefficient; Constructing a posterior distribution using the weight coefficients and substituting the posterior distribution into an expectation-maximization algorithm, performing an expectation step by calculating the posterior expectation, performing a maximization step in combination with a regularization term, and iteratively optimizing the parameters of the Bayesian deep belief network; The hidden layer output of the optimized Bayesian deep belief network is combined with the pre-acquired observation features to construct a conditional probability model. Based on the conditional probability model, the joint distribution of the hidden layer output and the observation features is calculated and the joint distribution is marginalized and integrated to obtain the risk probability density function.

4. The method according to claim 1, wherein The comprehensive exposure index and developmental impact score are input into the fuzzy reasoning module. The fuzzy rule base is constructed and the fuzzy risk value is calculated based on the spatiotemporal characteristics of the critical period of embryonic development. The fuzzy risk value is subjected to time-frequency analysis, and the characteristic coefficients of different critical periods of development are extracted and combined with support vector regression to predict the fetal development risk score. The comprehensive exposure index and the developmental impact score are added as input variables to a pre-set fuzzy reasoning module, and Gaussian membership functions are established for fuzzy processing. The Gaussian membership function includes five linguistic variables: low, medium-low, medium, medium-high, and high. The central value and standard deviation of the Gaussian membership function are determined based on expert experience and historical data to obtain the membership values ​​of the comprehensive exposure index and the developmental impact score; Based on the membership value, a fuzzy rule base is constructed in combination with the spatiotemporal characteristics corresponding to the critical period of embryonic development. The fuzzy rules in the fuzzy rule base include an input fuzzy set of the comprehensive exposure index, an input fuzzy set of the developmental impact score, and an output fuzzy set. The fuzzy rules are fuzzy inferenced using the Mandani reasoning method, and are clarified using the center of gravity method to obtain a fuzzy risk value representing the risk status of fetal development. The critical period of embryonic development includes the organ formation period, the functional differentiation period, and the system maturation period. Performing a continuous wavelet transform on the fuzzy risk value, calculating the wavelet energy spectrum at different scales through scale parameters and translation parameters, and extracting characteristic coefficients corresponding to each critical period of embryonic development based on the wavelet energy spectrum, wherein the characteristic coefficients include the energy value, variance value and entropy value of the wavelet energy spectrum; The characteristic coefficients are combined to construct a characteristic vector, a support vector regression algorithm is used to establish a risk prediction model, a radial basis kernel function is selected and the penalty factor, kernel function parameters and fault tolerance parameters of the radial basis kernel function are optimized through cross-validation to obtain an optimized risk prediction model, the characteristic vector is input into the optimized risk prediction model to obtain a fetal development risk score.

5. A fetal development assessment system for maternal microplastic exposure, which does not involve the diagnosis and treatment of diseases and is used to implement the method described in any one of claims 1 to 4, characterized in that: include: The first unit is used to obtain historical medical examination data of the pregnant woman to be tested and perform standardized preprocessing according to the time series to generate a structured time series feature matrix, wherein the historical medical examination data includes microplastic detection data, fetal development indicator data, and maternal health status data; The second unit is used to establish multidimensional characterization indicators based on the types and particle size distribution of microplastics, classify and count the microplastic detection data in the structured time series feature matrix, calculate the time series cumulative concentration and short-term fluctuation amplitude of microplastics, calculate the bioaccumulation coefficient of microplastics and determine the comprehensive exposure index based on pre-collected maternal blood and urine indicators, construct a distribution characteristic map of microplastics based on tissue and organ specific analysis methods, establish a scoring matrix by calculating the penetration rate and residual time of microplastics in various tissues and organs, and calculate the developmental impact score; The third unit is used to input the comprehensive exposure index and the developmental impact score into the fuzzy reasoning module, build a fuzzy rule base and calculate the fuzzy risk value based on the spatiotemporal characteristics of the critical period of embryonic development, perform time-frequency analysis on the fuzzy risk value, extract the characteristic coefficients of different critical development periods, and combine support vector regression to predict the fetal development risk score.

6. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images