Electromagnetic radiation dose calculation method and system based on polynomial fitting

Abstract

The invention relates to an electromagnetic radiation dose calculation method and system based on polynomial fitting, and the method comprises the steps: carrying out the variable human body modeling based on a CT image, and obtaining a variable human body model; selecting physique parameters, and sampling according to the physique parameters to obtain a physique parameter sample; establishing a basic human body model based on the variable human body model and the physical parameter sample, and performing electromagnetic exposure simulation on the basic human body model to obtain an electromagnetic radiation sample; fitting, evaluating and screening a plurality of multi-order PCE models according to the electromagnetic radiation sample to obtain an electromagnetic radiation metering model; and performing human body electromagnetic radiation dose evaluation based on the electromagnetic radiation metering model to obtain a metering result. According to the invention, the individual difference of the human body structure can be reflected more accurately, so that the accuracy of electromagnetic radiation evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bioelectromagnetism, and in particular to a method and system for calculating electromagnetic radiation dose based on polynomial fitting. Background Art

[0002] When an organism is exposed to radiofrequency electromagnetic fields, some of the energy is reflected back by the body, while some of the energy is absorbed. This results in a complex pattern of electromagnetic fields in the body, which depends largely on the characteristics of the electromagnetic fields and the physical properties and shape of the organism. The induced electric field generated by the radiofrequency electromagnetic field in the human body exerts a force on polar molecules and charged particles, causing them to accelerate and interact to generate heat energy. This heat energy is called the electromagnetic radiation thermal effect and will have adverse effects on human health. To measure this health risk, electromagnetic radiation dose is currently commonly used to assess the extent to which biological tissues are affected by electromagnetic radiation.

[0003] The population electromagnetic field exposure assessment scheme based on random dosimetry requires different degrees of simplification of the problem to achieve the purpose of modeling. The Monte Carlo simulation method is generally used to characterize the influence of human morphology and structural characteristics on electromagnetic dose.

[0004] The disadvantages of this method are that it requires a large number of samples and a slow convergence speed. Considering that the construction of each sample requires human body modeling and electromagnetic field simulation, it takes a lot of time and is difficult to implement. Summary of the invention

[0005] The present invention provides an electromagnetic radiation dose calculation method and system based on polynomial fitting, so as to solve the defects of the prior art.

[0006] The present invention provides a method for calculating electromagnetic radiation dose based on polynomial fitting, comprising:

[0007] S1: Perform variable human body modeling based on CT images to obtain a variable human body model;

[0008] S2: Selecting physical parameters, and sampling according to the physical parameters to obtain physical parameter samples;

[0009] S3: establishing a basic human body model based on the variable human body model and the physical parameter sample, and performing electromagnetic exposure simulation on the basic human body model to obtain an electromagnetic radiation sample;

[0010] S4: fitting, evaluating and screening a plurality of multi-order PCE models according to the electromagnetic radiation sample to obtain an electromagnetic radiation metrology model;

[0011] S5: Evaluate the electromagnetic radiation dose of the human body based on the electromagnetic radiation dosing model to obtain dosing results.

[0012] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, step S1 further comprises:

[0013] S11: Collect CT image samples;

[0014] S12: segmenting the multiple CT image samples by a threshold method to obtain a first tissue and organ group and a second tissue and organ group;

[0015] S13: Modeling is performed based on the first tissue and organ group and the second tissue and organ group according to the corresponding grids to obtain a variable human body model.

[0016] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, step S13 further comprises:

[0017] S131: Mapping the tissues and organs in the second tissue and organ group to a human body atlas based on a thin plate spline interpolation method, using the aligned vertices of the adjacent tissues and organs in the first tissue and organ group as control points;

[0018] S132: Using a deep alignment network, align the landmarks of the tissues and organs in the first tissue and organ group and the mapped tissues and organs in the second tissue and organ group to obtain a variable human body model.

[0019] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, step S2 further comprises:

[0020] S21: Collecting physical parameters, and collecting physical data according to the physical parameters;

[0021] S22: Based on the mutual information method, respectively calculating mutual information values ​​of a plurality of physical parameters according to the physical data;

[0022] S23: Based on the rank sum ratio method, sorting the multiple physical parameters according to the mutual information values ​​to obtain a sorting result;

[0023] S24: obtaining a physical parameter by selecting from the physical parameters according to the sorting result;

[0024] S25: Based on the UQLab tool, stratified sampling is performed according to the physical parameters to obtain physical parameter samples.

[0025] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, the physical parameters in step S21 include: height, chest circumference, hip circumference, weight, waist circumference and BMI index, and the physical parameters selected in step S24 include: height, hip circumference, waist circumference.

[0026] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, the expression of the mutual information value calculated in step S22 is:

[0027] I(X l ,X k )=H(X l )+H(X k )-H(X l ,X k );

[0028] Where l is the index value of the first independent random variable, k is the index value of the second independent random variable, and X l is the lth first independent random variable, X k is the kth second independent random variable, I(X l ,X k ) is the first independent random variable X l With the second independent random variable X k The mutual information value of , H(·) is the marginal entropy of the independent random variables, and H(·,·) is the joint entropy of the two independent random variables.

[0029] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, in step S3, when electromagnetic exposure simulation is performed on the basic human body model, the basic human body model faces the incident direction of the electromagnetic wave, and the electric field direction is perpendicular to the height of the basic human body model.

[0030] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, step S4 further comprises:

[0031] S41: performing KLT transformation on the sample points of the electromagnetic radiation sample to obtain a random variable group including a plurality of random variables;

[0032] S42: bringing the random variable groups into multiple multi-order PCE models respectively, solving the model coefficients, and obtaining multiple calculation models;

[0033] S43: performing leave-one-out cross-validation on multiple calculation models respectively, and calculating and obtaining the model accuracy;

[0034] S44: Screen and obtain an electromagnetic radiation metrology model based on model accuracy.

[0035] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, the model accuracy expression in step S43 is:

[0036] Q 2 =1-ε;

[0037] Among them, Q 2is the calculated model accuracy, and ε is the estimated error of cross validation.

[0038] According to a method for calculating electromagnetic radiation dose based on polynomial fitting provided by the present invention, the electromagnetic radiation measurement model in step S44 is a third-order PCE model.

[0039] The present invention also provides an electromagnetic radiation dose calculation system based on polynomial fitting, comprising:

[0040] Modeling module: used for variable human body modeling based on CT images to obtain a variable human body model;

[0041] Sampling module: used to select physical parameters and perform sampling according to the physical parameters to obtain physical parameter samples;

[0042] Simulation module: used for establishing a basic human body model based on the variable human body model and the physical parameter sample, and performing electromagnetic exposure simulation on the basic human body model to obtain an electromagnetic radiation sample;

[0043] Screening module: used for fitting, evaluating and screening multiple multi-order PCE models according to the electromagnetic radiation samples to obtain an electromagnetic radiation metrology model;

[0044] Calculation module: used to evaluate the electromagnetic radiation dose of the human body based on the electromagnetic radiation measurement model to obtain the measurement result.

[0045] The present invention provides a method and system for calculating electromagnetic radiation dose based on polynomial fitting. Through variable human body modeling, it can more accurately reflect the individual differences in human body structure, thereby improving the accuracy of electromagnetic radiation assessment. In addition, by fitting and evaluating electromagnetic radiation samples through a polynomial chaos expansion (PCE) model, the uncertainty in the assessment process can be further reduced, and the reliability of the assessment results can be improved. Secondly, the present invention selects key physical parameters for sampling, and establishes a basic human body model based on these parameters, so that the assessment method can be applicable to people with different physical characteristics. The assessment method based on individual differences makes the assessment results closer to the actual situation, and enhances the applicability of the assessment method; and thirdly, by adopting the polynomial fitting method, the present invention can improve the assessment efficiency while ensuring the accuracy of the assessment, and can have the advantages of fast calculation speed and easy implementation, and can quickly give the electromagnetic radiation measurement assessment results, meeting the rapid assessment needs in practical applications.

[0046] The human electromagnetic radiation dose assessment method provided by the present invention can provide a scientific basis for electromagnetic radiation protection. By assessing the exposure of people with different physical characteristics in an electromagnetic radiation environment, data support can be provided for the formulation of electromagnetic radiation protection standards and measures, thereby protecting human health; the proposal of the present invention has promoted the development of electromagnetic radiation assessment technology. By introducing advanced technologies such as variable human body modeling and polynomial fitting, new ideas and methods are provided for electromagnetic radiation assessment, which is helpful to promote the development of electromagnetic radiation assessment technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A schematic flow chart of a method for calculating electromagnetic radiation dose based on polynomial fitting provided in an embodiment of the present invention;

[0049] Figure 2 A two-dimensional scatter diagram of LHS sampling data for males provided in an embodiment of the present invention;

[0050] Figure 3 A two-dimensional scatter diagram of LHS sampling data for women provided in an embodiment of the present invention;

[0051] Figure 4 Box plot comparison results of the electromagnetic simulation and the proxy model provided by the embodiment of the present invention;

[0052] Figure 5 A schematic diagram of the structure of an electromagnetic radiation dose calculation system based on polynomial fitting provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0054] In order to better understand the embodiments of the present invention, the research background of the present invention is first explained in detail below.

[0055] When an organism is exposed to radio frequency electromagnetic fields, some of the energy is reflected back by the body, while some of the energy is absorbed. This results in a complex pattern of electromagnetic fields in the body, which depends largely on the characteristics of the electromagnetic fields and the physical properties and shape of the organism. The induced electric field generated by the radio frequency electromagnetic field in the human body exerts a force on polar molecules and charged particles, causing them to accelerate and interact to generate heat energy. This heat energy is called the electromagnetic radiation thermal effect, which will have an adverse effect on human health.

[0056] To measure this health risk, electromagnetic radiation dose is currently used to assess the extent to which biological tissues are affected by electromagnetic radiation. Electromagnetic radiation testing can be done mainly using simulation technology and experimental measurement. Considering that experimental measurement can have adverse effects on the human body, most studies use numerical simulation technology to calculate the distribution of electromagnetic fields in biological tissues to calculate the electromagnetic energy absorbed by the human body. Specific Absorption Rate (SAR) is a widely used electromagnetic dose unit used to quantify the electromagnetic power absorbed by biological tissues.

[0057] With the development of computer technology and the refinement of human anatomy, researchers have established a human model library for electromagnetic dose simulation. In the early days, due to the lack of non-invasive measurement technology, geometric models such as spheres and cylinders were mainly used to simulate the human body. In recent years, with the breakthroughs in clinical medical imaging technology (such as MRI, CT) and information technology, research on high-resolution human anatomical models has made significant progress. In addition, the method based on Finite-Difference Time-Domain (FDTD) is also widely used in electromagnetic dose calculations. By comparing the SAR obtained by numerical calculation with the basic limit in the guidelines, it can be evaluated whether the electromagnetic exposure dose of the human body in different electromagnetic environments exceeds the standard and causes health risks.

[0058] At present, the applicability of this limit system based on a few individual models to populations with different characteristics has been controversial, and the dosimetric results from a few models cannot provide reliable dosimetric information for epidemiological studies. Although many studies have evaluated the electromagnetic radiation doses of existing human models, they are still insufficient to conduct electromagnetic exposure assessments for the general population in a wide range of exposure scenarios. Due to differences in body morphology and internal anatomical structures among different subjects, existing technologies show great uncertainty when applied to different populations. In addition, investigating the electromagnetic exposure doses of various organs and tissues in a specific population helps to assess their potential associations with health problems.

[0059] In recent years, researchers have proposed random dosimetry for evaluating human electromagnetic radiation dose. For example, one technique evaluates the WBSAR of children in an indoor wireless LAN environment, and constructs a proxy model based on a low-rank tensor approximation algorithm to characterize the relationship between WBSAR and the child's location. Another technique proposes an exposure assessment scheme that combines electromagnetic computing technology and polynomial chaos expansion (PCE) to estimate the SAR levels of different tissues of children under 1,000 different antenna beams with a low amount of calculation, and determines the elevation and azimuth ranges of the main antenna beam that may lead to the highest exposure level. The third technique uses a proxy model to study the dosimetric variability of individuals caused by displacement in the magnetic resonance coil. This method replaces the heavy numerical simulation with analytical equations, establishes a metamodel to obtain the distribution of quantities of interest (such as WBSAR), and can quantify the electromagnetic dosimetric impact of input variables through sensitivity analysis. These methods provide a new perspective for uncertainty analysis of electromagnetic doses.

[0060] The use of high-performance computing units (HPC) has many advantages in fitting and establishing proxy models. HPC can process large amounts of data and complex calculations in parallel, which significantly improves the speed of fitting and model training, and can use more complex models to improve the accuracy and reliability of the model. At the same time, HPC can run the training of multiple models or parameter combinations at the same time, which is very beneficial for hyperparameter adjustment and model selection, and can find the optimal solution more quickly. Through efficient computing power, especially when simulating complex physical phenomena, HPC can perform multiple iterative optimizations in a shorter time, improve the convergence speed of the model, and provide more real data for fitting, improving the accuracy of the proxy model.

[0061] One of the current existing technologies is that the traditional method of evaluating the safety of electromagnetic field exposure of the population is to generate a series of models that meet the above standards based on the typical percentiles of the reference human characteristic indicators published by the standard organization or published, and then obtain the corresponding electromagnetic field exposure dosimetry results.

[0062] The deformation of tissues and organs generated by this method is mostly based on independent adjustment or scaling of external shape parameters and internal organs, which largely ignores the constraint relationship between anatomical structure and morphological variables in the deformation; secondly, the degree of tissue deformation in the above-mentioned studies ignores individual differences and lacks statistical significance, resulting in a certain gap between the dosimetry results calculated using traditional methods and the dosimetry results of the real population.

[0063] Another existing technology is a population electromagnetic field exposure assessment scheme based on random dosimetry, which requires different degrees of simplification of the problem to achieve the purpose of modeling. Generally, the Monte Carlo simulation method is used to characterize the influence of human morphology and structural characteristics on electromagnetic dose.

[0064] The disadvantages of this method are that it requires a large number of samples and a slow convergence speed. Considering that the construction of each sample requires human body modeling and electromagnetic field simulation, it takes a lot of time and is difficult to implement.

[0065] The present invention solves the problem of human electromagnetic radiation dose assessment by developing a rapid assessment technology for human electromagnetic radiation dose based on polynomial fitting. Taking into account the problem that random dosimetry leads to a very large sample size and significantly increases the calculation time cost, the present invention conducts a rapid assessment of human electromagnetic radiation dose through measurement data such as height, weight, waist circumference, and hip circumference, and adopts a high-performance computing unit to speed up the fitting and proxy model creation speed.

[0066] The embodiments of the present invention are described below with reference to the accompanying drawings.

[0067] like Figure 1 As shown, the present invention provides a method for calculating electromagnetic radiation dose based on polynomial fitting, comprising:

[0068] S1: Perform deformable human body modeling based on CT images to obtain a deformable human body model.

[0069] Wherein, step S1 further comprises:

[0070] S11: Collect CT image samples.

[0071] In step S11, a certain number of CT image samples need to be collected. The samples come from different individuals to cover a wide range of human morphological and structural differences. CT images provide a detailed view of the internal structure of the human body, including bones, muscles, internal organs, etc. The purpose of collecting diverse CT image samples is to ensure that the subsequent modeling steps can capture the diversity of human body structure, thereby improving the accuracy and applicability of the model.

[0072] S12: Segmenting the plurality of CT image samples by a threshold method to obtain a first tissue organ group and a second tissue organ group.

[0073] The threshold method is an image processing technique that segments an image into different regions or tissues based on the range of pixel values. In this step, the threshold method is used to segment the CT image into two groups of tissues and organs: the first tissue and organ group (high-density tissue) and the second tissue and organ group (low-density organ). The purpose of segmentation is to distinguish different tissues and organs inside the human body so that they can be processed separately in the subsequent modeling steps, which helps to improve the refinement and accuracy of the model.

[0074] S13: Modeling is performed based on the first tissue and organ group and the second tissue and organ group according to the corresponding grids to obtain a variable human body model.

[0075] In step S13, a mesh (which may be a three-dimensional mesh) is used to define and construct a human body model, where each vertex of the mesh corresponds to a point on the human body, and the vertices are connected by edges and faces to form a three-dimensional shape. Based on the segmented first tissue and organ group and the second tissue and organ group, the corresponding tissues and organs are mapped to the mesh to construct a complete human body model.

[0076] Step S13 can create a variable human body model that reflects the overall structure of the human body and contains detailed information of internal tissues and organs through mesh modeling. The obtained model will be used in subsequent electromagnetic exposure simulation to evaluate the absorption and distribution of electromagnetic radiation by different human body structures.

[0077] Wherein, step S13 further comprises:

[0078] S131: Based on the thin plate spline interpolation method, the tissues and organs in the second tissue and organ group are mapped to the human body atlas by using the vertices of the aligned adjacent tissues and organs in the first tissue and organ group as control points.

[0079] Furthermore, thin plate spline interpolation is a mathematical method commonly used in image registration and shape deformation. It finds a smooth deformation field by minimizing an energy function (usually related to the curvature of the surface), which can transform one shape into another; the registered first tissue and organ group means that the relative position and shape between the first tissue and organ groups have been calibrated and unified through the automatic registration algorithm, and the control points are the vertices of adjacent tissues and organs selected in the first tissue and organ group. The vertices will be used as input of the thin plate spline interpolation method to define the deformation field; and the human body atlas is a standardized three-dimensional human body model that contains the anatomical structure and organ position information of the human body. The purpose of mapping to the human body atlas is to map the second tissue and organ group (such as low-density organs such as soft tissue) to the human body atlas through the deformation field, so as to be consistent with the first tissue and organ group in space and morphology.

[0080] S132: Using a deep alignment network, align the landmarks of the tissues and organs in the first tissue and organ group and the mapped tissues and organs in the second tissue and organ group to obtain a variable human body model.

[0081] Furthermore, the deep alignment network is an image registration method based on deep learning, which realizes image alignment by learning the correspondence between input images. In this step, the deep alignment network is used to further refine the alignment between the first tissue and organ group and the mapped second tissue and organ group; the landmark points are feature points with anatomical significance selected on the human body model, such as joints, organ edges, etc. The landmark points will be used as input of the deep alignment network to guide the alignment process of the model; the obtained variable human body model is that after the refinement and alignment of the deep alignment network, the first tissue and organ group and the second tissue and organ group will reach a high degree of consistency in the landmark points, thereby forming a complete and accurate variable human body model.

[0082] In a specific embodiment, in step S1, 767 CT images of Chinese adults are first collected, including 496 male CT images and 271 female CT images. The subjects are between 20 and 80 years old and weigh between 40 and 120 kg. The threshold method is used to semi-automatically segment high-contrast tissues and organs (such as bones, liver, spleen, kidneys, etc.) based on the CT images. The low-contrast organs (such as brain tissue and soft tissue) use the vertices of the aligned adjacent high-contrast tissues as control points and are mapped to the human body atlas based on the thin plate spline interpolation method. The deep alignment network is used to align the landmarks and align the tissues. At this time, the anatomical changes between the subjects are modeled as changes in the coordinates of the corresponding mesh vertices of the organs, thereby establishing the SSM, and adjusting the deformation parameters of the SSM to match the appearance of different models to achieve personalized deformable human body modeling.

[0083] S2: Selecting physical parameters, and sampling according to the physical parameters to obtain physical parameter samples.

[0084] Wherein, step S2 further comprises:

[0085] S21: Collect physical parameters, and collect physical data according to the physical parameters.

[0086] The physical parameters in step S21 include: height, chest circumference, hip circumference, weight, waist circumference and BMI index.

[0087] In step S21, basic physical information of the individual is collected to provide a data basis for subsequent analysis. Physical parameters generally include height, chest circumference, hip circumference, weight, waist circumference and BMI (body mass index), etc. The collected data is obtained by consulting health records, etc.

[0088] S22: Based on the mutual information method, calculate mutual information values ​​of multiple physical parameters according to the physical data.

[0089] The purpose of step S22 is to evaluate the correlation between different physical parameters to determine which parameters provide the most independent information. Mutual information is a method to measure the degree of mutual dependence between two random variables. Its expression is as follows, which involves marginal entropy (uncertainty of a single variable) and joint entropy (common uncertainty of two variables). The higher the mutual information value, the stronger the correlation between the two variables.

[0090] The expression of the mutual information value calculated in step S22 is:

[0091] I(X l ,X k )=H(X l )+H(X k )-H(X l ,X k );

[0092] Where l is the index value of the first independent random variable, k is the index value of the second independent random variable, and X l is the lth first independent random variable, X k is the kth second independent random variable, I(X l ,X k ) is the first independent random variable X l With the second independent random variable X k The mutual information value of , H(·) is the marginal entropy of the independent random variables, and H(·,·) is the joint entropy of the two independent random variables.

[0093] S23: Based on the rank sum ratio method, multiple physical parameters are sorted according to the mutual information values ​​to obtain a sorting result.

[0094] Step S23 mainly sorts the physical parameters according to the mutual information values ​​to determine which parameters provide the most independent and useful information. The rank sum ratio method is a sorting method used to deal with decision problems with multiple evaluation indicators.

[0095] S24: obtaining physical parameters by selecting from the physical parameters according to the sorting results.

[0096] The physical parameters selected in step S24 include: height, hip circumference, and waist circumference.

[0097] In step S24, the parameters that best represent the individual's physical characteristics are selected from the sorted physical parameters. In this embodiment, height, hip circumference and waist circumference are selected as physical parameters. The parameters obtained are based on their high mutual information values, indicating that they have low correlation with other parameters, thereby providing more independent information.

[0098] S25: Based on the UQLab tool, stratified sampling is performed according to the physical parameters to obtain physical parameter samples.

[0099] In step S25, a stratified sampling method is mainly used to extract a representative sample from the population for subsequent analysis or modeling. The UQLab tool is an open source toolbox for uncertainty quantification (UQ) and reliability analysis, and stratified sampling is provided based on the tool.

[0100] In a specific embodiment, the present invention first collects physical data of men and women aged 18 to 65 from 2006 to 2011 provided by the National Physical Fitness and Health Database (http: / / cnphd.bmicc.cn / chs / cn / analysis.php) in step S21. According to the characteristics of region, ethnicity and economic development, the survey samples of the database are mainly from multiple regions, and the sample size of each region is about 15,000 people, divided into four survey points in urban and rural areas. A total of 81,490 adults aged 18 to 65 (46,422 female subjects and 35,068 male subjects) are included in the analysis after removing extreme data. The survey population adopts a multiple, multi-region, stratified random sampling method according to the statistical sampling principle. The database mainly measures and records six physical data: height, chest circumference, hip circumference, weight, waist circumference and BMI, including their mean, standard deviation and other indicators.

[0101] In the process of building a proxy model based on PCE, the input variables need to be reduced and simplified first to avoid the dimensionality disaster. Assuming that all six body parameters contained in the database are introduced, when the polynomial order is d = 3 or d = 4, the number of samples required will reach 168 and 420, which takes a considerable amount of time. In order to reduce the computational cost, the number of input variables is first set to 3, and the polynomial order is set to 2. The corresponding number of samples is initially set to 20. In the case of insufficient accuracy, the sample size can be further expanded by increasing the polynomial order.

[0102] For the dimensionality reduction of input variables, we assume that highly correlated physical parameters can be eliminated from the input variables because they are represented by other variables. The present invention uses the mutual information method to simplify the input variables, that is, the mutual information (MI) of two random variables measures the degree of mutual dependence between the two variables, the information entropy represents the degree of uncertainty of the random variables, and the mutual information represents the amount of information contained in one variable of the other variable between the two random variables. Assuming that X and Y are independent random variables that obey the joint distribution p(x, y), the MI of X and Y can be expressed as the mutual information value expression calculated in step S22.

[0103] For l=1...m, k=1...n in the expression, m and n respectively represent the number of rows and columns of the matrix. In this embodiment, m=6, n=6 are taken. For the edge entropy and joint entropy, the specific expressions are as follows.

[0104] H(X l )=-∫dx l f(x l )log 2 f(x l );

[0105] H(X k )=-∫dx k f(x k )log 2 f(x k );

[0106] H(X l ,X k )=-∫∫dx l dx k f(x l ,x k )log 2 f(x l ,x k );

[0107] Among them, f(x l ) and f(x k ) represent x l and x k The probability distribution of f(x l ,x k ) represents the random variable x l 、x k The joint distribution of .

[0108] The analysis results of MI obtained by calculation are shown in Table 1.

[0109] Table 1 MI results of physical parameters

[0110]

[0111] Rank Sum Ratio (RSR) is a new method that combines the advantages of classical parametric statistics and modern nonparametric statistics. The basic idea of ​​RSR is to calculate the ratio of the actual value of each indicator in all evaluation items of the indicator, and convert the value of the evaluation indicator into a corresponding ranking according to the ratio. Therefore, according to the MI results in Table 1, the present invention uses RSR to rank MI.

[0112] The specific calculation is as follows:

[0113]

[0114] Among them, R lk is the rank of the MI between the elements in the lth row and the kth column, I lk is the MI between the elements in the lth row and the kth column, max(I (k) ) and min(I (k) ) are the maximum and minimum values ​​of the evaluation index, RSR l is the RSR result of the elements in row l.

[0115] Furthermore, the present invention performed RSR analysis on the MI results of female and male physical parameters, and the results are shown in Table 2.

[0116] Table 2 Ranking of physical parameters MI based on RSR

[0117]

[0118]

[0119] For RSR, there is no specific threshold used to indicate the degree of correlation. Specifically, a lower RSR indicates that less information is gained about that variable by observing other variables. Therefore, a specific variable tends to be independent of other variables, which helps in the subsequent analysis of the sensitivity of individual physical parameters to dosimetry results.

[0120] In addition, BMI is a derivative index calculated from height and weight, and these two parameters have been selected for further proxy modeling. Weight can be expressed as density multiplied by body size in three dimensions. Therefore, BMI and weight are composite indicators and have been eliminated from the input variables. Since fat content has a significant impact on dosimetry results, the present invention has used weight as a reference indicator for deformable human modeling to improve these models. Therefore, according to the results in Table 2, the present invention selected three physical parameters with relatively low RSR: height, hip circumference, and waist circumference.

[0121] Latin hypercube sampling (LHS) is a commonly used stratified sampling method. Compared with simple random sampling, LHS samples are approximately consistent with the original parameter distribution, the required sample size is much smaller than classical random sampling (MC), and the spatial coverage of the samples is higher. LHS can accurately reconstruct the input distribution through sampling with fewer iterations. In addition, LHS can add new sample points while retaining the original sampling points through resampling, so that the sampling points are expanded while still consistent with the original input distribution, which is important for subsequent experiments.

[0122] Substituting the extracted random variables into the sample expression of LHS sampling in turn, we can obtain the Latin hypercube sample x that obeys the relevant probability density function distribution. 1 ,…,x m .

[0123] UQLab is a universal uncertainty quantification tool, which consists of open source modules such as Monte Carlo simulation, sensitivity analysis, reliability analysis, and uncertainty analysis. Since the three physical parameters we selected obey the Gaussian distribution with known mean and standard deviation, the number of samples required for PCE of different orders is also known. Therefore, the present invention uses the LHS module provided by the MATLAB toolbox UQLab to perform stratified sampling on the relevant uncertain input variables (height, hip circumference, waist circumference). Figure 2 As shown in the figure, it is a two-dimensional scatter diagram of 2nd, 3rd, and 4th order LHS sampling data of male height and hip circumference, as shown in the figure. Figure 3 As shown, it is a two-dimensional scatter diagram of 2nd, 3rd, and 4th order LHS sampling data of women's height and hip circumference. The horizontal axis represents height, and the vertical axis represents hip circumference, and the unit is cm.

[0124] S3: establishing a basic human body model based on the variable human body model and the physical parameter sample, and performing electromagnetic exposure simulation on the basic human body model to obtain an electromagnetic radiation sample.

[0125] Since the electromagnetic exposure dose inside the human body is difficult to measure, it is necessary to use a numerical method based on a deformable human body model to estimate it. The modeling method proposed in the present invention is aimed at generating a human body model that meets both specific requirements of appearance and anatomical structure associations. According to the LHS sampling data, 40 male and female models with different parameters are generated. The model is in an upright posture and contains 51 tissues and organs including muscles, skin, and fat. In addition, the male and female models each contain a separate reproductive organ. Therefore, the male model contains a total of 57 tissues and the female model contains a total of 54 tissues. The resolution is 1mm×1mm×1mm.

[0126] Wherein, in step S3, when performing electromagnetic exposure simulation on the basic human body model, the basic human body model faces the incident direction of the electromagnetic wave, and the electric field direction is perpendicular to the height of the basic human body model.

[0127] The present invention adopts an electromagnetic environment with incident plane electromagnetic wave sources. Specifically, a human body model is placed facing the incident direction of the electromagnetic wave to maximize the area of ​​the human body model directly exposed to electromagnetic radiation. The electric field direction is parallel to the height direction of the model. Five frequencies widely used in second to fifth generation wireless communications, 0.9, 1.8, 2.4, 2.6 and 3.5 GHz, are calculated respectively. The human body model is separated from the boundary of the calculation area by 30 voxels, and PML is used as the absorption boundary condition. WBSAR is normalized to the reference level of plane wave incident power density proposed by ICNIRP.

[0128] Furthermore, for the simulations at 0.9 and 1.8 GHz (containing 61.5 M voxels), the resolution was set to 2 mm × 2 mm × 2 mm, and the simulation time for each simulation was about 10 minutes, while for the simulations at 2.4, 2.6, and 3.5 GHz (containing 145.9 M voxels), the resolution was set to 1.5 mm × 1.5 mm × 1.5 mm, and the simulation time for each simulation was about 40 minutes. The simulations were run on a high-performance workstation with a pair of Intel Xeon Gold 6230 processors, 48 ​​GB onboard memory, and 2 NVIDIA Quadro P5000s (GPU, memory 32 GB GDDR5) using the CUDA GPU acceleration library. The simulations were performed using the FDTD solver of the simulation platform SEMCAD v19.0, and the model material settings were made according to the conductivity, relative permittivity, and density of human tissue provided by Gabriel et al.

[0129] S4: Fitting, evaluating and screening a plurality of multi-order PCE models according to the electromagnetic radiation samples to obtain an electromagnetic radiation metrology model.

[0130] Wherein, step S4 further comprises:

[0131] S41: performing KLT transformation on the sample points of the electromagnetic radiation sample to obtain a random variable group including a plurality of random variables.

[0132] Furthermore, the KLT transformation (Karhunen-Loève Transform) is an orthogonal transformation that converts the original data into a set of new, unrelated random variables. In step S41, the present invention converts the sample points of the electromagnetic radiation sample into a random variable group through the KLT transformation, the purpose of which is to reduce the redundancy of the data and improve the efficiency of subsequent model fitting. The obtained random variables will be used as the input of the PCE model.

[0133] S42: respectively bring the random variable groups into multiple multi-order PCE models, solve the model coefficients, and obtain multiple calculation models.

[0134] Step S42 mainly substitutes the random variable group obtained in S41 into multiple PCE models of different orders. The order of the PCE model determines the complexity of the polynomial. The higher the order, the more complex the data features that the model can capture. By solving the coefficients of each model, we obtain multiple calculation models based on different orders. The model coefficients are obtained by the optimization method of minimizing the model prediction error.

[0135] S43: Perform leave-one-out cross-validation on each of the multiple computational models, and calculate the model accuracy.

[0136] In order to evaluate the performance of multiple calculation models obtained in S42, the Leave-One-Out Cross-Validation (LOOCV) method is adopted in step S43. LOOCV is a cross-validation technique that uses each sample point as a test set and the remaining sample points as training sets. This process is repeated until each sample point has been used as a test set, and a series of prediction errors can be calculated for each model.

[0137] The model accuracy expression in step S43 is:

[0138] Q 2 =1-ε;

[0139] Among them, Q 2 is the calculated model accuracy, and ε is the estimated error of cross validation.

[0140] S44: Screen and obtain an electromagnetic radiation metrology model based on model accuracy.

[0141] Based on the prediction error calculated in step S43, the accuracy index of each model can be calculated. The specific index expression is as follows. The obtained accuracy index is used to compare the performance of different models, so as to screen out the optimal electromagnetic radiation measurement model, that is, the model with the highest prediction accuracy.

[0142] The electromagnetic radiation measurement model in step S44 is a third-order PCE model.

[0143] In step S4, firstly, the second-order and third-order PCEs based on the three input variables are constructed respectively, and the estimation error is verified. Specifically, the number of coefficients to be calculated for the second-order and third-order are 10 and 20 respectively. Then, LHS sampling is performed, and the number of sample points is 20 and 40 respectively. Subsequently, the sample points obtained by sampling are transformed by KLT to obtain independent random variables, which are substituted into the PCE expansions of different orders to obtain the coefficients to be calculated, and the calculation model is obtained.

[0144] Based on the 2nd and 3rd order PCE proxy models and calculating the fitted values ​​calculated by the model Then, according to the above formula, LOOCV is performed on the PCE model to obtain its model accuracy Q 2 .

[0145] Table 3 Leave-one-out cross-validation results of the 2nd-order PCE model

[0146]

[0147] Table 4 Leave-one-out cross-validation results of the 3rd-order PCE model

[0148]

[0149] The cross-validation results of the 2nd-order and 3rd-order PCE proxy models are shown in Tables 3 and 4. The results show that for the 2nd-order PCE model, male The calculation accuracy ranges from a minimum of 93.62% (1.8GHz calculation) to a maximum of 98.72% (0.9GHz calculation). The calculation accuracy ranges from a minimum of 94.85% (0.9 GHz calculation) to a maximum of 99.35% (2.4 GHz calculation). The fitting accuracy of the 2nd-order PCE proxy model under different frequency conditions studied is > 94%. For the 3rd-order PCE model, male The calculation accuracy ranges from a minimum of 95.86% (1.8GHz calculation) to a maximum of 99.17% (2.4GHz calculation). The calculation accuracy ranges from a minimum of 95.02% (0.9GHz calculation) to a maximum of 99.28% (2.4GHz calculation). The fitting accuracy of the third-order PCE proxy model under different frequency conditions is >95%. After comparison, at the same frequency, the accuracy of the third-order PCE model is 0.94% higher than that of the second-order model on average. Due to the existence of numerical calculation errors and calculation model errors, errors are always unavoidable. Therefore, the third-order PCE model proposed in the present invention can meet the calculation accuracy requirements, so there is no need to increase the PCE order. The present invention will discuss the WBSAR fitting value based on the third-order PCE model.

[0150] The specific third-order PCE expansion can be expressed as:

[0151]

[0152] Among them, c 0 ,c 1 ,...,c 19 is the coefficient to be solved, x 1 ,x 2 ,x 3 is the sampling point.

[0153] S5: Evaluate the electromagnetic radiation dose of the human body based on the electromagnetic radiation dosing model to obtain dosing results.

[0154] The following will discuss the WBSAR fitting values ​​based on the third-order PCE model selected in step S4. Figure 4 As shown, Figure 4 Shows the WBSAR and 3rd-order PCE model fitting values ​​obtained by FDTD electromagnetic simulation at different frequencies Box plot comparison, Figure 4 The left picture shows the electromagnetic simulation (WBSAR) and proxy model of a man. The box plot comparison results are Figure 4 The right picture shows the electromagnetic simulation of women (WBSAR) and the proxy model Box plot comparison results, normalized to the reference level of plane wave incident power density proposed by ICNIRP.

[0155] The present invention normalizes WBSAR to the reference level of plane wave incident power density proposed by ICNIRP, that is, the plane electromagnetic wave incident at 0.9 GHz and 1.8 GHz is set to have a power density of 4.5 W / m 2 The incident plane electromagnetic waves at 2.4 GHz, 2.6 GHz and 3.5 GHz frequencies are set to have a power density of 10 W / m 2 .

[0156] The upper and lower edges of the box plot represent the maximum and minimum values ​​of the data, respectively. The center line of the box represents the median of the data. The upper and lower edges of the box represent the upper and lower quartiles of the data, respectively. The difference between the maximum values ​​is <4.2%, and the maximum difference occurs at 1.8 GHz, where the female WBSAR is The maximum values ​​of the difference between the male and female data are <1.8%, and the largest difference occurs at 2.4GHz. The difference between the minimum values ​​is <5.2%, and the maximum difference occurs at 3.5 GHz, where the WBSAR of female data is The difference between the minimum values ​​is <3.3% and the maximum difference occurs at 3.5 GHz frequency.

[0157] like Figure 5 As shown, the present invention also provides an electromagnetic radiation dose calculation system based on polynomial fitting, comprising:

[0158] Modeling module 100: used for performing variable human body modeling based on CT images to obtain a variable human body model;

[0159] Sampling module 200: used to select physical parameters, and perform sampling according to the physical parameters to obtain physical parameter samples;

[0160] Simulation module 300: used to establish a basic human body model based on the variable human body model and the physical parameter sample, and perform electromagnetic exposure simulation on the basic human body model to obtain an electromagnetic radiation sample;

[0161] Screening module 400: used to perform fitting evaluation and screening on multiple multi-order PCE models according to the electromagnetic radiation samples to obtain an electromagnetic radiation metrology model;

[0162] Calculation module 500: used to evaluate the electromagnetic radiation dose of the human body based on the electromagnetic radiation measurement model to obtain a measurement result.

[0163] The present invention provides a method and system for calculating electromagnetic radiation dose based on polynomial fitting, which solves the problem of human electromagnetic radiation dose assessment by developing a rapid assessment technology of human electromagnetic radiation dose based on polynomial fitting. The present invention takes into account the problem that the sample size is extremely large and the calculation time cost is significantly increased due to random dosimetry, and the rapid assessment of human electromagnetic radiation dose is performed through measurement data such as human height, weight, waist circumference, and hip circumference. A novel method is used to establish a deformable human body model with both individual tissue statistical shape differences and multi-tissue constraints. By expanding the modeling training set, the complex inter-tissue relationship in the individual deformation process will continue to approach the real situation. The input variables are reduced in dimension by using the mutual information method and the rank sum ratio method, and the base polynomial type of PCE is selected by non-parametric estimation, and the establishment process of the proxy model is optimized. The proxy model is established by using PCE to obtain the WBSAR of the population under different percentiles of appearance parameters, and the calculation accuracy of the proxy model is >95%.

[0164] The method and system proposed in the present invention establish a human meta-model for numerical evaluation of radio frequency electromagnetic radiation, and perform PEC model fitting through human body shape parameters. In the case of omitting human body scanning, modeling, simulation and other links, the human body WBSAR can be obtained only by measuring human body shape data. This not only greatly reduces the time cost of human electromagnetic radiation dosimetry evaluation, but also can further study the electromagnetic exposure level of the population through the distribution of WBSAR of the population.

[0165] 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for calculating electromagnetic radiation dose based on polynomial fitting, characterized in that: include: S1: Perform variable human body modeling based on CT images to obtain a variable human body model; S2: Selecting physical parameters, and sampling according to the physical parameters to obtain physical parameter samples; S3: establishing a basic human body model based on the variable human body model and the physical parameter sample, and performing electromagnetic exposure simulation on the basic human body model to obtain an electromagnetic radiation sample; S4: fitting, evaluating and screening a plurality of multi-order PCE models according to the electromagnetic radiation sample to obtain an electromagnetic radiation metrology model; S5: Evaluate the electromagnetic radiation dose of the human body based on the electromagnetic radiation dosing model to obtain dosing results.

2. The electromagnetic radiation dose calculation method based on polynomial fitting according to claim 1, characterized in that: Step S1 further comprises: S11: Collect CT image samples; S12: segmenting the multiple CT image samples by a threshold method to obtain a first tissue and organ group and a second tissue and organ group; S13: Modeling is performed based on the first tissue and organ group and the second tissue and organ group according to the corresponding grids to obtain a variable human body model.

3. The electromagnetic radiation dose calculation method based on polynomial fitting according to claim 2, characterized in that: Step S13 further comprises: S131: Mapping the tissues and organs in the second tissue and organ group to a human body atlas based on a thin plate spline interpolation method, using the aligned vertices of the adjacent tissues and organs in the first tissue and organ group as control points; S132: Using a deep alignment network, align the landmarks of the tissues and organs in the first tissue and organ group and the mapped tissues and organs in the second tissue and organ group to obtain a variable human body model.

4. The electromagnetic radiation dose calculation method based on polynomial fitting according to claim 1, characterized in that: Step S2 further comprises: S21: Collecting physical parameters, and collecting physical data according to the physical parameters; S22: Based on the mutual information method, respectively calculating mutual information values ​​of a plurality of physical parameters according to the physical data; S23: Based on the rank sum ratio method, sorting the multiple physical parameters according to the mutual information values ​​to obtain a sorting result; S24: obtaining a physical parameter by selecting from the physical parameters according to the sorting result; S25: Based on the UQLab tool, stratified sampling is performed according to the physical parameters to obtain physical parameter samples.

5. The method for calculating electromagnetic radiation dose based on polynomial fitting according to claim 4, characterized in that: The physical parameters in step S21 include: height, chest circumference, hip circumference, weight, waist circumference and BMI index, and the physical parameters selected in step S24 include: height, hip circumference and waist circumference.

6. The method for calculating electromagnetic radiation dose based on polynomial fitting according to claim 4, characterized in that: The expression of the mutual information value calculated in step S22 is: I(X l ,X k )=H(X l )+H(X k )-H(X l ,X k ); Where l is the index value of the first independent random variable, k is the index value of the second independent random variable, and X l is the lth first independent random variable, X k is the kth second independent random variable, I(X l ,X k ) is the first independent random variable X l With the second independent random variable X k The mutual information value of , H(·) is the marginal entropy of the independent random variables, and H(·,·) is the joint entropy of the two independent random variables.

7. The electromagnetic radiation dose calculation method based on polynomial fitting according to claim 1, characterized in that: In step S3, when electromagnetic exposure simulation is performed on the basic human body model, the basic human body model faces the incident direction of the electromagnetic wave, and the direction of the electric field is perpendicular to the height of the basic human body model.

8. The method for calculating electromagnetic radiation dose based on polynomial fitting according to claim 1, characterized in that: Step S4 further comprises: S41: performing KLT transformation on the sample points of the electromagnetic radiation sample to obtain a random variable group including a plurality of random variables; S42: bringing the random variable groups into multiple multi-order PCE models respectively, solving the model coefficients, and obtaining multiple calculation models; S43: performing leave-one-out cross-validation on multiple calculation models respectively, and calculating and obtaining the model accuracy; S44: Screen and obtain an electromagnetic radiation metrology model based on model accuracy.

9. The method for calculating electromagnetic radiation dose based on polynomial fitting according to claim 8, characterized in that: The electromagnetic radiation measurement model in step S44 is a third-order PCE model.

10. An electromagnetic radiation dose calculation system based on polynomial fitting, characterized in that: include: Modeling module: used for variable human body modeling based on CT images to obtain a variable human body model; Sampling module: used to select physical parameters and perform sampling according to the physical parameters to obtain physical parameter samples; Simulation module: used for establishing a basic human body model based on the variable human body model and the physical parameter sample, and performing electromagnetic exposure simulation on the basic human body model to obtain an electromagnetic radiation sample; Screening module: used for fitting, evaluating and screening multiple multi-order PCE models according to the electromagnetic radiation samples to obtain an electromagnetic radiation metrology model; Calculation module: used to evaluate the electromagnetic radiation dose of the human body based on the electromagnetic radiation measurement model to obtain the measurement result.

Images