A method and device for detecting motion load process based on linear mixed model

Abstract

The present invention relates to the field of sports physiology data analysis and health monitoring, in particular to a method and device for detecting a sports load process based on a linear mixed model, the method comprising the following steps: S1: collecting physiological index data by performing multiple measurements on the same group of individuals at multiple time points, and constructing longitudinal data based on the physiological index data; S2: constructing a linear mixed model based on the longitudinal data; S3: using the linear mixed model to analyze the longitudinal data, and obtaining the fixed effect estimate, random effect estimate and covariance matrix of the individual in different exercise stages; the fixed effect estimate is used to describe the average influence of all individuals; the random effect estimate and covariance matrix are used to quantify the physiological load changes of the individual in different exercise stages. The method of the present invention captures the dynamic changes of the individual's physiological indicators before and after exercise by analyzing the repeated measurement data at multiple time points, and is suitable for health monitoring and sports load assessment in sports physiology.

Description

Technical Field

[0001] The present invention relates to the field of sports physiology data analysis and health monitoring, and in particular to a method and device for detecting a sports load process based on a linear mixed model. Background Art

[0002] With the development of sports medicine and physiological monitoring technology, the impact of exercise load on individual physiological indicators has become a research focus. Traditional exercise load detection methods mainly measure static or dynamic indicators of exercisers before and after exercise, and use simple statistical analysis methods to evaluate the impact of exercise on individual physiological load. These methods usually rely on observation data at one or a few time points, ignoring the characteristics of individual dynamic changes over time and the differences between different individuals. The existing exercise load detection methods have the following problems and shortcomings: First, traditional methods mostly use static models such as linear regression or variance analysis, which cannot effectively process repeated measurement data that change over time and are difficult to capture the dynamic physiological changes of individuals; second, individual differences are not fully considered, and different responses of different individuals to the same exercise load cannot be accurately reflected; finally, the existing technology has poor flexibility when facing missing data or irregular time intervals, which limits the accuracy of the analysis results. Therefore, there is an urgent need for a model that can simultaneously handle individual differences and dynamic changes over time to achieve a comprehensive and accurate quantitative analysis of the impact of exercise load.

[0003] For example, in existing physiological load tests, commonly used indicators include blood routine, biochemical indicators, heart rate, etc. Researchers infer the effect of exercise load by collecting these indicator data at fixed time points, but these methods are difficult to fully and dynamically describe the changes in individual physiological load, especially when there are significant differences between individuals. Existing methods such as linear regression or variance analysis fail to effectively deal with hierarchical problems and individual differences in the data.

[0004] At present, existing exercise load detection and analysis technologies usually use simple linear models or single time point analysis. For example, the Chinese invention patent with announcement number CN109385747A discloses an exercise load detection method based on heart rate changes, which uses a linear regression analysis method for heart rate data. However, this method cannot handle the time correlation and random differences between individuals in repeated measurement data, and is particularly limited in the analysis of multi-time point measurement data. In addition, this method is difficult to deal with irregular intervals or missing observation data.

[0005] The Chinese invention patent with announcement number CN106477425A discloses a method and device for detecting exercise load, and describes a device for determining exercise load by measuring bioelectric and biophysical parameters. Although the device can monitor the individual's exercise status in real time and provide a simple analysis report, it does not introduce a dynamic statistical model and cannot fully capture the individual's complex physiological reactions that change over time, nor can it effectively distinguish the reaction patterns of different individuals. Summary of the invention

[0006] The present invention proposes a method and device for detecting the exercise load process based on a linear mixed model. The exercise load quantification model of the present invention considers both fixed effects and random effects during the analysis process through a linear mixed model (LMM), which can effectively capture the changes in dynamic physiological indicators of individuals at multiple time points and quantify the differences between individuals. Compared with the existing model that only considers fixed effects, the present invention has stronger flexibility, especially showing significant advantages in processing missing data and irregular time point data.

[0007] In order to achieve the above objectives, the following technical solutions are proposed:

[0008] A method for detecting a motion load process based on a linear mixed model comprises the following steps:

[0009] S1: Collect physiological index data by taking multiple measurements of the same group of individuals at multiple time points, and construct longitudinal data based on the physiological index data;

[0010] S2: Linear mixed models were constructed based on longitudinal data;

[0011] S3: The longitudinal data were analyzed using a linear mixed model to obtain the fixed effect estimates, random effect estimates and covariance matrix of individuals in different exercise stages; the fixed effect estimates were used to describe the average impact of all individuals; the random effect estimates and covariance matrix were used to quantify the changes in physiological load of individuals in different exercise stages.

[0012] Preferably, the longitudinal data include the response variable, the covariate of the response variable and repeated measurements of the response variable at different times.

[0013] Preferably, the specific steps of S2 are:

[0014] The physiological index data were preprocessed to test whether the response variables in the longitudinal data conformed to the normal distribution. If so, a linear mixed model was used;

[0015] Determine random effects and fixed effects based on longitudinal data to obtain fixed effect independent variables and random effect independent variables;

[0016] A linear mixed model was established based on random effect variables and fixed effect variables.

[0017] Preferably, the specific steps of S3 are:

[0018] The parameters of the linear mixed model are selected and estimated simultaneously using the restricted maximum likelihood estimation method or the Bayesian method;

[0019] Check whether the residual distribution of the linear mixed model meets the normality assumption and verify the goodness of fit of the model;

[0020] The fixed effect estimates, random effect estimates and covariance matrices of individuals in different exercise stages were obtained.

[0021] Preferably, the steps of the restricted maximum likelihood estimation method include:

[0022] Define the log-likelihood function for the linear mixed model;

[0023] The SCAD penalty term is introduced to optimize the log-likelihood function;

[0024] The log-likelihood function is solved by a numerical optimization algorithm to obtain the estimated values ​​of the fixed effects by restricted maximum likelihood estimation.

[0025] Preferably, the specific steps of using the random effect estimate and the covariance matrix to quantify the changes in physiological load of an individual at different exercise stages are:

[0026] Linear mixed models were used to compare random effect estimates of different individuals to identify individual differences in physiological indicators;

[0027] Analyze the correlation between different random effect independent variables in the covariance matrix;

[0028] The physiological load change curves of different individuals at different time points before and after exercise were drawn based on the fixed effect estimates, random effect estimates and covariance matrix.

[0029] Preferably, the calculation formula of the linear mixed model is:

[0030] Y=Xβ+ZB+∈

[0031] where Y is the response variable vector of dimension n×1; X is the fixed effect design matrix of dimension n×p; β is the fixed effect parameter vector of dimension p×1; Z is the random effect design matrix of dimension n×q; B is the random effect vector of dimension q×1; ∈ is the error term vector of dimension n×1, and p and q are positive integers.

[0032] Preferably, the calculation formula for the correlation between different random effect independent variables in the covariance matrix is:

[0033] (Y,B)~N((Xβ,0),(σ²ICov(Y,B)Cov(B,Y)ΣB))

[0034] Where Y is the n×1 response variable vector, B is the q×1 random effect vector, Xβ represents the effect of fixed effects on the response variable, σ 2 I is the covariance matrix of the response variable Y, Cov(Y,B) and Cov(B,Y) are the covariance matrices between Y and B, and ΣB is the covariance matrix of the random effect B.

[0035] Based on the same concept, a motion load process detection device based on a linear mixed model is also proposed, comprising at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any one of the above-mentioned motion load process detection methods based on a linear mixed model.

[0036] Compared with the prior art, the method of the present invention has the following beneficial effects: the method of the present invention introduces a linear mixed model, which combines fixed effects and random effects, and is used to analyze dynamic changes in repeated measurement data. This model solves the problem that traditional models cannot simultaneously handle temporal dynamic changes and individual differences, and shows significant advantages in processing missing data and irregular time point data. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of a method for detecting a motion load process based on a linear mixed model in Example 1;

[0038] Figure 2 is a framework diagram of the linear mixed model in Example 1;

[0039] Figure 3 is a flowchart of data cleaning, interpolation and model estimation in Example 1;

[0040] Figure 4 This is a flow chart of the experiment on the effect of exercise on pregnant women in Example 3;

[0041] Figure 5 It is the QQ diagram in Example 3 for determining whether the response variable obeys the normal distribution;

[0042] Figure 6 is a schematic diagram of the change of individual blood glucose level over time in Example 3;

[0043] Figure 7 It is a schematic diagram of the structure of a device described in Example 4 that utilizes the method described in Example 1. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

[0045] Example 1

[0046] A method and device for detecting motion load process based on linear mixed model, the flow chart is as follows Figure 1 As shown, the following steps are included:

[0047] S1: Collect physiological index data by taking multiple measurements of the same group of individuals at multiple time points, and construct longitudinal data based on the physiological index data;

[0048] S2: Linear mixed models were constructed based on longitudinal data;

[0049] S3: The longitudinal data were analyzed using a linear mixed model to obtain the fixed effect estimates, random effect estimates and covariance matrix of individuals in different exercise stages; the fixed effect estimates were used to describe the average impact of all individuals; the random effect estimates and covariance matrix were used to quantify the changes in physiological load of individuals in different exercise stages.

[0050] Furthermore, the data of the experiment of the present invention is repeated measurement data, which refers to multiple measurements of the same group of individuals at multiple time points, with time correlation and hierarchical structure. This data type can capture dynamic changes in time series and differences between individuals, and is often used to study the changing patterns and trends of individuals over time.

[0051] First, the physiological index data were preprocessed, and interpolation or multiple interpolation techniques were used to clean the missing data and irregular time intervals.

[0052] In an optional implementation, the locations of missing values ​​and irregular time intervals in the data are determined, and the specific steps include:

[0053] 1. Check data integrity and identify missing values ​​(`NaN` or null values);

[0054] 2. Calculate the percentage and distribution of missing values;

[0055] 3. Check whether the time interval is irregular by variable (such as the missing value of a physiological indicator) or by time point (such as the missing value at a certain moment);

[0056] 4. Check whether all individuals were observed at the same time point;

[0057] 5. Check for any jumping time points or uneven intervals.

[0058] The longitudinal data refers to the observations obtained by repeatedly measuring the research subjects at multiple time observation points, also known as repeated measurement data and panel data; the longitudinal data includes repeated measurements of the same subjects (or other "experimental units") over time. Longitudinal data can not only reflect the overall level of change of the research subjects and effectively describe the relationship between the response variables and covariates, but also study the differences within individuals.

[0059] The specific method of constructing longitudinal data is to find the response variable y of interest and its related covariate x, as well as repeated measurements of the response variable y at different times.

[0060] Often, we want to characterize trends over time both within and across subjects. Data will always include a response, a time covariate, and an indicator of the subject on whom the measurement was made. If other covariates are recorded, such as whether a subject was in a treatment or control group, we may want to relate within- and between-subject trends to such covariates. For longitudinal data with multiple subjects, the most important relationship is the trend of the response across subjects over time. Our models describe the relationship between the response variable and some covariate that has been measured or observed along with the response.

[0061] S2 As a specific embodiment, if the response variable in the experiment is a continuous variable, obeys a normal distribution, and the research object is repeatedly measured over time, consider using a linear mixed model to perform a regression simulation process.

[0062] The linear mixed model is an extended linear model used to deal with fixed effects and random effects in data. Determining the fixed effects and random effects, and the degree of their influence on the response variable is a key step in building a model. Generally, the key effects on the response variable are considered as fixed effects, while the changes between individuals are random effects. This model not only takes into account the fixed effects of individuals, but also captures the differences between individuals and their changes over time by introducing random intercepts and random slopes, which more comprehensively reflects the randomness and change trend of the data.

[0063] Furthermore, a quantitative model for detecting the exercise load process (i.e., a linear mixed model) is established to analyze the characteristics of each data. Figure 2 As shown, the basic form of the linear mixed model is:

[0064] Y = Xβ + ZB + ∈, (1)

[0065] In this model, Y is the n×1 response variable vector, containing all observations; X is the n×p fixed effect design matrix, containing the fixed effect independent variables; β is the p×1 fixed effect parameter vector; Z is the n×q random effect design matrix, containing the random effect independent variables; B is the q×1 random effect vector, usually assumed to be a multivariate normal distribution B~N(0,ΣB); ∈ is the n×1 error term vector, usually assumed to be an independent and identically distributed normal distribution ∈~N(0,σ 2 I), where I is the identity matrix, σ 2 is the variance of the error. This model can handle the hierarchical structure in the data and the correlation within the group, while taking into account the variability of individuals and the impact of fixed effects.

[0066] The linear model is described by the distribution of the response variable Y, which is described by the distribution of two vector-valued random variables Y (response variable) and B (random effect variable). In this model, the distribution of Y is a multivariate normal distribution.

[0067] Y~N(Xβ+∈,σ 2 I-1),(2)

[0068] Where Y is an n-dimensional response vector, I is a diagonal matrix of known prior weights, β is a p-dimensional coefficient vector, X is an n×p model matrix, and ∈ is a vector of known prior bias terms. The parameters of the model are the coefficients β and the scale parameter σ. In this model, given B=b, the conditional distribution of Y has the following form:

[0069] (Y|B=b) ~N(Xβ+Zb+∈,σ 2 I-1),(3)

[0070] where Z is the model matrix of size n×q, and the B vector is a random effect variable of dimension q, with B fixed at b. The unconditional distribution of B is also a multivariate normal distribution with mean zero and a covariance matrix of ΣB,

[0071] B~N(0,ΣB),(4)

[0072] In this model, the joint distribution of the response variable Y and the random effect B can be expressed as:

[0073] (Y,B) ~N((Xβ,0),(σ²ICov(Y,B)Cov(B,Y)ΣB))

[0074] Where Y is the n×1 response variable vector, B is the q×1 random effect vector, Xβ represents the effect of fixed effects on the response variable, σ 2I is the covariance matrix of the response variable Y, which represents the variance of the error term, Cov(Y,B) and Cov(B,Y) are the covariance matrices between Y and B, and ΣB is the covariance matrix of the random effect B, which represents the variance of the random effect. This joint distribution describes the overall statistical characteristics of the response variable Y and the random effect B, taking into account the influence of fixed effects and random effects and the correlation between them.

[0075] The linear effect model not only takes into account the individual fixed effect, but also captures the differences between individuals and their changes over time by introducing random intercepts and random slopes, which more comprehensively reflects the randomness and change trend of the data. Another key feature of repeated measurement data is the need to pay attention to both individual changes over time and differences between individuals. Since the linear mixed model has the ability to process repeated measurement data, the model can effectively process data at different time points and comprehensively capture the dynamic changes of individual physiological loads. It is especially suitable for motion detection and analysis at multiple time points. Compared with the existing single time point analysis method, the present invention greatly improves the accuracy and comprehensiveness of detection.

[0076] S3 is a specific embodiment. Figure 3 As shown, the present invention uses maximum likelihood estimation or Bayesian method to estimate parameters, and then verifies the model to check whether the residual distribution conforms to the normality assumption and verify the goodness of fit of the model; finally, the change trend and analysis report of individual physiological load are output, specifically including the estimated value of fixed effect (such as the average effect of time on blood sugar), the estimated value of random effect and the covariance matrix, so as to evaluate individual differences.

[0077] Among them, the Penalized Maximum Likelihood Estimation (PMLE) process includes:

[0078] 1. Fitting goal: The goal is to maximize the likelihood estimation function (MLE, Maximum Likelihood Estimation), that is, to find a set of model parameters that maximizes the probability of this set of parameters under given observation data.

[0079] 2. Define the log-likelihood function: Based on the distribution assumptions of the model, define the log-likelihood function of the model. The log-likelihood function is the natural logarithm of the likelihood function and is often used to simplify the calculation and optimization process.

[0080] 3. Introducing penalty function: In order to avoid overfitting of the model and effectively select important variables, the SCAD (Smoothly Clipped Absolute Deviation) penalty function is introduced. The SCAD penalty function can automatically select important variables during parameter estimation and shrink unimportant variables, thereby obtaining a sparser model.

[0081] 4. Optimize the objective function: Add the penalty term to the log-likelihood function to form the entire objective function. This objective function takes into account both the degree of fit of the data and the complexity of the model (controlled by the penalty term).

[0082] 5. Solve parameter estimates: Solve the maximum value of the objective function through numerical optimization algorithms (such as EM algorithm, gradient descent method, Newton method, etc.) to obtain the estimated values ​​of the model parameters. These estimates take into account the degree of data fit and avoid the risk of overfitting.

[0083] 6. Result evaluation: After obtaining the parameter estimates, the model can be further evaluated, such as calculating the predictive performance of the model, explaining the model parameters, etc. At the same time, methods such as cross-validation can also be used to evaluate the stability and generalization ability of the model.

[0084] By restricting the maximum likelihood estimation process, we can obtain a set of model parameters that both meet the data distribution assumptions and avoid the risk of overfitting, thereby building a more robust and reliable statistical model.

[0085] Example 2

[0086] The linear effects model is different from the traditional ANOVA regression model in that it does not require the observations to be independent of each other, and this model has greater flexibility in dealing with missing data and irregularly spaced repeated measurements.

[0087] As a specific example, take the mixed effect model experiment of changes in blood routine indicators of pregnant women under exercise load as an example. In this experiment, time can be regarded as a stratum, that is, the fixed effect is time, and exercise at different time points has an impact on all individuals participating in the experiment, and the random effect is the difference between individuals at the beginning.

[0088] The linear effect model was used to analyze the effects of exercise on blood routine indicators and to evaluate whether these effects were associated with the risk of gestational diabetes. The following are the detailed steps and the model representation in matrix form:

[0089] 1. Establish the data structure of this experiment and preprocess the experimental data:

[0090] Observation unit: blood routine indicators of each subject at different time points.

[0091] Time points: before exercise, immediately after exercise, 1 hour after exercise, 24 hours after exercise, 12th week of exercise, and 20th week of exercise.

[0092] The variables included are:

[0093] Response variables: blood routine indicators (such as red blood cell count, white blood cell count, etc.).

[0094] Fixed effect independent variables: time point (may be a categorical variable), whether or not there is gestational diabetes (categorical variable), and other influencing fixed effects (such as basic characteristics of pregnant women: age, weight, etc.).

[0095] Random effects independent variables: individual differences among subjects (random effects for each subject).

[0096] Preprocess the variables to identify and resolve the problem of missing values ​​and irregular time intervals in the data:

[0097] The first type: If there are few missing values ​​and the data is continuous (such as time series data), the steps to use interpolation to handle missing values ​​include:

[0098] 1. Select the interpolation method

[0099] Linear interpolation: Missing values ​​are linearly estimated from the values ​​at adjacent time points, which is suitable for smoothly changing data.

[0100] Polynomial interpolation: Uses high-order equations to fit, suitable for nonlinearly varying data.

[0101] Spline interpolation: Use piecewise polynomial fitting, suitable for complex and smooth data changes.

[0102] 2. Determine the location of missing values ​​in the time series

[0103] According to the interpolation method, the estimated value of the missing value is calculated. As a specific embodiment, the blood glucose value of a participant at time points `'t1, t2, t3, t4` is interpolated:

[0104] Original data: [4.5, NaN, 4.8, NaN]

[0105] Interpolation result: [4.5, 4.65, 4.8, 4.8]

[0106] The rationality of the interpolation result can be verified by drawing the data change curve before and after interpolation to check whether it conforms to the trend.

[0107] The second type: If there are many missing values ​​and the data dimension is high (such as containing multiple features), it is necessary to consider randomness and uncertainty and use multiple interpolation technology to deal with missing values. The steps to deal with missing values ​​include:

[0108] 1. Identify missing patterns: Determine whether missing values ​​are completely random (MCAR), conditionally random (MAR), or not random (MNAR).

[0109] 2. Select the interpolation method:

[0110] Mean imputation: Fill in with the mean (simple but may underestimate variance).

[0111] Regression imputation: predicting missing values ​​based on other variables (taking into account correlation).

[0112] Multiple imputation: Generate multiple imputation datasets and combine multiple imputation results.

[0113] 3. Interpolation process:

[0114] Choose an interpolation algorithm, such as Multiple Imputation by Chained Equations (MICE) or KNN (K-Nearest Neighbor) algorithm;

[0115] The input data may include participants' blood glucose levels, weight, and exercise time, and multiple random interpolations are performed on missing values ​​to generate multiple complete data sets; each data set is then analyzed separately and the results are finally merged.

[0116] 4. Verify the interpolation results:

[0117] Check whether the distribution of the interpolated data is consistent with the original data.

[0118] Statistical analysis was performed on the imputed data to assess their rationality.

[0119] The third type: There are situations where the time intervals are uneven, which may affect the temporal dynamic characteristics of the model. The specific steps include:

[0120] 1. Standardized time interval: Determine a uniform time interval (such as weekly or daily) and resample the data to a uniform time point.

[0121] 2. Interpolation to supplement missing time points: For unobserved time points, use interpolation to estimate the values ​​of physiological indicators.

[0122] As a specific embodiment, it is as follows:

[0123] Original time point: [t1=0,t2=2,t3=4,t4=7]`

[0124] Standard time points: [t1=0, t2=1, t3=2, t4=3, t5=4, t6=5, t7=6, t8=7]`

[0125] Interpolation result: original data: [4.5, NaN, 4.8, 5.0];

[0126] After completion: [4.5, 4.6, 4.7, 4.8, 4.85, 4.9, 4.95, 5.0].

[0127] 3. After the preprocessing is completed, the method for verifying the preprocessing results is as follows:

[0128] Check whether the trend of the interpolated data is consistent with reality (such as smoothness of change);

[0129] Draw a comparison chart of the data before and after interpolation: such as the time-physiological index change curve;

[0130] The mean square error or root mean square error is used as a statistical indicator to evaluate the interpolation effect; the consistency of data distribution is tested by the mean or variance.

[0131] By introducing data cleaning steps such as interpolation or multiple interpolation, the model can cope with the common data missing and irregular time interval problems in practical applications, ensuring data integrity and improving the stability and robustness of model analysis. This improvement overcomes the limitations of existing technologies when dealing with incomplete data.

[0132] 2. Establish a quantitative model for detecting the exercise load process:

[0133] Y=Xβ+Zb+∈

[0134] Here, Y is an n × 1 response variable vector, where n is the total number of observations (40 pregnant women × 4 time points = 160 observations);

[0135] X is an n × p fixed-effect design matrix containing fixed-effect independent variables (e.g., time point is a categorical variable, an indicator variable for whether or not the patient has gestational diabetes, etc.);

[0136] β is a p × 1 fixed effect parameter vector, representing the coefficient of the fixed effect;

[0137] Z is an n × q random effects design matrix containing the random effect independent variables (such as individual differences among pregnant women);

[0138] b is a q×1 random effect parameter vector, representing the coefficient of the random effect;

[0139] ∈ is an n×1 error term vector, usually assumed to be an independent and identically distributed normal distribution.

[0140] Because the model uses fully parametric modeling technology, it can efficiently integrate and analyze complex data of multiple variables and multiple time points, which greatly improves the model's computational efficiency. Compared with traditional methods, the time for modeling and analysis is significantly shortened, the efficiency of data processing is improved, and it meets the needs of large-scale data sets.

[0141] Example 3

[0142] The following is an example of analyzing the effects of exercise on pregnant women provided for this application. Further, whether there is a difference in the effects of acute exercise and chronic exercise on pregnant women is analyzed. The experimental flow chart is as follows Figure 4 shown.

[0143] Step 1: Prepare data

[0144] Subjects: 30 exercise group, 10 control group

[0145] Data categories:

[0146] Basic information: age, number of pregnancies, body mass index before pregnancy, and weight gain during pregnancy, etc.

[0147] Serology: complete blood lipid panel, blood sugar, insulin, free fatty acids, etc.

[0148] Multi-omics: targeted metabolomics, untargeted proteomics, transcriptomics

[0149] Data collection nodes:

[0150] Before exercise, immediately after exercise, one hour after exercise, 12 weeks of exercise, and 20 weeks of exercise.

[0151] Step 2: Data processing: construct a longitudinal data set through the data collected above;

[0152] Specifically, a longitudinal data set with blood glucose as the response variable was constructed, where the covariates were age, exercise, BMI, and weeks of pregnancy. Some of the data are shown in Table 1.

[0153] Table 1 Part of the longitudinal data of the experiment with blood glucose as the response variable

[0154]

[0155] Among them, BG is blood sugar; age represents age; sports is a binary variable, 1 represents exercise, and 0 represents no exercise; g-weeks is the gestational week of the pregnant woman when the blood sugar value is measured; week represents the week of experimental measurement, 0 represents measurement before exercise, 12 represents measurement after 12 weeks of exercise, and 20 represents measurement after 20 weeks of exercise.

[0156] Step 3: Build the model

[0157] (1) Check whether the response variable conforms to the normal distribution:

[0158] like Figure 5 As shown in the figure, the response variable of this experiment (i.e., the normality of blood sugar) is tested, and the QQ graph is drawn to determine whether the response variable obeys the normal distribution. The conclusion is that blood sugar obeys the normal distribution, so a linear model can be used.

[0159] The basic form of the linear mixed model for this experiment is:

[0160] General form:

[0161] ,

[0162] ,

[0163] Among them, Y i represents the response vector of individual i, X i is the known covariate matrix, β is the coefficient of the fixed effect, Z i represents the random effects design matrix for individual i, b i is the individual random effect, is a random residual vector, m is a positive integer;

[0164] represents the random effect b i It obeys q-dimensional normal distribution, with mean 0 and covariance matrix G;

[0165] The residual vector of individual i is n i dimensional normal distribution, where the mean is 0 and the covariance matrix is ​​R i , n i is the number of observations of individual i;

[0166] (2) Determine the random effects

[0167] In the longitudinal data above, age and whether exercised do not change over time, while BMI and BG change over time. Although the measurement time points remain consistent during the experiment, for each experimental individual, the measurement occurs at different weeks of pregnancy, which is very important for studying the effects of exercise on pregnant women, so consider introducing the week of pregnancy as a random effect.

[0168] (3) Determine the composition of the linear mixed model by understanding and analyzing the variables:

[0169] ,

[0170] in, represents the response variable of individual i at time j, represents the fixed effect intercept, is the coefficient of the corresponding fixed effect, indicating the influence of the fixed effect on the response variable. represents the random effect intercept of individual i, is the random effect coefficient, which indicates the influence of random effect on the response variable. is the random residual of individual i at measurement time j; age i is the age of the i-th pregnant woman; time ij is the number of weeks of pregnancy of the i-th pregnant woman at the j-th time; S i Indicates whether the i-th pregnant woman exercises, 1 for exercise and 0 for no exercise; BMI ij is the weight index of the i-th pregnant woman at the j-th time;

[0171] (4) Model fitting: The SCAD penalty function based on the restricted maximum likelihood function is used to simultaneously select and estimate the model parameters.

[0172] (5) The model estimation results are shown in Table 2;

[0173] Table 2 Specific estimated coefficients of fixed effects in the model

[0174]

[0175] (6) The SCAD penalized likelihood estimate for age is 0.01, so the effect of this variable can be ignored.

[0176] like Figure 6 As shown in the figure, it shows the changes of individual blood sugar over time, with the horizontal axis representing time (in weeks) and the vertical axis representing blood sugar level. Blood sugar will increase as the pregnancy time gets longer, but adding exercise will change this situation, and the blood sugar will obviously decrease.

[0177] Since the experiment measured blood and biochemical indicators at multiple time points, these time points constituted a hierarchical structure of the data, and this model also had multi-level characteristics. In addition, the model was also suitable for processing continuous variables such as large platelet ratio and mean corpuscular hemoglobin concentration.

[0178] In summary, the present invention not only improves the accuracy and efficiency of exercise load detection and analysis, but also overcomes the limitations of the prior art in processing repeated measurement data, individual differences and data missing, and has significant technical progress and application value.

[0179] Example 4

[0180] like Figure 7As shown, based on the same application concept, an embodiment of the present application also provides a motion load process detection device based on a linear mixed model, comprising at least one processor, and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any one of the above-mentioned motion load process detection methods based on a linear mixed model.

[0181] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), disks or optical disks, etc. Various media that can store program codes.

[0182] When the above-mentioned integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0183] Finally, it should be noted that the embodiments described in detail above are only the best practices of the invention and cannot be used to limit the scope of rights of the invention. Equivalent replacement of the technical solutions recorded in the aforementioned embodiments does not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the invention, and they should all be included in the scope of the claims and description of the invention.

Claims

1. A motion load process detection method based on a linear mixed model, characterized in that: The following steps are involved: S1: Collect physiological index data by performing multiple measurements on the same group of individuals at multiple time points, and construct longitudinal data based on the physiological index data; the longitudinal data includes the response variable, the covariate of the response variable, and the repeated measurement values ​​of the response variable at different times; S2: Linear mixed models were constructed based on longitudinal data; S3: Analyze longitudinal data using linear mixed models to obtain fixed effect estimates, random effect estimates, and covariance matrices for individuals at different exercise stages; the fixed effect estimates are used to describe the average impact of all individuals; the random effect estimates and covariance matrix are used to quantify changes in physiological load for individuals at different exercise stages; The specific steps of using the random effect estimate and covariance matrix to quantify the changes in individual physiological load at different exercise stages are: Linear mixed models were used to compare random effect estimates of different individuals to identify individual differences in physiological indicators; Analyze the correlation between different random effect independent variables in the covariance matrix; Based on the fixed effect estimate, random effect estimate and covariance matrix, the physiological load change curves of different individuals at different time points before and after exercise were drawn; The calculation formula of the linear mixed model is: Where Y is the response variable vector of dimension n×1; X is the fixed effect design matrix of dimension n×p; β is the fixed effect parameter vector of dimension p×1; Z is the random effect design matrix of dimension n×q; B is the random effect vector of dimension q×1; is an n×1 error term vector, where p and q are positive integers; The calculation formula for the correlation between different random effect independent variables in the covariance matrix is: (Y,B)~N((Xβ,0),(σ²ICov(Y,B)Cov(B,Y)ΣB)) Where Y is the n×1 response variable vector, B is the q×1 random effect vector, Xβ represents the effect of fixed effects on the response variable, σ 2 I is the covariance matrix of the response variable Y, Cov(Y,B) and Cov(B,Y) are the covariance matrices between Y and B, and ΣB is the covariance matrix of the random effect B.

2. A method for detecting a motion load process based on a linear mixed model as claimed in claim 1, characterized in that: The specific steps of S2 are: S11: Preprocess the physiological index data to check whether the response variables in the longitudinal data conform to the normal distribution. If so, use the linear mixed model; S12: Determine random effects and fixed effects based on longitudinal data to obtain fixed effect independent variables and random effect independent variables; S13: A linear mixed model was constructed based on random effect variables and fixed effect variables.

3. The method for detecting a motion load process based on a linear mixed model as claimed in claim 1, characterized in that: The specific steps of S3 are: The parameters of the linear mixed model are selected and estimated simultaneously using the restricted maximum likelihood estimation method or the Bayesian method; Check whether the residual distribution of the linear mixed model meets the normality assumption and verify the goodness of fit of the model; The fixed effect estimates, random effect estimates and covariance matrices of individuals in different exercise stages were obtained.

4. A method for detecting a motion load process based on a linear mixed model as claimed in claim 3, characterized in that: The steps of the restricted maximum likelihood estimation method include: Define the log-likelihood function for the linear mixed model; The SCAD penalty term is introduced to optimize the log-likelihood function; The log-likelihood function is solved by a numerical optimization algorithm to obtain the estimated values ​​of the fixed effects by restricted maximum likelihood estimation.

5. A motion load process detection device based on a linear mixed model, characterized in that: comprising at least one processor, and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a motion load process detection method based on a linear mixed model as described in any one of claims 1 to 4.

Images