Insulin resistance early detection method, system, equipment and medium

Abstract

The invention discloses an insulin resistance early detection method, system, equipment and medium, and the method comprises the following steps: screening an obtained insulin resistance potential factor and an insulin resistance result corresponding to a first to-be-detected sample individual to obtain a key detection factor; different to-be-screened neural network models are evaluated to determine a to-be-optimized detection model, and the evaluation process is designed to perform screening according to feedback of the different to-be-screened neural network models in clinical prediction. And based on the key detection factor and the insulin resistance result, carrying out interpretability analysis on the to-be-optimized detection model to obtain a target insulin resistance detection model, and applying the target insulin resistance detection model to actual insulin resistance early detection. The method provided by the invention is low in detection cost, simple in operation steps and high in detection accuracy, and the insulin resistance risk is identified in an early stage by analyzing conventional physical examination data.

Description

Technical Field

[0001] The present invention relates to the technical field of disease detection, and in particular to an early detection method, system, equipment and medium for insulin resistance. Background Art

[0002] Insulin resistance (IR) is an important feature of type 2 diabetes and a precursor to many metabolic diseases. Insulin resistance is manifested as a weakened cellular response to insulin, which in turn leads to dysregulation of glucose and lipid metabolism, which may eventually develop into type 2 diabetes, atherosclerosis, cardiovascular disease and other metabolic complications, and even cancer. Long-term insulin resistance can also significantly increase the risk of cardiovascular disease and all-cause mortality. Early diagnosis and intervention of insulin resistance can prevent type 2 diabetes, reduce the risk of cardiovascular disease, improve metabolic syndrome, enhance quality of life, delay the occurrence of complications, improve health prognosis, and reduce future medical expenses and resource consumption.

[0003] At present, direct detection of insulin resistance is mainly aimed at diseases with insulin resistance characteristics such as type 2 diabetes, atherosclerosis and cardiovascular diseases. The detection methods mainly include hyperinsulinemia-euglycemia clamp technology and intravenous glucose tolerance test. Hyperinsulinemia-normoglycemia clamp technology is considered to be the gold standard for measuring insulin sensitivity, but due to its complexity and high cost, it is mainly used in scientific research and is not currently promoted in clinical practice. Intravenous glucose tolerance test is to test blood sugar levels after oral glucose to assess insulin sensitivity. However, this method is very time-consuming and patients need to fast, and the detection process is inconvenient.

[0004] It can be seen that how to solve the high cost, complicated operation and inconvenient detection process of the existing insulin resistance detection has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the invention

[0005] The present invention provides a method, system, device and medium for early detection of insulin resistance to solve the technical problems of high cost, complex operation and inconvenient detection process in existing insulin resistance detection, and achieve the effect of early prediction of insulin resistance in a simple, efficient, high-accuracy and low-cost manner.

[0006] In a first aspect, the present invention provides a method for early detection of insulin resistance, the method comprising:

[0007] Obtaining the insulin resistance potential factor and the insulin resistance result corresponding to the first sample individual to be tested, screening the insulin resistance potential factor according to the influence weight of the insulin resistance potential factor on the insulin resistance result, and obtaining the key detection factor;

[0008] Evaluating different neural network models to be screened to determine a detection model to be optimized, wherein the evaluation process is designed to screen according to feedback from different neural network models to be screened in clinical prediction;

[0009] Based on the key detection factors and the insulin resistance results, an interpretability analysis is performed on the detection model to be optimized to obtain a target insulin resistance detection model;

[0010] In the actual detection process, the real-time detection data corresponding to the key detection factors of the individual to be detected is input into the target insulin resistance detection model for processing to obtain the insulin resistance detection result of the individual to be detected.

[0011] Preferably, the step of screening the potential insulin resistance factors according to the weight of their influence on the insulin resistance results to obtain key detection factors comprises:

[0012] Training a pre-selected LASSO regression model according to the insulin resistance potential factors and the insulin resistance results, and determining an optimal regularization parameter of the LASSO regression model;

[0013] The insulin resistance potential factor is input into the LASSO regression model using the optimal regularization parameter to obtain the coefficient corresponding to the insulin resistance potential factor, and the insulin resistance potential factor whose coefficient is not zero is selected as the key detection factor.

[0014] Preferably, the key detection factors include: age, body mass index, waist circumference, fasting blood glucose, glycosylated hemoglobin, triglycerides and high-density lipoprotein cholesterol.

[0015] Preferably, the evaluating of different neural network models to be screened to determine the detection model to be optimized includes:

[0016] Selecting several machine learning algorithms to construct several different neural network models to be screened, and based on the key detection factors and the insulin resistance results, sequentially using grid search and ten-fold cross validation to train and optimize the hyperparameters of several neural network models to be screened;

[0017] According to the key detection factors and the insulin resistance results, a functional evaluation is performed on the optimized neural network models to be screened to obtain result values ​​of functional evaluation parameters, and according to the result values, a plurality of neural network models to be screened are screened to obtain a specific number of detection models; the functional evaluation parameters include: area under the characteristic curve, accuracy, sensitivity, specificity, positive detection value, negative detection value, F1 score, Kappa statistic, Brier score and calibration graph;

[0018] A clinical application evaluation is performed on a specific number of the detection models through a clinical decision curve, and the detection model that is optimal in the clinical application evaluation is used as the detection model to be optimized.

[0019] Preferably, the several machine learning algorithms include: extreme gradient boosting, random forest, logistic regression, support vector machine, H2O deep neural network in R language, Gaussian Bayes classifier, k-nearest neighbor algorithm and decision tree.

[0020] Preferably, the detection model to be optimized is constructed based on an extreme gradient enhancement algorithm.

[0021] Preferably, based on the key detection factors and the insulin resistance results, the interpretability analysis of the detection model to be optimized is performed to obtain a target insulin resistance detection model, including:

[0022] Inputting the key detection factors, the insulin resistance results and the detection model to be optimized into a SHAP interpretable analysis model to obtain a SHAP value of each key detection factor for the insulin resistance result;

[0023] Optimizing the detection model to be optimized based on the SHAP value to obtain an optimized optimal detection model;

[0024] The optimized optimal detection model is externally verified using a pre-selected second test sample individual. If the optimal detection model fails the external verification, the detection model is reconstructed, evaluated and optimized until the optimal detection model passes the external verification. The optimal detection model that passes the external verification is used as the target insulin resistance detection model.

[0025] In a second aspect, the present invention also provides an early detection system for insulin resistance, the system comprising: a key detection factor screening unit, a model evaluation unit, a model optimization unit and a model detection unit;

[0026] The key detection factor screening unit is used to obtain the insulin resistance potential factor and the insulin resistance result corresponding to the first sample individual to be tested, and screen the insulin resistance potential factor according to the influence weight of the insulin resistance potential factor on the insulin resistance result to obtain the key detection factor;

[0027] The model evaluation unit is used to evaluate different neural network models to be screened to determine the detection model to be optimized, wherein the evaluation process is designed to screen according to the feedback of different neural network models to be screened in clinical prediction;

[0028] The model optimization unit is used to perform an interpretability analysis on the detection model to be optimized based on the key detection factors and the insulin resistance results to obtain a target insulin resistance detection model;

[0029] The model detection unit is used to input the real-time detection data corresponding to the key detection factors of the individual to be detected into the target insulin resistance detection model for processing during the actual detection process to obtain the insulin resistance detection result of the individual to be detected.

[0030] In a third aspect, the present invention also provides a computer device, comprising a memory, a processor and a transceiver, which are connected via a bus; the memory is used to store a set of computer program instructions and data, and transmit the stored data to the processor, and the processor executes the program instructions stored in the memory to execute the above-mentioned method for early detection of insulin resistance.

[0031] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed, the above-mentioned method for early detection of insulin resistance is implemented.

[0032] The present invention provides a method, system, device and medium for early detection of insulin resistance. Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following:

[0033] (1) Simple operation and low detection cost. The method for early detection of insulin resistance provided by the present invention utilizes conventional physical examination data and an early detection model of insulin resistance to achieve early detection of insulin resistance, with low detection cost and simple operation steps.

[0034] (2) The detection accuracy is high. The extreme gradient enhancement algorithm with balanced functionality and optimal clinical application is selected to construct the target insulin resistance detection model. Combined with interpretable analysis, it can not only obtain accurate prediction results, but also understand the contribution of each key detection factor to the test results, thereby improving the transparency and trust of the model.

[0035] (3) It overcomes the technical bias of only testing for insulin resistance after the onset of type 2 diabetes, atherosclerosis, and cardiovascular diseases, and is able to make early predictions for healthy individuals and identify the risk of insulin resistance at an early stage by analyzing routine physical examination data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a schematic diagram of the steps of an early detection method for insulin resistance provided by a preferred embodiment of the present invention;

[0037] Figure 2It is a schematic diagram of the steps of a method for screening key detection factors provided by a preferred embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of the steps of a method for evaluating different neural network models to be screened provided by a preferred embodiment of the present invention;

[0039] Figure 4 is the result value of the functional evaluation parameter of different detection models provided by a preferred embodiment of the present invention;

[0040] Figure 5 It is a schematic diagram of the method steps for performing interpretability analysis on the detection model to be optimized provided by a preferred embodiment of the present invention;

[0041] Figure 6 It is a schematic diagram of an externally validated ROC curve, a calibration curve and a clinical decision curve provided by a preferred embodiment of the present invention;

[0042] Figure 7 This is a schematic diagram of the structure of an early detection system for insulin resistance provided by a preferred embodiment of the present invention;

[0043] Figure 8 It is a schematic diagram of a computer device provided by a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following is a detailed explanation of the implementation mode of the present invention in conjunction with the accompanying drawings. The embodiments are provided only for illustrative purposes and cannot be understood as limiting the present invention. The accompanying drawings are only for reference and illustration purposes and do not constitute a limitation on the scope of patent protection of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In the description of the present invention, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more.

[0045] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0046] In the description of the present invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0047] See also Figure 1 In an embodiment of the present invention, a method for early detection of insulin resistance is provided, the method comprising:

[0048] S1. Obtain the insulin resistance potential factors and insulin resistance results corresponding to the first sample individual to be tested, screen the insulin resistance potential factors according to the influence weights of the insulin resistance potential factors on the insulin resistance results, and obtain key detection factors.

[0049] S2. Evaluate different neural network models to be screened to determine the detection model to be optimized, wherein the evaluation process is designed to perform screening based on the feedback of different neural network models to be screened in clinical prediction.

[0050] S3. Based on the key detection factors and the insulin resistance results, an explainability analysis is performed on the detection model to be optimized to obtain a target insulin resistance detection model.

[0051] S4. During the actual detection process, the real-time detection data corresponding to the key detection factors of the individual to be detected is input into the target insulin resistance detection model for processing to obtain the insulin resistance detection result of the individual to be detected.

[0052] Existing insulin resistance tests are mostly limited to people with type 2 diabetes, atherosclerosis, cardiovascular diseases, and non-diabetic people without considering other complications. However, people with type 2 diabetes, atherosclerosis, cardiovascular diseases, etc. often already have insulin resistance. It is necessary to detect insulin resistance in advance when the individual does not have type 2 diabetes, atherosclerosis, cardiovascular diseases, etc., in order to achieve the effect of early prevention.

[0053] In a preferred embodiment of the present invention, in the data collection stage, the insulin resistance potential factors and insulin resistance results corresponding to individuals without diabetes, atherosclerosis and cardiovascular diseases are obtained. The present invention collects data from the National Health and Nutrition Examination Survey (NHANES1999-2012) of the United States, including 6791 individuals without diabetes, atherosclerosis and cardiovascular diseases, wherein cardiovascular diseases include: hypertension, coronary heart disease, myocardial infarction, heart failure and stroke. The above data set is used as the first sample individual to be tested for the construction, evaluation and optimization of the detection model.

[0054] In a preferred embodiment of the present application, the collected data also includes a second sample individual to be tested, which is used to verify the recognition and detection ability and stability of the selected optimal detection model on external data, and to verify the reliability and generalization of the optimal detection model on different data sets. The second sample individual to be tested collects data from the Korean National Health and Nutrition Examination Survey (KNHANES2007-2021), including 1,462 individuals without diabetes, atherosclerosis, cardiovascular diseases, etc.

[0055] The data of the first sample to be tested and the second sample to be tested are preprocessed to remove outliers and noise to ensure data quality. The data are standardized to eliminate the influence of different dimensions and ensure consistency when the data is input into the model.

[0056] In a preferred embodiment of the present invention, 23 features are selected from the first sample individual to be detected and the second sample individual to be detected as potential factors of insulin resistance. Potential factors of insulin resistance include demographic factors, including age, gender, race and education level, clinical measurement factors, including body mass index (BMI), waist circumference, blood pressure, fasting blood glucose (FBG), glycosylated hemoglobin (HbA1c), triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), renal function markers (creatinine), liver markers (aspartate aminotransferase, alanine aminotransferase, albumin), hemoglobin, and dietary intake of energy, protein and carbohydrates. Although the above potential factors of insulin resistance cover a wide range, they will cause difficulties in obtaining test data in actual detection and complicated operation process. And most of the many potential factors of insulin resistance do not contribute much to insulin resistance, but a large amount of characteristic data affects each other, which will reduce the accuracy of the detection. Therefore, it is necessary to select a suitable number of key detection factors, which can not only ensure the accuracy of insulin resistance detection, but also do not need to collect a large amount of data in the actual detection process, and the operation is simple. In a preferred embodiment of the present invention, the insulin resistance potential factors are screened according to the weights of their influence on the insulin resistance results to obtain key detection factors. Specifically, the least absolute shrinkage and selection operator (LASSO) algorithm is used to screen the insulin resistance potential factors to obtain key detection factors, such as Figure 2 As shown, the following steps are included:

[0057] S101. Training a pre-selected LASSO regression model according to the insulin resistance potential factors and the insulin resistance results, and determining an optimal regularization parameter of the LASSO regression model.

[0058] S102, inputting the insulin resistance potential factor into the LASSO regression model using the optimal regularization parameter to obtain the coefficient corresponding to the insulin resistance potential factor, and selecting the insulin resistance potential factor whose coefficient is not zero as the key detection factor.

[0059] First, the 23 pre-processed potential factors of insulin resistance in the first sample individual to be tested and the second sample individual to be tested are constructed into a feature matrix x, and the corresponding individual's insulin resistance results are used as the objective function variable y. The feature matrix and objective function variables of each individual in the first sample individual to be tested and the second sample individual to be tested constitute the data set to be screened, and the potential factors of insulin resistance are screened using the LASSO regression model. The core of the LASSO regression model is to compress the coefficients of the potential factors of insulin resistance to achieve the purpose of variable selection and complexity adjustment, thereby improving the prediction accuracy and explanatory power of the LASSO regression model. In high-dimensional space, the traditional least squares regression algorithm will have problems such as difficulty in variable selection and model overfitting. The LASSO regression model introduces an adjustment parameter λ to penalize the absolute value of the coefficient, forcing some unimportant coefficient values ​​to become zero, thereby achieving the effect of automatically selecting important potential factors of insulin resistance. The loss function of the LASSO regression model is expressed as follows:

[0060]

[0061] Wherein, j represents the jth insulin resistance potential factor, i represents the ith individual, n represents the total number of the first sample individual to be tested and the second sample individual to be tested, p represents the number of insulin resistance potential factors, y i represents the insulin resistance result of the i-th individual, β j represents the coefficient vector of the jth insulin resistance potential factor, x ij represents the jth insulin resistance potential factor of the ith individual, λ represents the regularization parameter, β j | denotes β j The L1 norm of .

[0062] λ in the LASSO regression model is a key parameter, and its value directly affects the performance of the LASSO regression model. When λ is 0, the LASSO regression model degenerates into an ordinary least squares regression algorithm. As λ increases, more and more coefficients are compressed to zero, which helps to screen potential factors of insulin resistance and reduce the complexity of the model. However, if λ is too large, the LASSO regression model may be too simple, thus affecting the LASSO regression model's ability to screen potential factors of insulin resistance. Therefore, choosing a suitable λ value is the key to achieving the best performance of the LASSO regression model.

[0063] In a preferred embodiment of the present invention, the optimal regularization parameter is determined by cross-validation, and the data set to be screened is divided into a first training set and a first validation set. A series of different λ values ​​are used, the first training set is used to train the LASSO regression model, and the first validation set is used to evaluate the model performance corresponding to each λ value, and the λ value with the smallest validation error is selected as the optimal regularization parameter. In the embodiment of the present invention, the optimal regularization parameter is 0.0257.

[0064] Furthermore, the insulin resistance potential factors are input into the LASSO regression model with the optimal regularization parameter. The LASSO regression model will automatically reduce the coefficients of some insulin resistance potential factors to 0, and finally select the insulin resistance potential factors with coefficients not equal to 0 as the key detection factors, thereby realizing the screening of insulin resistance potential factors. The key detection factors finally screened in the preferred embodiment of the present invention include: age, body mass index, waist circumference, fasting blood sugar, glycosylated hemoglobin, triglycerides and high-density lipoprotein cholesterol.

[0065] In a preferred embodiment of the present invention, a LASSO regression model is used to select 8 key detection factors from 23 potential factors of insulin resistance, reduce the interaction between features, and select key detection factors with a significant impact on insulin resistance. While improving the accuracy of subsequent early detection of insulin resistance, in the actual detection process, there is no need to obtain a large amount of real-time detection data, which reduces the detection cost, simplifies the operation steps, and enables application in a wider population.

[0066] After selecting the key detection factors, different neural network models to be screened are evaluated to determine the detection model to be optimized, such as Figure 3 As shown, the following steps are included:

[0067] S201. Select several machine learning algorithms to construct several different neural network models to be screened. Based on the key detection factors and the insulin resistance results, grid search and ten-fold cross validation are used in turn to train and optimize the hyperparameters of several of the neural network models to be screened.

[0068] S202. Functionally evaluate the optimized neural network models to be screened according to the key detection factors and the insulin resistance results to obtain result values ​​of functional evaluation parameters, and screen the neural network models to be screened according to the result values ​​to obtain a specific number of detection models; the functional evaluation parameters include: area under the characteristic curve, accuracy, sensitivity, specificity, positive detection value, negative detection value, F1 score, Kappa statistic, Brier score and calibration chart.

[0069] S203, performing clinical application evaluation on a specific number of the detection models through a clinical decision curve, and using the detection model that is optimal in the clinical application evaluation as the detection model to be optimized.

[0070] In a preferred embodiment of the present invention, in order to achieve the screening of the optimal detection model, the pre-selected machine learning algorithms include: extreme gradient boosting (XGBoost), random forest (RF), logistic regression (LR), support vector machine (SVM), H2O deep neural network (H2ODNN) under R language, Gaussian Bayes classifier (GNB), k-nearest neighbor algorithm (KNN) and decision tree (DT). According to the above 8 machine learning algorithms, the corresponding 8 neural network models to be screened are constructed.

[0071] A first sample data set to be tested is constructed based on the key detection factors and insulin resistance results of each individual in the first sample individuals to be tested, and the first sample data set to be tested is divided into a second training set, a second validation set and a second test set in a ratio of 70:15:15.

[0072] For the above 8 detection models, based on the first sample data set to be detected, grid search and ten-fold cross validation are used in turn to train and optimize their hyperparameters to improve the generalization performance and robustness of the neural network model to be screened. The grid search step includes: defining the hyperparameter range and performing grid search. The definition of the hyperparameter range defines a reasonable hyperparameter search space for each detection model. The second training set is used for grid search. The grid search will perform an exhaustive search in the predefined hyperparameter space to find the initial optimal hyperparameter combination. The initial optimal hyperparameter combination is further verified and optimized by ten-fold cross validation. Ten-fold cross validation further verifies and optimizes the initial optimal hyperparameters obtained by grid search. Ten-fold cross validation is performed using the validation set, the performance index of each cross validation is calculated, and the average value is taken as the final performance index of each neural network model to be screened.

[0073] In a preferred embodiment of the present invention, a preliminary optimal hyperparameter combination is obtained by grid search, and then a ten-fold cross validation is performed. On the basis of the preliminary optimal hyperparameter combination, the preliminary optimal hyperparameter combination is verified and optimized to obtain the optimal hyperparameter combination as the hyperparameter of each neural network model to be screened. Ten-fold cross validation can better evaluate the generalization performance of the model, reduce the risk of overfitting, and more accurately evaluate the performance of each neural network model to be screened on the data, thereby improving the generalization performance and robustness of the neural network model to be screened.

[0074] After the hyperparameters of each neural network model to be screened are determined, the second test set is used to perform functional evaluation on each neural network model to be screened, and the result value of the functional evaluation parameter is obtained. The functional evaluation parameters include: area under the characteristic curve, accuracy, sensitivity, specificity, positive detection value, negative detection value, F1 score, Kappa statistic, Brier score and calibration graph.

[0075] Among them, the characteristic curve (ROC curve) is a graph that plots the relationship between the true positive rate and the false positive rate. Each point on the characteristic curve represents the performance under a threshold. The area under the characteristic curve (AUC) ranges between 0.5 and 1. The closer the AUC is to 1, the better the classification performance of the model. Comparing the AUC values ​​of different neural network models to be screened, the neural network models to be screened with higher AUC values ​​usually have better classification performance.

[0076] Precision is the ratio of correctly predicted samples to the total number of samples. The precision values ​​of different neural network models to be screened are compared. The neural network model to be screened with a higher precision value has better overall performance.

[0077] Sensitivity is the proportion of positive samples correctly identified. By comparing the sensitivity values ​​of different neural network models to be screened, the neural network model to be screened with a high sensitivity value is more capable of identifying positive samples.

[0078] Specificity is the proportion of negative samples correctly identified. By comparing the specificity values ​​of different neural network models to be screened, the neural network models to be screened with higher specificity values ​​are better able to identify negative samples.

[0079] The positive prediction value is the proportion of samples predicted to be positive that are actually positive. By comparing the positive prediction values ​​of different neural network models to be screened, the positive samples detected by the neural network model to be screened with a high positive prediction value are more accurate.

[0080] The negative prediction value is the proportion of samples predicted to be negative that are actually negative. By comparing the negative prediction values ​​of different neural network models to be screened, the negative samples predicted by the neural network model to be screened with a high negative prediction value are more accurate.

[0081] The F1 score is the harmonic mean of precision and sensitivity. By comparing the F1 scores of different neural network models to be screened, the neural network models to be screened with high F1 scores have achieved a better balance between precision and sensitivity.

[0082] The Kappa statistic is used to measure the consistency between the detection model prediction and the actual label. The influence of random consistency is taken into account and the Kappa values ​​of different neural network models to be screened are compared. The higher the Kappa value, the better the consistency of the neural network model to be screened.

[0083] The Brier score is used to measure the accuracy of the predicted probability. It is the mean square error between the predicted probability and the actual result. The Brier scores of different neural network models to be screened are compared. The lower the Brier score, the more accurate the predicted probability of the neural network model to be screened.

[0084] The calibration plot is used to compare the relationship between the predicted probability and the observed results. Ideally, the predicted probability should be consistent with the observed probability. By observing the calibration plot, the higher the match between the predicted probability and the actual result, the better the performance of the neural network model to be screened.

[0085] Furthermore, a number of neural network models to be screened are screened according to the result values ​​of the functional evaluation parameters to obtain a specific number of detection models.

[0086] Specifically, in a preferred embodiment of the present invention, an optimal value range is set for each functional evaluation parameter, and the result value of each functional evaluation parameter corresponding to each neural network model to be screened is compared with the corresponding optimal value range. If the result value is within the optimal value range, it indicates that the neural network model to be screened meets the requirements. If the result value is not within the optimal value range, it indicates that the neural network model to be screened does not meet the requirements, and the neural network model to be screened is eliminated to perform a preliminary selection on the neural network model to be screened, so as to obtain the neural network model to be screened after the preliminary selection.

[0087] For each functional evaluation parameter, the neural network models to be screened after the preliminary selection are sorted according to the result value, and each neural network model to be screened after the preliminary selection is assigned a value according to the sorting result. For example, for the area under the characteristic curve, the first neural network model to be screened is assigned a value of 10, the second neural network model to be screened is assigned a value of 9, and the third neural network model to be screened is assigned a value of 8, and so on for each neural network model to be screened after the preliminary selection. Therefore, each neural network model to be screened after the preliminary selection has 10 assignments, which are the area under the characteristic curve, accuracy, sensitivity, specificity, positive detection value, negative detection value, F1 score, Kappa statistic, Brier score and calibration chart. Finally, the assignments of each neural network model to be screened after the preliminary selection are added to obtain the final score of the neural network model to be screened after the preliminary selection. The neural network models to be screened after the preliminary selection are screened according to the final scores of each neural network model to be screened after the preliminary selection to obtain a specific number of detection models that meet the requirements.

[0088] After obtaining a specific number of detection models, the clinical application of the specific number of detection models is evaluated through the clinical decision curve. The clinical decision curve (Decision Curve Analysis, DCA) evaluates the clinical utility of the detection model by using the net benefit of the screened detection model under different threshold probabilities, and selects the detection model with the best feedback in the clinical application evaluation as the detection model to be optimized. Through the above-mentioned method of selecting the detection model to be optimized, the detection model to be optimized has balanced performance in various functionalities and the highest clinical utility.

[0089] In the preferred embodiment of the present application, the detection model to be optimized is finally selected based on the extreme gradient enhancement algorithm. Figure 4 The result values ​​of the functional evaluation parameters of the neural network model to be screened constructed by the 8 machine learning algorithms selected in this case. Figure 4It can be seen that the area under the characteristic curve of the neural network model to be screened constructed by the extreme gradient boosting algorithm is the highest in both the training set and the validation set, indicating that the neural network model to be screened constructed by the extreme gradient boosting algorithm has good ability in distinguishing positive and negative samples. The accuracy of the neural network model to be screened constructed by the extreme gradient boosting algorithm is the highest in the training set and the highest in the validation set, indicating that the neural network model to be screened constructed by the extreme gradient boosting algorithm has a strong ability to correctly classify. The precision of the neural network model to be screened constructed by the extreme gradient boosting algorithm is the highest in the training set and is higher in the validation set, indicating that the neural network model to be screened constructed by the extreme gradient boosting algorithm has a higher accuracy in predicting positive samples. The sensitivity of the neural network model to be screened constructed by the extreme gradient boosting algorithm is high in both the training set and the validation set, indicating that the neural network model to be screened constructed by the extreme gradient boosting algorithm has a strong ability to capture positive samples. Although the specificity of the neural network model to be screened constructed by the extreme gradient boosting algorithm is low in both the training set and the validation set, the comprehensive performance of the neural network model to be screened constructed by the extreme gradient boosting algorithm is good. The positive predictive value of the neural network model to be screened constructed by the extreme gradient boosting algorithm was the highest in the training set and higher in the validation set, indicating that the neural network model to be screened constructed by the extreme gradient boosting algorithm had a higher accuracy in predicting positive results. The negative predictive value of the neural network model to be screened constructed by the extreme gradient boosting algorithm was higher in both the training set and the validation set, indicating that the neural network model to be screened constructed by the extreme gradient boosting algorithm had a higher accuracy in predicting negative results. The F1 score of the neural network model to be screened constructed by the extreme gradient boosting algorithm was the highest in the training set and higher in the validation set, indicating that the neural network model to be screened constructed by the extreme gradient boosting algorithm had a better overall performance in processing unbalanced data. The Kappa statistic of the neural network model to be screened constructed by the extreme gradient boosting algorithm was the highest in the training set and higher in the validation set, indicating that the prediction results of the neural network model to be screened constructed by the extreme gradient boosting algorithm were more consistent with the actual situation. The Brier score of the neural network model to be screened constructed by the extreme gradient boosting algorithm was the lowest in the training set and lower in the validation set, indicating that the prediction probability of the neural network model to be screened constructed by the extreme gradient boosting algorithm was more accurate.

[0090] From the above analysis, it can be seen that the neural network model to be screened constructed using the extreme gradient boosting algorithm performs well in multiple key indicators, especially in AUC, accuracy, precision, sensitivity, F1 score and Brier score. Although the specific result value is low, this does not significantly affect its overall performance in practical applications. The detection model to be optimized selected by the preferred embodiment of the present invention can provide stable, comprehensive and accurate insulin resistance prediction, and can provide reliable prediction results in practical medical applications.

[0091] Furthermore, the SHAP interpretable analysis method is used to perform interpretability analysis on the detection model to be optimized, and the SHAP value of each key detection factor on the insulin resistance result is determined, revealing the impact of the key detection factor on the individual insulin resistance result. The detection model to be optimized is optimized based on the SHAP value to obtain the optimized optimal detection model. The second sample data set to be tested is constructed from the key detection factors and insulin resistance results of each individual in the second sample to be tested. Furthermore, the optimized optimal detection model is externally verified through the second sample data set to obtain the target insulin resistance detection model.

[0092] In a preferred embodiment of the present invention, based on key detection factors and insulin resistance results, an interpretability analysis is performed on the detection model to be optimized to obtain a target insulin resistance detection model, such as Figure 5 As shown, the following steps are included:

[0093] S301, inputting the key detection factors, the insulin resistance results and the detection model to be optimized into a SHAP interpretable analysis model to obtain the SHAP value of each key detection factor for the insulin resistance results.

[0094] S302: Optimize the detection model to be optimized based on the SHAP value to obtain an optimized optimal detection model.

[0095] S303. Use a pre-selected second test sample individual to perform external verification on the optimized optimal detection model. If the optimal detection model fails the external verification, reconstruct, evaluate and optimize the detection model until the optimal detection model passes the external verification. The optimal detection model that passes the external verification is used as the insulin resistance detection model.

[0096] In a preferred embodiment of the present invention, the final analysis result of the SHAP interpretable analysis model is that the average SHAP value of fasting blood glucose (FBG) is the highest, which is 0.669, followed by body mass index (BMI), with an average SHAP value of 0.626, triglycerides (TG) of 0.438, age of 0.393, waist circumference of 0.387, glycosylated hemoglobin (HbA1c) of 0.141, and high-density lipoprotein cholesterol (HDL-C) of 0.090. The above average SHAP values ​​indicate the contribution of the corresponding key detection factors to the individual insulin resistance results of a single sample. In addition, the optimal critical values ​​of fasting blood glucose (FBG), body mass index (BMI), triglycerides (TG), age, waist circumference, glycosylated hemoglobin (HbA1c) and high-density lipoprotein cholesterol (HDL-C) were further determined, which are 5.0 mmol / L, 27.6 kg / m 2 , 101.5 mg / dL, 34 years old, 96.8 cm, 5.3% and 67.9 mg / dL. Further, the above interpretable analysis results are theoretically verified to verify whether the interpretable analysis results conform to the theoretical laws, and then the performance of the detection model to be optimized is verified to optimize the detection model to obtain the optimized optimal detection model.

[0097] Furthermore, in order to verify the recognition and prediction ability and stability of the optimal detection model on external data and its reliability and generalization on different data sets, the optimized optimal detection model was externally verified using the second sample data set of the second sample individuals selected in advance. The final verification results were an area under the characteristic curve of 0.822, an accuracy of 0.903, a sensitivity of 0.970, an F1 score of 0.935, a Kappa of 0.317 and a Brier score of 0.117. Figure 6 As shown, it has excellent calibration curve and clinical decision curve. The above verification results show that the optimal detection model has reliable recognition and prediction capabilities and stability, and has good reliability and generalization on different data sets, which is worthy of clinical promotion.

[0098] If the verification result of the optimal detection model on the second sample individual to be tested is not ideal, it means that the selected optimal detection model cannot be verified by external data, and it is necessary to reselect the machine learning algorithm to construct, evaluate and optimize the detection model until the optimal detection model verified on the second sample individual to be tested is obtained and used as the target insulin resistance detection model.

[0099] In the actual testing process, the real-time test data corresponding to the key test factors of the individual to be tested is input into the target insulin resistance test model for processing to obtain the insulin resistance test results of the individual to be tested. For example, the age of an individual is 25 years old, and the BMI is 27.6 kg / m2 , waist circumference: 120 cm, FBG: 5.7 mmol / l, HbA1c: 5.4%, TG: 121 mg / dl and HDL-C: 53 mg / dl. Entering it into the insulin resistance detection model, the IR value obtained was 89.15%, which exceeded the threshold of 31.6%, indicating that the individual had insulin resistance.

[0100] In an embodiment of the present application, the insulin resistance potential factor and insulin resistance result corresponding to the first sample individual to be tested are obtained, and the insulin resistance potential factor is screened according to the weight of the influence of the insulin resistance potential factor on the insulin resistance result to obtain the key detection factor. Different neural network models to be screened are evaluated to determine the detection model to be optimized, wherein the evaluation process is designed to screen according to the feedback of different neural network models to be screened in clinical prediction. Based on the key detection factors and insulin resistance results, the detection model to be optimized is analyzed for interpretability to obtain the target insulin resistance detection model. In the actual detection process, the real-time detection data corresponding to the key detection factors of the individual to be detected is input into the target insulin resistance detection model for processing to obtain the insulin resistance detection result of the individual to be detected. The early detection method of insulin resistance provided by the present invention uses conventional physical examination data and the early detection model of insulin resistance to realize the early detection of insulin resistance, with low detection cost and simple operation steps. Selecting the extreme gradient enhancement algorithm with balanced functionality and optimal clinical application to construct the target insulin resistance detection model, combined with interpretability analysis, can not only obtain accurate prediction results, but also understand the contribution of each key detection factor to the test results, and improve the transparency and trust of the model. It overcomes the technical bias of only testing insulin resistance after diseases such as type 2 diabetes, atherosclerosis and cardiovascular diseases have occurred, and is able to make early predictions for healthy individuals and identify the risk of insulin resistance at an early stage by analyzing routine physical examination data.

[0101] Accordingly, if Figure 7 As shown, based on an early detection method for insulin resistance, an embodiment of the present invention further provides an early detection system for insulin resistance, including: a key detection factor screening unit 1, a model evaluation unit 2, a model optimization unit 3 and a model detection unit 4;

[0102] The key detection factor acquisition unit 1 is used to obtain the insulin resistance potential factor and the insulin resistance result corresponding to the first sample individual to be tested, and screen the insulin resistance potential factor according to the influence weight of the insulin resistance potential factor on the insulin resistance result to obtain the key detection factor.

[0103] The model evaluation unit 2 is used to evaluate different neural network models to be screened to determine the detection model to be optimized, wherein the evaluation process is designed to perform screening based on the feedback of different neural network models to be screened in clinical prediction.

[0104] The model optimization unit 3 is used to perform an interpretability analysis on the detection model to be optimized based on the key detection factors and the insulin resistance results, so as to obtain a target insulin resistance detection model.

[0105] The model detection unit 4 is used to input the real-time detection data corresponding to the key detection factors of the individual to be detected into the target insulin resistance detection model for processing during the actual detection process to obtain the insulin resistance detection result of the individual to be detected.

[0106] For the specific definition of an early detection system for insulin resistance, please refer to the above-mentioned definition of an early detection method for insulin resistance, which will not be repeated here. A person of ordinary skill in the art will appreciate that the various modules and steps described in conjunction with the embodiments disclosed in the present invention can be implemented in hardware, software, or a combination of both. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0107] like Figure 8 As shown, an embodiment of the present invention provides a computer device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the above-mentioned early detection method of insulin resistance are implemented, for example Figure 1 Steps S1 to S4 described in .

[0108] Those skilled in the art will understand that the Figure 8 It is only an example of a computer device and does not constitute a limitation of the computer device. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.

[0109] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer device, and uses various interfaces and lines to connect various parts of the entire computer device.

[0110] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the computer device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0111] Wherein, if the module integrated in the computer device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0112] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0113] Accordingly, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to perform the steps in the method for early detection of insulin resistance in the above embodiment, for example Figure 1 Steps S1 to S4 described in .

[0114] The present embodiment provides an early detection method, system, computer device and storage medium for insulin resistance, which is aimed at the technical problems that the existing insulin resistance detection has high cost, complex operation and low accuracy, as well as the limitations of the detection object, and cannot achieve early detection. The present invention obtains the insulin resistance potential factor and insulin resistance result corresponding to the first sample individual to be tested, and screens the insulin resistance potential factor according to the weight of the influence of the insulin resistance potential factor on the insulin resistance result to obtain the key detection factor. Different neural network models to be screened are evaluated to determine the detection model to be optimized, wherein the evaluation process is designed to screen according to the feedback of different neural network models to be screened in clinical prediction. Based on the key detection factors and insulin resistance results, the detection model to be optimized is analyzed for interpretability to obtain the target insulin resistance detection model. In the actual detection process, the real-time detection data corresponding to the key detection factors of the individual to be detected is input into the target insulin resistance detection model for processing to obtain the insulin resistance detection result of the individual to be detected. The early detection method of insulin resistance provided by the present invention uses conventional physical examination data and the early detection model of insulin resistance to achieve early detection of insulin resistance, with low detection cost and simple operation steps. The extreme gradient enhancement algorithm with balanced functionality and optimal clinical application is selected to construct the target insulin resistance detection model. Combined with interpretable analysis, it can not only obtain accurate prediction results, but also understand the contribution of each key detection factor to the test results, and improve the transparency and trust of the model. It overcomes the technical bias of only performing insulin resistance testing after the occurrence of diseases such as type 2 diabetes, atherosclerosis and cardiovascular diseases, and can make early predictions for healthy individuals. By analyzing routine physical examination data, the risk of insulin resistance can be identified at an early stage.

[0115] The above-mentioned embodiments only express several preferred implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principle of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for early detection of insulin resistance, characterized in that: The method comprises: Obtaining the insulin resistance potential factor and the insulin resistance result corresponding to the first sample individual to be tested, screening the insulin resistance potential factor according to the influence weight of the insulin resistance potential factor on the insulin resistance result, and obtaining the key detection factor; Evaluating different neural network models to be screened to determine a detection model to be optimized, wherein the evaluation process is designed to screen according to feedback from different neural network models to be screened in clinical prediction; Based on the key detection factors and the insulin resistance results, an interpretability analysis is performed on the detection model to be optimized to obtain a target insulin resistance detection model; In the actual detection process, the real-time detection data corresponding to the key detection factors of the individual to be detected is input into the target insulin resistance detection model for processing to obtain the insulin resistance detection result of the individual to be detected.

2. The method for early detection of insulin resistance according to claim 1, characterized in that: The step of screening the potential insulin resistance factors according to the weight of their influence on the insulin resistance results to obtain key detection factors includes: Training a pre-selected LASSO regression model according to the insulin resistance potential factors and the insulin resistance results, and determining an optimal regularization parameter of the LASSO regression model; The insulin resistance potential factor is input into the LASSO regression model using the optimal regularization parameter to obtain the coefficient corresponding to the insulin resistance potential factor, and the insulin resistance potential factor whose coefficient is not zero is selected as the key detection factor.

3. The method for early detection of insulin resistance according to claim 1, characterized in that: The key detection factors include: age, body mass index, waist circumference, fasting blood glucose, glycosylated hemoglobin, triglycerides and high-density lipoprotein cholesterol.

4. The method for early detection of insulin resistance according to claim 1, characterized in that: The different neural network models to be screened are evaluated to determine the detection model to be optimized, including: Selecting several machine learning algorithms to construct several different neural network models to be screened, and based on the key detection factors and the insulin resistance results, sequentially using grid search and ten-fold cross validation to train and optimize the hyperparameters of several neural network models to be screened; According to the key detection factors and the insulin resistance results, a functional evaluation is performed on the optimized neural network models to be screened to obtain result values ​​of functional evaluation parameters, and according to the result values, a plurality of neural network models to be screened are screened to obtain a specific number of detection models; the functional evaluation parameters include: area under the characteristic curve, accuracy, sensitivity, specificity, positive detection value, negative detection value, F1 score, Kappa statistic, Brier score and calibration graph; A clinical application evaluation is performed on a specific number of the detection models through a clinical decision curve, and the detection model that is optimal in the clinical application evaluation is used as the detection model to be optimized.

5. The method for early detection of insulin resistance according to claim 4, characterized in that: The several machine learning algorithms include: extreme gradient boosting, random forest, logistic regression, support vector machine, H2O deep neural network in R language, Gaussian Bayes classifier, k-nearest neighbor algorithm and decision tree.

6. The method for early detection of insulin resistance according to claim 5, characterized in that: The detection model to be optimized is constructed based on the extreme gradient enhancement algorithm.

7. The method for early detection of insulin resistance according to claim 6, characterized in that: The method of performing an interpretability analysis on the detection model to be optimized based on the key detection factors and the insulin resistance results to obtain a target insulin resistance detection model includes: Inputting the key detection factors, the insulin resistance results and the detection model to be optimized into a SHAP interpretable analysis model to obtain a SHAP value of each key detection factor for the insulin resistance result; Optimizing the detection model to be optimized based on the SHAP value to obtain an optimized optimal detection model; The optimized optimal detection model is externally verified using a pre-selected second test sample individual. If the optimal detection model fails the external verification, the detection model is reconstructed, evaluated and optimized until the optimal detection model passes the external verification. The optimal detection model that passes the external verification is used as the target insulin resistance detection model.

8. An insulin resistance detection system, characterized in that: The system comprises: a key detection factor screening unit, a model evaluation unit, a model optimization unit and a model detection unit; The key detection factor screening unit is used to obtain the insulin resistance potential factor and the insulin resistance result corresponding to the first sample individual to be tested, and screen the insulin resistance potential factor according to the influence weight of the insulin resistance potential factor on the insulin resistance result to obtain the key detection factor; The model evaluation unit is used to evaluate different neural network models to be screened to determine the detection model to be optimized, wherein the evaluation process is designed to screen according to the feedback of different neural network models to be screened in clinical prediction; The model optimization unit is used to perform an interpretability analysis on the detection model to be optimized based on the key detection factors and the insulin resistance results to obtain a target insulin resistance detection model; The model detection unit is used to input the real-time detection data corresponding to the key detection factors of the individual to be detected into the target insulin resistance detection model for processing during the actual detection process to obtain the insulin resistance detection result of the individual to be detected.

9. A computer device, characterized in that: The computer device includes a memory, a processor and a transceiver, which are connected via a bus; the memory is used to store a set of computer program instructions and data, and transmit the stored data to the processor, and the processor executes the program instructions stored in the memory to perform the early detection method for insulin resistance as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the method for early detection of insulin resistance according to any one of claims 1 to 7 is implemented.

Images