Method for constructing glucose-metabolism-related disease prediction model, method for constructing non-diabetic subtyping model, and method for predicting postprandial blood glucose

Abstract

The present application belongs to the technical field of artificial intelligence prediction of glucose-metabolism-related diseases, and relates to a method for constructing a glucose-metabolism-related disease prediction model, a method for constructing a non-diabetic subtyping model, and a method for predicating postprandial blood glucose. In the present application, large-scale unlabeled dynamic blood glucose concentration data is used to construct a pre-trained model, and the pre-trained model is fine-tuned on the basis of clinical diagnosis and clinical physiological data, thereby realizing the prediction of glucose-metabolism-related diseases. Secondly, in the present application, dynamic blood glucose feature data is captured on the basis of large-scale CGM data, and non-diabetic dynamic blood glucose distribution data is obtained by means of clustering, thereby realizing the prediction of non-diabetic subtypes of individuals by using individual CGM data and the non-diabetic dynamic blood glucose distribution data. Furthermore, in the present application, individual dynamic blood glucose data obtained by means of continuous glucose monitoring is encoded and embedded into a latent space by means of the pre-trained model to obtain low-dimensional dynamic blood glucose feature data, such that predicted postprandial blood glucose corresponding to dietary structure information can be obtained on the basis of preprandial immediate blood glucose data and the dietary structure information.

Description

Methods for constructing predictive models for glucose metabolism-related diseases, non-diabetic subtyping models, and predicting postprandial blood glucose.

[0001] Related applications

[0002] This application claims priority to the following Chinese patent applications filed on June 21, 2024: application number 2024108113834, entitled "Constructing a Predictive Model for Glucose Metabolism-Related Diseases, and a Method and Apparatus for Predicting Glucose Metabolism-Related Diseases"; application number 202410811375X, entitled "Constructing a Predictive Model for Postprandial Blood Glucose, and a Method and Apparatus for Predicting Postprandial Blood Glucose"; and application number 2024108113694, entitled "Constructing a Non-Diabetes Subtype Model and a Method and Apparatus for Predicting Non-Diabetes Subtypes". The entire contents of these patent applications are incorporated herein by reference. Technical Field

[0003] This application belongs to the field of artificial intelligence prediction technology related to glucose metabolism-related diseases. Specifically, it relates to a method for constructing a prediction model for glucose metabolism-related diseases, a non-diabetic subtyping model, and a method for predicting postprandial blood glucose. Background Technology

[0004] Diabetes mellitus is one of the three major chronic diseases in the world, and its complications, including retinopathy, are a significant threat to patients' lives and health. Diabetes is caused by abnormalities in glucose metabolism dynamics, such as insufficient insulin secretion and decreased insulin sensitivity, and exhibits strong inter-patient heterogeneity in clinical manifestations, treatment, and prognosis.

[0005] The prevalence of diabetes has increased rapidly in recent decades, becoming an increasingly serious public health problem. Early warning and stratified management of non-diabetic populations are crucial for the prevention and control of diabetes, especially for those with prediabetes, a subclinical condition that is asymptomatic, with a global prevalence of approximately 10% to 30%.

[0006] However, prediabetes can progress to type 2 diabetes, with an annual conversion rate of 5% to 10%. Type 2 diabetes is associated with morbidity and mortality. Therefore, it is crucial to classify non-diabetic individuals to identify those at risk of developing diabetes and to provide targeted, appropriate dietary and exercise intervention recommendations.

[0007] Traditional diabetes diagnosis primarily relies on clinical physiological and pathological examinations, using glucose metabolism indicators such as fasting blood glucose, insulin, C-peptide, and glycated hemoglobin to diagnose diabetes. Screening for diabetic complications typically begins in the fifth year after diagnosis, combining medical examinations, including fundus imaging, to assess complication risk and manage the progression of diabetes. However, clinical physiological and pathological examinations are time-consuming and labor-intensive. The screening and management of complications depend on complex medical equipment, are invasive procedures, require individuals to visit a hospital, have poor accessibility, and ignore patient heterogeneity. The screening and testing of a large diabetic population and high-risk individuals places significant pressure on the healthcare system.

[0008] Continuous Glucose Monitoring Systems (CGMS) provide users with comprehensive, real-time blood glucose level data, reflecting the full, rich, and complex dynamic characteristics of blood glucose. They are constantly evolving towards greater convenience, accuracy, minimal invasiveness, and even non-invasiveness. Currently, technologies for diagnosing diabetes and its complications based on CGM data mainly focus on two aspects: First, statistical indicators obtained from CGM data, such as mean, variance, and time proportion within a range, reflect an individual's dynamic blood glucose characteristics. However, these indicators only reflect a single dynamic blood glucose feature. Combining clinical experience and population data, abnormal glucose metabolism can be identified from these statistical indicators, achieving diabetes diagnosis based on CGM data. However, this method can only diagnose abnormal glucose metabolism, i.e., identifying whether a patient is diabetic or a normal control group, lacking interpretation of issues including diabetes type and complication risk, resulting in insufficient accuracy and limited application scenarios. The other method is to apply traditional machine learning algorithms, such as neural networks, for supervised training and prediction on a single, small dataset, but the prediction accuracy is not high.

[0009] Summary of the Invention

[0010] Based on this, one embodiment of this application provides a method for constructing a prediction model for glucose metabolism-related diseases, a non-diabetic classification model, and a method for predicting postprandial blood glucose.

[0011] One or more embodiments of this application provide a method for constructing a predictive model for glucose metabolism-related diseases, including:

[0012] Obtain dynamic blood glucose concentration data of the first sample set, and perform machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the first sample set to obtain a pre-trained model;

[0013] Obtain dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set; train the pre-trained model and the first fine-tuning layer based on the dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set to obtain a prediction model for impaired glucose metabolism; and / or

[0014] We obtained dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from the third sample set. Based on the dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from the third sample set, we trained the pre-trained model and the second fine-tuning layer to obtain a model for predicting the degree of glucose metabolism impairment.

[0015] The first and third sample sets include multiple samples with normal glucose tolerance, multiple samples with impaired glucose tolerance, and multiple samples with diabetes. The second sample set includes samples with normal glucose tolerance and samples with impaired glucose metabolism corresponding to the prediction task.

[0016] In one embodiment, the process of obtaining a pre-trained model by performing unsupervised machine learning on the dynamic blood glucose concentration data based on the first sample set includes:

[0017] The dynamic blood glucose concentration data of the first sample set are preprocessed to obtain the continuous blood glucose sequence data of the first sample set;

[0018] The continuous blood glucose sequence data of the first sample set is input into the initial model to obtain the dynamic blood glucose feature data of the first sample set;

[0019] The continuous blood glucose sequence data of the first sample set is masked, and the masked dynamic blood glucose sequence data of the first sample set is input into the initial model to obtain the masked dynamic blood glucose feature data of the first sample set.

[0020] The training loss of the initial model is determined based on the blood glucose dynamic feature data of the first sample set and the blood glucose dynamic feature data after masking the target sample. The parameters of the initial model are updated based on the training loss of the initial model. The pre-trained model is obtained based on the updated initial model parameters.

[0021] In one embodiment, the initial model includes a transformer model, which includes an initial encoding layer, a multi-head self-attention mechanism layer, and a feedforward neural network layer.

[0022] In one embodiment, the training of the glucose metabolism impairment disease prediction model based on dynamic blood glucose concentration data and clinical diagnostic labels from a second sample set includes:

[0023] The dynamic blood glucose concentration data of the second sample set is input into the pre-trained model to obtain the dynamic blood glucose feature data of the second sample set;

[0024] The dynamic blood glucose feature data of the second sample set is input into the first fine-tuning layer to obtain the predicted clinical diagnosis label of the second sample set.

[0025] Based on the predicted clinical diagnostic labels and clinical diagnostic labels of the second sample set, a first training loss is determined. The parameters of the pre-trained model and the parameters of the first fine-tuning layer are updated according to the first training loss. Based on the updated parameters of the pre-trained model and the updated parameters of the first fine-tuning layer, a prediction model for glucose metabolism impairment diseases corresponding to the prediction task is obtained.

[0026] In one embodiment, the second fine-tuning layer includes a linear coding layer and the prediction layer. The pre-trained model and the second fine-tuning layer are trained based on dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from a third sample set to obtain a glucose metabolism impairment prediction model, including:

[0027] The dynamic blood glucose concentration data of the third sample set is input into the pre-trained model to obtain the dynamic blood glucose feature data of the third sample set.

[0028] The dynamic blood glucose feature data of the third sample set is input into the linear coding layer to obtain the glucose metabolism impairment feature data of the third sample set;

[0029] Inputting the glucose metabolism impairment feature data of the third sample set into the prediction layer yields multiple predicted clinical physiological features for each sample in the third sample set;

[0030] The second training loss is determined based on multiple predicted clinical physiological features, clinical diagnostic labels and multiple clinical physiological data for each sample in the third sample set. The parameters of the linear coding layer and the prediction layer are updated according to the second training loss. The glucose metabolism impairment prediction model is obtained based on the pre-trained model and the updated second fine-tuning layer parameters.

[0031] One or more embodiments of this application provide a method for predicting glucose metabolism-related diseases, including:

[0032] Obtain dynamic blood glucose concentration data from the subjects;

[0033] The subject's continuous glucose concentration data are input into the glucose metabolism impairment prediction model constructed using the above method to obtain the subject's glucose metabolism impairment prediction results; and / or

[0034] The subject's dynamic blood glucose concentration data is input into the glucose metabolism impairment prediction model constructed by the above method to obtain the predicted result of the subject's glucose metabolism impairment.

[0035] One or more embodiments of this application provide an apparatus for constructing a predictive model for glucose metabolism-related diseases, comprising:

[0036] Pre-training module: used to acquire dynamic blood glucose concentration data of the first sample set, and to obtain a pre-trained model by performing machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the first sample set.

[0037] The first model fine-tuning module is used to acquire dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set, and to train the pre-trained model and the first fine-tuning layer based on the dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set to obtain a prediction model for impaired glucose metabolism; and / or

[0038] The second model fine-tuning module is used to acquire dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data of the third sample set. Based on the dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data of the third sample set, the pre-trained model and the second fine-tuning layer are trained to obtain a model for predicting the degree of glucose metabolism impairment.

[0039] The first and third sample sets include multiple samples with normal glucose tolerance, multiple samples with impaired glucose tolerance, and multiple samples with diabetes. The second sample set includes samples with normal glucose tolerance and samples with impaired glucose metabolism corresponding to the prediction task.

[0040] One or more embodiments of this application provide an apparatus for predicting glucose metabolism-related diseases, comprising:

[0041] Data acquisition module: used to acquire dynamic blood glucose concentration data of the subjects;

[0042] First prediction module: used to input the subject's dynamic blood glucose concentration data into the glucose metabolism impairment prediction model constructed by the above method, and obtain the subject's glucose metabolism impairment prediction result; and / or

[0043] The second prediction module is used to input the subject's dynamic blood glucose concentration data into the glucose metabolism impairment prediction model constructed by the above method, and obtain the prediction result of the subject's glucose metabolism impairment.

[0044] One or more embodiments of this application disclose a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0045] One or more embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0046] One or more embodiments of this application also provide a method for constructing a non-diabetic subtyping model, including:

[0047] Obtain dynamic blood glucose concentration data of the target sample, and perform machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample to obtain a pre-trained model;

[0048] Obtain the dynamic blood glucose feature data of multiple non-diabetic samples output by the pre-trained model, and cluster the dynamic blood glucose feature data of multiple non-diabetic samples to obtain the dynamic blood glucose distribution data of non-diabetic samples.

[0049] Among them, the non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic characteristic data clusters and their corresponding non-diabetic subtypes;

[0050] The target samples include multiple non-diabetic samples and multiple diabetic samples, and the multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0051] In one embodiment, the pre-trained model obtained by unsupervised machine learning based on the dynamic blood glucose concentration data of the target sample includes:

[0052] The dynamic blood glucose concentration data of the target sample is preprocessed to obtain the dynamic blood glucose sequence data of the target sample;

[0053] The dynamic blood glucose sequence data of the target sample is input into the initial model to obtain the dynamic blood glucose feature data of the target sample;

[0054] The dynamic blood glucose sequence data of the target sample is masked to obtain the masked dynamic blood glucose sequence data. The masked dynamic blood glucose sequence data of the target sample is then input into the initial model to obtain the dynamic blood glucose feature data of the target sample.

[0055] The training loss is determined based on the dynamic blood glucose feature data of the target sample and the dynamic blood glucose feature data after masking the target sample. The parameters of the initial model are updated based on the training loss. The pre-trained model is obtained based on the initial model with updated parameters.

[0056] In one embodiment, the initial model includes an initial encoding layer and an initial prediction layer, the initial prediction layer including a multi-head self-attention mechanism layer and a feedforward neural network layer.

[0057] In one embodiment, the method further includes:

[0058] Multiple clinical features corresponding to each blood glucose dynamic feature data cluster are obtained, and the risk of diabetes for each non-diabetic subtype is determined based on the multiple clinical features and the blood glucose fluctuation characteristics of each blood glucose dynamic feature data cluster.

[0059] One or more embodiments of this application provide a method for predicting non-diabetic subtypes, including:

[0060] Obtain dynamic blood glucose concentration data of the sample to be tested;

[0061] The non-diabetic subtype of the test sample is determined based on the dynamic blood glucose concentration data of the test sample;

[0062] The method is based on a non-diabetic subtyping model, which is used to perform the following steps:

[0063] The dynamic blood glucose concentration data of the sample to be tested is input into the pre-trained model to obtain the dynamic blood glucose characteristic data of the sample to be tested.

[0064] The non-diabetic subtype of the test sample is determined based on the dynamic blood glucose characteristics data and the non-diabetic blood glucose distribution data of the test sample.

[0065] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0066] The non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic feature data clusters, which are obtained by clustering the blood glucose dynamic feature data of multiple non-diabetic samples output by the pre-trained model.

[0067] In one embodiment, determining the non-diabetic classification of the test sample based on the dynamic blood glucose characteristic data and non-diabetic blood glucose dynamic distribution data includes:

[0068] Calculate the dynamic blood glucose characteristic data of the sample to be tested and the average distance of each dynamic blood glucose characteristic data cluster;

[0069] The non-diabetic classification of the test sample is determined based on the non-diabetic classification represented by the blood glucose dynamic characteristic data cluster with the smallest average distance.

[0070] Optionally, the pre-trained model includes a pre-trained encoding layer and a prediction layer, wherein the prediction layer includes a multi-head self-attention mechanism layer and a feedforward neural network layer.

[0071] One or more embodiments of this application provide an apparatus for constructing a non-diabetic subtyping model, comprising:

[0072] Pre-training module: used to acquire dynamic blood glucose concentration data of the target sample, and to obtain a pre-trained model by performing machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample;

[0073] Clustering module: used to obtain the dynamic blood glucose feature data of multiple non-diabetic samples output by the pre-trained model, and to cluster the dynamic blood glucose feature data of multiple non-diabetic samples to obtain the dynamic blood glucose distribution data of non-diabetic samples.

[0074] Among them, the non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic characteristic data clusters and their corresponding non-diabetic subtypes;

[0075] The target samples include multiple non-diabetic samples and multiple diabetic samples, and the multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0076] One or more embodiments of this application provide a method for predicting non-diabetic subtypes, including:

[0077] Data acquisition module: used to acquire dynamic blood glucose concentration data of the sample to be tested;

[0078] Non-diabetic genotyping module: used to determine the non-diabetic genotyping of the test sample based on the dynamic blood glucose concentration data of the test sample;

[0079] The device is based on a non-diabetic subtyping model, which is used to perform the following steps:

[0080] The dynamic blood glucose concentration data of the sample to be tested is input into the pre-trained model to obtain the dynamic blood glucose characteristic data of the sample to be tested.

[0081] The non-diabetic subtype of the test sample is determined based on the dynamic blood glucose characteristics data and the non-diabetic blood glucose distribution data of the test sample.

[0082] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0083] The non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic feature data clusters and their corresponding non-diabetic subtypes. These multiple blood glucose dynamic feature data clusters are obtained by clustering the blood glucose dynamic feature data of multiple non-diabetic samples output by a pre-trained model.

[0084] One or more embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0085] One or more embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0086] One or more embodiments of this application also provide a method for predicting postprandial blood glucose, including:

[0087] The subjects' dynamic blood glucose characteristics data, pre-meal real-time blood glucose data, and at least one dietary structure information are obtained, wherein the subjects' dynamic blood glucose characteristics data are obtained based on a pre-trained model;

[0088] The subject's dynamic blood glucose characteristics data, pre-meal real-time blood glucose data, and at least one dietary structure information are input into the post-meal blood glucose prediction model, and the predicted post-meal blood glucose of the subject is output based on the post-meal blood glucose prediction model.

[0089] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample. The target sample includes multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0090] The postprandial blood glucose prediction model is obtained by training a machine learning model based on the subject's dynamic blood glucose characteristics data, historical dietary structure information, and historical blood glucose data of the corresponding target time zone. The historical blood glucose data of the target time zone includes historical pre-meal blood glucose data and historical postprandial blood glucose data.

[0091] In one embodiment, the method includes:

[0092] Dietary prescription recommendations are provided based on the subject's predicted postprandial blood glucose levels;

[0093] Optionally, the machine learning model includes a long short-term memory network model;

[0094] Optionally, the machine learning model includes an encoding layer, a long short-term memory network layer, and a prediction layer.

[0095] This application also provides a method for constructing a postprandial blood glucose prediction model, including:

[0096] The subjects' dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone were obtained. The subjects' dynamic blood glucose characteristics data were obtained based on a pre-trained model.

[0097] A machine learning model was trained using the subjects' dynamic blood glucose characteristics data, historical dietary structure information, and historical blood glucose data of the corresponding target time zone to obtain a postprandial blood glucose prediction model.

[0098] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0099] In one embodiment, the steps for constructing the pre-trained model include:

[0100] The dynamic blood glucose concentration data of the target sample is preprocessed to obtain the dynamic blood glucose sequence data of the target sample;

[0101] The dynamic blood glucose sequence data of the target sample is input into the initial model to obtain the dynamic blood glucose feature data of the target sample;

[0102] The dynamic blood glucose sequence data of the target sample is masked to obtain the masked dynamic blood glucose sequence data of the target sample. The masked dynamic blood glucose sequence data of the target sample is then input into the initial model to obtain the dynamic blood glucose feature data of the target sample.

[0103] The first training loss is determined based on the dynamic blood glucose feature data of the target sample and the dynamic blood glucose feature data after masking the target sample. The parameters of the initial model are updated based on the first loss and the pre-trained model is obtained based on the initial model with updated parameters.

[0104] In one embodiment, the initial model includes an initial encoding layer and an initial prediction layer, the initial prediction layer including a multi-head self-attention mechanism layer and a feedforward neural network layer.

[0105] In one embodiment, a postprandial blood glucose prediction model is obtained by training a machine learning model using the subject's dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone, including:

[0106] The subjects' dynamic blood glucose characteristics, historical dietary structure information, and corresponding pre-meal blood glucose data are input into a machine learning model to obtain predicted blood glucose data for the target time zone.

[0107] The second training loss is determined based on the postprandial blood glucose prediction data of the target time zone and the historical blood glucose data of the target time zone. The parameters of the machine learning model are updated according to the second training loss. The postprandial blood glucose prediction model is obtained based on the machine learning model with updated parameters.

[0108] Optionally, the machine learning model includes a long short-term memory network model;

[0109] The machine learning model includes an encoding layer, a long short-term memory network layer, and a prediction layer.

[0110] One or more embodiments of this application also provide a device for predicting postprandial blood glucose, comprising:

[0111] First data acquisition module: used to acquire the subject's dynamic blood glucose characteristic data, pre-meal real-time blood glucose data and at least one dietary structure information, wherein the subject's dynamic blood glucose characteristic data is obtained based on a pre-trained model;

[0112] Blood glucose prediction module: used to input the subject's dynamic blood glucose characteristics data, pre-meal real-time blood glucose data and at least one dietary structure information into the post-meal blood glucose prediction model, and based on the output of the post-meal blood glucose prediction model, to predict the subject's post-meal blood glucose.

[0113] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample. The target sample includes multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0114] The postprandial blood glucose prediction model is obtained by training a machine learning model based on the subject's dynamic blood glucose characteristics data, historical dietary structure information, and historical blood glucose data of the corresponding target time zone. The historical blood glucose data of the target time zone includes historical pre-meal blood glucose data and historical postprandial blood glucose data.

[0115] One or more embodiments of this application also provide an apparatus for constructing a postprandial blood glucose prediction model, comprising:

[0116] The second data acquisition module is used to acquire the subject's dynamic blood glucose characteristic data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone. The subject's dynamic blood glucose characteristic data is obtained based on a pre-trained model.

[0117] Model training module: used to train a machine learning model using the subject's dynamic blood glucose characteristics data, historical dietary structure information and corresponding historical blood glucose data of the target time zone to obtain a postprandial blood glucose prediction model. The postprandial blood glucose prediction model is used to predict the subject's postprandial blood glucose corresponding to the dietary structure information.

[0118] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0119] One or more embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0120] One or more embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0121] Details of one or more embodiments of this application are set forth in the following description, and other features, objects, and advantages of this application will become apparent from the specification and its claims. Attached Figure Description

[0122] To more clearly illustrate the technical solutions in the embodiments of this application and to more completely understand this application and its beneficial effects, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0123] Figure 1 is a flowchart of a method for constructing a prediction model for glucose metabolism-related diseases according to an embodiment of this application;

[0124] Figure 2 is a flowchart of a method for predicting glucose metabolism-related diseases provided in an embodiment of this application;

[0125] Figure 3 is a structural block diagram of the device for constructing a prediction model for glucose metabolism-related diseases provided in one embodiment of this application;

[0126] Figure 4 is a structural block diagram of a device for predicting glucose metabolism-related diseases provided in an embodiment of this application;

[0127] Figure 5 is an internal structure diagram of a computer device provided in an embodiment of this application;

[0128] Figure 6 is a flowchart of the method for constructing a prediction model for glucose metabolism-related diseases provided in Embodiment 1 of this application;

[0129] Figure 7 is a schematic flowchart of the method for constructing a pre-trained model provided in Embodiment 1 of this application;

[0130] Figure 8 is a flowchart of the method for constructing a risk model of impaired glucose metabolism provided in Embodiment 1 of this application;

[0131] Figure 9 shows the comparison of the accuracy of different prediction models provided in Embodiment 2 of this application in predicting diabetes.

[0132] Figure 10 shows the comparison of the accuracy of different prediction models provided in Embodiment 2 of this application in predicting diabetic complications;

[0133] Figure 11 is a flowchart of a method for predicting non-diabetic subtypes provided in an embodiment of this application;

[0134] Figure 12 is a flowchart of a method for constructing a non-diabetic subtyping model according to an embodiment of this application;

[0135] Figure 13 is a structural block diagram of a device for predicting non-diabetic subtypes provided in an embodiment of this application;

[0136] Figure 14 is a structural block diagram of the device for constructing a non-diabetic subtyping model provided in one embodiment of this application;

[0137] Figure 15 is a schematic flowchart of the method for predicting non-diabetic subtypes provided in Embodiment 3 of this application;

[0138] Figure 16 is a schematic diagram of the structure of the pre-trained model provided in Embodiment 3 of this application;

[0139] Figure 17 is a schematic flowchart of the method for constructing a pre-trained model provided in Embodiment 3 of this application;

[0140] Figure 18 shows the consistency comparison results between the non-diabetic classification, Glucotype classification and clinical gold standard provided in Example 4 of this application;

[0141] Figure 19 is a flowchart of a method for predicting postprandial blood glucose according to an embodiment of this application;

[0142] Figure 20 is a flowchart of a method for constructing a postprandial blood glucose prediction model according to an embodiment of this application;

[0143] Figure 21 is a structural block diagram of a device for predicting postprandial blood glucose provided in one embodiment of this application;

[0144] Figure 22 is a structural block diagram of the device for constructing a postprandial blood glucose prediction model provided in one embodiment of this application;

[0145] Figure 23 is a schematic flowchart of the method for constructing a postprandial blood glucose model provided in Embodiment 5 of this application;

[0146] Figure 24 is a flowchart illustrating the training of the postprandial blood glucose prediction model provided in Embodiment 5 of this application;

[0147] Figure 25 shows the accuracy comparison results of the postprandial blood glucose prediction model and the baseline model provided in Embodiment 6 of this application in postprandial blood glucose prediction. Detailed Implementation

[0148] The present application will be further described in detail below with reference to the embodiments and examples. It should be understood that these embodiments and examples are for illustrative purposes only and are not intended to limit the scope of the present application. The purpose of providing these embodiments and examples is to enable a more thorough and comprehensive understanding of the disclosure of the present application. It should also be understood that the present application can be implemented in many different forms and is not limited to the embodiments and examples described herein. Those skilled in the art can make various modifications or alterations without departing from the spirit of the present application, and the equivalent forms obtained also fall within the protection scope of the present application. Furthermore, numerous specific details are set forth in the following description to provide a fuller understanding of the present application. It should be understood that the present application can be implemented without one or more of these details.

[0149] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0150] the term

[0151] Unless otherwise stated or in case of contradiction, the terms or phrases used herein shall have the following meanings:

[0152] The terms "and / or," "or / and," and "and / or" as used herein include any one of two or more of the related listed items, as well as any and all combinations of the related listed items. These arbitrary and all combinations include any two related listed items, any more related listed items, or a combination of all related listed items. It should be noted that when at least three items are connected by at least two conjunctions selected from "and / or," "or / and," and "and / or," it should be understood that in this application, the technical solution undoubtedly includes technical solutions connected by "logical AND," and also undoubtedly includes technical solutions connected by "logical OR." For example, "A and / or B" includes three parallel solutions: A, B, and A+B. For example, the technical solution of "A, and / or, B, and / or, C, and / or, D" includes any one of A, B, C, and D (that is, a technical solution that is connected by "logical OR"), as well as any and all combinations of A, B, C, and D, that is, combinations of any two or three of A, B, C, and D, and also combinations of all four of A, B, C, and D (that is, a technical solution that is connected by "logical AND").

[0153] In this application, the terms "multiple", "various", "multiple times", "multi-dimensional", etc., unless otherwise specified, refer to a quantity greater than or equal to 2. For example, "one or more" means one or more than or equal to two.

[0154] The terms “combinations of,” “any combination of,” and “any combination of” used in this article include all suitable combinations of any two or more of the listed items.

[0155] In this document, the term "suitable" as used in phrases such as "suitable combination," "suitable method," and "any suitable method" refers to the ability to implement the technical solution of this application, solve the technical problem of this application, and achieve the expected technical effect of this application.

[0156] In this application, terms such as "further," "even further," and "particularly" are used to describe purposes and indicate differences in content, but should not be construed as limiting the scope of protection of this application.

[0157] In this application, "optionally," "optionally," and "optional" mean that something is optional, that is, it means that it is selected from either "with" or "without." If there are multiple "optional" entries in a technical solution, unless otherwise specified, and there are no contradictions or mutual constraints, each "optional" entry shall be independent.

[0158] In this application, the technical features described in an open-ended manner include both closed technical solutions consisting of the listed features and open technical solutions that include the listed features.

[0159] In this application, numerical intervals (i.e., numerical ranges) are involved. Unless otherwise specified, the selected numerical distributions within the aforementioned numerical intervals are considered continuous and include the two endpoints (i.e., the minimum and maximum values) of the numerical range, as well as every value between these two endpoints. Unless otherwise specified, when a numerical interval refers only to integers within that interval, it includes the two endpoint integers of the numerical range, as well as every integer between the two endpoints. In this document, this is equivalent to directly listing every integer. For example, if t is an integer selected from 1 to 10, it means that t is any integer selected from the group of integers consisting of 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10. Furthermore, when multiple ranges are provided to describe features or characteristics, these ranges can be merged. In other words, unless otherwise specified, the ranges disclosed herein should be understood to include any and all subranges to which they are included.

[0160] Unless otherwise specified, the temperature parameters in this application are permitted to be either constant-temperature treatment or variations within a certain temperature range. It should be understood that the constant-temperature treatment allows temperature fluctuations within the precision range of the instrument control, such as ±5℃, ±4℃, ±3℃, ±2℃, or ±1℃.

[0161] In this application, % (w / w) and wt% both represent weight percentage, % (v / v) refers to volume percentage, and % (w / v) refers to mass-volume percentage.

[0162] All references to documents mentioned in this application are incorporated herein by reference as if each document were individually incorporated herein by reference. Unless they conflict with the inventive purpose and / or technical solution of this application, all cited documents are incorporated herein by reference in their entirety and for all purposes. When citing documents in this application, the definitions of relevant technical features, terms, nouns, phrases, etc., are also incorporated herein by reference. When citing documents in this application, examples and preferred embodiments of the cited technical features may also be incorporated herein by reference, but only to the extent that they enable the implementation of this application. It should be understood that when the cited content conflicts with the description in this application, this application shall prevail or modifications shall be made adaptably to the description in this application.

[0163] Traditional diagnosis of diabetes primarily relies on clinical, physiological, and pathological tests. For example, based on venous plasma glucose levels, glucose metabolism is categorized into: normal blood glucose (fasting blood glucose < 6.1 mmol / L, 2-hour post-glucose load blood glucose < 7.8 mmol / L), impaired fasting glucose (6.1 ≤ fasting blood glucose < 7.0 mmol / L, 2-hour post-glucose load blood glucose < 7.8 mmol / L), impaired glucose tolerance (fasting blood glucose < 7.0 mmol / L, 7.8-7.0 mmol / L ≤ 2-hour post-glucose load blood glucose < 11.1 mmol / L), and diabetes (fasting blood glucose ≥ 7.0 mmol / L, 2-hour post-glucose load blood glucose ≥ 11.1 mmol / L). The diagnostic criteria for diabetes in the "Guidelines for the Prevention and Treatment of Type 2 Diabetes in China (2020 Edition)" include: typical diabetic symptoms, plus fasting blood glucose ≥ 7.0 mmol / L, plus 2-hour post-glucose load blood glucose ≥ 11.1 mmol / L, or plus HbA1c (glycated hemoglobin) ≥ 6.5%. It is evident that the quantitative measurement of blood glucose concentration in the gold standard for the diagnosis of diabetes depends on two static time points: fasting and 2 hours after a glucose load. This obviously cannot fully and timely reflect the patient's blood glucose fluctuations, especially making it difficult to detect occult hyperglycemia and hypoglycemia, such as postprandial hyperglycemia and nocturnal asymptomatic hypoglycemia.

[0164] In addition, large fluctuations in blood sugar levels over a short period of time are also associated with a variety of health problems, including increased risk of heart attack and stroke due to blood vessel damage and inflammation; kidney damage and decreased kidney function due to damage to the glomeruli (tiny structures in the kidneys that filter blood); retinal damage and vision loss due to retinal blood vessel damage and leakage; and nerve damage, causing numbness, tingling, and pain due to nerve damage and inflammation.

[0165] To at least partially address the aforementioned technical problems, the first aspect of this application provides a method for constructing a predictive model for glucose metabolism-related diseases, comprising:

[0166] S10: Obtain dynamic blood glucose concentration data of the first sample set, and perform machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the first sample set to obtain a pre-trained model.

[0167] Specifically, dynamic glucose concentration data refers to data obtained by continuously monitoring the blood glucose concentration of a subject throughout the day, such as CGM data.

[0168] In some implementations, the pre-trained model includes a pre-trained encoding layer and a prediction layer, with the prediction layer comprising a multi-head self-attention mechanism layer and a feedforward neural network layer. The pre-trained model is used to predict corresponding blood glucose dynamics features by machine learning the dynamic blood glucose concentration data of the target sample in an unsupervised manner. This application utilizes continuous blood glucose monitoring data from various types of samples to construct a pre-trained model, which can effectively capture and store the dynamic blood glucose change features contained in the CGM (Continuous Glucose Metric).

[0169] In some implementations, the first sample set includes multiple samples with normal glucose tolerance, multiple samples with impaired glucose tolerance, and multiple samples with diabetes.

[0170] The diagnostic criteria for normal glucose tolerance samples are: when the subject's fasting blood glucose is <6.1mmol / L and the 2-hour postprandial blood glucose is <7.8mmol / L, it indicates that the body's ability to regulate blood glucose after ingesting glucose is normal.

[0171] The diagnostic criteria for impaired glucose tolerance are: when a subject performs an oral glucose tolerance test, his / her fasting blood glucose is normal (less than 6.0 mmol / L), but his / her blood glucose 2 hours after oral glucose administration is between 7.8 and 11.1 mmol / L. This value is greater than the normal value, but does not meet the diagnostic criteria for diabetes. This state is impaired glucose tolerance.

[0172] Diabetes includes type 1 diabetes, type 2 diabetes, gestational diabetes, and other special types of diabetes. The gold standard for diagnosing type 2 diabetes is the glucose tolerance test. According to the "Guidelines for the Prevention and Treatment of Type 2 Diabetes in China (2020 Edition)," the diagnostic criteria for diabetes include: typical symptoms of diabetes, plus fasting blood glucose ≥7.0 mmol / L, plus 2-hour post-glucose load blood glucose ≥11.1 mmol / L, or plus HbA1c (glycated hemoglobin) ≥6.5%.

[0173] In some implementation schemes, dynamic blood glucose concentration data of a first sample set is obtained, and a pre-trained model is obtained by unsupervised machine learning based on the dynamic blood glucose concentration data of the first sample set, including:

[0174] The dynamic blood glucose concentration data of the first sample set are preprocessed to obtain the continuous blood glucose sequence data of the first sample set;

[0175] The continuous blood glucose sequence data of the first sample set is input into the initial model to obtain the dynamic blood glucose feature data of the first sample set;

[0176] The continuous blood glucose sequence data of the first sample set is masked, and the masked dynamic blood glucose sequence data of the first sample set is input into the initial model to obtain the masked dynamic blood glucose feature data of the first sample set.

[0177] The training loss of the initial model is determined based on the dynamic blood glucose feature data of the first sample set and the dynamic blood glucose feature data of the target sample after masking. The parameters of the initial model are updated based on the training loss of the initial model, and the pre-trained model is obtained based on the updated initial model parameters.

[0178] Specifically, dynamic blood glucose sequence data refers to time series data obtained by segmenting and processing dynamic blood glucose concentration data.

[0179] In some specific implementation schemes, the preprocessing of dynamic blood glucose concentration data involves dividing the data into daily segments. Each day, from 0:00 to 24:00, contains 288 complete measurements, forming a sequence. Incomplete sequences are filtered out to obtain a full-day blood glucose time series. The blood glucose concentration value measured at each time point is uniformly discretized into 261 blood glucose levels, ranging from Gmin (40 mg / ml) to Gmax (300 mg / ml), defined as Gmax-Gmin+1.

[0180] In some specific implementations, for the collected CGM data, this application constructs a deep self-attention transformation network model with a Transformer architecture as the initial model and pre-trains it on the collected CGM data. Deep learning is used to effectively capture and store the dynamic blood glucose change patterns contained in the CGM, so that the computer system can understand, reason and process this information, and then build a high-precision diabetes-related disease prediction model based on this information.

[0181] The initial model refers to the sequence data processing model that has not been trained. In some specific implementations, the initial model is a Transformer model.

[0182] Transformer is a deep learning model architecture for natural language processing (NLP) and other sequence-to-sequence tasks. By introducing a self-attention mechanism, it excels at processing sequential data.

[0183] In some implementations, the initial model includes an initial encoding layer and an initial prediction layer. The initial prediction layer consists of a multi-head self-attention mechanism layer and a feedforward neural network layer. Specifically, the initial encoding layer is used to encode the positions and elements in the dynamic blood glucose sequence data to obtain blood glucose encoded feature data. The initial prediction layer is used to further learn the patterns of the dynamic blood glucose encoded feature data and output dynamic blood glucose feature data. During pre-training, this application randomly sets a mask for the input sequence and feeds the masked input sequence into the initial model. The elements at the masked positions are reconstructed to learn the features and patterns of the input sequence, and a loss function is calculated.

[0184] In some specific implementations, the prediction layer consists of four transformer units, each of which consists of a multi-head self-attention module (MSA) and a feedforward neural network module (MLP).

[0185] Multi-head attention: The self-attention mechanism in the Transformer unit is extended to multiple attention heads, each of which can learn different attention weights to better capture different types of relationships. Multi-head attention allows the model to process different information subspaces in parallel.

[0186] A feedforward neural network is a fully connected network containing one hidden layer, consisting of two linear transformations separated by a non-linear activation function (typically ReLU) as the hidden layer. The first linear transformation projects the input representation into a higher-dimensional space using a learnable weight matrix. The non-linear activation passes the output of the first linear transformation through a non-linear activation function (e.g., ReLU), introducing non-linearity into the model and enabling it to capture complex patterns and relationships in the data. The second linear transformation projects the output of the activation function back into the original dimensional space through another learnable weight matrix.

[0187] Specifically, the process of inputting dynamic blood glucose sequence data into the initial model for pre-training includes the following steps:

[0188] The position and elements of the dynamic blood glucose sequence data are encoded to output the blood glucose coding feature data corresponding to the dynamic blood glucose sequence data;

[0189] Calculate the multi-head self-attention vector of blood glucose coding feature data, and output the combination of multi-head self-attention vectors after concatenation and layer normalization.

[0190] The output results after concatenation and layer normalization are input into the feedforward neural network for transformation. The transformed results are then normalized by connecting the residuals to the next layer to obtain the output of the feedforward neural network layer.

[0191] Specifically, after the initial encoding layer encodes the dynamic blood glucose sequence data to obtain blood glucose encoded feature data, the multi-head self-attention mechanism layer encodes the blood glucose encoded feature data into low-dimensional triples, namely query vector, key vector, and value vector. The query vector and key vector are multiplied by a dot product and then softmax is performed to obtain the coefficients of the value vector, which are the self-attention vectors of each head. After the self-attention vectors of each head are calculated, the final result is a linear combination of different value vectors. The linear combination results of different value vectors are concatenated and then normalized to obtain the output of the multi-head self-attention mechanism layer. Furthermore, the feedforward neural network layer transforms the output of the multi-head self-attention mechanism layer and outputs it in the form of residual connections and normalization.

[0192] Residual connections involve directly adding the input to the output, passing some information directly to the next layer, thus avoiding information loss. This connection method makes the network easier to train and helps improve model performance. Layer normalization is a normalization technique that normalizes features in the input of each layer, accelerating network convergence and improving the model's generalization ability.

[0193] In this application, by pre-training the Transformer model on large-scale CGM data, it is possible to learn the dynamic features and patterns of blood glucose fluctuations from a large amount of unlabeled CGM data in an unsupervised manner, capture the dynamic features of individual blood glucose, and realize the mapping of CGM data to a low-dimensional vector space to capture its potential dynamic features of blood glucose.

[0194] S20: Obtain dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set. Based on the dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set, train the pre-trained model and the first fine-tuning layer to obtain a prediction model for impaired glucose metabolism.

[0195] Specifically, by designing prediction tasks related to glucose metabolism diseases, selecting dynamic blood glucose concentration data of target samples that match the prediction tasks, and combining the clinical diagnostic labels of the target samples, a pre-trained model and the first fine-tuning layer are trained to obtain a prediction model for glucose metabolism disorders.

[0196] Among them, the prediction task for glucose metabolism-related diseases can include any one of the following: predicting whether glucose metabolism is normal, predicting whether someone has diabetes, predicting the type of diabetes, predicting macrovascular complications, and predicting microvascular complications.

[0197] In some implementations, for the task of predicting whether glucose metabolism status is normal, the second sample set includes multiple normal glucose tolerance samples, multiple glucose regulation impairment samples, and multiple mixed glucose tolerance impairment samples.

[0198] In some implementations, for the task of predicting whether someone has diabetes, the second sample set includes multiple samples with diabetes and multiple samples without diabetes. In some implementations, for the task of predicting the type of diabetes, the second sample set includes samples with type 1 diabetes and samples with type 2 diabetes.

[0199] In some implementations, for the task of predicting macrovascular complications, the second sample set includes diabetic samples with macrovascular complications and diabetic samples without macrovascular complications.

[0200] In some implementations, for the task of predicting microvascular complications, the second sample set includes diabetic samples with microvascular complications and diabetic samples without microvascular complications.

[0201] In some implementations, a glucose metabolism impairment prediction model is obtained by training a pre-trained model and a first fine-tuning layer based on dynamic blood glucose concentration data and clinical diagnostic labels from a second sample set.

[0202] The dynamic blood glucose concentration data of the second sample set is input into the pre-trained model to obtain the dynamic blood glucose feature data of the second sample set;

[0203] The dynamic blood glucose feature data of the second sample set is input into the first fine-tuning layer to obtain the predicted clinical diagnosis label of the second sample set.

[0204] Based on the predicted clinical diagnostic labels and clinical diagnostic labels of the second sample set, the first training loss is determined. The parameters of the pre-trained model and the parameters of the first fine-tuning layer are updated according to the first training loss. Based on the updated parameters of the pre-trained model and the updated parameters of the first fine-tuning layer, a prediction model for glucose metabolism disorders corresponding to the prediction task is obtained.

[0205] Specifically, by adding a task-specific prediction layer after the pre-trained model, the parameters of the first fine-tuning layer and the parameters in the pre-trained model are adjusted according to the dynamic blood glucose concentration data of the second sample set corresponding to different prediction tasks, thereby obtaining multiple task-specific prediction models for glucose metabolism disorders, so as to realize diagnostic tasks in multiple medical scenarios.

[0206] In some exemplary implementations, the glucose metabolism-related disease prediction model includes any one of the following: an impaired glucose tolerance prediction model for predicting whether glucose tolerance is impaired, a diabetes prediction model for predicting whether a subject has diabetes, a diabetes type prediction model for predicting the type of diabetes a subject has, and a diabetes complication prediction model for predicting whether a subject has complications.

[0207] In some specific implementation plans, the types of diabetes include type 1 diabetes and type 2 diabetes.

[0208] In some specific implementation plans, diabetic complications include macrovascular complications and microvascular complications.

[0209] S30: Obtain dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from the third sample set. Based on the dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from the third sample set, train the pre-trained model and the second fine-tuning layer to obtain a model for predicting the degree of glucose metabolism impairment.

[0210] Specifically, multiple clinical physiological data include age, BMI, fasting blood glucose, fasting insulin, HOMA-IS, HOMA-B, 120-minute postprandial blood glucose, glycated hemoglobin, and high-density cholesterol.

[0211] In some implementations, the third sample set includes multiple samples with normal glucose tolerance, multiple samples with impaired glucose tolerance, and multiple samples with diabetes. In some implementations, the third sample set and the first sample set may be the same.

[0212] In some implementations, the second fine-tuning layer comprises multiple linear coding layers and multiple prediction layers. Each prediction layer consists of a linear transformation and an activation function. The prediction task of each prediction layer is different.

[0213] In some implementation schemes, a model for predicting the degree of impaired glucose metabolism is obtained by training a pre-trained model and a second fine-tuning layer based on dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from a third sample set.

[0214] The dynamic blood glucose concentration data of the third sample set is input into the pre-trained model to obtain the dynamic blood glucose feature data of the third sample set.

[0215] The dynamic blood glucose feature data of the third sample set is input into the linear coding layer to obtain the glucose metabolism impairment feature data of the third sample set;

[0216] Inputting the glucose metabolism impairment feature data of the third sample set into the prediction layer yields multiple predicted clinical physiological features for each sample in the third sample set;

[0217] The second training loss is determined based on multiple predicted clinical physiological features, clinical diagnostic labels, and multiple clinical physiological data for each sample in the third sample set. The parameters of the linear coding layer and the prediction layer are updated according to the second training loss. The prediction model of the degree of glucose metabolism impairment is obtained based on the pre-trained model and the updated second fine-tuning layer parameters.

[0218] Specifically, the glucose metabolism impairment prediction model is obtained by adding a second fine-tuning layer consisting of a multi-layer linear coding layer and multiple prediction layers after the pre-trained model. The multi-layer linear coding layer maps CGM data to a real number, which reflects the patient's diabetes course and complication risk. This real number is then input into multiple task-specific prediction layers to decode and predict multiple clinical physiological data. The model's loss function is defined by the mean squared error of multiple tasks, thus obtaining the glucose metabolism impairment prediction model.

[0219] In summary, this application collects large-scale CGM data and constructs a CGM pre-trained model using self-supervised learning on a large-sample CGM dataset. This model can capture the dynamic features of individual blood glucose in CGM. For each glucose metabolism-related disease diagnosis task, different fine-tuning layers are added to the pre-trained model, and the model is fine-tuned by combining the clinical diagnostic labels of CGM data, thereby obtaining a prediction model adapted to different glucose metabolism-related disease prediction tasks. Based on the pre-trained model, a multi-task learning model is designed, and the pre-trained model is fine-tuned by combining the clinical diagnostic labels of CGM data and multiple clinical physiological data, thereby obtaining a model for predicting the degree of glucose metabolism impairment and the risk of complications.

[0220] Furthermore, a second aspect of this application provides a method for predicting glucose metabolism-related diseases, the flowchart of which is shown in Figure 2, including:

[0221] S100: Acquire dynamic blood glucose concentration data of the subject;

[0222] S200: Input the subject's continuous glucose concentration data into the glucose metabolism impairment prediction model constructed by the above method to obtain the subject's glucose metabolism impairment prediction results; and / or

[0223] S300: Input the subject's dynamic blood glucose concentration data into the glucose metabolism impairment prediction model constructed by the above method to obtain the subject's glucose metabolism impairment prediction results.

[0224] Specifically, this application provides a method for predicting glucose metabolism-related diseases using a pre-trained model trained on dynamic blood glucose concentration data. It combines two stages: unsupervised learning in the pre-training process and supervised learning in the fine-tuning process, to predict glucose metabolism-impaired diseases using an individual's CGM data.

[0225] For input subject CGM data, the fine-tuned pre-trained model can combine the blood glucose dynamics characteristics obtained from large-sample unlabeled pre-training with the learning of labeled CGM data to achieve multi-task-specific diagnosis of patients, answering questions such as whether they have diabetes, whether they have impaired glucose tolerance, the type of diabetes, whether they have macrovascular complications, and whether they have microvascular complications, and providing suggestions for diagnosis, treatment and diabetes management.

[0226] Furthermore, this application also provides a method for assessing the degree of impaired glucose metabolism based on dynamic blood glucose concentration data. By combining large-sample clinical physiological data and adopting a multi-task learning approach, it can estimate the degree of impaired glucose metabolism in patients, thereby assisting in the diagnosis and treatment of patients and the daily management of diabetes.

[0227] Accordingly, a third aspect of this application provides an apparatus for constructing a predictive model for glucose metabolism-related diseases, as shown in Figure 3, comprising:

[0228] Pre-training module 10: Used to obtain dynamic blood glucose concentration data of the first sample set, and to obtain a pre-trained model by performing machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the first sample set.

[0229] First model fine-tuning module 20: used to acquire dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set, and to train the pre-trained model and the first fine-tuning layer based on the dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set to obtain a prediction model for impaired glucose metabolism; and / or

[0230] The second model fine-tuning module 30 is used to acquire dynamic blood glucose concentration data, clinical diagnostic labels and clinical physiological data of the third sample set, and to train the pre-trained model and the second fine-tuning layer based on the dynamic blood glucose concentration data, clinical diagnostic labels and clinical physiological data of the third sample set to obtain a model for predicting the degree of glucose metabolism impairment.

[0231] The first and third sample sets include multiple samples with normal glucose tolerance, multiple samples with impaired glucose tolerance, and multiple samples with diabetes. The second sample set includes samples with normal glucose tolerance and samples with impaired glucose metabolism corresponding to the prediction task.

[0232] A fourth aspect of this application provides an apparatus for predicting glucose metabolism-related diseases, as shown in Figure 4, comprising:

[0233] Data acquisition module 100: Used to acquire dynamic blood glucose concentration data of the subject;

[0234] First prediction module 200: used to input the subject's dynamic blood glucose concentration data into the glucose metabolism impairment prediction model constructed by the above method, and obtain the subject's glucose metabolism impairment prediction result; and / or

[0235] The second prediction module 300 is used to input the subject's dynamic blood glucose concentration data into the glucose metabolism impairment prediction model constructed by the above method to obtain the subject's glucose metabolism impairment prediction result.

[0236] Specific limitations regarding the construction of predictive models and devices for glucose metabolism-related diseases can be found in the above-mentioned limitations on the methods for constructing such models and predicting these diseases, and will not be repeated here. Each module in the aforementioned predictive models and devices for glucose metabolism-related diseases can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0237] In some embodiments, a computer device is provided, which can be a server 104 or a terminal 102, and its internal structure diagram can be as shown in Figure 5. The computer device includes a processor, memory, and a communication interface connected via a system bus. When the computer device is a terminal, it also includes a display screen and input devices connected to the system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for constructing a predictive model for glucose metabolism-related diseases and predicting glucose metabolism-related diseases. The display screen of the computer device can be a liquid crystal display screen or an e-ink display screen. The input devices of the computer device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse, etc.

[0238] Those skilled in the art will understand that the structure shown in Figure 5 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or may combine certain components, or may have different component arrangements.

[0239] This application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-described method for constructing a prediction model for glucose metabolism-related diseases and predicting glucose metabolism-related diseases.

[0240] This application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method for constructing a prediction model for glucose metabolism-related diseases and predicting glucose metabolism-related diseases.

[0241] This application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the methods described above for constructing a prediction model for glucose metabolism-related diseases and predicting glucose metabolism-related diseases.

[0242] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0243] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0244] In another embodiment, traditional diagnosis and classification of diabetes and impaired glucose tolerance mainly rely on fasting static blood glucose measurements or average measurements of total blood glucose, as well as oral glucose tolerance tests (OGTT). For example, classification based on the OGTT categorizes individuals into four groups based on fasting and 2-hour postprandial blood glucose levels: normal glucose tolerance (NGT), impaired fasting glucose (IFG), impaired glucose tolerance (IGT), combined impaired glucose tolerance (CGI), and diabetes.

[0245] Previous studies have proposed a pathophysiological-based classification method to categorize individuals' clinical glucose and lipid metabolism characteristics, characterizing the clinical features and risk stratification of each classification. Another study proposed a classification method based on CGM data, using spectral clustering to classify blood glucose fluctuations into low, moderate, and severe fluctuations. However, this method only considers the volatility of blood glucose dynamics, neglecting other characteristics including blood glucose levels and hyperglycemic and hypoglycemic events.

[0246] Traditional methods for classifying non-diabetic populations, from a data perspective, mostly rely on individual blood glucose measurement data, pathophysiological data, and clinical glucose and lipid metabolism characteristic data collection, which require individuals to rely on hospitals, resulting in poor accessibility. From the perspective of individual blood glucose measurement, most are based on static single-point measurement data, which cannot obtain the individual blood glucose regulation characteristics reflected by the complex dynamics of blood glucose. From the perspective of mining feature patterns from data, clustering algorithms based on the spatial representation characteristics of individual original data lack in-depth mining and representation of the nonlinear correlations contained in the data.

[0247] To address at least one of the aforementioned problems, a fifth aspect of this application provides a method for constructing a non-diabetic model, the flowchart of which is shown in Figure 11, including:

[0248] S40: Obtain dynamic blood glucose concentration data of the target sample, and perform machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample to obtain a pre-trained model.

[0249] S50: Obtain the dynamic blood glucose feature data of multiple non-diabetic samples output by the pre-trained model, and cluster the dynamic blood glucose feature data of multiple non-diabetic samples to obtain the dynamic blood glucose distribution data of non-diabetic samples.

[0250] Specifically, dynamic glucose concentration data refers to data obtained by continuously monitoring the blood glucose concentration of a subject throughout the day, such as CGM data.

[0251] In some implementations, the pre-trained model includes a pre-trained encoding layer and a prediction layer, with the prediction layer comprising a multi-head self-attention mechanism layer and a feedforward neural network layer. The pre-trained model is used to predict corresponding dynamic blood glucose feature data by machine learning from the dynamic blood glucose concentration data of the target sample in an unsupervised manner.

[0252] In some implementations, the target sample includes multiple non-diabetic samples and multiple diabetic samples, and the multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0253] The diagnostic criteria for normal glucose tolerance samples are: when the subject's fasting blood glucose is <6.1mmol / L and the 2-hour postprandial blood glucose is <7.8mmol / L, it indicates that the body's ability to regulate blood glucose after ingesting glucose is normal.

[0254] The diagnostic criteria for impaired glucose tolerance are: when a subject performs an oral glucose tolerance test, his / her fasting blood glucose is normal (less than 6.0 mmol / L), but his / her blood glucose 2 hours after oral glucose administration is between 7.8 and 11.1 mmol / L. This value is greater than the normal value, but does not meet the diagnostic criteria for diabetes. This state is impaired glucose tolerance.

[0255] The gold standard for diagnosing diabetes is the glucose tolerance test. According to the "Guidelines for the Prevention and Treatment of Type 2 Diabetes in China (2020 Edition)," the diagnostic criteria for diabetes include: typical symptoms of diabetes, plus fasting blood glucose ≥7.0 mmol / L, plus 2-hour post-glucose load blood glucose ≥11.1 mmol / L, or plus HbA1c (glycated hemoglobin) ≥6.5%.

[0256] In this application, the non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic feature data clusters obtained by clustering the blood glucose dynamic feature data of multiple non-diabetic samples output by a pre-trained model.

[0257] Clustering refers to grouping similar data together without considering the label of the group. The goal is to aggregate similar data together. Clustering is an unsupervised learning method.

[0258] Specifically, the dynamic blood glucose sequence data of non-diabetic samples is dimensionality-reduced using a pre-trained model to obtain dynamic blood glucose feature data of non-diabetic samples; further, the similarity of non-diabetic samples is measured based on the cosine similarity of the dynamic blood glucose feature data to obtain multiple dynamic blood glucose feature data clusters.

[0259] This application constructs a pre-training dataset based on continuous blood glucose monitoring data of multiple different types of samples, so that the pre-training process can learn the dynamic features and patterns of blood glucose fluctuations from a large amount of unlabeled CGM data in an unsupervised manner, thereby effectively capturing and storing the dynamic blood glucose change features contained in CGM, and embedding CGM data into a low-dimensional vector space to obtain blood glucose dynamic feature data.

[0260] Furthermore, based on the clustering results of the dynamic blood glucose characteristic data of non-diabetic populations, dynamic blood glucose distribution data of non-diabetic populations are obtained. Then, a non-diabetic population classification model based on the CGM pre-trained model is constructed to capture dynamic blood glucose characteristic data from individual dynamic blood glucose concentration data. Based on the dynamic blood glucose characteristic data of non-diabetic samples, the non-diabetic population is classified. Then, the risk of diabetes is predicted by combining the glucose metabolism characteristics and clinical data of non-diabetic populations.

[0261] In some specific implementation schemes, dynamic blood glucose concentration data of the target sample is obtained, and a pre-trained model is obtained by unsupervised machine learning based on the dynamic blood glucose concentration data of the target sample, including:

[0262] S5001: Obtain dynamic blood glucose concentration data of the target sample, and preprocess the dynamic blood glucose concentration data of the target sample to obtain dynamic blood glucose sequence data of the target sample.

[0263] S5002: Input the dynamic blood glucose sequence data of the target sample into the initial model to obtain the dynamic blood glucose feature data of the target sample;

[0264] S5003: Mask the dynamic blood glucose sequence data of the target sample to obtain the masked dynamic blood glucose sequence data of the target sample, and input the masked dynamic blood glucose sequence data of the target sample into the initial model to obtain the masked dynamic blood glucose feature data of the target sample.

[0265] S5004: Determine the training loss based on the dynamic blood glucose feature data of the target sample and the dynamic blood glucose feature data after masking the target sample, update the parameters of the initial model based on the training loss, and obtain the pre-trained model based on the initial model with updated parameters.

[0266] Specifically, dynamic blood glucose series data refers to full-day time series data obtained by segmenting and processing dynamic blood glucose concentration data.

[0267] In some specific implementation schemes, the segmentation of dynamic glucose concentration (CGM) data refers to dividing the multi-day CGM data of the sample into daily segments. Each day's 288 complete measurements from 0:00 to 24:00 are considered a sequence. Incomplete sequences from the data collection process are filtered out to obtain a complete 24-hour blood glucose time series. For each time point, the blood glucose concentration value is uniformly discretized into a series of values ​​from G... min (40mg / ml) to G max (300mg / ml) G max -G min +1 means 261 blood glucose levels.

[0268] The initial model refers to the sequence data processing model that has not been trained. For the collected CGM data, this application constructs a deep self-attention transformation network model as the initial model and pre-trains it on the collected CGM data. Deep learning is used to effectively capture and store the dynamic blood glucose change patterns contained in the CGM, so that the computer system can understand, reason, and process this information. Based on this information, the dynamic blood glucose feature data corresponding to different non-diabetic subtypes can be obtained to construct a non-diabetic subtype prediction model.

[0269] In some implementations, the initial model includes an initial encoding layer and an initial prediction layer. The initial prediction layer consists of a multi-head self-attention mechanism layer and a feedforward neural network layer. Specifically, the initial encoding layer is used to encode the positions and elements in the dynamic blood glucose sequence data to obtain blood glucose encoded feature data. The initial prediction layer is used to further learn the patterns of the dynamic blood glucose encoded feature data and output dynamic blood glucose feature data. During pre-training, this application randomly sets a mask for the input sequence and feeds the masked input sequence into the initial model. The elements at the masked positions are reconstructed to learn the features and patterns of the input sequence, and a loss function is calculated.

[0270] In some specific implementations, the prediction layer consists of four transformer units, each of which consists of a multi-head self-attention module (MSA) and a feedforward neural network module (MLP).

[0271] Multi-head attention: The self-attention mechanism in the Transformer unit is extended to multiple attention heads, each of which can learn different attention weights to better capture different types of relationships. Multi-head attention allows the model to process different information subspaces in parallel.

[0272] A feedforward neural network is a fully connected network containing one hidden layer, consisting of two linear transformations separated by a non-linear activation function (typically ReLU) as the hidden layer. The first linear transformation projects the input representation into a higher-dimensional space using a learnable weight matrix. The non-linear activation passes the output of the first linear transformation through a non-linear activation function (e.g., ReLU), introducing non-linearity into the model and enabling it to capture complex patterns and relationships in the data. The second linear transformation projects the output of the activation function back into the original dimensional space through another learnable weight matrix.

[0273] Specifically, inputting the dynamic blood glucose sequence data of the target sample into the initial model to obtain the dynamic blood glucose feature data of the target sample includes:

[0274] The dynamic blood glucose sequence data of the target sample is positionally encoded and elementally encoded to output the blood glucose encoding feature data of the target sample.

[0275] Calculate the multi-head self-attention vector of the blood glucose encoding feature data of the target sample, and output the combination of multi-head self-attention vectors after concatenation and layer normalization;

[0276] The output results after concatenation and layer normalization are input into the feedforward neural network for transformation. The transformed results are then normalized by connecting the residuals to the next layer to obtain the output of the feedforward neural network layer.

[0277] Specifically, after the initial encoding layer performs positional and elemental encoding on the dynamic blood glucose sequence data to obtain blood glucose encoded feature data, the multi-head self-attention mechanism layer encodes the blood glucose encoded feature data into low-dimensional triples, namely query vector, key vector, and value vector. The query vector and key vector are multiplied by a dot product and then softmax is performed to obtain the coefficients of the value vector, which are the self-attention vectors of each head. After the self-attention vectors of each head are calculated, the final result is a linear combination of different value vectors. The linear combination results of different value vectors are concatenated and then normalized to obtain the output of the multi-head self-attention mechanism layer. Furthermore, the feedforward neural network layer transforms the output of the multi-head self-attention mechanism layer and outputs it in the form of residual connections and normalization.

[0278] Residual connections involve directly adding the input to the output, passing some information directly to the next layer, thus avoiding information loss. This connection method makes the network easier to train and helps improve model performance. Layer normalization is a normalization technique that normalizes features in the input of each layer, accelerating network convergence and improving the model's generalization ability.

[0279] In some specific implementation plans, the methods also include, in order to achieve diabetes risk prediction for each non-diabetes subtype:

[0280] Multiple clinical features are obtained for each dynamic blood glucose feature data cluster. Based on the multiple clinical features and the blood glucose fluctuation characteristics of each dynamic blood glucose feature data cluster, the risk of diabetes for each non-diabetic subtype is determined.

[0281] Specifically, the blood glucose fluctuation characteristics of each cluster can be the statistical analysis results of the mean and standard deviation of the blood glucose dynamic characteristics from different individual samples in each cluster. The blood glucose fluctuation characteristics can be the mean and standard deviation of blood glucose concentration over the entire time period.

[0282] In practical applications, by combining individual clinical characteristics matched with CGM data, the dynamic characteristics of blood glucose and clinical characteristics of each subtype are characterized based on the global and time-segmented blood glucose fluctuation characteristics and clinical data of individuals in each cluster. This enables the assessment of the risk of diabetes in non-diabetic samples by combining subtype-specific dynamic blood glucose characteristics and clinical characteristics.

[0283] Accordingly, a sixth aspect of this application provides a method for predicting non-diabetic subtypes, the flowchart of which is shown in Figure 12, including:

[0284] S400: Acquire dynamic blood glucose concentration data of the sample to be tested;

[0285] S500: Determine the non-diabetic type of the test sample based on the dynamic blood glucose concentration data of the test sample;

[0286] Specifically, the method is based on a non-diabetic genotyping model, which includes a pre-trained model and dynamic blood glucose distribution data for non-diabetic individuals. The non-diabetic genotyping model is used to perform the following steps:

[0287] The dynamic blood glucose concentration data of the sample to be tested is input into the pre-trained model to obtain the dynamic blood glucose characteristic data of the sample to be tested.

[0288] The non-diabetic classification of the test sample is determined based on the dynamic blood glucose characteristics data and the non-diabetic blood glucose distribution data of the test sample.

[0289] The pre-trained model is obtained through unsupervised machine learning based on the dynamic blood glucose concentration data of the target sample, in order to predict the dynamic blood glucose characteristic data corresponding to the dynamic blood glucose concentration data. The non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic characteristic data clusters obtained by clustering the blood glucose dynamic characteristic data of multiple non-diabetic samples output by the pre-trained model.

[0290] Specifically, for a test sample with multiple days of CGM data, the average of the vectors segmented and mapped from the multiple days of CGM data to the latent space is used as the dynamic blood glucose feature data of the test sample.

[0291] This application uses continuous glucose monitoring to obtain individual dynamic blood glucose data, which is then encoded and embedded into the latent space by a pre-trained model to obtain dynamic blood glucose feature data. This data is combined with the distribution data of dynamic blood glucose features corresponding to different non-diabetic subtypes obtained by clustering the dynamic blood glucose feature data captured by the pre-trained model based on large-scale CGM data, thereby obtaining the subtype of individual samples.

[0292] For test samples from non-diabetic individuals, this application utilizes the powerful feature extraction capabilities of a pre-trained model to capture blood glucose dynamics, assess an individual's blood glucose metabolism status, and classify the non-diabetic population based on large-sample blood glucose dynamic feature distribution data. This helps to more accurately diagnose prediabetes, intervene and manage abnormal blood glucose metabolism earlier, and develop personalized diet, exercise, drug intervention and treatment plans.

[0293] In some implementation schemes, determining the non-diabetic classification of a test sample based on its dynamic glycemic characteristics and non-diabetic glycemic distribution data includes:

[0294] Calculate the dynamic blood glucose characteristic data of the sample to be tested and the average distance of each dynamic blood glucose characteristic data cluster;

[0295] The non-diabetic classification of the sample to be tested is determined based on the non-diabetic classification represented by the blood glucose dynamic characteristic data cluster with the smallest average distance.

[0296] Specifically, when classifying the test sample, the average distance between the dynamic blood glucose characteristic data of the test sample and each dynamic blood glucose characteristic data cluster is calculated, and the non-diabetic classification represented by the dynamic blood glucose characteristic data cluster with the smallest average distance is taken as the classification of the test sample.

[0297] In summary, traditional classification of non-diabetic populations relies on static clinical measurements. In particular, the diagnosis and classification of clinical diabetes and impaired glucose tolerance primarily depend on fasting static blood glucose measurements or average total blood glucose measurements, neglecting the individual blood glucose regulation characteristics reflected by the complex dynamics of blood glucose. Furthermore, the analysis of continuous glucose monitoring data often focuses on single statistics or single-dimensional clustering. For example, CGM-based blood glucose fluctuation classification only considers the volatility of blood glucose dynamics, ignoring other characteristics including blood glucose levels and hyperglycemic and hypoglycemic events.

[0298] Accordingly, a seventh aspect of this application provides an apparatus for constructing a non-diabetic subtyping model, as shown in Figure 13, comprising:

[0299] Pre-training module 40: Used to acquire dynamic blood glucose concentration data of the target sample, and to obtain a pre-trained model by performing machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample.

[0300] Clustering module 50: Used to obtain the dynamic blood glucose feature data of multiple non-diabetic samples output by the pre-trained model, and to cluster the dynamic blood glucose feature data of multiple non-diabetic samples to obtain multiple dynamic blood glucose feature data clusters and obtain the dynamic blood glucose distribution data of non-diabetic samples.

[0301] The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0302] Accordingly, a fourth aspect of this application provides an apparatus for constructing a non-diabetic subtyping model, as shown in Figure 14, comprising:

[0303] Data acquisition module 400: Used to acquire dynamic blood glucose concentration data of the sample to be tested;

[0304] Non-diabetic subtyping module 500: Used to determine the non-diabetic subtyping of the test sample based on the dynamic blood glucose concentration data of the test sample;

[0305] The device is based on a non-diabetic subtyping model, which is used to perform the following steps:

[0306] The dynamic blood glucose concentration data of the sample to be tested is input into the pre-trained model to obtain the dynamic blood glucose characteristic data of the sample to be tested.

[0307] The non-diabetic subtype of the test sample is determined based on the dynamic blood glucose characteristics data and the non-diabetic blood glucose distribution data of the test sample.

[0308] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0309] The non-diabetic blood glucose dynamic distribution includes multiple blood glucose dynamic feature data clusters, which are obtained by clustering the blood glucose dynamic feature data of multiple non-diabetic samples output by the pre-trained model of multiple blood glucose dynamic feature data clusters.

[0310] Specific limitations regarding the apparatus for predicting non-diabetic subtypes and / or constructing predictive models for non-diabetic subtypes can be found in the limitations regarding the methods for predicting non-diabetic subtypes and / or constructing predictive models for non-diabetic subtypes described above, and will not be repeated here. Each module in the aforementioned apparatus for predicting non-diabetic subtypes and / or constructing predictive models for non-diabetic subtypes can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0311] In some embodiments, a computer device is provided, which can be a server 104 or a terminal 102, and its internal structure diagram can be as shown in Figure 5. The computer device includes a processor, memory, and a communication interface connected via a system bus. When the computer device is a terminal, it also includes a display screen and input devices connected to the system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting non-diabetic subtypes and / or constructing a predictive model for non-diabetic subtypes. The display screen of the computer device can be a liquid crystal display (LCD) or an e-ink display. The input devices of the computer device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse, etc.

[0312] Those skilled in the art will understand that the structure shown in Figure 5 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or may combine certain components, or may have different component arrangements.

[0313] This application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described methods for predicting non-diabetic subtypes and / or constructing a model for predicting non-diabetic subtypes.

[0314] This application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the methods described above for predicting non-diabetic subtypes and / or constructing a model for predicting non-diabetic subtypes.

[0315] This application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the methods described above for predicting non-diabetic subtypes and / or constructing a model for predicting non-diabetic subtypes.

[0316] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0317] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0318] In another embodiment, diabetic patients typically rely on finger-prick glucose testing for daily blood glucose management. However, finger-prick glucose testing requires patient self-administration and only provides discrete, single-point blood glucose concentration data. In contrast, a Continuous Glucose Monitoring System (CGMS) can provide users with comprehensive, real-time blood glucose level data, reflecting the full, rich, and complex dynamic characteristics of blood glucose. Although some methods use time-series prediction algorithms for blood glucose prediction, these methods fail to fully consider the role of individual characteristics and external disturbances (such as specific dietary nutrients, exercise, etc.) in blood glucose dynamics, and therefore have certain limitations.

[0319] Some retrospective dietary advice systems provide recommendations based on postprandial blood glucose dynamics, but cannot offer pre-meal dietary advice. To better manage blood glucose levels in diabetic patients, further exploration and development of smarter and more personalized blood glucose prediction methods are needed to better guide patients' diets and lifestyles.

[0320] To address at least one of the aforementioned problems, an eighth aspect of this application provides a method for predicting postprandial blood glucose, the flowchart of which is shown in Figure 19, comprising:

[0321] S60: Obtain the subject's dynamic blood glucose characteristic data, pre-meal real-time blood glucose data, and at least one dietary structure information. The subject's dynamic blood glucose characteristic data is obtained based on a pre-trained model.

[0322] S70: Input the subject's dynamic blood glucose characteristics data, pre-meal real-time blood glucose data and at least one dietary structure information into the postprandial blood glucose prediction model, and output the subject's predicted postprandial blood glucose based on the postprandial blood glucose prediction model.

[0323] Specifically, continuous glucose concentration (CGM) data refers to data obtained from continuous monitoring of a subject's blood glucose concentration throughout the day, such as CGM data. Pre-trained models are obtained through unsupervised machine learning based on the CGM data of the target sample. Postprandial blood glucose prediction models are models used to predict a subject's postprandial blood glucose levels, pre-trained from the subject's dynamic blood glucose characteristics, historical dietary structure information, and historical blood glucose data for the corresponding target time zone.

[0324] In some implementation schemes, dynamic blood glucose sequence data refers to full-day time series data obtained by segmenting and processing continuous blood glucose monitoring data.

[0325] In some specific implementation schemes, the segmentation of dynamic glucose concentration (CGM) data refers to dividing the multi-day CGM data of the sample into daily segments. Each day's 288 complete measurements from 0:00 to 24:00 are considered a sequence. Incomplete sequences from the data collection process are filtered out to obtain a complete 24-hour blood glucose time series. For each time point, the blood glucose concentration value is uniformly discretized into a series of values ​​from G... min (40mg / ml) to G max (300mg / ml) G max -G min +1 means 261 blood glucose levels.

[0326] Specifically, for subjects with multiple days of CGM data, the subjects' multiple days of CGM data are divided into multiple days of dynamic blood glucose sequence data and input into the pre-trained model to obtain multi-day dynamic blood glucose feature data. The average of the subjects' multi-day dynamic blood glucose feature data is taken as the subjects' dynamic blood glucose feature data.

[0327] In some implementation schemes, historical dietary information includes five dimensions: calories (kcal), carbohydrates (g), protein (g), fat (g), and fiber (g). Based on historical dietary records, historical dietary information and historical blood glucose data for a target time zone matched with this information can be obtained. Specifically, the historical blood glucose data for the target time zone refers to blood glucose concentration data for a preset time period before and after meals.

[0328] In some specific implementation plans, the historical blood glucose data for the target time zone includes blood glucose concentration data from 1 hour before meals to 2 hours after meals.

[0329] In some implementation schemes, pre-meal real-time blood glucose data refers to blood glucose data within a preset time period before meals. In some specific implementation schemes, pre-meal real-time blood glucose data includes blood glucose concentration data 1 hour before meals.

[0330] In some implementations, the target sample includes multiple non-diabetic samples and multiple diabetic samples, and the multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0331] The diagnostic criteria for normal glucose tolerance samples are: when the subject's fasting blood glucose is <6.1mmol / L and the 2-hour postprandial blood glucose is <7.8mmol / L, it indicates that the body's ability to regulate blood glucose after eating is normal.

[0332] The diagnostic criteria for impaired glucose tolerance are: when a subject performs an oral glucose tolerance test, their fasting blood glucose is normal (less than 6.0 mmol / L), but their blood glucose 2 hours after oral glucose administration is between 7.8 and 11.1 mmol / L. This value is greater than the normal value, but does not meet the diagnostic criteria for diabetes. This state is impaired glucose tolerance.

[0333] The gold standard for diagnosing diabetes is the glucose tolerance test. According to the "Guidelines for the Prevention and Treatment of Type 2 Diabetes in China (2020 Edition)," the diagnostic criteria for diabetes include: typical symptoms of diabetes, plus fasting blood glucose ≥7.0 mmol / L, plus 2-hour post-glucose load blood glucose ≥11.1 mmol / L, or plus HbA1c (glycated hemoglobin) ≥6.5%.

[0334] This application uses a pre-trained model to learn the dynamic features and patterns of blood glucose fluctuations from a large amount of unlabeled CGM data in an unsupervised manner. It effectively captures and stores the dynamic blood glucose change features contained in CGM, and embeds the CGM data into a low-dimensional vector space to obtain dynamic blood glucose feature data. Combined with the subject's pre-meal real-time blood glucose data and dietary structure information, it can predict the subject's post-meal blood glucose, and then provide the subject with a personalized dietary prescription by combining simulated diet.

[0335] To implement the above-mentioned method for predicting postprandial blood glucose, the ninth aspect of this application provides a method for constructing a postprandial blood glucose prediction model, the flowchart of which is shown in Figure 20, including:

[0336] S600: Acquires the subject's dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone. The subject's dynamic blood glucose characteristics data are obtained based on a pre-trained model.

[0337] S700: A machine learning model is trained using the subject's dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone to obtain a postprandial blood glucose prediction model.

[0338] In this application, after pre-training on large-scale CGM data, the pre-trained model can learn the dynamic characteristics and patterns of blood glucose fluctuations from a large amount of unlabeled CGM data in an unsupervised manner, capture the dynamic characteristics of individual blood glucose, and then combine the individual's historical dietary structure information and the corresponding historical blood glucose data of the target time zone to construct a postprandial blood glucose prediction model.

[0339] In some specific implementation schemes, the steps for building a pre-trained model include:

[0340] S7001: Preprocess the dynamic blood glucose concentration data of the target sample to obtain the dynamic blood glucose sequence data of the target sample;

[0341] S7002: Input the dynamic blood glucose sequence data of the target sample into the initial model to obtain the dynamic blood glucose feature data of the target sample;

[0342] S7003: Mask the dynamic blood glucose sequence data of the target sample to obtain the masked dynamic blood glucose sequence data of the target sample, and input the masked dynamic blood glucose sequence data of the target sample into the initial model to obtain the masked dynamic blood glucose feature data of the target sample.

[0343] S7004: Determine the first training loss based on the dynamic blood glucose feature data of the target sample and the dynamic blood glucose feature data after masking the target sample. Update the parameters of the initial model based on the first training loss. Obtain the pre-trained model based on the initial model with updated parameters.

[0344] Specifically, the initial model refers to the untrained sequence data processing model. For the collected CGM data, this application constructs a deep self-attention transformation network model as the initial model and pre-trains it on the collected CGM data. Deep learning is used to effectively capture and store the dynamic blood glucose change patterns contained in the CGM, so that the computer system can understand, reason, and process this information. Based on this information, the dynamic blood glucose feature distribution data corresponding to different non-diabetic subtypes can be obtained to construct a predictive postprandial blood glucose model.

[0345] During the pre-training process, this application randomly sets a mask using dynamic blood glucose sequence data, inputs the masked sequence into the initial model, reconstructs the elements at the mask positions, learns the features and patterns of the dynamic blood glucose sequence, and calculates the loss function to update the parameters of the initial model.

[0346] In some implementations, the initial model includes an initial encoding layer and an initial prediction layer. The initial prediction layer consists of a multi-head self-attention mechanism layer and a feedforward neural network layer. Specifically, the initial encoding layer is used to encode the positions and elements in the dynamic blood glucose sequence data to obtain blood glucose encoded feature data. The initial prediction layer is used to further learn the patterns of the dynamic blood glucose encoded feature data and output dynamic blood glucose feature data.

[0347] Multi-head attention: The self-attention mechanism in the Transformer unit is extended to multiple attention heads, each of which can learn different attention weights to better capture different types of relationships. Multi-head attention allows the model to process different information subspaces in parallel.

[0348] A feedforward neural network is a fully connected network containing one hidden layer, consisting of two linear transformations separated by a non-linear activation function (typically ReLU) as the hidden layer. The first linear transformation projects the input representation into a higher-dimensional space using a learnable weight matrix. The non-linear activation passes the output of the first linear transformation through a non-linear activation function (e.g., ReLU), introducing non-linearity into the model and enabling it to capture complex patterns and relationships in the data. The second linear transformation projects the output of the activation function back into the original dimensional space through another learnable weight matrix.

[0349] In some specific implementations, the prediction layer consists of four transformer units, each of which consists of a multi-head self-attention module (MSA) and a feedforward neural network module (MLP).

[0350] Accordingly, the dynamic blood glucose sequence data of the target sample is input into the initial model to obtain the dynamic blood glucose feature data of the target sample, including:

[0351] Based on the initial coding layer, positional and elemental coding are performed on dynamic blood glucose sequence data to obtain blood glucose coding feature data;

[0352] The multi-head self-attention vector of blood glucose coding feature data is calculated based on the multi-head self-attention mechanism layer, and the combination of multi-head self-attention vectors is output after layer normalization.

[0353] The normalized output of the above layers is transformed based on the feedforward neural network layer, and then output is obtained through residual connections and layer normalization.

[0354] Specifically, after the initial encoding layer encodes the dynamic blood glucose sequence data to obtain blood glucose encoded feature data, the multi-head self-attention mechanism layer encodes the blood glucose encoded feature data into low-dimensional triples, namely query vector, key vector, and value vector. The query vector and key vector are multiplied by a dot product and then softmax is performed to obtain the coefficients of the value vector, which are the self-attention vectors of each head. After the self-attention vectors of each head are calculated, the final result is a linear combination of different value vectors. The linear combination results of different value vectors are concatenated and then normalized to obtain the output of the multi-head self-attention mechanism layer. Furthermore, the feedforward neural network layer transforms the output of the multi-head self-attention mechanism layer and outputs it in the form of residual connections and normalization.

[0355] Residual connections involve directly adding the input to the output, passing some information directly to the next layer, thus avoiding information loss. This connection method makes the network easier to train and helps improve model performance. Layer normalization is a normalization technique that normalizes features in the input of each layer, accelerating network convergence and improving the model's generalization ability.

[0356] Some implementation schemes utilize subjects' dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone to train machine learning models to obtain postprandial blood glucose prediction models, including:

[0357] The subjects' dynamic blood glucose characteristics, historical dietary structure information, and corresponding historical pre-meal blood glucose data are input into a machine learning model to obtain predicted blood glucose data for the target time zone.

[0358] The second training loss is determined based on the predicted blood glucose data and historical blood glucose data of the target time zone. The parameters of the machine learning model are updated based on the second training loss. The postprandial blood glucose prediction model is obtained based on the machine learning model with updated parameters.

[0359] In some specific implementations, the machine learning model includes an LSTM model. An LSTM model consists of an encoding layer, a long short-term memory network layer, and a prediction layer.

[0360] The basic process of an LSTM model includes:

[0361] Data preprocessing: First, the input data is preprocessed, including standardization and normalization, so that the model can better learn the features of the data.

[0362] Model Creation: In deep learning frameworks like Keras, LSTM models are typically created using the Sequential class. Then, LSTM layers and other necessary layers, such as Dense layers (fully connected layers), are added in the required order for output prediction. Network Structure Definition: In LSTM, the input at each time step passes through three main "gate" structures: the forget gate, the input gate, and the output gate. These gate structures, through specific calculation formulas and control mechanisms, determine which information should be retained and which should be forgotten. Forget Gate: Controls whether to forget the hidden cell state of the previous layer, determining with a certain probability which information needs to be retained. Input Gate: Responsible for processing the input at the current sequence position and updating the cell state. It consists of two parts: one part uses the sigmoid activation function to calculate the input gate value, and the other part uses the tanh activation function to calculate the candidate memory unit value. Output Gate: Calculates and outputs the hidden state of the current time step based on the updated cell state.

[0363] Model training: The model is trained using the preprocessed data. During training, the model learns how to adjust its parameters to minimize prediction error. This is typically achieved using the backpropagation algorithm and the gradient descent optimizer.

[0364] Model evaluation: After training, the model is evaluated using a validation set or test set to check its performance. Evaluation metrics may include accuracy, loss function value, etc.

[0365] Model Application: Once the model has been trained and evaluated and has achieved the expected performance requirements, it can be applied to practical tasks such as time series prediction and natural language processing.

[0366] In this application, the subject's dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical pre-meal blood glucose data are input into a machine learning model to obtain predicted post-meal blood glucose data, including:

[0367] The subjects' dynamic blood glucose feature data are embedded into a low-dimensional latent space to obtain dynamic blood glucose coding feature data;

[0368] The subjects' historical dietary structure information and corresponding historical pre-meal blood glucose data are combined into pre-meal blood glucose diet sequence data. The pre-meal blood glucose diet sequence data includes blood glucose diet sequences at multiple times before meals. Each blood glucose diet sequence includes blood glucose data and dietary structure information at that time.

[0369] For each moment before a meal, based on the subject's blood glucose coding dynamic feature data and the pre-meal blood glucose diet sequence at that moment, the predicted blood glucose coding dynamic feature data and the predicted blood glucose feature vector for the next moment are iteratively output. The predicted blood glucose feature vector is then linearly decoded to obtain the predicted blood glucose data.

[0370] For each moment after a meal, the system iteratively outputs the dynamic feature data of the blood glucose code and the predicted blood glucose diet sequence at that moment, based on the predicted blood glucose code dynamic feature data and the predicted blood glucose feature vector at that moment. The predicted blood glucose feature vector is linearly decoded to obtain the predicted blood glucose data. The predicted blood glucose diet sequence includes the predicted blood glucose data and dietary structure information at that moment.

[0371] Furthermore, a loss function can be obtained based on the mean square error of the predicted blood glucose data and the historical blood glucose data of the target time zone, and the model parameters can be updated based on the loss function.

[0372] In practical applications, by perturbing dietary structure information, the impact of different simulated diets on postprandial blood glucose dynamics can be predicted, thereby providing patients with personalized dietary prescription recommendations.

[0373] Accordingly, a tenth aspect of this application provides an apparatus for predicting postprandial blood glucose, as shown in FIG21, comprising:

[0374] First data acquisition module 60: used to acquire the subject's dynamic blood glucose characteristic data, pre-meal real-time blood glucose data and at least one dietary structure information, the subject's dynamic blood glucose characteristic data is obtained based on a pre-trained model;

[0375] Blood glucose prediction module 70: used to input the subject's dynamic blood glucose characteristics data, pre-meal real-time blood glucose data and at least one dietary structure information into the post-meal blood glucose prediction model, and based on the output of the post-meal blood glucose prediction model, predict the subject's post-meal blood glucose corresponding to each dietary structure information.

[0376] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0377] The postprandial blood glucose prediction model is obtained by training a machine learning model based on the subject's dynamic blood glucose characteristics data, historical dietary structure information, and historical blood glucose data of the corresponding target time zone. The historical blood glucose data of the target time zone includes historical pre-meal blood glucose data and historical postprandial blood glucose data.

[0378] Accordingly, the eleventh aspect of this application provides an apparatus for constructing a postprandial blood glucose prediction model, as shown in Figure 22, comprising:

[0379] The second data acquisition module 600 is used to acquire the subject's dynamic blood glucose characteristic data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone. The subject's dynamic blood glucose characteristic data is obtained based on a pre-trained model.

[0380] Model training module 700: Used to train a machine learning model to obtain a postprandial blood glucose prediction model by utilizing the subject's dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data of the target time zone;

[0381] The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

[0382] Specific limitations regarding the apparatus for predicting postprandial blood glucose and / or constructing a model for predicting postprandial blood glucose can be found in the limitations regarding the methods for predicting postprandial blood glucose and / or constructing a model for predicting postprandial blood glucose described above, and will not be repeated here. Each module in the aforementioned apparatus for predicting postprandial blood glucose and / or constructing a model for predicting postprandial blood glucose can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0383] In some embodiments, a computer device is provided, which can be a server 104 or a terminal 102, and its internal structure diagram can be as shown in Figure 5. The computer device includes a processor, memory, and a communication interface connected via a system bus. When the computer device is a terminal, it also includes a display screen and input devices connected to the system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting postprandial blood glucose and / or constructing a model for predicting postprandial blood glucose. The display screen of the computer device can be a liquid crystal display or an e-ink display. The input devices of the computer device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse, etc.

[0384] Those skilled in the art will understand that the structure shown in Figure 5 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or may combine certain components, or may have different component arrangements.

[0385] This application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-described methods for predicting postprandial blood glucose and / or constructing a model for predicting postprandial blood glucose.

[0386] This application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the methods described above for predicting postprandial blood glucose and / or constructing a model for predicting postprandial blood glucose.

[0387] This application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the methods described above for predicting postprandial blood glucose and / or constructing a model for predicting postprandial blood glucose.

[0388] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0389] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0390] The embodiments of this application will be described in detail below with reference to examples. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of this application. For experimental methods in the following embodiments where specific conditions are not specified, please refer to the guidelines given in this application, or follow experimental manuals or conventional conditions in the art, or follow the conditions recommended by the manufacturer, or refer to experimental methods known in the art.

[0391] In the specific embodiments described below, the measurement parameters involving raw material components may have slight deviations within the weighing accuracy range unless otherwise specified. Temperature and time parameters are subject to acceptable deviations due to instrument testing accuracy or operational precision.

[0392] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0393] Example 1

[0394] This embodiment provides a method for constructing a predictive model for glucose metabolism-related diseases. The flowchart is shown in Figure 6, and the specific steps of the method are as follows.

[0395] 1. Construct the CGM pre-trained model dataset

[0396] We acquired CGM data from 964 participants across 11 hospitals in 7 provinces and municipalities nationwide (a nationwide multicenter CGM dataset), and collected matching clinical information for each participant. This data covers three major groups with different glucose metabolism characteristics: normal glucose tolerance, impaired glucose tolerance, and diabetes. Each participant wore a CGM device for an average of 3 days, with blood glucose levels measured every 5 minutes. During data collection, participants' diets, exercise, and other lifestyle factors were strictly controlled, providing high-quality and comprehensive CGM data.

[0397] 2. Construction and training of pre-trained models

[0398] For the nationwide multi-center CGM data collected in step 1, the time series data is first segmented by day, with each day consisting of 288 complete measurements from 0:00 to 24:00 as a sequence. Incomplete sequences from the data collection process are filtered out to obtain a complete 24-hour blood glucose time series. For the blood glucose value measured at each time point, it is discretized into 261 blood glucose levels ranging from 40 mg / ml to 300 mg / dl. The complete 24-hour blood glucose concentration time series is transformed into a sequence S composed of blood glucose levels, ultimately resulting in the CGM pre-trained model corpus.

[0399] After obtaining the CGM-based corpus, this embodiment constructs and trains a deep self-attention transformation network with a Transformer architecture, which can learn the dynamic features and patterns of blood glucose fluctuations from a large amount of unlabeled CGM data in an unsupervised manner.

[0400] Specifically, the network consists of four transformer layers, each layer comprising a multi-head self-attention module (MSA) and a feedforward neural network module (MLP). The network takes a sequence S of length L as input, learns the pattern of blood glucose sequences before and after it, sets a mask, and predicts the blood glucose level at corresponding positions. First, the encoder encodes the positions and elements in the input sequence, transforming each position into a d-dimensional vector. Each sequence can be represented as... Next, the encoded sequence is trained using four transformer layers, each with eight self-attention heads. In each layer, the input Z of the previous layer... l-1 (Z 0 =S enc First, in the self-attention module, the data is encoded into triples (query vector Q, key vector K, and value vector V) via a linear transformation, i.e.:

[0401] Q = Zl-1 W Q K = Z l-1 W K V = Z l-1 W V ;

[0402] in, These are the learnable weight parameters in the three linear mapping layers, where k is the feature dimension of the triples. Subsequently, the self-attention mechanism (SA) can be expressed as:

[0403] MSA is an extension with m independent self-attention mechanism operations, mapping their joint output: MSA(Z l-1 )=[SA1(Z l-1 SA2(Z) l-1 );…;SA m ( Zl-1 )]W O ,in k is usually set to d / m.

[0404] Then, the MSA output is transformed by the MLP block, and normalized after residual connection as the layer output:

[0405] As shown in Figure 7, during the pre-training process, for a given unlabeled sequence S, some positions are first randomly set as masks to generate S. mask Then S mask The input is passed to the pre-trained model to obtain the output. By reconstructing the elements at the mask positions, the features and patterns of the input sequence are learned. The loss function during the pre-training stage can be expressed as:

[0406] Where nunk is the number of masks and s is the original element embedding representation at the mask position.

[0407] Using the output matrix S of the last layer out The average vector calculated by position is used as the feature vector of the sequence.

[0408] This pre-trained model learns individual blood glucose dynamics and patterns from a large amount of unlabeled data through self-supervised learning, thus mapping CGM data to a low-dimensional vector space and capturing its potential blood glucose dynamics.

[0409] 3. Collect fine-tuning dataset

[0410] In addition to nationwide multi-center CGM data, this embodiment collected CGM databases from published papers for fine-tuning the pre-trained model. For CGM data generated by different devices, it was segmented and filtered by day according to the same CGM data processing procedure, and padded to a sequence of length 288. The CGM data was matched with the patients' clinical diagnostic labels to construct fine-tuning datasets for different diagnostic tasks, as shown in Table 1. Normal glucose tolerance (NGT, FBG < 6.1 mmol / L, 2hPG < 7.8 mmol / L), impaired fasting glucose (IFG, 6.1 ≤ FBG < 7.0 mmol / L, 2hPG < 7.8 mmol / L), impaired glucose tolerance (IGT, FBG < 6.1 mmol / L, 7.8 ≤ 2hPG < 11.1 mmol / L), and combined glucose tolerance (CGI, 6.1 ≤ FBG < 7.0 mmol / L, 7.8 ≤ 2hPG < 11.1 mmol / L). FBG refers to fasting blood glucose, 2hPG refers to blood glucose 2 hours after a meal, and IFG and IGT are collectively referred to as impaired glucose regulation (IGR, i.e., prediabetes).

[0411] Table 1

[0412] 4. Fine-tuning the pre-trained model based on matched CGM data and clinical diagnosis.

[0413] This embodiment adds a task-specific prediction layer after the pre-trained model to fine-tune the parameters of the prediction layer and the parameters in the original model.

[0414] Specifically, for an input blood glucose sequence s, the pre-trained model encodes it as v S =Enc(S, Θ) Pretrain ), Θ Pretrain These are all the parameters in the pre-trained model. For each diagnostic task, the prediction layer consists of one fully connected layer and one softmax layer, outputting the predicted clinical diagnostic label, denoted as:

[0415] L S =Dec(v S Θ finetuning ) = softmax(NN(v S ));

[0416] The training loss is defined by the cross-entropy loss between the predicted and true labels, i.e.:

[0417] Where, p S L represents the true label distribution. S For the label distribution predicted by the model, during the training process, the pre-trained parameters Θ Pretrain With fine-tuning layer parameters Θ finetuning It is optimized at the same time.

[0418] In this embodiment, task-specific fine-tuning models were built and trained on multiple diagnostic tasks to achieve diagnostic tasks in multiple medical scenarios.

[0419] 5. Estimate the extent of impaired glucose metabolism and the risk of complications by combining physiological and pathological data.

[0420] To quantitatively estimate the extent of impaired glucose metabolism in patients from CGM data and predict the duration of diabetes and the risk of complications, this embodiment designs a multi-task learning model that combines clinical physiological and pathological data to estimate the extent of impaired glucose metabolism and the risk of complications, based on a pre-trained model. As shown in Figure 8, this model consists of a multi-layer linear encoding layer and multiple prediction layers added after the pre-trained model. The multi-layer linear encoding layer maps CGM data to a real number, which reflects the duration of diabetes and the risk of complications in patients.

[0421] Specifically, for an input blood glucose sequence S, the pre-trained model encodes it as v S =Enc(S, Θ) Pretrain First, it is encoded into a real number by a multilayer linear perceptron. Then, multiple task-specific prediction layers (composed of linear transformations and activation functions) decode and predict various clinical physiological and pathological data, including age, BMI, fasting blood glucose, fasting insulin, HOMA-IS, HOMA-B, 120-minute postprandial blood glucose, glycated hemoglobin, and high-density cholesterol. The loss function is defined by the mean squared error of the multi-task analysis, i.e.:

[0422] Among them, L s Based on the label definition of sample S, a normal glucose tolerance sample is defined as 0, a glucose tolerance impaired sample as 0.5, and a diabetic sample as 1. Dec j (C S ) represents the decoder for clinical physiological and pathological index j, O j (S) represents the true value of index j of sample s, and p is the control parameter between the two loss functions.

[0423] During training, some parameters Θ of the pre-trained model Pretrain The process is frozen, optimizing the multi-layer linear coding layer and task-specific prediction layers, and then outputting the results. As an estimate of the patient's impaired glucose metabolism and risk of complications.

[0424] Example 2

[0425] Based on the glucose metabolism-related disease prediction model constructed in Example 1, this example was tested on national multicenter CGM data and a publicly available dataset annotated with diabetic complications.

[0426] On the national multicenter CGM dataset, this embodiment uses the information of hospitals diagnosing individuals according to clinical diagnostic criteria as a third-party gold standard, and designs a variety of baseline control methods to compare diagnostic accuracy.

[0427] One baseline approach is to use statistically derived metrics based on a single CGM.

[0428] This embodiment uses the igluR package to calculate CGM-derived indicators for glucose control and glucose variability, including mean glucose (mean), standard deviation (SD), mean glycemic amplitude (MAGE), time within the target range (TIR), and time to hyperglycemia (TAR). TIR refers to the time (usually expressed in minutes) or percentage of the target range (typically 3.9–10.0 mmol / L, or 3.9–7.8 mmol / L) of glucose within 24 hours. Time within the target range (TIR) ​​is a new glycemic control indicator that has received considerable attention both domestically and internationally in recent years, and it has a strong correlation with the diagnosis and complications of diabetes. The "Guidelines for the Prevention and Treatment of Type 2 Diabetes in China (2020 Edition)" includes TIR in the glycemic control targets, and the international consensus on TIR recommends a TIR control target of >70% for patients with type 2 diabetes mellitus (T2DM). Another baseline approach is to use machine learning methods to predict CGM time series data.

[0429] This embodiment compares two commonly used neural network types, Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM), to construct a model for predicting glucose metabolism-related diseases. These two neural network types each have their own characteristics and are suitable for different application scenarios. MLP can handle some linearly inseparable problems, while LSTM can handle sequential data. In a binary classification scenario diagnosing T2D from IGR or NGT samples, compared with all baseline methods, the method for predicting diabetes and complications based on a pre-trained model of continuous glucose monitoring in this application showed the highest precision (0.88) and recall (0.86), as shown in Figure 9. The method in this application consistently outperforms diagnoses based on single indicators derived from CGM records, including mean glucose (mean), SD, MAGE, TIR, and TAR, and significantly outperforms machine learning methods based on LSTM and MLP.

[0430] Macrovascular or microvascular complications generally have a long course, and their etiologies and possible environmental influences are complex. Predicting type 2 diabetes complications solely from individual glucose monitoring data is a relatively difficult task. To test the performance of this application in predicting diabetes complications based on glucose monitoring data, this embodiment collected a publicly available dataset identifying macrovascular or microvascular complications published by Shanghai Fourth People's Hospital (Zhao, Q. et al. Chinese diabetes datasets for data-driven machine learning. Sci Data 10, 35 (2023).). Using the hospital-diagnosed complication information in this dataset as the gold standard, three prediction tasks were designed, including total complications, macrovascular complications, and microvascular complications.

[0431] The glucose metabolism-related disease prediction model of Example 1 achieved an accuracy of 0.8 in predicting microvascular complications and 0.7 in predicting macrovascular complications. This example also compared other machine learning models based on continuous blood glucose data with the glucose metabolism-related disease prediction model of this application. The glucose metabolism-related disease prediction model of Example 1 outperformed other machine learning models in both predicting microvascular and macrovascular complications. The comparison results of the accuracy in predicting microvascular complications are shown in Figure 10(A), and the comparison results of the accuracy in predicting macrovascular complications are shown in Figure 10(B).

[0432] Example 3

[0433] This embodiment provides a method for predicting non-diabetic subtypes, the flowchart of which is shown in Figure 15, and the specific steps of the method are as follows.

[0434] 1. Build and train the pre-trained model

[0435] (1) Construct fractals and obtain the reference distribution of fractals in the latent space.

[0436] We acquired CGM data from 964 participants across 11 hospitals in 7 provinces and municipalities nationwide (a nationwide multicenter CGM dataset), and collected matching clinical information for each participant. This data covers three major groups with different glucose metabolism characteristics: normal glucose tolerance, impaired glucose tolerance, and diabetes. Each participant wore a CGM device for an average of 3 days, with blood glucose levels measured every 5 minutes. During data collection, participants' diets, exercise, and other lifestyle factors were strictly controlled, providing high-quality and comprehensive CGM data.

[0437] (2) Construction and training of pre-trained models

[0438] For the nationwide multi-center CGM data collected in step 1, the time series data is first segmented by day, with each day consisting of 288 complete measurements from 0:00 to 24:00 as a sequence. Incomplete sequences from the data collection process are filtered out to obtain a complete 24-hour blood glucose time series. For the blood glucose value measured at each time point, it is discretized into 261 blood glucose levels ranging from 40 mg / ml to 300 mg / dl. The complete 24-hour blood glucose concentration time series is transformed into a sequence S composed of blood glucose levels, ultimately resulting in the CGM pre-trained model corpus.

[0439] After obtaining the CGM-based corpus, this embodiment constructs and trains a deep self-attention transformation network with a Transformer architecture, which can learn the dynamic features and patterns of blood glucose fluctuations from a large amount of unlabeled CGM data in an unsupervised manner.

[0440] Specifically, as shown in the figure, the network consists of four transformer layers, with each transformer layer comprising a multi-head self-attention module (MSA) and a feedforward neural network module (MLP). The network takes a sequence S of length L as input, learns the pattern of blood glucose sequences before and after the sequence, sets a mask, predicts the blood glucose level at the corresponding position, and calculates the model's loss function.

[0441] First, the encoder encodes the positions and elements in the input sequence, transforming each position in the sequence into a d-dimensional vector. Each sequence can be represented as... Next, the encoded sequence is trained using four transformer layers, each with eight self-attention heads. In each layer l, the input Z of the previous layer... l-1 (Z 0 =S enc First, in the self-attention module, the data is encoded into triples (query vector Q, key vector K, and value vector V) via a linear transformation, i.e.:

[0442] Q = Z l-1 W Q K = Z l-1 W K V = Z l-1 V V

[0443] in, These are the learnable weight parameters in the three linear mapping layers, where k is the feature dimension of the triples. Subsequently, the self-attention mechanism (SA) can be expressed as:

[0444] MSA is an extension with m independent self-attention mechanism operations, mapping their joint output: MSA(Z l-1 )=[SA1(Z l-1 SA2(Z) l-1 );…;SA m (Z l-1 )]W O ,in k is usually set to d / m.

[0445] Then, the MSA output is transformed by the MLP block, and normalized after residual connection as the layer output:

[0446] As shown in Figures 16 and 17, during the pre-training process, for a given unlabeled sequence S, some positions are first randomly set as masks to generate S. mask Then Smask The input is passed to the pre-trained model to obtain the output. By reconstructing the elements at the mask positions, the features and patterns of the input sequence are learned. The loss function during the pre-training stage can be expressed as:

[0447] Where, n unk is the number of masks, and x is the original element embedding representation at the mask position.

[0448] Using the output matrix S of the last layer out The average vector calculated by position is used as the feature vector of the sequence. This pre-trained model learns individual blood glucose dynamics and patterns from a large amount of unlabeled data through self-supervised learning, thus mapping CGM data to a low-dimensional vector space and capturing its potential blood glucose dynamics.

[0449] 2. Construct a non-diabetic subtyping model based on a CGM pre-trained model

[0450] (1) Construct a fractal model and obtain the reference distribution of each fractal in the latent space.

[0451] The model, pre-trained on a large amount of unlabeled data, can effectively encode the blood glucose data recorded by CGM into vectors in the latent space, capturing the dynamic features of individual blood glucose. The distance between the vectors in the latent space reflects the similarity of the dynamic features of blood glucose in the corresponding CGM data.

[0452] The CGM corpus, constructed from nationwide multi-center CGM data, is pre-trained and encoded into vectors in a 128-dimensional latent space. In this embodiment, the vectors corresponding to non-diabetic individuals are clustered to obtain a non-diabetic population genotyping based on a continuous glucose monitoring pre-trained model, denoted as P = {P...} t Let {t, t = 1, 2, ..., M}, where M is the number of clusters obtained from clustering. Each cluster t is represented by a vector representation of its corresponding sequence in the national multicenter dataset, depicting its distribution in the latent space.

[0453] P t ={v s ,T(s)=t}

[0454] Among them, v s Let T(s) be the vector representation of sample s, and T(s) be the fractal of sample s.

[0455] (2) Combining clinical data to characterize subtype clinical features and diabetes risk

[0456] Furthermore, by combining individual clinical characteristics matched with CGM data, and based on global and time-segmented glycemic dynamics and clinical data, the glycemic dynamics and clinicopathophysiological characteristics of each subtype are characterized. The risk of diabetes is then assessed by combining subtype-specific glucose metabolism characteristics and pathophysiological features.

[0457] 3. Provide diagnostic and treatment recommendations for non-diabetic subtypes based on CGM data.

[0458] (1) Map the input CGM data from the pre-trained model to the latent space.

[0459] For the input CGM data, this embodiment first segments it by day and converts it into a sequence, which is then transformed into a latent space vector by the pre-trained model described above. For a subject individual s with data from multiple days, the vector obtained by segmenting and mapping the individual's CGM data is... The mean value is used as the latent space vector representation of the subjects:

[0460] (2) Sample typing based on the representation of individual CGM data in the latent space

[0461] calculate The clustering in the latent space is taken as the average distance from the distribution of each morphology in the latent space to the clustering in the latent space. The morphology with the smallest distance is taken as the morphology of sample s, that is:

[0462] In practical applications, this embodiment combines the glucose metabolism characteristics and pathophysiological characteristics of different types to assess an individual's risk of diabetes and provide personalized diagnosis and treatment recommendations.

[0463] Example 4

[0464] This embodiment is based on the non-diabetic subtyping method provided in Embodiment 1, and was tested using nationwide multi-center CGM data. Hierarchical clustering was performed on the blood glucose dynamic feature vectors obtained from the pre-trained model of the nationwide multi-center CGM data, which divided the blood glucose dynamic change patterns of CGM into 6 groups, or 6 subtypes. Statistical analysis of the average blood glucose dynamic characteristics of these 6 subtypes revealed that the 6 groups of blood glucose dynamic change patterns effectively reflected the average blood glucose values, dynamic fluctuation amplitudes, and other characteristics of blood glucose dynamic changes at different stages, including fasting and postprandial periods, as shown in Table 2. Furthermore, the blood glucose dynamic change characteristics of these samples were correlated with matched clinical data, as shown in Table 3. This revealed that one normal subtype and five prediabetic subtypes could be subdivided, demonstrating a high degree of consistency between the clustering subtyping results and clinical data.

[0465] Table 2

[0466] Table 3

[0467] Comparing the above six subtypes with the glucotype classification proposed by the Stanford University team, the classification obtained in this embodiment is more refined. Overall, the classification in this embodiment can subdivide the moderate variability subtype obtained by glucotype classification into two groups (Pre_Ia, Pre_Ib), and the severe variability subtype obtained by glucotype classification into three groups (Pre_IIa, Pre_IIb, Pre_IIc), as shown in Table 4. Specifically, if clinical OGTT results are used as the third-party gold standard, compared to glucotype classification, the six subtypes in this embodiment can more accurately distinguish between normal glucose tolerance (NGT, FBG < 6.1 mmol / L, 2hPG < 7.8 mmol / L), impaired fasting glucose (IFG, 6.1 ≤ FBG < 7.0 mmol / L, 2hPG < 7.8 mmol / L), impaired glucose tolerance (IGT, FBG < 6.1 mmol / L, 7.8 ≤ 2hPG < 11.1 mmol / L), and combined impaired glucose tolerance. Tolerance (CGI, 6.1≤FBG<7.0mmol / L, 7.8≤2hPG<11.1mmol / L), where FBG refers to fasting blood glucose, 2hPG refers to blood glucose 2 hours after a meal, and IFG and IGT are collectively referred to as impaired glucose regulation (IGR, i.e., prediabetes), as shown in Table 5. As shown in Figure 18, compared with the glucotype classification, the classification of the six subtypes in this embodiment has a higher consistency with the classification in the OGTT. In Figure 18, ARI represents the Land index, and AMI represents the mutual information index.

[0468] Table 4

[0469] Table 5

[0470] Example 5

[0471] This embodiment provides a method for predicting postprandial blood glucose, the flowchart of which is shown in Figure 23. The specific steps of this method are as follows.

[0472] 1. Build and train the pre-trained model

[0473] (1) Construct fractals and obtain the reference distribution of fractals in the latent space.

[0474] We acquired CGM data from 964 participants across 11 hospitals in 7 provinces and municipalities nationwide (a nationwide multicenter CGM dataset), and collected matching clinical information for each participant. This data covers three major groups with different glucose metabolism characteristics: normal glucose tolerance, impaired glucose tolerance, and diabetes. Each participant wore a CGM device for an average of 3 days, with blood glucose levels measured every 5 minutes. During data collection, participants' diets, exercise, and other lifestyle factors were strictly controlled, providing high-quality and comprehensive CGM data.

[0475] (2) Construction and training of pre-trained models

[0476] For the nationwide multi-center CGM data collected in step 1, the time series data is first segmented by day, with each day consisting of 288 complete measurements from 0:00 to 24:00 as a sequence. Incomplete sequences from the data collection process are filtered out to obtain a complete 24-hour blood glucose time series. For the blood glucose value measured at each time point, it is discretized into 261 blood glucose levels ranging from 40 mg / ml to 300 mg / dl. The complete 24-hour blood glucose concentration time series is transformed into a sequence S composed of blood glucose levels, ultimately resulting in the CGM pre-trained model corpus.

[0477] After obtaining the CGM-based corpus, this embodiment constructs and trains a deep self-attention transformation network with a Transformer architecture, which can learn the dynamic features and patterns of blood glucose fluctuations from a large amount of unlabeled CGM data in an unsupervised manner.

[0478] Specifically, as shown in Figure 24, the network consists of four transformer layers, with each layer comprising a multi-head self-attention module (MSA) and a feedforward neural network module (MLP). The network takes a sequence S of length L as input, learns the pattern of blood glucose sequences before and after the sequence, sets a mask, predicts the blood glucose level at the corresponding position, and calculates the model's loss function.

[0479] First, the encoder encodes the positions and elements in the input sequence, transforming each position in the sequence into a d-dimensional vector. Each sequence can be represented as... Next, the encoded sequence is trained using four transformer layers, each with eight self-attention heads. In each layer l, the input Z of the previous layer... l-1 (Z 0 =S enc First, in the self-attention module, the data is encoded into triples (query vector Q, key vector K, and value vector V) via a linear transformation, i.e.:

[0480] Q = Z l-1 WQ K = Z l-1 W K V = Z l-1 W V ;

[0481] in, These are the learnable weight parameters in the three linear mapping layers, where k is the feature dimension of the triples. Subsequently, the self-attention mechanism (SA) can be expressed as:

[0482] MSA is an extension with m independent self-attention mechanism operations, mapping their joint output: MSA(Z l-1 )=[SA1(Z l-1 SA2(Z) l-1 );…;SA m (Z l-1 )]W O ,in k is usually set to d / m.

[0483] Then, the MSA output is transformed by the MLP block, and the residual connection is normalized to become the layer output:

[0484] During pre-training, for a given unlabeled sequence S, some positions are first randomly masked to generate S. mask Then S mask The input is passed to the pre-trained model to obtain the output. By reconstructing the elements at the mask positions, the features and patterns of the input sequence are learned. The loss function during the pre-training stage can be expressed as:

[0485] Where, n unk is the number of masks, and x is the original element embedding representation at the mask position.

[0486] Using the output matrix S of the last layer out The average vector calculated by position is used as the feature vector of the sequence. This pre-trained model learns individual blood glucose dynamics and patterns from a large amount of unlabeled data through self-supervised learning, thus mapping CGM data to a low-dimensional vector space and capturing its potential blood glucose dynamics.

[0487] 2. Collect matching dietary records and continuous glucose monitoring data.

[0488] This embodiment collected CGM data with dietary records, matched the pre- and post-meal blood glucose dynamics for each meal, and converted the dietary structure information into the nutritional components of the meal, including five dimensions: calories (kcal), carbohydrates (g), protein (g), fat (g), and fiber (g). For an individual's CGM records, we encode them using the pre-trained model trained in step 1, and take the average vector of records from multiple days as the feature vector of patient P, i.e.

[0489] in, This is the set of all 24-hour blood glucose sequences segmented from patient P's CGM data. Let S be the feature vector encoded by the pre-trained model for the blood glucose sequence S.

[0490] 3. Construct a postprandial blood glucose prediction model

[0491] This embodiment predicts postprandial blood glucose dynamics by combining the individual blood glucose dynamics reflected in the patient's past CGM data, pre-meal blood glucose, and dietary structure information.

[0492] Specifically, as shown in Figure 24, the model input consists of three parts: the CGM data of patient P is a feature vector encoded by the pre-trained model. Blood sugar one hour before meals Where t represents the number of blood glucose tests conducted by the CGM device within one hour, with t=12 at 5-minute intervals; and when the dietary structure information is available... The model first puts v P It is embedded from a linear layer into an l-dimensional latent space, i.e. Then combined with pre-meal blood glucose (G) B right Fine-tuning is used to reflect an individual's current blood glucose status, and dietary information is processed as an immediate input perturbation to iteratively predict postprandial blood glucose.

[0493] Specifically, the blood glucose sequence and dietary structure information are first combined into a continuous time series signal as input, and then defined as follows: Where T 1,. =G B ,T 2:6,t =M,T 2:6,1:t-1 =0. Definition have:

[0494] v k+1 ,o k+1 =LSTM(v k ,R .,k )

[0495] Iterate t times for 1≤k≤t to obtain This indicates the state of glucose metabolism before a meal, while Let D represent the immediate blood glucose dynamics after eating. The output vector o is decoded by a linear decoder to predict blood glucose levels.

[0496] Then, postprandial blood glucose is predicted iteratively. For k>t, we have:

[0497] in, The blood sugar predicted from the previous moment. The model's loss function is represented by the mean squared error between the predicted and actual blood glucose levels, i.e.

[0498] Among them G pred =(g k ) 1<k≤3t G P This represents the actual postprandial blood glucose level. The model predicts postprandial blood glucose levels using... get.

[0499] Furthermore, by inputting information from different simulated dietary structures into the model, postprandial blood glucose dynamics under different dietary structures can be predicted. To provide personalized dietary prescriptions for patients, for patient S, their existing CGM records are first encoded by the pre-trained model and extracted into a vector v representing their blood glucose dynamics. S Then, based on their immediate pre-meal blood glucose level (G), B By combining individual basal metabolic rate with fixed mealtime calories, adjusting the functional ratio of carbohydrates, proteins and fats, and simulating different dietary structure input models, as shown in Table 6, postprandial blood glucose is predicted under different dietary structures, and personalized dietary prescriptions are selected for patients based on postprandial blood glucose dynamics.

[0500] Table 6

[0501] Example 6

[0502] This embodiment tested the predictive performance of the postprandial blood glucose prediction model in Example 5, verifying the feasibility of the prediction method.

[0503] We collected a publicly available dataset from Shanghai Fourth People's Hospital containing matching dietary structure information and dynamic blood glucose monitoring data (Zhao, Q. et al. Chinese diabetes datasets for data-driven machine learning. Sci Data 10, 35 (2023)). By converting the dietary structure information in the dataset into nutrient components such as calories, carbohydrates, protein, fat, and dietary fiber, we matched dietary information with glycemic dynamics and trained and tested the model on this dataset. Specifically, we combined individual CGM data with pre-trained model-encoded glycemic dynamic features, pre-meal blood glucose data, and dietary perturbations in the latent space to predict postprandial blood glucose levels. The postprandial blood glucose concentration measured by dynamic blood glucose monitoring was used as the gold standard, and the Pearson correlation coefficient was calculated to quantify the accuracy of the prediction.

[0504] The baseline model LSTM has the same architecture as the postprandial blood glucose prediction model constructed in Example 6, but it does not use the blood glucose dynamic features encoded by the pre-trained model as input. Instead, it directly combines the individual's preprandial blood glucose data and dietary perturbations as input and outputs the predicted postprandial blood glucose.

[0505] As shown in Figure 25, the postprandial blood glucose prediction model constructed in Example 6 has accurate and robust postprandial blood glucose prediction performance. Throughout the 2-hour period (8 time points, each 15 minutes apart), the Pearson correlation coefficient with the actual blood glucose value is greater than 0.8. Furthermore, the postprandial blood glucose prediction model constructed in Example 5 significantly outperforms the baseline model LSTM, which has the same architecture but does not use the encoded blood glucose dynamics features from the pre-trained model as input, at all time points throughout the 2-hour period.

[0506] The embodiments described above merely illustrate several implementation methods of this application to facilitate a detailed understanding of the technical solutions of this application, but should not be construed as limiting the scope of protection of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Furthermore, it should be understood that after reading the above teachings of this application, those skilled in the art can make various alterations or modifications to this application, and the equivalent forms obtained also fall within the scope of protection of this application. It should also be understood that technical solutions obtained by those skilled in the art based on the technical solutions provided in this application through logical analysis, reasoning, or limited experimentation are all within the scope of protection of the appended claims. Therefore, the scope of protection of this patent application should be determined by the content of the appended claims, and the specification and drawings can be used to interpret the content of the claims.

Claims

1. A method for constructing a predictive model for glucose metabolism-related diseases, characterized in that, include: Obtain dynamic blood glucose concentration data of the first sample set, and perform machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the first sample set to obtain a pre-trained model; Obtain dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set; train the pre-trained model and the first fine-tuning layer based on the dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set to obtain a prediction model for impaired glucose metabolism; and / or We obtained dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from the third sample set. Based on the dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from the third sample set, we trained the pre-trained model and the second fine-tuning layer to obtain a model for predicting the degree of glucose metabolism impairment. The first and third sample sets include multiple samples with normal glucose tolerance, multiple samples with impaired glucose tolerance, and multiple samples with diabetes. The second sample set includes samples with normal glucose tolerance and samples with impaired glucose metabolism corresponding to the prediction task.

2. The method according to claim 1, characterized in that, The pre-trained model obtained by unsupervised machine learning based on the dynamic blood glucose concentration data of the first sample set includes: The dynamic blood glucose concentration data of the first sample set are preprocessed to obtain the continuous blood glucose sequence data of the first sample set; The continuous blood glucose sequence data of the first sample set is input into the initial model to obtain the dynamic blood glucose feature data of the first sample set; The continuous blood glucose sequence data of the first sample set is masked, and the masked dynamic blood glucose sequence data of the first sample set is input into the initial model to obtain the masked dynamic blood glucose feature data of the first sample set. The training loss of the initial model is determined based on the blood glucose dynamic feature data of the first sample set and the blood glucose dynamic feature data after masking the target sample. The parameters of the initial model are updated based on the training loss of the initial model. The pre-trained model is obtained based on the updated initial model parameters.

3. The method according to claim 2, characterized in that, The initial model includes a transformer model, which comprises an initial encoding layer, a multi-head self-attention mechanism layer, and a feedforward neural network layer.

4. The method according to any one of claims 1 to 3, characterized in that, The glucose metabolism impairment prediction model trained on the pre-trained model and the first fine-tuning layer based on dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set includes: The dynamic blood glucose concentration data of the second sample set is input into the pre-trained model to obtain the dynamic blood glucose feature data of the second sample set; The dynamic blood glucose feature data of the second sample set is input into the first fine-tuning layer to obtain the predicted clinical diagnosis label of the second sample set. Based on the predicted clinical diagnostic labels and clinical diagnostic labels of the second sample set, a first training loss is determined. The parameters of the pre-trained model and the parameters of the first fine-tuning layer are updated according to the first training loss. Based on the updated parameters of the pre-trained model and the updated parameters of the first fine-tuning layer, a prediction model for glucose metabolism impairment diseases corresponding to the prediction task is obtained.

5. The method according to any one of claims 1 to 4, characterized in that, The second fine-tuning layer includes a linear encoding layer and the prediction layer. The pre-trained model and the second fine-tuning layer are trained based on dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data from the third sample set to obtain a glucose metabolism impairment prediction model, including: The dynamic blood glucose concentration data of the third sample set is input into the pre-trained model to obtain the dynamic blood glucose feature data of the third sample set. The dynamic blood glucose feature data of the third sample set is input into the linear coding layer to obtain the glucose metabolism impairment feature data of the third sample set; Inputting the glucose metabolism impairment feature data of the third sample set into the prediction layer yields multiple predicted clinical physiological features for each sample in the third sample set; The second training loss is determined based on multiple predicted clinical physiological features, clinical diagnostic labels and multiple clinical physiological data for each sample in the third sample set. The parameters of the linear coding layer and the prediction layer are updated according to the second training loss. The glucose metabolism impairment prediction model is obtained based on the pre-trained model and the updated second fine-tuning layer parameters.

6. A method for predicting glucose metabolism-related diseases, characterized in that, include: Obtain dynamic blood glucose concentration data from the subjects; Input the subject's continuous glucose concentration data into the glucose metabolism impairment prediction model constructed by the method of any one of claims 1 to 5 to obtain the glucose metabolism impairment prediction result of the subject; and / or The subject's dynamic blood glucose concentration data are input into the glucose metabolism impairment prediction model constructed by the method of any one of claims 1 to 5 to obtain the predicted glucose metabolism impairment result of the subject.

7. An apparatus for constructing a predictive model for glucose metabolism-related diseases, characterized in that, include: Pre-training module: used to acquire dynamic blood glucose concentration data of the first sample set, and to obtain a pre-trained model by performing machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the first sample set. The first model fine-tuning module is used to acquire dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set, and to train the pre-trained model and the first fine-tuning layer based on the dynamic blood glucose concentration data and clinical diagnostic labels from the second sample set to obtain a prediction model for impaired glucose metabolism; and / or The second model fine-tuning module is used to acquire dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data of the third sample set. Based on the dynamic blood glucose concentration data, clinical diagnostic labels, and clinical physiological data of the third sample set, the pre-trained model and the second fine-tuning layer are trained to obtain a model for predicting the degree of glucose metabolism impairment. The first and third sample sets include multiple samples with normal glucose tolerance, multiple samples with impaired glucose tolerance, and multiple samples with diabetes. The second sample set includes samples with normal glucose tolerance and samples with impaired glucose metabolism corresponding to the prediction task.

8. A device for predicting glucose metabolism-related diseases, characterized in that, include: Data acquisition module: used to acquire dynamic blood glucose concentration data of the subjects; First prediction module: used to input the subject's dynamic blood glucose concentration data into the glucose metabolism impairment prediction model constructed by the method of any one of claims 1 to 5, and obtain the glucose metabolism impairment prediction result of the subject; and / or The second prediction module is used to input the subject's dynamic blood glucose concentration data into the glucose metabolism impairment prediction model constructed by the method of any one of claims 1 to 5, and obtain the prediction result of the subject's glucose metabolism impairment.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

11. A method for constructing a non-diabetic subtyping model, characterized in that, include: Obtain dynamic blood glucose concentration data of the target sample, and perform machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample to obtain a pre-trained model; Obtain the dynamic blood glucose feature data of multiple non-diabetic samples output by the pre-trained model, and cluster the dynamic blood glucose feature data of multiple non-diabetic samples to obtain the dynamic blood glucose distribution data of non-diabetic samples. Among them, the non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic characteristic data clusters and their corresponding non-diabetic subtypes; The target samples include multiple non-diabetic samples and multiple diabetic samples, and the multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

12. The method according to claim 11, characterized in that, The pre-trained model obtained by machine learning based on the dynamic blood glucose concentration data of the target sample in an unsupervised manner includes: The dynamic blood glucose concentration data of the target sample is preprocessed to obtain the dynamic blood glucose sequence data of the target sample; The dynamic blood glucose sequence data of the target sample is input into the initial model to obtain the dynamic blood glucose feature data of the target sample; The dynamic blood glucose sequence data of the target sample is masked to obtain the masked dynamic blood glucose sequence data. The masked dynamic blood glucose sequence data of the target sample is then input into the initial model to obtain the dynamic blood glucose feature data of the target sample. The training loss is determined based on the dynamic blood glucose feature data of the target sample and the dynamic blood glucose feature data after masking the target sample. The parameters of the initial model are updated based on the training loss. The pre-trained model is obtained based on the initial model with updated parameters.

13. The method according to claim 12, characterized in that, The initial model includes an initial encoding layer and an initial prediction layer, wherein the initial prediction layer includes a multi-head self-attention mechanism layer and a feedforward neural network layer.

14. The method according to any one of claims 11 to 13, characterized in that, The method further includes: Multiple clinical features corresponding to each blood glucose dynamic feature data cluster are obtained, and the risk of diabetes for each non-diabetic subtype is determined based on the multiple clinical features and the blood glucose fluctuation characteristics of each blood glucose dynamic feature data cluster.

15. A method for predicting non-diabetic subtypes, characterized in that, include: Obtain dynamic blood glucose concentration data of the sample to be tested; The non-diabetic subtype of the test sample is determined based on the dynamic blood glucose concentration data of the test sample; The method is based on a non-diabetic subtyping model, which is used to perform the following steps: The dynamic blood glucose concentration data of the sample to be tested is input into the pre-trained model to obtain the dynamic blood glucose characteristic data of the sample to be tested. The non-diabetic subtype of the test sample is determined based on the dynamic blood glucose characteristics data and the non-diabetic blood glucose distribution data of the test sample. The pre-trained model is obtained through unsupervised machine learning based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple normal glucose tolerance samples and multiple glucose tolerance samples. Damaged sample; The non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic feature data clusters, which are obtained by clustering the blood glucose dynamic feature data of multiple non-diabetic samples output by the pre-trained model.

16. The method according to claim 15, characterized in that, The determination of the non-diabetic subtype of the test sample based on the dynamic blood glucose characteristic data and non-diabetic blood glucose dynamic distribution data includes: Calculate the dynamic blood glucose characteristic data of the sample to be tested and the average distance of each dynamic blood glucose characteristic data cluster; The non-diabetic classification of the test sample is determined based on the non-diabetic classification represented by the blood glucose dynamic characteristic data cluster with the smallest average distance. Optionally, the pre-trained model includes a pre-trained encoding layer and a prediction layer, wherein the prediction layer includes a multi-head self-attention mechanism layer and a feedforward neural network layer.

17. An apparatus for constructing a non-diabetic subtyping model, characterized in that, include: Pre-training module: used to acquire dynamic blood glucose concentration data of the target sample, and to obtain a pre-trained model by performing machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample; Clustering module: used to obtain the dynamic blood glucose feature data of multiple non-diabetic samples output by the pre-trained model, and to cluster the dynamic blood glucose feature data of multiple non-diabetic samples to obtain the dynamic blood glucose distribution data of non-diabetic samples. Among them, the non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic characteristic data clusters and their corresponding non-diabetic subtypes; The target samples include multiple non-diabetic samples and multiple diabetic samples, and the multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

18. A method for predicting non-diabetic subtypes, characterized in that, include: Data acquisition module: used to acquire dynamic blood glucose concentration data of the sample to be tested; Non-diabetic genotyping module: used to determine the non-diabetic genotyping of the test sample based on the dynamic blood glucose concentration data of the test sample; The device is based on a non-diabetic subtyping model, which is used to perform the following steps: The dynamic blood glucose concentration data of the sample to be tested is input into the pre-trained model to obtain the dynamic blood glucose characteristic data of the sample to be tested. The non-diabetic subtype of the test sample is determined based on the dynamic blood glucose characteristics data and the non-diabetic blood glucose distribution data of the test sample. The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance. The non-diabetic blood glucose dynamic distribution data includes multiple blood glucose dynamic feature data clusters and their corresponding non-diabetic subtypes. These multiple blood glucose dynamic feature data clusters are obtained by clustering the blood glucose dynamic feature data of multiple non-diabetic samples output by a pre-trained model.

19. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 11 to 16.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 11 to 16.

21. A method for predicting postprandial blood glucose, characterized in that, include: The subjects' dynamic blood glucose characteristics data, pre-meal real-time blood glucose data, and at least one dietary structure information are obtained, wherein the subjects' dynamic blood glucose characteristics data are obtained based on a pre-trained model; The subject's dynamic blood glucose characteristics data, pre-meal real-time blood glucose data, and at least one dietary structure information are input into the post-meal blood glucose prediction model, and the predicted post-meal blood glucose of the subject is output based on the post-meal blood glucose prediction model. The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample. The target sample includes multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance. The postprandial blood glucose prediction model is obtained by training a machine learning model based on the subject's dynamic blood glucose characteristics data, historical dietary structure information, and historical blood glucose data of the corresponding target time zone. The historical blood glucose data of the target time zone includes historical pre-meal blood glucose data and historical postprandial blood glucose data.

22. The method according to claim 21, characterized in that, The method includes: Dietary prescription recommendations are provided based on the subject's predicted postprandial blood glucose levels; Optionally, the machine learning model includes a long short-term memory network model; Optionally, the machine learning model includes an encoding layer, a long short-term memory network layer, and a prediction layer.

23. A method for constructing a postprandial blood glucose prediction model, characterized in that, include: The subjects' dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone were obtained. The subjects' dynamic blood glucose characteristics data were obtained based on a pre-trained model. A machine learning model was trained using the subjects' dynamic blood glucose characteristics data, historical dietary structure information, and historical blood glucose data of the corresponding target time zone to obtain a postprandial blood glucose prediction model. The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

24. The method according to claim 23, characterized in that, The steps for constructing the pre-trained model include: The dynamic blood glucose concentration data of the target sample is preprocessed to obtain the dynamic blood glucose sequence data of the target sample; The dynamic blood glucose sequence data of the target sample is input into the initial model to obtain the dynamic blood glucose feature data of the target sample; The dynamic blood glucose sequence data of the target sample is masked to obtain the masked dynamic blood glucose sequence data of the target sample. The masked dynamic blood glucose sequence data of the target sample is then input into the initial model to obtain the dynamic blood glucose feature data of the target sample. The first training loss is determined based on the dynamic blood glucose feature data of the target sample and the dynamic blood glucose feature data after masking the target sample. The parameters of the initial model are updated based on the first loss and the pre-trained model is obtained based on the initial model with updated parameters.

25. The method according to claim 24, characterized in that, The initial model includes an initial encoding layer and an initial prediction layer, wherein the initial prediction layer includes a multi-head self-attention mechanism layer and a feedforward neural network layer.

26. The method according to claim 23, characterized in that, The postprandial blood glucose prediction model was obtained by training a machine learning model using the subjects' dynamic blood glucose characteristics data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone. This model includes: The subjects' dynamic blood glucose characteristics, historical dietary structure information, and corresponding pre-meal blood glucose data are input into a machine learning model to obtain predicted blood glucose data for the target time zone. The second training loss is determined based on the postprandial blood glucose prediction data of the target time zone and the historical blood glucose data of the target time zone. The parameters of the machine learning model are updated according to the second training loss. The postprandial blood glucose prediction model is obtained based on the machine learning model with updated parameters. Optionally, the machine learning model includes a long short-term memory network model; The machine learning model includes an encoding layer, a long short-term memory network layer, and a prediction layer.

27. A device for predicting postprandial blood glucose, characterized in that, include: First data acquisition module: used to acquire the subject's dynamic blood glucose characteristic data, pre-meal real-time blood glucose data and at least one dietary structure information, wherein the subject's dynamic blood glucose characteristic data is obtained based on a pre-trained model; Blood glucose prediction module: used to input the subject's dynamic blood glucose characteristics data, pre-meal real-time blood glucose data and at least one dietary structure information into the post-meal blood glucose prediction model, and based on the output of the post-meal blood glucose prediction model, to predict the subject's post-meal blood glucose. The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target sample. The target sample includes multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance. The postprandial blood glucose prediction model is obtained by training a machine learning model based on the subject's dynamic blood glucose characteristics data, historical dietary structure information, and historical blood glucose data of the corresponding target time zone. The historical blood glucose data of the target time zone includes historical pre-meal blood glucose data and historical postprandial blood glucose data.

28. An apparatus for constructing a postprandial blood glucose prediction model, characterized in that, include: The second data acquisition module is used to acquire the subject's dynamic blood glucose characteristic data, historical dietary structure information, and corresponding historical blood glucose data for the target time zone. The subject's dynamic blood glucose characteristic data is obtained based on a pre-trained model. Model training module: used to train a machine learning model using the subject's dynamic blood glucose characteristics data, historical dietary structure information and corresponding historical blood glucose data of the target time zone to obtain a postprandial blood glucose prediction model. The postprandial blood glucose prediction model is used to predict the subject's postprandial blood glucose corresponding to the dietary structure information. The pre-trained model is obtained by machine learning in an unsupervised manner based on the dynamic blood glucose concentration data of the target samples. The target samples include multiple non-diabetic samples and multiple diabetic samples. The multiple non-diabetic samples include multiple samples with normal glucose tolerance and multiple samples with impaired glucose tolerance.

29. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 21 to 26.

30. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 21 to 26.

Images