Diabetes management method and system based on digital twinning and storage medium

Abstract

The invention relates to the technical field of health management, in particular to a diabetes management method and system based on digital twinning and a storage medium, and the method comprises the steps: obtaining basic physiological architecture parameters; building a digital twinborn model based on the basic physiological architecture parameters in combination with an HDT architecture; collecting real-time health data; processing the real-time health data as an input digital twin model, and predicting a blood glucose change trend to output a blood glucose prediction result; performing data mining on the input and the output to obtain potential health modes and abnormal data, and generating an interpretation event for a blood glucose prediction result based on the health modes and the abnormal data; the optimal insulin infusion amount is simulated and predicted in the digital twin model based on the blood glucose prediction result and the interpretation event; and sending the optimal insulin infusion amount to an insulin pump to adjust the injection metering. The method has the effect of providing personalized medication schemes according to individual differences of patients.

Description

Technical Field

[0001] The present application relates to the technical field of health management, and in particular to a diabetes management method, system and storage medium based on digital twins. Background Art

[0002] The management of elderly patients with type 2 diabetes faces multiple challenges, including comorbidities, multiple medications, reduced activity levels, increased dietary influences, variability in nutrient absorption with age, hormonal changes, etc., which make blood glucose levels (BGLs) highly unpredictable and variable. Given the individual differences among elderly patients with type 2 diabetes, personalized treatment approaches are needed to accurately administer medication, especially insulin infusion.

[0003] Current treatments for type 2 diabetes include: Oral hypoglycemic agents (OHAs): For newly diagnosed or non-insulin-dependent patients, oral hypoglycemic agents are the recommended initial treatment; Insulin infusion: Over time, many people with type 2 diabetes eventually become dependent on insulin therapy, which can be delivered manually or automatically. The artificial pancreas (AP) system continuously monitors interstitial glucose levels (IGL) and uses a control algorithm to infuse insulin to maintain blood glucose in the range of 70–180 mg / dL.

[0004] However, traditional insulin therapy and many existing AP systems lack consideration of patient-specific data, resulting in insufficient personalization of treatment plans. Different patients have different individual differences and different complications, medication conditions, living habits, etc., and it is difficult to achieve individualized blood sugar prediction and real-time decision analysis. This makes it impossible for the existing solutions in the technology to provide personalized and accurate management plans and give precise and scientific insulin dosages, which not only leads to a deterioration in the patient's health management effect, but also fails to control medication costs. Summary of the invention

[0005] In order to provide personalized medication plans based on individual differences among patients, the present application provides a diabetes management method, system and storage medium based on digital twins.

[0006] In the first aspect, the present application provides a diabetes management method based on digital twins, which adopts the following technical solutions: A diabetes management method based on digital twins, comprising the following steps: Obtain basic physiological architecture parameters; Building a digital twin model based on the basic physiological architecture parameters combined with the HDT architecture; Collect real-time health data, including blood sugar, diet, activity, and work and rest; The real-time health data is used as input to the digital twin model for processing, and the blood sugar change trend is predicted to output a blood sugar prediction result; Performing data mining on the input and output to obtain potential health patterns and abnormal data, and generating explanation events for the blood glucose prediction results based on the health patterns and the abnormal data; Simulating and predicting an optimal insulin infusion amount in the digital twin model based on the blood glucose prediction result and the explained event; The optimal insulin infusion amount is sent to the insulin pump to adjust the injection dosage.

[0007] In some embodiments, building a digital twin model based on the basic physiological architecture parameters combined with the HDT architecture includes the following steps: Generating a twin model based on the basic physiological architecture parameters, wherein the twin model includes a prediction model and a verification model; Among them, both the prediction model and the verification model build a virtual human body framework based on basic physiological information, add a virtual human body information flow in the virtual human body framework based on historical medication information and complication information, and generate a virtual metabolic flow and a virtual pathological flow based on the virtual human body information flow.

[0008] In some embodiments, the real-time health data is used as input to the digital twin model for processing, and the blood sugar change trend is predicted to output a blood sugar prediction result, including the following steps: Based on the real-time health data, the current blood sugar status and time-space nodes are obtained, and the time-space nodes include before eating, after eating, before activity, after activity, before work and rest, and after work and rest; Integrate the current blood sugar state and the spatiotemporal nodes into a test sample set and upload it to the prediction model, and obtain a first blood sugar change trend based on an LSTM network; Adding random impact event samples to the test sample set to obtain a verification sample set, uploading the verification sample set to the verification model, and obtaining a second blood sugar change trend based on the LSTM network; Calculate the predicted difference sensitivity based on the first blood sugar change trend and the second blood sugar change trend; When the difference sensitivity is lower than a preset value, the first blood sugar change trend is used as the blood sugar prediction result and output; when the difference sensitivity is not lower than the preset value, the first blood sugar change trend is fitted with the second blood sugar change trend and the fitting result is used as the blood sugar prediction result and output.

[0009] In some embodiments, data mining is performed on the input and output to obtain potential health patterns and abnormal data, including the following steps: Collect input data and output data and extract input features and output features based on feature engineering; Acquire a global behavior path between the input feature and the output feature through a decision tree path, and acquire a decision logic represented by the global behavior path based on a local interpretation library; Based on the decision logic, a health association rule between the input data and the output data is identified to generate a health mode, and abnormal values ​​reflected by the data under the health mode are obtained, and data with abnormal values ​​higher than a preset value are defined as abnormal data.

[0010] In some of the embodiments, after data mining the input and output to obtain potential health patterns and abnormal data, the following steps are also included: If the abnormal data corresponding to the prediction model and the verification model under the same health mode are different; Determine the same abnormal data as accurate abnormal data, and determine the different abnormal data as questionable abnormal data; Matching the suspected abnormal data to obtain corresponding input data based on the health association rule, and regenerating a prediction sample set and a verification sample set including at least the input data for input; Secondary abnormal data are mined according to the output data of the prediction model and the verification model, and the secondary abnormal data are respectively compared with the suspected abnormal data to determine the credibility of the suspected abnormal data.

[0011] In some embodiments, generating an explanation event for the blood glucose prediction result based on the health pattern and the abnormal data comprises the following steps: evaluating the interpretability of the associated input features and output features; Determining an influence coefficient between the associated input feature and the output feature; The explanation event corresponding to the blood glucose prediction result is obtained based on the local explanation library, and the explanation event includes positive input features, negative input features, and corresponding single feature contributions and multiple interactions.

[0012] In some of the embodiments, generating the explanation event further includes the following steps: Determining whether there is a difference between the explanation event corresponding to the prediction model and the explanation event corresponding to the verification model; If so, it is determined that the difference is associated with the random sample of the influencing events in the verification model; When there is a correlation, the feasible trend of the influencing event sample is determined based on historical data and empirical algorithms, and when the feasible trend is greater than a preset value, the explanatory event corresponding to the difference is deemed valid.

[0013] In some of the embodiments, the following steps are also included: When the insulin pump performs an injection based on the optimal insulin infusion amount, a monitoring task of a preset duration is generated to detect changes in blood glucose levels; The blood sugar prediction result is fed back for verification and marked based on the blood sugar level change, and the mark includes a normal mark and an abnormal warning mark.

[0014] In the second aspect, the present application provides a diabetes management system based on digital twins, which adopts the following technical solutions: A diabetes management system based on digital twins, including: Basic parameter upload module, used to upload basic physiological structure parameters; An HDT architecture module, used to build a digital twin model based on the basic physiological architecture parameters combined with the HDT architecture; IoT devices are used to collect real-time health data, including blood sugar, diet, activity, and work and rest; A prediction module, used to process the real-time health data as input to the digital twin model, and predict the blood sugar change trend to output a blood sugar prediction result; A data diagnosis module, configured to perform data mining on the input and output to obtain potential health patterns and abnormal data, and generate explanation events for the blood glucose prediction results based on the health patterns and the abnormal data; A control module, configured to simulate and predict an optimal insulin infusion amount in the digital twin model based on the blood glucose prediction result and the explanation event; An insulin pump is used to adjust the injection dosage according to the optimal insulin infusion amount.

[0015] In some of these embodiments, In a third aspect, the present application provides a storage medium, which adopts the following technical solution: A storage medium stores at least one instruction, at least one program, a code set or an instruction set, wherein the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the above method.

[0016] The technical solution provided by the embodiments of the present application has the following technical effects: A digital twin model is built through basic physiological structure parameters, and the health pattern and anomalies in the blood sugar prediction results are analyzed through the digital twin model and real-time blood sugar prediction results as well as the mining of the model's input and output data features. Finally, the optimal insulin infusion volume is comprehensively predicted based on the explanation events of the integrated abnormal data, and the insulin pump is intelligently adjusted. In this way, users do not need to manually design the infusion volume, and at the same time, users can get more accurate and scientific medication dosages, thereby improving health management effects and controlling medication costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the steps of a diabetes management method based on digital twins provided in an embodiment of the present application.

[0018] Figure 2 This is a module connection diagram of a diabetes management system based on digital twins provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] To more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. However, it should be understood by those of ordinary skill in the art that the present application can be implemented without these details. In some cases, in order to avoid unnecessary descriptions that make the various aspects of the present application obscure, the well-known methods, processes, systems, components and / or circuits that have been described at a higher level will not be described in detail. For those of ordinary skill in the art, it is obvious that various changes can be made to the embodiments disclosed in the present application, and without departing from the principles and scope of the present application, the general principles defined in the present application can be applied to other embodiments and application scenarios. Therefore, the present application is not limited to the embodiments shown, but conforms to the broadest scope consistent with the scope claimed for protection of the present application.

[0020] It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in each embodiment of the present invention described below can be combined with each other as long as there is no conflict between them.

[0021] In the description of this application, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used to distinguish the technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0022] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples.

[0023] like Figure 1 As shown, the embodiment of the present application discloses a diabetes management method based on digital twins, comprising the following steps: S100, obtaining basic physiological structure parameters.

[0024] Physiological architecture parameters represent the basic physiological information of individual patients, which are used to generate and serve as the basic information flow for building a human digital twin model in the future.

[0025] Basic physiological structure parameters include height and weight, age, historical medical records, historical medication records, complications, etc.

[0026] S200, builds a digital twin model based on basic physiological architecture parameters combined with HDT architecture.

[0027] HDT means human digital twin technology. Health digital twin (HDT) is a virtual representation of a real individual that can be used to simulate human physiology, diseases, and drug effects. It is generated using multimodal individual patient data, group data, and real-time input of patient and environmental variables.

[0028] The digital twin model is constructed by using the basic physiological architecture parameters obtained above and the HDT structure. The model is used to simulate blood sugar trends, medication effects, etc. through subsequent real-time monitoring of blood sugar data and other health data of the real human body, and to generate personalized medication plans through the simulation results of virtual human bodies of different individuals.

[0029] S300 collects real-time health data, including blood sugar, diet, activity, and daily routine.

[0030] Real-time health data is collected by IoT devices based on IoT technology, such as health watches and blood glucose meters, which can detect the user's blood sugar fluctuations in real time and record information uploaded by the user, such as food intake, dietary structure, exercise type, exercise time, exercise volume, sleep time, etc.

[0031] Real-time health data serves as input parameters for subsequent blood sugar predictions through the digital twin model.

[0032] S400 processes the real-time health data as input to the digital twin model, and predicts the blood sugar change trend to output the blood sugar prediction result.

[0033] The collected real-time health data is input into the digital twin model and processed and analyzed with the corresponding neural network algorithm to predict the blood sugar change trend under different situations.

[0034] S500, data mining is performed on the input and output to obtain potential health patterns and abnormal data, and an explanation event for the blood sugar prediction result is generated based on the health patterns and abnormal data.

[0035] Leverage pattern discovery and explainable AI techniques to analyze data and provide insights into patient conditions, perform pattern recognition and interpret events.

[0036] The health model is characterized by the decision-making logic relationship between input and output features in the entire data mining process, such as the relationship between what medicine and what symptom improvement, the relationship between blood sugar fluctuations and what behavior, etc.

[0037] Abnormal data include data with abnormal blood sugar fluctuations, data with low credibility, etc.

[0038] Explanatory events are characterized by explanatory content for anomalies or key events identified by the model, such as understanding why a patient's blood sugar level suddenly rises, why a certain dose of insulin is more effective, etc.

[0039] S600, simulate and predict the optimal insulin infusion amount in the digital twin model based on the blood glucose prediction results and explanation events.

[0040] Based on the real-time blood glucose prediction results and corresponding explanations obtained from the patient model, the learning-based model predictive control (LB-MPC) algorithm is used to calculate the personalized insulin infusion amount. The blood glucose prediction results are used as the prediction basis of the machine, and the corresponding explanation events are used as the prediction conditions of the machine, so as to generate the optimal prediction results of the dosage before each insulin infusion.

[0041] S700, sending the optimal insulin infusion amount to the insulin pump to adjust the injection measurement.

[0042] The system sends the optimal insulin infusion volume to the insulin pump, which uses the infusion volume to automatically adjust the dosage of the drug infusion task that needs to be performed this time.

[0043] Through the above steps, a digital twin model is built through basic physiological structure parameters, and the digital twin model and real-time blood sugar prediction results are used as well as the input and output data features of the model to analyze the anomalies in health patterns and blood sugar prediction results. Finally, the optimal insulin infusion volume is comprehensively predicted based on the explanation events of the integrated abnormal data, and the insulin pump is intelligently adjusted. In this way, users do not need to manually design the infusion volume, and users can get more accurate and scientific medication dosages, thereby improving health management effects and controlling medication costs.

[0044] In some other embodiments, building a digital twin model based on basic physiological architecture parameters combined with HDT architecture includes the following steps: S210, generating a twin model based on basic physiological architecture parameters, wherein the twin model includes a prediction model and a verification model.

[0045] When generating a digital twin model based on basic physiological architecture parameters, two independent models are generated. The two models serve as a prediction model for blood glucose data and a verification model for verifying the prediction model, respectively. In this way, in the subsequent process of blood glucose prediction and data analysis, the two twin models are trained and calculated synchronously, and the two models are cross-validated after each step to improve processing accuracy.

[0046] S220, both the prediction model and the verification model build a virtual human body framework based on basic physiological information, add a virtual human body information flow to the virtual human body framework based on historical medication information and complication information, and generate a virtual metabolic flow and a virtual pathological flow based on the virtual human body information flow.

[0047] The two models use the same data in the construction process. Both models build a virtual human body framework through basic physiological information, and add virtual human body information flow to the virtual human body framework through medical data, medication data, complication information, etc. as the physiological context of the virtual human body. The virtual human body information flow serves as the metabolic flow and pathological flow information of the virtual human body. The metabolic flow and pathological flow are used to match the physiological parameters of the virtual human body model with the real human body model. Later, when performing blood sugar prediction and medication dosage prediction, this information serves as a reference and prediction condition for blood sugar fluctuations, drug resistance, drug receptivity, etc.

[0048] In other embodiments, real-time health data is processed as an input to a digital twin model, and a blood sugar change trend is predicted to output a blood sugar prediction result, including the following steps: S310, obtaining the current blood sugar status and time-space nodes based on real-time health data, the time-space nodes including before eating, after eating, before activity, after activity, before work and rest, and after work and rest.

[0049] The current blood sugar status is obtained in real time by detecting blood sugar data, and time-space nodes are generated based on the uploaded real-time health data and current events. For example, after uploading a meal, it is recognized as after a meal, etc. At different time-space nodes, the changing trend and change logic of blood sugar are different. For example, after a meal, blood sugar will increase within 4 hours, and blood sugar will continue to decrease after exercise, etc.

[0050] S320, integrating the current blood sugar state and the spatiotemporal nodes into a test sample set and uploading it to the prediction model, and obtaining a first blood sugar change trend based on the LSTM network.

[0051] The current blood sugar value and time-space nodes are integrated into a test sample set and uploaded to the prediction model. The LSTM network is used to predict the blood sugar change trend in the future, and the result is defined as the first blood sugar change trend.

[0052] S330, adding random impact event samples to the test sample set to obtain a verification sample set, uploading the verification sample set to the verification model, and obtaining a second blood sugar change trend based on the LSTM network.

[0053] Add random impact event samples to the test sample set. The impact event samples are characterized by parameters that have not occurred yet but are likely to occur in the future and may affect blood sugar data or human health status, such as exercise, diet, sleep, etc.

[0054] By adding the influencing event samples, a verification sample set is obtained, and the verification sample set is uploaded to the verification model. The LSTM network is also used to predict the blood sugar change trend in the future, and the result is defined as the second blood sugar change trend.

[0055] S340, calculating the predicted difference sensitivity based on the first blood sugar change trend and the second blood sugar change trend.

[0056] Because the input contents of the prediction model and the verification model are different, and the difference in the input contents is a random influencing event sample, when there is a predicted difference between the first blood sugar change trend and the second blood sugar change trend, it means that the difference is caused by the added influencing event sample. In order to independently consider the influencing event sample to determine whether it has a highly sensitive impact on the prediction logic of blood sugar changes, it is necessary to predict the sensitivity through the difference between the two change trends.

[0057] When there are large changes in the amplitude, maximum value, minimum value, mean value, etc. of two changing trends, it is considered that their difference sensitivity is large.

[0058] S350, when the difference sensitivity is lower than the preset value, the first blood sugar change trend is used as the blood sugar prediction result and output; when the difference sensitivity is not lower than the preset value, the first blood sugar change trend is fitted by the second blood sugar change trend and the fitting result is used as the blood sugar prediction result and output.

[0059] When the difference sensitivity is lower than the preset value, it means that the random impact event sample has little effect on the overall prediction logic of the blood sugar change trend. At this time, the prediction results of the first blood sugar change trend and the second blood sugar change trend are verified to match. At this time, the first blood sugar change trend generated by the prediction model can be used as the blood sugar prediction result and output.

[0060] When the difference sensitivity is higher than the preset value, it means that the random impact event samples have a high degree of contradiction between the overall prediction logic of the blood glucose change trend, which means that the current prediction algorithm of the prediction model has large deviations in the prediction results under different scenarios, and it cannot be applied to the state where there are random impact events. Then its own prediction logic and parameters are defective. Therefore, it is necessary to fit the second blood glucose change trend and the first blood glucose change trend, and use the fitting result as the blood glucose prediction result.

[0061] The fitting is characterized by matching the data of each point on the two prediction curves for similarity, fitting the same or highly similar values ​​into one value, and taking the median or Bayesian prediction for two values ​​with low similarity and large deviation to determine the value with a higher probability of being the correct value and fitting it as the target point.

[0062] In some other embodiments, data mining is performed on the input and output to obtain potential health patterns and abnormal data, including the following steps: S410, collecting input data and output data and extracting input features and output features based on feature engineering.

[0063] Feature engineering is an important step in data preprocessing, which aims to extract useful features from raw data to better represent the data.

[0064] Input feature extraction: Depending on the type of input data, the following features can be extracted: Numerical features: such as age, weight, height, blood sugar, etc.

[0065] Category characteristics: such as gender, disease type, living habits, etc.

[0066] Text features: Keywords or phrases extracted from text information such as medical reports and medical records.

[0067] Output feature extraction: Output features usually include the severity of the disease, health status level, etc.

[0068] S420, obtaining a global behavior path between input features and output features through a decision tree path, and obtaining a decision logic represented by the global behavior path based on a local interpretation library.

[0069] Decision tree is a commonly used machine learning algorithm that can classify or regress output features based on input features.

[0070] The global behavior path refers to the path from the root node to the leaf node in the decision tree, which represents the relationship between the input features and the output features. By analyzing the path in the decision tree, we can get the global behavior pattern between the input features and the output features.

[0071] The local explanation library refers to a series of rules and knowledge bases used to explain the paths in the decision tree. Based on the local explanation library, we can convert the global behavior path into easy-to-understand decision logic, such as "If you currently eat x grams of y food, it may cause blood sugar to rise to z value."

[0072] S430, based on the decision logic, identifying the health association rules between the input data and the output data to generate a health mode, and obtaining abnormal values ​​reflected by the data under the health mode, and defining data with abnormal values ​​higher than a preset value as abnormal data.

[0073] Choose appropriate XAI technology, such as LIME (Local Interpretable Model-Sensitive Explanations), SHAP (SHapleyAdditive exPlanations), etc. Based on the decision logic, we can identify the association rules between input data and output data, such as "the relationship between current blood sugar and the use of x food".

[0074] These association rules can be used to generate health patterns, for example, to develop preventive measures for patients with type 2 diabetes who are 70 years old, have x complications, and take y types of medications.

[0075] In the health model, outliers are found by analyzing the data distribution. Outliers usually refer to data points that do not conform to the health model. They may indicate potential disease risks or abnormal conditions.

[0076] In some other embodiments, after data mining is performed on the input and output to obtain potential health patterns and abnormal data, the following steps are also included: S440, if the abnormal data corresponding to the prediction model and the verification model under the same health mode are different.

[0077] S441, determining the same abnormal data as accurate abnormal data, and determining the different abnormal data as questionable abnormal data.

[0078] The above-mentioned data mining process is carried out in the prediction model and the verification model at the same time. At the same time, when the abnormal data predicted and output by the two models under the same health mode are different, the different abnormal data are regarded as doubtful abnormal data, and the doubtful abnormal data are further analyzed and judged, while the same abnormal data are directly confirmed as accurate abnormal data.

[0079] S442, matching the suspected abnormal data to obtain corresponding input data based on the health association rule, and regenerating a prediction sample set and a verification sample set including at least the input data for input.

[0080] S443, mining secondary abnormal data according to the output data of the prediction model and the verification model, and comparing the secondary abnormal data with the suspected abnormal data to determine the credibility of the suspected abnormal data.

[0081] For suspicious abnormal data, the input features corresponding to the data are obtained based on the health association rules, and the data of the input features are used as the prediction sample set. Random influencing event samples are re-added and the blood glucose data prediction of the two models is re-performed. The blood glucose change trend is re-data mined to determine whether the newly mined abnormal data matches the original suspicious abnormal data. If the newly mined abnormal data matches a certain suspicious abnormal data, it is considered that the credibility of the suspicious abnormal data is higher. Otherwise, if it does not match, its credibility is lower.

[0082] The suspicious abnormal data whose credibility is higher than the preset value is regarded as accurate abnormal data and outputted.

[0083] In some other embodiments, generating an explanation event for a blood glucose prediction result based on a health pattern and abnormal data comprises the following steps: S450, evaluating the interpretability between the associated input features and output features.

[0084] Use XAI techniques (such as SHAP, LIME, CAM, etc.) to evaluate whether the model's decision process is explainable.

[0085] S460: Determine the influence coefficient between the associated input features and output features.

[0086] Determine the degree of influence of each feature in the model on blood sugar prediction, use XAI technology to analyze the most important features in the model prediction, and understand which health indicators have the greatest impact on blood sugar prediction. The larger the influence coefficient, the greater the impact of a certain input feature on the output result.

[0087] S470, obtaining explanation events corresponding to blood glucose prediction results based on the local explanation library, where the explanation events include positive input features, negative input features, and corresponding single feature contributions and multiple interactions.

[0088] Explain individual predictions through local explanation libraries, such as using SHAP values ​​or LIME to understand the factors behind the predictions for a specific patient. Analyze which features have the greatest impact on the predictions and how these features affect the model's decisions.

[0089] Provide local explanations for anomalies or key events identified by the model, for example, understand why a patient's blood sugar level suddenly rises, and analyze the reasons behind the event, including the impact of patient behavior, environmental factors, or medical interventions. Analyze the overall behavior of the model to understand how the model processes different categories of patient data and makes decisions. Finally, use natural language processing technology to convert the model's analysis results into easy-to-understand natural language descriptions.

[0090] Among them, the positive input features and the negative input features respectively represent whether the input features have a positive impact or a negative impact on the final prediction result.

[0091] The contribution of a single feature is the contribution of each input feature analyzed by the XAI technology to the prediction result. For example, it may show that "carbohydrate intake at breakfast" is an important factor leading to increased blood sugar.

[0092] Multinomial interactions are characterized as interactions between features revealed by the AI ​​technique. For example, it might indicate that “dietary fat content” has a greater effect on blood sugar when “exercise is low”.

[0093] In some other embodiments, generating an explanation event further includes the following steps: S480, determining whether there is a difference between the explanation event corresponding to the prediction model and the explanation event corresponding to the verification model.

[0094] At the same time, the two models simultaneously obtain the prediction of the explanation event. After both models obtain the explanation event, it is also necessary to determine whether there is a difference between the two explanation events.

[0095] S481, if present, the difference is judged to be associated with the random sample of impact events in the validation model.

[0096] S482, when there is a correlation, the feasible trend of the influencing event sample is determined based on historical data and empirical algorithms. When the feasible trend is greater than a preset value, the explanatory event corresponding to the difference is deemed valid.

[0097] If there is a difference, determine whether the difference is related to the influencing event sample. If there is a correlation, it means that the difference in explanation comes from the different content when the two samples are predicted. For the influencing event sample, determine whether there is a larger feasible trend. The feasible trend is characterized by the fact that the influencing event sample has not occurred at present, but according to the biological basis information of the twin model, it is considered that the event is very likely to occur, and it has a larger feasible trend. Then, when the feasible trend of the influencing event sample is greater than the preset value, the explained event is deemed valid.

[0098] For example, if the interpretation event of the test sample is: "The fat content of the diet is high", the effect on blood sugar is the highest; The explanatory event for the validation sample was: "dietary fat content" had the highest effect on blood glucose under the condition of "less exercise".

[0099] Then the difference is "less exercise", and the impact time sample is after 1 hour of exercise. If the basic physiological information and real-time health information corresponding to the human twin model show that the patient does exercise very little, and the impact time sample has a large feasible trend, then the explanatory event corresponding to the difference is valid and can be output.

[0100] In some other embodiments, the following steps are also included: S800, when the insulin pump performs an injection based on the optimal insulin infusion amount, a monitoring task of a preset duration is generated to detect changes in blood glucose levels.

[0101] S810, performing feedback verification and marking on the blood sugar prediction result based on the blood sugar level change, the marking including a normal mark and an abnormal warning mark.

[0102] After the insulin pump injects the optimal amount of insulin to the patient, it observes the user's blood sugar changes for a period of time. If the blood sugar improvement does not correspond to the predicted blood sugar change trend, the invalid or ineffective injection will be marked and an abnormal alarm mark will be generated. On the contrary, if the blood sugar improvement is the same as the predicted blood sugar change trend, it will be marked as normal.

[0103] Both normal marks and abnormal alarm marks are used to subsequently optimize and update the model's training parameters and training algorithms.

[0104] like Figure 2 As shown, the embodiment of the present application also discloses a diabetes management system based on digital twins, including: Basic parameter upload module, used to upload basic physiological structure parameters; HDT architecture module, used to build a digital twin model based on basic physiological architecture parameters combined with HDT architecture; IoT devices are used to collect real-time health data, including blood sugar, diet, activity, and work and rest; The prediction module is used to process real-time health data as input to the digital twin model and predict the blood sugar change trend to output the blood sugar prediction result; A data diagnosis module, which is used to perform data mining on inputs and outputs to obtain potential health patterns and abnormal data, and to generate explanation events for blood glucose prediction results based on the health patterns and abnormal data; A control module, used to simulate and predict the optimal insulin infusion amount in the digital twin model based on the blood glucose prediction results and explanation events; Insulin pump, used to adjust the injection dosage according to the optimal insulin infusion amount.

[0105] A storage medium is also disclosed in an embodiment of the present application, which stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the above method.

[0106] The implementation principle is: A digital twin model is built through basic physiological structure parameters, and the health pattern and anomalies in the blood sugar prediction results are analyzed through the digital twin model and real-time blood sugar prediction results as well as the mining of the model's input and output data features. Finally, the optimal insulin infusion volume is comprehensively predicted based on the explanation events of the integrated abnormal data, and the insulin pump is intelligently adjusted. In this way, users do not need to manually design the infusion volume, and at the same time, users can get more accurate and scientific medication dosages, thereby improving health management effects and controlling medication costs.

[0107] It should be understood that although the steps in the flowchart of the accompanying drawings are shown in sequence according to the instructions of the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise clearly stated in this document, the execution of these steps is not strictly limited in order and can be performed in other orders.

[0108] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A diabetes management method based on digital twins, characterized in that: The following steps are involved: Obtain basic physiological architecture parameters; Building a digital twin model based on the basic physiological architecture parameters combined with the HDT architecture; Collect real-time health data, including blood sugar, diet, activity, and work and rest; The real-time health data is used as input to the digital twin model for processing, and the blood sugar change trend is predicted to output a blood sugar prediction result; Performing data mining on the input and output to obtain potential health patterns and abnormal data, and generating explanation events for the blood glucose prediction results based on the health patterns and the abnormal data; Simulating and predicting an optimal insulin infusion amount in the digital twin model based on the blood glucose prediction result and the explained event; The optimal insulin infusion amount is sent to the insulin pump to adjust the injection dosage.

2. The diabetes management method based on digital twins according to claim 1, characterized in that: Building a digital twin model based on the basic physiological architecture parameters combined with the HDT architecture includes the following steps: Generating a twin model based on the basic physiological architecture parameters, wherein the twin model includes a prediction model and a verification model; Among them, both the prediction model and the verification model build a virtual human body framework based on basic physiological information, add a virtual human body information flow in the virtual human body framework based on historical medication information and complication information, and generate a virtual metabolic flow and a virtual pathological flow based on the virtual human body information flow.

3. The diabetes management method based on digital twins according to claim 2, characterized in that: The real-time health data is used as input to the digital twin model for processing, and the blood sugar change trend is predicted to output a blood sugar prediction result, including the following steps: Based on the real-time health data, the current blood sugar status and time-space nodes are obtained, and the time-space nodes include before eating, after eating, before activity, after activity, before work and rest, and after work and rest; Integrate the current blood sugar state and the spatiotemporal nodes into a test sample set and upload it to the prediction model, and obtain a first blood sugar change trend based on an LSTM network; Adding random impact event samples to the test sample set to obtain a verification sample set, uploading the verification sample set to the verification model, and obtaining a second blood sugar change trend based on the LSTM network; Calculate the predicted difference sensitivity based on the first blood sugar change trend and the second blood sugar change trend; When the difference sensitivity is lower than a preset value, the first blood sugar change trend is used as the blood sugar prediction result and output; when the difference sensitivity is not lower than the preset value, the first blood sugar change trend is fitted with the second blood sugar change trend and the fitting result is used as the blood sugar prediction result and output.

4. The diabetes management method based on digital twins according to claim 3, characterized in that: Data mining of inputs and outputs to obtain potential health patterns and abnormal data includes the following steps: Collect input data and output data and extract input features and output features based on feature engineering; Acquire a global behavior path between the input feature and the output feature through a decision tree path, and acquire a decision logic represented by the global behavior path based on a local interpretation library; Based on the decision logic, a health association rule between the input data and the output data is identified to generate a health mode, and abnormal values ​​reflected by the data under the health mode are obtained, and data with abnormal values ​​higher than a preset value are defined as abnormal data.

5. The diabetes management method based on digital twins according to claim 4, characterized in that: After data mining the inputs and outputs to find potential healthy patterns and abnormal data, the following steps are also included: If the abnormal data corresponding to the prediction model and the verification model under the same health mode are different; Determine the same abnormal data as accurate abnormal data, and determine the different abnormal data as questionable abnormal data; Matching the suspected abnormal data to obtain corresponding input data based on the health association rule, and regenerating a prediction sample set and a verification sample set including at least the input data for input; Secondary abnormal data are mined according to the output data of the prediction model and the verification model, and the secondary abnormal data are respectively compared with the suspected abnormal data to determine the credibility of the suspected abnormal data.

6. The diabetes management method based on digital twins according to claim 2, characterized in that: Generating an explanation event for the blood sugar prediction result based on the health pattern and the abnormal data comprises the following steps: evaluating the interpretability of the associated input features and output features; Determining an influence coefficient between the associated input feature and the output feature; The explanation event corresponding to the blood glucose prediction result is obtained based on the local explanation library, and the explanation event includes positive input features, negative input features, and corresponding single feature contributions and multiple interactions.

7. The diabetes management method based on digital twins according to claim 6, characterized in that: Generating the explanation event further includes the following steps: Determining whether there is a difference between the explanation event corresponding to the prediction model and the explanation event corresponding to the verification model; If so, determining that the difference is associated with the random sample of the influencing events in the verification model; When there is a correlation, the feasible trend of the influencing event sample is determined based on historical data and empirical algorithms, and when the feasible trend is greater than a preset value, the explanatory event corresponding to the difference is deemed valid.

8. The diabetes management method based on digital twins according to claim 1, characterized in that: The following steps are also included: When the insulin pump performs an injection based on the optimal insulin infusion amount, a monitoring task of a preset duration is generated to detect changes in blood glucose levels; The blood sugar prediction result is fed back for verification and marked based on the blood sugar level change, and the mark includes a normal mark and an abnormal warning mark.

9. A diabetes management system based on digital twins, characterized in that: include: Basic parameter upload module, used to upload basic physiological structure parameters; An HDT architecture module, used to build a digital twin model based on the basic physiological architecture parameters combined with the HDT architecture; IoT devices are used to collect real-time health data, including blood sugar, diet, activity, and work and rest; A prediction module, used to process the real-time health data as input to the digital twin model, and predict the blood sugar change trend to output a blood sugar prediction result; A data diagnosis module, configured to perform data mining on the input and output to obtain potential health patterns and abnormal data, and generate explanation events for the blood glucose prediction results based on the health patterns and the abnormal data; A control module, configured to simulate and predict an optimal insulin infusion amount in the digital twin model based on the blood glucose prediction result and the explanation event; An insulin pump is used to adjust the injection dosage according to the optimal insulin infusion amount.

10. A storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method as described in any one of claims 1-8.

Images