T2DM-NAFLD personalized lifestyle management system based on digital twinning

Abstract

The invention discloses a T2DM-NAFLD personalized lifestyle management system based on digital twinning, and particularly relates to the technical field of chronic disease management.The T2DM-NAFLD personalized lifestyle management system comprises the steps that behavior and psychological data of a patient are collected and subjected to time sequence processing, and a digital twinning model fusing a physiological mechanism and a data-driven network is constructed; dynamically calculating a behavior compliance index based on a metabolic stability parameter and a data change trend output by the digital twinborn model; selecting a matched incentive strategy from a predefined strategy library according to the index, and adjusting exercise intensity or nutrition distribution parameters in the digital twin model according to the matched incentive strategy; generating personalized lifestyle suggestions by using the adjusted digital twinborn model and pushing the personalized lifestyle suggestions through a visual interface; closed-loop linkage of behavior intervention and metabolism simulation is achieved, and the problems that in existing health management, intervention measures are disjointed with the real-time metabolism state and psychological motivation of the patient, and consequently compliance is low, and the long-term effect is poor are solved.

Description

Technical Field

[0001] This invention relates to the field of chronic disease management technology, and more specifically, to a T2DM-NAFLD personalized lifestyle management system based on digital twins. Background Technology

[0002] T2DM-NAFLD, medically speaking, specifically refers to the clinical condition of type 2 diabetes mellitus combined with non-alcoholic fatty liver disease. In chronic disease management scenarios based on digital twin technology, the system provides personalized lifestyle guidance to patients with type 2 diabetes mellitus and NAFLD by constructing a virtual simulation model of the patient's metabolic state. Patients continuously record their dietary intake, exercise levels, and physiological indicators using smart devices, and the digital twin model generates personalized recommendations based on this data. However, in practical applications, the system has been found to lack a dynamic perception and response mechanism for the patient's behavioral and psychological state. When patients face decreased motivation, emotional fluctuations, or environmental stress, the system cannot promptly identify and adjust intervention strategies, leading to a gradual disconnect between the recommendations generated by the digital twin model and the patient's actual ability to implement them. Over time, as patients no longer experience timely feedback and adaptive support from the system, their participation continues to decline, ultimately affecting the long-term effectiveness of disease management.

[0003] Existing digital twin systems have significant shortcomings in chronic disease management: while they can collect data and simulate metabolism, they lack the ability to dynamically assess and regulate patient behavioral adherence. Specifically, current technologies fail to integrate the metabolic stability parameters output by the digital twin model with the multi-dimensional trends of patient behavioral and psychological data, and cannot quantify long-term patient engagement through dynamic weight allocation and attention mechanisms. Furthermore, the systems fail to establish an adaptive correlation between incentive strategies and model parameter adjustments, leading to a mismatch between intervention strategies and the patient's real-time state. These technological limitations mean that although the system possesses accurate physiological simulation capabilities, it struggles to achieve sustained and effective behavioral interventions, ultimately impacting the overall effectiveness of chronic disease management. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in the prior art, this invention provides a T2DM-NAFLD personalized lifestyle management system based on digital twins. By constructing a digital twin model that integrates physiological mechanisms and data-driven approaches, and dynamically calculating the compliance index by combining metabolic stability parameters and behavioral trends, a closed-loop adjustment mechanism for incentive strategies and model parameters is established to achieve accurate generation and delivery of personalized lifestyle recommendations, thereby solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a T2DM-NAFLD personalized lifestyle management system based on digital twins, comprising: The data acquisition and preprocessing module is used to collect patients' exercise data, diet data, emotional state data and heart rate variability data through mobile applications and smart wearable devices authorized by the user, and to clean, standardize and time-series align the multi-source data, dividing it into continuous time windows according to natural days to form a standardized dataset. The behavioral compliance index calculation module is used to calculate the behavioral compliance index based on the standardized dataset. This includes segmenting the data by time window, obtaining the standard deviation of the predicted blood glucose value sequence as a metabolic stability parameter through a digital twin model, dynamically allocating the evaluation weight of each data item based on the metabolic stability parameter, calculating the attention coefficient of each data item based on the changing trend through a feedforward neural network, and using the dynamically allocated evaluation weight and attention coefficient for weighted fusion and generating the behavioral compliance index through Sigmoid function mapping.

[0006] The incentive strategy selection module is used to select a personalized incentive strategy from a predefined strategy library based on the behavioral compliance index. This includes comparing the behavioral compliance index with an incentive threshold, selecting either a gamified task strategy or a progressive goal feedback strategy based on the comparison result, and outputting the selected strategy as a control signal. The predefined strategy library includes: a gamified task strategy, which is implemented by breaking down health goals into atomic health tasks and associating them with virtual rewards; and a progressive goal feedback strategy, which is implemented by generating a sequence of phased goals based on user historical data using a goal planning algorithm and displaying progress through a data visualization component. The selection logic is as follows: the behavioral compliance index is compared with an incentive threshold; if it is lower than the threshold, the gamified task strategy is selected; if it is higher than or equal to the threshold, the progressive goal feedback strategy is selected. The activated personalized incentive strategy will be output as a control signal to the model parameter adjustment module. The model parameter adjustment module receives incentive policy control signals and adjusts the dynamic parameters in the digital twin model. This includes lowering the exercise intensity parameter when the policy is a gamified task policy and optimizing the nutrient allocation parameter when the policy is a progressive goal feedback policy. The updated parameters are then loaded into the digital twin model. The adjustment rules are as follows: when the policy is a gamified task policy, the baseline value of the exercise intensity parameter is lowered, which is technically implemented by modifying the corresponding numerical field in the model's exercise-metabolic response submodule; when the policy is a progressive goal feedback policy, the nutrient allocation parameter is optimized, which is technically implemented by introducing an inhibitory factor negatively correlated with the policy execution intensity into the fat synthesis rate calculation formula of the model's liver metabolism submodule. It is recommended to generate and output a module for running metabolic dynamics simulations using the adjusted digital twin model, generating exercise plans including exercise type, duration, and intensity, and dietary arrangements including food type, portion size, and eating time based on the simulation results, and pushing them to users in a visual form through a mobile application interface.

[0007] Preferably, the behavioral compliance index calculation module includes: Align and segment behavioral data and psychological indicator data according to time windows; The patient's continuous blood glucose monitoring data, nutritional intake data, and energy consumption data are input into the digital twin model to obtain its output predicted blood glucose value sequence, and the standard deviation of the sequence is used as a parameter to characterize metabolic stability. Based on the parameters characterizing metabolic stability obtained in the previous simulation cycle, evaluation weights are dynamically assigned to each data item. The change trend of each data item in the current time window relative to the previous time window is input into a feedforward neural network. After nonlinear transformation of the hidden layer and Softmax normalization of the output layer, the attention coefficient corresponding to each data item is calculated. Using the dynamically assigned evaluation weights and the attention coefficients, the segmented data items are weighted and fused, and the behavioral compliance index is generated by mapping through the Sigmoid function.

[0008] Preferably, the Sigmoid function takes metabolic stability parameters as the main input. When the behavioral compliance index is higher than a first threshold, a preset multiple of the basic weight is assigned to the exercise-related data items as the evaluation weight; when it is lower than the first threshold but the psychological stress score is higher than a second threshold, the preset multiple of the basic weight is assigned to the psychological indicator data items as the evaluation weight.

[0009] Preferably, the construction of the digital twin model includes: Construct a hybrid architecture consisting of a physiological mechanism model and a recurrent neural network connected in series; The patient's continuous blood glucose data, nutritional intake data, and energy consumption data are input into the physiological mechanism model, which outputs a preliminary prediction of blood glucose trajectory by solving ordinary differential equations. The intermediate state variables calculated by the physiological mechanism model are input into the recurrent neural network, and the recurrent neural network outputs compensation values ​​for individualized compensation and correction of the preliminary predicted blood glucose trajectory. The compensation value is used to adjust the preliminary predicted blood glucose trajectory to generate the final predicted blood glucose sequence; The standard deviation of the final predicted blood glucose sequence is calculated and output as a parameter characterizing metabolic stability.

[0010] Preferably, the dynamic weight allocation in the behavior compliance index calculation module includes: Set a first threshold and a second threshold; when the predicted blood glucose fluctuation coefficient is higher than the first threshold, assign a preset multiple of its basic weight to the exercise-related data item as an evaluation weight; when the predicted blood glucose fluctuation coefficient is lower than the first threshold and the psychological stress score is higher than the second threshold, assign the preset multiple of its basic weight to the psychological indicator data item as an evaluation weight.

[0011] Preferably, the incentive strategy selection module includes: The behavioral compliance index is compared with at least one preset incentive threshold. When the behavioral compliance index is lower than the incentive threshold, a gamified task strategy is selected and executed from a predefined incentive strategy library. The gamified task strategy includes breaking down health goals into atomic health tasks and associating virtual rewards with completed tasks. When the behavioral compliance index is higher than or equal to the incentive threshold, a progressive goal feedback strategy is selected and executed from a predefined incentive strategy library. The progressive goal feedback strategy includes generating a sequence of phased goals based on user historical data and visually displaying the progress.

[0012] Preferably, in the model parameter adjustment module, adjusting the dynamic parameters related to lifestyle intervention in the digital twin model based on the selected personalized incentive strategy includes: When the personalized incentive strategy is a gamified task strategy, the motion intensity parameter in the digital twin model is reduced. When the personalized incentive strategy is an incremental goal feedback strategy, the nutrient allocation parameters in the digital twin model are optimized, wherein the optimization includes introducing a parameter factor associated with the strength of strategy execution to suppress the simulation of liver fat synthesis.

[0013] Preferably, in the model parameter adjustment module, lowering the motion intensity parameter when the strategy is a gamified task strategy includes: Obtain the baseline values ​​of the current motion intensity parameters stored in the digital twin model; Multiply the reference value by a preset scaling factor to obtain the adjusted motion intensity parameters; Update the adjusted exercise intensity parameters to the exercise metabolic response submodule of the model.

[0014] Preferably, in the model parameter adjustment module, optimizing nutrient allocation parameters includes: The inhibition factor is calculated based on the policy execution intensity value using a linear mapping function. The inhibitory factor was introduced into the liver fat synthesis simulation calculation module; When calculating the rate of liver fat synthesis, the original synthesis rate is multiplied by 1 and the complement of the inhibitory factor to achieve quantitative control of liver fat synthesis simulation.

[0015] Preferably, the suggestion generation and output module includes: Metabolic kinetic simulations were run using the adjusted digital twin model to obtain predicted blood glucose curves and predicted trends in liver fat load. Based on the predicted blood glucose curve and the predicted liver fat load trend, an exercise plan including exercise type, duration, and intensity level is generated, as well as a dietary plan including food type, intake amount, and eating time window. The exercise plan and diet plan are pushed to the user's device in a visual format through the mobile application interface.

[0016] The technical effects and advantages of this invention are as follows: This invention achieves precise quantitative assessment of patients' physiological state and behavioral motivation by constructing a digital twin model that integrates metabolic mechanisms and individual differences, and dynamically calculating the behavioral compliance index using multi-source time-series data. Based on feedforward neural network analysis of data change trends to generate attention coefficients, a compliance index that truly reflects patients' long-term engagement and motivation intensity is obtained through weighted fusion and nonlinear mapping. This effectively solves the key problem of existing health management systems that rely solely on single or static data and are unable to accurately assess and predict patients' behavioral compliance.

[0017] This invention further achieves the dynamic generation and precise execution of personalized intervention strategies by establishing a closed-loop linkage mechanism between incentive strategies and digital twin model parameters; it automatically matches gamified tasks or progressive goal feedback strategies based on compliance index, and transforms the strategy signals into specific adjustment instructions for exercise intensity parameters or nutrient allocation parameters in the model; it uses the adjusted model to perform metabolic simulation, generates exercise plans and dietary arrangements that are both physiologically reasonable and behaviorally feasible, and pushes them through a visual interface, effectively solving the problem of the disconnect between lifestyle recommendations and patients' real-time metabolic status and psychological motivation in traditional methods. Attached Figure Description

[0018] Figure 1 This is a structural block diagram of the T2DM-NAFLD personalized lifestyle management system of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0020] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0021] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0022] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0023] Example 1, see Figure 1 The T2DM-NAFLD personalized lifestyle management system structure diagram provided by this invention is as follows. Figure 1 The T2DM-NAFLD personalized lifestyle management system based on digital twins, as shown, includes: Data acquisition and preprocessing module: Collects patients' behavioral and psychological data through smart devices, and preprocesses and aligns the multi-source heterogeneous data according to time series. Specifically, the behavioral data includes exercise frequency, exercise heart rate, and dietary logs from smart wearable devices; the psychological indicator data includes emotional state scale data reported by the user through a mobile application, and heart rate variability data collected by device sensors; the collection process is completed on the user's personal mobile device, through an application that the user actively authorizes and opens, with the purpose, method, and retention period of data processing clearly stated in an independent interactive interface, and implemented after obtaining the user's individual consent; at the same time, the system provides alternative solutions that do not rely on biometric information; After collection, the multi-source data is cleaned and standardized to eliminate differences in units and outliers between devices. The data is then aligned according to a unified timestamp and divided into continuous time windows based on natural days to form a standardized dataset for subsequent analysis. Behavioral compliance index calculation module: Based on the behavioral data and psychological indicator data, the behavioral compliance index is calculated. The behavioral compliance index reflects the patient's long-term participation and motivation intensity through multi-dimensional data fusion. Specifically, this includes: aligning and segmenting the behavioral data and psychological indicator data according to time windows; dynamically allocating evaluation weights for each data item based on the metabolic stability parameters output by the digital twin model in the previous simulation cycle; calculating the corresponding attention coefficient for each data item based on its changing trend in the current time window relative to the previous time window; and using the dynamically allocated evaluation weights and the attention coefficients to perform weighted fusion on the segmented data to generate the behavioral compliance index.

[0024] The behavioral compliance index is obtained through a dynamic evaluation method that integrates feedback from the digital twin model with data time-series trends. The specific calculation process is as follows: The preprocessed behavioral and psychological data from the data acquisition and preprocessing module are divided into continuous evaluation time windows by calendar day and aligned to a unified time series. The behavioral data includes daily exercise duration and exercise heart rate, while the psychological data includes daily self-reported mood state scores and stress level scores calculated from heart rate variability; parameters characterizing metabolic stability are acquired. The patient's latest continuous blood glucose monitoring data, carbohydrate and fat intake data derived from food image recognition, and basal energy expenditure data estimated by a smart wearable device are input into a pre-trained digital twin model. This digital twin model is a hybrid architecture, containing a physiologically based glucose-insulin metabolism mechanism model and a long short-term memory recurrent neural network for learning individual physiological differences. The input to this recurrent neural network is the intermediate state variables of the mechanism model and external input data; its output is used to compensate and correct the final prediction results of the mechanism model. After the digital twin model runs the simulation, it outputs a sequence of predicted blood glucose values ​​for the next few hours. The parameter characterizing metabolic stability is the standard deviation of the predicted blood glucose value sequence. This standard deviation is obtained by calculating the dispersion between the predicted blood glucose value and its sequence mean, and is defined as the predicted blood glucose fluctuation coefficient. Dynamic weight allocation is performed based on the predicted blood glucose fluctuation coefficient output by the digital twin model in the previous simulation cycle. The specific allocation logic is as follows: when the predicted blood glucose fluctuation coefficient is high, the patient is judged to have poor metabolic stability. At this time, the exercise duration and exercise heart rate data items are assigned higher evaluation weights. When the predicted blood glucose fluctuation coefficient is low but the synchronously input data indicates a high stress level, the emotional state score and stress level score data items are assigned higher evaluation weights. Attention coefficients are calculated for each data item to quantify the significance of its recent changes. This coefficient is calculated using a lightweight feedforward neural network. The network's input is the difference between the data value in the current time window and the same data value in the previous time window, representing the trend of change. The network has a hidden layer that performs a non-linear transformation on the input trend information, and finally, the output layer is normalized using a Softmax function, outputting an attention coefficient between zero and one for each data item; a data item showing a rapid recent decline will therefore receive a higher attention coefficient. Weighted fusion and index generation are performed: The normalized value of each data item within the current time window is multiplied by its corresponding dynamically assigned evaluation weight and attention coefficient to obtain the final contribution value of that data item. The final contribution values ​​of all data items are summed, and the summation result is then mapped to the interval between zero and one using a Sigmoid function. The resulting value is the behavioral compliance index for this time window.

[0025] Furthermore, the specific calculation rules for the dynamic weight allocation are implemented through a predefined mapping function. This function uses the predicted blood glucose fluctuation coefficient as the main input. When the coefficient is higher than a first threshold set by the system, the mapping function assigns a preset multiple of the base weight to the exercise duration and exercise heart rate data items as their evaluation weight. When the coefficient is lower than the first threshold but the psychological stress comprehensive score calculated by the system simultaneously is higher than a second threshold, the preset multiple of the base weight is assigned to the emotional state score and stress level score data items as their evaluation weight.

[0026] Furthermore, the construction of the digital twin model includes: Construct a hybrid architecture consisting of a physiological mechanism model and a recurrent neural network connected in series; The patient's continuous blood glucose data, nutritional intake data, and energy consumption data are input into the physiological mechanism model, which outputs a preliminary prediction of blood glucose trajectory by solving ordinary differential equations. The intermediate state variables calculated by the physiological mechanism model are input into the recurrent neural network, and the recurrent neural network outputs compensation values ​​for individualized compensation and correction of the preliminary predicted blood glucose trajectory. The compensation value is used to adjust the preliminary predicted blood glucose trajectory to generate the final predicted blood glucose sequence; The standard deviation of the final predicted blood glucose sequence is calculated and output as a parameter characterizing metabolic stability.

[0027] The digital twin model is a hybrid architecture used to simulate the dynamics of glucose-lipid metabolism in patients. It consists of a physiologically based mechanism model and a data-driven recurrent neural network, specifically including the following: Mechanism Model Construction and Function: This mechanism model is constructed using a set of ordinary differential equations (ODEs) that describe the fundamental kinetics of blood glucose concentration changes, insulin secretion and sensitivity, and hepatic fat synthesis and breakdown. The model's inputs include: current blood glucose concentration values ​​from a continuous glucose monitor, carbohydrate and fat intake rates derived from food image recognition and converted from a nutritional database, and energy expenditure rates estimated from a wearable device based on heart rate and activity levels. By solving the set of differential equations, the mechanism model outputs a preliminary predicted blood glucose trajectory for the next few hours. Integration and Function of Recurrent Neural Networks: A recurrent neural network containing long short-term memory units is placed after the mechanism model to perform individualized compensation correction on the model's predictions. The input layer of this network receives intermediate state variables from the mechanism model, including but not limited to simulated insulin sensitivity and hepatic glucose output rate. The recurrent neural network learns the complex mapping relationship between these intermediate variables and real individual physiological differences, and outputs a compensation value. This compensation value is used to adjust the initial predicted blood glucose trajectory output by the mechanism model. Model Training: The training of the digital twin model is divided into two stages. First, population baseline values ​​are set for the parameters of the ordinary differential equations in the mechanism model based on publicly available physiological knowledge. Second, supervised learning training is performed on the integrated recurrent neural network using de-identified patient cohort data. During training, historical time-series data is input into the mechanism model to obtain its preliminary prediction trajectory and intermediate state variables, which are then used as input to the recurrent neural network. The training objective is to minimize the mean squared error between the final predicted blood glucose sequence after compensation and correction by the recurrent neural network and the actual measured blood glucose sequence. The weights and bias parameters in the recurrent neural network are optimized using the backpropagation algorithm, and the training process stops when the loss function on the validation set converges. Model Application and Parameter Generation: In the application phase, the latest multi-source data of the target individual is input into the trained digital twin model. The model first runs a mechanism model to obtain preliminary predictions, and then a recurrent neural network compensates for these predictions based on the individual's characteristic data, ultimately outputting a personalized sequence of predicted blood glucose values ​​for a specific future time period. The parameter characterizing metabolic stability, namely the predicted blood glucose fluctuation coefficient, is obtained by calculating the standard deviation of this final predicted sequence. This standard deviation quantifies the degree of dispersion of the predicted blood glucose values ​​around its sequence mean, thus serving as a basis for assessing metabolic stability.

[0028] Furthermore, in the supervised learning training, the loss function used is mean squared error loss, and the optimizer is the Adam optimizer. During the training iteration, after each complete training cycle, the loss is calculated on an independent validation set. When the validation set loss no longer decreases for several consecutive training cycles, an early stopping mechanism is triggered, training is terminated, and the weights and bias parameters of the recurrent neural network at this time are saved.

[0029] Incentive strategy selection module: Based on the behavioral compliance index, selects a matching personalized incentive strategy from a predefined incentive strategy library, which contains various intervention methods based on behavioral economics principles; the activated personalized incentive strategy will be output as a control signal to the model parameter adjustment module; The explanation is as follows: the predefined incentive strategy library is stored in the system database, and its construction process includes: Based on knowledge in the field of chronic disease management, different behavioral intervention modes are abstracted into technical templates that can be systematically scheduled. The technical templates include at least gamified task strategies and progressive goal feedback strategies. The technical implementation of gamified task strategies includes breaking down macro-level health goals (such as losing 2 kg per month) into a series of atomic health tasks with a logical order (such as walking 8,000 steps daily or completing a blood glucose log). The system associates a virtual reward value with each atomic task and triggers dynamic visual feedback (such as badge lighting up or progress bars advancing) through the user interface when the user completes the task. The technical implementation of the progressive goal feedback strategy includes: dynamically generating a phased goal sequence based on the user's historical behavior data using a goal planning algorithm. The goal values ​​in this sequence gradually approach the ideal health value over time. The system uses a data visualization component to plot the goal achievement progress curve in the user interface, and generates and highlights corresponding electronic achievement badges when the user achieves a phased goal. The strategy selection logic is implemented through the following operations: Threshold comparison: The behavioral compliance index calculated by the behavioral compliance index calculation module is compared with at least one preset incentive threshold in real time. Strategy Mapping: Based on the comparison results, a preset mapping rule is executed; when the behavioral compliance index is lower than the incentive threshold, a gamified task strategy is automatically invoked and activated, aiming to quickly increase user engagement through high-frequency, small-step rewards. When the behavioral compliance index is higher than or equal to the incentive threshold, a progressive goal feedback strategy is automatically invoked and activated, aiming to maintain user motivation by showcasing long-term trends and macro achievements. Strategy Execution: After the selected strategy is instantiated, its output parameters (such as atomic task list, virtual reward value, target sequence, etc.) will be used as input and passed to the subsequent digital twin model adjustment module (model parameter adjustment module) and suggestion generation and output module, thereby completing the dynamic closed-loop adjustment of the entire personalized management strategy.

[0030] Model parameter adjustment module: Based on the selected personalized incentive strategy, adjust the dynamic parameters related to lifestyle intervention in the digital twin model to optimize the simulation accuracy of the digital twin model for the real-time state of the patient; Furthermore, in the model parameter adjustment module, based on the selected personalized incentive strategy, the dynamic parameters related to lifestyle intervention in the digital twin model are adjusted, including: When the personalized incentive strategy is a gamified task strategy, the motion intensity parameter in the digital twin model is reduced. When the personalized incentive strategy is an incremental goal feedback strategy, the nutrient allocation parameters in the digital twin model are optimized, wherein the optimization includes introducing a parameter factor associated with the strength of strategy execution to suppress the simulation of liver fat synthesis.

[0031] The specific execution process of the model parameter adjustment module is explained below: The dynamic parameters related to lifestyle interventions in the digital twin model are key variables used within the model to simulate the metabolic response of patients to exercise and nutrient intake. Specifically, these include: exercise intensity parameters, which are used to simulate the instantaneous changes in energy consumption rate and insulin sensitivity in the model during exercise of different intensities; and nutrient allocation parameters, which are used to simulate the allocation ratio of glucose absorption rate and liver fat synthesis pathway in the model after ingestion of different macronutrients.

[0032] Establish a mapping and adjustment rule between incentive strategies and dynamic parameters. This rule will make targeted adjustments to the aforementioned dynamic parameters based on the personalized incentive strategy type selected by the incentive strategy selection module. When the selected strategy is a gamified task strategy: This indicates that the patient's current compliance is low and the task threshold needs to be lowered. In this case, the adjustment process is to lower the baseline value of the exercise intensity parameter in the digital twin model. For example, reduce the simulated energy consumption rate corresponding to moderate-intensity exercise in the model by a preset percentage, so that the model can generate lower-intensity and easier-to-complete exercise suggestions for the patient in subsequent simulations. When the selected strategy is a progressive goal feedback strategy: this indicates that the patient's current compliance is high, but long-term metabolic health needs to be monitored. In this case, the adjustment process involves optimizing the nutrient allocation parameters in the digital twin model. Specifically, when simulating the lipid metabolism pathway, an inhibitory factor associated with the strategy's execution intensity is introduced. This factor acts on the simulation calculation module for liver fat synthesis, causing the model to predict higher health risks for the metabolic consequences of high-fat foods in subsequent simulations, thus generating nutritional recommendations that favor a low-fat diet. Update the adjusted dynamic parameters to the digital twin model.

[0033] The explanation is as follows: Lowering the baseline value of the exercise intensity parameter in the digital twin model is achieved by: In the exercise-metabolic response simulation submodule of the digital twin model, searching for the data structure storing the current exercise intensity parameter, and multiplying the numerical fields associated with moderate-intensity exercise in this data structure by a scaling factor less than one. Introducing an inhibitory factor is achieved by: In the liver metabolism simulation submodule of the digital twin model, adding a product term negatively correlated with the intensity of the strategy execution to the formula for calculating the liver fat synthesis rate; this product term is the inhibitory factor.

[0034] It is recommended to generate and output a module for running metabolic dynamics simulations using the adjusted digital twin model, generating exercise plans including exercise type, duration, and intensity, and dietary arrangements including food type, portion size, and eating time based on the simulation results, and pushing them to users in a visual form through a mobile application interface.

[0035] Furthermore, the suggestion generation and output module includes: Metabolic kinetic simulations were run using the adjusted digital twin model to obtain predicted blood glucose curves and predicted trends in liver fat load. Based on the predicted blood glucose curve and the predicted liver fat load trend, an exercise plan including exercise type, duration, and intensity level is generated, as well as a dietary plan including food type, intake amount, and eating time window; the exercise plan and dietary plan are pushed to the user's device in a visual form through a mobile application interface.

[0036] The following explanation and suggested steps are provided for the generation and output modules: Multi-source data, including current physiological indicators, past behavioral data, and dynamic parameters adjusted by the model parameter adjustment module, are input into the adjusted digital twin model. This drives the model to run its internal metabolic kinetics simulation program, simulating the trajectory of metabolite changes in patients under different hypothetical life scenarios within a specific future time range. The core outputs include predicted blood glucose curves, predicted energy balance status, and predicted liver fat load trends. The output data from the above simulation is analyzed and used to generate structured lifestyle recommendations: By comparing the metabolic benefits under different exercise intensities and durations in simulated scenarios, the exercise program that minimizes the predicted blood glucose fluctuation coefficient and maximizes energy consumption is selected. The final exercise plan clearly includes the exercise type, recommended exercise duration, and intensity level.

[0037] By comparing the simulated metabolic responses after consuming different food combinations, the nutritional ratio that can simultaneously maintain a stable predicted blood glucose curve and minimize the predicted increase in liver fat load is selected. The generated dietary plan clearly includes recommended food types, suggested intake amounts, and optimal eating time windows. Visualized push notifications are delivered via the user interface. Personalized lifestyle suggestions are encapsulated as data objects and transmitted to the user's mobile application. The mobile application interface utilizes its built-in visualization rendering engine to present the received exercise plan as a daily task list and animated demonstration, and the dietary plan as an interactive plate graphic and timeline chart. Furthermore, the core summary information of the suggestions is sent to the device notification center via the operating system's push service, ensuring timely delivery and operability of the suggestions.

[0038] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital twin based T2DM-NAFLD personalized lifestyle management system, characterized in that, The application comprises: a data acquisition and preprocessing module: behavior data and psychological indicator data of patients are collected by intelligent devices, and are preprocessed and time-aligned; a behavior adherence index calculation module: based on the behavior data and psychological indicator data, the behavior adherence index is used to reflect the long-term participation and motivation intensity of patients; including: recording each time period of behavior data and psychological indicator data as each data item, dynamically assigning evaluation weights of each data item based on metabolic stability parameters output by the digital twin model in the last simulation cycle; according to the change trend of each data item in the current time window relative to the previous time window, the attention coefficient corresponding to each data item is calculated; the segmented data is weighted and fused by using the dynamically assigned evaluation weights and the attention coefficient, and the behavior adherence index is generated; an incentive strategy selection module, used for selecting a personalized incentive strategy from a pre-defined strategy library according to the behavior adherence index, including comparing the behavior adherence index with an incentive threshold, selecting a gamification task strategy or a progressive goal feedback strategy based on the comparison result, and outputting the selected strategy as a control signal; a model parameter adjustment module, used for receiving the incentive strategy control signal and adjusting the dynamic parameters in the digital twin model, including lowering the exercise intensity parameter when the strategy is the gamification task strategy, and optimizing the nutrient distribution parameter when the strategy is the progressive goal feedback strategy, and loading the updated parameters into the digital twin model; a suggestion generation and output module, used for running metabolic kinetics simulation using the adjusted digital twin model, generating a movement plan including movement type, duration and intensity, and a diet arrangement including food type, portion and eating time based on the simulation result, and pushing the plan and the arrangement to the user in a visual form through a mobile application interface.

2. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 1, characterized in that, The behavior adherence index calculation module comprises: time alignment and segmentation of behavior data and psychological indicator data according to time windows; inputting continuous blood glucose monitoring data, nutrition intake data and energy consumption data of the patient into the digital twin model to obtain a predicted blood glucose value sequence output by the model, and taking the standard deviation of the sequence as a parameter representing metabolic stability; dynamically assigning evaluation weights to each data item based on the parameter representing metabolic stability obtained in the last simulation cycle; inputting the change trend of each data item in the current time window relative to the previous time window into a feedforward neural network, and calculating the attention coefficient corresponding to each data item through nonlinear transformation of the hidden layer of the network and Softmax normalization processing of the output layer; weighted fusion of each segmented data item by using the dynamically assigned evaluation weights and the attention coefficient, and generation of the behavior adherence index through Sigmoid function mapping.

3. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 2, characterized in that, The Sigmoid function takes the metabolic stability parameter as the main input, assigns a preset multiple of the basic weight as the evaluation weight for the exercise-related data item when the behavior adherence index is higher than the first threshold, and assigns the preset multiple of the basic weight as the evaluation weight for the psychological indicator data item when the behavior adherence index is lower than the first threshold and the psychological stress score is higher than the second threshold.

4. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 3, characterized in that, The construction of the digital twin model comprises: A hybrid architecture composed of a physiological mechanism model and a recurrent neural network is constructed; Continuous blood glucose data, nutrition intake data, and energy consumption data of the patient are input into the physiological mechanism model, which outputs a preliminary predicted blood glucose trajectory by solving ordinary differential equations; Intermediate state variables calculated by the physiological mechanism model are input into the recurrent neural network, which outputs compensation values for individualized compensation correction of the preliminary predicted blood glucose trajectory; The preliminary predicted blood glucose trajectory is adjusted using the compensation values to generate a final predicted blood glucose sequence; The standard deviation of the final predicted blood glucose sequence is calculated and output as a parameter representing metabolic stability.

5. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 4, characterized in that, The dynamic weight allocation in the behavior compliance index calculation module includes: A first threshold and a second threshold are set; when the predicted blood glucose fluctuation coefficient is higher than the first threshold, the motion-related data item is assigned a preset multiple of its base weight as the evaluation weight; when the predicted blood glucose fluctuation coefficient is lower than the first threshold and the psychological stress score is higher than the second threshold, the psychological indicator data item is assigned the preset multiple of its base weight as the evaluation weight.

6. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to any one of claims 1-5, characterized in that, The incentive strategy selection module includes: The behavior compliance index is compared with at least one preset incentive threshold; when the behavior compliance index is lower than the incentive threshold, a gamified task strategy is selected and executed from a predefined incentive strategy library, wherein the gamified task strategy includes decomposing health goals into atomic health tasks and associating virtual rewards with completed tasks; when the behavior compliance index is higher than or equal to the incentive threshold, a progressive goal feedback strategy is selected and executed from a predefined incentive strategy library, wherein the progressive goal feedback strategy includes generating a phased goal sequence based on user historical data and visually displaying progress.

7. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 6, characterized in that, In the model parameter adjustment module, based on the selected personalized incentive strategy, adjusting the dynamic parameters related to lifestyle intervention in the digital twin model includes: When the personalized incentive strategy is a gamified task strategy, the exercise intensity parameter in the digital twin model is lowered; When the personalized incentive strategy is a progressive goal feedback strategy, the nutrition distribution parameter in the digital twin model is optimized.

8. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 7, characterized in that, In the model parameter adjustment module, when the strategy is a gamified task strategy, the exercise intensity parameter is lowered, including: Obtain the current exercise intensity parameter baseline value stored in the digital twin model; Multiply the baseline value by a preset scaling factor to obtain the adjusted exercise intensity parameter; Update the adjusted exercise intensity parameter to the exercise metabolism response submodule of the model.

9. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 8, characterized in that, In the model parameter adjustment module, optimizing the nutrition distribution parameter includes: Based on the strategy execution intensity value, calculate the inhibition factor through a linear mapping function; Introduce the inhibition factor into the liver fat synthesis simulation calculation module; When calculating the liver fat synthesis rate, multiply the original synthesis rate by 1 and the complement of the inhibition factor to realize quantitative regulation of liver fat synthesis simulation.

10. The digital-twin-based T2DM-NAFLD personalized lifestyle management system according to claim 1, wherein, The suggestion generation and output module includes: running a metabolic kinetics simulation using the adjusted digital twin model to obtain a predicted blood glucose profile and a predicted liver fat load trend; based on the predicted blood glucose profile and the predicted liver fat load trend, generating an exercise plan including an exercise type, a duration, an intensity level, and generating a diet arrangement including a food category, an intake amount, and an eating time window; and pushing the exercise plan and the diet arrangement to a user device in a visualized form through a mobile application interface.

Images