Digital twinning-based insulin closed-loop management method

Abstract

The invention discloses an insulin closed-loop management method based on digital twinning, and relates to the crossing field of medical information technology and medical instruments, and the method comprises the following steps: S1, a data fusion step; s2, constructing and updating a personalized digital metabolic twinborn model; step S3, performing prospective disturbance prediction; s4, performing multi-step rolling optimization control; s5, performing dynamic safety verification and intervention; step S6, an instruction execution step; according to the insulin closed-loop management method based on digital twinning, the digital metabolic twinning model of an individual is constructed, and foresight prediction and active safety verification are fused, so that automatic blood glucose control which is more stable, safer and more suitable for the living mode of the individual is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical information technology and medical devices, in particular to an insulin closed-loop management method based on digital twinning. BACKGROUND

[0002] Insulin pump is an important tool for implementing intensive insulin therapy, and closed-loop insulin delivery system (artificial pancreas) is a hot and difficult point of current research. Its goal is to simulate the physiological function of healthy beta cells in the pancreas through continuous glucose monitoring (CGM), control algorithm and insulin pump, and automatically adjust insulin infusion.

[0003] Although the existing commercial closed-loop systems have made significant progress, they still have the following key defects: hysteresis and passive reaction: the mainstream system is based on proportional-integral-derivative (PID) or its variant algorithm, which essentially reacts to the current or past blood glucose deviation. Due to the delay of subcutaneous insulin absorption and action (usually 40-60 minutes), this "after-correction" mode is difficult to effectively smooth the postprandial sharp blood glucose fluctuations. Limited personalization and adaptability: system parameters (such as basal rate, insulin sensitivity coefficient) are usually preset by doctors, or can only be slowly adjusted through limited rules. It cannot adapt to the dynamic changes of insulin sensitivity caused by diet, exercise, stress, disease, menstrual cycle, etc. in real time and automatically. Single information dimension, insufficient decision-making intelligence: control decisions rely heavily on blood glucose time series data, and fail to fully utilize multi-modal information such as exercise, heart rate provided by wearable devices, and user input meals, schedules, etc. leading to the system's inability to "understand" the user's behavior intentions and make proactive decisions. Passive safety mechanism: safety protection relies heavily on setting static blood glucose threshold alarms (such as alarms below 4.4 mmol / L). This is a passive and after-the-fact protection that cannot intervene at the beginning of a rapid blood glucose decline trend, and has limited effect on preventing hypoglycemia. SUMMARY

[0004] The purpose of the present application is to provide an insulin closed-loop management method based on digital twinning, which realizes more stable, safer and more individualized life mode blood glucose automatic control by constructing individual digital metabolic twin model and integrating proactive prediction and active safety verification.

[0005] The present application provides an insulin closed-loop management method based on digital twinning, comprising the following steps: Step S1, data fusion, continuously acquiring a set of time series data of the user, the time series data set including blood glucose data from a continuous glucose monitor, physiological and behavioral activity data from a wearable device, and environmental context data from a user interaction interface; Step S2, personalized digital metabolic twin model construction and updating, based on the time series data set, establish and continuously calibrate a personalized digital model reflecting the user's individual glucose-insulin metabolic dynamics; Step S3, forward-looking disturbance prediction, analyzing environmental context data and behavior activity data, identifying predetermined or high-probability imminent metabolic disturbance events, and quantifying the impact of the events on future blood glucose trajectory; Step S4, multi-step rolling optimization control, in each control period; taking the current system state as the initial value, using the personalized digital model, predicting the open-loop blood glucose trajectory in a predetermined optimization time domain in the future without intervention; taking the minimization of the deviation of the predicted blood glucose value from the target range as the optimization objective, and at the same time imposing a smoothness constraint on the insulin infusion rate, solving the optimal insulin infusion sequence in the future control time domain; Step S5, dynamic safety check and intervention, before executing the first infusion instruction of the optimal insulin infusion sequence, based on the personalized digital model and the current physiological state, dynamically calculating a time-varying safe insulin infusion rate boundary, comparing the optimal instruction with the safe boundary, if the instruction exceeds the boundary, modifying it to the boundary value or triggering a safety alarm; Step S6, instruction execution, sending the insulin infusion instruction checked in step S5 to the insulin pump for execution.

[0006] Preferably, in step S3, identifying metabolic disturbance events includes predicting future meal times and carbohydrate intake by analyzing schedule, geographic location information and historical behavior patterns on the user terminal; detecting the start of exercise, identifying the type of exercise and predicting its duration and intensity, and further predicting energy consumption levels by analyzing real-time motion sensor data and heart rate data of wearable devices.

[0007] Preferably, in step S2, the personalized digital model is a gated recurrent unit network model based on attention mechanism, which can model long-range temporal dependencies in the time series data set and output attention weights of different input data features on current blood glucose change.

[0008] Preferably, in step S5, dynamically calculating the safe insulin infusion rate boundary, specifically including: using the personalized digital model, taking the current blood glucose state as the starting point, reverse simulation to calculate the maximum insulin infusion rate allowed at the current time under the constraint condition of not triggering hypoglycemia in the entire prediction time domain, and setting this value as the safe boundary.

[0009] Preferably, in step S4, the optimization objective function also contains a term that penalizes the predicted hypoglycemia index, which is a function of the degree and duration of blood glucose values below the safety threshold in the prediction time domain.

[0010] Preferably, in step S2, the construction and updating of the personalized digital model are performed on a cloud server, while the prediction, optimization and security verification tasks in steps S4 and S5, which require low latency, are performed on an edge computing device local to the user, and the local and cloud servers synchronize model parameters through a secure communication protocol.

[0011] Therefore, the present application adopts the above-mentioned insulin closed-loop management method based on digital twinning, realizes more stable, safer and more individualized life mode-compliant automatic blood glucose control by constructing an individual digital metabolic twinning model and combining prospective prediction and active security verification.

[0012] The technical solutions of the present application will be further described in detail below through examples. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 FIG. 1 is a schematic diagram of the overall structure of the present application, which is an insulin closed-loop management method based on digital twinning. DETAILED DESCRIPTION

[0014] The technical solutions of the present application will be further described in detail below through examples.

[0015] Unless otherwise defined, the technical terms or scientific terms used in the present application shall have the usual meanings understood by those with ordinary skills in the art to which the present application belongs.

[0016] The terms "first", "second", and similar terms used in the present application do not indicate any order, number or importance, but are only used to distinguish different components. The terms "include" or "contain" and similar terms mean that the elements or objects before the terms cover the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right" and the like only indicate relative positional relationships, which may change accordingly when the absolute positions of the described objects change.

[0017] Example 1 As shown in FIG. 1, the present application is an insulin closed-loop management method based on digital twinning, which includes the following steps: Figure 1 Step S1, data fusion, continuously acquiring a time series data set of a user, the time series data set including blood glucose data from a continuous glucose monitor, physiological and behavioral activity data from a wearable device, and environmental context data from a user interface; Step S1, data fusion, continuously acquiring a time series data set of a user, the time series data set including blood glucose data from a continuous glucose monitor, physiological and behavioral activity data from a wearable device, and environmental context data from a user interface; Step S2, personalized digital metabolic twin model construction and updating, based on the time series data set, a personalized digital model reflecting the user's individual glucose-insulin metabolism dynamics is established and continuously calibrated; calibration is based on recursive estimation algorithm, using the latest data to dynamically adjust the model parameters, so that the model output approximates the user's real physiological state.

[0018] In step S2, the personalized digital model is a gated recurrent unit network model based on attention mechanism, which can model the long-range temporal dependence relationship in the time series data set, and output the attention weight of different input data features on the current blood glucose change.

[0019] In step S2, the construction and updating task of the personalized digital model is deployed on the cloud server for execution, while the prediction, optimization and safety check tasks required in steps S4 and S5 with low delay are deployed on the user's local edge computing device for execution. The local and cloud synchronize the model parameters through a secure communication protocol.

[0020] The system continuously receives blood glucose data from CGM, physiological data from smartwatch, and environmental data from mobile phone APP. The personalized digital metabolic twin model calibrates the model state to the current optimal through recursive update algorithm according to the latest time series data. The model estimates the user's current insulin sensitivity factor and carbohydrate ratio.

[0021] Step S3, prospective perturbation prediction, analyzes environmental context data and behavior activity data, identifies predetermined or high-probability imminent metabolic perturbation events, and quantifies the impact of the events on future blood glucose trajectory; metabolic perturbation events include at least carbohydrate intake events and / or physical exertion events.

[0022] In step S3, identifying metabolic perturbation events includes predicting future meal times and carbohydrate intake by analyzing schedule, geographic location information and historical behavior patterns on the user terminal; detecting the start of exercise, identifying exercise type and predicting its duration and intensity, and then predicting energy consumption level by analyzing real-time motion sensor data and heart rate data of wearable devices.

[0023] The meal and behavior prediction unit is activated. It comprehensively analyzes the current time, geographic location and schedule information, and matches with the user's historical behavior pattern library. The unit makes a prediction, which is quantified as an input perturbation signal and passed to the next module.

[0024] Step S4, multi-step rolling optimization control: in each control cycle; with the current system state as the initial value, the personalized digital model is used to predict the open-loop blood glucose trajectory in a predetermined optimization time domain in the future under the condition of no intervention; the optimal insulin infusion sequence in the future control time domain is solved by minimizing the deviation of the predicted blood glucose value from the target range and at the same time imposing a smoothness constraint on the change rate of insulin infusion as the optimization objective; wherein the solving process fuses the predicted metabolic disturbance event in step S3 as feedforward information.

[0025] In step S4, the optimization objective function also contains a term for punishing the predicted hypoglycemia index, which is a function of the degree and duration of the blood glucose value below the safety threshold in the prediction time domain.

[0026] The multi-step rolling optimization unit starts a new control cycle, open-loop prediction: the personalized digital model is used to predict the blood glucose trajectory in the future 3 hours with the current blood glucose as the starting point. If no action is taken, the model prediction shows that the blood glucose will soar after the meal; optimization solving: the optimization unit takes the meal disturbance predicted in step S3 as known feedforward information. The optimization objective is to find an insulin infusion strategy that makes the future blood glucose trajectory as smooth as possible within the target range. After calculation, the optimization algorithm proposes a preliminary infusion instruction with a duration of 15 minutes. This instruction aims to pre-infuse a part of insulin before food absorption to match the postprandial blood glucose rise.

[0027] Step S5, dynamic safety check and intervention: before executing the first infusion instruction of the optimal insulin infusion sequence, a time-varying safety insulin infusion rate boundary is dynamically calculated based on the personalized digital model and the current physiological state, the optimal instruction is compared with the safety boundary, and if the instruction exceeds the boundary, it is modified to the boundary value or a safety warning is triggered.

[0028] In step S5, the safety insulin infusion rate boundary is dynamically calculated, which specifically includes: using the personalized digital model, taking the current blood glucose state as the starting point, and performing reverse simulation to calculate the maximum insulin infusion rate allowed at the current time under the condition that the low blood glucose constraint is not triggered in the entire prediction time domain, and setting this value as the safety boundary.

[0029] Before executing the above instruction, the safety constraint and intervention module first performs real-time checking; dynamic safety boundary calculation: the module uses the same personalized digital model to perform a reverse safety simulation. Starting from the current state, to ensure that there is no hypoglycemia in the future 4 hours. Instruction checking and decision-making: compare the optimization instruction with the dynamic safety boundary. Since the optimization instruction is less than the dynamic safety boundary, the instruction is determined to be safe.

[0030] Step S6, instruction execution, the insulin infusion instruction verified in step S5 is sent to the insulin pump for execution. The final instruction verified is sent to the insulin pump by the execution and communication module; the insulin pump starts on time and infuses the insulin at a precise rate until completion.

[0031] Therefore, the present application adopts the above-mentioned insulin closed-loop management method based on digital twinning, constructs a digital metabolic twinning model of an individual, and realizes more stable, safer, and more individual life mode-adapted automatic blood glucose control through the fusion of prospective prediction and active safety verification.

[0032] The above embodiments are only used to illustrate the technical solutions of the present application but not to limit it, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can still be modified or replaced by equivalents, and these modifications or replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A closed-loop insulin management method based on digital twins, characterized in that, Includes the following steps: Step S1: Data fusion, continuously acquiring the user's time-series data set, which includes blood glucose data from a continuous glucose monitor, physiological and behavioral activity data from wearable devices, and environmental context data from the user interface; Step S2: Construction and updating of personalized digital metabolic twin model. Based on time series data set, establish and continuously calibrate a personalized digital model that reflects the individual glucose-insulin metabolic dynamics of the user. Step S3: Prospective perturbation prediction. Analyze environmental context data and behavioral activity data to identify predetermined or high-probability metabolic perturbation events, and quantify and predict the impact of these events on future blood glucose trajectories. Step S4: Multi-step rolling optimization control. In each control cycle, using the current system state as the initial value, a personalized digital model is used to predict the open-loop blood glucose trajectory in a predetermined future optimization time domain under no-intervention conditions. The optimal insulin infusion sequence in a future control time domain is solved with the optimization objective of minimizing the deviation between the predicted blood glucose value and the target range, while simultaneously applying a smoothness constraint on the rate of change of insulin infusion. Step S5, Dynamic Safety Verification and Intervention: Before executing the first infusion command of the optimal insulin infusion sequence, a time-varying safe insulin infusion rate boundary is dynamically calculated based on the personalized digital model and the current physiological state. The optimal command is compared with the safety boundary. If the command exceeds the boundary, it is corrected to the boundary value or a safety alarm is triggered. Step S6: Instruction execution. The insulin infusion instruction verified in step S5 is sent to the insulin pump for execution.

2. The insulin closed-loop management method based on digital twin according to claim 1, characterized in that, In step S3, identifying metabolic disturbance events includes predicting future meal times and carbohydrate intake by analyzing the user's schedule, geographic location information, and historical behavioral patterns on the user terminal; and detecting the start of exercise, identifying the type of exercise, and predicting its duration and intensity by analyzing real-time motion sensor data and heart rate data from wearable devices, thereby predicting energy consumption levels.

3. A closed-loop insulin management method based on digital twins according to claim 1, characterized in that, In step S2, the personalized digital model is a gated recurrent unit network model based on an attention mechanism, which can model long-range temporal dependencies in a time series dataset and output attention weights for the influence of different input data features on current blood glucose changes.

4. A closed-loop insulin management method based on digital twins according to claim 1, characterized in that, In step S5, the safe insulin infusion rate boundary is dynamically calculated. Specifically, this includes using a personalized digital model to simulate and calculate the maximum insulin infusion rate that can be allowed at the current moment under the condition that the hypoglycemia constraint is not triggered throughout the entire prediction time domain, starting from the current blood glucose state, and setting this value as the safe boundary.

5. A closed-loop insulin management method based on digital twins according to claim 1, characterized in that, In step S4, the objective function for optimization also includes a term that penalizes the predicted hypoglycemia index, which is a function of the degree and duration of the predicted blood glucose value being below a safe threshold in the time domain.

6. A closed-loop insulin management method based on digital twins according to claim 1, characterized in that, In step S2, the task of building and updating the personalized digital model is deployed on the cloud server, while the low-latency prediction, optimization and security verification tasks in steps S4 and S5 are deployed on the user's local edge computing device. The local and cloud devices synchronize model parameters through a secure communication protocol.

Images