Machine learning-based low blood sugar induction closed-loop control method and system

Abstract

The present application relates to the field of machine learning assisted medical closed-loop regulation, a low blood sugar induced closed-loop regulation method and system based on machine learning, comprising: receiving a closed-loop regulation instruction, confirming a to-be-regulated object and a closed-loop regulation system based on the closed-loop regulation instruction, acquiring a continuous blood glucose sequence and electrocardio monitoring data according to a preset control period by using a continuous blood glucose monitoring unit and an electrocardio monitoring unit, acquiring historical infusion data of the to-be-regulated object, a target low blood sugar interval and individualized sensitivity parameters, executing an insulin infusion instruction by using an infusion control unit to obtain an actual infusion amount and an actual blood glucose response, and updating the individualized sensitivity parameters and a blood glucose prediction model based on the actual infusion amount and the actual blood glucose response. The present application can realize safe and accurate control of low blood sugar induction.

Description

Technical Field

[0001] This invention relates to the field of machine learning-assisted medical closed-loop regulation, and in particular to a machine learning-based method and system for hypoglycemia-induced closed-loop regulation. Background Technology

[0002] In the field of medical monitoring and automated infusion control, hypoglycemia induction regulation requires guiding the blood glucose level of the subject to a target hypoglycemic range under specific therapeutic purposes, while avoiding safety risks such as excessive blood glucose drop and electrocardiographic abnormalities during the regulation process. With the development of continuous blood glucose monitoring, electrocardiographic monitoring, and infusion pump control technologies, the hypoglycemia induction process now has the basic conditions for real-time acquisition of physiological data and execution of drug infusion.

[0003] However, existing hypoglycemia induction and regulation methods largely rely on the experience of healthcare professionals to set dosages and make manual adjustments. The control process is usually based on the current blood glucose level or a fixed infusion rule, making it difficult to fully consider blood glucose trends, insulin lag, and differences in insulin sensitivity among different individuals. When blood glucose decreases rapidly or there is a risk of future blood glucose exceeding the limit, traditional methods often cannot make decisions to reduce the dosage or stop the infusion in advance, resulting in a lag in response.

[0004] Furthermore, existing methods do not adequately utilize ECG monitoring data, historical infusion data, and actual blood glucose responses, lacking a mechanism to integrate multi-source physiological data, predicted blood glucose results, and individualized parameters into closed-loop regulation. During treatment, if individualized sensitivity parameters and prediction models cannot be continuously updated based on actual responses, it can easily lead to a mismatch between subsequent control strategies and the actual response of the patient. Therefore, how to achieve blood glucose trend prediction, infusion risk assessment, safety verification, and individualized closed-loop updates during hypoglycemia induction has become an urgent technical problem to be solved. Summary of the Invention

[0005] This invention provides a machine learning-based closed-loop regulation method for hypoglycemia induction and a computer-readable storage medium, the main purpose of which is to achieve safe and precise control of hypoglycemia induction.

[0006] To achieve the above objectives, this invention provides a machine learning-based closed-loop regulation method for hypoglycemia induction, comprising: Receive closed-loop control instructions, and identify the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. Using a continuous blood glucose monitoring unit and an electrocardiogram (ECG) monitoring unit, continuous blood glucose sequences and ECG monitoring data are acquired according to a preset control cycle. The current blood glucose value and blood glucose change rate are obtained based on the continuous blood glucose sequence, and the ECG safety index is obtained based on the ECG monitoring data. The system acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and a pre-built blood glucose prediction model, it obtains a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters. The hypoglycemia risk index is obtained based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set and ECG safety index. Insulin infusion instructions are obtained based on the hypoglycemia risk index, target hypoglycemia range and individualized sensitivity parameters. Using a safety protection unit, the insulin infusion command is safety-verified based on the hypoglycemia risk index to obtain a verification result, which is either pass or fail. After confirming that the verification result is pass, the insulin infusion command is executed by the infusion control unit to obtain the actual infusion volume and the actual blood glucose response. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, and the process returns to the step of obtaining continuous blood glucose sequences and electrocardiogram monitoring data according to the preset control cycle, thereby realizing closed-loop regulation of hypoglycemia induction in the target subject.

[0007] Optionally, the step of obtaining the current blood glucose value and the rate of change of blood glucose based on a continuous blood glucose sequence includes: The continuous blood glucose sequence is subjected to outlier removal and time alignment operations to obtain an effective blood glucose sequence, wherein the effective blood glucose sequence includes multiple blood glucose measurement values, and the latest blood glucose measurement value is extracted from the effective blood glucose sequence as the current blood glucose value; Using a preset rate analysis window, a rate analysis sequence is extracted from the effective blood glucose sequence. The rate analysis sequence is then subjected to least squares linear fitting to obtain the fitting slope, which is used as the rate of blood glucose change.

[0008] Optionally, obtaining the ECG safety index based on ECG monitoring data includes: Feature extraction is performed on the electrocardiogram (ECG) monitoring data to obtain ECG feature nodes, which include: corrected QT interval, heart rate fluctuation, and arrhythmia event identifiers. The abnormal QT component is obtained based on the corrected QT interval and the preset QT risk mapping function; the heart rate fluctuation component is obtained based on the heart rate fluctuation and the preset heart rate risk mapping function; and the arrhythmia component is obtained based on the arrhythmia event identifier and the preset arrhythmia mapping function. The QT abnormal component, heart rate fluctuation component, and arrhythmia component are weighted and fused to obtain the ECG safety index.

[0009] Optionally, the step of obtaining the predicted blood glucose value set based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters includes: The current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters are concatenated to obtain the input feature vector; Using a preset set of prediction time points, the input feature vector is input into the blood glucose prediction model to obtain a set of predicted blood glucose values. The set of predicted blood glucose values ​​includes multiple predicted blood glucose values, and the predicted blood glucose values ​​correspond one-to-one with the prediction time points in the set of prediction time points.

[0010] Optionally, obtaining the hypoglycemia risk index based on the current blood glucose level, blood glucose rate of change, predicted blood glucose value set, and electrocardiogram safety index includes: Obtain risk threshold nodes, which include: the lower limit of safe blood glucose and the upper limit of blood glucose rate of decline; The current blood glucose value is compared with the lower limit of blood glucose safety to obtain a blood glucose safety indicator. The blood glucose safety indicator is either above or below the limit. If the blood glucose safety indicator is above the limit, the difference between the lower limit of blood glucose safety and the current blood glucose value is calculated to obtain the blood glucose deviation margin. The ratio of the blood glucose deviation margin to the lower limit of blood glucose safety is confirmed as the absolute blood glucose risk component. Otherwise, zero is confirmed as the absolute blood glucose risk component. The rate of change of blood glucose is compared with zero to obtain the rate direction indicator, which is either decreasing or not decreasing. After confirming that the rate direction indicator is decreasing, the absolute value of the rate of change of blood glucose is taken to obtain the magnitude of the decrease in blood glucose. The ratio of the magnitude of the decrease in blood glucose to the upper limit of the rate of decrease in blood glucose is confirmed as the rate decrease risk component. Otherwise, zero is confirmed as the rate decrease risk component. The minimum predicted blood glucose value is extracted from the set of predicted blood glucose values ​​to obtain the lowest predicted blood glucose value; The predicted minimum blood glucose value is compared with the lower limit of blood glucose safety to obtain a predicted safety indicator, which is either exceeded or not exceeded. After confirming that the predicted safety indicator is exceeded, the difference between the lower limit of blood glucose safety and the predicted lowest blood glucose value is calculated to obtain the predicted exceedance range. The ratio of the predicted exceedance range to the lower limit of blood glucose safety is confirmed as the predicted hypoglycemia risk component; otherwise, zero is confirmed as the predicted hypoglycemia risk component. The ECG safety index is input into a preset ECG risk mapping function to obtain the ECG abnormality risk component; The hypoglycemia risk index is obtained by weighting and fusing the absolute risk component of blood glucose, the risk component of decreased rate of blood glucose, the risk component of predicted hypoglycemia, and the risk component of electrocardiographic abnormality. The calculation formula is shown below: in, Indicates the risk index of hypoglycemia. This indicates the absolute risk component of blood glucose. This indicates the risk component of the rate of decline. This indicates the component representing the predicted risk of hypoglycemia. Indicates the risk component of electrocardiogram abnormalities. , , , All of these represent preset weight values.

[0011] Optionally, obtaining insulin infusion instructions based on the hypoglycemia risk index, target hypoglycemia range, and individualized sensitivity parameters includes: The upper and lower bounds of the target hypoglycemia range are obtained; Compare the current blood glucose level with the target upper and lower bounds; If the current blood glucose level is higher than the target upper limit, the difference between the current blood glucose level and the target upper limit is calculated to obtain the blood glucose deviation. The blood glucose deviation is then divided by the individualized sensitivity parameter to obtain the basic infusion volume. If the current blood glucose level is between the target lower bound and the target upper bound, the preset maintenance infusion volume will be confirmed as the basic infusion volume. If the current blood glucose level is below the target lower limit, then zero will be considered as the basic infusion volume; Calculate the difference between the predicted minimum blood glucose value and the preset prediction safety lower limit, and input the difference into the preset prediction constraint mapping function to obtain the prediction constraint coefficient, wherein the prediction safety lower limit is not lower than the blood glucose safety lower limit; The hypoglycemia risk index is input into a preset risk correction mapping function to obtain the risk correction coefficient; The target infusion rate is obtained by multiplying the baseline infusion volume, the risk correction coefficient, and the prediction constraint coefficient, and an insulin infusion command is generated based on the target infusion rate.

[0012] Optionally, the step of performing a safety verification on the insulin infusion command based on the hypoglycemia risk index to obtain the verification result includes: By using a preset risk level threshold and the hypoglycemia risk index for classification and matching, a risk level identifier is obtained, wherein the risk level identifier is a safety level identifier or a danger level identifier. Compare the predicted minimum blood glucose value with the predicted safety lower limit to obtain a predicted limit violation indicator, wherein the predicted limit violation indicator is either predicted limit violation or predicted no limit violation; After confirming that the risk level identifier is a safety level identifier and the predicted limit exceedance identifier is a predicted no limit exceedance, the insulin infusion instruction is confirmed to have passed the safety verification, and the verification result is confirmed as passed; Otherwise, the insulin infusion command is confirmed to have failed the security check, and the check result is confirmed as unsuccessful.

[0013] Optionally, updating the individualized sensitivity parameters and blood glucose prediction model based on the actual infusion volume and actual blood glucose response includes: Based on a preset insulin action hysteresis window, the hysteresis sampling value is extracted from the actual blood glucose response; Calculate the difference between the current blood glucose value and the lagged sample value to obtain the blood glucose response difference; When the actual infusion volume is greater than the preset minimum effective infusion threshold, the ratio of the blood glucose response difference to the actual infusion volume is calculated to obtain the instantaneous sensitivity. The updated individualized sensitivity parameters are calculated based on the instantaneous sensitivity and individualized sensitivity parameters, using the following formula: in, This indicates the updated individual sensitivity parameter. This represents an individualized sensitivity parameter. Indicates the difference in blood glucose response. Indicates the actual infusion volume. Indicates instantaneous sensitivity. Indicates the preset reference sensitivity. This represents the minimum effective infusion threshold. This represents the preset smoothing coefficient, and ; The blood glucose prediction model is updated based on the actual infusion volume and the actual blood glucose response, resulting in the updated blood glucose prediction model.

[0014] Optionally, updating the blood glucose prediction model based on the actual infusion volume and actual blood glucose response to obtain the updated blood glucose prediction model includes: At each prediction time point in the prediction time point set, the actual blood glucose response is sampled to obtain multiple actual blood glucose sampling values, wherein each actual blood glucose sampling value corresponds one-to-one with a prediction time point. For each of the multiple actual blood glucose samples, perform the following operation: Using the actual blood glucose sampling value, the corresponding reference blood glucose value is retrieved from the predicted blood glucose value set, wherein the prediction time point corresponding to the reference blood glucose value is the same as the prediction time point corresponding to the actual blood glucose sampling value. Calculate the difference between the actual blood glucose sample value and the reference blood glucose value to obtain the prediction residual value; Summarize the predicted residual values ​​to obtain the predicted residual value set; Based on the predicted residual set, actual infusion volume, and historical infusion data, the model parameters of the blood glucose prediction model are updated online incrementally to obtain the updated blood glucose prediction model.

[0015] To achieve the above objectives, the present invention also provides a machine learning-based closed-loop regulation system for hypoglycemia induction, comprising: The object confirmation module is used to receive closed-loop control instructions and confirm the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. The monitoring and prediction module is used to acquire continuous blood glucose sequence and ECG monitoring data according to a preset control cycle using the continuous blood glucose monitoring unit and ECG monitoring unit, to obtain the current blood glucose value and blood glucose change rate based on the continuous blood glucose sequence, and to obtain the ECG safety index based on the ECG monitoring data. The system acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and a pre-built blood glucose prediction model, it obtains a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters. The risk decision module is used to obtain a hypoglycemia risk index based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set and ECG safety index, and to obtain insulin infusion instructions based on the hypoglycemia risk index, target hypoglycemia range and individualized sensitivity parameters. The verification and update module is used to perform safety verification on insulin infusion commands based on the hypoglycemia risk index using the safety protection unit, and obtain the verification result, which is either pass or fail. After confirming that the verification result is pass, the infusion control unit executes the insulin infusion command to obtain the actual infusion volume and the actual blood glucose response. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, and the process returns to the step of obtaining continuous blood glucose sequences and electrocardiogram monitoring data according to the preset control cycle, thereby realizing closed-loop regulation of hypoglycemia induction in the target subject.

[0016] To address the above problems, the present invention also provides an electronic device, the electronic device comprising: Memory, storing at least one instruction; The processor executes the instructions stored in the memory to implement the aforementioned machine learning-based closed-loop regulation method for hypoglycemia induction.

[0017] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the aforementioned machine learning-based hypoglycemia-induced closed-loop regulation method.

[0018] To address the problems described in the background art, this invention receives closed-loop control commands and identifies the object to be controlled and the closed-loop control system based on these commands. The closed-loop control system includes a continuous blood glucose monitoring unit, an electrocardiogram (ECG) monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. Using the continuous blood glucose monitoring unit and the ECG monitoring unit, continuous blood glucose sequences and ECG monitoring data are acquired according to a preset control cycle. The current blood glucose value and rate of change are obtained based on the continuous blood glucose sequence, and an ECG safety index is obtained based on the ECG monitoring data. This invention obtains the ECG safety index by first extracting core ECG features, then mapping and weighting them separately, enabling subsequent risk calculations to simultaneously reflect the QT interval and cardiac function using a single value. This invention assesses the electrocardiographic safety of the subject from three aspects: rate of blood glucose fluctuation, arrhythmia, and other factors. It acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters. Using a control processing unit and a pre-built blood glucose prediction model, a predicted blood glucose value set is obtained based on the current blood glucose value, rate of blood glucose change, electrocardiographic safety index, historical infusion data, and individualized sensitivity parameters. This invention, by concatenating multiple physiological and therapeutic state features into a fixed-format vector and simultaneously outputting multi-time-point predictions through a machine learning model, enables subsequent risk calculation and control decisions to adjust in advance based on future blood glucose trends. This compensates for the shortcomings of lagging control relying solely on the current blood glucose value. The invention obtains a hypoglycemia risk index based on the current blood glucose value, rate of blood glucose change, predicted blood glucose value set, and electrocardiographic safety index. Based on the hypoglycemia risk index, target hypoglycemia range, and individualized sensitivity parameters, this invention obtains insulin infusion instructions. It is evident that this invention achieves a balance between therapeutic efficacy and safety control by first determining the baseline infusion volume based on range deviation and individualized sensitivity parameters, and then adjusting the baseline infusion volume using risk correction coefficients and prediction constraint coefficients. This approach drives blood glucose levels towards the target hypoglycemia range while actively reducing the risk of future hypoglycemia. A safety protection unit performs a safety verification of the insulin infusion instructions based on the hypoglycemia risk index, obtaining a verification result (pass or fail). Upon confirming a pass result, the infusion control unit executes the insulin infusion instructions, obtaining the actual infusion volume and actual blood glucose response. Therefore, this invention... This invention introduces a safety protection unit independent of the control processing unit to perform two checks on the hypoglycemia risk index and the predicted minimum blood glucose value. This ensures that even if the control processing unit generates excessive insulin infusion commands due to model bias, it cannot directly drive the insulin infusion pump, thus improving the safety of hypoglycemia induction therapy. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, returning to the steps of obtaining continuous blood glucose sequences and ECG monitoring data according to the preset control cycle. This achieves closed-loop control of hypoglycemia induction for the target population. It is evident that this invention uses prediction deviations at multiple prediction time points within each control cycle to correct the blood glucose prediction model and performs online small-step updates in conjunction with the actual infusion volume and historical infusion data.This invention enables the blood glucose prediction model to continuously approximate the actual blood glucose changes of the target individual during treatment, forming a complete closed-loop learning process together with the updating of individualized sensitivity parameters. Therefore, this invention can achieve safe and precise control of hypoglycemia-induced hypoglycemia. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a machine learning-based closed-loop regulation method for hypoglycemia induction provided in an embodiment of the present invention; Figure 2 A functional block diagram of a hypoglycemia-induced closed-loop regulation system based on machine learning is provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device that implements the machine learning-based hypoglycemia-induced closed-loop regulation method according to an embodiment of the present invention.

[0020] Explanation of reference numerals in the attached figures: 10. Electronic device; 11. Processor; 12. Memory; 13. Bus.

[0021] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] This application provides a machine learning-based closed-loop regulation method for hypoglycemia induction. The executing entity of this machine learning-based closed-loop regulation method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the machine learning-based closed-loop regulation method for hypoglycemia induction can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0024] Reference Figure 1 The diagram shown is a flowchart illustrating a machine learning-based closed-loop regulation method for hypoglycemia induction according to an embodiment of the present invention. In this embodiment, the machine learning-based closed-loop regulation method for hypoglycemia induction includes: S1. Receive closed-loop control instructions, and confirm the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit.

[0025] It should be explained that the closed-loop control command is a control command issued by medical staff through a bedside treatment terminal to initiate the hypoglycemia induction treatment process, carrying the identification number of the patient to be controlled and the treatment task parameters. The patient to be controlled is a patient receiving hypoglycemia induction adjunctive therapy. The closed-loop control system is an integrated hardware and software platform deployed around the bedside to complete blood glucose collection, electrocardiogram monitoring, control calculation, safety protection, and insulin infusion. The continuous blood glucose monitoring unit is an electrochemical glucose sensor device attached to the skin of the patient to be controlled, which can output interstitial fluid glucose measurement data at preset sampling intervals. The electrocardiogram monitoring unit is a multi-lead electrocardiogram monitor that collects the electrocardiogram waveform of the patient to be controlled through surface electrodes, and can output the electrocardiogram waveform and clinical electrocardiogram parameters derived from the waveform in real time. The control processing unit is a computing module with an embedded processor as its core, running the blood glucose prediction model and risk calculation program, and undertaking the tasks of data reception, feature generation, model calculation, and command generation. The safety protection unit is a hardware watchdog module independent of the control processing unit, which only undertakes the safety judgment responsibility of allowing or blocking subsequent insulin infusion commands. The infusion control unit is an insulin infusion pump drive module with a stepper motor and a flow feedback sensor. It pushes the needle to inject insulin according to the insulin infusion command and reports the actual injection volume.

[0026] For example, when a doctor initiates an insulin tolerance test in a ward, the patient's blood glucose needs to be actively regulated to the target hypoglycemic range. Therefore, a closed-loop regulation command is issued on the bedside treatment terminal, which immediately identifies the patient as the subject to be regulated and the closed-loop regulation system, consisting of a subcutaneous glucose sensor (continuous blood glucose monitoring unit), a 12-lead electrocardiogram monitor (electrocardiogram monitoring unit), an embedded control host (control processing unit), an independent watchdog board (safety protection unit), and a micro-insulin infusion pump (infusion control unit), assists the doctor in completing the hypoglycemic induction treatment task.

[0027] S2. Using a continuous blood glucose monitoring unit and an electrocardiogram (ECG) monitoring unit, acquire continuous blood glucose sequences and ECG monitoring data according to a preset control cycle. Based on the continuous blood glucose sequence, acquire the current blood glucose value and the rate of blood glucose change. Based on the ECG monitoring data, acquire the ECG safety index.

[0028] It should be understood that the control cycle is a fixed time interval used to trigger each round of closed-loop calculation in this method. The value of the control cycle can be configured based on the insulin onset lag and the interstitial fluid glucose delay. The continuous blood glucose sequence is a set of blood glucose measurement data continuously output by the continuous blood glucose monitoring unit at its sampling interval within the time window corresponding to the control cycle, sorted from earliest to latest by sampling time. The electrocardiogram (ECG) monitoring data is the ECG waveform and waveform derived parameters synchronously output by the ECG monitoring unit within the time window corresponding to the control cycle. The current blood glucose value is the latest blood glucose measurement result of the subject to be regulated within the current control cycle. The rate of blood glucose change is the speed and direction of blood glucose change over time within the current control cycle. The ECG safety index is a quantitative indicator used to reflect the safety of the ECG state of the subject to be regulated for hypoglycemic induction treatment within the current control cycle; a larger value indicates a higher degree of ECG abnormality.

[0029] It is understandable that the raw blood glucose data output by the continuous blood glucose monitoring unit may contain sensor drift points and inconsistent sampling times, and a single blood glucose measurement cannot reflect the rate of blood glucose decline. Therefore, the step of obtaining the current blood glucose value and the rate of blood glucose change based on a continuous blood glucose sequence includes: The continuous blood glucose sequence is subjected to outlier removal and time alignment operations to obtain an effective blood glucose sequence, wherein the effective blood glucose sequence includes multiple blood glucose measurement values, and the latest blood glucose measurement value is extracted from the effective blood glucose sequence as the current blood glucose value; Using a preset rate analysis window, a rate analysis sequence is extracted from the effective blood glucose sequence. The rate analysis sequence is then subjected to least squares linear fitting to obtain the fitting slope, which is used as the rate of blood glucose change.

[0030] It should be explained that the outlier removal is the operation of identifying and removing isolated points in the continuous blood glucose sequence that significantly deviate from neighboring measurements. Optionally, an outlier detection method based on the moving median and moving absolute deviation can be used to achieve this purpose. The time alignment operation refers to the operation of resampling the blood glucose measurements after outlier removal to a unified time point according to the nominal sampling interval of the continuous blood glucose monitoring unit. Optionally, a linear interpolation method can be used to achieve this purpose. The effective blood glucose sequence is a sequence of blood glucose measurements arranged at a unified time point after outlier removal and time alignment operations. The blood glucose measurement value is a single data point in the effective blood glucose sequence, reflecting the interstitial fluid glucose concentration of the object to be regulated at the corresponding sampling time. The rate analysis window is the time length for calculating the rate by truncating a continuous measurement value from the end of the effective blood glucose sequence. The value of the rate analysis window can be configured according to the stability of blood glucose changes. The rate analysis sequence is a subsequence composed of the most recent blood glucose measurements truncated from the end of the effective blood glucose sequence by the rate analysis window. The least squares linear fitting process involves finding a linear function that minimizes the sum of squared deviations, with sampling time as the independent variable and blood glucose measurement as the dependent variable. The fitting slope is the coefficient of the linear function with respect to sampling time.

[0031] For example, assuming the control cycle is set to 5 minutes and the sampling interval of the continuous blood glucose monitoring unit is 1 minute, five raw blood glucose measurement data are obtained in each control cycle. Assuming the raw data are 5.0 mmol / L, 4.9 mmol / L, 9.9 mmol / L, 4.8 mmol / L, and 4.8 mmol / L respectively, the third measurement value of 9.9 mmol / L is significantly deviated from the adjacent values ​​and is identified and discarded. Then, linear interpolation time alignment is performed at 1-minute intervals to obtain the effective blood glucose sequence as 5.0 mmol / L, 4.9 mmol / L, 4.85 mmol / L, 4.8 mmol / L, and 4.8 mmol / L respectively. The latest blood glucose measurement value of 4.8 mmol / L is extracted from the end of the effective blood glucose sequence as the current blood glucose value. The rate analysis window is set to 5 minutes. All five blood glucose measurements from the effective blood glucose sequence are extracted from the end of the effective blood glucose sequence as the rate analysis sequence. A linear fit is performed with sampling time as the independent variable and blood glucose measurement as the dependent variable, and the fitted slope is -0.05 mmol / L / min. The fitted slope is used as the rate of blood glucose change.

[0032] Furthermore, the ECG monitoring data includes both the original ECG waveform and clinical parameters derived from the waveform. It is necessary to first select ECG features from the ECG monitoring data that directly reflect the risk of hypoglycemia-inducing treatment, and then synthesize them into a single index. Therefore, obtaining the ECG safety index based on ECG monitoring data includes: Feature extraction is performed on the electrocardiogram (ECG) monitoring data to obtain ECG feature nodes, which include: corrected QT interval, heart rate fluctuation, and arrhythmia event identifiers. The abnormal QT component is obtained based on the corrected QT interval and the preset QT risk mapping function; the heart rate fluctuation component is obtained based on the heart rate fluctuation and the preset heart rate risk mapping function; and the arrhythmia component is obtained based on the arrhythmia event identifier and the preset arrhythmia mapping function. The QT abnormal component, heart rate fluctuation component, and arrhythmia component are weighted and fused to obtain the ECG safety index.

[0033] It should be understood that the feature extraction is a process of identifying QRS complexes from the ECG monitoring data, calculating beat intervals, and inferring clinical ECG parameters accordingly. Optionally, this can be achieved using the Pan-Tompkins QRS detection algorithm combined with Bazett's formula correction. The ECG feature node is a data unit formed by combining several ECG feature fields within the current control cycle, storing all key parameters of the ECG state during the control cycle. The corrected QT interval is the interval value obtained by correcting the QT interval measured within the current control cycle according to the current heart rate, used to reduce the influence of heart rate on QT interval length. The heart rate fluctuation is a quantitative representation of the standard deviation of continuous beat intervals within the current control cycle, used to reflect the degree of heart rate fluctuation. The arrhythmia event identifier is a flag indicating whether arrhythmia events such as premature ventricular contractions or ventricular tachycardia are detected within the current control cycle. The QT risk mapping function is a piecewise linear function that maps the corrected QT interval to abnormal QT components. The segmentation point parameters in the QT risk mapping function are set according to the clinical safety protocols for hypoglycemia induction therapy. The heart rate risk mapping function is a piecewise linear function that maps heart rate variability to heart rate variability components. The segmentation points in the heart rate risk mapping function are also set according to the clinical safety protocols for hypoglycemia induction therapy. The arrhythmia mapping function is a piecewise function that outputs a high-risk constant when the arrhythmia event is identified as present and zero when it is not. The abnormal QT component is a risk value calculated by the QT risk mapping function, reflecting the degree to which the corrected QT interval deviates from the normal range. The heart rate variability component is a risk value calculated by the heart rate risk mapping function, reflecting the magnitude of heart rate variability. The arrhythmia component is a risk value calculated by the arrhythmia mapping function, reflecting the degree of impact of the arrhythmia event. The weighted fusion is an operation that sums the QT abnormal component, the heart rate fluctuation component, and the arrhythmia component according to their respective weights, and restricts the summation result to between 0 and 1. The weights used for weighting are calculated from historical ECG monitoring samples. Specifically, the ECG safety index labeled by experts is used as the target value, and the QT abnormal component, the heart rate fluctuation component, and the arrhythmia component are used as inputs. The weights are solved under the condition that the weights are not less than zero, and then processed so that the sum of the weights is one before being saved and used.

[0034] For example, the ECG monitoring unit outputs a 12-lead ECG waveform and corresponding derived parameters within the current control cycle. After feature extraction, ECG feature nodes are obtained, where the corrected QT interval is 455ms, heart rate fluctuation is 40ms, and arrhythmia events are marked as non-existent. The QT risk mapping function maps intervals not exceeding 440ms to 0, intervals between 440 and 500ms linearly to 0 and 1, and intervals not less than 500ms to 1. Therefore, the corrected QT interval of 455ms is mapped to a QT abnormality component of 0.25. The heart rate risk mapping function maps intervals not exceeding 20ms to 0, intervals between 20 and 80ms linearly to 0 and 1, and intervals not less than 80ms to 1. Therefore, the heart rate fluctuation of 40ms is mapped to a heart rate fluctuation component of 0.33. The arrhythmia mapping function outputs 0 for the absence of arrhythmia events, therefore the arrhythmia component is 0. The preset weights for the QT abnormality component, the heart rate fluctuation component, and the arrhythmia component are 0.4, 0.2, and 0.4, respectively. After weighted fusion, the resulting ECG safety index is 0.4 x 0.25 + 0.2 x 0.33 + 0.4 x 0 = 0.166, approximately 0.17. Therefore, the ECG safety index of the subject to be regulated within this control period is 0.17. This embodiment of the invention obtains the ECG safety index by first extracting core ECG features, then mapping and weighted fusion them separately. This allows subsequent risk calculations to simultaneously reflect the ECG safety status of QT interval, heart rate fluctuation, and arrhythmia using a single value.

[0035] S3. Obtain historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and the pre-built blood glucose prediction model, obtain a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters.

[0036] It should be explained that the historical infusion data is read from the treatment log of the infusion control unit, representing several insulin infusion records that have occurred from the start of this hypoglycemic induction treatment to the current moment. Each record includes the infusion time and the actual infusion volume. The target hypoglycemic range is a target blood glucose range set by medical staff, which is the range to which the patient's blood glucose is expected to be maintained during this treatment. It is read from the treatment task parameters written into the control processing unit by the bedside treatment terminal. The individualized sensitivity parameter is a patient-specific parameter used to characterize the decrease in blood glucose corresponding to each 1U of insulin infusion. At the start of treatment, medical staff input the initial value into the bedside treatment terminal based on the patient's previous insulin usage records or basic information such as weight and height. During treatment, this method automatically updates the value based on the actual response in each round. The blood glucose prediction model is a machine learning model pre-trained on a historical treatment case dataset. Optionally, a long short-term memory network can be used to achieve this purpose. The predicted blood glucose value set is a summary of predicted blood glucose values ​​corresponding to several prediction time points, calculated by the blood glucose prediction model based on the current input features.

[0037] Understandably, the blood glucose prediction model can only accept a feature vector in a uniform format as input. It needs to organize the current blood glucose value, blood glucose rate of change, ECG safety index, historical infusion data, and individualized sensitivity parameters into a single feature vector. Therefore, obtaining the predicted blood glucose value set based on the current blood glucose value, blood glucose rate of change, ECG safety index, historical infusion data, and individualized sensitivity parameters includes: The current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters are concatenated to obtain the input feature vector; Using a preset set of prediction time points, the input feature vector is input into the blood glucose prediction model to obtain a set of predicted blood glucose values. The set of predicted blood glucose values ​​includes multiple predicted blood glucose values, and the predicted blood glucose values ​​correspond one-to-one with the prediction time points in the set of prediction time points.

[0038] In detail, the feature concatenation is the operation of joining the input fields end-to-end into a one-dimensional real vector according to a pre-agreed fixed order. For fields like historical infusion data, where the number of records is variable, the most recent infusion records are first extracted according to a preset infusion review window and arranged in reverse chronological order. Then, the actual infusion amount in each record is sequentially filled into the corresponding slot of the vector. If the number of actual records is less than the length of the review window, it is padded with zeros. The input feature vector is a one-dimensional real vector with a fixed dimension obtained through feature concatenation, and it is the unique input to the blood glucose prediction model. The prediction time point set is a set of several time points extending backward relative to the current time, used to indicate the future time that the blood glucose prediction model needs to predict. The prediction time point is a single time point in the prediction time point set. The predicted blood glucose value is the blood glucose estimate output by the blood glucose prediction model for a single prediction time point at the corresponding time.

[0039] For example, the prediction time point set is set to three prediction time points: 10 minutes, 20 minutes, and 30 minutes after the current time. The infusion review window is set to the three most recent infusion records. Within the current control period, the current blood glucose value is 4.8 mmol / L, the blood glucose change rate is -0.05 mmol / L / min, the ECG safety index is 0.17, the individualized sensitivity parameter is 1.5 mmol / L / U, and the actual infusion volumes of the three most recent infusion records, in reverse chronological order, are 0.8, 0.7, and 0.6 U. Features are concatenated according to a pre-agreed field order to obtain the input feature vector: 4.8, -0.05, 0.17, 0.8, 0.7, 0.6, 1.5. The input feature vector is fed into a blood glucose prediction model based on a long short-term memory network. The model outputs predicted blood glucose values ​​of 4.2 mmol / L, 3.9 mmol / L, and 3.6 mmol / L at 10 min, 20 min, and 30 min, respectively, resulting in a set of predicted blood glucose values ​​of 4.2 mmol / L, 3.9 mmol / L, and 3.6 mmol / L. This embodiment of the invention concatenates multiple physiological and therapeutic state features into a fixed-format vector and feeds it into a machine learning model to simultaneously output multi-time-point predictions. This allows subsequent risk calculations and control decisions to be adjusted in advance based on future blood glucose trends, compensating for the shortcomings of lagging control that relies solely on current blood glucose values.

[0040] S4. Obtain a hypoglycemia risk index based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set, and ECG safety index. Obtain insulin infusion instructions based on the hypoglycemia risk index, target hypoglycemia range, and individualized sensitivity parameters.

[0041] It should be explained that the hypoglycemia risk index is a unified quantitative indicator that comprehensively reflects the probability of a severe hypoglycemic event in the subject under control within the current control cycle; a higher value indicates a closer proximity to a dangerous state. The insulin infusion command is a control command generated by the control processing unit based on the overall status within the current control cycle, used to instruct the infusion control unit to execute insulin infusion, and includes at least a target infusion rate field.

[0042] Understandably, relying solely on a single numerical value cannot comprehensively account for the impact of current blood glucose levels, the trend of blood glucose decline, the future risk of hypoglycemia, and ECG abnormalities. It is necessary to construct corresponding risk components separately and then fuse them into a unified index. Therefore, the method of obtaining a hypoglycemia risk index based on current blood glucose levels, blood glucose change rate, predicted blood glucose value set, and ECG safety index includes: Obtain risk threshold nodes, which include: the lower limit of safe blood glucose and the upper limit of blood glucose rate of decline; The current blood glucose value is compared with the lower limit of blood glucose safety to obtain a blood glucose safety indicator. The blood glucose safety indicator is either above or below the limit. If the blood glucose safety indicator is above the limit, the difference between the lower limit of blood glucose safety and the current blood glucose value is calculated to obtain the blood glucose deviation margin. The ratio of the blood glucose deviation margin to the lower limit of blood glucose safety is confirmed as the absolute blood glucose risk component. Otherwise, zero is confirmed as the absolute blood glucose risk component. The rate of change of blood glucose is compared with zero to obtain the rate direction indicator, which is either decreasing or not decreasing. After confirming that the rate direction indicator is decreasing, the absolute value of the rate of change of blood glucose is taken to obtain the magnitude of the decrease in blood glucose. The ratio of the magnitude of the decrease in blood glucose to the upper limit of the rate of decrease in blood glucose is confirmed as the rate decrease risk component. Otherwise, zero is confirmed as the rate decrease risk component. The minimum predicted blood glucose value is extracted from the set of predicted blood glucose values ​​to obtain the lowest predicted blood glucose value; The predicted minimum blood glucose value is compared with the lower limit of blood glucose safety to obtain a predicted safety indicator, which is either exceeded or not exceeded. After confirming that the predicted safety indicator is exceeded, the difference between the lower limit of blood glucose safety and the predicted lowest blood glucose value is calculated to obtain the predicted exceedance range. The ratio of the predicted exceedance range to the lower limit of blood glucose safety is confirmed as the predicted hypoglycemia risk component; otherwise, zero is confirmed as the predicted hypoglycemia risk component. The ECG safety index is input into a preset ECG risk mapping function to obtain the ECG abnormality risk component; The hypoglycemia risk index is obtained by weighting and fusing the absolute risk component of blood glucose, the risk component of decreased rate of blood glucose, the risk component of predicted hypoglycemia, and the risk component of electrocardiographic abnormality. The calculation formula is shown below: in, Indicates the risk index of hypoglycemia. This indicates the absolute risk component of blood glucose. This indicates the risk component of the rate of decline. This indicates the component representing the predicted risk of hypoglycemia. Indicates the risk component of electrocardiogram abnormalities. , , , All of these represent preset weight values.

[0043] It should be understood that the risk threshold node is a threshold dataset composed of several blood glucose safety limits, pre-stored in the control processing unit, serving as a reference value for hypoglycemia risk calculation. It is configured by medical staff on the bedside treatment terminal according to the clinical safety procedures for hypoglycemia induction treatment. The lower blood glucose safety limit is the lowest blood glucose value allowed during treatment; falling below this limit is considered entering a severe hypoglycemia zone. The upper limit of blood glucose decline rate is the fastest blood glucose decline rate allowed during treatment; exceeding this limit is considered an excessively rapid decline in blood glucose. The blood glucose safety indicator is a flag indicating whether the current blood glucose value is below the lower blood glucose safety limit. The blood glucose deviation margin is the non-negative difference obtained by subtracting the current blood glucose value from the lower blood glucose safety limit when the current blood glucose value is below it. The absolute blood glucose risk component is a risk value reflecting the absolute degree of the current blood glucose value's low position. The rate direction indicator is a flag indicating whether the blood glucose change rate is negative. The blood glucose decline magnitude is a non-directional value obtained by taking the absolute value of the blood glucose change rate when it is negative. The rate of decline risk component is a risk value used to quantify the extent to which the rate of blood glucose decline exceeds the upper limit of the rate of blood glucose decline. The predicted minimum blood glucose value is the smallest predicted blood glucose value in the set of predicted blood glucose values, representing the lowest possible blood glucose value within the time period covered by the future prediction time point set. The predicted safety indicator is a flag used to indicate whether the predicted minimum blood glucose value has already fallen below the lower limit of blood glucose safety. The predicted exceedance range is the non-negative difference obtained by subtracting the predicted minimum blood glucose value from the lower limit of blood glucose safety when the predicted minimum blood glucose value is lower than the lower limit of blood glucose safety. The predicted hypoglycemia risk component is a risk value used to quantify the degree to which the predicted minimum blood glucose value may fall below the lower limit of blood glucose safety in the future time period. The ECG risk mapping function is a piecewise linear function that proportionally maps the ECG safety index to the ECG abnormality risk component. The ECG abnormality risk component is the ECG-side risk value calculated by the ECG risk mapping function and participating in the weighted fusion of the hypoglycemia risk index. The weighted fusion is an operation in which the absolute risk component of blood glucose, the risk component of decreased rate of blood glucose, the predicted risk component of hypoglycemia, and the risk component of electrocardiographic abnormality are multiplied by their respective weight values, summed, and then divided by the sum of all weight values. The weight values ​​are coefficients that control the influence of each risk component on the hypoglycemia risk index, and are configured by medical staff in the bedside treatment terminal based on clinical experience in hypoglycemia induction treatment. In this embodiment, the four weight values ​​are equal.

[0044] For example, in the risk threshold node, the lower limit of blood glucose safety is set to 2.5 mmol / L, and the upper limit of blood glucose decrease rate is set to 0.1 mmol / L / min. The current blood glucose value is 4.8 mmol / L. Comparing the current blood glucose value with the lower limit of blood glucose safety of 2.5 mmol / L, the blood glucose safety indicator is found to be within the limit; therefore, 0 is identified as the absolute blood glucose risk component. The blood glucose change rate is -0.05 mmol / L / min. Comparing the blood glucose change rate with 0, the rate direction indicator is decreased. Taking the absolute value of the blood glucose change rate yields a blood glucose decrease magnitude of 0.05 mmol / L / min. Dividing the blood glucose decrease magnitude by the upper limit of the blood glucose decrease rate yields the rate decrease risk component: 0.05 divided by 0.1 equals 0.5. The predicted blood glucose values ​​are 4.2 mmol / L, 3.9 mmol / L, and 3.6 mmol / L. The minimum predicted blood glucose value is extracted from this set, resulting in a predicted minimum blood glucose value of 3.6 mmol / L. Comparing this predicted minimum blood glucose value with the lower limit of blood glucose safety (2.5 mmol / L), the predicted safety indicator is "not exceeded," therefore 0 is confirmed as the predicted hypoglycemia risk component. The ECG safety index is 0.17, and the ECG risk mapping function is an equivalent mapping function; therefore, the ECG abnormality risk component is 0.17. The four preset weight values ​​are 1, 1, 1, and 1, respectively. The hypoglycemia risk index is calculated using the formula: 1 x 0 + 1 x 0.5 + 1 x 0 + 1 x 0.17 = 0.67. Dividing this by the sum of the four weight values ​​(4) yields a hypoglycemia risk index of 0.1675, approximately 0.17.

[0045] Furthermore, after obtaining the hypoglycemia risk index, the hypoglycemia risk index, the target hypoglycemia range, and the individualized sensitivity parameters need to be used together to calculate and convert it into an insulin infusion command that can be executed by the infusion control unit. The generation process should simultaneously consider three factors: range deviation, risk magnitude, and prediction constraints. Therefore, obtaining the insulin infusion command based on the hypoglycemia risk index, the target hypoglycemia range, and the individualized sensitivity parameters includes: The upper and lower bounds of the target hypoglycemia range are obtained; Compare the current blood glucose level with the target upper and lower bounds; If the current blood glucose level is higher than the target upper limit, the difference between the current blood glucose level and the target upper limit is calculated to obtain the blood glucose deviation. The blood glucose deviation is then divided by the individualized sensitivity parameter to obtain the basic infusion volume. If the current blood glucose level is between the target lower bound and the target upper bound, the preset maintenance infusion volume will be confirmed as the basic infusion volume. If the current blood glucose level is below the target lower limit, then zero will be considered as the basic infusion volume; Calculate the difference between the predicted minimum blood glucose value and the preset prediction safety lower limit, and input the difference into the preset prediction constraint mapping function to obtain the prediction constraint coefficient, wherein the prediction safety lower limit is not lower than the blood glucose safety lower limit; The hypoglycemia risk index is input into a preset risk correction mapping function to obtain the risk correction coefficient; The target infusion rate is obtained by multiplying the baseline infusion volume, the risk correction coefficient, and the prediction constraint coefficient, and an insulin infusion command is generated based on the target infusion rate.

[0046] Specifically, the target upper bound is the boundary with the larger value within the target hypoglycemia range, representing the higher boundary that allows the blood glucose level of the subject to be regulated to be maintained. The target lower bound is the boundary with the smaller value within the target hypoglycemia range, representing the lower boundary that allows the blood glucose level of the subject to be regulated to be maintained. The blood glucose deviation is the positive difference obtained by subtracting the target upper bound from the current blood glucose value when the current blood glucose value is higher than the target upper bound, reflecting the extent to which the current blood glucose level needs to be further reduced from the target hypoglycemia range. The basal infusion volume is the initial infusion volume value obtained only based on the current blood glucose deviation and individualized sensitivity parameters, without considering risk correction and predictive constraints. The maintenance infusion volume is the preset small-dose infusion value used to maintain blood glucose within the target hypoglycemia range when the current blood glucose value is already within the target hypoglycemia range, set by medical staff in the bedside treatment terminal based on the subject's previous maintenance treatment experience. The predicted safety lower limit is a lower limit value used to constrain future blood glucose trends. It is set by medical staff in the bedside treatment terminal according to the clinical safety procedures for hypoglycemia induction therapy, and is not lower than the blood glucose safety lower limit, so that the future blood glucose trend is constrained in advance before reaching the blood glucose safety lower limit. The predicted constraint mapping function is a piecewise linear function that maps the difference between the predicted minimum blood glucose value and the predicted safety lower limit to a predicted constraint coefficient. The output is 0 when the difference is not greater than 0, linearly when the difference is greater than 0 and not greater than a preset safety margin, and 1 when the difference is greater than the safety margin. The safety margin is set by medical staff in the bedside treatment terminal according to the clinical safety procedures for hypoglycemia induction therapy. The predicted constraint coefficient is a coefficient calculated by the predicted constraint mapping function to reduce the risk of future hypoglycemia. The risk correction mapping function is a piecewise linear function that maps the hypoglycemia risk index to a risk correction coefficient. In this embodiment, the risk correction mapping function is in the form of 1 minus the hypoglycemia risk index. The risk correction coefficient is a coefficient calculated by the risk correction mapping function to actively reduce the infusion volume when the risk increases. The target infusion rate is the product of the baseline infusion volume, the risk correction factor, and the prediction constraint factor, reflecting the insulin infusion rate ultimately determined within the current control cycle. The insulin infusion command is a control command generated by the infusion control unit, with the target infusion rate as the main field, along with the current control cycle number and a timestamp.

[0047] For example, the target hypoglycemia range is set to 2.8 mmol / L to 3.3 mmol / L, from which the upper limit is 3.3 mmol / L and the lower limit is 2.8 mmol / L. The current blood glucose level is 4.8 mmol / L. Comparing the current blood glucose level with the upper limit of 3.3 mmol / L and the lower limit of 2.8 mmol / L, the current blood glucose level is higher than the upper limit. Therefore, the blood glucose deviation is calculated by subtracting the upper limit from the current blood glucose level: 4.8 mmol / L - 3.3 mmol / L = 1.5 mmol / L. Dividing this deviation by the individualized sensitivity parameter of 1.5 mmol / L / U yields a baseline infusion rate of 1.0 U / cycle. The predicted safety lower limit is set to 2.8 mmol / L, and the safety margin is preset to 1.0 mmol / L. The difference between the predicted lowest blood glucose value of 3.6 mmol / L and the predicted safety lower limit of 2.8 mmol / L is 0.8 mmol / L. This difference is input into the prediction constraint mapping function. Since the difference is greater than 0 and not greater than the safety margin, 0.8 is output linearly as the prediction constraint coefficient. The hypoglycemia risk index is 0.1675. The risk correction mapping function takes the form of 1 minus the hypoglycemia risk index; therefore, the risk correction coefficient is 1 minus 0.1675, which equals 0.8325. Multiplying the basal infusion rate, risk correction factor, and prediction constraint factor yields a target infusion rate of 1.0 x 0.8325 x 0.8, which equals 0.666 U / cycle. In this embodiment, following the rule of retaining two decimal places, we take 0.67 U / cycle. Insulin infusion instructions are generated based on the target infusion rate, and the current control cycle number and current timestamp are marked. This embodiment of the invention, by first obtaining the basal infusion rate based on interval deviation and individualized sensitivity parameters, and then adjusting the basal infusion rate using the risk correction factor and prediction constraint factor, actively reduces the risk of future hypoglycemia while driving blood glucose towards the target hypoglycemic range, thus achieving a balance between therapeutic efficacy and safety control.

[0048] S5. Using the safety protection unit, the insulin infusion command is safely verified based on the hypoglycemia risk index to obtain the verification result, which is either pass or fail. After confirming that the verification result is pass, the insulin infusion command is executed using the infusion control unit to obtain the actual infusion volume and the actual blood glucose response.

[0049] It should be explained that the safety verification is a secondary safety check process performed by a safety protection unit independent of the control processing unit, before the insulin infusion command is actually issued to the infusion control unit. The verification result is the result output by the safety protection unit after performing the safety verification, indicating whether the insulin infusion command is allowed to be issued, either passing or failing. The actual infusion volume is the total amount of insulin actually injected in this control cycle, reported by the flow feedback sensor built into the infusion control unit after the insulin infusion command is executed. The actual blood glucose response is a response curve composed of several continuous blood glucose measurement data reported by the continuous blood glucose monitoring unit at the end of the insulin lag window and subsequent predicted time points after the completion of the insulin infusion in this control cycle.

[0050] Understandably, the insulin infusion command generated by the control processing unit may still pose a risk of exacerbating hypoglycemia due to an increased risk of hypoglycemia or a lower predicted minimum blood glucose value. Therefore, a secondary verification by an independent safety protection unit is required. Thus, the safety verification of the insulin infusion command based on the hypoglycemia risk index, and the resulting verification results, include: By using a preset risk level threshold and the hypoglycemia risk index for classification and matching, a risk level identifier is obtained, wherein the risk level identifier is a safety level identifier or a danger level identifier. Compare the predicted minimum blood glucose value with the predicted safety lower limit to obtain a predicted limit violation indicator, wherein the predicted limit violation indicator is either predicted limit violation or predicted no limit violation; After confirming that the risk level identifier is a safety level identifier and the predicted limit exceedance identifier is a predicted no limit exceedance, the insulin infusion instruction is confirmed to have passed the safety verification, and the verification result is confirmed as passed; Otherwise, the insulin infusion command is confirmed to have failed the security check, and the check result is confirmed as unsuccessful.

[0051] It should be understood that the risk level threshold is a tiered threshold pre-stored in the safety protection unit, used to map the hypoglycemia risk index to a risk level identifier. This threshold is configured by medical personnel in the bedside treatment terminal according to the clinical safety procedures for hypoglycemia induction therapy. The tiered matching compares the hypoglycemia risk index with the risk level threshold. A safety level identifier is output when the hypoglycemia risk index is less than the risk level threshold, and a danger level identifier is output when it is not less than the risk level threshold. The risk level identifier is a flag bit output by the tiered matching, representing the overall hypoglycemia risk level within the current control period, and its value is either a safety level identifier or a danger level identifier. The predicted limit exceedance identifier is a flag bit used to indicate whether the predicted minimum blood glucose value is lower than the predicted safe lower limit, and its value is either predicted limit exceedance or predicted no limit exceedance. When the risk level identifier is a safety level identifier and the predicted limit exceedance identifier is predicted no limit exceedance, the safety protection unit allows the insulin infusion command and confirms the verification result as passed. Otherwise, the safety protection unit prevents the insulin infusion command from being issued, confirms the verification result as failed, and simultaneously triggers the infusion control unit to perform a deceleration, pause, or pump stop alarm action.

[0052] For example, the risk level threshold is set to 0.5. The hypoglycemia risk index for this control cycle is 0.17. Comparing the hypoglycemia risk index with the risk level threshold of 0.5, the hypoglycemia risk index is less than the risk level threshold, therefore the risk level identifier is confirmed as a safety level identifier. Assuming the predicted minimum blood glucose value is 3.6 mmol / L and the predicted lower safety limit is 2.8 mmol / L, comparing the predicted minimum blood glucose value with the predicted lower safety limit, the predicted minimum blood glucose value is not lower than the predicted lower safety limit, therefore the predicted exceedance identifier is confirmed as predicted not exceedance. Based on the combined judgment that the risk level identifier is a safety level identifier and the predicted exceedance identifier is predicted not exceedance, the insulin infusion command is confirmed to have passed the safety verification, and the verification result is passed. The safety protection unit releases the insulin infusion command to the infusion control unit, which drives the micro-insulin infusion pump to perform infusion at the target infusion rate of 0.67 U / cycle. At the end of this control cycle, the flow feedback sensor reports the actual infusion volume as 0.67 U. If the insulin hysteresis window is set to 15 minutes, the continuous glucose monitoring unit will report continuous glucose measurement data of 4.7 mmol / L, 4.5 mmol / L, 4.44 mmol / L, 4.45 mmol / L, and 4.4 mmol / L at 5 minutes, 10 minutes, 15 minutes, 20 minutes, and 30 minutes after the current time, respectively. The actual glucose response curve is then constructed from these continuous glucose measurement data. This embodiment of the invention introduces a safety protection unit independent of the control processing unit to perform two checks on the hypoglycemia risk index and the predicted minimum glucose value. This ensures that even if the control processing unit generates an excessively large insulin infusion command due to model bias, it cannot directly drive the insulin infusion pump, thus improving the safety of hypoglycemia induction therapy.

[0053] S6. Based on the actual infusion volume and actual blood glucose response, update the individualized sensitivity parameters and blood glucose prediction model, and return to the step of obtaining continuous blood glucose sequence and electrocardiogram monitoring data according to the preset control cycle to realize the closed-loop control of hypoglycemia induction in the target object.

[0054] It should be explained that updating the individualized sensitivity parameter is a process of reversing the individualized sensitivity parameter used in the next control cycle based on the actual decrease in blood glucose caused by the actual insulin injection within the current control cycle. Updating the blood glucose prediction model is a process of making small, online adjustments to the internal parameters of the blood glucose prediction model based on the deviation between the predicted blood glucose value and the actual measured blood glucose value within the current control cycle. Returning to the step of obtaining continuous blood glucose sequences and ECG monitoring data according to the preset control cycle means that after updating the parameters and model for the current control cycle, returning to step S2, and continuing into the next control cycle with the updated individualized sensitivity parameter and the updated blood glucose prediction model, thus forming a complete closed-loop control process for hypoglycemia induction.

[0055] Understandably, due to differences in insulin response among different subjects and the lag in insulin action, the individualized sensitivity parameters input at the start of treatment may not be accurate and need to be corrected using the actual response after each control cycle. Therefore, the updating of individualized sensitivity parameters and blood glucose prediction models based on actual infusion volume and actual blood glucose response includes: Based on a preset insulin action hysteresis window, the hysteresis sampling value is extracted from the actual blood glucose response; Calculate the difference between the current blood glucose value and the lagged sample value to obtain the blood glucose response difference; When the actual infusion volume is greater than the preset minimum effective infusion threshold, the ratio of the blood glucose response difference to the actual infusion volume is calculated to obtain the instantaneous sensitivity. The updated individualized sensitivity parameters are calculated based on the instantaneous sensitivity and individualized sensitivity parameters, using the following formula: in, This indicates the updated individual sensitivity parameter. This represents an individualized sensitivity parameter. Indicates the difference in blood glucose response. Indicates the actual infusion volume. Indicates instantaneous sensitivity. Indicates the preset reference sensitivity. This represents the minimum effective infusion threshold. This represents the preset smoothing coefficient, and ; The blood glucose prediction model is updated based on the actual infusion volume and the actual blood glucose response, resulting in the updated blood glucose prediction model.

[0056] It should be understood that the insulin action hysteresis window is a parameter reflecting the time required for insulin to stabilize blood glucose levels after subcutaneous injection, and is configured by healthcare professionals in the bedside treatment terminal based on the insulin action hysteresis characteristics. The hysteresis sampling value is the blood glucose measurement value at the end of the insulin action hysteresis window from the actual blood glucose response curve. The blood glucose response difference is the difference obtained by subtracting the hysteresis sampling value from the current blood glucose value, reflecting the magnitude of blood glucose decrease caused by insulin infusion within the current control cycle. The minimum effective infusion threshold is the lower limit for determining whether the actual infusion volume in the current control cycle is sufficient to induce a credible blood glucose response, and is configured by healthcare professionals in the bedside treatment terminal based on the minimum resolvable dose of the insulin infusion pump. The instantaneous sensitivity is the ratio of the blood glucose response difference to the actual infusion volume within the current control cycle, reflecting only the local sensitivity of a single infusion within the current control cycle. The reference sensitivity is a backup sensitivity value pre-input by healthcare professionals in the bedside treatment terminal based on the basic information of the subject to be regulated, used when the actual infusion volume in the current control cycle is too small to reliably estimate the instantaneous sensitivity. The smoothing coefficient is used to smooth the weighting of historical individualized sensitivity parameters and instantaneous sensitivity in the current control cycle. It is configured by medical staff in the bedside treatment terminal based on the stability requirements of closed-loop control. The updated individualized sensitivity parameter is the latest sensitivity parameter obtained by weighting historical individualized sensitivity parameters and instantaneous sensitivity in the current control cycle according to the formula. It will be used as the individualized sensitivity parameter in step S3 of the next control cycle to obtain the predicted blood glucose value set and in step S4 to calculate the basal infusion volume, thus participating in the control calculation of the next control cycle.

[0057] For example, the insulin hysteresis window is set to 15 minutes, the minimum effective infusion threshold is set to 0.05 U, the reference sensitivity is set to 1.5 mmol / L / U, and the smoothing coefficient is set to 0.7. The actual infusion volume within this control period is 0.67 U, which is greater than the minimum effective infusion threshold of 0.05 U. The blood glucose measurement value of 4.44 mmol / L at the end of the insulin hysteresis window is extracted from the actual blood glucose response curve as the hysteresis sampling value. The difference between the current blood glucose value of 4.8 mmol / L and the hysteresis sampling value of 4.44 mmol / L is calculated to obtain a blood glucose response difference of 0.36 mmol / L. Since the actual infusion volume is greater than the minimum effective infusion threshold, the instantaneous sensitivity calculated using the upper branch of the formula is approximately 0.537 mmol / L / U (0.36 divided by 0.67). Multiply the historical individualized sensitivity parameter of 1.5 by the smoothing factor of 0.7 to get 1.05. Multiply the instantaneous sensitivity without retaining decimal places by 1 and subtract the difference of 0.3 from the smoothing factor of 0.7 to get approximately 0.161. Add the two results to get the updated individualized sensitivity parameter of approximately 1.211 mmol / L / U. Following the rule of retaining two decimal places, we take 1.21 mmol / L / U as the individualized sensitivity parameter used in subsequent control periods.

[0058] Furthermore, simply updating the individualized sensitivity parameter is insufficient to correct the estimation bias of the blood glucose prediction model for future blood glucose levels. It is necessary to utilize the difference between the predicted blood glucose value and the actual blood glucose measurement value at the corresponding time point within the current control period to perform small-step online adjustments to the internal parameters of the blood glucose prediction model. Therefore, the step of updating the blood glucose prediction model based on the actual infusion volume and the actual blood glucose response to obtain the updated blood glucose prediction model includes: At each prediction time point in the prediction time point set, the actual blood glucose response is sampled to obtain multiple actual blood glucose sampling values, wherein each actual blood glucose sampling value corresponds one-to-one with a prediction time point. For each of the multiple actual blood glucose samples, perform the following operation: Using the actual blood glucose sampling value, the corresponding reference blood glucose value is retrieved from the predicted blood glucose value set, wherein the prediction time point corresponding to the reference blood glucose value is the same as the prediction time point corresponding to the actual blood glucose sampling value. Calculate the difference between the actual blood glucose sample value and the reference blood glucose value to obtain the prediction residual value; Summarize the predicted residual values ​​to obtain the predicted residual value set; Based on the predicted residual set, actual infusion volume, and historical infusion data, the model parameters of the blood glucose prediction model are updated online incrementally to obtain the updated blood glucose prediction model.

[0059] In detail, the sampling of the actual blood glucose response is based on the current blood glucose value collection time of the current control cycle as the reference time. The blood glucose measurement value is read from the actual blood glucose response curve according to the offset of each predicted time point relative to the reference time in the predicted time point set. The actual blood glucose sample value is the true blood glucose measurement value corresponding to a single predicted time point obtained by sampling the actual blood glucose response. The reference blood glucose value is a predicted blood glucose value retrieved from the predicted blood glucose value set according to a completely identical correspondence at the predicted time point, and corresponds one-to-one with the actual blood glucose sample value at the same predicted time point. The prediction residual value is the difference between the actual blood glucose sample value and the reference blood glucose value at the same predicted time point, reflecting the prediction deviation of the blood glucose prediction model at the predicted time point within the current control cycle. The prediction residual value set is a collection of prediction residual values ​​at all predicted time points. The online incremental update is an operation that uses the prediction residual value set, the actual infusion volume of the current control cycle, and the historical infusion data to construct a set of training samples, and performs an incremental update of the internal parameters of the blood glucose prediction model according to a preset small learning rate. Optionally, a stochastic gradient descent algorithm can be used to achieve this purpose. The updated blood glucose prediction model, obtained after a single online incremental update with corrected prediction bias, will be used as the blood glucose prediction model in step S3 of the next control cycle. This utilizes the prediction bias from the current control cycle to improve the prediction accuracy of the next cycle. After the update, the closed-loop control process jumps back to step S2, using the updated individualized sensitivity parameters and the updated blood glucose prediction model to continue acquiring continuous blood glucose sequences and ECG monitoring data according to the control cycle, entering the next round of closed-loop control.

[0060] For example, the three predicted time points in the predicted time point set are 10 minutes, 20 minutes, and 30 minutes backward from the current time. Under the premise that the insulin lag window is 15 minutes and the continuous glucose monitoring unit sampling interval is 1 minute within this control cycle, samples are taken from the actual glucose response curve with offsets of 10 minutes, 20 minutes, and 30 minutes backward from the current glucose value collection time, resulting in three actual glucose sampling values ​​of 4.5 mmol / L, 4.4 mmol / L, and 4.4 mmol / L respectively. The predicted glucose value set corresponding to this control cycle is 4.2 mmol / L, 3.9 mmol / L, and 3.6 mmol / L. The corresponding reference glucose values ​​retrieved at the same predicted time point are 4.2 mmol / L, 3.9 mmol / L, and 3.6 mmol / L respectively. The difference between the actual blood glucose sample value and the reference blood glucose value was calculated point by point, resulting in three predicted residual values ​​of 0.3 mmol / L, 0.55 mmol / L, and 0.8 mmol / L, respectively. These were then aggregated to form a set of predicted residual values ​​of 0.3 mmol / L, 0.55 mmol / L, and 0.8 mmol / L. Using the actual infusion volume of 0.67 U in the current control cycle and the three most recent infusion records (0.8, 0.7, and 0.6 U) from historical infusion data as input conditions for the training samples, and combining this with the predicted residual value set, a stochastic gradient descent iteration was performed at a preset learning rate of 0.001 to update the parameters of the Long Short-Term Memory network step-by-step, resulting in the updated blood glucose prediction model. After completing the update of the individualized sensitivity parameters and blood glucose prediction model for this control cycle, the closed-loop control system jumps back to step S2, using the updated individualized sensitivity parameters of 1.21 mmol / L / U and the updated blood glucose prediction model to continue acquiring continuous blood glucose sequences and ECG monitoring data for subsequent control cycles, entering subsequent closed-loop control until the bedside treatment terminal issues a treatment end command. This embodiment of the invention uses prediction deviations at multiple prediction time points within each control cycle to correct the blood glucose prediction model, and combines this with online small-step updates based on actual infusion volume and historical infusion data. This ensures that the blood glucose prediction model continuously approximates the actual blood glucose changes of the subject during treatment progression, forming a complete closed-loop learning process together with the update of individualized sensitivity parameters.

[0061] To address the problems described in the background art, this invention receives closed-loop control commands and identifies the object to be controlled and the closed-loop control system based on these commands. The closed-loop control system includes a continuous blood glucose monitoring unit, an electrocardiogram (ECG) monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. Using the continuous blood glucose monitoring unit and the ECG monitoring unit, continuous blood glucose sequences and ECG monitoring data are acquired according to a preset control cycle. The current blood glucose value and rate of change are obtained based on the continuous blood glucose sequence, and an ECG safety index is obtained based on the ECG monitoring data. This invention obtains the ECG safety index by first extracting core ECG features, then mapping and weighting them separately, enabling subsequent risk calculations to simultaneously reflect the QT interval and cardiac function using a single value. This invention assesses the electrocardiographic safety of the subject from three aspects: rate of blood glucose fluctuation, arrhythmia, and other factors. It acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters. Using a control processing unit and a pre-built blood glucose prediction model, a predicted blood glucose value set is obtained based on the current blood glucose value, rate of blood glucose change, electrocardiographic safety index, historical infusion data, and individualized sensitivity parameters. This invention, by concatenating multiple physiological and therapeutic state features into a fixed-format vector and simultaneously outputting multi-time-point predictions through a machine learning model, enables subsequent risk calculation and control decisions to adjust in advance based on future blood glucose trends. This compensates for the shortcomings of lagging control relying solely on the current blood glucose value. The invention obtains a hypoglycemia risk index based on the current blood glucose value, rate of blood glucose change, predicted blood glucose value set, and electrocardiographic safety index. Based on the hypoglycemia risk index, target hypoglycemia range, and individualized sensitivity parameters, this invention obtains insulin infusion instructions. It is evident that this invention achieves a balance between therapeutic efficacy and safety control by first determining the baseline infusion volume based on range deviation and individualized sensitivity parameters, and then adjusting the baseline infusion volume using risk correction coefficients and prediction constraint coefficients. This approach drives blood glucose levels towards the target hypoglycemia range while actively reducing the risk of future hypoglycemia. A safety protection unit performs a safety verification of the insulin infusion instructions based on the hypoglycemia risk index, obtaining a verification result (pass or fail). Upon confirming a pass result, the infusion control unit executes the insulin infusion instructions, obtaining the actual infusion volume and actual blood glucose response. Therefore, this invention... This invention introduces a safety protection unit independent of the control processing unit to perform two checks on the hypoglycemia risk index and the predicted minimum blood glucose value. This ensures that even if the control processing unit generates excessive insulin infusion commands due to model bias, it cannot directly drive the insulin infusion pump, thus improving the safety of hypoglycemia induction therapy. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, returning to the steps of obtaining continuous blood glucose sequences and ECG monitoring data according to the preset control cycle. This achieves closed-loop control of hypoglycemia induction for the target population. It is evident that this invention uses prediction deviations at multiple prediction time points within each control cycle to correct the blood glucose prediction model and performs online small-step updates in conjunction with the actual infusion volume and historical infusion data.This invention enables the blood glucose prediction model to continuously approximate the actual blood glucose changes of the target individual during treatment, forming a complete closed-loop learning process together with the updating of individualized sensitivity parameters. Therefore, this invention can achieve safe and precise control of hypoglycemia-induced hypoglycemia.

[0062] like Figure 2 The diagram shown is a functional block diagram of a hypoglycemia-inducing closed-loop regulation system based on machine learning, provided in an embodiment of the present invention.

[0063] The hypoglycemia-inducing closed-loop regulation system 100 based on machine learning described in this invention can be installed in an electronic device. Depending on the functions implemented, the hypoglycemia-inducing closed-loop regulation system 100 may include an object identification module 101, a monitoring and prediction module 102, a risk decision-making module 103, and a verification and update module 104. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0064] The object confirmation module 101 is used to receive closed-loop control instructions and confirm the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. The monitoring and prediction module 102 is used to acquire continuous blood glucose sequence and ECG monitoring data according to a preset control cycle using a continuous blood glucose monitoring unit and an ECG monitoring unit, acquire the current blood glucose value and blood glucose change rate based on the continuous blood glucose sequence, and acquire the ECG safety index based on the ECG monitoring data. The system acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and a pre-built blood glucose prediction model, it obtains a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters. The risk decision module 103 is used to obtain a hypoglycemia risk index based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set and electrocardiogram safety index, and to obtain an insulin infusion instruction based on the hypoglycemia risk index, target hypoglycemia range and individualized sensitivity parameters. The verification and update module 104 is used to perform a safety verification of the insulin infusion command based on the hypoglycemia risk index using the safety protection unit, and obtain a verification result, wherein the verification result is either pass or fail. After confirming that the verification result is pass, the insulin infusion command is executed using the infusion control unit to obtain the actual infusion volume and the actual blood glucose response. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, and the process returns to the step of obtaining continuous blood glucose sequences and electrocardiogram monitoring data according to the preset control cycle, thereby realizing closed-loop regulation of hypoglycemia induction in the target subject.

[0065] In detail, the modules in the machine learning-based hypoglycemia-inducing closed-loop regulation system 100 described in this embodiment of the invention employ the same methods as described above. Figure 1 The technique is the same as the machine learning-based hypoglycemia-induced closed-loop regulation method described in the article, and can produce the same technical effect, so it will not be repeated here.

[0066] like Figure 3 The diagram shown is a schematic representation of an electronic device that implements a machine learning-based hypoglycemia-induced closed-loop regulation method according to an embodiment of the present invention.

[0067] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a machine learning-based hypoglycemia-induced closed-loop regulation method program.

[0068] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as the portable hard drive of the electronic device 1. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 1. Furthermore, the memory 11 includes both internal storage units and external storage devices of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of a machine learning-based hypoglycemia-induced closed-loop regulation method, but also to temporarily store data that has been output or will be output.

[0069] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., a machine learning-based hypoglycemia-induced closed-loop regulation method program), and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0070] The bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0071] Figure 3 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0072] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0073] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices.

[0074] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0075] The program for a machine learning-based hypoglycemia-induced closed-loop regulation method stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When run in the processor 10, it can achieve the following: Receive closed-loop control instructions, and identify the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. Using a continuous blood glucose monitoring unit and an electrocardiogram (ECG) monitoring unit, continuous blood glucose sequences and ECG monitoring data are acquired according to a preset control cycle. The current blood glucose value and blood glucose change rate are obtained based on the continuous blood glucose sequence, and the ECG safety index is obtained based on the ECG monitoring data. The system acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and a pre-built blood glucose prediction model, it obtains a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters. The hypoglycemia risk index is obtained based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set and ECG safety index. Insulin infusion instructions are obtained based on the hypoglycemia risk index, target hypoglycemia range and individualized sensitivity parameters. Using a safety protection unit, the insulin infusion command is safety-verified based on the hypoglycemia risk index to obtain a verification result, which is either pass or fail. After confirming that the verification result is pass, the insulin infusion command is executed by the infusion control unit to obtain the actual infusion volume and the actual blood glucose response. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, and the process returns to the step of obtaining continuous blood glucose sequences and electrocardiogram monitoring data according to the preset control cycle, thereby realizing closed-loop regulation of hypoglycemia induction in the target subject.

[0076] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 3The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0077] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0078] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following: Receive closed-loop control instructions, and identify the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. Using a continuous blood glucose monitoring unit and an electrocardiogram (ECG) monitoring unit, continuous blood glucose sequences and ECG monitoring data are acquired according to a preset control cycle. The current blood glucose value and blood glucose change rate are obtained based on the continuous blood glucose sequence, and the ECG safety index is obtained based on the ECG monitoring data. The system acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and a pre-built blood glucose prediction model, it obtains a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters. The hypoglycemia risk index is obtained based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set and ECG safety index. Insulin infusion instructions are obtained based on the hypoglycemia risk index, target hypoglycemia range and individualized sensitivity parameters. Using a safety protection unit, the insulin infusion command is safety-verified based on the hypoglycemia risk index to obtain a verification result, which is either pass or fail. After confirming that the verification result is pass, the insulin infusion command is executed by the infusion control unit to obtain the actual infusion volume and the actual blood glucose response. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, and the process returns to the step of obtaining continuous blood glucose sequences and electrocardiogram monitoring data according to the preset control cycle, thereby realizing closed-loop regulation of hypoglycemia induction in the target subject.

[0079] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and actual implementations may have other classification methods.

[0080] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0082] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A machine learning-based closed-loop regulation method for hypoglycemia induction, characterized in that, The method includes: Receive closed-loop control instructions, and identify the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. Using a continuous blood glucose monitoring unit and an electrocardiogram (ECG) monitoring unit, continuous blood glucose sequences and ECG monitoring data are acquired according to a preset control cycle. The current blood glucose value and blood glucose change rate are obtained based on the continuous blood glucose sequence, and the ECG safety index is obtained based on the ECG monitoring data. The system acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and a pre-built blood glucose prediction model, it obtains a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters. The hypoglycemia risk index is obtained based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set and ECG safety index. Insulin infusion instructions are obtained based on the hypoglycemia risk index, target hypoglycemia range and individualized sensitivity parameters. Using a safety protection unit, the insulin infusion command is safety-verified based on the hypoglycemia risk index to obtain a verification result, which is either pass or fail. After confirming that the verification result is pass, the insulin infusion command is executed by the infusion control unit to obtain the actual infusion volume and the actual blood glucose response. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, and the process returns to the step of obtaining continuous blood glucose sequences and electrocardiogram monitoring data according to the preset control cycle, thereby realizing closed-loop regulation of hypoglycemia induction in the target subject.

2. The hypoglycemia-induced closed-loop regulation method based on machine learning as described in claim 1, characterized in that, The method of obtaining the current blood glucose value and the rate of change of blood glucose based on a continuous blood glucose sequence includes: The continuous blood glucose sequence is subjected to outlier removal and time alignment operations to obtain an effective blood glucose sequence, wherein the effective blood glucose sequence includes multiple blood glucose measurement values, and the latest blood glucose measurement value is extracted from the effective blood glucose sequence as the current blood glucose value; Using a preset rate analysis window, a rate analysis sequence is extracted from the effective blood glucose sequence. The rate analysis sequence is then subjected to least squares linear fitting to obtain the fitting slope, which is used as the rate of blood glucose change.

3. The hypoglycemia-induced closed-loop regulation method based on machine learning as described in claim 2, characterized in that, The method of obtaining the ECG safety index based on ECG monitoring data includes: Feature extraction is performed on the electrocardiogram (ECG) monitoring data to obtain ECG feature nodes, which include: corrected QT interval, heart rate fluctuation, and arrhythmia event identifiers. The abnormal QT component is obtained based on the corrected QT interval and the preset QT risk mapping function; the heart rate fluctuation component is obtained based on the heart rate fluctuation and the preset heart rate risk mapping function; and the arrhythmia component is obtained based on the arrhythmia event identifier and the preset arrhythmia mapping function. The QT abnormal component, heart rate fluctuation component, and arrhythmia component are weighted and fused to obtain the ECG safety index.

4. The hypoglycemia-induced closed-loop regulation method based on machine learning as described in claim 3, characterized in that, The method for obtaining a predicted blood glucose value set based on current blood glucose level, rate of blood glucose change, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters includes: The current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters are concatenated to obtain the input feature vector; Using a preset set of prediction time points, the input feature vector is input into the blood glucose prediction model to obtain a set of predicted blood glucose values. The set of predicted blood glucose values ​​includes multiple predicted blood glucose values, and the predicted blood glucose values ​​correspond one-to-one with the prediction time points in the set of prediction time points.

5. The hypoglycemia-induced closed-loop regulation method based on machine learning as described in claim 4, characterized in that, The method for obtaining a hypoglycemia risk index based on current blood glucose level, rate of blood glucose change, predicted blood glucose value set, and electrocardiogram safety index includes: Obtain risk threshold nodes, which include: the lower limit of safe blood glucose and the upper limit of blood glucose rate of decline; The current blood glucose value is compared with the lower limit of blood glucose safety to obtain a blood glucose safety indicator. The blood glucose safety indicator is either above or below the limit. If the blood glucose safety indicator is above the limit, the difference between the lower limit of blood glucose safety and the current blood glucose value is calculated to obtain the blood glucose deviation margin. The ratio of the blood glucose deviation margin to the lower limit of blood glucose safety is confirmed as the absolute blood glucose risk component. Otherwise, zero is confirmed as the absolute blood glucose risk component. The rate of change of blood glucose is compared with zero to obtain the rate direction indicator, which is either decreasing or not decreasing. After confirming that the rate direction indicator is decreasing, the absolute value of the rate of change of blood glucose is taken to obtain the magnitude of the decrease in blood glucose. The ratio of the magnitude of the decrease in blood glucose to the upper limit of the rate of decrease in blood glucose is confirmed as the rate decrease risk component. Otherwise, zero is confirmed as the rate decrease risk component. The minimum predicted blood glucose value is extracted from the set of predicted blood glucose values ​​to obtain the lowest predicted blood glucose value; The predicted minimum blood glucose value is compared with the lower limit of blood glucose safety to obtain a predicted safety indicator, which is either exceeded or not exceeded. After confirming that the predicted safety indicator is exceeded, the difference between the lower limit of blood glucose safety and the predicted lowest blood glucose value is calculated to obtain the predicted exceedance range. The ratio of the predicted exceedance range to the lower limit of blood glucose safety is confirmed as the predicted hypoglycemia risk component; otherwise, zero is confirmed as the predicted hypoglycemia risk component. The ECG safety index is input into a preset ECG risk mapping function to obtain the ECG abnormality risk component; The hypoglycemia risk index is obtained by weighting and fusing the absolute risk component of blood glucose, the risk component of decreased rate of blood glucose, the risk component of predicted hypoglycemia, and the risk component of electrocardiographic abnormality. The calculation formula is shown below: in, Indicates the risk index of hypoglycemia. This indicates the absolute risk component of blood glucose. This indicates the risk component of the rate of decline. This indicates the component representing the predicted risk of hypoglycemia. Indicates the risk component of electrocardiogram abnormalities. , , , All of these represent preset weight values.

6. The hypoglycemia-induced closed-loop regulation method based on machine learning as described in claim 5, characterized in that, The process of obtaining insulin infusion instructions based on the hypoglycemia risk index, target hypoglycemia range, and individualized sensitivity parameters includes: The upper and lower bounds of the target hypoglycemia range are obtained; Compare the current blood glucose level with the target upper and lower bounds; If the current blood glucose level is higher than the target upper limit, the difference between the current blood glucose level and the target upper limit is calculated to obtain the blood glucose deviation. The blood glucose deviation is then divided by the individualized sensitivity parameter to obtain the basic infusion volume. If the current blood glucose level is between the target lower bound and the target upper bound, the preset maintenance infusion volume will be confirmed as the basic infusion volume. If the current blood glucose level is below the target lower limit, then zero will be considered as the basic infusion volume; Calculate the difference between the predicted minimum blood glucose value and the preset prediction safety lower limit, and input the difference into the preset prediction constraint mapping function to obtain the prediction constraint coefficient, wherein the prediction safety lower limit is not lower than the blood glucose safety lower limit; The hypoglycemia risk index is input into a preset risk correction mapping function to obtain the risk correction coefficient; The target infusion rate is obtained by multiplying the baseline infusion volume, the risk correction coefficient, and the prediction constraint coefficient, and an insulin infusion command is generated based on the target infusion rate.

7. The hypoglycemia-induced closed-loop regulation method based on machine learning as described in claim 6, characterized in that, The safety verification of insulin infusion instructions based on the hypoglycemia risk index yields the following verification results: By using a preset risk level threshold and the hypoglycemia risk index for classification and matching, a risk level identifier is obtained, wherein the risk level identifier is a safety level identifier or a danger level identifier. Compare the predicted minimum blood glucose value with the predicted safety lower limit to obtain a predicted limit violation indicator, wherein the predicted limit violation indicator is either predicted limit violation or predicted no limit violation; After confirming that the risk level identifier is a safety level identifier and the predicted limit exceedance identifier is a predicted no limit exceedance, the insulin infusion instruction is confirmed to have passed the safety verification, and the verification result is confirmed as passed; Otherwise, the insulin infusion command is confirmed to have failed the security check, and the check result is confirmed as unsuccessful.

8. The hypoglycemia-induced closed-loop regulation method based on machine learning as described in claim 7, characterized in that, The method of updating individualized sensitivity parameters and blood glucose prediction models based on actual infusion volume and actual blood glucose response includes: Based on a preset insulin action hysteresis window, the hysteresis sampling value is extracted from the actual blood glucose response; Calculate the difference between the current blood glucose value and the lagged sample value to obtain the blood glucose response difference; When the actual infusion volume is greater than the preset minimum effective infusion threshold, the ratio of the blood glucose response difference to the actual infusion volume is calculated to obtain the instantaneous sensitivity. The updated individualized sensitivity parameters are calculated based on the instantaneous sensitivity and individualized sensitivity parameters, using the following formula: in, This indicates the updated individualized sensitivity parameter. This represents an individualized sensitivity parameter. Indicates the difference in blood glucose response. Indicates the actual infusion volume. Indicates instantaneous sensitivity. Indicates the preset reference sensitivity. This represents the minimum effective infusion threshold. This represents the preset smoothing coefficient, and ; The blood glucose prediction model is updated based on the actual infusion volume and the actual blood glucose response, resulting in the updated blood glucose prediction model.

9. The machine learning-based closed-loop regulation method for hypoglycemia induction as described in claim 8, characterized in that, The updated blood glucose prediction model, obtained by updating the blood glucose prediction model based on the actual infusion volume and actual blood glucose response, includes: At each prediction time point in the prediction time point set, the actual blood glucose response is sampled to obtain multiple actual blood glucose sampling values, wherein each actual blood glucose sampling value corresponds one-to-one with a prediction time point. For each of the multiple actual blood glucose samples, perform the following operation: Using the actual blood glucose sampling value, the corresponding reference blood glucose value is retrieved from the predicted blood glucose value set, wherein the prediction time point corresponding to the reference blood glucose value is the same as the prediction time point corresponding to the actual blood glucose sampling value. Calculate the difference between the actual blood glucose sample value and the reference blood glucose value to obtain the prediction residual value; Summarize the predicted residual values ​​to obtain the predicted residual value set; Based on the predicted residual set, actual infusion volume, and historical infusion data, the model parameters of the blood glucose prediction model are updated online incrementally to obtain the updated blood glucose prediction model.

10. A hypoglycemia-induced closed-loop regulation system based on machine learning, characterized in that, The system includes: The object confirmation module is used to receive closed-loop control instructions and confirm the object to be controlled and the closed-loop control system based on the closed-loop control instructions. The closed-loop control system includes: a continuous blood glucose monitoring unit, an electrocardiogram monitoring unit, a control processing unit, a safety protection unit, and an infusion control unit. The monitoring and prediction module is used to acquire continuous blood glucose sequence and ECG monitoring data according to a preset control cycle using the continuous blood glucose monitoring unit and ECG monitoring unit, to obtain the current blood glucose value and blood glucose change rate based on the continuous blood glucose sequence, and to obtain the ECG safety index based on the ECG monitoring data. The system acquires historical infusion data, target hypoglycemia range, and individualized sensitivity parameters of the target object. Using the control processing unit and a pre-built blood glucose prediction model, it obtains a set of predicted blood glucose values ​​based on the current blood glucose value, blood glucose change rate, electrocardiogram safety index, historical infusion data, and individualized sensitivity parameters. The risk decision module is used to obtain a hypoglycemia risk index based on the current blood glucose value, blood glucose change rate, predicted blood glucose value set and ECG safety index, and to obtain insulin infusion instructions based on the hypoglycemia risk index, target hypoglycemia range and individualized sensitivity parameters. The verification and update module is used to perform safety verification on insulin infusion commands based on the hypoglycemia risk index using the safety protection unit, and obtain the verification result, which is either pass or fail. After confirming that the verification result is pass, the infusion control unit executes the insulin infusion command to obtain the actual infusion volume and the actual blood glucose response. Based on the actual infusion volume and actual blood glucose response, the individualized sensitivity parameters and blood glucose prediction model are updated, and the process returns to the step of obtaining continuous blood glucose sequences and electrocardiogram monitoring data according to the preset control cycle, thereby realizing closed-loop regulation of hypoglycemia induction in the target subject.

Images