A blood glucose analysis method and edge computing device based on CGM dynamic graph

Abstract

The present invention provides a blood sugar analysis method and edge computing device based on CGM dynamic graphs, belonging to the technical field of intelligent medical devices. The present invention includes: using daily CGM graph data to predict the daily blood sugar fluctuation risk probability; using the daytime CGM dynamic graph data of the day and the nighttime CGM dynamic graph data of the previous day to predict the risk probability of hypoglycemia at night on the same day; for long-term CGM dynamic graph data, predict the long-term blood sugar fluctuation risk probability, and give the risk level of blood sugar control effect according to the long-term blood sugar fluctuation risk probability and daily analysis results. The present invention includes an edge computing device, which utilizes a blood sugar analysis method based on CGM dynamic graphs as described above. The present invention can effectively prevent the risk of nocturnal hypoglycemia, reflect the patient's blood sugar control situation to a certain extent, and provide risk warnings in a timely manner.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medical devices, and in particular to a blood glucose analysis method and an edge computing device based on CGM dynamic graphs. Background Art

[0002] In recent years, due to the influence of many factors such as living standards, dietary structure and sub-healthy lifestyle, the prevalence of diabetes has increased year by year. In order to effectively control blood sugar, blood sugar monitoring related technologies have gradually developed in the direction of intelligence and informationization. Therefore, the dynamic blood glucose monitoring (CGM) system has gradually attracted widespread attention from people in related fields. It can detect postprandial hyperglycemia, nocturnal hypoglycemia, dawn phenomenon and Somogyi phenomenon, which is of great significance in the clinical treatment of diabetes. The CGM system can generate a dynamic glucose profile (AGP), and can output blood sugar values, blood sugar graphs and blood sugar graph summaries (blood sugar fluctuations, average values, normal range time (TIR), time above normal range (TAR), time below normal range (TBR) and area under the blood sugar curve, etc.) at fixed sampling intervals every day; the index information provided by the profile can better adjust the treatment plan for diabetes and evaluate the control effect of blood sugar.

[0003] However, the utilization rate of the existing CGM system's blood glucose dynamic graph is low and the technical threshold is high. It is difficult for patients to interpret the graph report personally. After discharge, they can only control blood glucose from the alarm prompts of the APP corresponding to the CGM system; in addition, some patients have poor compliance and the CGM alarm is frequent, which makes the user alarm fatigue. Patients will customize the blood glucose alarm threshold settings to ignore the alarm information, resulting in a reduced application effect of the CGM system. Usually, daytime users will pay attention to their blood glucose changes, and may ignore the hidden risk of hypoglycemia at night, which poses a threat to the user's life and health. If the probability of hypoglycemia risk at night can be predicted based on the daytime and historical nighttime blood glucose of the day, and given to users as a reference, it will be more effective to prevent the risk of hypoglycemia at night. In addition, there are differences in blood glucose volatility over time, which can reflect the patient's blood glucose control to a certain extent. The current dynamic blood glucose graph fails to provide these differences. In addition, under the current conditions, the evaluation of blood glucose control effect and the risk prompts for hospitalization are not timely enough.

[0004] Chinese invention patent CN111110249B discloses a blood sugar fluctuation evaluation method and evaluation device, which is a blood sugar fluctuation calculation method based on gradient variation of CGM data. Although the proposed gradient variation parameter can simultaneously characterize the blood sugar fluctuation amplitude and blood sugar fluctuation frequency, some information in the CGM data may still be lost; while the graph analysis information of the present invention not only includes the blood sugar fluctuation amplitude and blood sugar fluctuation frequency, but also enables the information in the CGM data to be dynamically analyzed.

[0005] Chinese invention patent CN113948207A discloses a blood glucose data processing method for hypoglycemia warning, which uses a pre-trained support vector machine model to obtain a blood glucose prediction value corresponding to the real-time blood glucose data. Although there is a correlation between the real-time blood glucose data and time, which makes the output result of the warning model more accurate than relying on a single variable, it is only based on the real-time blood glucose data and time for warning, and does not distinguish between night and day alarms. The present invention not only warns based on each CGM graph data, but also provides a separate warning for hypoglycemia at night. Summary of the invention

[0006] In order to solve the technical problems existing in the prior art, the present invention provides a blood glucose analysis method based on CGM dynamic graphs: using daily CGM graph data to predict the daily blood glucose fluctuation risk probability; using the daytime CGM dynamic graph data of the day and the nighttime CGM dynamic graph data of the previous day to predict the risk probability of hypoglycemia at night on the same day; for the long-term CGM dynamic graph data of the first set time period, predict the long-term blood glucose fluctuation risk probability, and according to the long-term blood glucose fluctuation risk probability and the daily analysis results of the second set time period, give the risk level of the blood glucose control effect of the second set time period. The present invention provides an edge computing device, which utilizes a blood glucose analysis method based on CGM dynamic graphs as described above. The present invention can more effectively prevent the risk of nocturnal hypoglycemia, reflect the patient's blood glucose control situation to a certain extent, and can provide risk warnings in a timely manner.

[0007] The present invention provides a blood glucose analysis method based on CGM dynamic graph, comprising the following steps:

[0008] Convert user information and CGM dynamic graph data into structured data through data preprocessing;

[0009] Get atlas information;

[0010] Use daily CGM graph data and combine it with graph information to predict the risk probability of daily blood sugar fluctuations;

[0011] Utilize the CGM dynamic graph data of the day and the CGM dynamic graph data of the previous night, and combine the graph information to predict the risk probability of hypoglycemia at night on the same day;

[0012] Convert daily CGM dynamic graph data and daily analysis results into structured data;

[0013] According to the long-term CGM dynamic graph data of the first set time period, the probability of long-term blood sugar fluctuation risk is predicted in combination with the graph information;

[0014] According to the long-term blood sugar fluctuation risk probability and the daily analysis results of the second set time period, the risk level of the blood sugar control effect of the second set time period is given.

[0015] Preferably, the daily analysis results include the daily blood sugar fluctuation risk probability and the overnight hypoglycemia fluctuation risk probability.

[0016] Preferably, the sampling time of daily CGM dynamic graph data is 24 hours, and the sampling interval is 5 minutes.

[0017] Preferably, the first set time period is not less than 14 days, the second set time period is not less than 7 days, the daytime of the current day is from 6:00 to 24:00, the night of the previous day is from 0:00 to 6:00, and the last three hours of the daytime of the current day are from 21:00 to 24:00.

[0018] Preferably, the graph information includes graph analysis information and graph categories; the graph analysis information includes: the maximum value Max of the time period graph glucose, the minimum value Min of the time period graph glucose, the slope Slope of the time period graph, the coefficient of variation GV of the time period graph glucose, the standard deviation SD of the time period graph, the variance Var of the time period graph, the average value Mean of the time period graph glucose, the average blood glucose fluctuation amplitude MAGE of the time period graph and the average blood glucose absolute difference MODD of the time period graph; the graph categories include daytime graphs, nighttime graphs, all-time graphs and optional time period graphs.

[0019] Preferably, for the full-time profile, the profile analysis information includes the time percentages corresponding to the time when all glucose concentration segments appear in a day.

[0020] Preferably, the glucose concentration segments are divided into segments according to the number of milligrams of glucose concentration contained in each deciliter.

[0021] Preferably, the prediction of daily blood sugar fluctuation risk probability is completed by a daily blood sugar fluctuation evaluation model; the daily blood sugar fluctuation evaluation model obtains the daily blood sugar fluctuation risk probability according to the input data x, where x is as follows:

[0022] x=[SD,Mean,GV,MAGE,MODD].

[0023] Preferably, the daily blood sugar volatility evaluation model is as follows:

[0024]

[0025] Among them, p represents the daily blood sugar fluctuation risk probability, y represents the manually calibrated risk result, the values ​​of y of 1 and 0 represent high risk and low risk of blood sugar fluctuation, respectively, F(x) includes the random forest classification model and the logistic regression classification model, and ∈ represents the error term.

[0026] Preferably, the prediction of the risk probability of hypoglycemia at night on the same day is completed by a hypoglycemia risk warning model at night on the same day; the hypoglycemia risk warning model at night on the same day obtains the risk probability of hypoglycemia at night on the same day according to the input data X, and X is as follows:

[0027] X=[Gpremin 夜间 ,Gmin 日间 ,G_max_re,Gmin -3h ,Gmean -3h ,Gslope -3h ];

[0028] Among them, G_max_re represents the maximum value of the glucose value of the daytime dynamic spectrum G_max 日间 The maximum value of the glucose value in the previous day's dynamic graph G_Pre_max 日间 The ratio of Gpremin 夜间 Indicates the minimum value of the glucose value in the dynamic graph at night on the previous day, Gmin 日间 Indicates the minimum value of the glucose value in the daytime dynamic graph, and Gmin -3h 、Gmean -3h and Gslope -3h They respectively represent the minimum, average and slope of the glucose values ​​in the first three hours of the daytime dynamic spectrum.

[0029] Preferably, the risk warning model for hypoglycemia at night is as follows:

[0030]

[0031] Among them, p 低 represents the probability of hypoglycemia at night, S represents the hypoglycemia result at night of the manually calibrated data, the values ​​of S 1 and 0 represent hypoglycemia and no hypoglycemia at night, respectively, F(X) includes random forest classification model, Logistic regression classification model, XGboost classification model and LightGBM classification model, ∈ represents the error term.

[0032] Preferably, the prediction of the risk probability of long-term blood sugar fluctuation is completed by a long-term blood sugar fluctuation evaluation model; the long-term blood sugar fluctuation evaluation model is based on X 长 , obtain the risk probability of long-term blood sugar fluctuation, X 长 as follows:

[0033] X 长 =[Clu_index mean ,SD mean ,MEAN mean ,GV mean,MAGE mean ,MODD mean ];

[0034] Among them, Clu_index mean , SD mean 、MEAN mean , GV mean 、MAGE mean and MODD mean They are respectively the average values ​​of the clustering category indexes Clu_index, SD, MEAN, GV, MAGE and MODD of the daily CGM dynamic graph data within the set period.

[0035] Preferably, the cluster category index Clu_index of the daily CGM dynamic graph data is obtained using a hierarchical clustering model.

[0036] Preferably, the long-term blood sugar volatility evaluation model is as follows:

[0037]

[0038] Among them, p 长 is the risk probability of long-term blood sugar fluctuation, F(X 长 ) includes random forest classification model and logistic regression classification model, ∈ represents the error term.

[0039] Preferably, the risk level of the blood sugar control effect in the second set time period is obtained using a decision tree model.

[0040] The present invention provides an edge computing device based on CGM dynamic graph analysis, which uses the blood glucose analysis method based on CGM dynamic graph as described in any of the previous items to perform blood glucose analysis. The edge computing device includes an MCU controller and a data communication module, and the MCU controller is connected to the data communication module; the data communication module transmits data between the edge computing device and the terminal and the cloud data processing center; the MCU controller performs data preprocessing on the CGM dynamic graph data received from the terminal, converts it into structured data, and performs analysis; wherein the MCU controller includes a blood glucose volatility evaluation unit, a nighttime hypoglycemia risk warning unit, and a blood glucose control periodicity evaluation unit;

[0041] The blood sugar volatility assessment unit predicts the risk probability of daily blood sugar fluctuations and the risk probability of long-term blood sugar fluctuations;

[0042] The nighttime hypoglycemia risk warning unit predicts the probability of nighttime hypoglycemia risk on the same day;

[0043] The blood sugar control periodic evaluation unit evaluates the risk level of blood sugar control effect.

[0044] Preferably, the cloud data processing center stores the graph information.

[0045] Preferably, the risk probability of hypoglycemia at night on the same day is transmitted to the terminal by the data communication module; the risk level of blood sugar control effect is transmitted to the terminal and the cloud data processing center by the data communication module, and the cloud data processing center determines whether the terminal user needs to be hospitalized.

[0046] Preferably, the MCU controller also includes a slice priority evaluation unit, which classifies users into 5G network slices with different network speeds according to the risk level of blood sugar control effect, wherein users with worse blood sugar control effect are assigned to network slices with higher network speed priority in the corresponding 5G network slice.

[0047] Preferably, the MCU controller also includes a CGM graph data structuring processing unit to convert the CGM dynamic graph data into structured data.

[0048] Preferably, the MCU controller also includes an expandable unit for accessing new units.

[0049] Preferably, the data communication module includes a 5G module and a communication interface unit. The 5G module transmits data between the edge computing device and the cloud data processing center and the terminal supporting 5G communication. The communication interface unit transmits data between the edge computing device and the cloud data processing center and the terminal supporting non-5G communication.

[0050] Preferably, the edge computing device also includes a memory, a power supply, an indicator light, a buzzer and a touch screen connected to the MCU controller; the memory is used to store local data; the power supply is used to power the edge computing device; the indicator light is used to display the operating status of the equipment; the buzzer is used to alarm for equipment abnormalities; the touch screen is used to display and input parameter settings of the edge computing device, and to view the power status, space capacity, memory consumption and process status information of the edge computing device.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] 1. The present invention provides a blood glucose analysis method and edge computing device based on CGM dynamic graph, which uses the CGM dynamic graph data of the day and the CGM dynamic graph data of the previous night to predict the risk probability of hypoglycemia at night on the current day, and sends the risk probability of hypoglycemia at night on the current day to the terminal, so as to avoid the risk of threatening the user's life and health because the user may ignore the hidden risk of hypoglycemia at night. By predicting the risk probability of hypoglycemia at night for the user as a reference, the risk of hypoglycemia at night can be more effectively prevented.

[0053] 2. The present invention provides a blood glucose analysis method and edge computing device based on CGM dynamic graphs, which utilize daily CGM graph data to predict the daily blood glucose fluctuation risk probability. The difference in blood glucose volatility over time can reflect the patient's blood glucose control to a certain extent.

[0054] 3. The blood glucose analysis method and edge computing device based on CGM dynamic graph of the present invention predict the risk probability of long-term blood glucose fluctuation for long-term CGM dynamic graph data greater than a set time period, and give the risk level of blood glucose control effect, so that users can obtain blood glucose control effect evaluation and receive risk warnings for hospitalization in a timely manner.

[0055] 4. The blood glucose analysis method and edge computing device based on CGM dynamic graph of the present invention support 5G network communication and provide more favorable transmission conditions for CGM dynamic graph information collection.

[0056] 5. The present invention provides a blood glucose analysis method and edge computing device based on CGM dynamic graphs, which calculates the data in the edge computing device and then transmits it to the cloud data processing center, thereby reducing the pressure on the cloud and speeding up the data processing speed and the efficiency of risk prompts; improving the utilization rate of CGM dynamic graph data, effectively mining the potential value of graph data, effectively identifying blood glucose risks, improving user experience and thus effectively controlling blood glucose, and improving the actual application effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flow chart of a blood glucose analysis method based on CGM dynamic graph according to an embodiment of the present invention;

[0058] Figure 2 A flow chart of a blood glucose analysis system integrating a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device according to an embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of the connection structure of an edge computing device based on CGM dynamic graph analysis, a terminal supporting 5G communication, a continuous glucose sensor module and a cloud data processing center according to an embodiment of the present invention;

[0060] Figure 4 This is a schematic diagram of the structure of an edge computing device based on CGM dynamic graph analysis according to an embodiment of the present invention;

[0061] Figure 5 This is a schematic diagram of the structure of each module in an MCU controller of an edge computing device based on CGM dynamic graph analysis according to an embodiment of the present invention.

[0062] In the figure, 1. MCU controller; 2. Memory; 3. Data communication module; 31. Communication interface unit; 32. 5G module; 4. Power supply; 5. Indicator light; 6. Buzzer; 7. Touch screen; 11. CGM graph data structured processing unit; 12. Blood glucose volatility evaluation unit; 13. Nocturnal hypoglycemia risk warning unit; 14. Blood glucose control periodicity evaluation unit; 15. Slice priority evaluation unit; 16. Expandable unit. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] Example 1

[0065] Combination Figure 1 As shown, according to a specific embodiment of the present invention, a blood glucose analysis method based on CGM dynamic graph of the present invention is described in detail below.

[0066] The present invention provides a blood glucose analysis method based on CGM dynamic graph, comprising the following steps:

[0067] Convert user information and CGM dynamic graph data into structured data through data preprocessing;

[0068] Acquire graph information, which includes graph analysis information and graph categories; the graph analysis information includes: the maximum value Max of the time period graph glucose, the minimum value Min of the time period graph glucose, the slope Slope of the time period graph, the coefficient of variation GV of the time period graph glucose, the standard deviation SD of the time period graph, the variance Var of the time period graph, the average value Mean of the time period graph glucose, the average blood sugar fluctuation amplitude MAGE of the time period graph, and the average blood sugar absolute difference MODD of the time period graph; the graph categories include daytime graphs, nighttime graphs, all-time graphs, and optional time period graphs;

[0069] Use daily CGM graph data and combine it with graph information to predict the risk probability of daily blood sugar fluctuations;

[0070] Utilize the CGM dynamic graph data of the day and the CGM dynamic graph data of the previous night, and combine the graph information to predict the risk probability of hypoglycemia at night on the same day;

[0071] Convert daily CGM dynamic graph data and daily analysis results into structured data;

[0072] According to the long-term CGM dynamic graph data of the first set time period, the probability of long-term blood sugar fluctuation risk is predicted in combination with the graph information;

[0073] According to the long-term blood sugar fluctuation risk probability and the daily analysis results of the second set time period, the risk level of the blood sugar control effect of the second set time period is given.

[0074] Example 2

[0075] Combination Figure 2-3 As shown, according to a specific embodiment of the present invention, a system of the present invention that integrates a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device is described in detail below.

[0076] The present invention provides a system integrating a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device. Before performing blood glucose analysis, the following steps need to be performed:

[0077] S1: The user opens the blood glucose monitoring APP on the terminal to register, enters basic information and sets whether to allow access to the edge computing device;

[0078] Among them, basic information includes user information and blood glucose sensor information; basic information includes name, age, height, weight, type of diabetes and recent fasting blood glucose value; sensor information includes sensor production date, ID number, daytime blood glucose map collection time and nighttime blood glucose map collection time. Blood glucose sensor information can be obtained by scanning the QR code through the APP.

[0079] S2: If access is allowed, the user information, blood glucose sensor information, and daily CGM dynamic graph data are transmitted to the edge computing device;

[0080] Among them, the daily graph data contains 24h, 5min sampling interval, and a total of 288 data points of sensor output blood glucose values;

[0081] The edge computing device requests the transmission of CGM dynamic graph data in two batches, the first for daytime blood sugar collection and the second for nighttime blood sugar collection.

[0082] The default collection time range for daytime blood sugar is 6:00-24:00 every day. Users can set the daytime blood sugar collection time range according to their daily routine, such as 5:30-23:30;

[0083] The default collection time for nighttime blood glucose is from 24:00 on the current day to 6:00 on the next day. The collection time for nighttime blood glucose is calculated based on the set daytime blood glucose collection time range. After the nighttime blood glucose collection is completed, the customized 24h CGM graph data transmission is allowed to be completed.

[0084] Example 3

[0085] Combination Figure 2-3 As shown, according to a specific embodiment of the present invention, a system of the present invention that integrates a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device is described in detail below.

[0086] The present invention provides a system integrating a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device, and when performing blood glucose analysis, the system includes the following steps:

[0087] S3: MCU controller 1 converts user information and CGM map data into structured data and stores it in the cloud data processing center.

[0088] Among them, a user information table, a sensor information table, a blood glucose value record table, a graph analysis record table and a user blood glucose control risk assessment table are established in the cloud data processing center, as shown in Table 1, Table 2, Table 3, Table 4 and Table 5 below.

[0089] As shown in Table 1, the fields of the user information table include user ID, user information, network access category, network priority, creation time, modification time and extensible fields, and the primary key is user ID;

[0090] The network access category includes non-5G communication protocols and 5G communication protocols, which are represented by integers 1 and 2 respectively;

[0091] The network priority includes four slices with different network speed priority levels, which are represented by integers 0, 1, 2, and 3 respectively. 0 represents the highest priority network slice.

[0092] Table 1 User information table

[0093]

[0094] As shown in Table 2, the fields of the sensor information table include sensor ID, sensor attributes, user ID, sensor status and extensible fields. The primary key is the sensor ID, and the foreign key is the user ID. The user information table can be associated with the user ID.

[0095] Table 2 Sensor information table

[0096]

[0097]

[0098] As shown in Table 3, the fields of the blood glucose value record table include record number, blood glucose value, time, sensor ID and expandable fields. The primary key is the record number, and the foreign key is the sensor ID. The sensor information table can be associated with the sensor ID.

[0099] Table 3 Blood glucose value record

[0100]

[0101] As shown in Table 4, the fields of the spectrum analysis record table include: record number, spectrum analysis information, time, spectrum category, sensor ID and expandable fields. The primary key is the record number, and the foreign key is the sensor ID. The sensor information table can be associated with the sensor ID.

[0102] The graph analysis information includes: the maximum value Max of the time period graph glucose, the minimum value Min of the time period graph glucose, the slope Slope of the time period graph, the coefficient of variation GV of the time period graph glucose, the standard deviation SD of the time period graph, the variance Var of the time period graph, the average value Mean of the time period graph glucose, the average blood sugar fluctuation amplitude MAGE of the time period graph and the average blood sugar absolute difference MODD of the time period graph.

[0103] For the full-segment profile, the profile analysis information also includes the time percentage corresponding to the time of day when all glucose concentration segments occur: TAR1 (181-250 mg / dL), TAR2 (>250 mg / dL), TIR (70-180 mg / dL), TBR1 (54-69 mg / dL) and TBR2 (<54 mg / dL).

[0104] The types of maps include: daytime maps, nighttime maps, all-time maps and optional time-period maps (maps selected according to a certain time period).

[0105] Table 4 Spectrum analysis record

[0106]

[0107] As shown in Table 5, the fields of the user blood glucose control risk assessment table include record number, assessment type, assessment time, assessment result, remarks, user ID and expandable fields. The primary key is the record number, and the foreign key is the user ID. The user information table can be associated with the user ID. The types and results of various types of user blood glucose assessments will be written into the user blood glucose control risk assessment table.

[0108] Table 5 User blood sugar risk assessment table

[0109]

[0110] Example 4

[0111] Combination Figure 2-3 As shown, according to a specific embodiment of the present invention, a system of the present invention that integrates a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device is described in detail below.

[0112] The present invention provides a system integrating a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device, and when performing blood glucose analysis, the system includes the following steps:

[0113] S4: Evaluate daily blood glucose fluctuation based on daily CGM dynamic graph data;

[0114] For daily CGM dynamic graph data, standard deviation SD, segment glucose mean Mean, segment glucose variation coefficient GV, average blood glucose fluctuation amplitude MAGE and average blood glucose absolute difference MODD are used as data feature inputs, and the results of manual calibration are high risk as 1 and low risk as 0 as target outputs. The daily blood glucose fluctuation evaluation model is established as follows:

[0115]

[0116] Among them, p represents the risk probability of daily blood sugar fluctuation, F(x) includes random forest classification model and logistic regression classification model, ∈ represents the error term, x represents the feature data of daily CGM dynamic graph input, x = [SD, MEAN, GV, MAGE, MODD].

[0117] Furthermore, the random forest classification model and the logistic regression classification model are implemented through the RandomForestClassifier module and the LogisticRegression module in the sklearn library in python, and the model training is carried out using manually calibrated data.

[0118] Furthermore, the daily blood glucose volatility evaluation model has been trained in the cloud data processing center and transmitted to the blood glucose volatility evaluation unit 14 of the edge computing device for calling, and outputs the daily blood glucose fluctuation risk probability p.

[0119] Furthermore, the segmental glucose coefficient of variation GV formula is as follows:

[0120]

[0121] Furthermore, the formula for average blood sugar fluctuation range MAGE is as follows:

[0122]

[0123] Among them, C max Indicates the peak glucose value at a fluctuation point of the blood glucose curve, C min Indicates the minimum value of a fluctuation corresponding to the trough. An effective fluctuation refers to (C max -C min)>1SD, n represents the number of effective fluctuations within 24 hours.

[0124] Furthermore, the mean absolute difference in blood glucose MODD formula is as follows:

[0125]

[0126] Among them, C i represents the glucose observation value at time i on the day, C i-24h represents the glucose observation value at time i on the previous day, and k is the number of observations.

[0127] Furthermore, when the user replaces the blood glucose sensor, the mean blood glucose absolute difference MODD is calculated using the data closest to the day of replacement.

[0128] Example 5

[0129] Combination Figure 2-3 As shown, according to a specific embodiment of the present invention, a system of the present invention that integrates a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device is described in detail below.

[0130] The present invention provides a system integrating a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device, and when performing blood glucose analysis, the system includes the following steps:

[0131] S5: Use the nighttime graph data of the previous day and the daytime graph data of the current day to predict the risk of hypoglycemia at night. If the probability of hypoglycemia risk exceeds the set threshold, the risk warning will be transmitted to the terminal.

[0132] Among them, the input data feature extracts the minimum value of the glucose value G_Pre_min of the dynamic graph of the previous night (default 00:00-6:00) 夜间 , the maximum value of glucose value G_max in the daytime dynamic spectrum (6:00-24:00) 日间 The maximum value of the glucose value in the previous day's dynamic graph G_Pre_max 日间 The ratio G_max_re of the daytime spectrum is the minimum value Gmin of the spectrum in the first three hours of the daytime spectrum (default 21:00-24:00) -3h 、mean value Gmean -3h and segment slope Gslope -3h .

[0133] Furthermore, the formula of G_max_re is as follows:

[0134]

[0135] Furthermore, if the user customizes the time range of the nighttime segment and the daytime segment according to his living habits, the above data features can be summarized as the following input feature data X:

[0136] X=[Gpremin 夜间 ,Gmin 日间 ,G_max_re,Gmin -3h ,Gmean -3h ,Gslope -3h ].

[0137] Furthermore, the characteristic data X is input as input, and the result of the manual calibration data is 1 for hypoglycemia at night and 0 for no hypoglycemia as the target output, and a risk warning model for hypoglycemia at night is established, which is expressed as follows:

[0138]

[0139] Among them, p 低 represents the probability of hypoglycemia at night on the same day. F(X) includes random forest classification model, logistic regression classification model, XGboost classification model and LightGBM classification model. ∈ represents the error term.

[0140] Furthermore, the XGboost classification model and LightGBM regression classification model are implemented through the xgboost module and lightgbm module in Python, and the model training is carried out using manually calibrated data.

[0141] Furthermore, the nighttime hypoglycemia warning model has been trained in the cloud data processing center and transmitted to the nighttime hypoglycemia risk warning unit 13 of the edge computing device for invocation, and the nighttime hypoglycemia risk probability p of the day is output. 低 .

[0142] Example 6

[0143] Combination Figure 2-3 As shown, according to a specific embodiment of the present invention, a system of the present invention that integrates a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device is described in detail below.

[0144] The present invention provides a system integrating a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device, and when performing blood glucose analysis, the system includes the following steps:

[0145] S6: Organize the daily CGM graph data and daily analysis results into structured data and save them to the cloud data processing center. The daily analysis results include the daily blood sugar fluctuation risk probability p and the nighttime hypoglycemia probability p of the same day. 低 .

[0146] Furthermore, the full-segment graph data and graph analysis information of each user on the day, the daily blood sugar fluctuation risk probability p, the daily hypoglycemia risk probability p 低 , sorted into json string format and transmitted to the cloud data processing center.

[0147] S7: for the daily CGM dynamic graph data of the first set time period (not less than 14 days), provide a long-term blood sugar volatility evaluation;

[0148] Furthermore, TAR1, TAR2, TIR, TBR1 and TBR2 of the daily CGM dynamic graph data of the first set time period (not less than 14 days) are extracted as feature input data, and the clustering category index Clu_index of the daily CGM dynamic graph data is obtained using a hierarchical clustering model.

[0149] Furthermore, in the cloud data processing center, the hierarchical clustering model uses the characteristic data of TAR1, TAR2, TIR, TBR1 and TBR2 of many patients every day as the input of the hierarchical clustering model, and divides the TIR ratio from high to low into 35 categories as Clu_index. The larger the Clu_index, the more obvious the fluctuation of blood sugar.

[0150] Furthermore, the hierarchical clustering model is trained by the cloud data processing center and saved to the blood glucose volatility evaluation unit of the edge computing device for call.

[0151] Further, according to the hierarchical clustering model, the Clu_index of each day within the first set time period (not less than 14 days) is obtained, and the Clu_index within the first set time period (not less than 14 days) is averaged to obtain the Clu_index mean .

[0152] Furthermore, the SD, MEAN, GV, MAGE and MODD obtained every day within the first set time period (not less than 14 days) are averaged to obtain SD mean 、MEAN mean , GV mean 、MAGE mean and MODD mean .

[0153] Further, the data is organized into input feature data X 长 as follows:

[0154] X 长 =[Clu_index mean ,SD mean ,MEAN mean ,GV mean,MAGE mean ,MODD mean ];

[0155] Furthermore, using the input feature data X 长 , the long-term blood glucose fluctuations manually calibrated as high risk is 1, and low risk is 0 as the target output, and the following long-term blood glucose fluctuation evaluation model is established as follows:

[0156]

[0157] Among them, p 长 is the probability of long-term blood sugar fluctuation, F(X 长 ) includes random forest classification model and logistic regression classification model.

[0158] Example 7

[0159] Combination Figure 2-3 As shown, according to a specific embodiment of the present invention, a system of the present invention that integrates a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device is described in detail below.

[0160] The present invention provides a system integrating a blood glucose analysis method based on CGM dynamic graph analysis and an edge computing device, and when performing blood glucose analysis, the system includes the following steps:

[0161] S8: Based on the long-term blood sugar fluctuation probability and the daily analysis results of the second set time period (not less than 7 days), the blood sugar control risk level of the second set time period is given; the daily analysis results include the daily blood sugar fluctuation risk probability and the overnight hypoglycemia fluctuation risk probability.

[0162] Furthermore, according to p 长 , within the second set time period p 低 The average value p 低_ean , the average glucose value GV in the second set time period mean Glycated hemoglobin (G HBA1C ) and the calculated GMI index as the characteristic data column X input as follows:

[0163] X input =[p 长 ,p 低_ean ,G HBA1C ,GMI];

[0164] Furthermore, G HBA1C The calculation formula is as follows:

[0165] G HBA1C (%) = (GV mean( / L)+2.59) / 1.59;

[0166] Going further, the calculation formula of GMI is as follows:

[0167] GMI(%)=3.31+0.02392*GV mean ( / dL);

[0168] Furthermore, the decision tree model is used to give the blood sugar control risk level, which is divided into four levels: low risk, medium risk, medium-high risk and high risk, corresponding to the output of the four categories of 0, 1, 2 and 3 of the decision tree model.

[0169] Furthermore, the decision tree model is implemented by DecisionTreeClassifier in the sklearn library in python.

[0170] Furthermore, the blood sugar control risk assessment model is trained by the cloud data processing center and saved to the blood sugar control periodic assessment unit 14 of the edge computing device for call.

[0171] S9: For users with medium-high risk and high risk levels, the edge computing device sends the risk information to the terminal and transmits the risk information to the cloud data processing center.

[0172] S10: The cloud data processing center receives the risk warning and further determines whether hospitalization is required. If so, the user is notified to go to the hospital in time.

[0173] S11: If the terminal supports the 5G communication protocol, the user is classified into the corresponding 5G slice network according to the blood sugar control risk level in the second set time period (not less than 7 days).

[0174] Furthermore, the slice priority evaluation unit 15 will establish a connection between the blood glucose control risk level of a second set time period (not less than 7 days) and the corresponding 5G network slice, extract the blood glucose control risk level from the user blood glucose risk evaluation table, and modify the network priority field value in the user information table as an identifier of the network slicing division.

[0175] Example 8

[0176] Combination Figure 3-5 As shown, according to a specific embodiment of the present invention, an edge computing device based on CGM dynamic graph analysis of the present invention is described in detail below.

[0177] The present invention provides an edge computing device based on CGM dynamic graph analysis, which performs blood glucose analysis using a blood glucose analysis method based on CGM dynamic graph as described in any of the previous embodiments; the device comprises: an MCU controller 1, a memory 2, a data communication module 3, a power supply 4, an indicator light 5, a buzzer 6 and a touch screen 7;

[0178] The MCU controller controls the operation between the memory, power supply, indicator light, buzzer, touch screen and data communication module, and pre-processes the CGM dynamic graph data transmitted from the terminal into structured data and analysis.

[0179] Memory 2 stores local data, including CGM dynamic graph data, algorithm model and parameter configuration file sent from the terminal.

[0180] The data communication module 3 is used for data transmission between the edge computing device and the terminal and the cloud data processing center.

[0181] The power supply 4 is used to power the edge computing device.

[0182] Indicator light 5 is used to display the operating status of the equipment.

[0183] Buzzer 6 is used to warn of abnormal equipment.

[0184] The touch screen 7 is used to display and input parameter settings of the edge computing device, and to view the power status, space capacity, memory consumption and process status information of the edge computing device.

[0185] Example 9

[0186] Combination Figure 3-5 As shown, according to a specific embodiment of the present invention, an edge computing device based on CGM dynamic graph analysis of the present invention is described in detail below.

[0187] The present invention provides an edge computing device based on CGM dynamic graph analysis, the edge computing device performs blood glucose analysis using the blood glucose analysis method based on CGM dynamic graph as described in any of the previous embodiments; the edge computing device includes an MCU controller 1; the MCU controller 1 includes: a CGM graph data structuring processing unit 11, a blood glucose volatility evaluation unit 12, a nocturnal hypoglycemia risk warning unit 13, a blood glucose control periodicity evaluation unit 14, a slice priority evaluation unit 15, an expandable unit 16 and a database.

[0188] The CGM graph data structured processing unit 11 converts the CGM dynamic graph data into structured data.

[0189] The blood glucose volatility evaluation unit 12 predicts the daily blood glucose fluctuation risk probability and the long-term blood glucose fluctuation risk probability.

[0190] The nighttime hypoglycemia risk warning unit 13 predicts the risk probability of nighttime hypoglycemia on the current day.

[0191] The blood sugar control periodicity evaluation unit 14 evaluates the risk level of the blood sugar control effect in the second set time period (not less than 7 days).

[0192] The slice priority evaluation unit 15 classifies users into 5G network slices with different network speeds according to the risk level of blood sugar control effect, wherein users with worse blood sugar control effect are allocated to network slices with higher network speed priority in the corresponding 5G network slice.

[0193] The expandable unit 16 can be used to connect new units.

[0194] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A blood glucose analysis method based on CGM dynamic graph, characterized in that: The following steps are involved: Convert user information and CGM dynamic graph data into structured data through data preprocessing; Get atlas information; Using daily CGM dynamic graph data and combining graph information: time period graph standard deviation SD, time period graph glucose average Mean, time period graph glucose variation coefficient GV, time period graph average blood sugar fluctuation amplitude MAGE, time period graph average blood sugar absolute difference MODD, predict daily blood sugar fluctuation risk probability, which is obtained using daily blood sugar fluctuation evaluation model; Using the CGM dynamic graph data of the first three hours of the day and the CGM dynamic graph data of the previous night, combined with the graph information: the minimum value of the glucose value of the dynamic graph of the previous night Gpremin 夜间 、The minimum value of the glucose value of the daytime dynamic graph Gmin 日间 、The maximum value of the glucose value of the day's dynamic spectrum G_max 日间 The maximum value of the glucose value in the previous day's dynamic graph G_Pre_max 日间 The ratio G_max_re of the daytime dynamic spectrum is the minimum value of the glucose value in the first three hours of the day Gmin -3h , the average value of glucose value in the first three hours of the daytime dynamic spectrum Gmean -3h The slope of the glucose value in the last three hours of the daytime dynamic graph Gslope -3h , predict the risk probability of hypoglycemia at night on the same day, and the risk probability of hypoglycemia at night on the same day is obtained using the hypoglycemia risk warning model at night on the same day; Convert daily CGM dynamic graph data and daily analysis results into structured data; daily analysis results include daily blood sugar fluctuation risk probability and nighttime hypoglycemia fluctuation risk probability; According to the long-term CGM dynamic graph data of the first set time period, and combined with the graph information: the cluster category index Clu_index of each day in the first set time period, the standard deviation SD of the time period graph, the average glucose value MEAN of the time period graph, the glucose coefficient of variation GV of the time period graph, the average blood glucose fluctuation amplitude MAGE of the time period graph, and the average blood glucose absolute difference MODD of the time period graph are averaged, and the long-term blood glucose fluctuation risk probability is predicted based on the average value. The long-term blood glucose fluctuation risk probability is obtained using the long-term blood glucose fluctuation evaluation model, wherein the glucose concentration of the daily CGM dynamic graph data in the first set time period is greater than or equal to 181 mg / dl and less than or equal to 250 mg / dl in one day The percentage of time TAR1 corresponding to the time when the glucose concentration is greater than 250 mg / dl in a day, the percentage of time TAR2 corresponding to the time when the glucose concentration is greater than or equal to 70 mg / dl and less than or equal to 180 mg / dl in a day, the percentage of time TIR corresponding to the time when the glucose concentration is greater than or equal to 54 mg / dl and less than or equal to 69 mg / dl in a day, and the percentage of time TBR1 corresponding to the time when the glucose concentration is less than 54 mg / dl in a day are used as feature data input, and the hierarchical clustering model is used to calculate the clustering category index Clu_index for each day in the first set time period; According to the long-term blood sugar fluctuation risk probability and the average value of the daily hypoglycemia risk probability in the second set time period, the glycosylated hemoglobin value converted from the average glucose value in the second set time period, and the calculated GMI index, the risk level of the blood sugar control effect in the second set time period is given, and the risk level of the blood sugar control effect in the second set time period is obtained by using a decision tree model; The daily blood glucose volatility evaluation model, the same-day night hypoglycemia risk warning model, and the long-term blood glucose volatility evaluation model include the random forest classification model, the logistic regression classification model, the XGboost classification model, and the LightGBM classification model.

2. A blood glucose analysis method based on CGM dynamic graph according to claim 1, characterized in that: The sampling time of the daily CGM dynamic graph data is 24 hours, and the sampling interval is 5 minutes.

3. A blood glucose analysis method based on CGM dynamic graph according to claim 1, characterized in that: The first set time period is not less than 14 days, the second set time period is not less than 7 days, the daytime of the current day is from 6:00 to 24:00, the night of the previous day is from 0:00 to 6:00, and the last three hours of the daytime of the current day are from 21:00 to 24:

00.

4. A blood glucose analysis method based on CGM dynamic graph according to claim 3, characterized in that: The graph information includes graph analysis information and graph categories; the graph analysis information includes: the maximum value Max of the time period graph glucose, the minimum value Min of the time period graph glucose, the slope Slope of the time period graph, the coefficient of variation GV of the time period graph glucose, the standard deviation SD of the time period graph, the variance Var of the time period graph, the average value Mean of the time period graph glucose, the average blood glucose fluctuation amplitude MAGE of the time period graph and the average blood glucose absolute difference MODD of the time period graph; the graph categories include daytime graphs, nighttime graphs, all-time graphs and optional time period graphs.

5. A blood glucose analysis method based on CGM dynamic graph according to claim 4, characterized in that: For the full-time graph, the graph analysis information also includes the time percentage corresponding to the time when all glucose concentration segments appear in a day.

6. A blood glucose analysis method based on CGM dynamic graph according to claim 5, characterized in that: The glucose concentration sections are divided into sections according to the amount of glucose concentration in milligrams per deciliter.

7. A blood glucose analysis method based on CGM dynamic graph according to claim 4, characterized in that: The prediction of daily blood sugar fluctuation risk probability is completed by a daily blood sugar fluctuation evaluation model; the daily blood sugar fluctuation evaluation model obtains the daily blood sugar fluctuation risk probability according to the input data x, where x is as follows: x=[SD,Mean,GV,MAGE,MODD].

8. A blood glucose analysis method based on CGM dynamic graph according to claim 7, characterized in that: The daily blood sugar volatility evaluation model is shown in the following formula: Among them, p represents the daily blood sugar fluctuation risk probability, y represents the manually calibrated risk result, the values ​​of y of 1 and 0 represent high risk and low risk of blood sugar fluctuation, respectively, F(x) includes the random forest classification model and the logistic regression classification model, and ∈ represents the error term.

9. The blood glucose analysis method based on CGM dynamic graph according to claim 5, characterized in that: The prediction of the risk probability of hypoglycemia at night on the same day is completed by the hypoglycemia risk warning model at night on the same day; the hypoglycemia risk warning model at night on the same day obtains the risk probability of hypoglycemia at night on the same day according to the input data X, and X is as follows: X=[Gpremin 夜间 ,Municipalities 日间 ,G_max_re,Gmin -3h ,Gmean -3h ,Gslope -3h ]; Among them, G_max_re represents the maximum value of the glucose value of the daytime dynamic spectrum G_max 日间 The maximum value of the glucose value in the previous day's dynamic graph G_Pre_max 日间 The ratio of Gpremin 夜间 Indicates the minimum value of the glucose value in the dynamic graph at night on the previous day, Gmin 日间 Indicates the minimum value of the glucose value in the daytime dynamic graph, and Gmin -3h 、Gmean -3h and Gslope -3h They respectively represent the minimum, average and slope of the glucose values ​​in the first three hours of the daytime dynamic spectrum.

10. A blood glucose analysis method based on CGM dynamic graph according to claim 9, characterized in that: The hypoglycemia risk warning model for the night of the same day is as follows: Among them, p 低 represents the probability of hypoglycemia at night, S represents the hypoglycemia result at night of the manually calibrated data, the values ​​of S 1 and 0 represent hypoglycemia and no hypoglycemia at night, respectively, F(X) includes random forest classification model, Logistic regression classification model, XGboost classification model and LightGBM classification model, ∈ represents the error term.

11. The blood glucose analysis method based on CGM dynamic graph according to claim 5, characterized in that: The prediction of the risk probability of long-term blood sugar fluctuation is completed by a long-term blood sugar fluctuation evaluation model; the long-term blood sugar fluctuation evaluation model is based on X 长 , obtain the risk probability of long-term blood sugar fluctuation, X 长 as follows: X 长 =[Clu_index mean , SD mean , MEAN mean ,GV mean ,MAGE mean , MODE mean ]; Among them, Clu_index mean , SD mean 、MEAN mean , GV mean 、MAGE mean and MODD mean They are respectively the average values ​​of the clustering category indexes Clu_index, SD, MEAN, GV, MAGE and MODD of the daily CGM dynamic graph data within the set period.

12. A blood glucose analysis method based on CGM dynamic graph according to claim 11, characterized in that: The cluster category index Clu_index of the daily CGM dynamic graph data is obtained using a hierarchical clustering model.

13. A blood glucose analysis method based on CGM dynamic graph according to claim 11, characterized in that: The long-term blood sugar volatility evaluation model is shown in the following formula: Among them, p 长 is the risk probability of long-term blood sugar fluctuation, F(X 长 ) includes random forest classification model and logistic regression classification model, ∈ represents the error term.

14. An edge computing device based on CGM dynamic graph analysis, characterized in that: Blood glucose analysis is performed using the blood glucose analysis method based on CGM dynamic graph according to any one of claims 1 to 13, wherein the edge computing device includes an MCU controller and a data communication module, and the MCU controller is connected to the data communication module; the data communication module transmits data between the edge computing device and the terminal and the cloud data processing center; the MCU controller performs data preprocessing on the CGM dynamic graph data received from the terminal, converts it into structured data, and performs analysis; wherein the MCU controller includes a blood glucose volatility evaluation unit, a nighttime hypoglycemia risk warning unit, and a blood glucose control periodicity evaluation unit; The blood sugar volatility assessment unit predicts the daily blood sugar volatility risk probability and the long-term blood sugar volatility risk probability; The nighttime hypoglycemia risk warning unit predicts the risk probability of nighttime hypoglycemia on the same day; The blood sugar control periodic evaluation unit evaluates the risk level of the blood sugar control effect.

15. The edge computing device based on CGM dynamic graph analysis according to claim 14, characterized in that: The cloud data processing center stores the graph information.

16. The edge computing device based on CGM dynamic graph analysis according to claim 14, characterized in that: The risk probability of hypoglycemia at night on the same day is transmitted to the terminal by the data communication module; the risk level of the blood sugar control effect is transmitted to the terminal and the cloud data processing center by the data communication module, and the cloud data processing center determines whether the user of the terminal needs to be hospitalized.

17. The edge computing device based on CGM dynamic graph analysis according to claim 14, characterized in that: The MCU controller also includes a slice priority evaluation unit, which classifies users into 5G network slices with different network speeds according to the risk level of the blood sugar control effect, wherein users with worse blood sugar control effects are allocated to network slices with higher network speed priority levels in the corresponding 5G network slices.

18. The edge computing device based on CGM dynamic graph analysis according to claim 14, characterized in that: The MCU controller also includes a CGM dynamic graph data structuring processing unit, which converts the CGM dynamic graph data into structured data.

19. The edge computing device based on CGM dynamic graph analysis according to claim 14, characterized in that: The MCU controller also includes an expandable unit for accessing new units.

20. The edge computing device based on CGM dynamic graph analysis according to claim 14, characterized in that: The data communication module includes a 5G module and a communication interface unit. The 5G module transmits data between the edge computing device and the cloud data processing center and the terminal supporting 5G communication. The communication interface unit transmits data between the edge computing device and the cloud data processing center and the terminal supporting non-5G communication.

Images