AGP intelligent interpretation and insulin adjustment method based on expert system

Abstract

The present invention belongs to the technical field of blood sugar control, and relates to an AGP intelligent interpretation and insulin adjustment method based on an expert system, the method comprising: establishing a knowledge base in a reasoning mechanism; constructing an interpretation and decision support expert system based on the knowledge base using a simplified expert system structure; constructing a patient problem analysis tree of three dimensions of hypoglycemia, blood sugar fluctuation and hyperglycemia; using expert AGP reading graphs and empirical data to expand each rule; constructing a basal insulin dosage adjustment rule and a meal insulin dosage adjustment rule based on an interval type II fuzzy expert system; adjusting the patient problem analysis tree node according to the patient's group based on the interpretation and decision support expert system, and providing a decision suggestion in combination with the basal insulin dosage adjustment rule and the meal insulin dosage adjustment rule. Its beneficial effect is that a safe, universal and relatively accurate insulin adjustment decision suggestion is provided to the patient in decision making, and no specific initial insulin treatment plan for the patient is required.

Description

Technical Field

[0001] The present invention relates to the technical field of blood sugar control, and in particular to an AGP intelligent interpretation and insulin adjustment method based on an expert system. Background Art

[0002] The ambulatory glucose profile (AGP) is a standardized glucose test report recommended by domestic and international guidelines. It contains comprehensive blood sugar information for diabetic patients, and accurate interpretation can guide clinical treatment for patients. However, since the glucose profile and index summary omit some key information, analysts may ignore the real pattern of patients. It is necessary to carefully check the daily profile to mark and analyze events, which has a contradiction between accuracy and time cost, and requires a lot of expert knowledge.

[0003] Several expert clinical guidelines have been formulated for the analysis of AGP maps at home and abroad. At the same time, digital therapies are being continuously developed and applied to assist clinical decision-making, such as blood sugar fluctuation parameters for distinguishing type 1 and type 2 diabetes, and automatic meal detection and carbohydrate estimation methods based on artificial intelligence models, which show the application prospects of artificial intelligence in dynamic glucose maps.

[0004] In the process of interpreting AGP using the clinical "five-step method" proposed by the latest domestic guidelines, the patient's insulin and other medications are adjusted by evaluating the overall blood sugar compliance, hypoglycemia risk, blood sugar volatility, and hyperglycemia risk, and suggestions for improving lifestyle are made. However, the current atlas interpretation and decision support have the following problems:

[0005] Although the glucose map tool allows doctors and patients to more intuitively understand the blood sugar status over multiple days, it still misses some of the patient's blood sugar information and requires careful analysis of the daily map. This process requires accurate expert knowledge and a lot of time and energy. In the process, the patient's blood sugar level needs to be judged by strict thresholds, which has a certain degree of absoluteness, but in many cases, the expert's decision is flexible.

[0006] There are few studies on automated systems that help users interpret AGP profiles and optimize management recommendations. Since there are significant differences in lifestyle, physiological status, insulin sensitivity and other factors among patients, blood sugar levels are affected by many factors, including diet, exercise, and drug therapy. Traditional universal treatment methods cannot fully consider the individual differences of patients and thus cannot make personalized treatment decisions.

[0007] In the process of reviewing data to make personalized treatment decisions, doctors and patients are not only concerned about whether the insulin dose should be adjusted, but also how much the insulin dose should be adjusted. Improper insulin administration can lead to transient and severe hypoglycemia and hyperglycemia, and even diabetic ketoacidosis. In the existing domestic expert guidelines, there are no clear rules for adjusting insulin doses based on patients' multi-day blood sugar monitoring data. Summary of the invention

[0008] Technical issues to be solved

[0009] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides an AGP intelligent interpretation and insulin adjustment method based on an expert system, which solves the technical problem that there are no clear rules for adjusting insulin dosage according to the patient's multi-day blood glucose monitoring data in the existing expert guidelines.

[0010] Technical Solution

[0011] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0012] In a first aspect, the present invention provides an AGP intelligent interpretation and insulin adjustment method based on an expert system, comprising:

[0013] The knowledge base in the reasoning mechanism is established based on the AGP graph and the top-down forward reasoning strategy;

[0014] Construct interpretation and decision support expert system based on knowledge base using simplified expert system structure;

[0015] Based on the fault tree analysis method and expert guidelines, a patient problem analysis tree with three dimensions of hypoglycemia, blood sugar fluctuation and hyperglycemia was constructed. Each branch of the problem analysis tree corresponds to one or more rules.

[0016] Use expert AGP map reading and experience data to develop each rule;

[0017] The Kmeans algorithm was used to divide the patients into groups;

[0018] Based on interval type-2 fuzzy expert system, basal insulin dosage adjustment rules and meal insulin dosage adjustment rules are constructed;

[0019] Based on the interpretation and decision support expert system, the patient problem analysis tree nodes are adjusted according to the group to which the patient belongs, and decision suggestions are provided in combination with the basal insulin dose adjustment rules and meal insulin dose adjustment rules.

[0020] Optionally, the simplified expert system structure includes a knowledge base, a database, a human-computer interface, and a reasoning mechanism;

[0021] The knowledge base is used to store patient interpretation rules and decision rules based on AGP atlas;

[0022] The database is used to store patients’ personal information and blood sugar data;

[0023] The human-machine interface is used to transfer patient information into the system and display AGP reports;

[0024] The reasoning mechanism is used to utilize the rules and indicators calculated from the data in the knowledge base to infer and output the patient's AGP report content.

[0025] Optionally, establishing a knowledge base in the reasoning mechanism includes:

[0026] Check data adequacy, review overall blood sugar compliance, assess hypoglycemia risk, assess blood sugar volatility, and assess hyperglycemia risk to obtain the knowledge base in the reasoning mechanism.

[0027] Optionally, using the Kmeans algorithm to divide the patients into groups includes:

[0028] CGM Matrix for Patients Extract GSP features based on domain knowledge;

[0029] Using GSP features as input, the K-means clustering algorithm is used to obtain possible clusters of patients. K =4 as the final number of categories, and input the GSP features and K value into the kmeans algorithm;

[0030] Until the algorithm converges and outputs the cluster to which each patient belongs, the patients are finally divided into four subtype groups: T1, T2, T3, and T4 based on clinical significance.

[0031] Optionally, constructing a basal insulin dosage adjustment rule based on an interval type-2 fuzzy expert system includes:

[0032] According to the comprehensive basal insulin adjustment rules, the median of fasting blood sugar, the proportion of days with low blood sugar at night, the proportion of days with high blood sugar at night, and the proportion of days with high blood sugar at night are used as system input indicators;

[0033] 7.2mmol / L was used as the cut-off point for high fasting blood glucose in patients with type 1 and type 2 diabetes. Daily fasting blood glucose was calculated according to [6, min(t z , 8)], and the system output is used to adjust the basic insulin dose, where t z The breakfast start time is predicted by a meal detection algorithm, wherein the meal detection algorithm generates and preprocesses a data set, trains a neural network to obtain an optimal model, and applies the model to a test set to filter and output the most likely meal point as the breakfast start time;

[0034] The interval type II fuzzy set with triangle as the main and trapezoid as the auxiliary is used to establish the membership function of each index;

[0035] Based on the membership function, fuzzy logic is used to formulate rules for adjusting the patient's basal insulin dose. Among them, rules are formulated based on the probability of high and low blood sugar occurring at night. On this basis, rules are formulated for dose adjustment based on the median fasting blood glucose.

[0036] Optionally, constructing meal insulin dosage adjustment rules based on interval type-2 fuzzy expert system includes:

[0037] Comprehensive meal insulin adjustment rules, using the median of the difference between blood sugar levels 2 hours after meals and before meals, the proportion of days with low blood sugar levels before the next meal or before bedtime, and the proportion of days with high fasting blood sugar levels before the next meal or before bedtime as system input indicators;

[0038] The difference between the blood sugar level 2 hours after the meal and before the meal is calculated based on the difference between the blood sugar level 2 hours after the meal and the blood sugar level 30 minutes before the meal. The system output is the insulin adjustment dose during the meal.

[0039] The interval type II fuzzy set with triangle as the main and trapezoid as the auxiliary is used to establish the membership function of each index;

[0040] Based on the membership function, fuzzy logic is used to formulate rules for adjusting the insulin dosage during meals for diabetic patients. Among them, the rules are formulated based on the probability of high or low blood sugar before the next meal or before bedtime. On this basis, the dosage adjustment rules are formulated according to the median difference between the blood sugar value 2 hours after the meal and before the meal.

[0041] Optionally, the method further comprises:

[0042] If the patient adopted the insulin dose adjustment recommendation in the previous cycle, adjust the basal insulin αIU and the mealtime insulin βIU;

[0043] Then, in this cycle, the interval type-II fuzzy expert system of the previous cycle is used to calculate the recommended basal insulin λIU and mealtime insulin γIU;

[0044] The parameters of the membership function are updated based on λ and γ, and the insulin dosage adjustment recommendation for the current cycle is determined based on the updated membership function.

[0045] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the expert system-based AGP intelligent interpretation and insulin adjustment method described in any one of the first aspects above.

[0046] In a third aspect, the present invention provides a storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the expert system-based AGP intelligent interpretation and insulin adjustment method described in any one of the first aspects above.

[0047] Beneficial Effects

[0048] The beneficial effects of the present invention are as follows: the expert system-based AGP intelligent interpretation and insulin adjustment method of the present invention combines the expert knowledge in the medical field to quickly locate the patient's problem and establish an expert system that assists doctors in making decisions and helps patients understand their blood sugar status more clearly. The system can provide highly personalized intelligent decision-making suggestions for patients with different blood sugar control conditions, and at the same time provide patients with a safe, universal, and relatively accurate insulin adjustment decision suggestion in the decision-making process, without requiring the patient's specific initial insulin treatment plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of the structure of an interpretation and decision support expert system constructed for an embodiment of the present invention;

[0050] Figure 2 A schematic diagram of a five-step process in the reasoning mechanism of an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the overall structure of the intelligent patient problem analysis tree according to an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of the overall construction process of the intelligent patient problem analysis tree according to an embodiment of the present invention;

[0053] Figure 5 It is a schematic diagram of the structures corresponding to the four subtype groups in the intelligent patient problem analysis tree in an embodiment of the present invention;

[0054] Figure 6 It is a membership function diagram of four input indicators in the fuzzy expert system for adjusting the basal insulin dosage according to an embodiment of the present invention;

[0055] Figure 7 It is a membership function diagram of basal insulin dosage adjustment in the fuzzy expert system for basal insulin dosage adjustment according to an embodiment of the present invention;

[0056] Figure 8 It is a membership function diagram of three input indicators in the meal insulin dosage adjustment fuzzy expert system according to an embodiment of the present invention;

[0057] Fig. 9 This is a membership function diagram of the mealtime insulin dosage adjustment in the mealtime insulin dosage adjustment fuzzy expert system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0059] An embodiment of the present invention proposes an AGP intelligent interpretation and insulin adjustment method based on an expert system, including: construction of an interpretation and decision support expert system, an intelligent patient problem analysis tree, and a "basal + mealtime" insulin dosage adjustment rule based on an interval type-2 fuzzy expert system.

[0060] Compared with the related art, the present invention intelligently interprets the patient's continuous glucose monitoring chart based on the patient's historical blood sugar data and provides personalized auxiliary decision support. First of all, this method can facilitate experts and patients to more intuitively understand the patient's blood sugar status for many days, and will not miss some of the patient's blood sugar information. The process does not require more complex expert knowledge and a lot of time and energy, and the judgment of the patient's blood sugar level in the process has a certain flexibility; secondly, it enhances the relevance of decision-making, and also brings customized health management plans to each patient, thereby optimizing the treatment effect, avoiding redundant decision-making suggestions and improving the patient's quality of life. Finally, this method can better adapt to the blood sugar control of different patients or the same patient under different blood sugar states while ensuring safety, and can be widely promoted in the field of blood sugar management of diabetic patients.

[0061] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0062] In a first aspect, this embodiment provides an AGP intelligent interpretation and insulin adjustment method based on an expert system, comprising:

[0063] S1, the knowledge base in the reasoning mechanism is established based on the AGP graph and the top-down forward reasoning strategy.

[0064] S2, constructs an interpretation and decision support expert system based on the knowledge base using a simplified expert system structure.

[0065] The system can guide analysts to interpret the graph step by step. Based on the core clinical indicators and carefully selected other indicators, it introduces statistical indicators based on scenario time periods and daily high and low blood sugar event discrimination, while solving the problem of AGP graph dependence on data integrity and the difficulty in explaining AGP volatility caused by differences in daily meal times.

[0066] S3, based on the fault tree analysis method and expert guidelines, constructs a patient problem analysis tree with three dimensions of hypoglycemia, blood sugar fluctuation and hyperglycemia. Each branch of the problem analysis tree corresponds to one or more rules.

[0067] Based on the above expert system, combined with the latest AGP reading expert consensus and diabetes treatment plan adjustment rules, a patient problem analysis tree with intelligently adjustable nodes is constructed to output personalized decision suggestions. Taking into account the individual differences between patients and the targetedness of decision-making, this method can intelligently adjust the nodes of the patient problem analysis tree for patients with different blood sugar control conditions in order to provide highly personalized intelligent decision suggestions. This method not only enhances the relevance of decisions, but also brings customized health management plans to each patient, thereby optimizing the treatment effect, avoiding redundant decision suggestions and improving the quality of life of patients.

[0068] S4, using expert AGP image reading and experience data to develop each rule. The experience data is the experience data adjusted according to the cause of the disease in the past.

[0069] S5, use Kmeans algorithm to divide patients into groups.

[0070] S6, construct basal insulin dosage adjustment rules and meal insulin dosage adjustment rules based on interval type-2 fuzzy expert system.

[0071] S7, based on the interpretation and decision support expert system, adjusts the nodes of the patient problem analysis tree according to the group to which the patient belongs, and provides decision suggestions in combination with the basal insulin dose adjustment rules and meal insulin dose adjustment rules.

[0072] Fuzzy reasoning is introduced on the basis of the proposed insulin dosage adjustment rules, and the concept of uncertainty is introduced as a fuzzy rule to provide patients with a safe, universal, and relatively accurate insulin adjustment decision suggestion, which does not require the patient's specific initial insulin treatment plan. On this basis, a simple incremental learning method is used, and with multiple periodic adjustments, the parameters of the fuzzy reasoning system can be adjusted according to the patient's blood sugar improvement.

[0073] Optionally, the simplified expert system structure includes a knowledge base, a database, a human-computer interface, and a reasoning mechanism.

[0074] The knowledge base is used to store patient interpretation rules and decision rules based on AGP atlas.

[0075] The database is used to store patients' personal information and blood sugar data.

[0076] The human machine interface is used to transfer patient information into the system and to display AGP reports.

[0077] The reasoning mechanism is used to utilize the rules and indicators calculated from the data in the knowledge base to infer and output the patient's AGP report content.

[0078] Optionally, establishing a knowledge base in the reasoning mechanism includes:

[0079] Check data adequacy, review overall blood sugar compliance, assess hypoglycemia risk, assess blood sugar volatility, and assess hyperglycemia risk to obtain the knowledge base in the reasoning mechanism.

[0080] Optionally, using the Kmeans algorithm to divide the patients into groups includes:

[0081] CGM Matrix for Patients Extract GSP features based on domain knowledge;

[0082] Using GSP features as input, the K-means clustering algorithm is used to obtain possible clusters of patients. K =4 as the final number of categories, and input the GSP features and K value into the kmeans algorithm;

[0083] Until the algorithm converges and outputs the cluster to which each patient belongs, the patients are finally divided into four subtype groups: T1, T2, T3, and T4 based on clinical significance.

[0084] Optionally, constructing a basal insulin dosage adjustment rule based on an interval type-2 fuzzy expert system includes:

[0085] Combining two basal insulin adjustment rules (basal insulin adjustment rules are shown in Tables 17 and 18 below), the median of fasting blood glucose, the proportion of days with low blood glucose at night, the proportion of days with high blood glucose at night, and the proportion of days with high blood glucose at night are used as system input indicators;

[0086] 7.2mmol / L was used as the cut-off point for high fasting blood glucose in patients with type 1 and type 2 diabetes. Daily fasting blood glucose was calculated according to [6, min(t z , 8)], and the system output is used to adjust the basic insulin dose, where t z The breakfast start time is predicted by a meal detection algorithm, wherein the meal detection algorithm generates and preprocesses a data set, trains a neural network to obtain an optimal model, and applies the model to a test set to filter and output the most likely meal point as the breakfast start time;

[0087] The interval type II fuzzy set with triangle as the main and trapezoid as the auxiliary is used to establish the membership function of each index;

[0088] Based on the membership function, fuzzy logic is used to formulate rules for adjusting the patient's basal insulin dose. Among them, rules are formulated based on the probability of high and low blood sugar occurring at night. On this basis, rules are formulated for dose adjustment based on the median fasting blood glucose.

[0089] Optionally, constructing meal insulin dosage adjustment rules based on interval type-2 fuzzy expert system includes:

[0090] Combining two meal insulin adjustment rules (see Table 20 and Table 21 for meal insulin adjustment rules), the median of the difference between blood glucose values ​​2 hours after and before meals, the proportion of days with low blood glucose before the next meal or before bedtime, and the proportion of days with high fasting blood glucose before the next meal or before bedtime are used as system input indicators;

[0091] The difference between the blood sugar level 2 hours after the meal and before the meal is calculated based on the difference between the blood sugar level 2 hours after the meal and the blood sugar level 30 minutes before the meal. The system output is the insulin adjustment dose during the meal.

[0092] The interval type II fuzzy set with triangle as the main and trapezoid as the auxiliary is used to establish the membership function of each index;

[0093] Based on the membership function, fuzzy logic is used to formulate rules for adjusting the insulin dosage during meals for diabetic patients. Among them, the rules are formulated based on the probability of high or low blood sugar before the next meal or before bedtime. On this basis, the dosage adjustment rules are formulated according to the median difference between the blood sugar value 2 hours after the meal and before the meal.

[0094] Optionally, the method further comprises:

[0095] If the patient adopted the insulin dose adjustment recommendation in the previous cycle, adjust the basal insulin αIU and the mealtime insulin βIU;

[0096] Then, in this cycle, the interval type-II fuzzy expert system of the previous cycle is used to calculate the recommended basal insulin λIU and mealtime insulin γIU;

[0097] The parameters of the membership function are updated based on λ and γ, and the insulin dosage adjustment recommendation for the current cycle is determined based on the updated membership function.

[0098] The following is a specific example to further illustrate an AGP intelligent interpretation and insulin adjustment method based on an expert system of the present application:

[0099] 1. Construction of interpretation and decision support expert system

[0100] like Figure 1As shown, this method organizes the existing expert consensus, analyzes the AGP spectrum from the perspective of daily spectrum to supplement the information missed by the AGP spectrum, quickly locates patient problems, and establishes an expert system to assist doctors in decision-making and help patients better understand their blood sugar status.

[0101] A simplified expert system structure is adopted, including knowledge base, database, human-computer interface, and reasoning mechanism, such as Figure 1 As shown in the figure, the knowledge base is used to store the patient interpretation rules and decision rules based on the AGP map; the database is used to store the patient's personal information and blood sugar data; the human-computer interface is mainly responsible for transmitting the patient's information to the system and displaying the AGP report; the reasoning mechanism uses the rules combined with the indicators calculated by the data in the database to infer the patient's AGP report content and output it.

[0102] The forward reasoning strategy from top to bottom is adopted, which is to match the corresponding rules in the knowledge base according to the known facts. The forward reasoning engine of PyKnow is used. The whole reasoning process is as follows: Figure 2 shown.

[0103] The specific analysis rules are as follows:

[0104] (1) Check data adequacy issues and analysis:

[0105] Let d be the number of days the patient wears CGM (days), and e be the proportion of valid CGM data (%). Table 1 shows the evaluation of data adequacy issues and analysis rules.

[0106] Table 1

[0107]

[0108] (2) Check the overall blood sugar standard and analyze the following issues:

[0109] set up The time when the median falls outside the interval range is shown in Table 2. Indicates the TIR threshold.

[0110] Table 2

[0111]

[0112] (3) Assessment of hypoglycemia risk issues and analysis:

[0113] With reference to the expert's ideas, the hypoglycemia risk assessment rules were designed as shown in Table 3 0301-0308. Individualized TBR and thresholds were used for different types of diabetic patients. .

[0114] Table 3

[0115]

[0116] In addition, in order to enhance the information provided by the AGP map, the daily hypoglycemia period is judged based on the scene information. , , The definitions of the daily pre-meal, nighttime, and bedtime periods are shown in the following table. The problem analysis rules are shown in 0309-0318 in the above table. The definitions of the daily pre-meal, nighttime, and bedtime periods are shown in Table 4.

[0117] Table 4

[0118]

[0119] (4) Assessment and analysis of blood sugar volatility:

[0120] Blood sugar fluctuations can be divided into two aspects: intraday blood sugar fluctuations and interday blood sugar fluctuations. Common indicators for evaluating blood sugar fluctuations in continuous blood sugar monitoring include SDBG, MAGE, LAGE, CV, and MODD.

[0121] Expert consensus recommends evaluating intraday blood sugar fluctuations based on the fluctuations in the median during the night, breakfast, lunch, and dinner periods, and discussing them in sections to facilitate tracking of the cause of the problem. are the average values ​​of SDBG, LAGE, and CV calculated during the night time of each day. ; ; The average values ​​of SDBG, LAGE, and CV calculated for the breakfast, lunch, and dinner periods of each day are shown in Table 5 below. The nighttime and breakfast, lunch, and dinner periods are divided according to the Guidelines for the Clinical Application of Continuous Glucose Monitoring in China (2017 Edition).

[0122] Table 5

[0123]

[0124] On the other hand, for daily fluctuations, it is recommended to view the interquartile range (IQR) and inter-twentieth range (IVR) in segments. However, the IQR and IVR are set to visualize the continuous blood glucose data for multiple days for the convenience of doctors to view, and there is no clear threshold. Therefore, MODD is adopted, and the is the average MODD. The questions and analysis rules for evaluating blood glucose variability are shown in Table 6 below.

[0125] Table 6

[0126]

[0127] (5) Assessment of hyperglycemia risk issues and analysis:

[0128] The design idea of ​​the hyperglycemia risk assessment rule refers to the hypoglycemia risk assessment in (3), assuming that the time when the 20th digit line is in the hyperglycemia range is . Refer to different types of diabetic patients to use individualized TAR and thresholds. The rules for evaluating hyperglycemia risk are shown in Table 7 below. In addition, the statistics and analysis of common categories of hyperglycemic events are supplemented according to the definition of common hyperglycemic events, as shown in Table 8 below. This work will improve the efficiency of analysts' decision-making and trace the specific causes of the hyperglycemic area in AGP. For different types of diabetic patients, the fasting and postprandial blood sugar control targets are different. The upper limits of the blood sugar control range are set as follows: If fasting and postprandial hyperglycemia exceed As a definition. Refer to the expert consensus on the clinical application of dynamic glucose profile reports (2023 edition), the guidelines for the diagnosis and treatment of hyperglycemia in pregnancy (2022), the status of blood glucose management and coping strategies for hospitalized patients in China, and Johnson et al. set 130 mg / dL as the fasting or pre-meal blood glucose threshold for patients with type 2 diabetes. The upper limit of the blood glucose control range for patients with type 12 diabetes, elderly high-risk diabetes, and pregnant type 1 diabetes , as shown in Table 9 below, can also be customized according to the patient's actual situation . To determine whether these six types of hyperglycemia events occur every day, the design calculates the early noon The number of days with hyperglycemia after dinner was Insufficient insulin use at night Dawn phenomenon The number of days for the Somujie phenomenon events is .

[0129] Table 7

[0130]

[0131] Table 8

[0132]

[0133] Table 9

[0134]

[0135] The meaning or calculation method of the special design indicators in the rules are explained as follows:

[0136] Days with hyperglycemia :

[0137] The numbers of days with six types of hyperglycemia events during the monitoring period are shown in Table 10.

[0138] Table 10

[0139]

[0140] 2. Construction of intelligent patient problem analysis tree

[0141] ①GSP feature extraction

[0142] First, for the patient CGM matrix Calculate the patient's GSP characteristics , M is the number of patients, N is the number of blood glucose values ​​collected by the patient. The GSP features are used as input to obtain the possible clusters of patients using the K-means clustering algorithm.

[0143] The feature extraction algorithm process is described as follows:

[0144] The calculation formula is as follows:

[0145] ;

[0146] ;

[0147] in Indicates that the CGM sequence is in i The blood sugar pattern represented by the moment, Representation Mode P No. k blood sugar patterns, and In the k The positions have the same meaning, and the function Used to identify whether the symbol status at the current moment is consistent with If they are the same, they are counted; if they are different, they are assigned a value of 0.

[0148] The algorithm pseudo code is shown in Table 11.

[0149] Table 11

[0150]

[0151] ②Kmeans clustering

[0152] Before using the algorithm, you must first determine the number of clusters KTo ensure the accuracy of the results, four evaluation criteria are selected. K .Will K =4 as the final number of categories, and the GSP features and K values ​​were input into the kmeans algorithm until the algorithm converged and output the cluster to which each patient belonged. Finally, they were divided into the following four categories based on clinical significance, as shown in Table 12.

[0153] Table 12

[0154] Clinical significance Adjustment strategy T1 Blood sugar control is relatively good, with more IS patterns, slight fluctuations and a small amount of hypoglycemia Conservative treatment strategy T2 Hyperglycemia fluctuations are more pronounced; Precision treatment strategies T3 Most of them are in high blood sugar, but the fluctuation is smaller than T2; Precision treatment strategies T4 Best blood sugar control Conservative treatment strategy

[0155] ③Problem analysis tree design

[0156] like Figure 3 As shown, the program adjustment part requires AGP chart reading and patient treatment integration, which is not easy to directly design rules. Referring to the fault tree analysis method and expert guidelines, a problem analysis tree with three dimensions of patient hypoglycemia, blood sugar fluctuation, and hyperglycemia is constructed.

[0157] Each branch of the problem analysis tree corresponds to one or more rules. Experts use their experience in AGP reading and adjusting based on the cause to expand each rule, and organize the meaning of events and corresponding adjustment plans as shown in Table 13 below.

[0158] Table 13

[0159] event significance Adjustment plan A1 Not eating enough Meals should be taken at regular times and in fixed amounts. If the amount of food eaten is reduced, the dose of hypoglycemic drugs should be reduced accordingly. If meals are likely to be missed, preparations should be made in advance (a1) A2 Binge drinking and drinking on an empty stomach Binge drinking and drinking on an empty stomach should be avoided (a2) A3 Increased exercise Choose a suitable exercise method according to your condition and physical fitness, and increase your carbohydrate intake before exercise (a3) A4 other If there are other possible causes (such as vomiting, diarrhea, autonomic dysfunction, liver and kidney dysfunction), make targeted adjustments (a4) B1 Basal insulin Reduce basal insulin (b1) B2 More insulin at mealtime Reduce breakfast insulin (b2-1) / Reduce lunch insulin (b2-2) / Reduce dinner insulin (b2-3) C1 Sleep late Encourage early bedtime (c1) C2 Less insulin at mealtime Increase breakfast insulin (c2-1) / Increase lunch insulin (c2-2) / Increase dinner insulin (c2-3) C3 High carbohydrate ratio Reduce the carbohydrate ratio of breakfast (c3-1) / Reduce the carbohydrate ratio of lunch (c3-2) / Reduce the carbohydrate ratio of dinner (c3-3) D1,D2, D3 Irregular sleep / eating / exercise Regular sleep (d1) / diet (d2) / exercise (d3) E1 Fear of hypoglycemia Provide psychological support to patients to prevent fear of hypoglycemia (e1) F1 Low basal insulin Increase basal insulin (f1) G1 Less insulin at mealtime Increase breakfast insulin (g1-1) / Increase lunch insulin (g1-2) / Increase dinner insulin (g1-3) H1,I1, J1 Dawn phenomenon, Somogyi phenomenon, other hyperglycemic events Adjust according to specific circumstances K1,K2 Insulin dose adjustment recommendations / conservative insulin adjustment recommendations Accurate dosage adjustment based on fuzzy inference output / appropriate reduction or increase based on blood sugar levels

[0160] Combining the above table with AGP chart reading, we get the rule conditions and conclusions, as shown in Tables 14, 15 and 16. The rule conclusions correspond to the plan adjustments in the above table, and the rule conditions correspond to the performance of the event on the indicator. Assume that the median fluctuations during breakfast, lunch and dinner are , , . Assume that the proportion of days with hypoglycemia in the four periods of night, before lunch, before dinner, and before bed is The proportion of days with high blood sugar is , the proportion of days with high blood sugar is , the rules give priority to hypoglycemia adjustment, thus obtaining insulin adjustment rules 0605-0608, 0702-0707, 0804-0808.

[0161] Table 14

[0162]

[0163] Table 15

[0164]

[0165] Table 16

[0166]

[0167] ④ Problem analysis tree intelligent adjustment design

[0168] Although patients can obtain intelligent decision support through the established patient problem analysis tree, this method often cannot provide sufficient pertinence and personalized suggestions. Therefore, the present invention adopts a clustering method, which divides blood glucose symbol pattern (GSP) into four subtype groups, T1, T2, T3, and T4, using the Kmeans algorithm based on the characteristics of blood glucose symbol pattern (GSP). In this way, the expert system can provide more personalized and targeted intelligent decision suggestions according to the group to which the patient belongs by intelligently adjusting the problem tree nodes. The specific analysis flow chart is as follows Figure 4 The analysis tree used for each group is shown in Figure 5 shown.

[0169] 3. Construction of “basal + mealtime” insulin dosage adjustment method based on interval type-2 fuzzy expert system

[0170] ①Fuzzy expert system for adjusting basal insulin dosage

[0171] Indicator selection and fuzzification are shown in Table 17 and Table 18:

[0172] Table 17

[0173] Fasting blood glucose (mmol / L) Basal insulin dose adjustment (IU) <4.4 -2 4.4~6.1 0 6.2~7.8 +2 7.9~10.0 +4 >10 +6

[0174] Table 18

[0175] Fasting blood sugar Basal insulin dose adjustment (IU) < 70 mg / dL on more than two occasions -4 < 70 mg / dL (1 mL) 0 No <70 mg / dL and >130 mg / dL on more than three occasions +2 No <70 mg / dL and >180 mg / dL on more than three occasions +4 Other situations 0

[0176] Combining the two basic insulin adjustment rules in the above two tables, the median of fasting blood sugar for multiple days , the proportion of days with low blood sugar level at night , the proportion of days with high blood sugar at night , the proportion of days with high blood sugar levels at night As the system input indicator (at least two CGM data numbers are within the set range during the time period), the membership function diagram of the above indicators is as follows Figure 6 Assume that the system output is the basic insulin adjustment dose .

[0177] In order to reduce the overall computational complexity, interval type-2 fuzzy sets with triangle as the main and trapezoid as the auxiliary are used to establish the membership functions of various indicators. Set low on the domain ,middle , slightly higher , higher , very high Five fuzzy sets are set with reference indicators for type 1 and type 2 diabetes patients, and the membership functions are as follows:

[0178]

[0179] For different types of patients, The parameters of the fuzzy set membership function on the domain are shown in Table 19:

[0180] Table 19

[0181]

[0182] Set low on the domain ,middle ,high Three fuzzy sets. The membership functions are as follows:

[0183]

[0184] Set low ,high Two fuzzy sets. The membership functions are as follows:

[0185]

[0186] Output Set the domain to reduce more , reduce less , no change , less increase , increased more , increase a lot Six fuzzy sets. The membership function diagram is as follows Figure 7 The membership function is as follows:

[0187]

[0188] Rule design:

[0189] Fuzzy logic is used to formulate rules for adjusting the basal insulin dose for diabetic patients. Rules are formulated based on the probability of high and low blood sugar levels occurring at night, and on this basis, rules are formulated for the dosage adjustment based on the median of fasting blood sugar levels over multiple days.

[0190] ② Fuzzy expert system for adjusting meal insulin dosage

[0191] Indicator selection and fuzzification are shown in Table 20 and Table 21:

[0192] Table 20

[0193] Difference between blood sugar level 2 hours after meal and blood sugar level before meal Mealtime insulin adjustment dose (IU) <1.7mmol / L -1~-4 >3.3mmol / L 2~4

[0194] Table 21

[0195] Blood sugar level before the next meal / before bed Mealtime insulin adjustment dose (IU) < 70 mg / dL on more than two occasions -2 < 70 mg / dL (1.5 mmol / L) 0 No <70 mg / dL and >130 mg / dL on more than three occasions +2 Other situations 0

[0196] Combining the meal insulin adjustment rules in the above two tables, the median of the difference between the blood glucose values ​​2 hours after meals and before meals for multiple days is calculated. , the proportion of days when blood sugar is in the low blood sugar range before the next meal ( / before going to bed) , the proportion of days when blood sugar before the next meal ( / before bedtime) is high in fasting blood sugar (More than two CGM data within a fixed range within a time period are considered to be events). As the system input indicator, the membership function diagram of the indicator is as follows: Figure 8 Assume that the system output is the insulin dose adjustment during meals .

[0197] The idea of ​​setting the membership function is the same as that of adjusting the basal insulin dose. Set low on the domain ,middle ,high Three fuzzy sets. The membership functions are as follows:

[0198]

[0199] Set low on the domain ,middle ,high Three fuzzy sets, . Set low on the domain ,high Two fuzzy sets, .

[0200] Set the domain to reduce more , reduce less , no change , less increase , increased more Five fuzzy sets. The membership function diagram is as follows Fig. 9 The membership function is as follows:

[0201]

[0202] Rule design:

[0203] Fuzzy logic is used to formulate rules for adjusting insulin dosage during meals for diabetic patients. Rules are formulated based on the probability of taking the insulin before the next meal (or before going to bed) first, and then the dosage adjustment is formulated based on the median difference between the blood sugar values ​​2 hours after meals and before meals for multiple days.

[0204] ③Incremental learning of rule consequent membership function

[0205] The insulin dosage adjustment fuzzy inference system designed in the previous article can provide general safety recommendations for patients' continuous blood glucose monitoring for multiple days. With multiple periodic adjustments, the parameters of the fuzzy inference system can be adjusted according to the patient's blood glucose improvement. Here, a simple incremental learning method is considered to adjust the membership function parameters of the rule consequence.

[0206] If the patient adopted the insulin adjustment suggestion of the fuzzy inference system in the previous cycle, assuming that the basal insulin is adjusted IU (including positive and negative signs), adjust mealtime insulin IU (the adjustment rules for multiple meals are the same, and the concept of mealtime insulin is used here to refer to it). In this cycle, the original fuzzy reasoning system is used to recommend adjusting the basal insulin IU, mealtime insulin IU. It can be understood that in the last periodic adjustment, the deviations were and . Based on this, adjust the membership function parameters of the fuzzy inference system rule consequent. To ensure safety, set the maximum parameter adjustment range to 1. Suppose the membership function diagram of the basal insulin adjustment dose , , , , When the membership degree is m, the values ​​on the corresponding domain are , , , , , the membership function diagram of the mealtime insulin dose adjustment , , , When the membership degree is m, the values ​​on the corresponding domain are , , , The parameter update method formula is as follows:

[0207] ;

[0208]

[0209] That is, the membership function of the fuzzy set of the rule consequent is moved holistically according to the output of the original reasoning system, and the domain is also expanded accordingly.

[0210] In a second aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements an expert system-based AGP intelligent interpretation and insulin adjustment method as described in any one of the first aspects above.

[0211] In a third aspect, an embodiment of the present invention provides a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, an expert system-based AGP intelligent interpretation and insulin adjustment method as described in any one of the first aspects above is implemented.

[0212] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0213] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

[0214] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for intelligent interpretation of AGP and insulin adjustment based on an expert system, characterized in that: include: The knowledge base in the reasoning mechanism is established based on the AGP graph and the top-down forward reasoning strategy; Construct interpretation and decision support expert system based on knowledge base using simplified expert system structure; Based on the fault tree analysis method and expert guidelines, a patient problem analysis tree with three dimensions of hypoglycemia, blood sugar fluctuation and hyperglycemia was constructed. Each branch of the problem analysis tree corresponds to one or more rules. Use expert AGP map reading and experience data to develop each rule; The Kmeans algorithm was used to divide the patients into groups; Based on interval type-2 fuzzy expert system, basal insulin dosage adjustment rules and meal insulin dosage adjustment rules are constructed; Based on the interpretation and decision support expert system, the patient problem analysis tree nodes are adjusted according to the patient's group, and decision suggestions are provided in combination with the basal insulin dose adjustment rules and meal insulin dose adjustment rules; The simplified expert system structure includes a knowledge base, a database, a human-computer interface, and a reasoning mechanism; The knowledge base is used to store patient interpretation rules and decision rules based on AGP atlas; The database is used to store patients’ personal information and blood sugar data; The human-machine interface is used to transfer patient information into the system and display AGP reports; The reasoning mechanism is used to utilize the rules and indicators calculated from the data in the knowledge base to infer and output the patient's AGP report content.

2. The method for intelligent interpretation of AGP and insulin adjustment based on an expert system according to claim 1, characterized in that: The knowledge base in establishing the reasoning mechanism includes: Check data adequacy, review overall blood sugar compliance, assess hypoglycemia risk, assess blood sugar volatility, and assess hyperglycemia risk to obtain the knowledge base in the reasoning mechanism.

3. The method for intelligent interpretation of AGP and insulin adjustment based on an expert system according to claim 1, characterized in that: The use of Kmeans algorithm to divide patients into groups includes: CGM Matrix for Patients Extract GSP features based on domain knowledge; Using GSP features as input, the K-means clustering algorithm is used to obtain possible clusters of patients. K =4 as the final number of categories, and input the GSP features and K value into the kmeans algorithm; Until the algorithm converges and outputs the cluster to which each patient belongs, the patients are finally divided into four subtype groups: T1, T2, T3, and T4 based on clinical significance.

4. The method of AGP intelligent interpretation and insulin adjustment based on expert system according to claim 1, characterized in that: The basal insulin dosage adjustment rules based on interval type-2 fuzzy expert system include: According to the comprehensive basal insulin adjustment rules, the median of fasting blood sugar, the proportion of days with low blood sugar at night, the proportion of days with high blood sugar at night, and the proportion of days with high blood sugar at night are used as system input indicators; 7.2mmol / L was used as the cut-off point for high fasting blood glucose in patients with type 1 and type 2 diabetes. Daily fasting blood glucose was calculated according to [6, min(t z , 8)], and the system output is used to adjust the basic insulin dose, where t z The breakfast start time is predicted by a meal detection algorithm, wherein the meal detection algorithm generates and preprocesses a data set, trains a neural network to obtain an optimal model, and applies the model to a test set to filter and output the most likely meal point as the breakfast start time; The interval type II fuzzy set with triangle as the main and trapezoid as the auxiliary is used to establish the membership function of each index; Based on the membership function, fuzzy logic is used to formulate rules for adjusting the patient's basal insulin dose. Among them, rules are formulated based on the probability of high and low blood sugar occurring at night. On this basis, rules are formulated for dose adjustment based on the median fasting blood glucose.

5. The method of AGP intelligent interpretation and insulin adjustment based on expert system according to claim 1, characterized in that: The meal insulin dosage adjustment rules based on interval type-2 fuzzy expert system include: Comprehensive meal insulin adjustment rules, using the median of the difference between blood sugar levels 2 hours after meals and before meals, the proportion of days with low blood sugar levels before the next meal or before bedtime, and the proportion of days with high fasting blood sugar levels before the next meal or before bedtime as system input indicators; The difference between the blood sugar level 2 hours after the meal and before the meal is calculated based on the difference between the blood sugar level 2 hours after the meal and the blood sugar level 30 minutes before the meal. The system output is the insulin adjustment dose during the meal. The interval type II fuzzy set with triangle as the main and trapezoid as the auxiliary is used to establish the membership function of each index; Based on the membership function, fuzzy logic is used to formulate rules for adjusting the insulin dosage during meals for diabetic patients. Among them, the rules are formulated based on the probability of high or low blood sugar before the next meal or before bedtime. On this basis, the dosage adjustment rules are formulated according to the median difference between the blood sugar value 2 hours after the meal and before the meal.

6. The method for intelligent interpretation of AGP and insulin adjustment based on an expert system according to claim 1, characterized in that: The method further comprises: If the patient adopted the insulin dose adjustment recommendation in the previous cycle, adjust the basal insulin αIU and the mealtime insulin βIU; Then, in this cycle, the interval type-II fuzzy expert system of the previous cycle is used to calculate the recommended basal insulin λIU and mealtime insulin γIU; The parameters of the membership function are updated based on λ and γ, and the insulin dosage adjustment recommendation for the current cycle is determined based on the updated membership function.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for intelligent interpretation of AGP and adjustment of insulin based on an expert system as described in any one of claims 1 to 6 is implemented.

8. A storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, characterized in that: When the processor executes the computer program, the method for intelligent interpretation of AGP and adjustment of insulin based on an expert system as described in any one of claims 1 to 6 is implemented.

Images