Intelligent insulin recommendation method, system and electronic device

Abstract

The present invention belongs to the field of physiological signal processing technology, and discloses an intelligent insulin recommendation method, system and electronic device, including: obtaining a number of historical clinical case data in a database; extracting the historical clinical indicators associated with the newly input clinical indicators from the historical clinical medical record data as associated clinical indicators; training the basic clinical indicators and their corresponding historical clinical diagnosis results as training objects one by one; comparing and combining the first set with the reference file to locate the first-level indicators of the insulin recommendation scheme; calculating the initial total daily insulin dose in the insulin recommendation scheme, and establishing a corresponding insulin dose adjustment scheme; based on the previous insulin injection site data, establishing a corresponding insulin injection site rotation adjustment scheme. The present invention can quickly locate an effective insulin recommendation scheme, and improve the user's comfort on the basis of implementing effective blood sugar control.

Description

Technical Field

[0001] The present invention relates to the technical field of physiological signal processing, and in particular to an intelligent insulin recommendation method, system and electronic equipment. Background Art

[0002] Hyperglycemia during pregnancy mainly includes gestational diabetes mellitus (GDM), overt diabetes mellitus (ODM) and pre-gestational diabetes mellitus (PGDM). With the rapid growth of diabetes patients in my country and the increase of older pregnant women after the adjustment of fertility policy, hyperglycemia during pregnancy has become the most common complication of pregnancy, with an incidence of about 17.5%. Studies on the relationship between hyperglycemia and adverse pregnancy outcomes have shown that even mild hyperglycemia during pregnancy will increase the risk of adverse pregnancy outcomes such as large-for-gestational-age infants, cord blood C-peptide greater than the 90th percentile, cesarean section delivery and neonatal hypoglycemia, and the offspring may develop metabolic syndrome such as hyperinsulinemia, obesity, and hypertension in adulthood after delivery. Therefore, the American Diabetes Association (ADA) recommends strict control of blood sugar throughout pregnancy.

[0003] Common methods for controlling blood sugar include lifestyle intervention and medical nutrition therapy. Usually, lifestyle intervention is performed immediately after the diagnosis of GDM is established. However, insulin therapy can be started if ketosis recurs after diet control, or blood sugar reaches the standard but weight loss occurs, blood sugar exceeds the standard after increasing calorie intake, or blood sugar test is still not up to standard after one week of diet and exercise therapy. Insulin is the only hormone in the human body that can lower blood sugar. Together with other glucose-raising hormones, it maintains blood sugar in the normal range in the body. It is a macromolecular substance that does not pass through the placenta and can effectively control blood sugar. It is the first choice for treating hyperglycemia during pregnancy. At the same time, the secretion and regulation of insulin in pregnant women will be affected by pregnancy. After pregnancy, the peak of postprandial insulin secretion increases and arrives later, and the recovery to normal levels is also slowed down, and even insulin resistance (IR) may occur. Therefore, timely and effective insulin therapy is particularly important in the process of controlling blood sugar in pregnant women.

[0004] In recent years, remote or mobile medical technologies have been increasingly used in the field of hyperglycemia during pregnancy, and the use of technology to remotely and accurately manage and monitor blood sugar in pregnant women has become a general trend. However, the current insulin treatment plan for patients with hyperglycemia during pregnancy is still mainly calculated manually by endocrinologists. Due to the lack of standardized clinical pathways and decision-making criteria, and the inability to take into account the differences between patients, even under effective monitoring conditions, it is still difficult or delayed for patients to obtain correct treatment, resulting in unnecessary increased risks to mothers and infants. On the other hand, insulin therapy is a long-term and sophisticated method of blood sugar control. However, communication between doctors and patients is currently lacking, and it still mainly relies on face-to-face consultations. This requires patients to go to the hospital regularly and multiple times in the process of controlling blood sugar, which consumes a lot of time and energy, and at the same time cannot get effective management and help in a very timely manner. Summary of the invention

[0005] The present invention provides an intelligent insulin recommendation method, system and electronic equipment, which can timely and effectively provide appropriate drug use decision-making plans for patients with hyperglycemia during pregnancy.

[0006] The present invention is achieved through the following technical solutions:

[0007] An intelligent insulin recommendation method, including:

[0008] S1. Obtaining a number of historical clinical case data in a database, each of the historical clinical case data includes a number of historical clinical indicators and their corresponding historical clinical diagnosis results;

[0009] S2, extracting the historical clinical indicators associated with the newly input clinical indicators from the plurality of historical clinical medical record data as associated clinical indicators;

[0010] S3, calculating a first correlation coefficient between the associated clinical indicator and the corresponding historical clinical diagnosis result, and extracting the associated clinical indicator with a higher first correlation coefficient as a basic clinical indicator;

[0011] S4, using the basic clinical indicators and their corresponding historical clinical diagnosis results as training objects for training one by one, so as to obtain prediction results of insulin type, insulin use time, and insulin injection method corresponding to each training object;

[0012] S5, calculating the second correlation coefficient between each training object and its corresponding prediction result, and extracting the prediction results whose second correlation coefficient is greater than the second threshold, establishing a first set of prediction results and corresponding historical clinical case data, and then comparing and combining the first set with the reference file to locate and obtain the first-level indicator of the insulin recommendation scheme, wherein the first-level indicator of the insulin recommendation scheme includes insulin type, insulin use time, and insulin injection method;

[0013] S6. Calculate the total initial daily insulin dose in the secondary index of the insulin recommendation scheme, and establish a corresponding insulin dose adjustment scheme based on the reference file, the total initial daily insulin dose combined with the relevant basic clinical indicators and the corresponding historical clinical diagnosis results, to obtain the secondary index of the insulin recommendation scheme, wherein the secondary index of the insulin recommendation scheme includes the total initial daily insulin dose and the insulin dose adjustment scheme;

[0014] S7. Based on the data of the previous insulin injection site, by comparing and combining the reference file and the corresponding basic clinical index and the corresponding historical clinical diagnosis result, a corresponding third-level index of the insulin recommendation scheme is established, wherein the third-level index of the insulin recommendation scheme is an insulin injection site rotation adjustment scheme;

[0015] S8, training the insulin recommendation scheme primary index, insulin recommendation scheme secondary index, insulin recommendation scheme tertiary index and their corresponding historical clinical diagnosis results one by one as training objects, so as to obtain the insulin recommendation scheme prediction results corresponding to each training object;

[0016] S9. Storing and outputting the insulin recommendation plan, wherein the insulin recommendation plan includes a first-level indicator of the insulin recommendation plan, a second-level indicator of the insulin recommendation plan, and a third-level indicator of the insulin recommendation plan.

[0017] As an optimization, in S3, the basic clinical indicators include age, gestational age, gestational weight, body mass index, insulin sensitivity coefficient, carbohydrate coefficient, diabetes classification, diabetes course, blood routine test data, blood sugar level, blood lipid level, history of hypoglycemia, diabetic complications, combined use of hypoglycemic drugs, insulin secretion function, creatinine-blood, creatinine-urine, alanine aminotransferase, aspartate aminotransferase, blood pressure, diabetes symptoms and insulin type, insulin use time, insulin injection method, initial total insulin dose, initial total daily insulin dose, daily increase in insulin dose, and insulin injection site.

[0018] As an optimization, the specific steps of S4 are:

[0019] S4.1. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the logistic regression model for training, and output a first prediction result for predicting the type of insulin, the duration of insulin use, and the method of insulin injection, wherein the first prediction result includes the type of insulin, the duration of insulin use, and the method of insulin injection;

[0020] S4.2, inputting the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the support vector machine model for training, and outputting a second prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the second prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection;

[0021] S4.3, inputting the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the K nearest neighbor classification model for training, and outputting a third prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the third prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection;

[0022] S4.4. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and output a fourth prediction result for predicting the type of insulin, insulin usage time, and insulin injection method. The fourth prediction result includes the type of insulin, insulin usage time, and insulin injection method.

[0023] As an optimization, the specific implementation steps of S5 are:

[0024] S5.1, respectively calculating the second correlation coefficient between each of the training objects and its corresponding first prediction result, second prediction result, third prediction result, and fourth prediction result;

[0025] S5.2, extracting prediction results corresponding to higher second correlation coefficients among a number of second correlation coefficients, and combining the prediction results with corresponding historical clinical case data to form a first set;

[0026] S5.3. Compare and combine the first set with the reference file to locate and obtain the first-level indicators of the insulin recommendation plan.

[0027] As an optimization, the types of insulin include: mealtime insulin, basal insulin, and mealtime + basal insulin combination. The insulin usage times include: basal insulin 7:00, i.e. b7; basal insulin 16:00, i.e. b16; basal insulin 22:00, i.e. b22; mealtime + basal insulin combination 7:30, i.e. p7:30; mealtime + basal insulin combination 16:30, i.e. p16:30; mealtime insulin 6:30, i.e. s6:30; mealtime insulin 11:30, i.e. s11:30; mealtime insulin 16:30, i.e. s16:30. The insulin injection methods include: subcutaneous injection and intravenous drip.

[0028] As an optimization, the specific steps of S6 are:

[0029] S6.1. Input the initial total insulin dose in the basic clinical indicators and the historical clinical diagnosis results corresponding to the initial total insulin dose into the deep deterministic policy gradient model for machine learning training, and establish the function of the initial total insulin dose and the initial total daily insulin dose: y=k+R, wherein y represents the initial total daily insulin dose, k represents the initial total insulin dose, ranging from 0.2 to 1.0 U / (kg*d), and R represents the daily increase in insulin dose, ranging from -4 to -2 U or 2 to 4 U, or R≤0.2y. If the patient's blood sugar is stable, R is 0;

[0030] S6.2. Based on the reference file, the basic clinical indicators in each clinical medical record data and the corresponding historical clinical diagnosis results are input into the deep deterministic policy gradient model for machine learning training to obtain an insulin dosage adjustment plan.

[0031] As an optimization, the specific implementation steps of S7 include:

[0032] S7.1. Based on the reference file, the basic clinical indicators in each clinical medical record data and the corresponding historical clinical diagnosis results are input into the deep deterministic strategy gradient model for machine learning training to obtain an insulin injection site rotation adjustment plan.

[0033] As an optimization, the specific implementation steps of S8 include:

[0034] S8.1. Input the insulin recommendation regimen and its corresponding historical clinical diagnosis results as input parameters into the logistic regression model for training, and output a fifth prediction result for predicting the first-level index of the insulin recommendation regimen, the second-level index of the insulin recommendation regimen, and the third-level index of the insulin recommendation regimen, wherein the fifth prediction result includes the first-level index of the insulin recommendation regimen, the second-level index of the insulin recommendation regimen, and the third-level index of the insulin recommendation regimen;

[0035] S8.2, inputting the insulin recommendation scheme and its corresponding historical clinical diagnosis results as input parameters into the support vector machine model for training, and outputting a sixth prediction result of the insulin recommendation scheme primary index, the insulin recommendation scheme secondary index, and the insulin recommendation scheme tertiary index, wherein the sixth prediction result includes the insulin recommendation scheme primary index, the insulin recommendation scheme secondary index, and the insulin recommendation scheme tertiary index;

[0036] S8.3, input the insulin recommendation scheme and its corresponding historical clinical diagnosis results as input parameters into the K nearest neighbor classification model for training, and output a seventh prediction result for predicting the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme, wherein the seventh prediction result includes the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme;

[0037] S8.4. Input the insulin recommendation scheme and its corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and output the eighth prediction result for predicting the primary indicators of the insulin recommendation scheme, the secondary indicators of the insulin recommendation scheme, and the tertiary indicators of the insulin recommendation scheme. The eighth prediction result includes the primary indicators of the insulin recommendation scheme, the secondary indicators of the insulin recommendation scheme, and the tertiary indicators of the insulin recommendation scheme.

[0038] The present invention also discloses an intelligent insulin recommendation system, comprising:

[0039] Database, used to store historical clinical medical record data;

[0040] An input module, for receiving new input clinical indicators;

[0041] A data acquisition module, used for acquiring the historical clinical case data from a database;

[0042] A related index extraction module, used for comparing the new clinical index inputted in the input module with the historical clinical case data acquired by the data acquisition module, and acquiring the historical clinical index associated with the new clinical index as the related clinical index;

[0043] A basic index extraction module, used for calculating a first correlation coefficient between the associated clinical index and the corresponding historical clinical diagnosis result, and extracting the associated clinical index with a higher first correlation coefficient;

[0044] Training module 1, for training the basic clinical indicators and their corresponding historical clinical diagnosis results one by one as training objects, so as to obtain prediction results of insulin type, insulin use time and insulin injection method corresponding to each training object;

[0045] A matching screening module is used to calculate the second correlation coefficient between each training object and its corresponding prediction result, and extract the prediction result with a higher second correlation coefficient, establish a first set of prediction results and corresponding clinical case data, and then compare and combine the first set with the reference file to locate the first-level indicator of the insulin recommendation plan;

[0046] The initial total dose module is used to calculate the initial total daily insulin dose in the secondary indicators of the insulin recommendation plan;

[0047] An insulin dosage adjustment module, used for adjusting the dosage of insulin injection based on the initial total daily insulin dosage;

[0048] An injection site rotation module is used to obtain the injection site of each insulin injection, and the injection site of the next injection is different from the previous injection;

[0049] The training module 2 is used to train the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, the third-level index of the insulin recommendation scheme and their corresponding historical clinical diagnosis results as training objects one by one, so as to obtain the prediction results of the insulin recommendation scheme corresponding to each training object;

[0050] A memory bank, used to store the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, the third-level index of the insulin recommendation scheme, and the training process of obtaining the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme;

[0051] The prediction module uses the first-level index, second-level index and third-level index of the stored insulin recommendation scheme as the insulin recommendation scheme, uses artificial intelligence technology to predict the effect of artificial blood sugar control, and matches it with the blood sugar control target value to complete a recommendation;

[0052] The review module combines the insulin recommendation scheme and the prediction result of the prediction module with the expert review opinions to complete the secondary recommendation;

[0053] The output module is used to output insulin recommendation plans and expert review opinions related to the newly input clinical indicators.

[0054] The present invention also discloses an electronic device, comprising at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute an intelligent insulin recommendation method as described above.

[0055] The present invention also discloses a storage medium storing a computer program, wherein the computer program realizes the above-mentioned intelligent insulin recommendation method when executed by a processor.

[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0057] 1. The present invention can quickly locate an effective insulin recommendation plan, make timely adjustments based on the user's physical indicators, accurately locate the daily insulin dosage, and ensure that a more accurate and appropriate insulin recommendation plan is provided to the user;

[0058] 2. The present invention fully considers the rotation of insulin injection sites, which can effectively reduce the occurrence of nodules and infections on the local skin due to long-term injections, and improve the user's comfort on the basis of implementing effective blood sugar control;

[0059] 3. The present invention recommends the insulin recommendation scheme given in the prediction module twice, optimizes the remote or mobile medical technology and the doctor's medication scheme, ensures the effectiveness and feasibility of the insulin recommendation scheme, and greatly improves the treatment effect and diagnosis efficiency;

[0060] 4. The present invention fully considers individual differences among users, establishes a data file dedicated to each user, comprehensively combines the actual situation of the user and provides a one-to-one insulin recommendation plan, which can effectively control blood sugar and reduce the occurrence of adverse pregnancy outcomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0062] Figure 1 A flowchart of an intelligent insulin recommendation method according to the present invention;

[0063] Figure 2 It is a structural schematic diagram of an intelligent insulin recommendation system of the present invention;

[0064] Figure 3 Schematic diagram of the connection structure between the basic indicator extraction module, training module 1 and screening matching module. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0066] Example 1

[0067] When recommending an insulin regimen for a user, the user's basic clinical indicators (such as age, gestational age, body mass index, gestational weight, blood lipid level, glycosylated hemoglobin, urine ketone bodies, blood sugar control target, blood sugar level, diabetes classification) are used; according to the method of the present invention, an insulin recommendation regimen is automatically given to the user, including daily insulin type, usage time, injection method, initial dose, dose adjustment regimen and injection site rotation suggestions, which are sent to the user after review. The specific recommendation method is as follows:

[0068] like Figure 1 As shown, an intelligent insulin recommendation method includes:

[0069] S1. Acquire a number of historical clinical case data from a database, each of which includes a number of historical clinical indicators and their corresponding historical clinical diagnosis results.

[0070] S2. Extracting the historical clinical indicators associated with the newly input clinical indicators from the plurality of historical clinical medical record data as associated clinical indicators.

[0071] The historical clinical cases contained in the database of the present invention include all clinical indicators of patients with hyperglycemia during pregnancy. By inputting new clinical indicators, corresponding historical clinical cases can be obtained.

[0072] Read several historical clinical case data in the database, and compare them with the newly input clinical indicators to find the historical clinical case data associated with the newly input clinical indicators. The newly input clinical indicators refer to the clinical indicators that the user needs to enter, including the values ​​corresponding to the indicators, such as blood sugar after lunch and its corresponding values ​​or urine ketones and whether it is positive; enter these indicators into the database, search for the historical clinical indicators corresponding to the input clinical indicators, and extract them to achieve the purpose of association, and the clinical indicators extracted from the database are the associated clinical indicators associated with the newly input clinical indicators.

[0073] S3. Calculate the first correlation coefficient between the associated clinical indicator and the corresponding historical clinical diagnosis result, and extract the associated clinical indicator whose first correlation coefficient is greater than the first threshold as the basic clinical indicator. The correlation coefficient is a prior art and will not be described here. The first threshold here is 0.7. Since there are many associated clinical indicators extracted, the values ​​or contents of each associated clinical indicator and its corresponding clinical diagnosis result are similar or different, so it is necessary to select those with higher correlation coefficients (correlation coefficients greater than 0.7) as basic clinical indicators.

[0074] In this embodiment, the basic clinical indicators include age, gestational age, gestational weight, body mass index, insulin sensitivity coefficient, carbohydrate coefficient, diabetes classification, diabetes course, blood routine test data, blood sugar level, blood lipid level, history of hypoglycemia, diabetic complications, combined use of hypoglycemic drugs, insulin secretion function, creatinine-blood, creatinine-urine, alanine aminotransferase, aspartate aminotransferase, blood pressure, diabetes symptoms and insulin type, insulin use time, insulin injection method, initial total insulin dose, initial total daily insulin dose, daily increase in insulin dose, and insulin injection site.

[0075] For the aforementioned basic clinical indicators, when extracting associated clinical indicators, the indicator and its corresponding value or the indicator and whether it is positive will be input. For example, when extracting associated clinical indicators, indicators represented by numbers such as age and carbohydrate coefficient are directly input as numbers, that is, the input is the indicator and its corresponding description method.

[0076] S4, using the basic clinical indicators and their corresponding historical clinical diagnosis results as training objects for training one by one, so as to obtain prediction results of insulin type, insulin use time, and insulin injection method corresponding to each training object;

[0077] In this embodiment, the specific steps of S4 are:

[0078] S4.1. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the logistic regression model for training, and output a first prediction result for predicting the type of insulin, insulin usage time, and insulin injection method. The first prediction result includes the type of insulin, insulin usage time, and insulin injection method.

[0079] The training process is as follows: collect the feature vector x of the new sample (basic clinical indicators and their corresponding historical clinical diagnosis results), substitute it into the logistic regression model to find the decision function and The probability value of , where w is the logistic regression model parameter, select the item with a large probability, and the value of this item y is the label of the new sample, that is, insulin type, insulin use time, and insulin injection method.

[0080] S4.2. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the support vector machine model for training, and output a second prediction result for predicting the type of insulin, insulin usage time, and insulin injection method. The second prediction result includes the type of insulin, insulin usage time, and insulin injection method.

[0081] The training process is: collect the feature vector x of the new sample i(basic clinical indicators and their corresponding historical clinical diagnosis results), substituted into the support vector machine model, and the prediction formula

[0082]

[0083] Calculate the second prediction result, y i The values ​​of insulin, namely, the type of insulin, the duration of insulin use, and the method of insulin injection.

[0084] S4.3. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the K nearest neighbor classification model for training, and output a third prediction result for predicting the type of insulin, insulin usage time, and insulin injection method. The third prediction result includes the type of insulin, insulin usage time, and insulin injection method.

[0085] The training process is as follows: input the test samples (basic clinical indicators and their corresponding historical clinical diagnosis results) into the K nearest neighbor classification model, and find multiple training samples in the training set that are closest to the test samples based on the Euclidean distance. The number of multiple training samples here is set according to actual conditions, and then predict the type of insulin, insulin usage time, and insulin injection method based on the information of these multiple training samples.

[0086] S4.4. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and output a fourth prediction result for predicting the type of insulin, insulin usage time, and insulin injection method. The fourth prediction result includes the type of insulin, insulin usage time, and insulin injection method.

[0087] The training process is as follows: the test samples (basic clinical indicators and their corresponding historical clinical diagnosis results) are input into the deep deterministic policy gradient model, and the output is generated through a multi-layer deep neural network composed of multiple neurons connected with weights through step function processing on neurons, and finally the type of insulin, insulin usage time, and insulin injection method are obtained.

[0088] S5, calculating the second correlation coefficient between each training object and its corresponding prediction result, extracting the prediction result whose second correlation coefficient is greater than the second threshold, establishing a first set of prediction results and corresponding historical clinical case data, and then comparing and combining the first set with the reference file to locate and obtain the first-level index of the insulin recommendation scheme, wherein the first-level index of the insulin recommendation scheme includes insulin type, insulin use time, and insulin injection method. The second threshold here is also 0.7.

[0089] In this embodiment, the specific implementation steps of S5 are:

[0090] S5.1, respectively calculating the second correlation coefficient between each of the training objects and its corresponding first prediction result, second prediction result, third prediction result, and fourth prediction result;

[0091] S5.2, extracting prediction results corresponding to higher second correlation coefficients among a number of second correlation coefficients, and combining the prediction results with corresponding historical clinical case data to form a first set;

[0092] S5.3. Compare and combine the first set with the reference file to locate and obtain the first-level indicators of the insulin recommendation plan.

[0093] Compare the first-level indicators of the insulin recommendation scheme in the first set with the reference documents to search for the same or similar insulin usage, and extract them, so as to locate the first-level indicators of the insulin recommendation scheme in the reference documents. The similarity or sameness here means that the type of insulin, the time of insulin use and the method of insulin injection are the same, and the difference in insulin use time is less than A hours, where A can be 0.1 or other values, which are set according to the actual situation. The reference documents are specifically guidelines for hyperglycemia in pregnancy / expert consensus / systematic reviews, etc. (generally journal documents);

[0094] Specifically, each data combination in the first set consists of the first-level indicators of the insulin recommendation scheme and its corresponding clinical case data.

[0095] In this embodiment, the types of insulin include: mealtime insulin, basal insulin, and mealtime + basal insulin combination. The insulin usage time is divided into eight types according to different injection times: basal insulin 7:00, i.e. b7; basal insulin 16:00, i.e. b16; basal insulin 22:00, i.e. b22; mealtime + basal insulin combination 7:30, i.e. p7:30; mealtime + basal insulin combination 16:30, i.e. p16:30; mealtime insulin 6:30, i.e. s6:30; mealtime insulin 11:30, i.e. s11:30; mealtime insulin 16:30, i.e. s16:30. The insulin injection methods include: subcutaneous injection and intravenous drip.

[0096] S6. Calculate the total initial daily insulin dose in the secondary index of the insulin recommendation scheme, and establish a corresponding insulin dose adjustment scheme based on the reference file, the total initial daily insulin dose combined with the relevant basic clinical indicators and the corresponding historical clinical diagnosis results, to obtain the secondary index of the insulin recommendation scheme, wherein the secondary index of the insulin recommendation scheme includes the total initial daily insulin dose and the insulin dose adjustment scheme;

[0097] In this embodiment, the specific steps of S6 are:

[0098] S6.1. Input the initial total insulin dose in the basic clinical indicators and the historical clinical diagnosis results corresponding to the initial total insulin dose into the deep deterministic policy gradient model for machine learning training, and establish the function of the initial total insulin dose and the initial total daily insulin dose: y=k+R, wherein y represents the initial total daily insulin dose, k represents the initial total insulin dose, ranging from 0.2 to 1.0 U / (kg*d), and R represents the daily increase in insulin dose, ranging from -4 to -2 U or 2 to 4 U, or R≤0.2y. If the patient's blood sugar is stable, R is 0;

[0099] S6.2. Based on the reference file, the basic clinical indicators in each of the clinical medical record data and the corresponding historical clinical diagnosis results are input into the deep deterministic policy gradient model for machine learning training to obtain an insulin dose adjustment plan. The insulin dose adjustment plan refers to allocating the initial total daily insulin dose according to the daily insulin type, insulin use time, and insulin injection method based on the initial total daily insulin dose.

[0100] The training process is as follows: input the test data (basic clinical indicators and their corresponding historical clinical diagnosis results) into the deep deterministic policy gradient model, design a reward function and evaluate the data, verify the effectiveness of the insulin dose, and obtain the feedback reward given by the reward function. Positive rewards are given to those with good effectiveness, and penalties are given to those with poor effectiveness. All information samples are stored in the experience pool. When the number of samples reaches the predetermined number, the samples are randomly extracted, and the parameters of the deep deterministic policy gradient model are updated using gradient descent to continuously adjust the insulin dose.

[0101] S7. Based on the data of the previous insulin injection site, by comparing and combining the reference file and the corresponding basic clinical index and the corresponding historical clinical diagnosis result, a corresponding third-level index of the insulin recommendation scheme is established, wherein the third-level index of the insulin recommendation scheme is an insulin injection site rotation adjustment scheme;

[0102] By comparing with the reference file, the data of the previous insulin injection site is searched and extracted to provide guidance on the next insulin injection site and establish an adjustment plan for the rotation of insulin injection sites.

[0103] In this embodiment, the specific steps of S7 include:

[0104] S7.1. Based on the reference file, the basic clinical indicators in each clinical medical record data and the corresponding historical clinical diagnosis results are input into the deep deterministic strategy gradient model for machine learning training to obtain an insulin injection site rotation adjustment plan.

[0105] The training process is as follows: input the test data (basic clinical indicators and their corresponding historical clinical diagnosis results) into the deep deterministic policy gradient model. Design a reward function and evaluate the data, verify the effectiveness of the injection site, and obtain the feedback reward given by the reward function. Give positive rewards for good effectiveness and penalties for poor effectiveness. All information samples are stored in the experience pool. When the number of samples reaches the predetermined number, the samples are randomly extracted, and the parameters of the model are updated using gradient descent to continuously adjust the injection site.

[0106] The insulin injection site rotation adjustment plan includes the circular rotation method for limb insulin injection and the two-part, four-zone, five-finger rotation method for insulin injection. The circular rotation method for limb insulin injection is as follows: before each injection, the previous insulin injection site is determined to select the injection site for this time. The circle can be clockwise or counterclockwise. In the two-part, four-zone, five-finger rotation method for insulin injection, the two parts refer to the lower limbs and upper arms; the four zones refer to the left lower limb, right lower limb, left upper arm, and right upper arm; the five fingers refer to the five fingers of the operator.

[0107] All of the above areas can be injected without a mandatory order of priority. The circular rotation method and the two-part, four-zone, five-finger rotation plan must be used in combination to ensure that the same injection point is not used repeatedly within one month.

[0108] Recommended injection method: Inject the same site (such as the left lower limb) at the same time every day. Each injection point should be at least 1 cm away from the last injection point, that is, the circular rotation method. Rotate the injection site about every week (such as from the left lower limb to the right lower limb), that is, the two-part, four-zone, five-finger rotation plan.

[0109] S8, training the insulin recommendation scheme primary index, insulin recommendation scheme secondary index, insulin recommendation scheme tertiary index and their corresponding historical clinical diagnosis results one by one as training objects, so as to obtain the insulin recommendation scheme prediction results corresponding to each training object;

[0110] In this embodiment, the specific implementation steps of S8 are:

[0111] S8.1. Input the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, the third-level indicators of the insulin recommendation scheme and their corresponding historical clinical diagnosis results as input parameters into the logistic regression model for training, and output the fifth prediction result for predicting the insulin recommendation scheme, wherein the fifth prediction result includes the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, and the third-level indicators of the insulin recommendation scheme.

[0112] The training process is as follows: collect the feature vector x of the new sample (first-level indicator of insulin recommendation scheme, second-level indicator of insulin recommendation scheme, third-level indicator of insulin recommendation scheme and its corresponding historical clinical diagnosis results), substitute it into the logistic regression model to find the decision function and The probability value of , where w is the logistic regression model parameter, select the item with a large probability. The value of this item y is the label of the new sample, that is, the insulin recommendation plan.

[0113] S8.2. Input the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, the third-level indicators of the insulin recommendation scheme and their corresponding historical clinical diagnosis results as input parameters into the support vector machine model for training, and output the sixth prediction result for predicting the insulin recommendation scheme. The sixth prediction result includes the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, and the third-level indicators of the insulin recommendation scheme.

[0114] The training process is: collect the feature vector x of the new sample i (first-level indicators of insulin recommendation scheme, second-level indicators of insulin recommendation scheme, third-level indicators of insulin recommendation scheme and their corresponding historical clinical diagnosis results) are substituted into the support vector machine model, and the prediction formula is used.

[0115]

[0116] The sixth prediction result, y, is calculated i The value of is the recommended insulin regimen.

[0117] S8.3. Input the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, the third-level indicators of the insulin recommendation scheme and their corresponding historical clinical diagnosis results as input parameters into the K nearest neighbor classification model for training, and output the seventh prediction result for predicting the insulin recommendation scheme. The seventh prediction result includes the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, and the third-level indicators of the insulin recommendation scheme.

[0118] The training process is as follows: input the test samples (first-level indicators of insulin recommendation scheme, second-level indicators of insulin recommendation scheme, third-level indicators of insulin recommendation scheme and their corresponding historical clinical diagnosis results) into the K nearest neighbor classification model, and find multiple training samples in the training set that are closest to the test samples based on the Euclidean distance. The number of multiple training samples here is set according to actual conditions, and then the insulin recommendation scheme is predicted based on the information of these multiple training samples.

[0119] S8.4. Input the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, the third-level indicators of the insulin recommendation scheme and their corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and output the eighth prediction result for predicting the insulin recommendation scheme. The eighth prediction result includes the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, and the third-level indicators of the insulin recommendation scheme.

[0120] The training process is as follows: the test samples (first-level indicators of insulin recommendation scheme, second-level indicators of insulin recommendation scheme, third-level indicators of insulin recommendation scheme and their corresponding historical clinical diagnosis results) are input into the deep deterministic policy gradient model, and the output is generated through a multi-layer deep neural network composed of multiple neurons connected with weights through step function processing on neurons, and finally the insulin recommendation scheme is obtained.

[0121] S9. Storing and outputting the insulin recommendation plan, wherein the insulin recommendation plan includes a first-level indicator of the insulin recommendation plan, a second-level indicator of the insulin recommendation plan, and a third-level indicator of the insulin recommendation plan.

[0122] Example 2

[0123] like Figure 2-3 As shown, the present invention also discloses an intelligent insulin recommendation system, comprising:

[0124] Database, used to store historical clinical medical record data;

[0125] An input module, for receiving new input clinical indicators;

[0126] A data acquisition module, used for acquiring the historical clinical case data from a database;

[0127] A related index extraction module, used for comparing the new clinical index inputted in the input module with the historical clinical case data acquired by the data acquisition module, and acquiring the historical clinical index associated with the new clinical index as the related clinical index;

[0128] A basic index extraction module, used for calculating a first correlation coefficient between the associated clinical index and the corresponding historical clinical diagnosis result, and extracting the associated clinical index with a higher first correlation coefficient;

[0129] The training module 1 is used to train the basic clinical indicators and their corresponding historical clinical diagnosis results one by one as training objects, so as to obtain the prediction results of the insulin type, insulin use time and insulin injection method corresponding to each training object.

[0130] Specifically, the training module 1 includes a first training unit, a second training unit, a third training unit, and a fourth training unit.

[0131] The first training unit is used to input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the logistic regression model for training, and output a first prediction result for predicting the type of insulin, the duration of insulin use, and the method of insulin injection;

[0132] The second training unit is used to input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the support vector machine model for training, and output a second prediction result for predicting the type of insulin, insulin use time, and insulin injection method;

[0133] The third training unit is used to input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the K nearest neighbor classification model for training, and output a third prediction result for predicting the type of insulin, the duration of insulin use, and the method of insulin injection;

[0134] The fourth training unit is used to input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and output a fourth prediction result for predicting the type of insulin, insulin usage time, and insulin injection method.

[0135] A matching screening module is used to calculate the second correlation coefficient between each training object and its corresponding prediction result, and extract the prediction result with a higher second correlation coefficient, establish a first set of prediction results and corresponding clinical case data, and then compare and combine the first set with the reference file to locate the first-level indicator of the insulin recommendation plan;

[0136] The initial total dose module is used to calculate the initial total daily insulin dose in the secondary indicators of the insulin recommendation plan;

[0137] An insulin dosage adjustment module, used for adjusting the dosage of insulin injection based on the initial total daily insulin dosage;

[0138] An injection site rotation module is used to obtain the injection site of each insulin injection, and the injection site of the next injection is different from the previous injection;

[0139] Training module two is used to train the first-level indicators of the insulin recommendation scheme, the second-level indicators of the insulin recommendation scheme, the third-level indicators of the insulin recommendation scheme and their corresponding historical clinical diagnosis results as training objects one by one to obtain the prediction results of the insulin recommendation scheme corresponding to each training object; the training method of training module two is the same as the training method of training module one, which will not be repeated here.

[0140] The memory bank is used to store the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, the third-level index of the insulin recommendation scheme, and the training process of obtaining the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme.

[0141] Specifically, the memory bank is respectively connected to the training module one, the training module two, the screening and matching module, the initial total dose module, the insulin dose adjustment module, and the injection site rotation module, and is used to store the parameters of the training module one, the training module two, the screening set module, the matching module, the initial total dose module, the insulin dose adjustment module, and the injection site rotation module, as well as the training process of the parameters.

[0142] The prediction module uses the first-level index, second-level index and third-level index of the stored insulin recommendation scheme as the insulin recommendation scheme, uses artificial intelligence technology to predict the effect of artificial blood sugar control, and matches it with the blood sugar control target value to complete a recommendation;

[0143] The indicators entered by the user are input, and the computer simulates the physiological environment after the user uses the insulin recommendation plan generated according to the above process of the present invention, compares it with the normal physiological environment, and predicts the feasibility of the insulin recommendation plan.

[0144] The review module combines the insulin recommendation scheme and the prediction result of the prediction module with the expert review opinions to complete the secondary recommendation;

[0145] The output module is used to output the insulin recommendation plan related to the newly input clinical indicators and the expert review opinions.

[0146] Example 3

[0147] The present invention also discloses an electronic device, comprising at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute an intelligent insulin recommendation method as described above.

[0148] Example 4

[0149] The present invention also discloses a storage medium storing a computer program, wherein the computer program realizes the above-mentioned intelligent insulin recommendation method when executed by a processor.

[0150] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent insulin recommendation method, characterized in that: include: S1. Obtaining a number of historical clinical case data in a database, each of the historical clinical case data includes a number of historical clinical indicators and their corresponding historical clinical diagnosis results; S2, extracting the historical clinical indicators associated with the newly input clinical indicators from the plurality of historical clinical medical record data as associated clinical indicators; S3, calculating a first correlation coefficient between the associated clinical indicator and the corresponding historical clinical diagnosis result, and extracting the associated clinical indicator whose first correlation coefficient is greater than a first threshold as a basic clinical indicator; S4, using the basic clinical indicators and their corresponding historical clinical diagnosis results as training objects for training one by one, so as to obtain prediction results of insulin type, insulin use time, and insulin injection method corresponding to each training object; The specific steps of S4 are: S4.

1. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the logistic regression model for training, and output a first prediction result for predicting the type of insulin, the duration of insulin use, and the method of insulin injection, wherein the first prediction result includes the type of insulin, the duration of insulin use, and the method of insulin injection; S4.2, inputting the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the support vector machine model for training, and outputting a second prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the second prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection; S4.3, inputting the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the K nearest neighbor classification model for training, and outputting a third prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the third prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection; S4.

4. Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and output a fourth prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the fourth prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection; S5, calculating the second correlation coefficient between each training object and its corresponding prediction result, and extracting the prediction results whose second correlation coefficient is greater than the second threshold, establishing a first set of prediction results and corresponding historical clinical case data, and then comparing and combining the first set with the reference file to locate and obtain the first-level indicator of the insulin recommendation scheme, wherein the first-level indicator of the insulin recommendation scheme includes insulin type, insulin use time, and insulin injection method; The specific implementation steps of S5 are: S5.1, respectively calculating the second correlation coefficient between each of the training objects and its corresponding first prediction result, second prediction result, third prediction result, and fourth prediction result; S5.

2. Extract prediction results corresponding to second correlation coefficients greater than the second threshold value from among the plurality of second correlation coefficients, and combine the prediction results with corresponding historical clinical case data to form a first set; S5.

3. Compare and combine the first set with the reference file to locate and obtain the first-level indicator of the insulin recommendation plan; S6. Calculate the total initial daily insulin dose in the secondary index of the insulin recommendation scheme, and establish a corresponding insulin dose adjustment scheme based on the reference file, the total initial daily insulin dose combined with the relevant basic clinical indicators and the corresponding historical clinical diagnosis results, to obtain the secondary index of the insulin recommendation scheme, wherein the secondary index of the insulin recommendation scheme includes the total initial daily insulin dose and the insulin dose adjustment scheme; S7. Based on the data of the previous insulin injection site, by comparing and combining the reference file and the corresponding basic clinical index and the corresponding historical clinical diagnosis result, a corresponding three-level index of the insulin recommendation scheme is established, wherein the three-level index of the insulin recommendation scheme is an insulin injection site rotation adjustment scheme; S8, training the insulin recommendation scheme primary index, insulin recommendation scheme secondary index, insulin recommendation scheme tertiary index and their corresponding historical clinical diagnosis results one by one as training objects, so as to obtain the prediction result of the insulin recommendation scheme corresponding to each training object; S9. Storing and outputting the insulin recommendation plan, wherein the insulin recommendation plan includes a first-level indicator of the insulin recommendation plan, a second-level indicator of the insulin recommendation plan, and a third-level indicator of the insulin recommendation plan.

2. An intelligent insulin recommendation method according to claim 1, characterized in that: In S3, the basic clinical indicators include age, gestational age, gestational weight, body mass index, insulin sensitivity coefficient, carbohydrate coefficient, diabetes classification, diabetes course, blood routine test data, blood sugar level, blood lipid level, history of hypoglycemia, diabetic complications, combined use of hypoglycemic drugs, insulin secretion function, creatinine-blood, creatinine-urine, alanine aminotransferase, aspartate aminotransferase, blood pressure, diabetes symptoms and insulin type, insulin use time, insulin injection method, initial total insulin dose, initial total daily insulin dose, daily increase in insulin dose, and insulin injection site.

3. An intelligent insulin recommendation method according to claim 1, characterized in that: The types of insulin include: mealtime insulin, basal insulin, and mealtime + basal insulin combination. The insulin usage times include: basal insulin 7:00, i.e. b7; basal insulin 16:00, i.e. b16; basal insulin 22:00, i.e. b22; mealtime + basal insulin combination 7:30, i.e. p7:30; mealtime + basal insulin combination 16:30, i.e. p16:30; mealtime insulin 6:30, i.e. s6:30; mealtime insulin 11:30, i.e. s11:30; mealtime insulin 16:30, i.e. s16:

30. The insulin injection methods include: subcutaneous injection and intravenous drip.

4. The intelligent insulin recommendation method according to claim 1, characterized in that: The specific steps of S6 are: S6.

1. Input the initial total insulin dose in the basic clinical indicators and the historical clinical diagnosis results corresponding to the initial total insulin dose into the deep deterministic policy gradient model for machine learning training, and establish the function of the initial total insulin dose and the initial total daily insulin dose: y=k+R, wherein y represents the initial total daily insulin dose, k represents the initial total insulin dose, ranging from 0.2 to 1.0 U / (kg*d), R represents the daily increase in insulin dose, ranging from -4 to -2 U or 2 to 4 U, or R≤0.2y. If the patient's blood sugar is stable, R is 0; S6.

2. Based on the reference file, the basic clinical indicators in each clinical medical record data and the corresponding historical clinical diagnosis results are input into the deep deterministic policy gradient model for machine learning training to obtain an insulin dosage adjustment plan.

5. The intelligent insulin recommendation method according to claim 1, characterized in that: The specific implementation steps of S7 include: S7.

1. Based on the reference file, the basic clinical indicators in each clinical medical record data and the corresponding historical clinical diagnosis results are input into the deep deterministic strategy gradient model for machine learning training to obtain an insulin injection site rotation adjustment plan.

6. The intelligent insulin recommendation method according to claim 1, characterized in that: The specific implementation steps of S8 include: S8.

1. Input the insulin recommendation regimen and its corresponding historical clinical diagnosis results as input parameters into the logistic regression model for training, and output a fifth prediction result for predicting the first-level index of the insulin recommendation regimen, the second-level index of the insulin recommendation regimen, and the third-level index of the insulin recommendation regimen, wherein the fifth prediction result includes the first-level index of the insulin recommendation regimen, the second-level index of the insulin recommendation regimen, and the third-level index of the insulin recommendation regimen; S8.2, inputting the insulin recommendation scheme and its corresponding historical clinical diagnosis results as input parameters into the support vector machine model for training, and outputting a sixth prediction result for predicting the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme, wherein the sixth prediction result includes the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme; S8.3, input the insulin recommendation scheme and its corresponding historical clinical diagnosis results as input parameters into the K nearest neighbor classification model for training, and output a seventh prediction result for predicting the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme, wherein the seventh prediction result includes the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme; S8.

4. Input the insulin recommendation scheme and its corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and output the eighth prediction result for predicting the primary indicators of the insulin recommendation scheme, the secondary indicators of the insulin recommendation scheme, and the tertiary indicators of the insulin recommendation scheme. The eighth prediction result includes the primary indicators of the insulin recommendation scheme, the secondary indicators of the insulin recommendation scheme, and the tertiary indicators of the insulin recommendation scheme.

7. An intelligent insulin recommendation system, characterized in that: include: Database, used to store historical clinical medical record data; An input module, for receiving new input clinical indicators; A data acquisition module, used for acquiring the historical clinical case data from a database; A related index extraction module, used for comparing the new clinical index inputted in the input module with the historical clinical case data acquired by the data acquisition module, and acquiring the historical clinical index associated with the new clinical index as the related clinical index; A basic index extraction module, used for calculating a first correlation coefficient between the associated clinical index and the corresponding historical clinical diagnosis result, and extracting the associated clinical index whose first correlation coefficient is greater than a first threshold as a basic clinical index; Training module 1, for training the basic clinical indicators and their corresponding historical clinical diagnosis results one by one as training objects, so as to obtain prediction results of insulin type, insulin use time and insulin injection method corresponding to each training object; The specific process of obtaining the prediction results is as follows: Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into a logistic regression model for training, and output a first prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the first prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection; Inputting the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into a support vector machine model for training, and outputting a second prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the second prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection; Input the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into a K-nearest neighbor classification model for training, and output a third prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the third prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection; Inputting the basic clinical indicators and their corresponding historical clinical diagnosis results as input parameters into the deep deterministic policy gradient model for training, and outputting a fourth prediction result for predicting the type of insulin, the time of insulin use, and the method of insulin injection, wherein the fourth prediction result includes the type of insulin, the time of insulin use, and the method of insulin injection; A screening matching module is used to calculate a second correlation coefficient between each training object and its corresponding prediction result, and extract the prediction results whose second correlation coefficient is greater than a second threshold, establish a first set of prediction results and corresponding clinical case data, and then compare and combine the first set with the reference file to locate and obtain the first-level indicator of the insulin recommendation plan; The specific process of locating the first-level indicators of the insulin recommendation plan is as follows: Respectively calculating the second correlation coefficient between each of the training objects and its corresponding first prediction result, second prediction result, third prediction result, and fourth prediction result; Extracting prediction results corresponding to second correlation coefficients greater than a second threshold value from among a plurality of second correlation coefficients, and combining the prediction results with corresponding historical clinical case data to form a first set; Compare and combine the first set with the reference file to locate and obtain the first-level indicator of the insulin recommendation plan; The initial total dose module is used to calculate the initial total daily insulin dose in the secondary indicators of the insulin recommendation plan; An insulin dosage adjustment module, used for adjusting the dosage of insulin injection based on the initial total daily insulin dosage; An injection site rotation module is used to obtain the injection site of each insulin injection, and the injection site of the next injection is different from the previous injection; Training module 2 is used to train the insulin recommendation scheme primary index, insulin recommendation scheme secondary index, insulin recommendation scheme tertiary index and their corresponding historical clinical diagnosis results as training objects one by one, so as to obtain the insulin recommendation scheme prediction results corresponding to each training object; A memory bank, used to store the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, the third-level index of the insulin recommendation scheme, and the training process of obtaining the first-level index of the insulin recommendation scheme, the second-level index of the insulin recommendation scheme, and the third-level index of the insulin recommendation scheme; The prediction module uses the first-level index, second-level index, and third-level index of the stored insulin recommendation plan as the insulin recommendation plan, uses artificial intelligence technology to predict the blood sugar control effect, and matches it with the blood sugar control target value to complete a recommendation; The review module combines the insulin recommendation scheme and the prediction result of the prediction module with the expert review opinions to complete the secondary recommendation; The output module is used to output insulin recommendation plans and expert review opinions related to the newly input clinical indicators.

8. An electronic device, characterized in that: It includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute an intelligent insulin recommendation method as described in any one of claims 1 to 6.

Images