Diabetes adjuvant therapy cloud platform system based on artificial intelligence model

Abstract

The invention relates to a diabetes adjuvant therapy cloud platform system based on an artificial intelligence model, and relates to the technical field of health-related information systems and cloud platforms, and the diabetes adjuvant therapy cloud platform system comprises a correlation analysis module which is used for determining a plurality of patient classification factors based on medical records of a plurality of historical diabetic patients, determining a plurality of diabetic patient types and the complication risk of each diabetic patient type; the feature extraction module is used for determining the type of the diabetic patient of the current patient according to the medical record of the current patient and the plurality of patient classification factors, and extracting complication risk features from the medical record of the current patient; the auxiliary treatment module is used for determining similar historical diabetic patients according to the complication risk characteristics of the current patient and determining an auxiliary treatment scheme of the current patient based on the auxiliary treatment scheme of the similar historical diabetic patients and the complication risk characteristics of the current patient, and the diabetes diagnosis and treatment efficiency and the treatment effect are improved.

Description

Technical Field

[0001] This invention relates to the field of health-related information systems and cloud platform technology, and in particular to a cloud platform system for diabetes adjuvant treatment based on an artificial intelligence model. Background Technology

[0002] Diabetes mellitus is a metabolic disease characterized by chronic hyperglycemia. Its essence lies in insufficient insulin secretion or impaired insulin action (insulin resistance), leading to metabolic disorders of nutrients such as carbohydrates, fats, and proteins. Over the long term, it can cause multi-system damage, affecting organs and tissues including the eyes, kidneys, nerves, heart, and blood vessels. The harm of diabetes extends beyond abnormal blood sugar levels; its complications involve multiple organ systems, including the cardiovascular, renal, retinal, and nervous systems. For example, diabetic retinopathy is the leading cause of blindness in working-age individuals, and diabetic nephropathy accounts for more than 30% of end-stage renal disease cases worldwide.

[0003] Existing treatment guidelines are mostly based on population data and do not fully consider individual patient differences. For example, among type 2 diabetes patients, obese and lean individuals have different insulin resistance mechanisms, resulting in a three-fold difference in the efficacy of metformin; elderly patients with declining liver and kidney function need to reduce the dosage of sulfonylureas by 50%. However, due to a lack of individualized analysis tools, primary care physicians often use a "one-size-fits-all" medication regimen, resulting in a blood glucose control rate of less than 30%.

[0004] Therefore, there is a need to provide a cloud platform system for diabetes assisted treatment based on artificial intelligence models to improve the efficiency and treatment effect of diabetes diagnosis and treatment. Summary of the Invention

[0005] This invention provides a cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model, comprising: a correlation analysis module for acquiring medical records of multiple historical diabetes patients, determining multiple patient classification factors based on these records, determining multiple diabetes patient types based on the patient classification factors and the medical records of multiple historical diabetes patients, and determining the complication risk for each diabetes patient type based on the medical records of multiple historical diabetes patients; a feature extraction module for determining the current patient's diabetes patient type based on the current patient's medical records and multiple patient classification factors, and extracting complication risk features from the current patient's medical records based on the complication risk risk of the current patient's diabetes patient type; and an adjuvant therapy module for identifying similar historical diabetes patients among the historical diabetes patients included in the current patient's diabetes patient type based on the current patient's complication risk features, and determining the current patient's adjuvant therapy plan based on the adjuvant therapy plans of similar historical diabetes patients and the current patient's complication risk features.

[0006] Furthermore, the association analysis module identifies multiple patient classification factors based on the medical records of multiple historical diabetic patients, including: identifying multiple candidate factors and multiple diabetes diagnostic factors; for each candidate factor and each diabetes diagnostic factor, based on the medical records of multiple historical diabetic patients, determining the feature values ​​of the candidate factor corresponding to each historical diabetic patient and the feature values ​​of the diabetes diagnostic factor corresponding to each historical diabetic patient, and calculating the correlation coefficient between the candidate factor and the diabetes diagnostic factor; for each candidate factor, calculating the classification key value of the candidate factor based on the correlation coefficient between the candidate factor and each diabetes diagnostic factor; and based on the classification key value of each candidate factor, identifying multiple patient classification factors from the multiple candidate factors.

[0007] Furthermore, the association analysis module determines multiple types of diabetes patients based on multiple patient classification factors and medical records of multiple historical diabetes patients. This includes: for each historical diabetes patient, extracting a type feature vector based on the patient's medical record and multiple patient classification factors, wherein the type feature vector consists of the feature values ​​of the historical diabetes patient corresponding to each patient classification factor; calculating the vector distance between the type feature vectors of any two historical diabetes patients; grouping the multiple patient classification factors using a clustering algorithm based on the vector distance between the type feature vectors of any two historical diabetes patients; and determining multiple types of diabetes patients based on the grouping results.

[0008] Furthermore, the association analysis module determines the complication risk for each type of diabetes patient based on the medical records of multiple historical diabetes patients, including: determining the incidence probability of multiple complications corresponding to each type of diabetes patient based on the medical records of multiple historical diabetes patients included in the diabetes patient type; determining the first complication of the diabetes patient type based on the incidence probability of multiple complications corresponding to each type of diabetes patient; determining the co-occurrence probability of each other complication with each first complication of the diabetes patient type based on the medical records of multiple historical diabetes patients included in the diabetes patient type; and determining the second complication of the diabetes patient type based on the co-occurrence probability of each other complication with each first complication of the diabetes patient type, wherein the complication risk of the diabetes patient type includes both the first and second complications.

[0009] Furthermore, the feature extraction module determines the current patient's diabetes patient type based on the current patient's medical history and multiple patient classification factors, including: for each diabetes patient type, determining a type feature vector of the diabetes patient type based on the type feature vectors of historical diabetes patients included in the diabetes patient type; determining key patient classification factors for each diabetes patient type based on the type feature vectors of each diabetes patient type; generating key type feature vectors for each diabetes patient type based on the key patient classification factors and the type feature vectors of each diabetes patient type; generating key type feature vectors for each diabetes patient type corresponding to the current patient based on the current patient's medical history and the key patient classification factors for each diabetes patient type; and determining the current patient's diabetes patient type based on the key type feature vectors for each diabetes patient type and the key type feature vectors for each diabetes patient type corresponding to the current patient.

[0010] Furthermore, based on the type feature vector of each type of diabetes patient, the key patient classification factors for each type of diabetes patient are determined, including: for each patient classification factor, calculating the discrete value of the patient classification factor corresponding to each type of diabetes patient based on the type feature vector of each type of diabetes patient; determining the initial patient classification factor for each type of diabetes patient based on the discrete value of each patient classification factor corresponding to each type of diabetes patient; establishing an objective function, wherein the objective function is related to the discrete value of each patient classification factor corresponding to each type of diabetes patient; and iteratively optimizing the objective function and the initial patient classification factor for each type of diabetes patient until the optimization conditions are met, thereby determining the key patient classification factors for each type of diabetes patient.

[0011] Furthermore, the feature extraction module extracts complication risk features from the current patient's medical record based on the complication risk of the current patient's diabetes type, including: identifying key diagnostic factors for each complication; and extracting complication risk features from the current patient's medical record based on the key diagnostic factors for the complication risk of the current patient's diabetes type.

[0012] Furthermore, the auxiliary treatment module identifies similar historical diabetes patients from among those with a history of diabetes included in the current patient's diabetes patient type based on the current patient's complication risk characteristics, including: calculating the complication risk similarity between the current patient's complication risk characteristics and the complication risk characteristics of those with a history of diabetes included in the current patient's diabetes patient type; and identifying similar historical diabetes patients from among those with a history of diabetes included in the current patient's diabetes patient type based on the complication risk similarity.

[0013] Furthermore, the adjunctive therapy module determines the adjunctive therapy plan for the current patient based on the adjunctive therapy plans of patients with similar historical diabetes and the current patient's complication risk characteristics. This includes: determining the adjunctive therapy plan for the current patient through an adjunctive therapy model based on the adjunctive therapy plans of patients with similar historical diabetes, complication risk characteristics, key type feature vectors, the current patient's complication risk characteristics, and key type feature vectors.

[0014] Furthermore, the auxiliary treatment model includes an input layer, a first feature extraction branch, a second feature extraction branch, a third feature extraction branch, a feature fusion layer, and a fully connected unit. The first feature extraction branch is used to extract features of auxiliary treatment plans for patients with similar historical diabetes. The second feature extraction branch is used to perform feature extraction after concatenating the complication risk features and key type feature vectors of patients with similar historical diabetes. The third feature extraction branch is used to perform feature extraction after concatenating the complication risk features and key type feature vectors of the current patient. The feature fusion layer is used to fuse the outputs of the first, second, and third feature extraction branches. The fully connected unit is used to output the auxiliary treatment plan for the current patient.

[0015] Compared to existing technologies, the cloud platform system for diabetes adjuvant treatment based on an artificial intelligence model provided in this specification has at least the following beneficial effects: 1. The correlation analysis module extracts patient classification factors from multi-dimensional medical record data and categorizes patients, accurately identifying individual characteristics of diabetic patients and quantifying the risk of various types of complications, providing a scientific basis for personalized treatment. The feature extraction module, based on patient type and risk, selectively extracts key complication risk features from medical records, avoiding irrelevant information interference and improving the correlation between features and treatment plans. The auxiliary treatment module matches similar historical patients, integrates their effective treatment plans with current patient risk characteristics, and generates treatment strategies tailored to individual needs, improving treatment effectiveness and safety. 2. By calculating the correlation coefficients and classification key values ​​between candidate factors and diabetes diagnostic factors, key factors with high contribution to classification are accurately screened from massive amounts of data, improving the accuracy and reliability of patient classification. Based on vector distance and clustering algorithms using type feature vectors, the characteristics and patterns of patient groups are automatically mined, achieving objective and differentiated classification of diabetes patient types, laying the foundation for personalized treatment; 3. By calculating the similarity of complication risks, individuals with highly matched disease characteristics to the current patient can be quickly screened from historical patients of the same type, improving the targeting of treatment references. The auxiliary treatment model integrates historical treatment plans, risk characteristics, and type characteristics. Through multi-branch feature extraction and fusion, it comprehensively captures individual patient differences and treatment patterns, enhancing the scientific rigor of the plan. Based on the experience of similar patients and the real-time characteristics of the current patient, the model automatically outputs highly adaptable auxiliary treatment plans, reducing human intervention bias and improving treatment efficiency and accuracy. Attached Figure Description

[0016] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a block diagram of a cloud platform system for diabetes adjuvant treatment based on an artificial intelligence model, as shown in one embodiment of this application; Figure 2 This is a flowchart illustrating key patient classification factors for determining each type of diabetes patient in one embodiment of this application; Figure 3 This is a structural diagram of an adjunctive therapy model shown in one embodiment of this application. Detailed Implementation

[0017] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0018] Figure 1 This is a block diagram of a cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model, as shown in one embodiment of this application. This system belongs to the category of health-related information systems and cloud platforms, such as... Figure 1 As shown, the cloud platform system for diabetes assisted treatment based on artificial intelligence models can include a correlation analysis module, a feature extraction module, and an assisted treatment module.

[0019] The association analysis module can be used to obtain the medical records of multiple patients with a history of diabetes, determine multiple patient classification factors based on the medical records of multiple patients with a history of diabetes, determine multiple types of diabetes patients based on the multiple patient classification factors and the medical records of multiple patients with a history of diabetes, and determine the risk of complications for each type of diabetes patient based on the medical records of multiple patients with a history of diabetes.

[0020] Specifically, the correlation analysis module needs to integrate multi-dimensional data such as hospital electronic medical records, wearable device monitoring data (e.g., blood glucose fluctuation curves), and genetic testing reports. For example, by establishing a diabetes-specific database, outpatient records, inpatient medical records, and follow-up data from 100,000 patients were integrated to form a structured dataset. To address issues such as inconsistent terminology (e.g., "type 2 diabetes" vs. "T2DM") and missing values ​​(e.g., some patients did not record their glycated hemoglobin values), natural language processing technology was used to extract key information, and interpolation or multiple imputation methods were used to handle missing values.

[0021] The medical records of patients with a history of diabetes include structured or unstructured medical records containing multi-dimensional data such as patient demographic information (e.g., age, gender), clinical indicators (e.g., blood glucose, blood pressure, blood lipids), diagnostic information (e.g., type of diabetes, type of complication), and treatment records (e.g., medication history, surgical history).

[0022] In some embodiments, the association analysis module determines multiple patient classification factors based on the medical records of multiple historical diabetes patients, including: Identify multiple candidate factors and multiple diagnostic factors for diabetes; For each candidate factor and each diabetes diagnostic factor, based on the medical records of multiple historical diabetes patients, the characteristic values ​​of the candidate factor and the characteristic values ​​of the diabetes diagnostic factor for each historical diabetes patient are determined, and the correlation coefficient between the candidate factor and the diabetes diagnostic factor is calculated. For each candidate factor, the classification key value of the candidate factor is calculated based on the correlation coefficient between the candidate factor and each diabetes diagnostic factor. Based on the classification key value of each candidate factor, multiple patient classification factors are determined from multiple candidate factors.

[0023] Specifically, candidate factors are variables that are potentially related to diabetes classification extracted from medical records, such as age, sex, body mass index (BMI), smoking history, drinking history, and exercise frequency.

[0024] Diabetes diagnostic factors are standard variables used to define the diagnosis or classification of diabetes, such as blood glucose levels, glycated hemoglobin, insulin, and C-peptide measurements.

[0025] The characteristic values ​​of candidate factors and diabetes diagnostic factors for each patient with a history of diabetes can be the specific values ​​of the candidate factors and diabetes diagnostic factors in a single patient with a history of diabetes. For example: Patient A has a BMI of 28 kg / m², so the characteristic value of BMI is 28.

[0026] Patient B's HbA1c is 9.2%, so the characteristic value of HbA1c is 9.2.

[0027] The correlation coefficient between the candidate factor and the diabetes diagnostic factor can be calculated by substituting the characteristic values ​​of the candidate factor and the diabetes diagnostic factor corresponding to each historical diabetic patient into the formula for calculating the Pearson correlation coefficient or Spearman rank correlation coefficient.

[0028] For each diabetes diagnostic factor, the overall standard deviation of the characteristic values ​​corresponding to each patient with a history of diabetes can be calculated.

[0029] For each candidate factor, the overall standard deviation of the correlation coefficient between the candidate factor and each diabetes diagnostic factor can be calculated.

[0030] Based on the correlation coefficient between candidate factors and each diabetes diagnostic factor, the overall standard deviation of the characteristic values ​​of each historical diabetes patient corresponding to the diabetes diagnostic factor, and the overall standard deviation of the correlation coefficient between candidate factors and each diabetes diagnostic factor, the classification key value of the candidate factors is calculated. For example, the classification key value of candidate factors can be calculated using the following formula:

[0031] in, Let be the classification key value of the i-th candidate factor. Let be the overall standard deviation of the correlation coefficient between the i-th candidate factor and each diabetes diagnostic factor. Let be the overall standard deviation of the characteristic values ​​for each patient with a history of diabetes corresponding to the nth diabetes diagnostic factor. Let be the correlation coefficient between the i-th candidate factor and the n-th diabetes diagnostic factor. This represents the total number of diagnostic factors for diabetes.

[0032] Candidate factors whose classification key values ​​are greater than the classification key value threshold (e.g., 0.2) can be used as patient classification factors.

[0033] Understandably, for each diabetes diagnostic factor, the overall standard deviation of its historical diabetic patient characteristic values ​​is calculated to reflect the dispersion of that factor within the patient population. Greater dispersion indicates a stronger potential impact on classification, thus warranting higher weighting. For each candidate factor, the overall standard deviation of its correlation coefficient with all diabetes diagnostic factors is calculated to measure the volatility of the correlation between the candidate factor and the diagnostic factors. As the overall standard deviation of the correlation coefficient between the i-th candidate factor and each diabetes diagnostic factor increases, the inhibition coefficient monotonically decreases, weakening the score of highly volatile candidate factors. High volatility may be caused by measurement errors or random associations (e.g., a candidate factor is abnormally correlated with diagnostic factors in a small number of patients). The inhibition coefficient reduces the risk of noisy factors being mistakenly selected as classification factors by lowering their scores.

[0034] In some embodiments, the association analysis module determines multiple types of diabetes patients based on multiple patient classification factors and medical records of multiple historical diabetes patients, including: For each patient with a history of diabetes, a type feature vector is extracted based on their medical history and multiple patient classification factors. This type feature vector consists of the feature values ​​corresponding to each patient classification factor for the patient with a history of diabetes. For example, for each patient P... i Extract the feature values ​​of M categorical factors from the medical records to form a type feature vector v. i =(v i1 ,v i2 ,…,v iM ), where v ij Patient P i In classification factor F j The value on; Calculate the vector distance between the type feature vectors of any two historical diabetes patients. For example, calculate the Euclidean distance between the type feature vectors of two historical diabetes patients as the vector distance. Clustering algorithms (e.g., K-means, hierarchical clustering) are used to group multiple patient classification factors based on the vector distance between the type feature vectors of any two historical diabetes patients. Based on the grouping results, various types of diabetes patients are determined.

[0035] Specifically, after clustering is completed, each patient is assigned to a cluster (i.e., a type), and patients within the same cluster have similar type feature vectors.

[0036] In some embodiments, the association analysis module determines the risk of complications for each type of diabetes based on the medical records of multiple historical diabetes patients, including: Based on the medical records of multiple historical diabetic patients included in the diabetic patient types, the incidence probability of various complications corresponding to diabetic patient types was determined. Based on the incidence probability of various complications corresponding to different types of diabetic patients, the primary complication of a diabetic patient type is determined. Based on the medical records of multiple historical diabetic patients included in the diabetic patient type, determine the co-occurrence probability of each other complication with each first complication of the diabetic patient type; The second complication of a diabetes patient type is determined based on the co-occurrence probability of each other complication with each first complication of the diabetes patient type, wherein the complication risk of the diabetes patient type includes both the first and second complications.

[0037] Specifically, based on the incidence rate of complications, the most common complications in patients with a certain type of diabetes are identified.

[0038] For each complication, for each historical diabetic patient included in the diabetic patient type, we count whether they have the complication and calculate the proportion of patients with the complication in the diabetic patient type as the incidence probability of the corresponding diabetic patient type.

[0039] Complications with an incidence rate greater than a threshold (e.g., 50%) are considered the first complication of a diabetes patient type.

[0040] For any other complication and any first complication of a diabetes patient type, calculate the ratio of the number of patients in the diabetes patient type who simultaneously have the other complication and the first complication to the number of patients with the complication. Use this ratio as the co-occurrence probability of the other complication and the first complication. Other complications with a co-occurrence probability greater than a co-occurrence probability threshold (e.g., 0.5) are designated as second complications.

[0041] By calculating the co-occurrence probability of other complications with the first complication, complications strongly associated with the first complication (i.e., "second complications") are identified, thereby constructing a risk chain of "primary concurrency - secondary concurrency".

[0042] The feature extraction module can be used to determine the type of diabetes in the current patient based on the current patient's medical record and multiple patient classification factors, and to extract complication risk features from the current patient's medical record based on the complication risk of the current patient's type of diabetes.

[0043] In some embodiments, the feature extraction module determines the current patient's diabetes patient type based on the current patient's medical history and multiple patient classification factors, including: For each type of diabetes patient, the type feature vector of the diabetes patient type is determined based on the type feature vector of the historical diabetes patients included in the diabetes patient type. For example, the mean of the type feature vectors of the historical diabetes patients included in the diabetes patient type is used as the type feature vector of the diabetes patient type. Based on the type feature vector of each type of diabetes patient, the key patient classification factors for each type of diabetes patient are determined; Based on the key patient classification factors and the type feature vector of each type of diabetes patient, a key type feature vector is generated for each type of diabetes patient. Specifically, for each type of diabetes patient, the values ​​of the elements corresponding to the non-key patient classification factors in the type feature vector of the type of diabetes patient are set to zero, while the values ​​of the elements corresponding to the key patient classification factors are retained. For example, if the type feature vector of the diabetes patient is (1, 2, 3, 4), where the first and third elements are key patient classification factors, then the key type feature vector of the diabetes patient is (1, 0, 3, 0). Based on the current patient's medical records and key patient classification factors for each type of diabetes, key type feature vectors for each type of diabetes are generated for the current patient. Specifically, the method for generating key type feature vectors for each type of diabetes is similar to that for generating key type feature vectors for each type of diabetes, and will not be elaborated here. Based on the key type feature vector of each type of diabetes patient and the key type feature vector of the current patient corresponding to each type of diabetes patient, the diabetes patient type of the current patient is determined.

[0044] Figure 2 This is a flowchart illustrating key patient classification factors for determining each type of diabetes patient in one embodiment of this application, such as... Figure 2 As shown, in some embodiments, key patient classification factors for each type of diabetes patient are determined based on the type feature vector of each type of diabetes patient, including: For each patient classification factor, the discrete value of the patient classification factor corresponding to each type of diabetes patient is calculated based on the type feature vector of each type of diabetes patient. Based on the discrete value of each patient classification factor corresponding to each type of diabetes patient, the initial patient classification factor for each type of diabetes patient is determined. For example, the patient classification factor with a discrete value greater than the discrete value threshold (e.g., 0, 7) is used as the initial patient classification factor for the type of diabetes patient. Establish an objective function, whereby the objective function is related to the discrete values ​​of each patient classification factor corresponding to each type of diabetes patient; The objective function and the initial patient classification factors for each type of diabetes are iteratively optimized until the optimization conditions are met, thereby determining the key patient classification factors for each type of diabetes.

[0045] Specifically, the discrete values ​​of the patient classification factor for each type of diabetes patient can be calculated using the following formula:

[0046] in, Let be the discrete value of the classification factor for the nth patient corresponding to the i-th type of diabetes. Let be the value of the element corresponding to the nth patient classification factor in the type feature vector of the kth type of diabetes patient. Let be the value of the element corresponding to the nth patient classification factor in the type feature vector of the i-th type of diabetes patient. This represents the total number of types of diabetes patients.

[0047] Multiple individuals are initialized based on initial patient classification factors for each type of diabetes patient, wherein each individual includes key patient classification factors for each type of diabetes patient, wherein the key patient classification factors are randomly selected from the initial patient classification factors for each type of diabetes patient.

[0048] For each individual, a key type feature vector of the diabetes patient type is generated based on the key patient classification factor of the diabetes patient type. The objective function is also related to the similarity of the key type feature vectors of any two diabetes patient types.

[0049] For example, the objective function could be:

[0050] in, Let be the objective function. For the kth type of diabetes patient, the corresponding number is... Discrete values ​​of key patient classification factors Let be the total number of key patient classification factors corresponding to the k-th type of diabetes patient. Let i be the key type feature vector of the i-th type of diabetes patient. Let j be the key type feature vector of the j-th type of diabetes patient. Let be the cosine similarity between the key type feature vector of the i-th type of diabetes patient and the key type feature vector of the j-th type of diabetes patient.

[0051] Understandably, the above formula, by calculating the sum of the dispersion of classification factors, encourages significant differences in key classification factors among different patient types. The larger the dispersion value, the greater the contribution of that factor to distinguishing types. During optimization, the formula tends to preserve or enhance the differences in these factors. It also calculates the reciprocal of the feature vector similarity to penalize the similarity of feature vectors from different types, encouraging vector direction differentiation. If the feature vectors of two patient types are similar, this value will be significantly reduced, thus forcing feature vector separation during optimization and improving classification discrimination. This formula maximizes the dispersion of classification factors, making the differences in key classification factors among different patient types significant, enhancing the factor's discriminative power, and minimizes the similarity of feature vectors, ensuring that the feature vector directions of different patient types are significantly different, avoiding type confusion.

[0052] The objective function can be used as the fitness function, and a genetic algorithm can be used to iteratively optimize the objective function until the optimization conditions are met (e.g., the number of iterations reaches the maximum value or the objective function value converges). Based on the individual with the largest objective function value, the key patient classification factor for each type of diabetes patient can be determined.

[0053] Calculate the cosine similarity between the key type feature vector of each type of diabetes patient and the key type feature vector of the current patient corresponding to each type of diabetes patient. The type of diabetes patient with the highest cosine similarity is taken as the type of diabetes patient of the current patient.

[0054] In some embodiments, the feature extraction module extracts complication risk features from the current patient's medical record based on the complication risk of the current patient's diabetes patient type, including: Identify the key diagnostic factors for each complication; Based on the key diagnostic factors for complication risk of the current patient’s diabetes type, complication risk characteristics are extracted from the current patient’s medical records.

[0055] Specifically, key diagnostic factors are core indicators that directly reflect the risk of specific complications. Patients at different risk levels must show significant differences in their values ​​for this factor (e.g., high-risk patients have significantly higher levels of urinary microalbumin than low-risk patients). Statistical analysis (such as chi-square test and logistic regression) is used to screen factors significantly associated with complications from historical medical records as key diagnostic factors for those complications.

[0056] Complication risk characteristics may include feature values ​​of key diagnostic factors for each first and second complication of the current patient’s diabetes patient type, extracted from the current patient’s medical records.

[0057] The adjunctive therapy module can be used to identify patients with similar historical diabetes from among those with a history of diabetes, based on the current patient's complication risk characteristics. Then, based on the adjunctive therapy plans for these patients and the current patient's complication risk characteristics, the module can determine the current patient's adjunctive therapy plan.

[0058] In some embodiments, the adjunctive therapy module identifies patients with similar historical diabetes history from among those with a history of diabetes included in the current patient's diabetes patient type, based on the current patient's complication risk characteristics, including: Calculate the complication risk similarity between the current patient's complication risk characteristics and the complication risk characteristics of historical diabetic patients included in the current patient's diabetes patient type. The method for determining the complication risk characteristics of historical diabetic patients included in the current patient's diabetes patient type is similar to the method for determining the current patient's complication risk characteristics, and will not be repeated here. Based on the similarity of complication risk, identify patients with similar historical diabetes from among those with a history of diabetes included in the current patient's diabetes patient type.

[0059] Specifically, based on the similarity of complication risk, the current patient's diabetes patient type and historical diabetes patients can be sorted, and the top n (e.g., 3, 4, etc.) historical diabetes patients in the sorted list are considered as similar historical diabetes patients.

[0060] In some embodiments, the adjunctive therapy module determines the current patient's adjunctive therapy plan based on adjunctive therapy plans of patients with similar historical diabetes and the current patient's complication risk characteristics, including: The adjuvant therapy model is used to determine the adjuvant therapy plan for the current patient based on the adjuvant therapy plans, complication risk characteristics, and key type feature vectors of patients with similar history of diabetes, as well as the complication risk characteristics and key type feature vectors of the current patient.

[0061] Figure 3 This is a structural diagram of an adjunctive therapy model shown in one embodiment of this application, as follows: Figure 3 As shown, the adjunctive treatment model includes an input layer, a first feature extraction branch, a second feature extraction branch, a third feature extraction branch, a feature fusion layer, and a fully connected unit. The first feature extraction branch is used to extract features of adjunctive treatment plans for patients with similar historical diabetes. The second feature extraction branch is used to perform feature extraction by concatenating the complication risk features and key type feature vectors of patients with similar historical diabetes. The third feature extraction branch is used to perform feature extraction by concatenating the complication risk features and key type feature vectors of the current patient. The feature fusion layer is used to fuse the outputs of the first, second, and third feature extraction branches. The fully connected unit is used to output the adjunctive treatment plan for the current patient.

[0062] Specifically, the first feature extraction branch uses an embedding layer to encode discrete treatment operations (such as drug names) into dense vectors. It captures time-series patterns of treatment regimens (such as medication order and dosage adjustment patterns) through 1D convolutions or LSTMs. It outputs a pattern feature vector of historical treatment regimens (e.g., an encoded representation of "insulin + metformin combination therapy"). The second feature extraction branch learns the interaction between risk features and type features through fully connected layers (MLP) or attention mechanisms (e.g., "elderly type 2 diabetic patients are more sensitive to kidney damage"), outputting a risk-feature association vector for historical patients (e.g., a latent rule representation of "hyperglycemia + abnormal kidney function → preferential use of SGLT-2 inhibitors"). The third feature extraction branch uses a graph neural network (GNN) or Transformer to model complex relationships between features (e.g., the synergistic risk of "elderly blood pressure" and "diabetic retinopathy"). It outputs a personalized risk-feature vector for the current patient (e.g., a specific representation of "current patient: hyperglycemia + mild kidney disease → need to avoid nephrotoxic drugs"). The feature fusion layer assigns branch importance through learnable weights and outputs a fused comprehensive feature vector. The fully connected unit maps the comprehensive feature vector to a specific adjunctive treatment plan by gradually transforming the features through 2-3 fully connected layers (e.g., from 512 dimensions → 256 dimensions → 128 dimensions), and the linear layer outputs continuous values ​​(e.g., insulin dose, exercise duration).

[0063] For example, input data: Current patient: Type 2 diabetes, high cardiovascular risk, HbA1c = 8.5%, blood pressure = 150 / 95 mmHg. Similar patient history: 3 patients with type 2 diabetes, all treated with metformin + SGLT-2 inhibitors, with well-controlled blood pressure.

[0064] Branch output: First feature extraction branch: Extract "metformin + SGLT-2 inhibitor" as a high-frequency treatment regimen.

[0065] The second feature extraction branch discovered the association that "hypertension + hyperglycemia → SGLT-2 inhibitors are more effective".

[0066] The third feature extraction branch: captures the personalized risk of the current patient's "blood pressure not reaching the target".

[0067] Final output: Recommendation: "Metformin (1g / d) + Dapagliflozin (10mg / d) + Antihypertensive medication adjustment", with the basis being "the efficacy of similar patients in history" and "current blood pressure control needs".

[0068] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model, characterized in that, include: The association analysis module is used to obtain the medical records of multiple patients with a history of diabetes, determine multiple patient classification factors based on the medical records of multiple patients with a history of diabetes, determine multiple types of diabetes patients based on the multiple patient classification factors and the medical records of multiple patients with a history of diabetes, and determine the risk of complications for each type of diabetes patient based on the medical records of multiple patients with a history of diabetes. The feature extraction module is used to determine the type of diabetes in the current patient based on the current patient's medical record and multiple patient classification factors, and to extract complication risk features from the current patient's medical record based on the complication risk of the current patient's type of diabetes. The adjunctive therapy module is used to identify patients with similar historical diabetes from among those with a history of diabetes, based on the current patient's complication risk characteristics. Then, based on the adjunctive therapy plans for these patients and the current patient's complication risk characteristics, the module determines the current patient's adjunctive therapy plan.

2. The cloud platform system for diabetes adjuvant treatment based on an artificial intelligence model according to claim 1, characterized in that, The association analysis module identifies multiple patient classification factors based on the medical records of multiple historical diabetes patients, including: Identify multiple candidate factors and multiple diagnostic factors for diabetes; For each candidate factor and each diabetes diagnostic factor, based on the medical records of multiple historical diabetes patients, the characteristic values ​​of the candidate factor and the characteristic values ​​of the diabetes diagnostic factor for each historical diabetes patient are determined, and the correlation coefficient between the candidate factor and the diabetes diagnostic factor is calculated. For each candidate factor, the classification key value of the candidate factor is calculated based on the correlation coefficient between the candidate factor and each diabetes diagnostic factor. Based on the classification key value of each candidate factor, multiple patient classification factors are determined from multiple candidate factors.

3. The cloud platform system for diabetes adjuvant treatment based on an artificial intelligence model according to claim 1, characterized in that, The association analysis module identifies various types of diabetes patients based on multiple patient classification factors and the medical records of multiple historical diabetes patients, including: For each patient with a history of diabetes, a type feature vector is extracted based on the patient's medical history and multiple patient classification factors. The type feature vector consists of the feature values ​​of the patient with a history of diabetes for each patient classification factor. Calculate the vector distance between the type feature vectors of any two historical diabetes patients; By using a clustering algorithm based on the vector distance between the type feature vectors of any two historical diabetes patients, multiple patient classification factors are grouped, and various types of diabetes patients are determined based on the grouping results.

4. The cloud platform system for diabetes adjuvant treatment based on an artificial intelligence model according to claim 1, characterized in that, The association analysis module determines the risk of complications for each type of diabetes patient based on the medical records of multiple historical diabetes patients, including: Based on the medical records of multiple historical diabetic patients included in the diabetic patient types, the incidence probability of various complications corresponding to diabetic patient types was determined. Based on the incidence probability of various complications corresponding to different types of diabetic patients, the primary complication of a diabetic patient type is determined. Based on the medical records of multiple historical diabetic patients included in the diabetic patient type, determine the co-occurrence probability of each other complication with each first complication of the diabetic patient type; The second complication of a diabetes patient type is determined based on the co-occurrence probability of each other complication with each first complication of the diabetes patient type, wherein the complication risk of the diabetes patient type includes both the first and second complications.

5. A cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model according to any one of claims 1-4, characterized in that, The feature extraction module determines the current patient's diabetes type based on the current patient's medical history and multiple patient classification factors, including: For each type of diabetes patient, the type feature vector of the diabetes patient type is determined based on the type feature vector of the historical diabetes patients included in the diabetes patient type. Based on the type feature vector of each type of diabetes patient, the key patient classification factors for each type of diabetes patient are determined; Based on the key patient classification factors and the type feature vectors of each type of diabetes patient, a key type feature vector for each type of diabetes patient is generated. Based on the current patient’s medical records and key patient classification factors for each type of diabetes, generate key type feature vectors for each type of diabetes corresponding to the current patient. Based on the key type feature vector of each type of diabetes patient and the key type feature vector of the current patient corresponding to each type of diabetes patient, the diabetes patient type of the current patient is determined.

6. A cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model according to claim 5, characterized in that, Based on the type feature vector of each type of diabetes patient, key patient classification factors for each type of diabetes patient are determined, including: For each patient classification factor, the discrete value of the patient classification factor corresponding to each type of diabetes patient is calculated based on the type feature vector of each type of diabetes patient. The initial patient classification factor for each type of diabetes is determined based on the discrete value of each patient classification factor corresponding to each type of diabetes patient. Establish an objective function, whereby the objective function is related to the discrete values ​​of each patient classification factor corresponding to each type of diabetes patient; The objective function and the initial patient classification factors for each type of diabetes are iteratively optimized until the optimization conditions are met, thereby determining the key patient classification factors for each type of diabetes.

7. A cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model according to any one of claims 1-4, characterized in that, The feature extraction module extracts complication risk features from the current patient's medical record based on the complication risk of the current patient's diabetes type, including: Identify the key diagnostic factors for each complication; Based on the key diagnostic factors for complication risk of the current patient’s diabetes type, complication risk characteristics are extracted from the current patient’s medical records.

8. A cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model according to any one of claims 1-4, characterized in that, The adjunctive treatment module identifies patients with similar historical diabetes history from among those with a history of diabetes, based on the current patient's complication risk characteristics. These include: Calculate the similarity of complication risk characteristics between the current patient's complication risk characteristics and the complication risk characteristics of historical diabetic patients included in the current patient's diabetes patient type; Based on the similarity of complication risk, identify patients with similar historical diabetes from among those with a history of diabetes included in the current patient's diabetes patient type.

9. A cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model according to claim 5, characterized in that, The adjunctive therapy module determines the current patient's adjunctive therapy plan based on adjunctive therapy plans for patients with similar historical diabetes and the current patient's complication risk characteristics, including: The adjuvant therapy model is used to determine the adjuvant therapy plan for the current patient based on the adjuvant therapy plans, complication risk characteristics, and key type feature vectors of patients with similar history of diabetes, as well as the complication risk characteristics and key type feature vectors of the current patient.

10. A cloud platform system for diabetes adjuvant therapy based on an artificial intelligence model according to claim 9, characterized in that, The adjunctive treatment model includes an input layer, a first feature extraction branch, a second feature extraction branch, a third feature extraction branch, a feature fusion layer, and a fully connected unit. The first feature extraction branch is used to extract features of adjunctive treatment plans for patients with similar historical diabetes. The second feature extraction branch is used to perform feature extraction after concatenating the complication risk features and key type feature vectors of patients with similar historical diabetes. The third feature extraction branch is used to perform feature extraction after concatenating the complication risk features and key type feature vectors of the current patient. The feature fusion layer is used to fuse the outputs of the first, second, and third feature extraction branches. The fully connected unit is used to output the adjunctive treatment plan for the current patient.

Images