Artificial intelligence agent acquisition method and application of multi-stage diabetic foot disease management
By constructing an AI-powered intelligent agent to integrate diabetic foot disease management data, the problems of data fragmentation and heterogeneity have been solved, enabling efficient and complete disease management and improving the accuracy and consistency of diagnosis and treatment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU HOSPITAL OF TRADITIONAL CHINESE MEDICINE
- Filing Date
- 2026-03-31
- Publication Date
- 2026-06-26
AI Technical Summary
Multi-stage management of diabetic foot faces challenges such as fragmented follow-up visit data, heterogeneous and incomplete multimodal data, and a lack of standardized assessment methods, resulting in low treatment efficiency and poor outcome consistency.
By constructing an AI intelligent agent, integrating multiple disease management data, updating disease task templates, optimizing the data integration process, and using large language models to evaluate and optimize data optimization actions, the convenience and completeness of data are achieved.
It improves the convenience and data integrity of multi-stage management of diabetic foot, provides good data technology support for diagnosis and treatment, and enhances diagnostic accuracy and treatment plan formulation.
Smart Images
Figure CN122290925A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of diabetes disease course data management technology, specifically to an AI agent acquisition method and application for multi-stage diabetic foot disease course management. Background Technology
[0002] Diabetic foot is one of the most serious chronic complications of diabetes. Its progression involves multi-stage and multi-dimensional pathological changes, including lower extremity arteriosclerosis, peripheral nerve damage, skin and subcutaneous tissue ulcers, infection, and osteomyelitis. It is characterized by three core features: frequent follow-up visits, multimodal data heterogeneity, and significant individual differences in visit frequency, posing a significant challenge to dynamic clinical assessment and precise intervention. Lower extremity arterial ultrasound, as the preferred non-invasive imaging method for assessing vascular lesions, can dynamically monitor the inner diameter, intimal thickness, plaque status, and blood flow parameters of key vessels such as the dorsalis pedis artery, anterior tibial artery, and posterior tibial artery. By integrating morphological and hemodynamic information, it improves diagnostic accuracy and provides core evidence for disease grading and treatment planning. In addition, clinical assessment also requires the integration of unstructured text medical records (ulcer location, exudation, etc.), laboratory indicators (glycated hemoglobin, C-reactive protein, white blood cell count, etc.), and other imaging data such as X-rays.
[0003] In summary, the current management of diabetic foot disease faces three major bottlenecks: ① Fragmented and scattered follow-up data, with patients' follow-up intervals ranging from two weeks to several years, involving numerous visits. The results of each examination are stored in a scattered manner in the medical record system, making manual integration time-consuming, labor-intensive, and prone to missing information about disease progression trends; ② Heterogeneous and incomplete multimodal data, with some examination data missing at different stages of treatment due to changes in the condition or limitations in medical resources; ③ Lack of standardized assessment methods, with physicians relying on personal experience for assessment, resulting in poor consistency of results. Summary of the Invention
[0004] The purpose of this application is to overcome the shortcomings and deficiencies in the existing technology and provide an AI agent acquisition method and application for multi-stage disease management of diabetic foot. This AI agent can integrate the patient's disease management data, improve the convenience and data integrity of multi-stage disease management of diabetic foot, and provide good data technology support for the diagnosis and treatment of diabetic foot.
[0005] The first aspect of this application provides a method for acquiring an AI agent for multi-stage management of diabetic foot, including:
[0006] Obtain multiple training data and multiple validation data for the multi-stage management of diabetic foot.
[0007] The initial disease course task template is updated based on the multiple disease course management training data to obtain candidate disease course task templates; the candidate disease course task templates are used to cooperate with the initial intelligent agent to acquire data and optimize actions.
[0008] Based on the candidate disease course task template and the initial intelligent agent, obtain the verification data optimization actions for each disease course management verification data;
[0009] If the overall score of the optimization action of all the validation data of the disease management validation data is greater than or equal to the preset score threshold, the candidate disease task template is determined as the target disease task template; otherwise, the candidate disease task template is updated according to the multiple disease management training data.
[0010] Based on the target disease course task template and the initial intelligent agent, an AI intelligent agent is constructed to optimize actions by outputting data based on the current disease course management data.
[0011] As one implementation method, the step of updating the initial disease course task template based on the multiple disease course management training data to obtain candidate disease course task templates includes:
[0012] Using the initial disease course task template and the initial intelligent agent, obtain the training data optimization action for the current disease course management training data;
[0013] By using disease course assessment templates and large language assessment models, multi-dimensional training scores are obtained to optimize actions based on training data of the current disease course management training data.
[0014] Based on the multi-dimensional training score of the current disease course management training data, the large language optimization model is called to optimize the initial disease course task template to obtain the optimized disease course task template.
[0015] Based on the optimized disease course task template and the initial agent, the training data optimization action for the next disease course management training data is obtained. Based on the multi-dimensional training score of the training data optimization action for the next disease course management training data, the large language optimization model is called to optimize the optimized disease course task template, so as to continue to obtain the training data optimization action for other disease course management training data, until the multiple disease course management training data are traversed, and the finally optimized optimized disease course task template is used as the candidate disease course task template.
[0016] As one implementation method, the step of optimizing the initial disease task template by calling a large language optimization model based on the multi-dimensional training score of the current disease course management training data to obtain the optimized disease task template includes:
[0017] If any dimension training score in the multi-dimensional training score is lower than the preset score threshold, the large language optimization model is invoked to optimize the initial disease course task template, thereby obtaining the optimized disease course task template.
[0018] If the training scores of all dimensions of the multi-dimensional training score are greater than the preset score threshold, the initial disease course task template is determined as the optimized disease course task template.
[0019] As one implementation method, the step of optimizing the multi-dimensional training score of the action based on the training data of the next course of disease management training data, and calling the large language optimization model to optimize the optimized course of disease task template, includes:
[0020] If the multi-dimensional training score of the training data optimization action of the next course of disease management training data is lower than the preset score threshold, the large language optimization model is called to optimize the optimized course of disease task template to obtain a new optimized course of disease task template.
[0021] If all dimensions of the multi-dimensional training score of the training data optimization action of the next course of disease management training data are greater than the preset score threshold, the optimized course of disease task template is retained.
[0022] As one implementation method, the steps for obtaining the comprehensive score of the validation data optimization actions for all disease management validation data include:
[0023] By using disease course assessment templates and big language assessment models, we obtain multi-dimensional validation scores for validation data optimization actions for each disease course management validation data;
[0024] The comprehensive score is obtained by averaging the multi-dimensional validation scores of all disease management validation data.
[0025] As one implementation method, the comprehensive score is obtained using the following formula:
[0026]
[0027] in, For the aforementioned comprehensive score, The amount of data used to validate disease management. The multi-dimensional validation score is used to validate the data for the management of the disease course.
[0028] Compared to related technologies, the AI agent acquisition method for multi-stage diabetic foot disease management in this application updates the initial disease task template with multiple disease management training data to obtain candidate disease task templates. Then, based on the candidate disease task templates and the initial agent, it obtains the verification data optimization actions for each disease management verification data. Next, based on the comprehensive score of the verification data optimization actions for all disease management verification data, it determines whether to continue updating the candidate disease task templates until a target disease task template is obtained where the comprehensive score of the verification data optimization actions is greater than or equal to a preset score threshold. Thus, based on the target disease task template and the initial agent, an AI agent is constructed to output data optimization actions based on the current disease management data. This facilitates the optimization and integration of the patient's current disease management data based on the data optimization actions output by the AI agent, improving the convenience and data integrity of multi-stage diabetic foot disease management.
[0029] The second aspect of this application provides a method for applying an AI agent in the multi-stage management of diabetic foot, including:
[0030] The patient's current disease management data is input into the AI agent, and the first data optimization action is output by the AI agent.
[0031] Based on the first data optimization action, the current disease management data is optimized to obtain optimized disease management data;
[0032] The optimized disease management data is input into the AI agent. If the AI agent outputs a second data optimization action, the optimized disease management data is optimized according to the second data optimization action. The AI agent continues to obtain new second data optimization actions corresponding to the optimized disease management data after optimization, until the AI agent no longer outputs new second data optimization actions, and then outputs the final optimized disease management data.
[0033] In one implementation, the first data optimization action and / or the second data optimization action includes querying patient pathology data.
[0034] In one implementation, the first data optimization action and / or the second data optimization action includes calling a data processing tool; the data processing tool includes a medical record query tool, a network search tool, an API call tool, a calculation tool, and a multimodal data processing tool.
[0035] As one implementation, after the step of outputting the final optimized disease management data, the method further includes:
[0036] By summarizing the final optimized disease management data through a large language model, the patient's disease management data summary results are obtained.
[0037] Compared with related technologies, the AI agent application method for multi-stage management of diabetic foot in this application can optimize actions based on the data output by the AI agent, optimize and integrate the patient's current disease management data, improve the convenience and data integrity of multi-stage management of diabetic foot, and provide good data technology support for the diagnosis and treatment of diabetic foot.
[0038] To provide a clearer understanding of this application, the specific embodiments of this application will be described below in conjunction with the accompanying drawings. Attached Figure Description
[0040] Figure 1 This is a flowchart illustrating an embodiment of the AI agent acquisition method for multi-stage management of diabetic foot disease according to this application.
[0041] Figure 2 This is a flowchart illustrating an embodiment of the AI agent application method for multi-stage management of diabetic foot disease according to this application.
[0042] Figure 3 This is a diagram illustrating a database of historical medical records.
[0043] Figure 4 This is a schematic diagram illustrating the application process of an AI agent for multi-stage management of diabetic foot according to an embodiment of this application.
[0044] Figure 5 This is a schematic diagram illustrating the scoring differences between an AI agent and two native large models according to one embodiment of this application. Detailed Implementation
[0046] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.
[0047] It should be understood that the described embodiments are merely some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of the embodiments of this application.
[0048] In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances. The singular forms "a," "the," and "the" used in this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. The word "if" as used herein can be interpreted as "when," "when," or "in response to determination."
[0049] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0050] Please see Figure 1 The first embodiment of this application discloses a method for obtaining an AI agent for multi-stage management of diabetic foot, including:
[0051] S01: Obtain multiple training data and multiple validation data for the multi-stage management of diabetic foot;
[0052] S02: Update the initial disease course task template based on the multiple disease course management training data to obtain candidate disease course task templates; the candidate disease course task templates are used to cooperate with the initial intelligent agent to acquire data and optimize actions;
[0053] S03: Based on the candidate disease course task template and the initial intelligent agent, obtain the verification data optimization action for each disease course management verification data;
[0054] S04: If the comprehensive score of the optimization action of all the validation data of the disease management validation data is greater than or equal to the preset score threshold, the candidate disease task template is determined as the target disease task template; otherwise, the candidate disease task template is updated according to the multiple disease management training data.
[0055] S05: Based on the target disease course task template and the initial intelligent agent, construct an AI intelligent agent for optimizing actions by outputting data based on the current disease course management data.
[0056] In a feasible embodiment, S021: The step of updating the initial disease course task template based on the multiple disease course management training data to obtain candidate disease course task templates includes:
[0057] S021: Obtain the training data optimization action for the current disease course management training data through the initial disease course task template and the initial intelligent agent;
[0058] S022: Using the disease course assessment template and the big language assessment model, obtain the multi-dimensional training score of the training data optimization action for the current disease course management training data;
[0059] S023: Based on the multi-dimensional training score of the current disease course management training data, call the large language optimization model to optimize the initial disease course task template to obtain the optimized disease course task template;
[0060] S024: Based on the optimized disease course task template and the initial agent, obtain the training data optimization action for the next disease course management training data, and based on the multi-dimensional training score of the training data optimization action for the next disease course management training data, call the large language optimization model to optimize the optimized disease course task template, so as to continue to obtain the training data optimization action for other disease course management training data, until the multiple disease course management training data are traversed, and use the finally optimized optimized disease course task template as the candidate disease course task template.
[0061] In a feasible embodiment, S023: The step of optimizing the initial disease task template by calling a large language optimization model based on the multi-dimensional training score of the current disease course management training data to obtain the optimized disease task template includes:
[0062] S0231: If there is a dimension training score that is lower than the preset score threshold in the multi-dimensional training score, the large language optimization model is called to optimize the initial disease course task template to obtain the optimized disease course task template.
[0063] S0232: If all dimensions of the multi-dimensional training score are greater than the preset score threshold, the initial disease course task template is determined as the optimized disease course task template.
[0064] In a feasible embodiment, S024: The step of optimizing the multi-dimensional training score of the action based on the training data of the next course of disease management training data, and calling the large language optimization model to optimize the optimized course of disease task template, includes:
[0065] S0241: If the multi-dimensional training score of the training data optimization action of the next course of disease management training data is lower than the preset score threshold, the large language optimization model is called to optimize the optimized course of disease task template to obtain a new optimized course of disease task template.
[0066] S0242: If all dimensions of the multi-dimensional training score of the training data optimization action of the next course of disease management training data are greater than the preset score threshold, the optimized course of disease task template is retained.
[0067] In a feasible embodiment, S04: The step of obtaining the comprehensive score of the validation data optimization action for all disease management validation data includes:
[0068] S041: Obtain multi-dimensional validation scores for validation data optimization actions for each disease course management validation data through the disease course assessment template and the big language assessment model;
[0069] S042: The comprehensive score is obtained by averaging the scores of the multi-dimensional validation scores of all disease management validation data.
[0070] In a feasible embodiment, the comprehensive score is obtained by the following formula:
[0071]
[0072] in, For the aforementioned comprehensive score, The amount of data used to validate disease management. The multi-dimensional validation score is used to validate the data for the management of the j-th disease course.
[0073] The process of the AI agent acquisition method for multi-stage management of diabetic foot in this application can be implemented through the following training loop:
[0074] 1. Input: Training set Validation set Maximum number of iterations Preset rating threshold Among them, the training set Includes multiple disease management training data ; Validation set Includes multiple disease management verification data.
[0075] 2. Loop optimization: when At that time, perform the following steps:
[0076] a. Training set evaluation and optimization (steps i to iii): Evaluate and optimize the training set. Each piece of data :
[0077] i. Generation: Using a disease progress task template The agent generates output training data and optimizes actions. ;
[0078] ii. Assessment: Use the course assessment template and large language assessment model right An evaluation was conducted to obtain score vectors for each dimension. ,in, Let be the score of the i-th dimension in the t-th iteration, and k be the upper bound of the dimension. The optimization action is performed on the same training data. The corresponding score vectors for all dimensions are the multi-dimensional training scores.
[0079] iii. Inspection and Optimization: If the score for any dimension falls below the threshold... Then the large language optimization model is invoked. Generate optimization suggestions and update the disease progress task template. Receive the disease progress task template for the next cycle. .otherwise, ← .
[0080] b. Validation set evaluation: using the final disease progression task template. In the validation set The evaluation was conducted, and all rating vectors were collected. . Represents all score vectors of the validation data for the j-th disease course management.
[0081] c. Termination judgment: Calculate the average composite score on the validation set. .like Then stop iterating and output the disease course task template. Otherwise, let And continue the cycle.
[0082] The dimensions mentioned include at least whether they comply with medical guidelines, the completeness of the report, and its conciseness.
[0083] Agent: A small executor with parameter θ, which remains fixed during training;
[0084] Disease progress task template This includes task instructions, character settings, and output formats, and is the core object of optimization.
[0085] Disease course assessment template Used to guide the evaluation of LLM to perform multi-dimensional quantitative scoring of agent output;
[0086] Large Language Assessment Model As an "automatic reviewer," based on the disease progress assessment template... Score the output of the intelligent agent;
[0087] Large Language Optimization Model As a "Prompt Engineer," based on the large language evaluation model... Based on feedback, the disease progress task template was iteratively modified and optimized. .
[0088] Compared to related technologies, the AI agent acquisition method for multi-stage diabetic foot disease management in this application updates the initial disease task template with multiple disease management training data to obtain candidate disease task templates. Then, based on the candidate disease task templates and the initial agent, it obtains the verification data optimization actions for each disease management verification data. Next, based on the comprehensive score of the verification data optimization actions for all disease management verification data, it determines whether to continue updating the candidate disease task templates until a target disease task template is obtained where the comprehensive score of the verification data optimization actions is greater than or equal to a preset score threshold. Thus, based on the target disease task template and the initial agent, an AI agent is constructed to output data optimization actions based on the current disease management data. This facilitates the optimization and integration of the patient's current disease management data based on the data optimization actions output by the AI agent, improving the convenience and data integrity of multi-stage diabetic foot disease management.
[0089] Please see Figure 2 The second embodiment of this application provides a method for applying an AI agent in the multi-stage management of diabetic foot, including:
[0090] S11: Input the patient's current disease management data into the AI agent to obtain the first data optimization action output by the AI agent;
[0091] S12: Optimize the current disease management data according to the first data optimization action to obtain optimized disease management data;
[0092] S13: Input the optimized disease management data into the AI agent. If the AI agent outputs a second data optimization action, optimize the optimized disease management data according to the second data optimization action. Then, through the AI agent, continue to obtain a new second data optimization action corresponding to the optimized disease management data after optimization, until the AI agent no longer outputs a new second data optimization action, and output the final optimized disease management data.
[0093] For example, the AI agent determines that it needs to obtain the patient's historical medical records based on the patient's current disease management data. At this time, the first data optimization action can be to call the "query patient medical history" tool to update and optimize the current disease management data to obtain optimized disease management data.
[0094] Then, the AI agent determines that the patient has had an ultrasound examination before based on the optimized disease management data, and it is necessary to compare the differences between the two ultrasound examination results. At this time, the corresponding second data optimization action can be to call the "ultrasound examination result comparison" tool to compare the ultrasound examination results and update and optimize the optimized disease management data based on the comparison results.
[0095] Then, based on the optimized disease management data, the AI agent determines that there are still differences in the patient's ultrasound examination that require further indication. At this point, the corresponding second data optimization action could be to call RAG to obtain ultrasound-related knowledge... until the AI agent does not output any new second data optimization actions, and finally obtains the patient's current condition, changes compared to the previous diagnosis, and treatment recommendations.
[0096] In one implementation, the first data optimization action and / or the second data optimization action includes querying patient pathology data.
[0097] The patient pathology data retrieval can be achieved through document extraction technology and RAG knowledge base retrieval technology. Document extraction technology: The MinerU tool is used to parse complex document formats, accurately identifying tables, formulas, and text paragraphs, solving problems such as content fragmentation and table extraction failures in traditional extraction tools. RAG knowledge base retrieval: During retrieval, a query vector is first generated using Qwen3~Embedding~0.6B, and then compared with the knowledge base vector using cosine distance to select the top 20 relevant segments. Next, the Qwen3~Reranker~0.6B model performs semantic rearrangement, filtering out irrelevant content, and finally retaining the top 5 core knowledge points for input into the LLM, ensuring the evidence-based nature of the conclusions.
[0098] In one implementation, the first data optimization action and / or the second data optimization action includes calling a data processing tool; the data processing tool includes a medical record query tool, a network search tool, an API call tool, a calculation tool, and a multimodal data processing tool.
[0099] The medical record retrieval tool allows users to input their medical card number and the number of historical records. Based on data from the historical medical record database, it returns structured JSON formatted medical record data. The historical medical record database includes, for example,... Figure 3As shown; Online search tool: supports keyword search on search engines, limiting the number of results; API calling tool: connects to medical database APIs such as PubMed to obtain the latest research evidence; Calculation tool: realizes the calculation of indicator change rate (such as the proportion of ulcer area reduction, blood glucose fluctuation amplitude); Multimodal data processing tool: integrates heterogeneous data from ultrasound reports, other imaging reports, laboratory indicators and text medical records.
[0100] As one implementation, after the step of outputting the final optimized disease management data, the method further includes:
[0101] S14: Summarize the final optimized disease management data through a large language model to obtain the summary results of the patient's disease management data.
[0102] Compared with related technologies, the AI agent application method for multi-stage management of diabetic foot in this application can optimize actions based on the data output by the AI agent, optimize and integrate the patient's current disease management data, improve the convenience and data integrity of multi-stage management of diabetic foot, and provide good data technology support for the diagnosis and treatment of diabetic foot.
[0103] In summary, the AI agent application method for multi-stage management of diabetic foot in this application can be understood as a data processing method for multi-stage management of diabetic foot, including:
[0104] Acquire disease management data of the target patient at multiple stages of medical treatment, wherein the disease management data includes at least text medical record data, laboratory indicator data and medical examination data;
[0105] The disease management data of the multiple treatment stages are subjected to structured parsing to obtain structured disease data corresponding to each treatment stage. The structured disease data includes at least the treatment time identifier, lesion information, vascular assessment information, infection assessment information, and metabolic indicator information.
[0106] Based on patient identifiers and consultation time identifiers, structured disease progress data at each consultation stage are correlated across stages to obtain a time-series disease progress dataset for the target patient.
[0107] The time-series disease dataset is subjected to integrity and consistency checks to identify missing data items, conflicting data items, and data items to be supplemented.
[0108] For the missing data items, conflicting data items, and data items to be supplemented, the corresponding data processing tools are invoked to perform at least one of the following processes: data supplementation, historical medical record retrieval, indicator comparison, medical knowledge retrieval, and result reordering, to obtain the updated disease course dataset;
[0109] Based on the updated disease course dataset, a structured disease course management result is generated, which includes information on disease course evolution, changes in abnormal indicators, changes in lesions, and sources of evidence.
[0110] In the above steps, the processes of integrity detection, consistency verification, calling data processing tools, and generating structured disease management results containing information on disease progression, abnormal indicator changes, lesion changes, and evidence sources are all implemented by the trained AI agent.
[0111] Please see Figure 4 After obtaining the patient's current disease management data, the AI intelligent caller will output corresponding actions, such as calling pathology tools to query historical medical records, calling the CLTI knowledge base for CLTI definitions and diagnostic criteria, calling the ultrasound database to obtain the patient's ultrasound data, and calling search tools to find treatment plans and nursing suggestions.
[0112] For the AI agent in this application, a blinded evaluation method can be used for assessment. The core indicators include: ① Historical data matching accuracy: the accuracy rate of automatically associating different medical data of the same patient, calculated as "number of correctly matched cases / total number of cases × 100%"; ② Multi-course analysis accuracy: the degree of consistency between the description of disease progression, assessment of lesion changes and clinical reality, using the Kappa consistency test; ③ Guideline citation accuracy: the relevance of the guideline provisions cited in the generated conclusions to the clinical scenario, calculated as "number of accurately cited cases / total number of cases × 100%"; ④ Clinical usability score: using a 1-10 Likert scale to assess the report's auxiliary value for clinical decision-making; ⑤ Hallucination incidence rate: the proportion of generated content that does not conform to real data or clinical common sense, calculated as "number of cases with hallucinations / total number of cases × 100%"; ⑥ Disease course data integration time: the total time taken from data retrieval to generating a complete report (unit: min).
[0113] The performance differences between the AI agent of this application and the two native large models (Qwen3~max, Baichuan~M1~14B) are compared below:
[0114] Please see Figure 5The performance of the AI agent in 34 cases was compared with that of two native large models. The results showed that the clinical usability score of the AI agent was 8.29±0.91, which was higher than that of Qwen3~max (7.56±0.70, t=4.19, P<0.001) and Baichuan~M1~14B (7.82±0.67, t=3.67, P<0.001); the score of Baichuan~M1~14B was higher than that of Qwen3~max (t=-2.18, P=0.04). Of the 34 case scores, 55.88% (19 / 34) of the AI agent scores were higher than the two native large models (comparing the scores of the same case, the same below); 11.76% (4 / 34) of the Qwen3 ~ max scores were higher than the other two; 14.71% (5 / 34) of the Baichuan~M1~14B scores were higher than the other two; and 17.62% (6 / 34) of the scores of the three were the same.
[0115] After grouping by the number of disease courses, the AI agent scored higher than the two native large models in each group, and the score difference was greater for the number of disease courses (as shown in Table 1 below):
[0116]
[0117] Table 1
[0118] Pearson correlation analysis shows (as shown in Table 2 below):
[0119]
[0120] Table 2
[0121] Tables 1 and 2 show that the AI agent score is positively correlated with the number of disease courses (r=0.405, P<0.01), indicating that the more disease courses there are, the higher the AI agent score tends to be; while the Qwen3~max score is not correlated with the number of disease courses (r=-0.197, P=0.264), and the Baichuan~M1~14B score has a weak correlation with the number of disease courses (r=0.254, P=0.148).
[0122] In addition, regarding the improvement of staging diagnosis accuracy by RAG (Retrieval Enhancement Generation Technology), the AI agent in this application has a staging accuracy rate of 94.1%, which is higher than Qwen3~max's 70.6% and Baichuan~M1~14B's 82.4%. Both native large models are prone to misjudging the pre-CLTI stage, while the AI agent can accurately identify the early CLTI stage through RAG retrieval.
[0123] Regarding the reduction of hallucinations, the hallucination incidence rate of the AI agent in this application is 8.7%, which is lower than 27.4% for Qwen3~max and 20.6% for Baichuan~M1~14B. For complex cases with multiple follow-up visits, the AI agent can automatically filter redundant data and focus on key changes in the patient's condition.
[0124] Regarding the enhanced efficiency of tool invocation, the automatic recognition accuracy of key indicators of the AI intelligent agent in this application reaches 95.7%, which is higher than Qwen3~max's 64.7% and Baichuan~M1~14B's 73.5%; and both native large models require explicit labeling of indicators in the prompt words.
[0125] In terms of other performance indicators, the AI agent in this application achieved a historical data matching accuracy of 95.7%, a multi-pathology analysis accuracy of 89.3%, and a guideline citation accuracy of 92.1%, all of which are superior to the two native large models (P < 0.05). Specifically, the core performance indicators of the three models are compared in Table 3 below:
[0126]
[0127] Table 3
[0128] For example, in a clinical application case: Patient Zhang San, 65 years old, with a 10-year history of diabetes and a 3-year history of diabetic foot, whose condition has worsened in the past 2 months, has visited the hospital a total of 8 times (high disease duration group). The clinical application cases of the AI agent and the two native large models are compared in Table 4 below:
[0129]
[0130] Table 4
[0131] In comparison, it is evident that the AI agent in this application yields significantly more complete multi-stage disease management data for diabetic foot than the two native large models. The Wagner classification, mentioned in Table 4, was first proposed by Meggitt in 1976. Wagner later popularized it and it is currently the most widely used classification method in clinical practice and research. The Wagner classification for diabetic foot is as follows:
[0132] Grade 0 - There are risk factors for developing foot ulcers, but no ulcers are currently present; Grade 1 - Superficial ulcers on the foot surface, without signs of infection, with neuropathic ulcers as the most prominent feature;
[0133] Grade 2 - Deeper ulcer, often complicated by soft tissue infection, without osteomyelitis or deep abscess;
[0134] Grade 3 - Deep ulcer, with abscess or osteomyelitis;
[0135] Grade 4 - Localized gangrene (toes, heels, or dorsum of the forefoot), characterized by ischemic gangrene, usually accompanied by neuropathy;
[0136] Grade 5 - Gangrene of the entire foot.
[0137] The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.
[0138] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0139] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function selected in one or more boxes.
[0140] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function selected in one or more boxes.
[0141] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0142] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0143] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0144] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0145] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for acquiring an AI agent for multi-stage management of diabetic foot, characterized in that, include: Obtain multiple training data and multiple validation data for the multi-stage management of diabetic foot. The initial disease course task template is updated based on the multiple disease course management training data to obtain candidate disease course task templates; the candidate disease course task templates are used to cooperate with the initial intelligent agent to acquire data and optimize actions. Based on the candidate disease course task template and the initial intelligent agent, obtain the verification data optimization actions for each disease course management verification data; If the overall score of the optimization action of all the validation data of the disease management validation data is greater than or equal to the preset score threshold, the candidate disease task template is determined as the target disease task template; otherwise, the candidate disease task template is updated according to the multiple disease management training data. Based on the target disease course task template and the initial intelligent agent, an AI intelligent agent is constructed to optimize actions by outputting data based on the current disease course management data.
2. The method for acquiring an AI agent for multi-stage management of diabetic foot according to claim 1, characterized in that, The steps for updating the initial disease course task template based on the multiple disease course management training data to obtain candidate disease course task templates include: Using the initial disease course task template and the initial intelligent agent, obtain the training data optimization action for the current disease course management training data; By using disease course assessment templates and large language assessment models, multi-dimensional training scores are obtained to optimize actions based on training data of the current disease course management training data. Based on the multi-dimensional training score of the current disease course management training data, the large language optimization model is called to optimize the initial disease course task template to obtain the optimized disease course task template. Based on the optimized disease course task template and the initial agent, the training data optimization action for the next disease course management training data is obtained. Based on the multi-dimensional training score of the training data optimization action for the next disease course management training data, the large language optimization model is called to optimize the optimized disease course task template, so as to continue to obtain the training data optimization action for other disease course management training data, until the multiple disease course management training data are traversed, and the finally optimized optimized disease course task template is used as the candidate disease course task template.
3. The method for acquiring an AI agent for multi-stage management of diabetic foot according to claim 2, characterized in that, The steps of optimizing the initial disease task template by calling a large language optimization model based on the multi-dimensional training score of the current disease course management training data to obtain the optimized disease task template include: If any dimension training score in the multi-dimensional training score is lower than the preset score threshold, the large language optimization model is invoked to optimize the initial disease course task template, thereby obtaining the optimized disease course task template. If the training scores of all dimensions of the multi-dimensional training score are greater than the preset score threshold, the initial disease course task template is determined as the optimized disease course task template.
4. The method for acquiring an AI agent for multi-stage management of diabetic foot according to claim 2, characterized in that, The steps of optimizing the multi-dimensional training score of actions based on the training data of the next course of disease management training data, and calling the large language optimization model to optimize the optimized course of disease task template, include: If the multi-dimensional training score of the training data optimization action of the next course of disease management training data is lower than the preset score threshold, the large language optimization model is called to optimize the optimized course of disease task template to obtain a new optimized course of disease task template. If all dimensions of the multi-dimensional training score of the training data optimization action of the next course of disease management training data are greater than the preset score threshold, the optimized course of disease task template is retained.
5. The method for acquiring an AI agent for multi-stage management of diabetic foot according to claim 2, characterized in that, The steps for obtaining the comprehensive score of the validation data optimization actions for all disease management validation data include: By using disease course assessment templates and big language assessment models, we obtain multi-dimensional validation scores for validation data optimization actions for each disease course management validation data; The comprehensive score is obtained by averaging the multi-dimensional validation scores of all disease management validation data.
6. The method for acquiring an AI agent for multi-stage management of diabetic foot according to claim 5, characterized in that, The overall score is obtained using the following formula: in, For the aforementioned comprehensive score, The amount of data used to validate disease management. The multi-dimensional validation score is used to validate the data for the management of the j-th disease course.
7. A method for applying an AI agent in the multi-stage management of diabetic foot, characterized in that, include: The patient's current disease management data is input into the AI agent, and the first data optimization action is output by the AI agent. Based on the first data optimization action, the current disease management data is optimized to obtain optimized disease management data; The optimized disease management data is input into the AI agent. If the AI agent outputs a second data optimization action, the optimized disease management data is optimized according to the second data optimization action. The AI agent continues to obtain new second data optimization actions corresponding to the optimized disease management data after optimization, until the AI agent no longer outputs new second data optimization actions, and then outputs the final optimized disease management data.
8. The application method of the AI agent for multi-stage management of diabetic foot according to claim 7, characterized in that, The first data optimization action and / or the second data optimization action includes querying patient pathology data.
9. The application method of the AI agent for multi-stage management of diabetic foot according to claim 7, characterized in that, The first data optimization action and / or the second data optimization action includes calling data processing tools; the data processing tools include medical record query tools, online search tools, API call tools, calculation tools, and multimodal data processing tools.
10. The application method of the AI agent for multi-stage management of diabetic foot according to claim 7, characterized in that, After the step of outputting the final optimized disease management data, the method further includes: By summarizing the final optimized disease management data through a large language model, the patient's disease management data summary results are obtained.