A chronic disease medical data collection, analysis and management method and cloud platform

Abstract

The application discloses a chronic disease medical data collection, analysis, management method and cloud platform, comprising the following steps: firstly, acquiring the first chronic disease symptom description of the patient side and the second chronic disease symptom description of the doctor side through an interactive terminal respectively, obtaining a target chronic disease description through a preset large language model integration, and determining the target chronic disease type through a target chronic disease analysis model. Then, the feedback information of the doctor for the type is acquired, the feedback information and the target chronic disease type are input into the large language model to generate a management suggestion, and the management suggestion is pushed to a preset interactive interface of the interactive terminal, so that the chronic disease data is effectively processed and managed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart medical treatment, in particular to a chronic disease medical data collection, analysis and management method and cloud platform. BACKGROUND

[0002] With the increasing number of chronic disease patients, it is increasingly important to effectively collect, analyze and manage chronic disease medical data. Under the traditional mode, data collection is scattered and analysis is not accurate enough, making it difficult to provide comprehensive and targeted management suggestions for patients. SUMMARY

[0003] The present application relates to the technical field of smart medical treatment, in particular to a chronic disease medical data collection, analysis and management method and cloud platform.

[0004] In a first aspect, the present application provides a chronic disease medical data collection, analysis and management method, comprising:

[0005] Obtaining a first chronic disease symptom description on the patient side and a second chronic disease symptom description on the doctor side through an interactive terminal;

[0006] Inputting the first chronic disease symptom description and the second chronic disease symptom description into a pre-set large language model for integration to obtain a target chronic disease description;

[0007] Inputting the target chronic disease description into a pre-trained target chronic disease analysis model to obtain a target chronic disease type corresponding to the target chronic disease description;

[0008] Obtaining feedback information on the target chronic disease type from the doctor side through the interactive terminal;

[0009] Inputting the target chronic disease type and the feedback information into the pre-set large language model to obtain a management suggestion for the target chronic disease type, and pushing the management suggestion to a pre-set interactive interface of the interactive terminal.

[0010] In a second aspect, the present application provides a cloud platform, which is used to execute the method of the first aspect.

[0011] Compared with the prior art, the present application provides the following beneficial effects: by using the chronic disease medical data collection, analysis and management method and cloud platform disclosed in the present application, the first chronic disease symptom description on the patient side and the second chronic disease symptom description on the doctor side are obtained through an interactive terminal, a target chronic disease description is obtained by integrating the pre-set large language model, and a target chronic disease type is determined by a target chronic disease analysis model. Then, feedback information on the type from the doctor is obtained, which is input into the large language model together with the target chronic disease type to generate a management suggestion, which is pushed to a pre-set interactive interface of the interactive terminal, thereby realizing effective processing and management of chronic disease data. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the steps of the chronic disease medical data collection, analysis, and management method provided in an embodiment of the present invention.

[0014] Figure 2 This is a schematic block diagram of the cloud platform provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0016] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the chronic disease medical data collection, analysis, and management method provided in this embodiment. The method will now be described in detail.

[0018] Step S201: Obtain the first chronic disease symptom description from the patient's side and the second chronic disease symptom description from the doctor's side through the interactive terminal;

[0019] Step S202: Input the first chronic disease symptom description and the second chronic disease symptom description into a preset large language model for integration to obtain the target chronic disease description;

[0020] Step S203: Input the target chronic disease description into the pre-trained target chronic disease analysis model to obtain the target chronic disease type corresponding to the target chronic disease description;

[0021] Step S204: Obtain feedback information from the doctor regarding the target chronic disease type through the interactive terminal;

[0022] Step S205: Input the target chronic disease type and the feedback information into the preset large language model to obtain management suggestions for the target chronic disease type, and push the management suggestions to the preset interactive interface of the interactive terminal.

[0023] In this embodiment of the invention, for example, the server maintains a communication connection with the interactive terminal (such as a mobile app used by the patient or a self-service terminal in the hospital lobby). After the patient opens the relevant medical application and logs in to their account, the server receives a request from the terminal and begins to provide symptom input guidance for the patient. The server first pushes a visual symptom input interface to the interactive terminal. This interface has multiple options for common chronic disease symptoms, such as "headache," "fatigue," "prolonged cough," "joint pain," etc., as well as a custom symptom input box. Suppose a patient named Mr. Li has been feeling unwell recently and opens this medical application. Mr. Li sees "joint pain" among the common chronic disease symptoms and remembers that his knees have been hurting lately, so he clicks on this option. At this step, the server obtains the part of the symptoms selected by the patient as a partial description of the first chronic disease symptom. However, Mr. Li feels that his knee pain seems to have some special characteristics, so he enters the text "knee pain, especially when going up and down stairs, and sometimes it wakes me up at night" in the custom symptom input box. After receiving this information entered by Mr. Li in the custom symptom input box, the server uses it as a supplementary description of the first chronic disease symptom. Finally, the server integrates the previously obtained partial description of the first chronic disease symptom (joint pain) and the supplementary description of the first chronic disease symptom (knee pain, especially aggravated when going up and down stairs, sometimes waking up at night due to pain), resulting in a complete description of the first chronic disease symptom: "Joint pain, knee pain, especially aggravated when going up and down stairs, sometimes waking up at night due to pain." After the patient completes the symptom input and submits it on the interactive terminal, the server pushes the patient's relevant information (such as basic identity information, preliminary symptom description, etc.) to the doctor's interactive terminal (which may be an internal hospital diagnostic system terminal, etc.). After receiving this information, the doctor will conduct further inquiries and examinations on the patient. Assuming Dr. Zhang sees Mr. Li, after inquiring about Mr. Li's medical history in detail and examining relevant body parts, Dr. Zhang inputs his own description of Mr. Li's symptoms on the doctor's interactive terminal. For example, Dr. Zhang inputs: "The patient reports joint pain; examination revealed slight swelling in the knee joint, a grinding sound during movement, and a significant recent decline in physical strength." The server receives this input from Dr. Zhang and obtains the second chronic disease symptom description from the doctor's perspective. Before inputting the descriptions of the first and second chronic disease symptoms into the preset large language model, the server will first perform semantic cleanup operations on the two descriptions respectively.For Mr. Li's first chronic symptom description, "joint pain, knee pain, especially aggravated when going up and down stairs, sometimes waking him up at night due to pain," the server will check for duplicate expressions (for example, if Mr. Li accidentally entered a duplicate expression like "knee pain, knee very painful," the server will remove the redundant part), correct any typos (if Mr. Li mistyped a word), and standardize medical terminology (for example, further standardizing Mr. Li's common phrase "knee pain" to "knee joint pain"). For Dr. Zhang's second chronic symptom description, "patient reports joint pain, examination reveals slight swelling in the knee joint, grinding sound during movement, and recent significant decline in physical strength," the server will also perform the above semantic cleanup operations, such as standardizing some expressions to conform to standard medical terminology format. Then, the server will input the semantically cleaned first and second chronic symptom descriptions into a preset format conversion module. This format conversion module will convert these descriptions into appropriate formats according to the input requirements of the large language model. For example, the large language model requires the input text to have a specific encoding format and certain paragraph division rules. The format conversion module will convert the text according to these requirements to ensure that the input content can be correctly processed by the large language model. After semantic cleaning and format conversion, Mr. Li's symptom description and Dr. Zhang's symptom description are input into the preset large language model. Assuming that this large language model is trained based on deep learning algorithms, it can understand and analyze the input text content and integrate these descriptions according to its internal language logic and knowledge system. After receiving these two descriptions, the large language model will comprehensively consider the patient's self-perception (Mr. Li's description) and the doctor's professional examination results (Dr. Zhang's description) to generate a more comprehensive and accurate description of the target chronic disease. For example, for Mr. Li's case, the large language model may generate the following description of the target chronic disease: "The patient has knee pain symptoms, which are aggravated when going up and down stairs, and sometimes wakes up at night due to pain. At the same time, the examination revealed slight swelling in the knee joint and a grinding sound when moving. Recently, the patient's physical strength has decreased significantly, and it is preliminarily judged that there may be a joint-related chronic disease." Before actual application, the server has completed the training of the target chronic disease analysis model. This training process involves acquiring a large dataset of chronic disease analysis models and sample instances. For example, the sample instance dataset contains various chronic disease description instances, including target chronic disease description instances (instances whose corresponding chronic disease type is clearly known) and potential chronic disease description instances (instances whose specific chronic disease type is not yet known). For instance, for joint-related chronic diseases, the target chronic disease description instance might be "a patient experiences long-term knee pain, accompanied by swelling, stiffness, and limited mobility, diagnosed as rheumatoid arthritis," where the preset target chronic disease type is explicitly configured as "rheumatoid arthritis."For each chronic disease description instance, the server performs a segmentation operation. For example, the description instance of rheumatoid arthritis, mentioned above, is segmented into multiple symptom description texts, such as "long-term knee pain," "accompanied by swelling," "stiffness," and "limited mobility." Then, through the chronic disease analysis model, feature extraction operations are performed on the chronic disease description instance itself and each of its symptom description texts. This yields the overall descriptive features of the chronic disease description instance (e.g., the overall features extracted for the entire rheumatoid arthritis description instance may be related to some features of joint inflammation) and the partial descriptive features corresponding to each symptom description text (e.g., the partial features extracted for the symptom description text "long-term knee pain" may be related to the degree and duration of pain). Based on the above series of operations and the target chronic disease type value of the target chronic disease description instance, the server optimizes the model through a series of complex steps including aggregation, inference, and tuning, as mentioned earlier, ultimately obtaining the target chronic disease analysis model. When the server inputs the previously generated description of Mr. Li's condition ("The patient experiences knee pain, which worsens when going up and down stairs, and sometimes wakes him up at night due to pain. Examination also reveals slight swelling in the knee joint, a grinding sound during movement, and a recent significant decline in physical strength; a preliminary assessment suggests a possible joint-related chronic disease") into the pre-trained target chronic disease analysis model, the model analyzes and identifies this description based on its learned knowledge and feature patterns. After calculation and judgment, the model concludes that Mr. Li's condition likely corresponds to "osteoarthritis." The server, having obtained the target chronic disease type (e.g., "osteoarthritis"), pushes this result to the doctor's interactive terminal. Dr. Zhang, upon seeing this result, will further consider and judge the diagnosis based on his professional knowledge and clinical experience. Suppose that after careful consideration, Dr. Zhang believes this preliminary judgment has some merit but requires further examination for confirmation, such as a joint fluid examination. Dr. Zhang entered his feedback on the doctor's interactive terminal: "A preliminary assessment suggests osteoarthritis is a possibility, but further synovial fluid examination is needed for confirmation. The patient is advised to rest and reduce joint stress in the near future." The server received this feedback from Dr. Zhang, thus completing the acquisition of feedback information for the target chronic disease type. The server then input the previously obtained target chronic disease type ("osteoarthritis") and Dr. Zhang's feedback ("A preliminary assessment suggests osteoarthritis is a possibility, but further synovial fluid examination is needed for confirmation. The patient is advised to rest and reduce joint stress in the near future") into a pre-set large language model. This large language model will comprehensively analyze and process the input content based on its internally stored extensive medical knowledge, clinical experience, and relevant rules for chronic disease management.For example, the large language model might generate management suggestions like this: "For patients who may have osteoarthritis, while awaiting further test results, in addition to resting and reducing joint stress as advised by the doctor, they can also engage in some gentle joint exercises, such as walking, but the intensity of the activity should not be too high; in terms of diet, they can appropriately increase their intake of foods rich in calcium and vitamin D, such as milk and fish, which helps maintain joint health; if joint pain worsens, they can take painkillers prescribed by the doctor to relieve symptoms in a timely manner." After receiving the management suggestions generated by the large language model, the server will push these management suggestions to the preset interactive interface of the interactive terminal. If the push is sent to the patient's interactive terminal, Mr. Li will see the push notification of these management suggestions on his mobile application, and after opening it, he can view the specific management suggestions for his condition in detail, thereby better cooperating with treatment and conducting self-health management. If the push is sent to the doctor's interactive terminal, Dr. Zhang can also see these management suggestions, so as to refer to and further improve the treatment plan in subsequent diagnosis and treatment.

[0024] In one possible implementation, the target chronic disease analysis model is obtained in the following manner.

[0025] Obtain a chronic disease analysis model and a sample instance dataset. The chronic disease description instances in the sample instance dataset include target chronic disease description instances and potential chronic disease description instances. The target chronic disease description instances are configured with preset chronic disease type target values.

[0026] For each chronic disease description instance, the chronic disease description instance is segmented to obtain multiple symptom description texts for the chronic disease description instance;

[0027] Using the chronic disease analysis model, feature extraction operations are performed on the chronic disease description instance and each symptom description text of the chronic disease description instance to obtain the overall description features of the chronic disease description instance and the partial description features corresponding to each symptom description text.

[0028] Based on the target value of the chronic disease type of the target chronic disease description instance, an aggregation operation is performed on the overall description features of each chronic disease description instance in the sample instance dataset to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance.

[0029] For each symptom description text of the chronic disease description instance, based on the partial description features corresponding to the symptom description text and the overall description features of the chronic disease description instance to which the symptom description text belongs, the homogeneity information between the symptom description text and the chronic disease description instance to which it belongs is confirmed.

[0030] Based on the homogenized information and the target value of the chronic disease type corresponding to the potential chronic disease description instance, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model. The target chronic disease analysis model is used to identify chronic diseases in the target chronic disease description to obtain the target value of the chronic disease type in the target chronic disease description.

[0031] The target chronic disease type corresponding to the target chronic disease description is determined based on the target value of the chronic disease type.

[0032] In this embodiment of the invention, for example, the server first initiates a relevant model acquisition program, which connects to a pre-defined model repository. This repository can be a specific folder on the server's local machine or a remote storage area based on cloud services, containing various medical analysis models for different purposes and types. The server searches for chronic disease analysis models in the repository according to specific search rules and identifiers. For example, it filters by model name tags, version numbers, applicable disease areas, etc. Suppose that in the repository, the server finds a model named "General Chronic Disease Analysis Model V2.0," which was previously developed by a professional medical data team based on a large amount of medical data and advanced machine learning algorithms, and is suitable for preliminary analysis of various common chronic diseases. The server extracts this model from the repository, preparing it for subsequent processing and optimization operations. At the same time, the server also acquires a sample instance dataset. This dataset is also stored in a specific location, possibly in a different folder on the same server as the model repository, or in another dedicated data storage server. The server establishes a connection with the storage location of the dataset through a data access interface, and then operates according to predetermined dataset acquisition instructions. For example, it sends a request containing information such as the dataset name, version number, and data source. Suppose the dataset to be acquired is named "Chronic Disease Description Instance Dataset 2023," which is compiled from a large amount of real chronic disease patient case information provided by multiple hospitals and medical research institutions. This dataset contains chronic disease description instances, including target chronic disease description instances and potential chronic disease description instances. For example, a target chronic disease description instance might be a case description where the specific chronic disease type is known (here, pulmonary tuberculosis is the preset target chronic disease type), such as "The patient has a persistent fever for two weeks, accompanied by cough and sputum; a chest X-ray shows shadows in the lungs; and has been diagnosed with pulmonary tuberculosis." A potential chronic disease description instance, on the other hand, might be a case description where the specific chronic disease type is unclear, such as "The patient experiences intermittent headaches, sometimes accompanied by dizziness, and poor sleep quality, but the specific chronic disease diagnosis is not yet clear." After successfully acquiring this dataset, the server can perform subsequent processing operations on the instances. After acquiring the sample instance dataset, the server will perform a segmentation operation on each chronic disease description instance. Suppose the server is processing the previously mentioned chronic disease description instance of tuberculosis: "The patient has had a persistent fever for two weeks, accompanied by cough and sputum; a chest X-ray showed shadows in the lungs, and the patient was diagnosed with tuberculosis." The server will identify key information and delimiters in the sentence according to certain rules, such as using punctuation marks like commas and pauses as initial segmentation criteria, while also combining medical terminology and common symptom descriptions. For this instance, the server will first segment it into the following parts: "The patient has had a persistent fever for two weeks," "accompanied by cough," "sputum," and "a chest X-ray showed shadows in the lungs," etc.These segmented portions constitute the multiple symptom description texts of the chronic disease description instance. For example, regarding the potential chronic disease description instance, "The patient experiences intermittent headaches, sometimes accompanied by dizziness, and poor sleep quality, but the specific chronic disease diagnosis is still unclear," the server similarly segments it according to the above rules, obtaining symptom description texts such as "the patient experiences intermittent headaches," "sometimes accompanied by dizziness," and "poor sleep quality." Through this segmentation operation, the server can refine each chronic disease description instance into multiple more specific symptom description texts, facilitating subsequent more in-depth feature extraction and analysis. After completing the segmentation operation, the server inputs each chronic disease description instance and each segmented symptom description text into the previously acquired chronic disease analysis model for feature extraction. For the target chronic disease description instance of pulmonary tuberculosis, "The patient has a persistent fever for two weeks, accompanied by cough and sputum; chest X-ray shows shadows in the lungs, and has been diagnosed with pulmonary tuberculosis," when it is input as a whole into the chronic disease analysis model, the model analyzes the overall text content of this instance based on its internal neural network structure and algorithmic logic. Assuming the model comprehensively processes the symptom information (persistent fever, cough, sputum, lung shadows, etc.), disease diagnosis (tuberculosis), and their interrelationships within this instance, the extracted overall descriptive features may include feature vectors related to lung inflammation, such as the degree of lung inflammation and the source of infection. These overall descriptive features can reflect some key characteristics of this chronic disease description instance as a whole. Simultaneously, for each symptom description text segmented from this target chronic disease description instance of tuberculosis, such as "patient has had persistent fever for two weeks," when it is individually input into the chronic disease analysis model, the model will also perform feature extraction operations according to its internal algorithm. For the symptom description text "patient has had persistent fever for two weeks," the model may extract descriptive features including the duration of fever and the degree of fever (e.g., low-grade fever, high fever, etc.). Similarly, for the symptom description text "accompanied by cough," the model may extract descriptive features including the frequency and severity of cough. By performing this feature extraction operation on each symptom description text, the server can obtain partial descriptive features corresponding to each symptom description text. These partial descriptive features can reflect the specific characteristics of each symptom in more detail. After obtaining the overall and partial descriptive features of all chronic disease description instances, the server begins the aggregation operation. First, the server determines the target value of the chronic disease type for the target chronic disease description instance. For example, for the previously mentioned target chronic disease description instance of pulmonary tuberculosis, its target value of chronic disease type is "pulmonary tuberculosis". Then, the server iterates through each chronic disease description instance in the sample instance dataset to obtain their overall descriptive features.Suppose that, in addition to the previously mentioned tuberculosis case, the sample instance dataset also contains a potential chronic disease description instance: "A patient has a persistent cough, occasionally accompanied by low-grade fever; chest X-ray shows no obvious abnormalities, but lung disease is suspected." The server will obtain the overall descriptive features of this instance, which may include lung-related feature vectors such as lung function features and respiratory symptom features. A common aggregation operation is based on matching degree. The server will pre-define some aggregation cluster representatives, which can be obtained by extracting and organizing typical overall descriptive features of some known chronic disease types. For example, for chronic diseases related to lung disease, there might be one aggregation cluster representative representing typical overall descriptive features for tuberculosis, and another aggregation cluster representative representing typical overall descriptive features for chronic bronchitis, etc. The server will then verify the matching degree between the overall descriptive features of each chronic disease description instance in the sample instance dataset and each aggregation cluster representative. For the previously mentioned potential chronic disease description instance, "The patient has a long-term cough, occasionally accompanied by low-grade fever, and chest X-ray shows no obvious abnormalities, but lung disease is suspected," the server calculates its overall descriptive features and their matching degree with the pulmonary tuberculosis cluster representative, as well as its matching degree with the chronic bronchitis cluster representative. Assuming that the calculated matching degree with the pulmonary tuberculosis cluster representative is relatively high, based on this matching degree and the target chronic disease type (pulmonary tuberculosis) of the target chronic disease description instance, the server determines the target cluster representative (in this case, the pulmonary tuberculosis cluster representative) corresponding to this chronic disease description instance from each cluster representative, and then adds this chronic disease description instance to the chronic disease description set corresponding to the target cluster representative. Next, for each cluster representative's corresponding chronic disease description set, the server selects chronic disease description instances that meet the preset cluster representative conditions as new cluster representatives. For example, if the preset criteria for a cluster representative are instances with the highest matching degree and a number reaching a certain threshold, the server will select instances that meet this criterion from the chronic disease description set corresponding to the tuberculosis cluster representative as the new tuberculosis cluster representative. The server will repeatedly confirm the matching degree between the overall descriptive features of each chronic disease description instance in the sample instance dataset and each cluster representative until the determined cluster representative meets the aggregation completion status. For example, aggregation is considered complete when the number of instances in the chronic disease description set corresponding to each cluster representative no longer changes significantly, or when the preset maximum number of iterations is reached. Based on the target chronic disease type target value (here, the target chronic disease type target value for tuberculosis) of the target chronic disease description instance in the target chronic disease description set corresponding to the cluster representative that meets the aggregation completion status, the target chronic disease type target value of the potential chronic disease description instance in the target chronic disease description set is determined.In this example, if the potential chronic disease description instance is ultimately identified in the chronic disease description set represented by the tuberculosis cluster, then its corresponding chronic disease type target value could be inferred as "tuberculosis" or a disease type related to tuberculosis. Another aggregation method is based on a chronic disease association graph. The server first determines the sample matching degree between pairs of chronic disease description instances based on their overall descriptive features. For example, for the tuberculosis instance in the sample instance dataset and the aforementioned potential chronic disease description instance, "The patient has a long-term cough, occasionally accompanied by low-grade fever, and chest X-ray shows no obvious abnormalities, but lung disease is suspected," the server calculates the sample matching degree between them. Based on these sample matching degrees, the server constructs a chronic disease association graph. This chronic disease association graph includes the entity corresponding to each chronic disease description instance (e.g., the entity corresponding to the tuberculosis instance is "tuberculosis," and the entity corresponding to the potential chronic disease description instance is the unknown disease condition it describes) and the connections between entities, indicating the mutual influence relationship between the two connected entities. Based on this chronic disease association graph, the server performs target value knowledge reasoning on the target value of the chronic disease type (tuberculosis) of the target chronic disease description instance to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance. Specifically, in the chronic disease association graph, target value knowledge reasoning is performed on the target value of the chronic disease type of the target chronic disease description instance through the connections between entities to determine the baseline target value of the chronic disease type for the potential chronic disease description instance. For each potential chronic disease description instance, based on the entities directly connected to the potential chronic disease description instance in the chronic disease association graph, the baseline target value of the potential chronic disease description instance is recursively optimized until the target values ​​of the entities in the chronic disease association graph are optimized, thus obtaining the target value of the chronic disease type corresponding to the potential chronic disease description instance. For example, in this case, if the potential chronic disease description instance has a direct connection with the tuberculosis entity in the chronic disease association graph, then by recursively optimizing its baseline target value of the chronic disease type, it may ultimately be determined that its corresponding chronic disease type target value is "tuberculosis" or a disease type related to tuberculosis. After completing the aggregation operation, the server verifies the homogeneity of the symptom description texts for each chronic disease description instance. Taking the previously mentioned chronic disease description instance of pulmonary tuberculosis, "The patient has had a persistent fever for two weeks, accompanied by cough and sputum; chest X-ray shows shadows in the lungs, and the patient has been diagnosed with pulmonary tuberculosis," as an example, for the segmented symptom description text "The patient has had a persistent fever for two weeks," the server combines the partial descriptive features corresponding to this symptom description text (such as the duration and severity of fever) with the overall descriptive features of the chronic disease description instance to which it belongs (such as feature vectors related to lung inflammation, including the severity of lung inflammation and the source of infection). The server analyzes the relationships between these features using specific algorithms and rules.For example, if there is a correlation between the duration of fever and the degree of lung inflammation (e.g., the longer the fever lasts, the higher the degree of lung inflammation may be), then it can be considered that there is a certain degree of homogeneity between the symptom description text "patient has had a persistent fever for two weeks" and the chronic disease description instance to which it belongs. Similarly, for other symptom description texts such as "accompanied by cough" or "sputum production," the server will also use a similar method to confirm the homogeneity between them and the target chronic disease description instance of tuberculosis. Taking the potential chronic disease description instance "patient has a long-term cough, occasionally accompanied by low-grade fever, chest X-ray shows no obvious abnormalities, but lung disease is suspected" as an example, for the segmented symptom description text "patient has a long-term cough," the server will combine its corresponding partial description features (such as the frequency and severity of cough) with the overall description features of the chronic disease description instance to which it belongs (such as some feature vectors related to the lungs, including lung function features and respiratory symptom features), and analyze the relationship between these features to confirm the homogeneity between the symptom description text "patient has a long-term cough" and the chronic disease description instance to which it belongs. After confirming the homogenized information and the target values ​​of chronic disease types corresponding to the potential chronic disease description instances, the server prepares to optimize the model parameters of the chronic disease analysis model. First, the server determines the partial category cost parameters for each symptom description text based on the partial descriptive features of each symptom description text and the target values ​​of the chronic disease types of the chronic disease description instances to which they belong. For example, for the symptom description text "patient has had a persistent fever for two weeks" of the target chronic disease description instance for pulmonary tuberculosis, the partial category cost parameters for this symptom description text are calculated using a specific formula (such as based on an error function) based on its partial descriptive features (duration of fever, degree of fever, etc.) and the target value of the chronic disease type for pulmonary tuberculosis. Simultaneously, the server also calculates the overall cost parameters based on the target values ​​of the chronic disease types corresponding to the potential chronic disease description instances obtained in the aggregation operation, as well as other relevant information (such as overall descriptive features). For example, for a set of potential chronic disease description instances related to pulmonary tuberculosis, the overall cost parameters are calculated using a specific formula based on their overall descriptive features and the inferred target values ​​of the chronic disease types. The server optimizes the model parameters of the chronic disease analysis model based on the overall cost parameters and partial category cost parameters. A common optimization method is to calculate the error between the partial type distribution characteristics of the symptom description text and the preset standard distribution to obtain the discreteness parameter corresponding to the symptom description text.For example, for the symptom description text "patient has had a persistent fever for two weeks" in the chronic disease description instance of pulmonary tuberculosis, the server calculates the error between its partial type distribution characteristics (such as the distribution of fever duration and fever severity) and a preset standard distribution (such as the standard distribution of fever based on a large number of normal fever cases and pulmonary tuberculosis fever cases), obtaining the dispersion parameter corresponding to this symptom description text. Then, based on the homogeneity information corresponding to the symptom description text, the partial category cost parameter and the dispersion parameter are integrated to obtain the partial cost parameter of the symptom description text. For example, for the symptom description text "patient has had a persistent fever for two weeks," its partial category cost parameter and the previously calculated dispersion parameter are integrated through a weighted average or other method to obtain the partial cost parameter of this symptom description text. Based on the overall cost parameter and the partial cost parameter corresponding to each symptom description text of the chronic disease description instance, the model parameters of the chronic disease analysis model are optimized. For example, by adjusting the neural network weights, biases, and other parameters in the chronic disease analysis model, the model can more accurately identify the type of chronic disease in subsequent predictions. After multiple iterations and optimizations until the preset optimization stopping conditions are met (such as error less than a certain threshold, iteration count reaching a certain upper limit, etc.), the target chronic disease analysis model is finally obtained. This target chronic disease analysis model can then be used to identify chronic diseases from the descriptions of target chronic diseases, obtaining the target value of the chronic disease type for each description. When the server uses the optimized target chronic disease analysis model to identify chronic diseases from the descriptions, it obtains the target value of the chronic disease type for each description. For example, suppose the server analyzes a new chronic disease description, "Patient has recurrent cough, accompanied by wheezing, and chest X-ray shows increased lung texture," and inputs it into the target chronic disease analysis model. After the model's calculation and analysis, the target value of the chronic disease type for this description is "chronic bronchitis." Based on this target value, the server can determine that the target chronic disease type corresponding to this description is "chronic bronchitis." In this way, the server completes the entire process from acquiring the model and dataset to finally determining the target chronic disease type, providing an important basis for the accurate diagnosis and subsequent management of chronic diseases.

[0033] In this embodiment of the invention, the step of performing an aggregation operation on the overall descriptive features of each chronic disease description instance in the sample instance dataset based on the target value of the chronic disease type of the target chronic disease description instance to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance can be implemented through the following example.

[0034] Obtain multiple preset cluster representatives and confirm the matching degree between the overall descriptive features of each chronic disease description instance in the sample instance dataset and each cluster representative;

[0035] For each chronic disease description instance, based on the matching degree and the target value of the chronic disease type of the target chronic disease description instance, a target cluster representative corresponding to the chronic disease description instance is determined from each cluster representative, and the chronic disease description instance is added to the chronic disease description set corresponding to the target cluster representative;

[0036] For each cluster representative corresponding to a set of chronic disease descriptions, a chronic disease description instance that meets the preset cluster representative conditions is selected from the set of chronic disease descriptions as a new cluster representative;

[0037] Repeat the step of confirming the matching degree between the overall descriptive features of each chronic disease description instance in the sample instance dataset and each cluster representative until the determined cluster representative meets the aggregation completion state; based on the chronic disease type target value of the target chronic disease description instance in the target chronic disease description set corresponding to the cluster representative that meets the aggregation completion state, determine the chronic disease type target value of the potential chronic disease description instance in the target chronic disease description set.

[0038] In this embodiment of the invention, for example, the server first retrieves multiple pre-defined aggregated cluster representatives from its local storage area or a dedicated model configuration database. These aggregated cluster representatives are carefully selected and organized by professional medical personnel and data processing experts, and they correspond to the typical manifestations of different types of chronic diseases in terms of overall descriptive features. For example, for lung-related chronic diseases, there may be an aggregated cluster representative specifically constructed for the typical overall descriptive features of tuberculosis. This representative may include the manifestation of common symptom combinations of tuberculosis at the overall descriptive feature level, such as lung inflammation-related features (e.g., lung shadow conditions, degree of inflammation, etc.), and the manifestation of symptoms such as fever, cough, and sputum in the overall features. In addition, there is an aggregated cluster representative for chronic bronchitis, which covers the presentation of typical respiratory symptoms and lung function characteristics of chronic bronchitis in the overall descriptive features. The server extracts these different aggregated cluster representatives, preparing them for subsequent calculation of the matching degree with chronic disease description instances in the sample instance dataset. The server then begins to traverse each chronic disease description instance in the sample instance dataset. Taking a specific sample instance dataset as an example, it contains chronic disease descriptions of many different patients. Suppose the current chronic disease description instance being processed is "a patient with a persistent cough, occasionally accompanied by low-grade fever, with no obvious abnormalities on chest X-ray, but suspected lung disease." The server first obtains the overall descriptive features of this instance, which may include lung-related feature vectors, such as lung function features (e.g., vital capacity, dyspnea), and respiratory symptom features (e.g., frequency and nature of cough). Then, the server checks the matching degree between the overall descriptive features of this instance and each previously obtained cluster representative. For cluster representatives targeting tuberculosis, the server uses a specific algorithm to calculate the matching degree between them. This algorithm may comprehensively consider the similarity across various feature dimensions, such as the matching of lung inflammation-related features, and the correspondence of symptoms like fever, cough, and sputum in the overall features. Similarly, a similar matching degree calculation is performed for cluster representatives targeting chronic bronchitis. In this way, the server can accurately determine the matching degree between the overall descriptive features of each chronic disease description instance and each cluster representative. Let's continue with the example of the chronic disease description mentioned above: "The patient has a long-term cough, occasionally accompanied by a low-grade fever, and the chest X-ray shows no obvious abnormalities, but lung disease is suspected." Assume that the target chronic disease type of the target chronic disease description example has been clearly identified as "pulmonary tuberculosis" (for example, obtained from a typical case description of a patient who has been diagnosed with pulmonary tuberculosis).After calculating the matching degree between this chronic disease description instance and various cluster representatives (such as the tuberculosis cluster representative and the chronic bronchitis cluster representative), the server determines the target cluster representative corresponding to the chronic disease description instance based on these matching degrees and the target value of the chronic disease type ("tuberculosis") of the target chronic disease description instance. If the calculation shows that the instance has a relatively high matching degree with the tuberculosis cluster representative, the server will determine the tuberculosis cluster representative as the target cluster representative corresponding to this chronic disease description instance. Once the target cluster representative (here, the tuberculosis cluster representative) is determined, the server will add this chronic disease description instance of "patient with long-term cough, occasional low-grade fever, no obvious abnormalities seen on chest X-ray, but suspected lung disease" to the chronic disease description set corresponding to the tuberculosis cluster representative. This set may already contain other chronic disease description instances that are somewhat similar to tuberculosis in overall descriptive features. By continuously adding new instances, this set will gradually become richer for further analysis and processing. For the chronic disease description set corresponding to the tuberculosis cluster representative, the server selects a new representative according to preset cluster representative conditions. Suppose the preset cluster representative conditions are: select chronic disease description instances with the highest matching degree and a number reaching a certain threshold (e.g., 10) as the new cluster representative. The server analyzes and evaluates each chronic disease description instance in the set again, calculating their matching degree with the original tuberculosis cluster representative in terms of overall descriptive features. Then, based on these matching degrees and quantity requirements, it selects qualified chronic disease description instances from the set. For example, after screening, if 10 chronic disease description instances are found to have outstanding matching degrees and meet the quantity requirements, the server will determine these 10 instances as the new tuberculosis cluster representative. The purpose of this is to continuously optimize the cluster representative, making it more accurately represent the typical situation of the corresponding type of chronic disease in terms of overall descriptive features. After completing the above-mentioned new representative selection operation, the server does not stop, but continues to repeat the step of confirming the overall descriptive features of each chronic disease description instance in the sample instance dataset with the new (updated) matching degree of each cluster representative. Because with the selection of new cluster representatives, some chronic disease description instances that were initially poorly matched may now have a higher match with the new cluster representatives, or the match of instances that were initially highly matched may change. Therefore, this process needs to be repeated continuously to dynamically adjust the cluster representatives to which each chronic disease description instance belongs and the corresponding chronic disease description set. For example, after the first selection of a new cluster representative for pulmonary tuberculosis, a new chronic disease description instance, "Patient has a cough with a small amount of sputum, a prolonged low-grade fever, and a slight shadow on the lungs on chest X-ray," is added to the sample instance dataset.The server then needs to recalculate the overall descriptive features of this new instance and their matching degree with the new tuberculosis cluster representative and other cluster representatives (such as the chronic bronchitis cluster representative). Based on the matching degree, subsequent operations are performed, such as determining its corresponding target cluster representative and adding it to the appropriate set. So how is the aggregation completion status determined? Generally, the server sets some criteria. For example, when the number of instances in the chronic disease description set corresponding to each cluster representative no longer changes significantly—that is, after repeating the above steps multiple times, the number of instances in the set remains relatively stable in several consecutive checks—this can be considered a sign of aggregation completion. Alternatively, a maximum number of iterations can be set; for example, after repeating the above steps 10 times, regardless of whether the number of instances in the set is still changing, aggregation is considered complete. When these aggregation completion conditions are met, the server stops repeating the above steps and proceeds to the next step. Assume that after the above series of operations, the aggregation completion status has been met. Now let's look at the target chronic disease description set corresponding to the tuberculosis cluster representative. This set contains both target chronic disease description instances that are clearly identified as pulmonary tuberculosis (with a target chronic disease type value of "pulmonary tuberculosis") and potential chronic disease description instances (instances whose specific chronic disease type is not yet clear). The server determines the target chronic disease type value of the potential chronic disease description instances based on the target chronic disease description instances' target chronic disease type value ("pulmonary tuberculosis") in this set. Because these potential chronic disease description instances are grouped into this set based on their similarity to the target chronic disease description instances in overall descriptive features, it is reasonable to infer that their chronic disease type may be related to pulmonary tuberculosis. For example, for a potential chronic disease description instance "patient cough with a small amount of sputum, prolonged low-grade fever, chest X-ray showing slight shadows in the lungs," in this set, since the target chronic disease type value of other target chronic disease description instances in the set is "pulmonary tuberculosis," the server can preliminarily infer that the target chronic disease type value of this potential chronic disease description instance may also be "pulmonary tuberculosis" or a disease type closely related to pulmonary tuberculosis. In this way, through this aggregation operation and the inference of the target value of chronic disease type based on the target chronic disease description instance, the server can determine a relatively reasonable target value of chronic disease type for the potential chronic disease description instance, providing an important basis for subsequent chronic disease analysis and diagnosis.

[0039] In this embodiment of the invention, the step of performing an aggregation operation on the overall descriptive features of each chronic disease description instance in the sample instance dataset based on the target value of the chronic disease type of the target chronic disease description instance to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance can be implemented through the following example.

[0040] Based on the overall descriptive features of each pair of chronic disease description instances, determine the sample matching degree between each pair of chronic disease description instances;

[0041] Based on the sample matching degree, a chronic disease association map is constructed, which indicates the mutual influence relationship between each chronic disease description instance;

[0042] Based on the chronic disease association map, target value knowledge reasoning is performed on the target value of the chronic disease type of the target chronic disease description instance to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance.

[0043] In this embodiment of the invention, exemplarily, the server first acquires a sample instance dataset, which contains numerous chronic disease description instances. For example, instance A: "The patient has a long-term cough with sputum, occasional low-grade fever, and chest X-ray shows increased lung texture," and instance B: "The patient has recurrent cough, severe wheezing, occasional chest tightness, and abnormal pulmonary function tests." For each pair of chronic disease description instances, the server extracts their overall descriptive features. For instance A, the overall descriptive features may involve lung symptom features, fever, etc.; for instance B, the overall descriptive features include cough and wheezing characteristics, chest tightness, and lung function-related features. Then, using a specific algorithm (such as a feature vector-based similarity calculation algorithm), the server determines the sample matching degree between the pairs of chronic disease description instances based on these overall descriptive features. Taking instances A and B as an example, the server will compare their similarity in lung symptoms, respiratory status, and other features in detail, calculating a specific sample matching degree value, such as 0.6 (the value is only an example and may vary depending on the algorithm), to quantify their similarity. After obtaining the sample matching degrees between all pairs of chronic disease description instances, the server begins to construct a chronic disease association graph. In this graph, each chronic disease description instance is considered an entity. For example, instance A and instance B are two different entities. Based on the calculated sample matching degree, when the sample matching degree reaches a certain threshold (e.g., 0.5), a connection is drawn between the corresponding two entities to indicate that there is a mutual influence relationship between these two chronic disease description instances. For example, if the sample matching degree between instance A and instance C is 0.7, exceeding the threshold, then a connection is drawn between the entities corresponding to instance A and instance C in the graph. As the matching degree between all instances is processed, a complete chronic disease association graph is gradually constructed, clearly showing the association between each chronic disease description instance. Assume the target chronic disease description instance is instance D, and its target value for chronic disease type is "chronic bronchitis". The server performs target value knowledge inference in the chronic disease association graph, starting from instance D. It infers the mutual influence relationship represented by the connections between instance D and other entities. For example, if instance D and instance E are connected in the graph, it indicates that they have a mutual influence relationship. Based on the target value of chronic disease type "chronic bronchitis" for instance D, and combined with the knowledge contained in the graph (such as the evolution pattern of disease types in similar associations in past data), the server initially infers the baseline target value of chronic disease type for the potential chronic disease description instance corresponding to instance E. Then, for each potential chronic disease description instance (such as instance E), based on the entities directly connected to it in the chronic disease association graph (which may be connected to other instances besides instance D), the server recursively optimizes its baseline chronic disease type target value.For example, instance E is also connected to instance F. By combining relevant information from instance F, the inference of the chronic disease type of instance E is further improved until the target value of the entity in the chronic disease association graph is optimized. Finally, the target value of the chronic disease type corresponding to potential chronic disease description instances such as instance E is obtained, which may also be "chronic bronchitis" or a related disease type. Through the above steps, the server completes the reasoning and determination process from the target value of the chronic disease type of the target chronic disease description instance to the target value of the chronic disease type corresponding to the potential chronic disease description instance with the help of the chronic disease association graph.

[0044] In this embodiment of the invention, the chronic disease association map includes an entity corresponding to each chronic disease description instance and a connection between entities, wherein the connection indicates the mutual influence relationship between two connected entities;

[0045] The step of performing target value knowledge reasoning on the target value of the chronic disease type of the target chronic disease description instance based on the chronic disease association map to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance can be implemented through the following example.

[0046] In the chronic disease association graph, target value knowledge reasoning is performed on the target value of chronic disease type of the target chronic disease description instance by connecting the entities, so as to determine the baseline target value of chronic disease type of potential chronic disease description instances;

[0047] For each potential chronic disease description instance, based on the entities directly connected to the potential chronic disease description instance in the chronic disease association graph, the target value of the baseline chronic disease type of the potential chronic disease description instance is recursively optimized until the target value of the entities in the chronic disease association graph is optimized, thus obtaining the target value of the chronic disease type corresponding to the potential chronic disease description instance.

[0048] In this embodiment of the invention, for example, the server first obtains a pre-constructed chronic disease association graph. Each chronic disease description instance in the graph corresponds to an entity, and the lines between entities represent mutual influence relationships. Assume the entity corresponding to the target chronic disease description instance is A, and its target chronic disease type is "diabetes". The server examines other entities connected to entity A, such as entity B and entity C. These connections indicate mutual influence relationships between them. Based on accumulated medical knowledge and analysis of associations in numerous similar cases, when entity A is connected to entity B, the server infers based on entity A's "diabetes" type target value. For example, it is known that poor long-term blood sugar control in diabetic patients may lead to kidney-related problems. If the chronic disease description instance corresponding to entity B exhibits some characteristics of abnormal kidney function, the server can preliminarily infer that the baseline chronic disease type target value of the potential chronic disease description instance corresponding to entity B may be related to "diabetic nephropathy". Similarly, for entity C connected to entity A, if its corresponding chronic disease description instance includes descriptions related to eye diseases, and considering the knowledge that diabetes can cause eye complications, the server can infer that the baseline chronic disease type target value for the potential chronic disease description instance corresponding to entity C is likely to be "diabetic retinopathy". Take the previously inferred potential chronic disease description instance corresponding to entity B (with a baseline chronic disease type target value of "diabetic nephropathy") as an example. The server further examines other entities directly connected to entity B in the chronic disease association graph, assuming there are entities D, E, etc. If the chronic disease description instance corresponding to entity D has hypertension-related features, and diabetic nephropathy patients often have hypertension as a complication, then the server will optimize the baseline chronic disease type target value for the potential chronic disease description instance corresponding to entity B based on this situation. For example, it might further clarify that its chronic disease type target value is "diabetic nephropathy with hypertension". Next, consider entity E. If its corresponding chronic disease description instance exhibits new symptom characteristics, such as worsening edema, the server, combining the existing type "diabetic nephropathy with hypertension" and related medical knowledge, further optimizes it, continuously refining the target value of the chronic disease type for this potential chronic disease description instance. The server will perform this process on all potential chronic disease description instances in the graph, continuously recursively optimizing the target value based on their directly connected entities and related features, until the target value optimization of all entities in the entire chronic disease association graph is completed. Ultimately, the server accurately obtains the target value of the chronic disease type corresponding to each potential chronic disease description instance. For example, the final target value of the chronic disease type for entity B is determined to be "diabetic nephropathy with hypertension and worsening edema." Through this process, the server, using the chronic disease association graph, completes the accurate reasoning and optimization determination of the target value of the chronic disease type corresponding to potential chronic disease description instances.

[0049] In this embodiment of the invention, the process of determining the homogeneity information between the symptom description text and the chronic disease description instance to which the symptom description text belongs, based on the partial description features corresponding to the symptom description text and the overall description features of the chronic disease description instance to which the symptom description text belongs, can be implemented through the following example.

[0050] Based on the overall descriptive features of each chronic disease description instance, an overall descriptive feature set is constructed, and based on the partial descriptive features of the symptom description text of each chronic disease description instance, a partial descriptive feature set is constructed.

[0051] For each chronic disease description instance, at least one similar overall description feature corresponding to the overall description feature of the chronic disease description instance is extracted from the overall description feature set. Based on the target value of the chronic disease type of the chronic disease description instance to which the similar overall description feature belongs, the set of undetermined target values ​​for the chronic disease description instance is determined.

[0052] For each symptom description text of the chronic disease description instance, at least one similar partial description feature corresponding to the partial description feature of the symptom description text is extracted from the partial description feature set. Based on the chronic disease type target value of the chronic disease description instance to which the similar partial description feature belongs, the set of undetermined target values ​​of the symptom description text is determined.

[0053] Based on the set of undetermined target values ​​for the symptom description text and the set of undetermined target values ​​for the chronic disease description instance to which the symptom description text belongs, the homogeneity information between the symptom description text and the chronic disease description instance to which it belongs is confirmed.

[0054] In this embodiment of the invention, for example, the server first obtains a large number of chronic disease description instances, which come from a previously collected and organized sample instance dataset. For example, one chronic disease description instance is about a diabetic patient: "The patient experiences polydipsia, polyphagia, and polyuria, with gradual weight loss, persistently high blood sugar, mild blurred vision, and occasional numbness in the feet." For this instance, the server first extracts its overall descriptive features. For the aforementioned diabetic instance, the overall descriptive features may include features related to metabolic disorders (such as abnormal blood sugar), physical symptom features (such as polydipsia, polyphagia, and polyuria), and some complication-related features (such as blurred vision and foot numbness). After organizing the overall descriptive features of this instance, it is placed into an overall descriptive feature set. Next, the server performs a segmentation operation on the diabetic instance to obtain multiple symptom description texts, such as "the patient experiences polydipsia," "the patient experiences polyphagia," "the patient experiences polyuria," "persistently high blood sugar," "accompanied by mild blurred vision," and "occasionally numbness in the feet." Then, a feature extraction operation is performed on each symptom description text using a chronic disease analysis model to obtain partial descriptive features for each symptom description text. For example, the partial descriptive features of the symptom description text "patient with excessive thirst" might involve quantitative features of water intake, water intake frequency, etc. After organizing the partial descriptive features of each symptom description text, they are placed into a partial descriptive feature set. The server repeats the above operation for each chronic disease description instance in the sample instance dataset, gradually constructing a complete overall descriptive feature set and a partial descriptive feature set. Taking the diabetes chronic disease description instance as an example again, the server searches within the already constructed overall descriptive feature set. It uses a specific similarity calculation algorithm to find at least one similar overall descriptive feature corresponding to the overall descriptive features of the diabetes instance. For example, it might find that the overall descriptive features of another instance have a high similarity to the overall features of the diabetes instance in terms of abnormal blood sugar, physical symptoms, etc. Suppose that the chronic disease description instance to which this similar overall descriptive feature belongs is about another type of diabetes (such as type 2 diabetes, and its chronic disease type target value is explicitly "type 2 diabetes"). Since such a similar overall descriptive feature has been found, the server will add "type 2 diabetes" as a potential target value for this diabetes chronic disease description instance to the potential target value set. The server continues to search for other possible similar overall descriptive features in the overall descriptive feature set, and continuously adds new pending target values ​​to the pending target value set based on the chronic disease type target value of the chronic disease description instance to which it belongs. For example, if another instance similar to the diabetes instance in certain symptoms and metabolic characteristics is found, and its chronic disease type target value is "special type of diabetes with complications", this value will also be added to the pending target value set. For the symptom description text "patient with polydipsia" in the diabetes instance, the server searches in the partial descriptive feature set.The server uses a similarity calculation algorithm to find at least one similar partial descriptive feature corresponding to the partial descriptive features of the symptom description "patient with excessive thirst". Suppose a similar partial descriptive feature is found to belong to a chronic disease description instance, and its chronic disease type target value is "a certain type of diabetes with worsening excessive thirst". Then, the server will add this chronic disease type target value as a potential target value to the symptom description text "patient with excessive thirst" and place it in the potential target value set. Similarly, for other symptom description texts such as "patient with excessive hunger" and "patient with excessive urination", the server will also search for corresponding similar partial descriptive features in the partial descriptive feature set and determine the potential target value set for each symptom description text based on the chronic disease type target value of the corresponding chronic disease description instance. Taking the symptom description text "patient with excessive thirst" as an example, the server has already determined its potential target value set, which may include phrases such as "a certain type of diabetes with worsening excessive thirst" and "another type of diabetes with frequent excessive thirst". Simultaneously, the chronic disease description instance to which this symptom description text belongs also has its own potential target value set, which may include phrases such as "type 2 diabetes" and "a special type of diabetes with complications". The server will analyze the relationship between these two potential target value sets. If two sets contain many identical or similar target values ​​for chronic disease types—for example, both sets have target values ​​related to a certain type of diabetes and involving polydipsia—then it can be considered that there is a high degree of homogeneity between the symptom description text "patient polydipsia" and the corresponding chronic disease description instance of diabetes. Conversely, if the target values ​​in the two sets differ significantly, with almost no overlap or similarity, then the homogeneity between them is weak. In this way, the server analyzes each symptom description text for each chronic disease description instance, accurately confirming the homogeneity between the symptom description text and the corresponding chronic disease description instance, providing crucial information for subsequent model optimization and other operations.

[0055] In this embodiment of the invention, the process of confirming the homogeneity information between the symptom description text and the chronic disease description instance to which the symptom description text belongs, based on the set of undetermined target values ​​of the symptom description text and the set of undetermined target values ​​of the chronic disease description instance to which the symptom description text belongs, can be implemented through the following example.

[0056] Statistical analysis is performed on the target values ​​of chronic disease types in the set of undetermined target values ​​of the symptom description text to determine the distribution characteristics of the target values ​​of the symptom description text.

[0057] Statistical analysis is performed on the target values ​​of chronic disease types in the set of undetermined target values ​​of the chronic disease description instances to which the symptom description text belongs, to determine the distribution characteristics of the target values ​​of the chronic disease description instances to which the symptom description text belongs;

[0058] Based on the target value distribution characteristics of the symptom description text and the target value distribution characteristics of the chronic disease description instance to which the symptom description text belongs, the homogeneity information between the symptom description text and the chronic disease description instance to which it belongs is confirmed.

[0059] In this embodiment of the invention, for example, the server first obtains a set of pending target values ​​for a symptom description text. For instance, for the symptom description text "patient has a persistent cough," its set of pending target values ​​may include target values ​​for chronic disease types such as "chronic bronchitis," "pneumonia," and "cough variant asthma." The server performs statistical analysis on the chronic disease type target values ​​in this set. It counts the frequency of each chronic disease type target value. For example, the statistics show that "chronic bronchitis" appears 5 times, "pneumonia" appears 3 times, and "cough variant asthma" appears 2 times. Simultaneously, the server also analyzes the distribution of these target values, such as whether they are concentrated in a few types or are relatively evenly distributed. In this example, the frequency of "chronic bronchitis" is relatively high, indicating that this type is more likely to appear in the pending analysis related to "patient's persistent cough." Therefore, it is determined that the target value distribution characteristic of the symptom description text "patient's persistent cough" is relatively concentrated in "chronic bronchitis," but other possible types also exist. Suppose the symptom description text "patient with persistent cough" belongs to a chronic disease description instance that is a detailed description of a patient's long-term respiratory discomfort. Its set of pending target values ​​includes target values ​​for chronic disease types such as "chronic bronchitis," "bronchiectasis," and "chronic obstructive pulmonary disease." The server also performs statistical analysis on the target values ​​for chronic disease types in this set. The frequency of each target value is counted; for example, "chronic bronchitis" appears 4 times, "bronchiectasis" appears 2 times, and "chronic obstructive pulmonary disease" appears 3 times. Further analysis of their distribution reveals that the frequencies of "chronic bronchitis" and "chronic obstructive pulmonary disease" are relatively similar and the overall distribution is relatively even. Therefore, it is determined that the target value distribution characteristics of this chronic disease description instance are relatively balanced among these common respiratory chronic disease types, although "chronic bronchitis" is slightly more likely. The server then compares the target value distribution characteristics of the symptom description text "patient with persistent cough" and its corresponding chronic disease description instance. The target value distribution characteristics of the symptom description text are relatively concentrated on "chronic bronchitis," while the target value distribution characteristics of the chronic disease description instances show a more balanced distribution across several common respiratory chronic disease types, but "chronic bronchitis" is slightly more likely. It can be seen that both have a high probability of pointing to "chronic bronchitis," indicating a certain degree of homogeneity between the symptom description text "patient's persistent cough" and the chronic disease description instance to which it belongs. That is, among the various possible chronic disease types involved in this chronic disease description instance, the symptom has a relatively strong association with "chronic bronchitis." In this way, the server performs similar analysis on each symptom description text and its associated chronic disease description instance, thereby accurately confirming the homogeneity between them and providing a basis for subsequent related processing.

[0060] In this embodiment of the invention, the optimization of the model parameters of the chronic disease analysis model based on the homogenized information and the target value of the chronic disease type corresponding to the potential chronic disease description instance to obtain the target chronic disease analysis model can be implemented through the following example.

[0061] Based on the overall descriptive features of the chronic disease description instance, chronic disease identification is performed on the chronic disease description instance to obtain the overall type distribution characteristics corresponding to the chronic disease description instance;

[0062] Based on the partial descriptive features corresponding to the symptom description text, chronic disease identification is performed on the symptom description text to obtain the partial type distribution characteristics corresponding to the symptom description text.

[0063] Based on the overall type distribution characteristics, the partial type distribution characteristics, the homogenization information, and the target value of the chronic disease type corresponding to the potential chronic disease description instance, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model.

[0064] In this embodiment of the invention, for example, the server obtains a chronic disease description instance, such as "the patient has a long-term cough with sputum, occasional low-grade fever, chest X-ray shows increased lung texture, and some decline in lung function." The server extracts the overall descriptive features of this instance, including lung symptom features, fever, and lung function-related features. Then, using an existing chronic disease analysis model (in the basic version before optimization), the server identifies the chronic disease based on these overall descriptive features. After analysis and calculation by the model, the overall type distribution characteristics corresponding to the instance are obtained. For example, the model determines that the probability of this instance belonging to "chronic bronchitis" is 30%, the probability of belonging to "pneumonia" is 25%, and the probability of belonging to "other respiratory chronic diseases" is 45%, etc., thus clarifying the distribution of different chronic disease types at the overall level, i.e., the overall type distribution characteristics. For the above chronic disease description instance, the server segments it into multiple symptom description texts, such as "the patient has a long-term cough," "with sputum," "occasionally low-grade fever," "chest X-ray shows increased lung texture," "some decline in lung function," etc. For each symptom description text, the server extracts its corresponding partial descriptive features. For example, partial descriptive features for "patient with chronic cough" might include the duration, frequency, and severity of the cough. Then, using a chronic disease analysis model, chronic disease identification is performed on each symptom description text based on these partial descriptive features. Taking "patient with chronic cough" as an example, the model analysis reveals that the partial type distribution characteristics corresponding to this symptom description text might be: 40% probability of belonging to "chronic bronchitis," 30% probability of belonging to "cough variant asthma," and 30% probability of belonging to "other respiratory-related diseases," thus clarifying the distribution of each symptom description text across different chronic disease types. The server then comprehensively considers all of the above information. For example, regarding the previously mentioned chronic disease description instance, we know its overall type distribution characteristics, the partial type distribution characteristics of each symptom description text, and the homogeneity information between the symptom description text and the chronic disease description instance to which it belongs (assuming that the analysis in the previous steps shows that the symptom description text "patient has a long-term cough" has high homogeneity with the entire instance in terms of chronic disease type). We also know the target value of the chronic disease type corresponding to the potential chronic disease description instance (assuming that through previous aggregation and other operations, it is inferred that the instance may belong to the "chronic bronchitis" related type). Based on this information, the server adjusts the model parameters of the chronic disease analysis model using a specific optimization algorithm.For example, if a symptom description text is found to have a high probability of pointing to "chronic bronchitis" in certain type distribution characteristics, and its homogeneity information with the overall instance also indicates a close association with "chronic bronchitis," but the model currently has a slightly lower probability of classifying the instance as "chronic bronchitis," then the weights and other parameters related to the symptom description text and the overall instance in the model will be appropriately adjusted. This will allow the model to more accurately identify the likelihood of the instance belonging to "chronic bronchitis" in subsequent analyses. By continuously performing such analyses and parameter adjustments on different chronic disease description instances, and through multiple rounds of iterative optimization, until the model achieves good accuracy and stability in the identification of various chronic diseases, the target chronic disease analysis model is finally obtained. This model can more accurately identify and analyze new chronic disease descriptions and arrive at more accurate chronic disease type judgments.

[0065] In this embodiment of the invention, the optimization of the model parameters of the chronic disease analysis model based on the overall type distribution characteristics, the partial type distribution characteristics, the homogenization information, and the target value of the chronic disease type corresponding to the potential chronic disease description instance to obtain the target chronic disease analysis model can be implemented through the following example.

[0066] For each chronic disease description instance, based on the partial type distribution characteristics of each symptom description text in the chronic disease description instance and the homogenization information corresponding to each symptom description text, the target value of the chronic disease type of the chronic disease description instance is optimized to obtain the optimized target value of the chronic disease type of the chronic disease description instance.

[0067] Based on the overall type distribution characteristics of each chronic disease description instance, the optimized chronic disease type target values ​​of each chronic disease description instance are integrated to obtain the overall cost parameter.

[0068] Based on the overall cost parameter, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model.

[0069] In this embodiment of the invention, for example, the server obtains a specific chronic disease description instance, such as "The patient has recurrent cough, accompanied by wheezing, and the sputum is sometimes yellow. Chest X-ray shows increased and blurred lung texture, and mildly impaired lung function." First, the server segments this instance into multiple symptom description texts, such as "the patient has recurrent cough," "accompanied by wheezing," "the sputum is sometimes yellow," "chest X-ray shows increased and blurred lung texture," and "mildly impaired lung function." For each symptom description text, the server has already obtained its partial type distribution characteristics through previous steps. For example, the partial type distribution characteristics corresponding to "the patient has recurrent cough" might be: a 40% probability of belonging to "chronic bronchitis," a 30% probability of belonging to "asthma," and a 30% probability of belonging to "other respiratory diseases." Simultaneously, the homogeneity information between each symptom description text and the chronic disease description instance is also clarified. For example, the symptom description text "the patient has recurrent cough" has a high degree of homogeneity with the entire instance in pointing to "chronic bronchitis," that is, from the perspective of the overall symptom combination and correlation, this symptom is relatively consistent with the overall instance in suggesting possible chronic bronchitis. Based on these factors, the server comprehensively considers the partial type distribution characteristics and homogeneity information of each symptom description text to optimize the target value of the chronic disease type for each chronic disease description instance. For example, if the previously inferred target value for the chronic disease type of this instance was "suspected chronic bronchitis or other respiratory disease," but analysis reveals that several key symptom description texts, such as "patient's recurrent cough," strongly point to "chronic bronchitis" in terms of partial type distribution characteristics and are supported by homogeneity information, then the target value for the chronic disease type of this instance is optimized to "chronic bronchitis," resulting in the optimized target value. The server has multiple chronic disease description instances; for example, in addition to the respiratory symptom instance mentioned earlier, there is also a chronic disease description instance related to cardiovascular diseases. For each chronic disease description instance, the optimized target value for the chronic disease type has already been obtained in the previous step. Simultaneously, the server also understands the overall type distribution characteristics corresponding to each chronic disease description instance. For example, the respiratory instance that was just optimized to "chronic bronchitis" might have the following overall type distribution characteristics: a 70% probability of "chronic bronchitis," a 20% probability of "asthma," and a 10% probability of "other respiratory diseases." The server integrates the optimized chronic disease type target values ​​for each chronic disease description instance based on these overall type distribution characteristics. For instance, it comprehensively considers the distribution of different instances across their respective optimized chronic disease type target values ​​to calculate an overall cost parameter. This overall cost parameter might involve a comprehensive measurement of factors such as the overall distribution differences of different chronic disease types across all instances, and the degree of deviation between the optimized target value and the original inferred value. A specific numerical value is derived using a specific calculation formula to represent the overall cost.After receiving the calculated overall cost parameter, the server uses it to optimize the model parameters of the chronic disease analysis model. Assume the chronic disease analysis model has some key model parameters, such as weights and biases in the neural network. If the overall cost parameter indicates a significant deviation in the current model's judgment of chronic disease types (e.g., a large overall cost parameter value suggests a large difference between the optimized target value for chronic disease types in each instance and the actual situation), then the server will adjust these model parameters using specific optimization algorithms. For example, if it is found that the overall cost parameter is mainly due to inaccurate processing of certain symptom description texts, the server will specifically adjust the model parameters (such as weights) related to these symptom description texts, enabling the model to more accurately determine the chronic disease type based on the symptom description text and overall instance features in subsequent processing. By continuously adjusting and optimizing the model parameters based on the overall cost parameters, and through multiple iterations, until the overall cost parameters reach a relatively ideal level (e.g., the value is small enough, indicating that the model has good accuracy and stability in judging chronic disease types), the target chronic disease analysis model is finally obtained. This model can more accurately analyze and judge new chronic disease descriptions.

[0070] In this embodiment of the invention, the optimization of the model parameters of the chronic disease analysis model based on the overall cost parameter to obtain the target chronic disease analysis model can be implemented through the following example.

[0071] Based on the partial type distribution characteristics of the symptom description text and the target value of the chronic disease type of the chronic disease description instance to which the symptom description text belongs, the partial category cost parameters of the symptom description text are identified.

[0072] Based on the overall cost parameter and the partial category cost parameter, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model.

[0073] In an embodiment of the invention, for example, the server obtains a chronic disease description instance, such as "The patient has a persistent cough with sputum, occasional low-grade fever, and chest X-ray shows increased lung texture." A segmented symptom description text is "The patient has a persistent cough." Previously, partial type distribution characteristics of "The patient has a persistent cough" have been obtained, such as a 40% probability of belonging to "chronic bronchitis," a 30% probability of belonging to "pneumonia," and a 30% probability of belonging to "other respiratory diseases." Simultaneously, the target value of the chronic disease type of the chronic disease description instance to which this symptom description text belongs is known to be "chronic bronchitis." The server uses a specific calculation method to determine the partial category cost parameter of the symptom description text "The patient has a persistent cough." For example, if the cost is measured by the degree of difference between the predicted result and the target value, when the probability of predicting "chronic bronchitis" is 40%, and the target value is "chronic bronchitis," the difference is relatively small, and the calculated partial category cost parameter value is relatively small; if the probability of predicting other disease types is high, and the difference from the target value is large, then the partial category cost parameter value will be large. Through this calculation, a partial category cost parameter is determined for each symptom description text. The server has already calculated the partial category cost parameters for each symptom description text, as well as the overall cost parameter (which is obtained by integrating the optimized chronic disease type target values ​​for each chronic disease description instance based on the overall type distribution characteristics corresponding to each chronic disease description instance). Assuming the chronic disease analysis model is a neural network-based model, it contains many model parameters such as weights and biases. The server optimizes these model parameters based on the overall cost parameter and the partial category cost parameter. If the overall cost parameter shows the overall prediction bias, and the partial category cost parameter indicates the prediction bias related to a specific symptom description text, then the model parameters are adjusted accordingly. For example, if a partial category cost parameter for a certain symptom description text is found to be large, it indicates that the prediction performance of that symptom description text in the model is poor. The server will then look for model parameters related to the processing of that symptom description text, such as the corresponding neural network weights. Through a specific optimization algorithm, the value of this weight is appropriately reduced or increased, so that the model can more accurately predict the chronic disease type to which the symptom description text belongs when processing it subsequently. A comprehensive analysis of all partial category cost parameters and overall cost parameters is conducted, and the model parameters are continuously adjusted. After multiple rounds of iterative optimization, the overall cost parameters and partial category cost parameters reach a relatively ideal level (i.e., the model has good accuracy and stability in predicting chronic disease types). Finally, the target chronic disease analysis model is obtained, which can more accurately analyze and judge new chronic disease descriptions.

[0074] In this embodiment of the invention, the optimization of the model parameters of the chronic disease analysis model based on the overall cost parameter and the partial category cost parameter to obtain the target chronic disease analysis model can be implemented through the following example.

[0075] Error calculation is performed between the partial type distribution characteristics of the symptom description text and the preset standard distribution to obtain the discreteness parameter corresponding to the symptom description text;

[0076] Based on the homogenization information corresponding to the symptom description text, the partial category cost parameters and the discreteness parameters are integrated to obtain the partial cost parameters of the symptom description text;

[0077] Based on the overall cost parameter and the partial cost parameter corresponding to each symptom description text of the chronic disease description instance, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model.

[0078] In this embodiment of the invention, for example, the server selects symptom description text from a chronic disease description instance. For instance, in the instance "The patient has recurrent coughing, accompanied by wheezing, sometimes with yellow sputum, chest X-ray shows increased and blurred lung texture, and mildly impaired lung function," "the patient has recurrent coughing" is taken as the example symptom description text. The partial type distribution characteristics of this "patient's recurrent cough" have been previously obtained, assuming a 40% probability of belonging to "chronic bronchitis," a 30% probability of belonging to "asthma," and a 30% probability of belonging to "other respiratory diseases." The server has a preset standard distribution, which is a standard distribution of cough symptoms among various chronic diseases derived from a large amount of accurately diagnosed case data. For example, in the preset standard distribution, "chronic bronchitis" accounts for 50%, "asthma" accounts for 25%, and "other respiratory diseases" accounts for 25%. The server compares and calculates the partial type distribution characteristics of "patient's recurrent cough" with the preset standard distribution using a specific error calculation method (such as mean squared error). For example, the differences in the "chronic bronchitis" category and the differences in other categories are calculated. These differences are then combined to obtain a specific numerical value, which is the dispersion parameter corresponding to the symptom description text "patient with recurrent cough." This value reflects the degree of deviation between the actual distribution characteristics of the symptom description text and the preset standard distribution. Continuing with the example of "patient with recurrent cough," the server has already determined some of its category cost parameters (assuming that the value calculated in the previous steps is a specific value, reflecting the degree of difference between the predicted result of the symptom description text and the target value of the chronic disease type of the corresponding chronic disease description instance). The dispersion parameter, which was just calculated, is also available. Furthermore, the homogeneity information between the symptom description text "patient with recurrent cough" and the chronic disease description instance to which it belongs has been identified (for example, analysis shows that the symptom and the entire instance have a high degree of homogeneity in pointing to "chronic bronchitis"). Based on this homogeneity information, the server uses a specific integration method (such as weighted average) to integrate the partial category cost parameters and the dispersion parameter. For example, if homogenized information indicates that the symptom is closely related to the overall instance in terms of chronic disease type judgment, then some category cost parameters may be given relatively high weights during integration. After integration calculation, a partial cost parameter is obtained for the symptom description text "patient has recurrent cough," which comprehensively considers factors such as prediction differences, distribution deviations, and homogenized information. The server has an overall cost parameter (which is obtained by integrating the optimized chronic disease type target values ​​of each chronic disease description instance based on the overall type distribution characteristics corresponding to each chronic disease description instance), and a partial cost parameter corresponding to each symptom description text in each chronic disease description instance (such as the partial cost parameter for "patient has recurrent cough" just calculated). Assume that the chronic disease analysis model is based on a neural network architecture and contains many model parameters such as weights and biases.The server optimizes model parameters based on the overall cost parameters and the partial cost parameters corresponding to each symptom description text. If the overall cost parameters indicate a certain bias in the overall model's judgment of chronic disease types, and the partial cost parameters of a certain symptom description text are large, it indicates that the prediction effect related to that symptom description text is poor. For example, if the partial cost parameter for "patients have recurrent coughs" is found to be large, the server will look for model parameters related to processing this symptom description text, such as the corresponding neural network weights. Through specific optimization algorithms, these weights and other model parameters are adjusted, such as appropriately decreasing or increasing weight values, so that the model can more accurately predict the chronic disease type when processing this symptom description text in subsequent processes. A comprehensive analysis of all partial cost parameters and overall cost parameters is performed, and the model parameters are continuously adjusted. After multiple rounds of iterative optimization, until both the overall cost parameters and partial cost parameters reach a relatively ideal level (i.e., the model has good accuracy and stability in chronic disease type prediction), the target chronic disease analysis model is finally obtained, which can more accurately analyze and judge new chronic disease descriptions.

[0079] In addition, a visual symptom input interface is provided on interactive terminals (including smartphones, tablets, desktop computers, and smart wearable devices) to obtain the patient's primary chronic disease symptom description. This interface offers multiple common chronic disease symptom options and a custom symptom input box. First, the common chronic disease symptom options guide the patient to make a selection, and the selected symptom becomes part of the primary chronic disease symptom description. Then, the text content entered by the patient in the custom symptom input box is obtained as a supplementary primary chronic disease symptom description. Finally, these two parts are integrated to obtain a complete primary chronic disease symptom description. A dedicated chronic disease diagnosis and recording interface is set up on the terminal device used by doctors, including a symptom description text box, symptom severity labeling options, and a portal for uploading relevant examination results. Doctors enter their observations and analyses of the patient's chronic disease symptoms in the symptom description text box as the text portion. Based on the severity of the patient's symptoms, they label the symptoms in the labeling options to obtain severity labeling information and associate it with the text portion. If relevant examination results are available, they are uploaded through the portal. The examination results, text portion, and severity labeling information are integrated to obtain a complete secondary chronic disease symptom description. Before integrating the descriptions of the first and second chronic disease symptoms, the pre-defined large language model performs semantic cleanup on both, including removing duplicate expressions, correcting typos, and standardizing medical terminology. The semantically cleaned symptom descriptions are then input into a pre-defined format conversion module. This module sets corresponding format conversion rules based on different large language models, converting the symptom descriptions into a format that meets the input requirements of the large language model. The integrated target chronic disease description is then input into a pre-trained target chronic disease analysis model to obtain the corresponding target chronic disease type. The target chronic disease type is then compared and verified with a pre-defined chronic disease type standard library (which stores standard definitions and feature information for various chronic diseases). If discrepancies exist, the target chronic disease description is re-input into the target chronic disease analysis model for analysis, while recording relevant information about the discrepancies from the first analysis. Finally, the target chronic disease type is updated based on the results of the re-analysis, and the update status and related records are pushed to the pre-defined management interface on the interactive terminal for viewing. A dedicated feedback input interface is provided for doctors on the interactive terminal, including input boxes for treatment plans, medication suggestions, dietary precautions, and rehabilitation guidance. Doctors input specific treatment plans, medication recommendations, dietary precautions, and rehabilitation guidance for their target chronic disease type into these input boxes. This information is then integrated to generate doctor-side feedback for the target chronic disease type. The target chronic disease type and doctor-side feedback are then input into a pre-defined large language model to obtain management recommendations for the target chronic disease type.Management suggestions are categorized into treatment management suggestions, medication management suggestions, dietary management suggestions, and rehabilitation management suggestions, with different priority weights assigned to each category. For example, critical treatment management suggestions involving patient life safety are given the highest priority weight. The suggestions are then sorted according to their priority weights and pushed to the preset interactive interface of the interactive terminal in priority order. A reminder function is set in the preset interactive interface, allowing for timed reminders based on information such as time nodes and follow-up cycles in the management suggestions. When the set time is reached, a reminder window pops up on the interactive terminal displaying the relevant reminder content, and the reminder information is sent to the terminal device of the patient's family member or guardian to assist the patient in performing relevant operations and following precautions. After receiving the management suggestions, the preset large language model compares and analyzes them with management suggestions for similar chronic disease cases in the historical case database (which stores a large amount of processed chronic disease case information and corresponding management suggestions). It extracts management suggestions for similar cases, statistically analyzes the differences, optimizes and adjusts the management suggestions based on the results, and then pushes them back to the preset interactive interface. Different types of interactive terminals have different preset interactive interface designs. Smartphones or tablets adopt an interface design adapted to mobile devices, featuring convenient touch operation and a clear information display layout; desktop computers adopt an interface design adapted to computer screens, with a complete functional operation area and a detailed information display area; smart wearable devices adopt a simplified interface design, highlighting the display of key information and simple operation commands to adapt to the characteristics of small screens.

[0080] In summary, this method for collecting, analyzing, and managing chronic disease medical data encompasses obtaining symptom descriptions from both patients and doctors, processing them through a series of steps such as semantic cleaning, format conversion, and analysis models to determine the type of chronic disease, generating management suggestions based on doctor feedback, and optimizing, sorting, pushing, and setting reminders for these suggestions through various methods. Furthermore, it fully considers the characteristics of different interactive terminals for adaptation, aiming to provide comprehensive, accurate, and convenient services for the diagnosis and management of chronic diseases.

[0081] This invention provides a cloud platform 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the cloud platform 100 executes the aforementioned method for collecting, analyzing, and managing chronic disease medical data. Figure 2 As shown, Figure 2 This is a structural block diagram of a cloud platform 100 provided in an embodiment of the present invention. The cloud platform 100 includes a memory 111, a processor 112, and a communication unit 113. To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0082] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A method for collecting, analyzing, and managing medical data related to chronic diseases, characterized in that, include: The system obtains the patient's first chronic disease symptom description and the doctor's second chronic disease symptom description through an interactive terminal. The first chronic disease symptom description and the second chronic disease symptom description are input into a preset large language model and integrated to obtain the target chronic disease description; The target chronic disease description is input into a pre-trained target chronic disease analysis model to obtain the target chronic disease type corresponding to the target chronic disease description; The interactive terminal is used to obtain feedback information from the doctor regarding the target chronic disease type. The target chronic disease type and the feedback information are input into the preset large language model to obtain management suggestions for the target chronic disease type, and the management suggestions are pushed to the preset interactive interface of the interactive terminal.

2. The method according to claim 1, characterized in that, The target chronic disease analysis model is obtained through the following methods: Obtain a chronic disease analysis model and a sample instance dataset. The chronic disease description instances in the sample instance dataset include target chronic disease description instances and potential chronic disease description instances. The target chronic disease description instances are configured with preset chronic disease type target values. For each chronic disease description instance, the chronic disease description instance is segmented to obtain multiple symptom description texts for the chronic disease description instance; Using the chronic disease analysis model, feature extraction operations are performed on the chronic disease description instance and each symptom description text of the chronic disease description instance to obtain the overall description features of the chronic disease description instance and the partial description features corresponding to each symptom description text. Based on the target value of the chronic disease type of the target chronic disease description instance, an aggregation operation is performed on the overall description features of each chronic disease description instance in the sample instance dataset to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance. Based on the overall descriptive features of each chronic disease description instance, an overall descriptive feature set is constructed, and based on the partial descriptive features of the symptom description text of each chronic disease description instance, a partial descriptive feature set is constructed. For each chronic disease description instance, at least one similar overall description feature corresponding to the overall description feature of the chronic disease description instance is extracted from the overall description feature set. Based on the target value of the chronic disease type of the chronic disease description instance to which the similar overall description feature belongs, the set of undetermined target values ​​for the chronic disease description instance is determined. For each symptom description text of the chronic disease description instance, at least one similar partial description feature corresponding to the partial description feature of the symptom description text is extracted from the partial description feature set. Based on the chronic disease type target value of the chronic disease description instance to which the similar partial description feature belongs, the set of undetermined target values ​​of the symptom description text is determined. Statistical analysis is performed on the target values ​​of chronic disease types in the set of undetermined target values ​​of the symptom description text to determine the distribution characteristics of the target values ​​of the symptom description text. Statistical analysis is performed on the target values ​​of chronic disease types in the set of undetermined target values ​​of the chronic disease description instances to which the symptom description text belongs, to determine the distribution characteristics of the target values ​​of the chronic disease description instances to which the symptom description text belongs; Based on the target value distribution characteristics of the symptom description text and the target value distribution characteristics of the chronic disease description instance to which the symptom description text belongs, the homogeneity information between the symptom description text and the chronic disease description instance to which it belongs is confirmed. Based on the overall descriptive features of the chronic disease description instance, chronic disease identification is performed on the chronic disease description instance to obtain the overall type distribution characteristics corresponding to the chronic disease description instance; Based on the partial descriptive features corresponding to the symptom description text, chronic disease identification is performed on the symptom description text to obtain the partial type distribution characteristics corresponding to the symptom description text. For each chronic disease description instance, based on the partial type distribution characteristics of each symptom description text in the chronic disease description instance and the homogenization information corresponding to each symptom description text, the target value of the chronic disease type of the chronic disease description instance is optimized to obtain the optimized target value of the chronic disease type of the chronic disease description instance. Based on the overall type distribution characteristics of each chronic disease description instance, the optimized chronic disease type target values ​​of each chronic disease description instance are integrated to obtain the overall cost parameter. Based on the overall cost parameter, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model. The target chronic disease analysis model is used to identify chronic diseases in the target chronic disease description to obtain the target value of the chronic disease type in the target chronic disease description. The target chronic disease type corresponding to the target chronic disease description is determined based on the target value of the chronic disease type.

3. The method according to claim 2, characterized in that, The target value of the chronic disease type based on the target chronic disease description instance involves performing an aggregation operation on the overall descriptive features of each chronic disease description instance in the sample instance dataset to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance, including: Obtain multiple preset cluster representatives and confirm the matching degree between the overall descriptive features of each chronic disease description instance in the sample instance dataset and each cluster representative; For each chronic disease description instance, based on the matching degree and the target value of the chronic disease type of the target chronic disease description instance, a target cluster representative corresponding to the chronic disease description instance is determined from each cluster representative, and the chronic disease description instance is added to the chronic disease description set corresponding to the target cluster representative; For each cluster representative corresponding to a set of chronic disease descriptions, a chronic disease description instance that meets the preset cluster representative conditions is selected from the set of chronic disease descriptions as a new cluster representative; Repeat the steps of confirming the matching degree between the overall descriptive features of each chronic disease description instance in the sample instance dataset and each cluster representative until the determined cluster representative meets the aggregation completion state; based on the chronic disease type target value of the target chronic disease description instance in the target chronic disease description set corresponding to the cluster representative that meets the aggregation completion state, determine the chronic disease type target value of the potential chronic disease description instance in the target chronic disease description set.

4. The method according to claim 2, characterized in that, The target value of the chronic disease type based on the target chronic disease description instance involves performing an aggregation operation on the overall descriptive features of each chronic disease description instance in the sample instance dataset to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance, including: Based on the overall descriptive features of each pair of chronic disease description instances, determine the sample matching degree between each pair of chronic disease description instances; Based on the sample matching degree, a chronic disease association map is constructed, which indicates the mutual influence relationship between each chronic disease description instance; Based on the chronic disease association map, target value knowledge reasoning is performed on the target value of the chronic disease type of the target chronic disease description instance to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance.

5. The method according to claim 4, characterized in that, The chronic disease association map includes entities corresponding to each chronic disease description instance, as well as connections between entities, with the connections indicating the mutual influence relationship between two connected entities; The step of performing target value knowledge reasoning on the target value of the chronic disease type of the target chronic disease description instance based on the chronic disease association map to obtain the target value of the chronic disease type corresponding to the potential chronic disease description instance includes: In the chronic disease association graph, target value knowledge reasoning is performed on the target value of chronic disease type of the target chronic disease description instance by connecting the entities, so as to determine the baseline target value of chronic disease type of potential chronic disease description instances; For each potential chronic disease description instance, based on the entities directly connected to the potential chronic disease description instance in the chronic disease association graph, the target value of the baseline chronic disease type of the potential chronic disease description instance is recursively optimized until the target value of the entities in the chronic disease association graph is optimized, thus obtaining the target value of the chronic disease type corresponding to the potential chronic disease description instance.

6. The method according to claim 2, characterized in that, The optimization of the model parameters of the chronic disease analysis model based on the overall cost parameter to obtain the target chronic disease analysis model includes: Based on the partial type distribution characteristics of the symptom description text and the target value of the chronic disease type of the chronic disease description instance to which the symptom description text belongs, the partial category cost parameters of the symptom description text are identified. Based on the overall cost parameter and the partial category cost parameter, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model.

7. The method according to claim 6, characterized in that, The optimization of the model parameters of the chronic disease analysis model based on the overall cost parameter and the partial category cost parameters to obtain the target chronic disease analysis model includes: Error calculation is performed between the partial type distribution characteristics of the symptom description text and the preset standard distribution to obtain the discreteness parameter corresponding to the symptom description text; Based on the homogenization information corresponding to the symptom description text, the partial category cost parameters and the discreteness parameters are integrated to obtain the partial cost parameters of the symptom description text; Based on the overall cost parameter and the partial cost parameter corresponding to each symptom description text of the chronic disease description instance, the model parameters of the chronic disease analysis model are optimized to obtain the target chronic disease analysis model.

8. The method according to claim 1, characterized in that, The process of obtaining the patient's first chronic disease symptom description via an interactive terminal includes: The interactive terminal provides a visual symptom input interface, which includes multiple options for common chronic disease symptoms and a custom symptom input box. The patient is guided to select symptoms by the common chronic disease symptom options, and the selected symptoms are obtained as part of the description of the primary chronic disease symptoms. Obtain the text content entered by the patient in the custom symptom input box as a supplement to the description of the first chronic disease symptom; By integrating the partial description of the first chronic disease symptoms and the supplementary description of the first chronic disease symptoms, a complete description of the first chronic disease symptoms is obtained.

9. The method according to claim 1, characterized in that, Before integrating the first chronic disease symptom description and the second chronic disease symptom description, the preset large language model also includes: Semantic cleaning operations are performed on the first chronic disease symptom description and the second chronic disease symptom description, respectively. The semantic cleaning operations include removing duplicate expressions, correcting typos, and standardizing the expression of medical terminology. The semantically cleaned descriptions of the first and second chronic disease symptoms are input into a preset format conversion module to convert them into a format that meets the input requirements of the large language model.

10. A cloud platform, characterized in that, The cloud platform is used to perform the method described in any one of claims 1-9.

Images