A large model-based respiratory chronic disease intelligent diagnosis and treatment management method and system

Abstract

The application relates to the technical field of data processing, and discloses a respiratory chronic disease intelligent diagnosis and treatment management method and system based on a large model. The method comprises the following steps: collecting and processing multi-dimensional data of respiratory chronic disease patients to form a structured data set; constructing a large model and a knowledge graph by using a double coding framework; performing risk stratification evaluation based on the large model; generating a personalized diagnosis and treatment scheme according to the evaluation result; constructing a medical collaboration framework to perform efficacy monitoring; customizing health education content and continuously optimizing the system. The application realizes accurate stratification of respiratory chronic disease patients, personalized treatment scheme formulation, multi-level medical collaboration management and dynamic efficacy monitoring, thereby overcoming the defects of insufficient diagnosis and treatment accuracy, low resource allocation efficiency and discontinuous patient management in the prior art, and improving the overall efficiency of respiratory chronic disease diagnosis and treatment management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a respiratory chronic disease intelligent diagnosis and treatment management method and system based on a large model. BACKGROUND

[0002] Respiratory chronic diseases include chronic obstructive pulmonary disease (COPD), asthma, pulmonary fibrosis, etc., which have the characteristics of long course, repeated attacks, and long-term management, and have become a major global public health problem. Traditional respiratory chronic disease management mainly relies on regular outpatient follow-up and patient self-reporting, and uneven distribution of medical resources leads to a significant gap in diagnosis and treatment capabilities between primary medical institutions and specialized hospitals. Existing technologies attempt to improve respiratory chronic disease monitoring and management through telemedicine and wearable devices, such as physiological parameter remote monitoring systems based on the Internet of Things, medication reminder systems with simple rule engines, and disease risk prediction tools based on statistical models. Some research also explores patient education platforms based on mobile Internet and primary artificial intelligence assisted diagnosis systems, trying to improve disease management efficiency and patient compliance.

[0003] However, the existing technology has obvious deficiencies in practical application. First, the traditional respiratory chronic disease management mode lacks precision and individualization, and cannot dynamically adjust the treatment plan according to the individual situation of the patient. Second, although existing remote monitoring systems can collect data, they lack effective data analysis and knowledge mining capabilities, making it difficult to extract valuable clinical decision-making information from massive multi-modal medical data. Third, the information silo problem between medical institutions is serious, and the hierarchical diagnosis and treatment is difficult to be substantially implemented, resulting in inefficient allocation of high-quality medical resources. In addition, existing patient education tools generally adopt a "one-size-fits-all" approach, ignoring the differences in patients' cognitive levels and acceptance abilities, and the education effect is limited. Most importantly, the existing technology lacks an effective way to integrate medical professional knowledge and advanced artificial intelligence technology, and cannot realize medical knowledge-driven intelligent decision support. SUMMARY

[0004] The present application provides a respiratory chronic disease intelligent diagnosis and treatment management method and system based on a large model, which is used to realize precise stratification of respiratory chronic disease patients, individualized treatment plan formulation, multi-level medical collaborative management, and dynamic efficacy monitoring, thereby overcoming the deficiencies of insufficient diagnosis and treatment precision, low resource allocation efficiency, and discontinuous patient management in the prior art, and improving the overall efficiency of respiratory chronic disease diagnosis and treatment management.

[0005] In a first aspect, the application provides a large model-based intelligent diagnosis and treatment management method for respiratory chronic diseases, which comprises: collecting and processing multi-dimensional data of patients with respiratory chronic diseases to obtain a structured patient data set; inputting the structured patient data set and a respiratory system disease professional knowledge base into a double coding pre-training and contrast learning fine-tuning framework to obtain a respiratory chronic disease large model and a respiratory chronic disease knowledge graph; performing semantic analysis and multi-dimensional calculation on patient symptom descriptions and evaluation scale answers based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph to generate a patient risk stratification evaluation result; retrieving clinical guideline rules using the respiratory chronic disease large model and combining patient individual characteristics to generate an individualized diagnosis and treatment plan and a matching score according to the patient risk stratification evaluation result; constructing a multi-level medical institution collaboration framework and performing dynamic efficacy monitoring according to the individualized diagnosis and treatment plan and the matching score to generate efficacy indicators and treatment response indices; and customizing patient health education content and self-management tools according to the efficacy indicators and the treatment response indices, while collecting system operation data for model updating and process optimization.

[0006] In a second aspect, the application provides a large model-based intelligent diagnosis and treatment management system for respiratory chronic diseases, which comprises:

[0007] A collection module for collecting and processing multi-dimensional data of patients with respiratory chronic diseases to obtain a structured patient data set;

[0008] An input module for inputting the structured patient data set and a respiratory system disease professional knowledge base into a double coding pre-training and contrast learning fine-tuning framework to obtain a respiratory chronic disease large model and a respiratory chronic disease knowledge graph;

[0009] A calculation module for performing semantic analysis and multi-dimensional calculation on patient symptom descriptions and evaluation scale answers based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph to generate a patient risk stratification evaluation result;

[0010] A generation module for retrieving clinical guideline rules using the respiratory chronic disease large model and combining patient individual characteristics to generate an individualized diagnosis and treatment plan and a matching score according to the patient risk stratification evaluation result;

[0011] A construction module for constructing a multi-level medical institution collaboration framework and performing dynamic efficacy monitoring according to the individualized diagnosis and treatment plan and the matching score to generate efficacy indicators and treatment response indices;

[0012] A customization module for customizing patient health education content and self-management tools according to the efficacy indicators and the treatment response indices, while collecting system operation data for model updating and process optimization.

[0013] In a third aspect, a large model-based respiratory chronic disease intelligent diagnosis and treatment management device is provided, comprising a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to enable the large model-based respiratory chronic disease intelligent diagnosis and treatment management device to perform the above-mentioned large model-based respiratory chronic disease intelligent diagnosis and treatment management method.

[0014] In a fourth aspect, a computer-readable storage medium is provided, the computer-readable storage medium storing instructions, when running on a computer, enabling the computer to perform the above-mentioned large model-based respiratory chronic disease intelligent diagnosis and treatment management method.

[0015] In the technical scheme provided in the present application, structured patient data sets are formed through multi-dimensional data collection and processing, realizing comprehensive acquisition and standardized processing of respiratory chronic disease related information. Secondly, the structured patient data sets and respiratory system disease professional knowledge base are processed by using a dual coding pre-training and contrast learning fine-tuning framework, and the generated respiratory chronic disease large model and respiratory chronic disease knowledge graph have deep medical understanding ability and professional reasoning ability, which are significantly superior to traditional machine learning models, can understand complex medical context and extract key clinical features therefrom. Thirdly, based on the respiratory chronic disease large model and the knowledge graph, semantic analysis and multi-dimensional calculation are performed on patient symptom descriptions and evaluation scale answers, and the generated patient risk stratification evaluation results have high accuracy, can capture subtle disease feature differences, and realize precise stratification, thereby optimizing medical resource allocation. Fourthly, according to the patient risk stratification evaluation results, the respiratory chronic disease large model is used to retrieve clinical guideline rules and generate personalized diagnosis and treatment schemes and matching scores combined with patient individual characteristics, which fully considers patient individual differences, makes the treatment scheme comply with the principle of evidence-based medicine and is highly personalized, and improves treatment accuracy and compliance. Fifthly, the multi-level medical institution collaboration framework and dynamic efficacy monitoring system constructed according to the personalized diagnosis and treatment scheme and the matching score break the medical information island, realize reasonable allocation of high-quality medical resources and continuous medical services, and reduce repeated examinations and treatment errors. The patient health education content and self-management tools customized according to the efficacy indicators and treatment response index, combined with continuous collection of system operation data and model updating and process optimization, form a closed-loop quality improvement system, so that the system continuously evolves and improves. It is particularly emphasized that the large model algorithm applied in the present scheme makes great contribution to the field of respiratory chronic disease diagnosis and treatment management, its context understanding ability enables the model to extract key clinical information from unstructured medical text, the knowledge enhanced reasoning mechanism realizes automatic support for complex medical decisions, the multi-modal data fusion algorithm can integrate patient information from different sources to form a comprehensive view, and the adaptive learning characteristic enables the system to continuously optimize diagnosis and treatment strategies according to new data. These specific algorithm characteristics jointly construct a highly specialized intelligent auxiliary decision system for respiratory system diseases, greatly improving the efficiency and accuracy of respiratory chronic disease management. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings based on these drawings without creative labor.

[0017] Figure 1 An embodiment schematic diagram of the large model based intelligent diagnosis and treatment management method for respiratory chronic diseases in the embodiments of the present application;

[0018] Figure 2 FIG. 1 is a schematic diagram of an embodiment of a respiratory chronic disease intelligent diagnosis and treatment management system based on a large model in the present application;

[0019] Figure 3 FIG. 1 is a schematic diagram of an embodiment of a respiratory chronic disease intelligent diagnosis and treatment management system based on a large model in the present application; DETAILED DESCRIPTION

[0020] The present application provides a respiratory chronic disease intelligent diagnosis and treatment management method and system based on a large model. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the specific process of the embodiments of the present application is described below. Please refer to Figure 1 An embodiment of a respiratory chronic disease intelligent diagnosis and treatment management method based on a large model in the present application includes:

[0022] Step S101, multi-dimensional data acquisition and processing are performed on respiratory chronic disease patients to obtain a structured patient data set;

[0023] Step S102, the structured patient data set and the respiratory system disease professional knowledge base are input into a double coding pre-training and contrast learning fine-tuning framework to obtain a respiratory chronic disease large model and a respiratory chronic disease knowledge graph;

[0024] Step S103, based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph, semantic analysis and multi-dimensional calculation are performed on patient symptom descriptions and evaluation scale answers to generate patient risk stratification evaluation results;

[0025] Step S104, according to the patient risk stratification evaluation results, the respiratory chronic disease large model is used to retrieve clinical guideline rules and combine patient individual characteristics to generate individualized diagnosis and treatment plans and matching score;

[0026] Step S105, according to the individualized diagnosis and treatment plans and the matching score, a multi-level medical institution collaboration framework is constructed and dynamic efficacy monitoring is performed to generate efficacy indicators and treatment response indexes;

[0027] Step S106, according to the therapeutic index and treatment response index, customizing patient health education content and self-management tools, while collecting system operation data for model updating and process optimization.

[0028] It can be understood that the execution subject of the present application can be a respiratory chronic disease intelligent diagnosis and treatment management system based on a large model, and can also be a terminal or a server, which is not limited here. The server is taken as an example for illustration in the embodiments of the present application.

[0029] Specifically, starting from the multi-dimensional data collection of respiratory chronic disease patients, data is collected through three main channels: clinical input of medical institutions, self-reporting of patient mobile terminals, and automatic collection of intelligent wearable devices. These data include the patient's basic information, disease history, medication, symptom assessment, and physiological monitoring data. The collected raw data are heterogeneous and need to be preprocessed, including standardization and normalization of structured data, word segmentation and entity standardization of text data, and noise reduction and outlier processing of time series data. Subsequently, a unique patient identifier is established, and all data are organized into a tree-like hierarchical structure, which, after integrity check and consistency verification, forms a structured patient data set.

[0030] The structured patient data set and the respiratory system disease knowledge base (including authoritative clinical guidelines and evidence-based medical literature) are input into a dual-encoding pre-training and contrastive learning fine-tuning framework. The framework uses a Transformer encoder structure, pre-trains a large-scale medical corpus through a masked language model and next sentence prediction task, and then fine-tunes it using a contrastive learning method to bring similar case representation vectors closer together and push different case representation vectors further apart, thereby obtaining a respiratory chronic disease large model. At the same time, through entity extraction and relationship extraction, medical concepts such as diseases, symptoms, and drugs and their associations are identified to form knowledge triples, and a respiratory chronic disease knowledge graph is constructed through graph embedding algorithms. Based on these two core components, the patient's symptom description and assessment scale answers are intelligently analyzed. The system obtains the patient's symptom description through a natural language interaction interface, extracts key symptom attributes and forms a symptom entity matrix; at the same time, it guides the patient to complete a structured symptom assessment questionnaire to obtain a questionnaire numerical vector. Combine these two parts of data with basic physiological monitoring data to construct a multi-dimensional feature space. Assign weight coefficients to the core symptoms, and obtain the symptom burden index through weighted calculation method; perform time series analysis on the symptom fluctuation trend, acute attack risk, etc. to generate the disease stability index; combine the basic disease severity classification and comorbidity situation to calculate the comprehensive risk score, and divide the patients into low, medium and high risk levels accordingly.

[0031] According to the risk stratification evaluation results, the respiratory chronic disease large model is used to retrieve the corresponding clinical guideline rules. For patients with chronic obstructive pulmonary disease, they are divided into A / B / C / D four groups according to the degree of airflow limitation and symptom score; for asthma patients, they are labeled based on control level and severity. Then, different drug regimens are matched according to the disease classification results to form a basic treatment framework. This framework is compared and analyzed with the patient's past treatment records, drug adverse reaction history, and the priority of drug options is adjusted to form a personalized drug regimen, supplemented with detailed drug administration information and non-drug treatment suggestions. Finally, through the calculation of the patient's past medication response, lifestyle, compliance factors, economic affordability and other matching scores, a percentage matching score is obtained.

[0032] According to the patient risk level and matching score, a multi-level medical institution collaboration framework is constructed, and low-risk patients are assigned to primary medical institutions, medium-risk patients are arranged for primary management and specialist follow-up, and high-risk patients are managed by specialist hospitals. Different levels of medical institutions are connected through an intelligent collaboration platform to realize data sharing. For patients treated in primary care, the system provides large model assisted decision support. For patients starting treatment, multi-dimensional monitoring is carried out to record symptom score changes, physiological index changes, medication compliance data, and quality of life assessment data, and through time series analysis to calculate the improvement of each dimension, generate multi-dimensional efficacy indicators, and calculate the comprehensive score of treatment response through weighted aggregation algorithm to form the time series treatment response index. According to the efficacy indicators and treatment response index, personalized health education content and self-management tools are customized. According to the patient's characteristics and efficacy, disease knowledge, symptom recognition, medication guidance and other content are screened and converted into multimedia resources, and the presentation depth is dynamically adjusted according to the patient's understanding level. Electronic medicine box reminder system, symptom monitoring diary and other self-management tools are developed and integrated into a multi-channel access platform. At the same time, the data of user behavior, feedback opinions, diagnosis and treatment effects during the use of the system are recorded and analyzed, which are used for large model knowledge base update, algorithm optimization and system process optimization.

[0033] In the embodiments of the present application, structured patient data sets are formed by multi-dimensional data collection and processing, comprehensive acquisition and standardized processing of respiratory chronic disease related information are achieved; secondly, the structured patient data sets and the respiratory system disease professional knowledge base are processed by using the dual coding pre-training and contrast learning fine-tuning framework, the generated respiratory chronic disease large model and respiratory chronic disease knowledge graph have deep medical understanding ability and professional reasoning ability, which are significantly better than traditional machine learning models, and can understand complex medical context and extract key clinical features from it; thirdly, based on the respiratory chronic disease large model and the knowledge graph, semantic analysis and multi-dimensional calculation are performed on the patient symptom description and the answer of the evaluation scale, the generated patient risk stratification evaluation result has high accuracy and can capture subtle disease feature differences, so as to realize accurate stratification and optimize medical resource allocation; fourthly, according to the patient risk stratification evaluation result, the respiratory chronic disease large model is used to retrieve clinical guideline rules and generate personalized diagnosis and treatment scheme and matching score combined with patient individual characteristics, which fully considers patient individual differences, so that the treatment scheme conforms to the principle of evidence-based medicine and is highly personalized, improving treatment accuracy and compliance; fifthly, the multi-level medical institution collaboration framework and dynamic efficacy monitoring system constructed according to the personalized diagnosis and treatment scheme and the matching score break the medical information island, realize the reasonable allocation of high-quality medical resources and continuous medical services, and reduce repeated examination and treatment errors; the patient health education content and self-management tool customized according to the efficacy index and treatment response index, combined with the continuous collection of system operation data and model updating and process optimization, form a closed-loop quality improvement system, so that the system continuously evolves and improves. It is especially worth emphasizing that the large model algorithm applied in the present scheme makes great contribution to the field of respiratory chronic disease diagnosis and treatment management, its context understanding ability enables the model to extract key clinical information from unstructured medical text, the knowledge enhanced reasoning mechanism realizes the automation support of complex medical decision, the multi-modal data fusion algorithm can integrate patient information from different sources to form a comprehensive view, and the adaptive learning characteristics enable the system to continuously optimize the diagnosis and treatment strategy according to new data, greatly improving the efficiency and accuracy of respiratory chronic disease management.

[0034] In a specific embodiment, the process of performing step S101 can specifically include the following steps:

[0035] The basic information, disease history, medication, symptom evaluation and physiological monitoring data of the patient are collected by the medical institution clinical input, the patient mobile terminal self-reporting and the intelligent wearable device, to form a multi-source heterogeneous data set;

[0036] The structured data in the multi-source heterogeneous data set is standardized and normalized, the text data is subjected to word segmentation, stop word removal and entity standardization operations, the time series data is subjected to noise reduction, missing value interpolation and outlier processing, to generate a preprocessed data set;

[0037] The pre-processed data set is labeled according to the respiratory drug classification method, a unique patient identifier is established, and an initial labeled data set is generated;

[0038] The initial labeled data set is organized in a tree hierarchy, and the top-level patient basic information, the secondary-level disease classification, and the bottom-level symptom monitoring indicators are associated with the corresponding patient files to obtain a hierarchical associated data set;

[0039] The hierarchical associated data set is subjected to integrity check, consistency verification and outlier marking to generate a quality-controlled data set;

[0040] The quality-controlled data set is stored in a secure database to form a structured patient data set.

[0041] Specifically, patient basic data is collected through medical institution clinical entry. Medical staff enters the patient's gender, age, occupation, living environment, and other demographic information into the system during diagnosis and treatment, as well as disease history information such as disease diagnosis date, number of previous acute attacks, hospitalization history, and current medication information such as drug type, dosage, and compliance record. The patient's mobile terminal self-reporting collects the patient's daily symptom changes and medication information through a specially designed mobile application. The patient fills out the assessment form regularly or when the symptoms change, including the degree of dyspnea, cough frequency, sputum character, and activity limitation. The smart wearable device automatically collects the patient's lung function test values, blood oxygen saturation, and respiratory rate, and other physiological monitoring data. These devices synchronize with the system in real time or periodically, forming continuous physiological parameter monitoring records. These data from different channels form a multi-source heterogeneous data set. Due to the diversity of data sources, formats and contents differ, and data preprocessing is required. Standardization processing is performed on structured data to convert numerical data of different dimensions to a unified standard, such as converting lung function test values from different laboratories to predicted value percentages. Normalization processing adjusts the data range to 0-1, making each indicator comparable. Text data is processed by performing word segmentation, breaking down the patient's symptom experience into basic semantic units; removing stop words, deleting virtual words and conjunctions that have no substantial meaning for medical judgment; and entity standardization, which maps synonymous expressions to medical terminology. The processing of time series data includes noise reduction, eliminating short-term fluctuations by methods such as sliding window averaging; missing value interpolation, making reasonable inferences based on the previous and subsequent values of the time series or similar patterns of similar patients; and outlier processing, identifying and correcting data points that deviate significantly from the normal range.

[0042] The preprocessed dataset needs to be labeled according to the respiratory medication classification. Respiratory medication classification is a classification system specifically for respiratory disease drugs, including bronchodilators, anti-inflammatory drugs, and mucolytic agents. According to the patient's medication history and current treatment plan, the data is labeled accordingly. At the same time, a unique patient identifier is established by synthesizing multiple personal information and encrypting it to ensure the uniqueness of each patient in the system, avoiding data confusion, forming an initial labeled dataset. The initial labeled dataset is organized in a tree-like hierarchical structure, with the top layer storing patient basic information, including demographic characteristics and unique identifiers; the next layer is disease classification, which classifies patients into specific respiratory disease types such as chronic obstructive pulmonary disease, asthma, and interstitial lung disease; the bottom layer is symptom monitoring indicators, including all disease-related symptom scores, physiological indicators, and medication records. Through the unique identifier, the three layers of data are associated with the corresponding patient file to form a patient view, and a clear reference relationship is established between the hierarchical data.

[0043] Quality control is performed on the hierarchical associated dataset to check for completeness, confirm that each patient's necessary information is complete, including basic demographic information, disease diagnosis, core symptom evaluation, and key physiological indicators; consistency verification checks for internal data conflicts, such as whether different records at the same time point conflict, and whether symptom descriptions match objective indicators; outlier labeling sets a reasonable range for each indicator and labels data outside the range, such as extreme lung function values or unreasonable symptom scores. The quality-controlled dataset is stored in a secure database using encryption storage and access control mechanisms to ensure patient data security, forming a structured patient dataset.

[0044] Take a chronic obstructive pulmonary disease patient data processing as an example: the original data is collected from three channels, the basic information entered by the medical institution shows that the patient is a 65-year-old male, and his occupation is a retired teacher; the disease history record shows that the COPD diagnosis time is 5 years ago, and there have been 2 acute attacks in the past year; the current medication includes salmeterol / fluticasone inhaler, twice a day. The symptom data reported by the mobile terminal includes the daily dyspnea score (mMRC scale) fluctuation in the past week, which is between 2-3 points, the cough frequency is 3-5 times a day, and the sputum is white and sticky. The average value of the night blood oxygen saturation recorded by the smart wearable device is 93%, with a temporary fluctuation to 88%. After preprocessing these heterogeneous data, the dyspnea score is averaged to 2.5 points, standardized as moderate; the blood oxygen data is denoised to eliminate one obvious device shedding false reading; the patient's report of "chest tightness and shortness of breath" is standardized to "dyspnea" medical terminology through text processing. The data is labeled according to the combined use of LABA+ICS, and after establishing a unique identifier, it is organized into a tree structure, with the top layer being personal basic information, the next layer being COPD diagnosis (GOLD classification B level), and the bottom layer being symptom and monitoring index details. Quality control found that there was a 30-minute gap in the night blood oxygen data, which was interpolated by the average value of the previous and subsequent periods; at the same time, one record of sudden exacerbation of self-reported dyspnea but no corresponding physiological indicator change was marked, which needed to be further verified. The structured data set contains patient respiratory chronic disease portraits.

[0045] In a specific embodiment, the process of performing step S102 can specifically include the following steps:

[0046] Integrating respiratory disease clinical guidelines, evidence-based medical literature library, expert consensus documents, typical case set and drug interaction database to build respiratory system disease professional knowledge base;

[0047] Inputting the structured patient data set and the medical text corpus in the respiratory system disease professional knowledge base into the Transformer encoder structure at the same time, using the mask language model and the next sentence prediction task for multi-round iterative training to obtain the initial large model parameters;

[0048] Based on the initial large model parameters, a contrast learning framework is constructed to pull similar case representation vectors and push different case representation vectors, and the model parameters are optimized by minimizing the contrast loss function to generate a respiratory chronic disease large model;

[0049] Performing entity extraction on the structured patient data set and the respiratory system disease professional knowledge base to identify diseases, symptoms, drugs, examinations, risk factors as medical concepts to obtain a medical entity set;

[0050] Relations are extracted from the set of medical entities to identify disease-symptom, disease-drug, drug-drug, symptom-severity, and examination-reference value as relationships between entities, forming a set of relation triples.

[0051] By using graph embedding algorithms to learn knowledge representations of medical entity sets and relation triple sets, a knowledge graph of chronic respiratory diseases is constructed. This graph is then linked to a large-scale model of chronic respiratory diseases through entity linking technology, supporting knowledge-enhanced reasoning.

[0052] Specifically, this involves integrating authoritative clinical guidelines for respiratory diseases, including the COPD guidelines published by the Global Initiative for Chronic Obstructive Lung Disease (GOLD), the asthma guidelines published by the Global Initiative for Asthma (GINA), and the guidelines for the diagnosis and treatment of interstitial lung disease. These clinical guidelines are organized in structured XML files, extracting core content such as disease definitions, diagnostic criteria, classification methods, treatment recommendations, and management strategies. The integration of evidence-based medicine literature covers high-quality research literature on respiratory diseases, retrieving relevant randomized controlled trials, systematic reviews, and meta-analyses published within the past five years from professional databases, extracting research findings, evidence-based conclusions, and levels of evidence. The integration of expert consensus documents includes disease management consensus published by respiratory professional societies from various countries, extracting expert experience and clinical recommendations. Typical case collections gather representative cases of various chronic respiratory diseases, including disease presentation, diagnostic process, treatment plans, and prognostic outcomes. The integration of drug interaction databases summarizes interaction information between commonly used respiratory drugs and between drugs and other systems, forming a drug interaction network. These multi-source data, after text extraction, topic modeling, and content indexing, are used to construct a structured professional knowledge base for respiratory diseases. The model was trained using a Transformer encoder structure, which incorporates a structured patient dataset and medical text corpora from a respiratory disease knowledge base. The Transformer encoder is a neural network architecture based on a self-attention mechanism, capable of capturing long-distance dependencies in text. The training process employs two key tasks: masked language modeling and next-sentence prediction. In the masked language modeling task, 15% of the words in the input text are randomly replaced with special markers, and the model is then trained to predict these masked words based on context. For example, masking the word "dyspnea" in "patients present with severe markings and excessive phlegm" allows the model to learn to infer appropriate medical terminology from the context. The next-sentence prediction task provides the model with two sentences, training it to determine whether the second sentence is a natural continuation of the first, thus learning the coherence and logical relationships of the text. Through multiple rounds of iterative training on these two tasks, the model gradually learns the representation of medical language and respiratory disease knowledge, forming the initial large model parameters.

[0053] Based on the initial large model parameters, a contrastive learning framework is constructed to further optimize the model. Contrastive learning is a method of learning representation by comparing the similarities and differences between samples. In this scheme, clinically similar respiratory chronic disease cases are used as positive sample pairs, and cases of different types or severity are used as negative sample pairs. Through data enhancement techniques, different expression variants of the case are generated, keeping the core medical content unchanged but diversifying the expression form. The large model is used to extract features from each case, resulting in a high-dimensional representation vector. Then the cosine similarity between the vectors is calculated, and a contrastive loss function is constructed to reduce the distance between the representation vectors of similar cases and increase the distance between the representation vectors of different cases. Through gradient descent algorithm, the contrastive loss function is minimized, and the model parameters are iteratively optimized to generate a large model sensitive to respiratory chronic diseases. Structured patient data sets and professional knowledge bases are used for entity extraction to identify key medical concepts. Using named entity recognition technology, biomedical dictionaries and contextual semantic analysis are used to identify medical entities such as diseases, symptoms, drugs, tests, and risk factors in the text. Each identified entity is assigned a unique identifier and category label to form a medical entity set.

[0054] The medical entity set is then subjected to relationship extraction to establish semantic connections between entities. Through dependency syntax analysis and semantic role labeling techniques, the relationships between entities are extracted from the text. For example, from the sentence "Asthma often presents as wheezing and shortness of breath," two disease-symptom relationships are extracted: "Asthma - presents as - wheezing" and "Asthma - presents as - shortness of breath." From the sentence "Combination of salbutamol and ipratropium bromide can enhance bronchodilation effect," the drug-drug relationship "Salbutamol - synergistic effect - ipratropium bromide" is extracted. In addition, the association between symptoms and severity and the association between tests and reference values are also extracted. Each extracted relationship forms a subject-relation-object triple structure, which is aggregated to form a relationship triple set. The medical entity set and the relationship triple set are converted into a knowledge graph through a graph embedding algorithm. The graph embedding algorithm maps each medical entity and relationship to a low-dimensional vector space, so that semantically similar concepts are closer in the vector space. Commonly used graph embedding algorithms include TransE, ComplEx, and RotatE, among others. In this scheme, the improved TransE algorithm is used, which learns the vector representation of entities and relationships by optimizing the objective function that the head entity vector plus the relationship vector is close to the tail entity vector. The generated vector representation preserves the semantic relationships between entities, forming a respiratory chronic disease knowledge graph. Through entity linking technology, the medical concepts mentioned in the text are mapped to the entity nodes in the knowledge graph, thereby realizing the fusion of the knowledge graph and the large model, supporting knowledge-enhanced medical reasoning.

[0055] Take the processing of chronic obstructive pulmonary disease (COPD) related knowledge as an example: extract the definition "COPD is a common, preventable and treatable disease characterized by persistent respiratory symptoms and airflow limitation" from the GOLD guideline, identify "COPD" as a disease entity, "persistent respiratory symptoms" and "airflow limitation" as symptom entities through entity extraction. The relationship extraction stage identifies two triples "COPD-features are-persistent respiratory symptoms" and "COPD-features are-airflow limitation". At the same time, extract "smoking is the main risk factor for COPD" from the evidence-based literature, identify "smoking" as a risk factor entity, and form the "smoking-is-a-risk factor for COPD" triple. Extract the COPD classification criteria from the clinical guidelines, such as "FEV1 / FVC <0.7 and FEV1 <30% predicted value for severe COPD", form "FEV1 / FVC <0.7-diagnostic criteria-COPD" and "FEV1 <30% predicted value-severity assessment-severe COPD" triples. The drug treatment part extracts "long-acting beta2 receptor agonists (LABA) and long-acting anticholinergic drugs (LAMA) are used for symptom severe COPD patients", forms "LABA- combination drug-LAMA" and "LABA+LAMA-treatment for-symptom severe COPD" relationships. These entities and relationships are mapped to vector space through graph embedding algorithm, for example, "COPD" is mapped to 128-dimensional vector, and related entities such as symptoms and drugs form clusters in vector space, constructing a local graph structure of COPD related knowledge. When new patient data is input into the system, the patient's symptoms "severe dyspnea" are associated with the corresponding nodes in the knowledge graph through entity linking, and the large model infers the possible diagnosis and appropriate treatment plan accordingly, realizing knowledge-enhanced intelligent diagnosis and treatment decision.

[0056] In a specific embodiment, the process of performing step S103 can specifically include the following steps:

[0057] Guide the patient to describe the current symptom experience through the natural language interaction interface, input the patient's natural language into the respiratory chronic disease large model, extract the symptom duration, severity, trigger, and relieving factors as key symptom attributes, and generate a symptom entity matrix;

[0058] Guide the patient to complete the respiratory distress score, cough frequency score, sputum character score, activity limitation degree score, sleep impact score, and medication compliance score as a structured symptom assessment questionnaire, and obtain a questionnaire numerical vector;

[0059] Fuse the symptom entity matrix and the questionnaire numerical vector, combine the lung function indicators and blood oxygen saturation values in the basic physiological monitoring data to construct a multi-dimensional feature space, and form a comprehensive symptom representation of the patient;

[0060] A weight coefficient is assigned to dyspnea, cough, and sputum, which are core symptoms in the comprehensive symptom representation of the patient. The symptom burden index is obtained by multiplying the severity values of each symptom by the corresponding weight coefficient and summing them up through a weighted calculation method.

[0061] The recent symptom fluctuation trend, acute attack risk, medication compliance value, and environmental factor influence are analyzed in time sequence. The disease stability index is generated through a stability calculation model. Meanwhile, the comprehensive risk score is calculated by combining the basic disease severity classification, comorbidity situation, and social support score.

[0062] According to the threshold interval of the comprehensive risk score, patients with a risk score below the first threshold are classified as low-risk level, patients with a risk score between the first threshold and the second threshold are classified as medium-risk level, and patients with a risk score higher than the second threshold are classified as high-risk level. The patient risk stratification assessment result is generated.

[0063] Specifically, the patient is guided to describe the current symptom experience through a natural language interaction interface. The interaction interface adopts a semi-structured inquiry approach, posing open-ended questions such as "Please describe your recent breathing difficulty situation" and supplemented with guiding questions such as "How long does this situation last?" "What factors can aggravate or alleviate symptoms?" to guide the patient to make a comprehensive statement. The natural language description input by the patient is transmitted to the respiratory chronic disease large model, and key symptom attributes are identified through named entity recognition and relation extraction techniques. The symptom duration attribute is extracted from the time expression in the text, such as "three days ago" to a specific time length value; the severity attribute identifies words that describe intensity, such as "mild", "obvious", "severe" to a three-level scale of light, medium, and heavy; the trigger attribute extracts symptom triggers, such as "after exercise", "exposure to pollen", "inhaling cold air", etc.; the relieving factor attribute captures conditions that improve symptoms, such as "after resting", "after using an inhaler", etc. These extracted attribute values are organized into a structured symptom entity matrix, with each row representing a symptom and each column representing an attribute dimension, and the elements in the matrix are the corresponding attribute values. At the same time, the patient is guided to complete a structured symptom assessment questionnaire, including multiple dimension rating scales. The breathing difficulty score uses the modified Medical Research Council (mMRC) scale, which ranges from 0 to indicate breathing difficulty only during strenuous activity to severe breathing difficulty to the extent that the patient cannot leave home or dress; the cough frequency score ranges from 0 to 3, indicating from no cough to persistent cough affecting daily activities; the sputum character score ranges from 0 to 3, indicating from no sputum to a large amount of purulent sputum; the activity limitation score ranges from 0 to 5, indicating from no limitation to complete bed rest; the sleep impact score ranges from 0 to 5, indicating from normal sleep to complete inability to sleep due to respiratory symptoms; the medication adherence score ranges from 0 to 2, indicating from complete compliance to basically not following medical advice. The patient completes the score by selecting the option that best fits their own situation, and these scores form a questionnaire numerical vector, with each element of the vector corresponding to a score dimension.

[0064] The symptom entity matrix and the questionnaire numerical vector are fused by feature concatenation and cross-encoding methods. Feature concatenation directly combines the two parts of data into a larger feature vector; cross-encoding calculates the interaction between the two parts of features to generate interaction features. For example, when the breathing difficulty duration in the symptom entity matrix is "long-term" and the breathing difficulty score in the questionnaire is 4, the interaction feature will strengthen the representation of this severe condition. Combined with the basic physiological monitoring data, which includes lung function indicators and blood oxygen saturation values in the embodiments of the present application, the feature space is further expanded to obtain a multi-dimensional comprehensive symptom representation of the patient.

[0065] The core symptoms in the patient's comprehensive symptom profile are weighted to calculate the symptom burden index. Dyspnea, as the most reflective indicator of patients' quality of life and disease severity, is given the highest weight; cough, as the most common symptom, is closely related to social function and sleep quality, and is given the second highest weight; sputum changes reflect airway inflammation and infection risk, and are also given appropriate weight. In the weighting calculation process, first, the scores of each symptom are standardized to the 0-1 interval, then multiplied by the corresponding weight coefficient, and finally summed to obtain the symptom burden index. For example, the standardized dyspnea score of 0.75 multiplied by the weight of 0.5, the cough score of 0.6 multiplied by the weight of 0.3, and the sputum score of 0.4 multiplied by the weight of 0.2, the sum of the three is 0.75x0.5+0.6x0.3+0.4x0.2=0.62 as the symptom burden index.

[0066] The patient's multiple time series data are analyzed to evaluate disease stability. The recent symptom fluctuation trend is calculated by the coefficient of variation and the trend slope of continuous monitoring data, and a large coefficient of variation or a positive slope indicates unstable symptoms; the acute attack risk is predicted based on the historical attack frequency, seasonal factors and current symptom changes, and is quantified as a risk value between 0 and 1; the medication compliance value is obtained from the electronic medicine box record and the patient's self-reported data, and is converted into a compliance percentage; the environmental factor influence considers meteorological conditions, air quality and personal sensitivity factors. These time series indicators are integrated to generate a disease stability index, with a higher value indicating more stable disease. At the same time, according to the disease severity classification of medical guidelines, the number and severity of patient comorbidities, and the social support score, the overall risk score is calculated.

[0067] The threshold setting of the risk score is based on large-scale clinical data analysis and expert consensus. By retrospectively analyzing the historical data of patients with respiratory diseases, the correlation between different risk scores and adverse events is statistically analyzed to determine two key dividing points: the first threshold is set to 40 points, and the historical data of patients below this value shows that the incidence of adverse events is less than 5%; the second threshold is set to 65 points, and the incidence of adverse events in patients above this value is more than 20%. Based on these two thresholds, patients are divided into three risk levels: low risk level (score <40), suitable for primary care management, only requiring basic intervention; medium risk level (score 40-65), requiring enhanced monitoring and more aggressive treatment strategies; high risk level (score >65), requiring specialist hospital intervention and close monitoring.

[0068] In a specific embodiment, the process of performing step S104 can specifically include the following steps:

[0069] The patient risk stratification evaluation result is input into a respiratory chronic disease large model, and the disease classification of the patient is identified. When the patient is a chronic obstructive pulmonary disease, the patient is divided into four groups of A / B / C / D according to the degree of airflow limitation and symptom score. When the patient is an asthma, the patient is marked based on the control level and severity, and a disease classification result is generated.

[0070] The disease classification result is matched with a clinical guideline rule. When the patient belongs to group A, a mild drug regimen is searched. When the patient belongs to group B, a moderate drug regimen is searched. When the patient belongs to group C or group D, a severe drug regimen is searched, and a basic treatment framework is obtained.

[0071] The basic treatment framework is compared and analyzed with the patient's previous treatment record and drug adverse reaction history, and a preliminary individualized drug regimen is generated.

[0072] Based on the preliminary individualized drug regimen, drug information is developed, including preferred drug class combination, specific drug selection, administration route, dosage and administration, course design, and a drug treatment regimen is generated.

[0073] Combined with the drug treatment regimen, non-drug treatment suggestions are created, including respiratory rehabilitation training plan, lifestyle adjustment suggestions, psychological health intervention measures, and disease self-management skill training. When the patient's risk level is high, an acute exacerbation emergency plan is developed, and a personalized diagnosis and treatment plan is formed.

[0074] The personalized diagnosis and treatment plan is matched and evaluated with the patient's characteristics, and the patient's previous drug response matching score, lifestyle matching score, compliance factor matching score, economic bearing matching score, and medical insurance coverage matching score are calculated. Through calculation, a percentage matching degree score is obtained.

[0075] Specifically, the patient risk stratification assessment results are input into the respiratory chronic disease large model for disease classification. The large model analyzes the patient's basic clinical data, symptom manifestations, and physiological indicators to identify specific disease types. For chronic obstructive pulmonary disease (COPD) patients, according to the Global Initiative for Chronic Obstructive Pulmonary Disease (GOLD) guideline standards, the degree of airflow limitation (expressed as a percentage of predicted FEV1) and symptom score (usually using CAT score or mMRC dyspnea score) are used as two key dimensions. When FEV1 ≥ 80% of the predicted value, it is mild limitation, 50% ≤ FEV1 < 80% of the predicted value is moderate limitation, 30% ≤ FEV1 < 50% of the predicted value is severe limitation, and FEV1 < 30% of the predicted value is extremely severe limitation. In terms of symptom burden, when CAT < 10 points or mMRC < 2 points, it is less symptomatic, and CAT ≥ 10 points or mMRC ≥ 2 points is more symptomatic. Combined with the history of acute exacerbation in the past year (0 times, 1 time of non-hospitalization exacerbation, ≥ 2 times of non-hospitalization exacerbation, or ≥ 1 time of hospitalization exacerbation), COPD patients are divided into A / B / C / D four groups: A group (low symptoms, low risk), B group (high symptoms, low risk), C group (low symptoms, high risk), and D group (high symptoms, high risk). When the patient is asthma, then according to the Global Initiative for Asthma (GINA) guidelines, based on symptom control level and acute attack risk assessment, classification is performed. The symptom control level is divided into good control (no daytime symptoms or ≤ 2 times per week, no nighttime symptoms, no activity limitation, no rescue medication or ≤ 2 times per week), partial control (1-2 indicators of poor control), and uncontrolled (≥ 3 indicators of poor control). At the same time, considering the severity of asthma, it is divided into mild, moderate, severe, and extremely severe, forming a standardized disease classification result.

[0076] According to the disease classification results, the large model retrieves matching clinical guideline rules to generate a basic treatment framework. For COPD patients, when the patient belongs to group A, the mild medication regimen is retrieved, mainly including the use of short-acting bronchodilators as needed; when the patient belongs to group B, the moderate medication regimen is retrieved, recommending regular use of long-acting bronchodilators; when the patient belongs to group C or group D, the severe medication regimen is retrieved, including the combined use of LABA and LAMA, and if necessary, the addition of inhaled corticosteroids (ICS) or phosphodiesterase-4 inhibitors (PDE4i). For asthma patients, the treatment regimen is matched according to the stepwise treatment principle, starting from low-dose ICS, and gradually adjusting to ICS / LABA combination therapy according to the control situation, and considering the addition of leukotriene receptor antagonists or biological agents for severe patients. Through knowledge graph retrieval of related drug indications, contraindications, and recommended doses, a basic treatment framework for specific disease classification is formed.

[0077] The basic treatment framework is compared with the patient's previous treatment records and adverse drug reaction history. The drug use, dose adjustment history, and treatment response records in the patient's previous treatment records are structured extracted to analyze drug effectiveness. At the same time, the adverse reaction history is checked to confirm whether the patient has an allergic reaction, intolerance, or other adverse reactions to a specific drug or drug component. By setting drug selection priority rules: when the patient has a good response to a certain class of drugs in the past, the priority of this class of drugs is increased; when the patient has an adverse reaction history to a certain class of drugs, the drug is deleted or reduced in priority from the recommended list; when the patient has a response to multiple drugs, the combination with the least adverse reactions and the best efficacy is selected. Based on the preliminary individualized drug regimen, detailed drug administration information is developed. The preferred drug class combination is determined, such as LAMA+LABA dual bronchodilator combination for COPD patients, and ICS+LABA combination for asthma patients. Then specific drugs are selected from each category, considering drug characteristics (such as onset time, duration of action), drug administration device type (such as metered-dose inhaler, dry powder inhaler, nebulizer), and patient use ability. The administration route is preferentially selected by inhalation to ensure that the drug directly acts on the respiratory tract and reduces systemic adverse reactions. The dosage is individualized according to the standard dose in the drug instructions combined with the patient's body weight, age, liver and kidney function, to develop accurate drug administration time points and frequency. The treatment course design considers the different needs of short-term intensive treatment and long-term maintenance treatment in the acute phase, including the initial dose phase, the stable phase, and the adjustment phase, with clear phase evaluation points and dose adjustment conditions.

[0078] On the basis of the drug treatment regimen, non-drug treatment recommendations are created. Respiratory rehabilitation training plans include respiratory control techniques, limb exercise prescriptions, and endurance training programs, which are designed individually based on the patient's basic physical fitness, activity tolerance, and comorbidities. Lifestyle adjustment recommendations include smoking cessation guidance, environmental control measures, dietary guidance, and occupational protection measures. Psychological health intervention measures mainly target the common anxiety and depression problems of respiratory disease patients, providing cognitive behavioral intervention strategies and coping skills. Disease self-management skills training includes symptom recognition, medication skills, correct use of inhalation devices, and daily monitoring points. For patients with a high risk level, an acute exacerbation emergency plan is additionally developed, with clear warning symptoms (such as significantly worsening dyspnea, increased sputum volume, and sputum color change), self-treatment measures (such as adjusting drug dosage, body position drainage), and judgment criteria for seeking medical treatment, to ensure that patients can respond to changes in their condition in a timely manner. These contents are integrated into an individualized diagnosis and treatment plan, covering comprehensive guidance for drug and non-drug interventions.

[0079] The personalized diagnosis and treatment plan is matched with the patient's characteristics for matching degree evaluation. The matching degree evaluation is divided into multiple dimensions: a past medication response matching score evaluates the consistency of the recommended drugs in the evaluation plan with the patient's historical medication response; a lifestyle matching score evaluates the adaptability of the treatment recommendation to the patient's daily activity patterns and life rules; an adherence factor matching score evaluates the patient's ability and willingness to execute a complex treatment plan; an economic affordability matching score evaluates the matching degree of treatment costs and the patient's economic situation; a medical insurance coverage matching score evaluates the medical insurance reimbursement of the recommended drugs and treatment programs. Each dimension is given different weights, and the weighted total score is calculated to obtain a matching degree score in percentage. When the matching degree score is lower than the preset threshold (usually 60 points), the diagnosis and treatment plan is adjusted until a satisfactory matching degree is achieved, ensuring that the plan meets both medical guidelines and is practical.

[0080] In a specific embodiment, the process of performing step S105 can specifically include the following steps:

[0081] According to the patient risk stratification evaluation results and the matching degree score, low-risk patients are assigned to primary medical institutions, medium-risk patients are arranged for management by primary medical institutions and regular follow-up by specialist hospitals, and high-risk patients are managed by specialist hospitals and followed up by primary medical institutions, to construct a hierarchical diagnosis and treatment path diagram;

[0082] Different levels of medical institutions are connected through an intelligent collaboration platform to share data on medical records, test results, and diagnosis and treatment opinions, forming a medical collaboration network;

[0083] For respiratory chronic disease patients treated at primary medical institutions, patient data and doctor's preliminary judgment are input into the respiratory chronic disease large model to analyze symptom characteristics, disease progression risk, and treatment difficulty, and generate diagnosis and treatment decision recommendations;

[0084] Multi-dimensional symptom monitoring is performed on patients starting treatment, recording changes in core symptom scores, physiological indicator changes, medication adherence data, and quality of life assessment data, to construct a full-dimensional monitoring data set;

[0085] Time series analysis is performed on the full-dimensional monitoring data set to calculate symptom improvement rate, physiological indicator recovery rate, medication adherence index, and quality of life improvement amplitude, to generate multi-dimensional efficacy indicators;

[0086] Based on the multi-dimensional efficacy indicators, a weighted aggregation algorithm is used to calculate a comprehensive treatment response score, and the trend of efficacy is determined according to the score change trend. When the score continuously decreases for more than three monitoring periods, the treatment plan adjustment process is triggered, and a time series treatment response index is formed.

[0087] Specifically, the rational distribution of patients was achieved by the hierarchical diagnosis and treatment pathway. According to the risk stratification assessment results and the matching score of diagnosis and treatment plan, a resource allocation decision matrix was established. Low-risk patients (risk score <40) were assigned to primary medical institutions, and routine management was performed by community health service centers or township health clinics. Medium-risk patients (risk score 40-65) were arranged to primary medical institutions for daily management, and followed up in specialized hospitals every 3 months to achieve two-way management. High-risk patients (risk score >65) were managed by the respiratory department of specialized hospitals, and primary medical institutions cooperated in monthly follow-up to ensure timely detection of changes in disease. For patients with a matching score of less than 70 points, regardless of the risk level, the first diagnosis was arranged in specialized hospitals, and the initial plan was developed by specialist doctors before being transferred to the corresponding level of medical institutions. The hierarchical diagnosis and treatment pathway was constructed in the form of a flowchart, which clearly defined the responsibility subjects, referral criteria, and information flow rules at each node, and converted abstract hierarchical management principles into executable specific processes. Data exchange and collaboration between multiple levels of medical institutions were achieved through an intelligent collaboration platform. The platform uses a federated learning architecture, with each medical institution retaining local data storage while sharing necessary information through a secure channel. Medical record information sharing includes disease diagnosis standardization, medication history structuring, and treatment response quantification, making medical records readable across institutions; examination result sharing uses a unified standard format, such as converting lung function test values to predicted value percentages to enable data comparability between different laboratories; and diagnosis and treatment opinion sharing records the judgments and recommendations of specialist doctors and primary care physicians according to standardized templates. During the data sharing process, differential privacy technology is used to protect patients' sensitive information, and blockchain technology is used to record data access and modification logs to ensure traceability throughout the process. Through shared data and interconnection mechanisms, a collaborative network connecting different levels of medical institutions is constructed, achieving resource integration and diagnosis and treatment collaboration.

[0088] For respiratory patients treated in primary medical institutions, primary care physicians enter patient baseline data and initial judgments, including chief complaints, physical examination findings, simple lung function test results, and initial diagnosis opinions. After standardization, these data are input into the respiratory disease large model, which assists in analysis based on the knowledge graph, calculates symptom pattern similarity, identifies typical or atypical manifestations, assesses treatment difficulty, and generates diagnosis and treatment decision recommendations. Decision recommendations include diagnosis confirmation or correction, examination suggestions, treatment adjustment suggestions, and referral indication judgments. When the large model detects complex situations that primary medical institutions cannot handle, such as severe complications, atypical symptom combinations, or poor treatment response, it automatically generates a referral application containing a key information summary and referral reasons and pushes it to the corresponding specialized hospital workstation.

[0089] Multi-dimensional symptom monitoring is performed on patients with respiratory chronic diseases who start treatment, and a dynamic tracking system is constructed. Changes in core symptom scores are collected through regular questionnaires, including dyspnea mMRC score, cough score, sputum score, etc. Patients fill out the questionnaire once a week, and the trend is recorded. Changes in physiological indicators are collected through portable monitoring devices, such as FEV1 values measured by a lung function meter and night-time blood oxygen fluctuation recorded by a blood oxygen saturation monitor. The data is automatically uploaded to the management platform. Medication adherence data is recorded by an intelligent medicine box, which records the actual medication time and frequency, combined with the patient's self-reported information, and is quantified into an adherence percentage. Quality of life assessment data is obtained by using respiratory disease-specific scales such as CAT or AQLQ to regularly assess changes in patient quality of life. After format unification and missing value processing, multi-source data is stored in the patient's electronic health record, forming a time series data set and constituting a full-dimensional monitoring data set.

[0090] Time series analysis is performed on the full-dimensional monitoring data set, and multiple efficacy indicators are calculated. Symptom improvement rate is calculated based on changes in core symptom scores. The difference between pre-treatment and post-treatment scores is divided by the initial score to obtain a standardized improvement percentage. Physiological indicator recovery rate is calculated by tracking the recovery of key physiological parameters such as FEV1 relative to expected normal values. The medication adherence index is calculated by dividing the actual number of medications by the prescribed number of medications, combined with the medication time accuracy score to calculate a comprehensive adherence index. The quality of life improvement amplitude is calculated by the difference between the pre- and post-standardized scale scores. For each indicator, a threshold is set, such as a symptom improvement rate > 30% for significant improvement, 15-30% for moderate improvement, and < 15% for slight improvement. These calculation results form structured multi-dimensional efficacy indicators, which directly reflect various aspects of treatment effectiveness.

[0091] Based on the multi-dimensional efficacy indicators, a comprehensive treatment response score is calculated by a weighted aggregation algorithm. The weights of each indicator are determined. For COPD patients, lung function indicators and dyspnea scores have higher weights. For asthma patients, symptom control rate and acute attack frequency have higher weights. Each indicator is standardized to the 0-100 point interval, and the weighted sum is obtained according to the weights to obtain the comprehensive treatment response score. The score is calculated once per period (usually 2 weeks) to form a time series data. Through time series analysis methods, including moving average, trend analysis and seasonal decomposition, the change trend of treatment response is evaluated. Monitoring warning rules are set: when the score continuously rises, it indicates good treatment effect; when the score is stable, it indicates stable disease; when the score fluctuates significantly, it indicates unstable control; when the score continuously decreases for more than three monitoring periods, the treatment regimen adjustment process is triggered. The comprehensive score and its change rate calculated each time are stored in the patient's time series database to form a time series treatment response index.

[0092] In a specific embodiment, the process of performing step S106 can specifically include the following steps:

[0093] Based on the disease type, severity, cognitive level, technical acceptance and efficacy indicators of the patient, the content of disease basic knowledge, symptom recognition method, medication guidance content, lifestyle adjustment strategy and acute exacerbation prevention measures is screened to generate a personalized health education content library;

[0094] The personalized health education content library is converted into multimedia resources in the form of text, pictures, videos and interactive question and answer, the presentation depth is dynamically adjusted according to the learning progress and content understanding level of the patient, and a hierarchical education resource package is constructed;

[0095] According to the individualized diagnosis and treatment scheme and the treatment response index, an electronic medicine box reminding system, a symptom self-monitoring diary, a breathing training guidance system and an environmental risk early warning system are developed as functional modules to form a patient self-management tool set;

[0096] According to the technical acceptance ability of the patient, the health education content and the self-management tool set are integrated into mobile applications, web versions and smart wearable device multi-channel access ports to generate a patient end interactive interface;

[0097] The operation behavior, feedback opinions, diagnosis and treatment effect evaluation and large model performance indicators of the patient and medical staff during the use of the system are recorded and analyzed to form a system optimization data set;

[0098] Based on the system optimization data set, the knowledge base of the respiratory chronic disease large model is updated and the algorithm is optimized, the efficiency of the system process nodes is analyzed and the resources are optimized, and the user interface and the interaction logic are optimized for human-computer interaction to complete the system iteration and process optimization.

[0099] Specifically, the personalized health education content is screened through patient characteristic parameter analysis, wherein the patient disease type is classified into respiratory chronic disease categories such as chronic obstructive pulmonary disease, asthma, bronchiectasis, etc.; the severity is classified according to clinical classification standards, such as GOLD classification of COPD or control level classification of asthma; the cognitive level is determined as primary, intermediate or advanced through simple knowledge evaluation; the technology acceptance is evaluated as low, medium and high levels through technology use habit questionnaire; and the efficacy index data calculated previously is integrated. Based on the five-dimensional feature vectors, relevant content is extracted from the respiratory system disease knowledge base, the etiology and mechanism, natural course of disease, prognosis factors and other content related to the specific disease type of the patient are selected for disease basic knowledge; the early warning symptom identification points and self-assessment tools suitable for the severity of the patient are extracted for symptom identification method; the medication guide content is directly associated with the individualized diagnosis and treatment plan of the patient, including the mechanism of action of the drug used, the correct use method, precautions, etc.; the lifestyle adjustment strategy selects appropriate exercise programs, dietary suggestions and environmental control measures based on the individual situation of the patient; and the acute exacerbation prevention measures are customized according to the past attack mode and risk factors of the patient. The content modules are matched with the patient feature vectors through a similarity matching algorithm to screen the most suitable content combination for the patient, and a personalized health education content library is formed. The personalized health education content library is converted into diversified multimedia resources to realize the acceptability and interactivity of the content. The text materials adopt hierarchical reading difficulty control, the vocabulary complexity and sentence structure are set according to the cognitive level of the patient, and professional term explanations are provided; the picture resources include medical illustrations, anatomical diagrams and operation diagrams, which intuitively present abstract concepts through visualization; the video resources focus on dynamic skill demonstration, such as inhaler use method, breathing exercise action and body position drainage technology; and the interactive question and answer is designed to have ladder-type difficulty knowledge detection questions to evaluate the understanding degree of the patient on the learned content. All resources are divided into three levels of basic, advanced and professional according to the content complexity, the initial presentation is basic level content, the learning behavior data of the patient is recorded, including reading time, video completion rate, question and answer correct rate and other indicators, the learning effect is analyzed through decision tree algorithm, when the question and answer correct rate of the basic level content is more than 80%, the advanced level content is automatically unlocked; when the advanced content test is passed, the professional level content is opened, the dynamic adjustment of education depth is realized, and a hierarchical education resource package is constructed.

[0100] According to the personalized diagnosis and treatment plan and the treatment response index, a set of complementary self-management tools is developed. The electronic medicine box reminder system sets accurate medication time points and dose reminders according to the patient's medication plan, records actual medication use, and calculates medication adherence indicators; when missing or taking the wrong medicine for three consecutive times, the intelligent reminder upgrade mechanism is triggered to ensure the regularity of medication through stronger prompt sound, relative contact, etc. The symptom self-monitoring diary designs a simple daily recording interface, guides patients to regularly assess changes in core symptoms, automatically generates symptom trend charts, and identifies abnormal fluctuations. The respiratory training guidance system provides personalized respiratory muscle training plans and whole-body aerobic exercise suggestions based on the patient's lung function status and physical fitness level, assesses training effectiveness with sensor feedback, and dynamically adjusts difficulty. The environmental risk warning system integrates real-time weather data, air quality index, and patient personal sensitivity factors to predict high-risk exposure and issue warnings 12-24 hours in advance, while providing corresponding protection measures recommendations.

[0101] According to the patient's technical acceptance ability, the health education content and self-management tools are integrated into various technical platforms to achieve seamless access. Technical acceptance ability is determined by questionnaire evaluation combined with age characteristics and actual device usage habits, divided into low, medium and high levels. For patients with high technical acceptance ability, a functional mobile application is provided, which contains all education resources and management tools, supports complex interaction and data visualization; for patients with medium technical acceptance ability, a simplified mobile application or web portal is provided, which highlights the core functions and simplifies the operation process; for patients with low technical acceptance ability, basic monitoring is mainly realized through smart wearable devices, supplemented by paper materials and telephone follow-up for auxiliary management. Data is synchronized and shared between platforms to ensure that patients can access consistent health management services at different portals. The interactive interface design follows the principle of intuitiveness, using large fonts, high-contrast colors and icons consistent with cognitive habits to reduce the threshold and improve patient acceptance.

[0102] A comprehensive data collection is conducted on the actual operation of the system to form the basis for iterative optimization. Patient operation behavior data includes usage frequency, function access path, dwell time and interaction depth, which are recorded in real time through the burying point technology; the operation behavior data of medical staff focuses on the use of clinical decision support functions and the adjustment frequency. Feedback opinions are collected through structured questionnaires and free text forms, covering system usability, content practicality and function integrity evaluation. Diagnosis and treatment effect evaluation data integrates disease control index changes, acute exacerbation frequency changes and life quality score changes to quantify the system intervention effect. The performance indicators of large models record the accuracy of diagnosis suggestions, the rationality of treatment plans and the satisfaction of patient problem answers to evaluate the performance of artificial intelligence core components. These multi-dimensional data are cleaned, standardized and correlated to form a structured system optimization data set to provide the basis for the next iteration.

[0103] Multi-level iterative updates are performed based on system optimization datasets. The knowledge base update of the respiratory chronic disease large model is performed through incremental learning, and new clinical guidelines, research findings and typical cases are added to the training corpus to expand the model knowledge. Algorithm optimization is performed by identifying case types with large prediction deviation through error analysis, and the weight of these samples in the training set is specifically strengthened to improve the accuracy in specific scenarios. System process node optimization is based on user behavior data to identify links with longer operation time or higher error rate, and the operation is simplified through process reengineering or auxiliary prompts. Resource allocation optimization adjusts the proportion of computing resource allocation according to the use frequency and importance of each functional module to improve the system response speed. User interface and interaction logic optimization uses A / B testing method to compare the user acceptance and operation efficiency of different design schemes, and selects the optimal solution. The whole iterative process is carried out in a closed loop according to the "collect data - analyze problems - implement optimization - effect evaluation", to ensure the continuous evolution of the system and adapt to the changing clinical needs and technical environment.

[0104] The above describes the respiratory chronic disease intelligent diagnosis and treatment management method based on a large model in the embodiments of the present application. The following describes the respiratory chronic disease intelligent diagnosis and treatment management system based on a large model in the embodiments of the present application. Please refer to Figure 2 An embodiment of the respiratory chronic disease intelligent diagnosis and treatment management system based on a large model in the embodiments of the present application includes:

[0105] The acquisition module 201 is configured to acquire and process multi-dimensional data of the respiratory chronic disease patient to obtain a structured patient dataset.

[0106] The input module 202 is configured to input the structured patient dataset and the respiratory system disease professional knowledge base into a double encoding pre-training and contrast learning fine-tuning framework to obtain a respiratory chronic disease large model and a respiratory chronic disease knowledge graph.

[0107] The calculation module 203 is configured to perform semantic analysis and multi-dimensional calculation on patient symptom description and evaluation scale answers based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph to generate a patient risk stratification evaluation result.

[0108] The generation module 204 is configured to retrieve clinical guideline rules using the respiratory chronic disease large model and combining patient individual characteristics to generate an individualized diagnosis and treatment scheme and a matching degree score according to the patient risk stratification evaluation result.

[0109] The construction module 205 is configured to construct a multi-level medical institution collaboration framework and perform dynamic efficacy monitoring according to the individualized diagnosis and treatment scheme and the matching degree score to generate efficacy indicators and treatment response indexes.

[0110] A customization module 206 is configured to customize patient health education content and self-management tools according to the therapeutic index and the treatment response index, while collecting system operation data for model updating and process optimization.

[0111] Through the synergistic cooperation of the above-mentioned various components, the comprehensive acquisition and standardized processing of respiratory chronic disease related information are realized by multi-dimensional data acquisition and processing to form a structured patient data set; secondly, the structured patient data set and the respiratory system disease professional knowledge base are processed by using the dual coding pre-training and contrast learning fine-tuning framework, and the generated respiratory chronic disease large model and respiratory chronic disease knowledge graph have deep medical understanding ability and professional reasoning ability, which are significantly better than traditional machine learning models, and can understand complex medical context and extract key clinical features therefrom; thirdly, based on the respiratory chronic disease large model and the knowledge graph, semantic analysis and multi-dimensional calculation are performed on the patient symptom description and the answer of the evaluation scale, and the generated patient risk stratification evaluation result has high accuracy and can capture subtle disease feature differences, thereby realizing precise stratification and optimizing medical resource allocation; fourthly, according to the patient risk stratification evaluation result, the respiratory chronic disease large model is used to retrieve clinical guideline rules and generate an individualized diagnosis and treatment scheme and a matching degree score in combination with the individual characteristics of the patient, which fully considers the individual differences of the patient, so that the treatment scheme conforms to the principle of evidence-based medicine and is highly individualized, thereby improving the treatment accuracy and compliance; fifthly, the multi-level medical institution collaboration framework and the dynamic efficacy monitoring system constructed according to the individualized diagnosis and treatment scheme and the matching degree score break the medical information island, realize the reasonable allocation of high-quality medical resources and continuous medical services, and reduce repeated examinations and treatment errors; the patient health education content and self-management tools customized according to the therapeutic index and the treatment response index are combined with the continuous collection of system operation data and model updating and process optimization to form a closed-loop quality improvement system, so that the system can continuously evolve and improve. It is particularly emphasized that the large model algorithm applied in the present scheme makes great contribution to the field of respiratory chronic disease diagnosis and treatment management, the context understanding ability of which enables the model to extract key clinical information from unstructured medical text, the knowledge enhanced reasoning mechanism realizes the automation support of complex medical decision making, the multi-modal data fusion algorithm can integrate patient information from different sources to form a comprehensive view, and the adaptive learning characteristic enables the system to continuously optimize the diagnosis and treatment strategy according to new data, and these specific algorithm characteristics jointly construct a highly specialized intelligent auxiliary decision making system for respiratory system diseases, which greatly improves the efficiency and accuracy of respiratory chronic disease management.

[0112] The above Figure 2 The large model based respiratory chronic disease intelligent diagnosis and treatment management system in the embodiment of the present application is described in detail from the perspective of modular functional entities, and the large model based respiratory chronic disease intelligent diagnosis and treatment management device in the embodiment of the present application is described in detail from the perspective of hardware processing.

[0113] Figure 3 is a structural schematic diagram of a large model-based respiratory chronic disease intelligent diagnosis and treatment management device provided by an embodiment of the present application. The large model-based respiratory chronic disease intelligent diagnosis and treatment management device 300 can have relatively large differences due to different configurations or performances, and can include one or more processors (central processing units, CPUs) 310 (for example, one or more processors) and a memory 320, one or more storage media 330 (for example, one or more mass storage device ends) storing application programs 333 or data 332. The memory 320 and the storage medium 330 can be temporary storage or persistent storage. The programs stored in the storage medium 330 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the large model-based respiratory chronic disease intelligent diagnosis and treatment management device 300. Further, the processor 310 can be configured to communicate with the storage medium 330 and execute a series of instruction operations in the storage medium 330 on the large model-based respiratory chronic disease intelligent diagnosis and treatment management device 300 to implement the steps of the above-mentioned large model-based respiratory chronic disease intelligent diagnosis and treatment management method.

[0114] The large model-based respiratory chronic disease intelligent diagnosis and treatment management device 300 can also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and the like. Those skilled in the art can understand that, Figure 3 The structure of the large model-based respiratory chronic disease intelligent diagnosis and treatment management device shown does not constitute a limitation on the large model-based respiratory chronic disease intelligent diagnosis and treatment management device provided by the present application, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0115] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium, and can also be a volatile computer readable storage medium. The computer readable storage medium has instructions stored therein, and when the instructions are run on a computer, the computer executes the steps of the large model-based respiratory chronic disease intelligent diagnosis and treatment management method.

[0116] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0117] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a large model-based respiratory chronic disease intelligent diagnosis and treatment management device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (read-only memory, ROM), a random access memory (random access memory, RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0118] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A large model-based respiratory chronic disease intelligent diagnosis and treatment management method, characterized in that, The method comprises: Multi-dimensional data acquisition and processing are performed on respiratory chronic disease patients to obtain a structured patient data set; The structured patient data set and a respiratory system disease professional knowledge base are input into a double coding pre-training and contrast learning fine-tuning framework to obtain a respiratory chronic disease large model and a respiratory chronic disease knowledge graph; Based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph, semantic analysis and multi-dimensional calculation are performed on patient symptom descriptions and evaluation scale answers to generate patient risk stratification evaluation results; According to the patient risk stratification evaluation results, the respiratory chronic disease large model is used to retrieve clinical guideline rules and combine patient individual characteristics to generate individualized diagnosis and treatment schemes and matching scores; According to the individualized diagnosis and treatment schemes and the matching scores, a multi-level medical institution collaboration framework is constructed and dynamic therapeutic effect monitoring is performed to generate therapeutic effect indicators and treatment response indexes; According to the therapeutic effect indicators and the treatment response indexes, patient health education content and self-management tools are customized, and system operation data are collected for model updating and process optimization; The structured patient data set and the respiratory system disease professional knowledge base are input into a double coding pre-training and contrast learning fine-tuning framework to obtain a respiratory chronic disease large model and a respiratory chronic disease knowledge graph, comprising: Respiratory disease clinical guidelines, evidence-based medicine literature databases, expert consensus documents, typical case sets and drug interaction databases are integrated to construct a respiratory system disease professional knowledge base; The structured patient data set and medical text corpus in the respiratory system disease professional knowledge base are simultaneously input into a Transformer encoder structure, a mask language model and a next sentence prediction task are used for multi-round iterative training to obtain initial large model parameters; Based on the initial large model parameters, a contrast learning framework is constructed to pull similar case representation vectors and push different case representation vectors, and model parameters are optimized by minimizing a contrast loss function to generate a respiratory chronic disease large model; Entity extraction is performed on the structured patient data set and the respiratory system disease professional knowledge base to identify diseases, symptoms, drugs, examinations and risk factors as medical concepts to obtain a medical entity set; Relationship extraction is performed on the medical entity set to determine disease-symptom, disease-drug, drug-drug, symptom-severity and examination-reference value as entity interrelation to form a relationship triple set; The medical entity set and the relationship triple set are subjected to knowledge representation learning through a graph embedding algorithm to construct a respiratory chronic disease knowledge graph, and are associated with the respiratory chronic disease large model through entity linking technology to support knowledge-enhanced reasoning; The graph embedding algorithm uses an improved TransE algorithm which learns entity and relationship vector representations by optimizing a target function that the head entity vector plus the relationship vector approaches the tail entity vector, and the generated vector representation retains the semantic relationship between entities to constitute a respiratory chronic disease knowledge graph; The semantic analysis and multi-dimensional calculation of patient symptom descriptions and evaluation scale answers based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph to generate patient risk stratification evaluation results comprise: The patient is guided to describe the current symptom experience through a natural language interaction interface, the patient's natural language input is transmitted into the respiratory chronic disease large model, the duration, severity, trigger, and relieving factors of the symptoms are extracted as key symptom attributes, and a symptom entity matrix is generated; The patient is guided to complete a structured symptom assessment questionnaire including dyspnea score, cough frequency score, sputum character score, activity limitation score, sleep impact score, and medication compliance score, and a questionnaire numerical vector is obtained; The symptom entity matrix and the questionnaire numerical vector are fused, and the lung function indicators and blood oxygen saturation values in the basic physiological monitoring data are combined to construct a multi-dimensional feature space and form a comprehensive symptom representation of the patient; The dyspnea, cough, and sputum in the comprehensive symptom representation of the patient are taken as core symptoms and weighted coefficients are assigned, the severity values of each symptom are multiplied by the corresponding weighted coefficients and summed through a weighted calculation method to obtain a symptom burden index; The recent symptom fluctuation trend, acute attack risk, medication compliance value, and environmental factor influence are analyzed in time sequence, a disease stability index is generated through a stability calculation model, and a comprehensive risk score is calculated based on the basic disease severity classification, comorbidity, and social support score; According to the threshold interval of the comprehensive risk score, patients with a risk score below the first threshold are classified as low-risk level, patients with a risk score between the first threshold and the second threshold are classified as medium-risk level, and patients with a risk score higher than the second threshold are classified as high-risk level, and a patient risk stratification evaluation result is generated; According to the patient risk stratification evaluation result, the respiratory chronic disease large model is used to retrieve clinical guideline rules and combine patient individual characteristics to generate a personalized diagnosis and treatment plan and a matching score, including: The patient risk stratification evaluation result is input into the respiratory chronic disease large model to identify the patient's disease classification, when the patient has chronic obstructive pulmonary disease, the patient is divided into four groups A / B / C / D according to the airflow limitation and symptom score, and when the patient has asthma, the patient is labeled based on the control level and severity to generate a disease classification result; The disease classification result is matched with clinical guideline rules, when the patient belongs to group A, a mild medication plan is retrieved, when the patient belongs to group B, a moderate medication plan is retrieved, and when the patient belongs to group C or D, a severe medication plan is retrieved to obtain a basic treatment framework; The basic treatment framework is compared and analyzed with the patient's previous treatment records and drug adverse reaction history to generate a preliminary individualized drug plan; Based on the preliminary individualized drug plan, drug information is developed, including preferred drug class combination, specific drug selection, administration route, dosage, treatment course design, and a drug treatment plan is generated; Combined with the drug treatment plan, non-drug treatment suggestions are created, including respiratory rehabilitation training plan, lifestyle adjustment suggestions, psychological health intervention measures, and disease self-management skill training, when the patient's risk level is high, an acute exacerbation emergency plan is developed to form a personalized diagnosis and treatment plan. The personalized diagnosis and treatment scheme is matched with patient characteristics for evaluation, and patient past medication reaction matching scores, life habit matching scores, compliance factor matching scores, economic affordability matching scores, and medical insurance coverage matching scores are calculated to obtain a matching degree score in percentage; According to the personalized diagnosis and treatment scheme and the matching degree score, a multi-level medical institution collaborative framework is constructed and dynamic therapeutic effect monitoring is performed to generate therapeutic effect indicators and treatment response indexes, including: According to the patient risk stratification evaluation results and the matching degree score, low-risk patients are assigned to primary medical institutions, medium-risk patients are arranged for management by primary medical institutions and regular follow-up by specialist hospitals, and high-risk patients are managed by specialist hospitals and followed up by primary medical institutions, thereby constructing a hierarchical diagnosis and treatment path diagram; Different levels of medical institutions are connected through an intelligent collaborative platform to share data such as medical records, test results, and diagnosis and treatment opinions, forming a medical collaborative network; For respiratory chronic disease patients treated in primary medical institutions, patient data and doctor's preliminary judgments are input into the respiratory chronic disease large model to analyze symptom characteristics, disease progression risk, and treatment difficulty, and generate diagnosis and treatment decision suggestions; For patients starting treatment, multi-dimensional symptom monitoring is performed to record changes in core symptom scores, physiological index changes, medication compliance data, and quality of life assessment data, and a full-dimensional monitoring data set is constructed; The full-dimensional monitoring data set is subjected to time series analysis to calculate symptom improvement rate, physiological index recovery rate, medication compliance index, and quality of life improvement amplitude, and generate multi-dimensional therapeutic effect indicators; Based on the multi-dimensional therapeutic effect indicators, a treatment response comprehensive score is calculated by a weighted aggregation algorithm, and the trend of therapeutic effect is determined according to the score change trend. When the score continuously decreases for more than three monitoring periods, the treatment scheme adjustment process is triggered, and a time series treatment response index is formed; According to the therapeutic effect indicators and the treatment response index, patient health education content and self-management tools are customized, and system operation data is collected for model updating and process optimization, including: Based on the patient's disease type, severity, cognitive level, technical acceptance, and therapeutic effect indicators, disease basic knowledge, symptom recognition methods, medication guidance content, lifestyle adjustment strategies, and acute exacerbation prevention measures are content-filtered to generate an individualized health education content library; The individualized health education content library is converted into multimedia resources in the form of text, pictures, videos, and interactive question and answer, and the presentation depth is dynamically adjusted according to the patient's learning progress and content understanding level to construct a hierarchical education resource package; According to the personalized diagnosis and treatment scheme and the treatment response index, an electronic medicine box reminder system, a symptom self-monitoring diary, a respiratory training guidance system, and an environmental risk warning system are developed as functional modules to form a patient self-management tool set; According to the patient's technical acceptance ability, the health education content and the self-management tool set are integrated into mobile applications, web versions, and smart wearable device multi-channel access ports to generate a patient-end interactive interface; Record and analyze the patient and medical staff operation behavior, feedback, diagnosis and treatment effect evaluation, and large model performance indicators during system use, to form a system optimization dataset; Based on the system optimization dataset, update the knowledge base and optimize the algorithm of the respiratory chronic disease large model, analyze the efficiency and optimize the resource allocation of the system process nodes, optimize the human-computer interaction of the user interface and interaction logic, and complete the system iteration and process optimization.

2. The large model-based respiratory chronic disease intelligent diagnosis and treatment management method according to claim 1, characterized in that, The multi-dimensional data of the respiratory chronic disease patient is collected and processed to obtain a structured patient dataset, which includes: Through medical institution clinical input, patient mobile terminal self-reporting, and intelligent wearable device collection of patient basic information, disease history, medication, symptom evaluation, and physiological monitoring data, a multi-source heterogeneous dataset is formed; Standardize and normalize the structured data in the multi-source heterogeneous dataset, perform word segmentation, stop word removal, and entity standardization operations on text data, and implement noise reduction, missing value interpolation, and outlier processing on time series data to generate a preprocessed dataset; The preprocessed dataset is labeled according to the respiratory system drug classification method, a unique patient identifier is established, and an initial labeled dataset is generated; The initial labeled dataset is organized in a tree hierarchy structure, and the top-level patient basic information, secondary-level disease classification, and bottom-level symptom monitoring indicators are associated to the corresponding patient archives to obtain a hierarchical associated dataset; Perform integrity check, consistency verification, and outlier marking on the hierarchical associated dataset to generate a quality controlled dataset; Store the quality controlled dataset in a secure database to form a structured patient dataset.

3. A large model-based respiratory chronic disease intelligent diagnosis and treatment management system, characterized in that, The system for implementing the large model-based respiratory chronic disease intelligent diagnosis and treatment management method according to any one of claims 1-2 includes: A collection module for collecting and processing multi-dimensional data of respiratory chronic disease patients to obtain a structured patient dataset; An input module for inputting the structured patient dataset and respiratory system disease knowledge base into a double encoding pre-training and contrast learning fine-tuning framework to obtain a respiratory chronic disease large model and a respiratory chronic disease knowledge graph; A calculation module for performing semantic analysis and multi-dimensional calculation on patient symptom descriptions and evaluation scale answers based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph to generate patient risk stratification evaluation results; A generation module for retrieving clinical guideline rules using the respiratory chronic disease large model and combining patient individual characteristics to generate personalized diagnosis and treatment plans and matching scores based on the patient risk stratification evaluation results; A construction module for constructing a multi-level medical institution collaboration framework and performing dynamic efficacy monitoring based on the personalized diagnosis and treatment plans and the matching scores to generate efficacy indicators and treatment response indices; A customization module for customizing patient health education content and self-management tools based on the efficacy indicators and the treatment response indices, and collecting system operation data for model updating and process optimization.

4. A large model-based respiratory chronic disease intelligent diagnosis and treatment management device, characterized in that, The computer program is stored in the memory and includes a computer program capable of running on the processor, and the processor executes the computer program to realize the large model-based intelligent diagnosis and treatment management method for respiratory chronic diseases in any one of claims 1-2.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is stored in the memory and includes a computer program capable of running on the processor, and the processor executes the computer program to realize the large model-based intelligent diagnosis and treatment management method for respiratory chronic diseases in any one of claims 1-2.

Images