Detailed Description
The embodiment of the application provides a breathing chronic disease intelligent diagnosis and treatment management method and system based on a large model. The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.
For easy understanding, the following describes a specific flow of an embodiment of the present application, referring to fig. 1, and an embodiment of a method for intelligent diagnosis and treatment management of respiratory distress disease based on a large model in the embodiment of the present application includes:
Step S101, carrying out multidimensional data acquisition and processing on a patient suffering from slow breathing to obtain a structured patient data set;
step S102, inputting a 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 slow disease large model and a respiratory slow disease knowledge map;
Step S103, carrying out semantic analysis and multidimensional calculation on the answers of the patient symptom description and the evaluation scale based on the respiratory slow disease large model and the respiratory slow disease knowledge graph to generate a patient risk layering evaluation result;
step S104, according to the risk stratification evaluation result of the patient, a clinical guideline rule is searched by using a respiratory chronic disease large model, and individual characteristics of the patient are combined to generate a personalized diagnosis and treatment scheme and a matching degree score;
step 105, constructing a multi-stage medical institution cooperation frame and executing dynamic curative effect monitoring according to the personalized diagnosis and treatment scheme and the matching degree score to generate a curative effect index and a curative response index;
And S106, customizing the health education content and the self-management tool of the patient according to the curative effect index and the treatment response index, and collecting system operation data to update the model and optimize the flow.
It can be understood that the execution subject of the application may be a large model-based intelligent diagnosis and treatment system for chronic respiratory disease, and may also be a terminal or a server, which is not limited herein. The embodiment of the application is described by taking a server as an execution main body as an example.
Specifically, data is collected through three main channels, namely clinical entry of a medical institution, self-reporting of a patient mobile end and automatic acquisition of intelligent wearable equipment, from multi-dimensional data acquisition of a patient suffering from slow breathing. These data include patient basic information, disease history, medication, symptom assessment, and physiological monitoring data. The collected original data has heterogeneity, pretreatment is needed, standardized and normalized processing is carried out on the structured data, word segmentation and entity standardization are carried out on the text data, and noise reduction and outlier processing are carried out on the time-series data. A unique patient identifier is then established, all data is organized into a tree hierarchy, and after integrity checking and consistency verification, a structured patient data set is formed.
The structured patient dataset and respiratory disease expertise library (including authoritative clinical guidelines, evidence-based medical literature) are entered into a dual coding pre-training and contrast learning fine tuning framework. The framework uses a transducer encoder structure to pretrain a large-scale medical corpus through a mask language model and a next sentence prediction task, then uses a contrast learning method to conduct fine tuning, and draws the representing vectors of similar cases closer together and pushes away the representing vectors of different cases, so as to obtain a large model of the respiratory slow disease. Meanwhile, medical concepts and associations of diseases, symptoms, medicines and the like are identified through entity extraction and relation extraction to form a knowledge triplet, and a knowledge graph of the chronic respiratory disease is constructed through a graph embedding algorithm. Based on the two core components, intelligent analysis is performed on symptom descriptions and evaluation scale answers of patients. The system acquires the symptom description of the patient through a natural language interactive interface, extracts key symptom attributes and forms a symptom entity matrix, and simultaneously guides the patient to complete a structured symptom evaluation questionnaire to obtain a questionnaire numerical vector. Combining the two parts of data with basic physiological monitoring data to construct a multidimensional feature space. The method comprises the steps of assigning weight coefficients to core symptoms, obtaining symptom burden indexes through a weighted calculation method, carrying out time sequence analysis on symptom fluctuation trend, acute episode risk and the like to generate disease stability indexes, calculating comprehensive risk scores by combining basic disease severity grading and complications, and dividing patients into three grades of low, medium and high risks according to the comprehensive risk scores.
And then according to the risk stratification evaluation result, searching corresponding clinical guideline rules by using the large respiratory slowness model. For patients with chronic obstructive pulmonary disease they were divided into four groups a/B/C/D according to airflow limitation and symptom score, and for asthmatic patients, they were labeled based on control level and severity. And then matching different medication schemes according to the disease classification result to form a basic treatment framework. The framework is compared and analyzed with the past treatment record of the patient and the history of adverse drug reaction, the priority of the drug options is adjusted, a personalized drug scheme is formed, and detailed administration information and non-drug treatment suggestions are supplemented. Finally, calculating the matching scores of the patients in terms of past medication response, living habits, compliance factors, economic tolerance and the like to obtain a percentage matching score.
And constructing a multi-level medical institution collaborative framework according to the risk level and the matching degree score of the patient, distributing the low-risk patient to the basic medical institution, arranging basic management and special follow-up visit for the risk patient, and leading the high-risk patient to be managed by a special hospital. The intelligent collaboration platform is connected with medical institutions of different levels to realize data sharing. For patients with basic-level consultation, the system provides large-model auxiliary decision support. The method comprises the steps of performing multidimensional monitoring on a patient starting treatment, recording symptom score change, physiological index change, medication compliance data and life quality evaluation data, calculating each dimensional improvement condition through time sequence analysis, generating a multidimensional curative effect index, calculating treatment response comprehensive scores through a weighted aggregation algorithm, and forming a time sequence treatment response index. And customizing personalized health education content and self-management tools according to the curative effect index and the curative response index. According to the characteristics and curative effect of the patient, the contents of disease knowledge, symptom identification, medication guidance and the like are screened and converted into multimedia resources, and the presentation depth is dynamically adjusted according to the understanding degree of the patient. Development of self-management tools such as an electronic medicine box reminding system, a symptom monitoring diary and the like, and integration of the self-management tools into a multi-channel access platform. And meanwhile, the data such as user behavior, feedback opinion, diagnosis and treatment effect and the like in the using process of the system are recorded and analyzed, and the system is used for updating a large model knowledge base, optimizing an algorithm and optimizing a system flow.
In the embodiment of the application, the structured patient data set is formed through multi-dimensional data acquisition and processing, so that the comprehensive acquisition and standardized processing of the information related to the chronic respiratory disease are realized; secondly, a structured patient data set and a respiratory system disease professional knowledge base are processed by adopting double coding pre-training and contrast learning fine-tuning framework, the generated respiratory disease large model and respiratory disease knowledge graph have deep medical understanding capability and professional reasoning capability, which are obviously superior to those of the traditional machine learning model, complex medical contexts can be understood and key clinical features can be extracted from the complex medical contexts, thirdly, semantic analysis and multidimensional calculation are carried out on patient symptom description and evaluation scale answers based on the respiratory disease large model and the knowledge graph, the accuracy of the generated patient risk layering evaluation result is high, fine disease feature differences can be captured, accurate layering is realized, and therefore medical resource allocation is optimized, fourth, the personalized diagnosis and treatment scheme and matching degree scores generated by utilizing the respiratory disease large model and combining individual characteristics of patients are fully considered, the treatment scheme is in accordance with the rule of evidence-based medical principles and is highly personalized, the treatment accuracy and compliance are improved, fifth, a multistage medical mechanism collaboration framework and a dynamic high-quality monitoring system constructed according to the personalized treatment accuracy and the matching degree score is broken, the medical resource allocation and continuous examination and continuous response of a medical resource allocation and a medical management tool are combined with a medical resource continuous and a medical management system is optimized, and a medical resource allocation and a medical resource is continuously used and a medical management system is updated, and a medical treatment index is continuously provided is updated, a closed-loop quality improvement system is formed, so that the system is continuously evolved and perfected. Particularly, the large model algorithm applied in the scheme has great contribution to the diagnosis and treatment management field of the respiratory chronic diseases, the context understanding capability enables the model to extract key clinical information from unstructured medical texts, the knowledge enhancement reasoning mechanism realizes automatic support of complex medical decisions, the multi-mode 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, so that the efficiency and the accuracy of the management of the respiratory chronic diseases are greatly improved.
In a specific embodiment, the process of executing step S101 may specifically include the following steps:
the method comprises the steps that a multi-source heterogeneous data set is formed by collecting basic information of a patient, medical history, medication condition, symptom assessment and physiological monitoring data through clinical input of a medical institution, self-report of a patient mobile terminal and intelligent wearable equipment;
carrying out standardization and normalization processing on structured data in the multi-source heterogeneous data set, carrying out word segmentation, stop word removal and entity standardization operation on text data, carrying out Shi Jiangzao, missing value interpolation and outlier processing on the time-series data, and generating a preprocessing data set;
carrying out data labeling on the preprocessed data set according to a respiratory system medication classification method, establishing a unique patient identifier, and generating an initial labeling data set;
performing tree hierarchy structure organization on the initial labeling data set, and associating the top-layer patient basic information, the sub-layer disease classification and the bottom-layer symptom monitoring index to corresponding patient files to obtain a hierarchy association data set;
Performing integrity check, consistency verification and outlier marking on the hierarchical association data set to generate a quality-controlled data set;
the quality controlled data set is stored in a secure database to form a structured patient data set.
Specifically, the acquisition of basic data of a patient is realized through clinical input of a medical institution, and medical staff inputs demographic information such as gender, age, occupation, living environment and the like of the patient into a system in the diagnosis and treatment process, and acquires disease history information such as disease diagnosis date, past acute episode times, admission history and the like, and medication conditions such as the type, dosage and compliance record of the currently used medicament. The self report of the mobile terminal of the patient collects daily symptom changes and medication conditions of the patient through specially designed mobile application, and the patient fills in evaluation forms of dyspnea degree, cough frequency, sputum character, activity limitation degree and the like at regular intervals or when the symptoms change. The intelligent wearing equipment automatically collects physiological monitoring data such as the lung function measured value, the blood oxygen saturation, the respiratory frequency and the like of the patient, and the equipment and the system are subjected to real-time or periodic data synchronization to form continuous physiological parameter monitoring records. These data from different sources form a multi-source heterogeneous data set, and due to the variety of data sources, the format and content are different, and data preprocessing is needed. The structured data is standardized, numerical data of different dimensions are converted into unified standards, for example, lung function measured values of different laboratories are unified into expected value percentages, and the data range is adjusted to be between 0 and 1 by the normalization, so that each index has comparability. The method comprises the steps of executing word segmentation operation on text data, decomposing symptom experience described by a patient into basic semantic units, removing stop words, deleting virtual words and connecting words which have no meaning on medical judgment, and mapping synonymous expressions into medical terms when entity standardization is carried out. The processing of time sequence data comprises noise reduction, short-term fluctuation interference elimination through methods such as sliding window averaging, missing value interpolation, reasonable deduction according to front and back values of a time sequence or similar modes of similar patients, abnormal value processing, and data point obviously deviating from a normal range identification and correction.
The preprocessed data set needs to be labeled according to a respiratory medication classification method. The respiratory system medication classification method is a classification system of drugs specially aiming at respiratory diseases, and comprises the categories of bronchodilators, anti-inflammatory drugs, mucus regulators and the like. The data are labeled accordingly based on the patient's history of medication and current treatment regimen. Meanwhile, a unique identifier of the patient is established, and the uniqueness of each patient in the system is ensured by synthesizing a plurality of pieces of personal information and carrying out encryption processing, so that data confusion is avoided, and an initial labeling data set is formed. The method comprises the steps of carrying out tree-like hierarchical structure organization on an initial labeling data set, wherein the top layer stores basic information of a patient, including demographic characteristics and unique identifiers, the second layer is disease classification, classifying the patient into specific respiratory system disease types, such as chronic obstructive pulmonary disease, asthma, interstitial lung disease and the like, and the bottom layer is symptom monitoring index, including all symptom scores, physiological indexes and medication records related to the disease. The three layers of data are associated with corresponding patient files through unique identifiers to form a patient view, and a clear reference relationship is established among the layers of data.
The method comprises the steps of performing quality control on a hierarchical association data set, performing integrity check to confirm whether necessary information of each patient is complete, including basic demographic information, disease diagnosis, core symptom assessment and key physiological indexes, checking whether contradiction exists in the data through consistency check, such as whether different records at a time point conflict, whether symptom descriptions are matched with objective indexes or not, and marking out-of-range data, such as extreme lung function values or abnormal symptom scores, by setting reasonable ranges of various indexes. The data set after quality control is stored in a safe database, and the safety of the patient data is ensured by adopting an encryption storage and access control mechanism, so that a structured patient data set is formed.
Taking the data processing of a patient with chronic obstructive pulmonary disease as an example, the original data is acquired from three channels, the basic information recorded by a medical institution shows that the patient is a 65-year old male and is taken as a retired teacher, the disease history record shows that 2 acute attacks exist in the last year before the COPD diagnosis time is 5 years, and the current medicine comprises salmeterol/fluticasone inhalant twice daily. Symptom data reported by the mobile terminal self include daily dyspnea score (mMRC scale) fluctuation of 2-3 minutes in the past week, cough frequency of 3-5 times daily, and white sticky sputum. The average night blood oxygen saturation recorded by the intelligent wearable device is 93 percent, and the fluctuation is reduced to 88 percent. After preprocessing, the respiratory difficulty score is averaged for 2.5 minutes and standardized to be moderate, blood oxygen data is subjected to noise reduction treatment to eliminate an obvious equipment falling false reading, and chest distress and shortness reported by a patient is standardized to be a medical term of respiratory difficulty through text processing. The data are marked by the combined drug of LABA and ICS, the data are organized into a tree structure after unique identifiers are established, the top layer is personal basic information, the lower layer is COPD diagnosis (GOLD grade B), and the bottom layer is symptom and monitoring index details. The quality control finds that 30 minutes of night blood oxygen data is missing, interpolation is carried out through the average value of the front and rear time periods, and meanwhile, a record of sudden exacerbation of dyspnea reported by one time is marked, but no corresponding physiological index change is recorded, and further verification is needed. The forming of the structured dataset comprises a patient respiratory slowness image.
In a specific embodiment, the process of executing step S102 may specifically include the following steps:
integrating a respiratory disease clinical guideline, a evidence-based medical document library, an expert consensus file, a typical case set and a drug interaction database to construct a respiratory disease professional knowledge base;
Simultaneously inputting the structured patient data set and the medical text corpus in the respiratory system disease professional knowledge base into a transducer encoder structure, and performing multiple rounds of iterative training by adopting a mask language model and a next sentence prediction task to obtain initial large model parameters;
Constructing a contrast learning frame based on the initial large model parameters, pulling similar case expression vectors to be close and pushing different case expression vectors to be far, optimizing model parameters by minimizing contrast loss functions, and generating a respiratory slow disease large model;
Entity extraction is carried out on the structured patient data set and the respiratory system disease professional knowledge base, and diseases, symptoms, medicines, examination and risk factors are identified as medical concepts, so that a medical entity set is obtained;
Extracting the relationship from the medical entity set, and determining the relationship between the disease-symptom, the disease-drug, the drug-drug, the symptom-severity and the examination-reference value as the relationship between the entities to form a relationship triplet set;
The medical entity set and the relation triplet set are subjected to knowledge representation learning through a graph embedding algorithm, a knowledge graph of the slow breathing disease is constructed, and the knowledge graph is associated with a large model of the slow breathing disease through an entity linking technology, so that knowledge enhancement reasoning is supported.
Specifically, by integrating the clinical guidelines for authoritative respiratory diseases, the clinical guidelines for Chronic Obstructive Pulmonary Disease (COPD) issued by the Global control of respiratory diseases initiative (GOLD), the clinical guidelines for asthma issued by the Global control of asthma initiative (GINA), the clinical guidelines for interstitial lung diseases, and the like are included. These clinical guideline data are organized in the form of structured XML files, extracting core content such as disease definitions, diagnostic criteria, hierarchical classification methods, treatment recommendations, and management policies. The integration of evidence-based medical literature library covers high-quality respiratory disease research literature, and related random control tests, system evaluation and meta analysis published in the last five years are searched through a professional database, so that research findings, evidence-based conclusions and evidence grades are extracted. Expert consensus file integration comprises disease treatment consensus issued by the respiratory professional society of each country, and expert experience and clinical advice are extracted. The typical case set then collects representative cases of various types of respiratory slowness, including disease manifestations, diagnostic processes, treatment regimens, prognosis results, and the like. And integrating the drug interaction database to collect interaction information between the common drugs in the respiratory system and between the common drugs in other systems so as to form a drug interaction network. These multisource data are subject to text extraction, topic modeling and content indexing to construct a structured respiratory disease expertise repository. And inputting the medical text corpus in the structured patient data set and the respiratory disease professional knowledge base into a transducer encoder structure to develop model training. The transducer encoder is a neural network structure based on a self-attention mechanism and can capture long-distance dependency in text. The training process adopts two key tasks, namely a mask language model and a next sentence prediction task. In the masking language model task, 15% of the tokens in the input text are randomly replaced with special tokens, and then the training model predicts these masked words based on context. For example, the term "dyspnea" in "patient presented with severe signs and sputum" is masked, allowing model learning to infer reasonable medical terms from context. The next sentence prediction task provides two sentences to the model, and trains it to determine whether the second sentence is a natural continuation of the first sentence, thereby learning the consistency and logical relationship of the text. Through multiple rounds of iterative training of the two tasks, the model gradually learns the representation of medical language and the expertise of respiratory diseases to form initial large model parameters.
Based on the initial large model parameters, a model is further optimized by constructing a comparison learning framework. Contrast learning is a method of learning a representation by comparing similarities and differences between samples. In this scenario, clinically similar cases of slow breathing are taken as positive pairs of samples, and cases of different types or severity are taken as negative pairs of samples. Different expression variants of the case are generated through data enhancement technology, and the core medical content is kept unchanged but the expression forms are diversified. And extracting the characteristics of each case by using a large model to obtain a high-dimensional representation vector. And then, calculating cosine similarity among vectors, constructing a contrast loss function, and enabling the representing vector distance of similar cases to be reduced and the representing vector distance of different cases to be increased. And minimizing a contrast loss function through a gradient descent algorithm, and iteratively optimizing model parameters to generate a special large model sensitive to the respiratory slow disease. And performing entity extraction on the structured patient data set and the professional knowledge base, and identifying key medical concepts. By adopting a named entity recognition technology, medical entities such as diseases, symptoms, medicines, inspection, risk factors and the like in the text are recognized through a biomedical special dictionary and contextual semantic analysis. Each identified entity is assigned a unique identifier and a category label, forming a set of medical entities.
And then, extracting the relation of the medical entity set, and establishing semantic relation among the entities. Relationships between entities are extracted from text by dependency syntax analysis and semantic role labeling techniques. For example, the two disease-symptom relations of "asthma-manifesting as-wheezing" and "asthma-manifesting as-dyspnea" are extracted from the sentence "asthma often manifesting as wheezing and dyspnea", and the drug-drug relation of "salbutamol-synergistic-ipratropium bromide" is extracted from the sentence "salbutamol and ipratropium bromide combination can enhance bronchodilatory effect". In addition, the correlation of symptoms with severity and the correlation with reference values are also extracted and examined. Each extracted relationship forms a subject-relationship-object triplet structure, and is summarized to form a relationship triplet set. And converting the medical entity set and the relation triplet set 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 concepts with similar semantics are closer in the vector space. Common graph embedding algorithms include TransE, complEx and RotatE, etc., in which a modified TransE algorithm is employed that learns the vector representations of entities and relationships by optimizing the objective function of head entity vectors plus relationship vectors in triples near tail entity vectors. The generated vector representation reserves the semantic relation among the entities and forms a knowledge graph of the respiratory slow disease. Through the entity linking technology, a mapping relation is established between the medical concepts mentioned in the text and entity nodes in the knowledge graph, so that the fusion of the knowledge graph and the large model is realized, and the medical reasoning of knowledge enhancement is supported.
Taking the treatment of Chronic Obstructive Pulmonary Disease (COPD) as an example of the knowledge related to COPD, extraction of COPD from GOLD guidelines is a common, preventable and treatable disease characterized by the definition of persistent respiratory symptoms and restricted airflow, by which COPD is identified as a disease entity and persistent respiratory symptoms and restricted airflow as symptom entities. The relationship extraction stage identifies two triplets, "COPD-characterized by persistent respiratory symptoms" and "COPD-characterized by airflow limitation". Meanwhile, smoking is extracted from evidence-based literature and is a main risk factor of COPD, smoking is identified as a risk factor entity, and a smoking-is-risk factor of COPD triplet is formed. COPD stratification criteria were extracted from clinical guidelines, such as "FEV1/FVC <0.7 and FEV1<30% predictive value is severe COPD", forming triplets of "FEV1/FVC < 0.7-diagnostic criteria-COPD" and "FEV1<30% predictive value-severity assessment-severe COPD". The medicine treatment part extracts 'long-acting beta 2 receptor agonist (LABA) and long-acting anticholinergic agent (LAMA) which are combined to be applied to patients with severe symptoms of COPD', and forms 'LABA-combined drug-LAMA' and 'LABA+LAMA-treatment to be applied to the patients with severe symptoms of COPD', and the like. These entities and relationships are mapped to a vector space by a graph embedding algorithm, for example, "COPD" is mapped to a 128-dimensional vector, and clusters are formed in the vector space with related entities such as symptoms, drugs, etc., to construct a local graph structure of knowledge related to COPD. When new patient data is input into the system, the patient symptom of serious dyspnea is associated with the corresponding node in the knowledge graph through entity link, and the large model deduces possible diagnosis and proper treatment scheme according to the result, so that intelligent diagnosis and treatment decision with enhanced knowledge is realized.
In a specific embodiment, the process of executing step S103 may specifically include the following steps:
guiding a patient to describe current symptom experience through a natural language interaction interface, inputting natural language of the patient into a large model of the respiratory chronic disease, extracting symptom duration, severity, cause and relief factors as key symptom attributes, and generating a symptom entity matrix;
guiding the patient to complete dyspnea score, cough frequency score, sputum shape score, activity restriction degree score, sleep influence score and medication compliance score as a structured symptom evaluation questionnaire to obtain a questionnaire numerical vector;
Feature fusion is carried out on the symptom entity matrix and the questionnaire numerical vector, and a multidimensional feature space is constructed by combining the lung function index and the blood oxygen saturation numerical value in the basic physiological monitoring data to form the symptom characterization of the patient;
Assigning weight coefficients to dyspnea, cough and sputum in the symptom characterization of the patient as core symptoms, multiplying the severity values of the symptoms by the corresponding weight coefficients through a weighted calculation method and summing to obtain symptom burden indexes;
Carrying out time sequence analysis on recent symptom fluctuation trend, acute episode risk, medication compliance value and environmental factor influence, generating a disease stability index through a stability calculation model, and simultaneously calculating a comprehensive risk score by combining basic disease severity grading, complications and social support scores;
and setting a threshold interval according to the comprehensive risk score, dividing patients with risk scores lower than a first threshold into low risk levels, dividing patients with risk scores between the first threshold and a second threshold into medium risk levels, dividing patients with risk scores higher than the second threshold into high risk levels, and generating a patient risk stratification assessment result.
Specifically, the patient is guided to describe the current symptomatic experience through a natural language interactive interface. The interactive interface adopts a semi-structured query mode to present open questions such as "please describe your recent dyspnea condition", and to assist with guiding questions such as "how long this condition persists". The natural language description input by the patient is transmitted into the large model of the respiratory chronic disease, and key symptom attributes are identified through named entity identification and relation extraction technology. The symptom duration attribute extracts time expression from text, such as 'three days before' is converted into specific time length value, the severity attribute identifies words describing intensity, such as 'slight', 'obvious', 'severe' is converted into light, medium and heavy three-level scales, the inducement attribute extracts symptom trigger factors, such as 'after exercise', 'contact with pollen', 'inhale cool air', and the like, and the remission factor attribute captures conditions for improving symptoms, such as 'after rest', 'after use of inhaler', and the like. The extracted attribute values are organized into a structured symptom entity matrix, each row represents a symptom, each column represents an attribute dimension, and the elements in the matrix are corresponding attribute values. The patient is simultaneously guided to complete a structured symptom assessment questionnaire comprising a multi-dimensional scoring scale. The dyspnea score uses the modified medical research committee (mMRC) scale, from 0 score, indicating from dyspnea felt only during strenuous activities to dyspnea severe to failure to leave home or wear clothes, a cough frequency score from 0 score to 3 score, indicating from no cough to sustained cough affecting daily activities, a sputum-like score from 0 score to 3 score, indicating from no sputum to a large amount of purulent sputum, an activity restriction degree score from 0 score to 5 score, indicating from unrestricted to complete bed, a sleep impact score from 0 score to 5 score, indicating from normal sleep to complete inability to fall asleep due to respiratory symptoms, a medication compliance score from 0 score to 2 score, indicating from complete compliance to substantially non-prescribed medication. The patient performs scoring by selecting the options that best fit the situation, the scores comprising a questionnaire numerical vector, each element of the vector corresponding to a scoring dimension.
And carrying out feature fusion on the symptom entity matrix and the questionnaire numerical vector, and adopting feature splicing and cross coding methods. And the cross coding calculates the interaction between the two parts of features to generate interactive features. For example, when the dyspnea duration in the symptomatic entity matrix is "long-term" and the dyspnea score in the questionnaire is 4 timescales, the interactive features may enhance the representation of this severe condition. In combination with the basic physiological monitoring data, in the embodiment of the application, the basic physiological monitoring data comprises a lung function index and a blood oxygen saturation value, and the feature space is further expanded to obtain the multidimensional patient syndrome characterization.
And (5) carrying out weighted calculation on the core symptoms in the comprehensive symptom characterization of the patient to obtain a symptom burden index. Dyspnea is the index that best reflects the quality of life and severity of the disease, cough is the most common symptom, is closely related to social function and sleep quality, is given the next highest weight, sputum changes reflect airway inflammation and infection risk, and is given proper weight. In the weighted calculation process, the symptom scores are normalized to the interval of 0-1, then multiplied by the corresponding weight coefficients, and finally summed to obtain the symptom burden index. For example, normalized dyspnea score 0.75 multiplied by weight 0.5, cough score 0.6 multiplied by weight 0.3, sputum score 0.4 multiplied by weight 0.2, and the sum of the three is 0.75x0.5+0.6x0.3+0.4x0.2=0.62 as a symptom burden index.
Multiple time series data of the patient are analyzed to evaluate disease stability. The recent symptom fluctuation trend is calculated through the variation coefficient and the trend slope of continuous monitoring data, the variation coefficient is large or the slope is positive to indicate that symptoms are unstable, the acute attack risk is predicted based on the historical attack frequency, seasonal factors and the current symptom change and quantified to be a risk value between 0 and 1, the medication compliance value is obtained from electronic medicine box records and patient self-reporting data and is converted into a compliance percentage, and environmental factors influence weather conditions, air quality and personal sensitivity factors. The time sequence indexes are comprehensively calculated to generate a disease stability index, and the higher the numerical value is, the more stable the disease state is. Meanwhile, the overall risk score is comprehensively calculated according to the disease severity grade, the number and severity of the patient complications and the social support score of the medical guideline.
The threshold setting for risk score is formulated based on extensive clinical data analysis and expert consensus. By retrospectively analyzing the historical data of patients suffering from slow breathing, the relevance of different risk scores and adverse events is counted, two key demarcation points are determined, wherein the first threshold value is set to 40 scores, the historical data of patients below the first threshold value shows that the occurrence rate of the adverse events is lower than 5%, and the second threshold value is set to 65 scores, and the occurrence rate of the adverse events of the patients above the first threshold value exceeds 20%. Based on these two thresholds, patients are classified into three risk levels, low risk level (score < 40), suitable for primary medical institution management, only basic intervention required, medium risk level (scores 40-65), need enhanced monitoring and more aggressive treatment strategies, high risk level (score > 65), need specialty hospital intervention and close monitoring.
In a specific embodiment, the process of executing step S104 may specifically include the following steps:
inputting a patient risk stratification assessment result into a respiratory chronic disease large model, identifying disease classification of a patient, dividing the patient into four groups A/B/C/D according to the air flow restriction degree and symptom score when the patient is chronic obstructive pulmonary disease, and marking the patient based on the control level and the severity degree when the patient is asthma to generate a disease classification result;
Clinical guideline rule matching is carried out on the disease classification result, a mild medication scheme is searched when the patient belongs to the group A, a moderate medication scheme is searched when the patient belongs to the group B, and a severe medication scheme is searched when the patient belongs to the group C or the group D, so that a basic treatment framework is obtained;
comparing and analyzing the basic treatment framework with the past treatment record of the patient and the adverse reaction history of the medicine to generate a primary personalized medicine scheme;
Formulating administration information based on the primary personalized medicine scheme, including preferred medicine category combination, specific medicine selection, administration route, usage amount and course of treatment design, and generating a medicine treatment scheme;
creating non-drug treatment suggestions by combining a drug treatment scheme, wherein the non-drug treatment suggestions comprise a respiratory rehabilitation training program, a life style adjustment suggestion, mental health intervention measures and disease self-management skill training, and when the risk level of a patient is high risk, making an acute exacerbation emergency plan to form a personalized diagnosis and treatment scheme;
and carrying out matching evaluation on the personalized diagnosis and treatment scheme and the characteristics of the patient, calculating the past medication response matching score, the life habit matching score, the compliance factor matching score, the economic tolerance matching score and the medical insurance coverage matching score of the patient, and calculating to obtain a percentage matching degree score.
Specifically, the patient risk stratification assessment results are input into a large model of the respiratory chronic disease for disease classification. The large model analyzes the patient's underlying clinical data, symptoms and physiological indicators to identify specific disease types. For Chronic Obstructive Pulmonary Disease (COPD) patients, airflow limitation (expressed as a percentage of FEV1 predicted values) and symptom score (typically CAT score or mMRC dyspnea score) are taken as two key dimensions according to global chronic obstructive pulmonary disease control initiative (GOLD) guidelines. When FEV1 is more than or equal to 80 percent of predicted values, the predicted values are slightly limited, 50 percent less than or equal to FEV1 is less than 80 percent of predicted values are moderately limited, 30 percent less than or equal to FEV1 is less than 50 percent of predicted values are severely limited, and the predicted values of FEV1 is less than 30 percent of predicted values are extremely severely limited. In terms of symptom burden, symptoms are lighter when CAT is less than 10 minutes or mMRC is less than 2 minutes, and the symptoms are heavier when CAT is more than or equal to 10 minutes or mMRC is more than or equal to 2 minutes. COPD patients were divided into four groups a/B/C/D, group a (low symptoms, low risk), group B (high symptoms, low risk), group C (low symptoms, high risk) and group D (high symptoms, high risk), in combination with a history of acute exacerbations over the past year (0, 1 non-hospitalization exacerbations, >2 non-hospitalization exacerbations or >1 hospitalization exacerbations). When the patient is asthma, classification is based on symptom control levels and seizure risk assessment according to the global asthma control initiative (GINA) guidelines. Symptom control levels are divided into good control (no daytime symptoms or no more than 2 times per week, no nighttime symptoms, no limitation of activity, no need of rescue drugs or no more than 2 times per week), partial control (1-2 poorly controlled indicators exist) and no control (no less than 3 poorly controlled indicators exist). While considering the severity of asthma, it is classified as mild, moderate, severe and extremely severe, forming a standardized disease classification result.
And according to the disease classification result, the large model is searched and matched with corresponding clinical guideline rules to generate a basic treatment framework. For COPD patients, a mild regimen is retrieved, which mainly comprises the use of a short-acting bronchodilator as needed, a moderate regimen is retrieved, which recommends long-term regular use of a long-acting bronchodilator, when the patient belongs to group A, and a severe regimen is retrieved, which comprises the combined use of LABA and LAMA, when the patient belongs to group C or D, optionally with the addition of an inhaled glucocorticoid (ICS) or phosphodiesterase-4 inhibitor (PDE 4 i). For asthmatics, 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, severe patients consider the addition of leukotriene receptor antagonists or biologies. The related medicine indications, contraindications and recommended doses are searched through the knowledge graph to form a basic treatment framework aiming at specific disease classification.
The basic treatment framework was analyzed in comparison with the patient's past treatment record and history of adverse drug reactions. The drug use condition, the dose adjustment history and the treatment response record of the patient in the previous treatment record are extracted in a structured way, and the drug effectiveness is analyzed. At the same time, the history of adverse reactions is checked to confirm whether the patient has allergic reactions, intolerance or other adverse reactions to the particular drug or drug component. By setting the priority rule of drug selection, the priority of a certain type of drug is improved when the patient has good response to the drug, the drug is deleted or reduced from the recommended list when the patient has adverse reaction history to the drug, and the combination with the least adverse reaction and the optimal curative effect is selected when the patient has response to various drugs. Based on the preliminary personalized medicine scheme, detailed medicine administration information is formulated. The preferred combination of drug classes is determined, for example, a combination of lama+laba with a double bronchodilator in COPD patients and ics+laba in asthma patients. The particular drug is then selected from each class, taking into account the drug characteristics (e.g., onset of action, duration of action), the type of drug delivery device (e.g., metered dose inhaler, dry powder inhaler, nebulizer) and patient-ability. The administration route is preferably inhalation administration, so that the medicine is ensured to directly act on the respiratory tract, and the adverse reaction of the whole body is reduced. The dosage of the dosage is individually adjusted according to the standard dosage of the drug instruction and the weight, age and liver and kidney functions of the patient, and the accurate administration time point and frequency are formulated. The course design considers the different requirements of acute phase short-term intensive therapy and long-term maintenance therapy, including initial dose phase, stabilization phase and adjustment phase, and defines phase evaluation points and dose adjustment conditions.
On the basis of the medication regimen, non-medication recommendations are created. The respiratory rehabilitation training program comprises a respiratory control technology, a limb exercise prescription and a endurance training scheme, and is individually designed according to basic physical ability, activity endurance and complications of a patient. Lifestyle modification advice includes smoking cessation guidelines, environmental control measures, diet guidelines, and occupational protection measures. Psychological health intervention measures mainly aim at anxiety and depression problems common to patients with slow breathing, and provide cognitive behavioral intervention strategies and coping skills. Disease self-management skill training includes symptom identification, medication skill, method of inhalation device proper use, daily monitoring points, and the like. For patients with high risk level, an acute exacerbation emergency plan is additionally formulated, and early warning symptoms (such as obviously aggravated dyspnea, increased sputum volume and change of sputum color), self-treatment measures (such as adjusting medicine dosage and posture drainage) and medical treatment opportunity judgment standards are definitely made, so that the patients can timely cope with the change of illness state. These are integrated into a personalized regimen covering all-round guidelines for drug and non-drug intervention.
And evaluating the matching degree of the personalized diagnosis and treatment scheme and the characteristics of the patient. The matching degree evaluation comprises multiple dimensions, namely consistency of recommended medicines in the traditional medicine response matching score evaluation scheme and historical medicine response of a patient, suitability of life habit matching score evaluation treatment advice and daily activity modes and life laws of the patient, capability and willingness of the patient to execute a complex treatment scheme are evaluated by compliance factor matching scores, matching degree of economic tolerance matching score evaluation treatment expense and economic condition of the patient is evaluated by economic tolerance matching score, and medical insurance coverage matching score evaluation of medical insurance reimbursement conditions of recommended medicines and treatment projects. And giving different weights to each dimension, and calculating a weighted total score to obtain a percentage matching degree score. When the matching score is below a preset threshold (typically 60 minutes), the treatment protocol is readjusted until a satisfactory matching is achieved, ensuring that the protocol is both consistent with medical guidelines and viable.
In a specific embodiment, the process of executing step S105 may specifically include the following steps:
Distributing low-risk patients to basic medical institutions according to the patient risk layering evaluation results and matching degree scores, arranging primary medical institutions for management and periodic special hospital follow-up for stroke patients, leading and managing high-risk patients by the special hospitals and collaborative follow-up for the basic medical institutions, and constructing a hierarchical diagnosis and treatment path diagram;
The intelligent collaboration platform is connected with medical institutions of different levels, and data sharing is carried out on medical record information, examination results and diagnosis and treatment opinions to form a medical collaboration network;
For patients with slow breathing, which are diagnosed by basic medical institutions, the data of the patients and the primary judgment of doctors are input into a large model of the slow breathing, and symptom characteristics, disease progression risks and treatment difficulty are analyzed to generate diagnosis and treatment decision suggestions;
The method comprises the steps of performing multidimensional symptom monitoring on a patient starting treatment, recording core symptom score change, physiological index change, medication compliance data and life quality evaluation data, and constructing a full-dimension monitoring data set;
Performing time sequence analysis on the full-dimension monitoring data set, calculating symptom improvement rate, physiological index recovery rate, medication compliance index and life quality improvement amplitude, and generating a multi-dimension curative effect index;
Based on the multidimensional curative effect index, calculating comprehensive curative response scores through a weighted aggregation algorithm, judging curative effect trend according to the score change trend, and triggering a treatment scheme adjustment flow when the scores continuously drop more than three monitoring periods to form a time sequence treatment response index.
Specifically, reasonable split flow of patients is realized through the hierarchical diagnosis and treatment path diagram. According to the patient risk layering evaluation result and the diagnosis and treatment scheme matching degree score, a resource allocation decision matrix is established, low-risk patients (risk score < 40) are allocated to basic medical institutions, the community health service center or the village and town health hospital is responsible for conventional management, medium-risk patients (risk scores 40-65) are arranged to the basic medical institutions to conduct daily management, special hospitals follow-up is conducted every 3 months, two-way management is achieved, high-risk patients (risk scores > 65) are managed by the special hospitals in a respiratory department leading mode, and the basic medical institutions cooperatively conduct monthly follow-up to ensure that the state of illness changes are found timely. For patients with matching degree scores lower than 70, the first diagnosis is arranged in a special hospital no matter the risk level, and the special doctor makes an initial scheme and then transfers to a medical institution with corresponding level. The hierarchical diagnosis and treatment path diagram is constructed in a flow chart mode, the responsibility main body, the referral standard and the information flow rule of each node are defined, and the abstract hierarchical management principle is converted into an executable concrete flow. The data intercommunication and collaboration among the multi-stage medical institutions are realized through an intelligent collaboration platform. The platform adopts a federal learning architecture, and each medical institution reserves data for local storage and shares necessary information through a secure channel. The medical record sharing comprises disease diagnosis standardization, medication history structuring and treatment response quantification, so that the medical record is readable across institutions, the examination result sharing adopts a unified standard format, such as the conversion of a lung function detection value into an expected value percentage, the comparison of data among different examination rooms is realized, and the diagnosis and treatment opinion sharing records the judgment and suggestion of a specialist doctor and a basic doctor according to a standardized template. In the data sharing process, a differential privacy technology is adopted to protect sensitive information of patients, and meanwhile, a blockchain technology is used for recording data access and modifying logs, so that the whole process is guaranteed to be traceable. And constructing a cooperative network for connecting medical institutions of different levels through sharing data and an interconnection mechanism, so as to realize resource integration and diagnosis and treatment cooperation.
For patients with slow breathing who are diagnosed by basic medical institutions, basic doctors input basic data and preliminary judgment of the patients, wherein the basic data comprises complaint symptoms, physical examination findings, simple lung function measurement results and preliminary diagnosis comments. The data are input into a large model of the respiratory chronic disease after standardized processing, the large model is based on knowledge graph auxiliary analysis, symptom mode similarity is calculated, typical or atypical performance is identified, treatment difficulty is evaluated, and diagnosis and treatment decision advice is generated. Decision advice includes diagnostic validation or revision, exam advice, treatment adjustment advice, and referral instruction judgment. When the large model detects the complex situation that the basic medical institution is difficult to treat, such as serious complications, atypical symptom combination or poor treatment response, the large model automatically generates a referral application, including a key information abstract and a referral reason, and the referral application is pushed to a corresponding special hospital workstation.
And (3) carrying out multidimensional symptom monitoring on the patients suffering from the respiratory depression, which start treatment, and constructing a dynamic tracking system. The method comprises the steps of collecting core symptom score changes through a periodic questionnaire, including dyspnea mMRC score, cough score, sputum score and the like, filling a patient once a week, recording change trend, collecting physiological index changes through portable monitoring equipment, such as a lung function meter, measuring FEV1 value, a blood oxygen saturation monitor, recording night blood oxygen fluctuation, automatically uploading data to a management platform, recording actual medicine taking time and frequency through an intelligent medicine box according to medication compliance data, quantifying the actual medicine taking time and frequency into compliance percentage according to self-report conditions of the patient, and periodically evaluating life quality evaluation data of the patient by adopting a respiratory system disease specificity scale such as CAT or AQLQ. The multi-source data is stored into the electronic health file of the patient after being processed by the format unification and the missing value, and a time sequence data set is formed, so that a full-dimension monitoring data set is formed.
And (3) carrying out time sequence analysis on the full-dimension monitoring data set, and calculating a plurality of curative effect indexes. The symptom improvement rate calculation is based on the change of the core symptom score, the difference between the scores before and after treatment is divided by the initial score to obtain a standardized improvement percentage, the physiological index recovery rate calculation is used for calculating the recovery degree of key physiological parameters such as FEV1 relative to an expected normal value and tracking a change curve of the key physiological parameters, the medication compliance index is used for calculating a comprehensive compliance index by calculating the ratio of the actual medication times to the prescribed times of medical advice and combining the medication time accuracy score, and the life quality improvement amplitude is calculated by the difference before and after the score of a standardized scale. A threshold is set for each indicator, e.g., symptom improvement >30% is a significant improvement, 15-30% is a moderate improvement, and <15% is a slight improvement. These calculations form a structured multidimensional therapeutic index that visually reflects various aspects of the therapeutic effect.
Based on the multidimensional curative effect index, the comprehensive treatment response score is calculated through a weighted aggregation algorithm. The weights of the indexes are determined, and the indexes of the lung function and the dyspnea score are higher in the patients with COPD, and the symptom control rate and the acute episode frequency are higher in the patients with asthma. And normalizing the indexes to be between 0 and 100 partitions, and obtaining the comprehensive score of the treatment response according to the weighted summation of the weights. The score is calculated once per cycle (typically 2 weeks) to form time series data. The trend of the change in the response to treatment was assessed by time series analysis methods, including moving average, trend analysis, and seasonal decomposition. Setting a monitoring and early warning rule, namely, when the score continuously rises to indicate good treatment effect, the score is stable to indicate stable illness state, the score fluctuation is obvious to indicate unstable control, and the score continuously falls for more than three monitoring periods to trigger the treatment scheme adjustment flow. The calculated comprehensive score and the change rate thereof are stored in a time sequence database of the patient each time to form a time sequence treatment response index.
In a specific embodiment, the process of executing step S106 may specifically include the following steps:
Based on disease types, severity, cognitive level, technical acceptance and curative effect indexes of patients, content screening is carried out on disease basic knowledge, symptom identification methods, medication guidance content, life style adjustment strategies and acute exacerbation preventive measures, and a personalized health education content library is generated;
Converting the personalized health education content library into multimedia resources in the forms of words, pictures, videos and interactive questions and answers, dynamically adjusting the presentation depth according to the learning progress and the content understanding degree of the patient, and constructing a hierarchical education resource package;
According to the personalized diagnosis and treatment scheme and the treatment response index, an electronic medicine box reminding system, a symptom self-monitoring diary, a respiratory training guidance system and an environmental risk early warning system are developed to serve as functional modules, so that a patient self-management tool set is formed;
Integrating health education content and a self-management tool set into mobile application, a webpage and an intelligent wearable device multichannel access port according to the technical acceptance of a patient, and generating a patient-side interactive interface;
Recording and analyzing operation behaviors, feedback comments, diagnosis and treatment effect evaluation and large model performance indexes of patients and medical staff in the using process of the system to form a system optimization data set;
Based on the system optimization data set, the knowledge base update and algorithm optimization are carried out on the respiratory chronic disease large model, the efficiency analysis and resource allocation optimization are carried out on the system flow nodes, the man-machine interaction optimization is carried out on the user interface and the interaction logic, and the system iteration and flow optimization are completed.
Specifically, personalized health education content screening is performed through patient characteristic parameter analysis, wherein patient disease types are classified into chronic obstructive pulmonary disease, asthma, bronchiectasis and other respiratory chronic disease types, severity is classified according to clinical classification standards, such as GOLD classification of COPD or control level classification of asthma, cognitive level is determined to be primary, medium or high through simple knowledge assessment, technical acceptance is assessed to be low, medium or high through technical use habit questionnaires, and previously calculated curative effect index data are integrated. Based on the five-dimensional feature vectors, relevant contents are extracted from a respiratory disease professional knowledge base, the contents of etiology mechanisms, natural course of disease, prognosis factors and the like related to specific disease types of patients are selected according to basic knowledge of the diseases, early warning symptom recognition points and self-evaluation tools suitable for the severity degree of the patients are extracted according to a symptom recognition method, medication guidance contents are directly related to personalized diagnosis and treatment schemes of the patients, the medication guidance contents comprise action mechanisms, correct use methods, notes and the like of the used medicines, lifestyle adjustment strategies select proper exercise schemes, diet suggestions and environmental control measures based on individual conditions of the patients, and acute exacerbation prevention measures are customized according to past attack modes and risk factors of the patients. And matching the content module with the characteristic vector of the patient through a similarity matching algorithm, and screening the content combination which is most suitable for the patient to form a personalized health education content library. The personalized health education content library is converted into diversified multimedia resources, so that the acceptability and interactivity of the content are realized. The text data adopts hierarchical reading difficulty control, sets word complexity and sentence structure according to the cognition level of a patient and is matched with professional terms for explanation, the picture resources comprise medical insets, anatomic drawings and operation schematic diagrams, abstract concepts are visually presented in a visual mode, video resources are focused on dynamic skill display, such as inhaler use methods, respiratory exercise actions and body position drainage technologies, interactive question and answer design knowledge detection questions of the hierarchical difficulty, and the understanding degree of the patient on the learned contents is estimated. All resources are divided into three levels of basic level content, advanced level content and professional level according to the complexity of the content, the basic level content is initially presented, the system records learning behavior data of a patient, the learning behavior data comprise indexes such as reading time, video completion rate, question-answer accuracy rate and the like, learning effects are analyzed through a decision tree algorithm, the advanced level content is automatically unlocked when the question-answer accuracy rate of the basic level content exceeds 80%, the professional level content is opened again after the advanced level content test is passed, dynamic adjustment of education depth is achieved, and a hierarchical education resource package is constructed.
According to the personalized diagnosis and treatment scheme and the treatment response index, a self-management tool set with complementary functions is developed. The electronic medicine box reminding system sets accurate medicine taking time points and dose reminding according to a medicine taking scheme of a patient, records actual medicine use conditions, calculates medicine taking compliance indexes, triggers an intelligent reminding upgrading mechanism when medicine is missed or misplaced for three times continuously, and ensures medicine taking regularity through modes of stronger prompt tone, relatives and the like. The symptom self-monitoring diary is designed into a simple daily recording interface, so that a patient is guided to evaluate the core symptom change regularly, a symptom trend chart is automatically generated, and abnormal fluctuation is identified. The respiratory training guidance system provides personalized respiratory muscle training plans and whole-body aerobic exercise suggestions based on the pulmonary function state and physical energy level of the patient, evaluates the training effect by combining sensor feedback, and dynamically adjusts the difficulty. The environmental risk early warning system integrates real-time meteorological data, air quality indexes and personal sensitive factors of patients, predicts high risk exposure conditions, gives early warning 12-24 hours in advance, and provides corresponding protective measures suggestions.
According to the technical acceptance of patients, health education content and self-management tools are integrated into a plurality of technical platforms, so that seamless access is realized. The technical acceptance is determined by questionnaire evaluation in combination with age group characteristics and actual equipment usage habits and is classified into three classes of low, medium and high. The method comprises the steps of providing a mobile application of functions for patients with higher technical acceptance, including all educational resources and management tools, supporting complex interaction and data visualization, providing a simplified mobile application or a webpage portal for patients with medium technical acceptance, highlighting core functions, simplifying operation flow, and performing basic monitoring and auxiliary management by matching paper materials and telephone follow-up for patients with lower technical acceptance mainly through intelligent wearing equipment. The data among the platforms are synchronously shared, so that the patient can acquire consistent health management services at different entrances. The design of the interactive interface follows the principle of intuitiveness, adopts icons with large fonts, high contrast colors and conforming to cognitive habits, reduces the use threshold and improves the acceptance of patients.
And carrying out omnibearing data acquisition on the actual running condition of the system to form the basis of iterative optimization. The operation behavior data of the patient comprises the use frequency, the function access path, the stay time and the interaction depth, the operation behavior data of the medical staff is recorded in real time through the embedded point technology, and the operation behavior data of the medical staff is focused on the use condition and the adjustment frequency of the clinical decision support function. Feedback ideas are collected through structured questionnaires and free text forms, covering system ease of use, content utility, and functional integrity assessment. The diagnosis and treatment effect evaluation data integrates the disease control index change, the acute exacerbation frequency change and the life quality score change, and the system intervention effect is quantified. The large model performance index records the accuracy of diagnosis advice, the rationality of treatment plan and the satisfaction of patient question answers, and evaluates the performance of the artificial intelligence core component. The multidimensional data are subjected to cleaning, standardization and association analysis to form a structured system optimization data set, and a basis is provided for the next iteration.
And carrying out multi-level iterative updating based on the system optimization data set. The knowledge base of the large model of the respiratory chronic disease is updated by an incremental learning mode, new clinical guidelines, research findings and typical cases are added to the training corpus to expand the knowledge surface of the model, the model is identified by error analysis when the algorithm is optimized, the weight of the samples in the training set is pertinently strengthened, and the accuracy of the samples in a specific scene is improved. The system flow node optimization is based on user behavior data, identifies links with longer operation time consumption or higher error rate, simplifies operation through flow reconstruction or auxiliary prompt, and adjusts and calculates resource allocation proportion according to the use frequency and importance of each functional module to improve the response speed of the system. The user interface and the interaction logic are optimized by adopting an A/B test method, and the user acceptability and the operation efficiency of different design schemes are compared, so that the optimal solution scheme is selected. The whole iterative process is performed according to a closed loop of data collection, problem analysis, optimization implementation and effect evaluation, so that the continuous evolution of the system is ensured, and the continuous evolution system is adapted to the continuously changing clinical requirements and technical environments.
The foregoing describes a method for intelligent diagnosis and treatment of respiratory depression based on a large model in the embodiment of the present application, and the following describes a system for intelligent diagnosis and treatment of respiratory depression based on a large model in the embodiment of the present application, referring to fig. 2, and one embodiment of the system for intelligent diagnosis and treatment of respiratory depression based on a large model in the embodiment of the present application includes:
The acquisition module 201 is used for acquiring and processing multidimensional data of a patient suffering from slow breathing to obtain a structured patient data set;
the input module 202 is configured to input the structured patient data set and the respiratory disease professional knowledge base into a dual coding pre-training and contrast learning fine tuning framework to obtain a respiratory disease large model and a respiratory disease knowledge map;
The calculation module 203 is configured to perform semantic analysis and multidimensional calculation on the patient symptom description and the evaluation scale answer based on the respiratory chronic disease large model and the respiratory chronic disease knowledge graph, and generate a patient risk stratification evaluation result;
The generating module 204 is configured to retrieve clinical guideline rules by using the large respiratory slowness model and combine individual characteristics of the patient according to the risk stratification evaluation result of the patient to generate a personalized diagnosis and treatment scheme and a matching degree score;
The construction module 205 is configured to construct a multi-level medical institution collaboration frame and perform dynamic efficacy monitoring according to the personalized diagnosis and treatment scheme and the matching degree score, so as to generate an efficacy index and a treatment response index;
The customizing module 206 is configured to customize the health education content and the self-management tool of the patient according to the therapeutic effect index and the therapeutic response index, and collect the system operation data to perform model updating and flow optimization.
The method comprises the steps of forming a structured patient data set through the collaborative cooperation of the components, realizing comprehensive acquisition and standardization of information related to the respiratory chronic disease through multi-dimensional data acquisition and processing, secondly, adopting a double coding pre-training and contrast learning fine tuning framework to process the structured patient data set and a respiratory system disease professional knowledge base, generating a respiratory chronic disease large model and a respiratory chronic disease knowledge map with deep medical understanding capability and professional reasoning capability, which are remarkably superior to the traditional machine learning model, capable of understanding complex medical contexts and extracting key clinical features from the complex medical context, thirdly, carrying out semantic analysis and multi-dimensional calculation on the patient symptom description and evaluation scale answers based on the respiratory chronic disease large model and the knowledge map, generating a patient risk layering evaluation result with high accuracy, capturing subtle disease feature differences, realizing accurate layering, optimizing medical resource allocation, fourthly, fully considering individual difference of a patient by utilizing a respiratory chronic disease large model to retrieve clinical guideline rules and combining personalized scheme and matching degree score generated by individual feature of the patient, improving the compliance of the medical guideline, and continuously optimizing and the medical diagnosis and treatment system according to the continuous response of the medical guideline, and the medical diagnosis and treatment system, and the method is realized by continuously optimizing and the medical system according to the continuous response of the personalized guideline of the medical system and the personalized examination and the medical system, a closed-loop quality improvement system is formed, so that the system is continuously evolved and perfected. Particularly, the large model algorithm applied in the scheme has great contribution to the diagnosis and treatment management field of the respiratory chronic diseases, the context understanding capability enables the model to extract key clinical information from unstructured medical texts, the knowledge enhancement reasoning mechanism realizes automatic support of complex medical decisions, the multi-mode data fusion algorithm can integrate patient information from different sources to form a comprehensive view, the adaptive learning characteristic enables the system to continuously optimize diagnosis and treatment strategies according to new data, and the specific algorithm features jointly construct a highly specialized intelligent auxiliary decision-making system for the respiratory system diseases, so that the efficiency and the accuracy of the management of the respiratory chronic diseases are greatly improved.
The embodiment of the invention of the intelligent diagnosis and treatment management system for the respiratory depression based on the large model is described in detail from the angle of modularized functional entities in fig. 2, and the embodiment of the invention of the intelligent diagnosis and treatment management system for the respiratory depression based on the large model is described in detail from the angle of hardware processing.
Fig. 3 is a schematic structural diagram of a large model-based breathing slow disease intelligent diagnosis and treatment device according to an embodiment of the present invention, where the large model-based breathing slow disease intelligent diagnosis and treatment device 300 may generate relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (e.g., one or more processors) and a memory 320, one or more storage mediums 330 (e.g., one or more mass storage devices) storing an application 333 or data 332. Wherein memory 320 and storage medium 330 may be transitory or persistent storage. The program stored on the storage medium 330 may include one or more modules (not shown), each of which may include a series of instruction operations in the large model-based respiratory slowness intelligent diagnostic management device 300. Still further, the processor 310 may 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 distress intelligent diagnosis and treatment management apparatus 300 to implement the steps of the large model-based respiratory distress intelligent diagnosis and treatment management method described above.
The large model-based respiratory slowness intelligent diagnostic management device 300 may 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. It will be appreciated by those skilled in the art that the large model-based breathing chronic disease intelligent diagnosis and treatment management apparatus structure shown in fig. 3 does not constitute a limitation on the large model-based breathing chronic disease intelligent diagnosis and treatment management apparatus provided by the present invention, and may include more or fewer components than illustrated, or may combine certain components, or may be arranged in different components.
The invention also provides a computer readable storage medium, which can be a nonvolatile computer readable storage medium, and can also be a volatile computer readable storage medium, wherein instructions are stored in the computer readable storage medium, and when the instructions run on a computer, the instructions cause the computer to execute the steps of the intelligent diagnosis and treatment management method based on the large model.
It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, systems and units may refer to the corresponding processes in the foregoing method embodiments, which are not repeated herein.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or partly in the form of a software product, or all or part of the technical solution, stored in a storage medium, comprising instructions for causing a large model-based intelligent diagnosis and treatment device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. The storage medium includes a U disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, an optical disk, or other various media capable of storing program codes.
The foregoing embodiments are merely for illustrating the technical solution of the present invention, but not for limiting the same, and although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that modifications may be made to the technical solution described in the foregoing embodiments or equivalents may be substituted for parts of the technical features thereof, and that such modifications or substitutions do not depart from the spirit and scope of the technical solution of the embodiments of the present invention in essence.