Prediction system for chronic obstructive pulmonary disease and storage medium

Abstract

The invention discloses a chronic obstructive pulmonary disease prediction system and a storage medium. The chronic obstructive pulmonary disease prediction system comprises an information collection module, a science popularization module, an illness state analysis module and an illness state diagnosis database, wherein the information collection module is used for collecting patient information; the science popularization module comprises a digital human model and is used for playing a science popularization video and interacting with a patient through the digital human model to generate interaction information; the illness state analysis module comprises a data arrangement module, a large model illness state diagnosis module and a self-adaptive reflection reasoning module; the data arrangement module is used for extracting a first knowledge point and a patient symptom according to the patient information, the science popularization video and the interaction information; the large model illness state diagnosis module comprises an illness state diagnosis model, the large model illness state diagnosis module generates a prediction thought, and the illness state diagnosis model obtains an illness state diagnosis result according to the patient information, the first knowledge point, the patient symptom and the prediction thought; and the self-adaptive reflection reasoning module comprises a similar medical record retrieval model, a disease development analysis model and a medical suggestion model.

Description

Technical Field

[0001] The present application relates to the field of medical artificial intelligence technology, and more specifically to a prediction system and storage medium for chronic obstructive pulmonary disease. Background Art

[0002] With the development of my country's social economy, the continuous acceleration of urbanization, and the changes in residents' living behaviors / styles, the prevalence of chronic non-communicable diseases (referred to as chronic diseases) represented by cancer and hypertension has increased year by year. Chronic diseases are seriously threatening the health of Chinese residents, including chronic obstructive pulmonary disease. According to the characteristics of chronic obstructive pulmonary disease, this is a chronic disease that requires long-term management, and self-management plays a vital role in chronic disease management. Traditional chronic disease management models mainly include biomedical management models, cognitive behavioral intervention models, and psychodynamic intervention models. They are effective models formed through active exploration in various fields of medicine for a long time. However, the traditional chronic disease management model still has certain shortcomings.

[0003] Individuals have weak self-management abilities, lack of knowledge about chronic diseases, and poor compliance, which weakens the effectiveness of health interventions. my country's public health knowledge penetration rate is low, education levels are low, and the awareness rate of chronic obstructive pulmonary disease is very low, especially among primary care patients. This not only causes patients to seek medical treatment only when symptoms are very severe or even reach an acute exacerbation state, which seriously affects prognosis, but also leads to poor compliance and inability to conduct effective long-term disease management.

[0004] Most prevention and management strategies are universal, and the differences in individual needs are not fully considered during formulation and implementation. The current shortage of medical resources, especially the shortage of doctors, nurses and other personnel who directly provide medical services, has always been a major problem that plagues the improvement of medical services. How to effectively utilize medical resources so that medical workers can release their energy from simple repetitive work has become one of the key tasks of medical information workers. Summary of the invention

[0005] This application is proposed to solve the above problems. According to one aspect of this application, a prediction system for chronic obstructive pulmonary disease is provided, including an information collection module, a popular science module, a disease analysis module and a disease diagnosis database: The information collection module is used to collect patient information; The science popularization module includes a digital human model, which is used to play science popularization videos and interact with patients to generate interactive information; The disease analysis module includes a data sorting module, a large model disease diagnosis module and an adaptive reflective reasoning module; The data sorting module is used to extract the first knowledge point and patient symptoms based on patient information, popular science videos and interactive information; The large model disease diagnosis module includes a disease diagnosis model. The large model disease diagnosis module generates a prediction idea, and the disease diagnosis model obtains a disease diagnosis result based on the patient information, the first knowledge point, the patient's symptoms, and the prediction idea. The adaptive reflection and reasoning module includes a similar medical record retrieval model, a disease development analysis model, and a medical advice model. The similar medical record retrieval model is used to retrieve a specified medical record according to the patient information, the patient's symptoms, and the disease diagnosis result. The disease development analysis model is used to propose an improvement method based on the disease diagnosis result and the historical diagnosis data in the disease diagnosis database. The medical advice model gives medical advice by combining the specified medical record and the improvement method.

[0006] In an embodiment of the present application, the large model disease diagnosis module further includes a reflection model and an optimization model. During the training process of the large model disease diagnosis module, the disease diagnosis model is trained through the reflection model and the optimization model: During the training process, the disease diagnosis model diagnoses the sample according to the patient information, the first knowledge point, and the patient's symptoms to obtain an incorrect sample diagnosis result and an incorrect idea. The reflection model analyzes the cause of the error based on the incorrect sample diagnosis result, the corresponding sample, and the incorrect idea, and proposes an improvement suggestion for the incorrect idea. The optimization model generates a prediction idea based on the cause of the error and the improvement suggestion for the incorrect idea.

[0007] In an embodiment of the present application, the similar medical record retrieval model includes an information extraction module, a feature enhancement module, and a time series optimization module: The hospital medical record library includes multiple hospital patient medical records. The information extraction module extracts the patient information, the patient's symptoms, and the disease diagnosis result from the hospital patient medical records. The feature enhancement module is used to enhance the features of the patient information, the patient's symptoms, the disease diagnosis result, the patient information, the patient's symptoms, and the disease diagnosis result in the hospital patient medical records, so as to better retrieve the specified medical record. The time series optimization module is used to optimize the features of the patient information, the patient's symptoms, and the disease diagnosis result after feature enhancement according to the historical diagnosis data, and optimize the patient information, the patient's symptoms, and the disease diagnosis result according to the time series data in the historical diagnosis data, so as to retrieve the matching specified medical record.

[0008] In an embodiment of the present application, the time series optimization module is used to optimize the features of the patient information, the patient's symptoms, and the disease diagnosis result after feature enhancement according to the historical diagnosis data, including: Encode and enhance the features of the patient information, patient symptoms, and disease diagnosis results in each historical diagnosis data with those in the current diagnosis to obtain the corresponding feature vectors , , , where refers to the feature vector of the th medical record, is the sequence length. Among them, , , are the feature vectors corresponding to the patient information, patient symptoms, and disease diagnosis results in the current diagnosis. Treat the medical record sequence in the historical diagnosis data as a time series and input it into a long short-term memory network for processing. The long short-term memory network updates the hidden state through the following formula: ; where is the hidden state at time step, LSTM is the long short-term memory network, is the hidden state of the last time step of the sequence, representing the development trend feature of the entire medical record sequence. Use this development trend feature to optimize the feature vector of the current diagnosis. The formula is as follows: ; where is the weight matrix, is the bias, represents vector concatenation, is the optimized feature vector. The other two features and similarly fuse the development trend of the disease condition to obtain the optimized feature vector .

[0009] In an embodiment of the present application, the similar medical record retrieval model is used to retrieve a specified medical record according to the similarity between the patient information, patient symptoms, and disease diagnosis results with optimized features and the patient information, patient symptoms, and disease diagnosis results in the hospital patient medical records with enhanced features, including: Calculate the fusion similarity between the patient information, patient symptoms, and disease diagnosis results with optimized features and the patient information, patient symptoms, and disease diagnosis results in the hospital patient medical records with enhanced features : + ; is the representation vector of patient information, is the representation vector of patient symptoms, is the representation vector of disease diagnosis results; is the representation vector of patient information in the medical records of hospital patients, is the representation vector of patient symptoms in the medical records of hospital patients, is the representation vector of the disease diagnosis result in the medical records of hospital patients; is the patient information weight, is the patient symptom weight, is the weight of the disease diagnosis result ; is a similarity function, a function used to calculate the similarity between two feature vectors; According to the specified fusion similarity , the specified medical record is retrieved.

[0010] In an embodiment of the present application, the science popularization module further includes a knowledge base, a key information extraction model, and a knowledge point extraction model: The knowledge base stores medical knowledge points; The key information extraction model is used to extract the patient's current symptoms, patient questions, and the main content of the video according to the patient information, interaction information, historical diagnosis data, and science popularization video; The knowledge point extraction model is used to extract the second knowledge point from the knowledge base according to the patient's current symptoms, patient questions, and the main content of the video; The digital human model determines the interaction direction according to the patient information, interaction information, historical diagnosis data, and the second knowledge point. The interaction direction includes two interaction directions. The first interaction direction is to recommend that the patient watch the science popularization video at the specified position, and the second interaction direction is to ask the patient relevant questions.

[0011] In an embodiment of the present application, the patient symptoms, disease diagnosis results, and medical suggestions obtained by the disease analysis module are stored in the disease diagnosis database.

[0012] In an embodiment of the present application, the disease development analysis model determines whether the expected treatment goal is achieved according to the disease diagnosis result and the historical diagnosis data in the disease diagnosis database; If the expected treatment goal is not achieved, improvement methods are proposed according to the expected treatment goal.

[0013] In an embodiment of the present application, the medical advice model obtains preliminary medical advice according to the disease diagnosis result, patient information, the first knowledge point, patient symptoms, and prediction ideas; The medical advice model optimizes the preliminary medical advice according to the medical advice of the specified medical record to obtain medical advice.

[0014] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored. When the computer program is run by a processor, the processor is caused to execute the above-mentioned prediction system for chronic obstructive pulmonary disease.

[0015] The prediction system for chronic obstructive pulmonary disease of the present invention interacts with patients through a science popularization module, strengthens patients' self-management, and improves users' awareness of the disease. Through the disease analysis and diagnosis module, in combination with the interaction information of the science popularization module, in-depth analysis and accurate diagnosis of patients are carried out, a disease diagnosis result is given, and rigorous and personalized medical advice is provided to help users conduct effective disease management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 FIG. shows a system block diagram of a prediction system for chronic obstructive pulmonary disease according to an embodiment of the present application; Figure 2 FIG. shows a training framework diagram of a large model disease diagnosis module of a prediction system for chronic obstructive pulmonary disease according to an embodiment of the present application; Figure 3 FIG. shows an adaptive reflection reasoning module diagram of a prediction system for chronic obstructive pulmonary disease according to an embodiment of the present application; Figure 4 FIG. shows a digital human interaction framework diagram of a prediction system for chronic obstructive pulmonary disease according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] In order to make the objectives, technical solutions, and advantages of the present application more obvious, exemplary embodiments according to the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein. Based on the embodiments of the present application described herein, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application.

[0019] First, reference is made to Figures 1-4 to describe the prediction system for chronic obstructive pulmonary disease for implementing the embodiments of the present invention. Figure 1 FIG. shows a system block diagram of a prediction system 100 for chronic obstructive pulmonary disease according to an embodiment of the present application. AsFigure 1 As shown in Figure 1 , the prediction system 100 for chronic obstructive pulmonary disease according to an embodiment of the present application includes an information collection module 110, a science popularization module 120, a disease condition analysis module 130, and a disease condition diagnosis database 140.

[0020] The information collection module 110 is used to collect patient information. When the user first enters the system, they need to fill in their personal basic information (gender, age, past medical history, etc.), and the information collection module 110 of the system will record this information in detail for subsequent use.

[0021] The science popularization module 120 includes a digital human model, which is used to play science popularization videos and interact with the user through the digital human model to generate interaction information. The digital human model is used to guide the user to conduct personalized interactions, provide detailed science popularization to the user in combination with the science popularization videos, and at the same time track the development of the user's disease symptoms. Specifically, the user can freely watch the existing science popularization videos or directly ask questions and get answers from the digital human.

[0022] The disease condition analysis module 130 includes a data sorting module, a large model disease condition diagnosis module, and an adaptive reflection and reasoning module. In the disease condition analysis module 130, the system will analyze all the interaction data obtained from the science popularization module 120, and then diagnose the user's disease condition and provide medical advice to the user. The user can enter this module after the video viewing ends or actively, or automatically enter this module when preparing to exit the system.

[0023] The data sorting module is used to extract the first knowledge points and patient symptoms based on the patient information, science popularization videos, and interaction information. The large model disease condition diagnosis module includes a disease condition diagnosis model. The large model disease condition diagnosis module generates a prediction idea, and the disease condition diagnosis model obtains a disease condition diagnosis result based on the patient information, the first knowledge points, the patient symptoms, and the prediction idea; the adaptive reflection and reasoning module includes a similar medical record retrieval model, a disease development analysis model, and a medical advice model; the similar medical record retrieval model is used to retrieve a specified medical record based on the patient information, the patient symptoms, and the disease condition diagnosis result; the disease development analysis model is used to propose improvement methods based on the disease condition diagnosis result and the historical diagnosis data in the disease condition diagnosis database 140; the medical advice model gives medical advice by combining the specified medical record and the improvement methods.

[0024] Specifically, the system accurately diagnoses the user's disease condition through the trained large model disease condition diagnosis module. Then, in the adaptive reflection and reasoning module, the system can refer to the user's treatment effect and the treatment plan of similar medical records in detail through the disease development analysis model and the similar medical record retrieval model, and finally give rigorous and personalized medical advice. Finally, these analysis results are saved in the disease condition diagnosis database 140 to provide personalized health advice and guidance for the user.

[0025] The prediction system for chronic obstructive pulmonary disease of the present invention interacts with patients through the science popularization module 120, strengthens patients' self-management, and improves users' awareness of the disease. Through the disease analysis and diagnosis module, in combination with the interaction information of the science popularization module 120, in-depth analysis and accurate diagnosis of patients are carried out, the disease diagnosis results are given, and rigorous and personalized medical advice is provided to help users effectively manage the disease.

[0026] The interaction information in the science popularization module 120 includes information such as users' questions, symptoms, and points of interest. All these interaction information will be uniformly sorted out in the disease analysis module 130 to obtain more detailed disease diagnosis results, and the digital human will also put forward targeted medical advice. First, the interaction information is standardized and sorted out by the data sorting module to obtain the patient's symptoms and the first knowledge points. Then these data are sent into the large model disease diagnosis module and the adaptive reflection and reasoning module, and then accurate disease diagnosis results and rigorous medical advice are obtained.

[0027] In the data sorting part, the data sorting module sorts out all the interaction information to obtain the patient's symptoms and the first knowledge points. The corresponding prompt words for this part are as follows: Prompt = "You are an information sorting expert, responsible for analyzing and extracting the patient's health information from the patient interaction data. Please perform the following tasks: Analyze <interaction information, patient information, and science popularization videos>, identify and sort out the patient's current symptoms and the first knowledge points. Please ensure the accuracy and relevance of the information to support subsequent diagnosis and treatment suggestions."

[0028] In order to improve the accuracy of disease diagnosis and make it closer to the diagnosis level of professional doctors, a large model disease diagnosis module and an adaptive reflection and reasoning module are designed. First, in the large model disease diagnosis module, professional medical record data is used as the training set to construct a collaborative framework for constructing a disease diagnosis model, a reflection model, and an optimization model to train the disease diagnosis model to improve the diagnosis level. Then, in the system application, the trained disease diagnosis model diagnoses the patient's condition, and the disease diagnosis results will be sent into an adaptive reflection and reasoning module. The adaptive reflection and reasoning module can give more rigorous and personalized medical advice by observing the treatment situation and referring to similar medical records.

[0029] In the embodiment of the present application, the large model disease diagnosis module further includes a reflection model and an optimization model. During the training process of the large model disease diagnosis module, the disease diagnosis model is trained through the reflection model and the optimization model: during the training process, the disease diagnosis model diagnoses the sample according to the patient information, the first knowledge point, and the patient's symptoms to obtain the wrong sample diagnosis result and the wrong thinking; the reflection model analyzes the reasons for the error according to the wrong sample diagnosis result, the corresponding sample, and the wrong thinking, and puts forward improvement suggestions for the wrong thinking; the optimization model generates the prediction thinking according to the reasons for the error and the improvement suggestions for the wrong thinking.

[0030] As Figure 2 shown, the details of the training framework for the large model disease diagnosis module are as follows: Before training, the disease diagnosis model only uses <patient information, first knowledge point, patient symptoms> as the basis for disease diagnosis. The large model itself is not rigorous enough in medical diagnosis, and there will be errors when diagnosing diseases only based on this information. Therefore, an information, prediction idea, is introduced. What is recorded in the prediction idea is the idea by which the disease diagnosis model can correctly diagnose diseases. Before training, the prediction idea is empty, and then during the training process, the reflection model and the optimization model continuously optimize the prediction idea. The prompt of the disease diagnosis model is as follows: Prompt = "You are a professional medical diagnosis expert for chronic obstructive pulmonary disease, responsible for analyzing the patient's disease information. Please make a diagnosis based on the following information: <patient information, first knowledge point, patient symptoms>. And, please refer to <prediction idea> to provide an accurate diagnosis result. The diagnosis result is a certain disease and its probability."

[0031] The disease diagnosis result is generally a certain disease and the probability of that disease. Here, is used as the confidence level of the diagnosis result of the sample , where is the probability of the diagnosed disease in the correct result. Here, a threshold is set to measure whether the diagnosis result is qualified. In the case of , it is considered a misdiagnosis; for a misdiagnosis, record the reasoning process based on which the model makes the diagnosis during the diagnosis, including the factors considered, the knowledge points applied, and the logical reasoning steps. This reasoning process is the wrong idea. During the training process, when the diagnosis result is incorrect, the prediction idea, patient information, first knowledge point, patient symptoms, misdiagnosed sample diagnosis result, and wrong idea are sent into the reflection model, and the reflection model analyzes the reasons for the misdiagnosis and puts forward improvement suggestions for the prediction idea.

[0032] Specifically, the reflection model uses two-step instruction prompts: 1. The reflection model is instructed to analyze the error samples to obtain the reasons for the mistakes made by the disease diagnosis model; 2. The reflection model is required to propose suggestions for solving these problems. The prompt for the reflection model is as follows: Prompt = "You are a reflection model responsible for analyzing the diagnostic results of error samples of the disease diagnosis model and proposing improvement suggestions. Please perform the following tasks: First step, analyze the error samples: Based on <wrong thinking, wrong diagnostic results, expert's diagnostic results> and the current diagnostic basis <patient information, first knowledge point, patient symptoms, prediction thinking>, analyze the reasons for the mistakes made by the disease diagnosis model. Please elaborate on the possible factors leading to errors, including but not limited to insufficient information, misuse of knowledge points, or logical reasoning errors. Second step, propose improvement suggestions: Based on the analysis in the first step, propose specific improvement suggestions to optimize the prediction thinking. These suggestions should include how to better utilize the existing information, supplement necessary knowledge points, or adjust the diagnostic logic to improve the diagnostic accuracy. Please ensure that your analysis and suggestions are clear and specific and can be directly applied to improving the diagnostic ability of the disease diagnosis model, limited to 1 to 3 sentences."

[0033] After the reflection model gives improvement suggestions for the error samples, the optimization model sorts them out and generates specific prediction thinking. Specifically, the optimization model will perform operations such as adding, deleting, or editing the prediction thinking according to the modification suggestions, and summarize the specific prediction thinking under specific diseases. The prompt for the optimization model is as follows: Prompt = "I am training a disease diagnosis model. You are an optimization model responsible for sorting out and generating new prediction thinking based on <the improvement suggestions provided by the reflection model> and referring to the current diagnostic basis. Please perform the following steps: First step, modify the original prediction thinking according to the suggestions of the reflection model, including adding, deleting, or editing relevant content to ensure that the prediction thinking is more accurate and comprehensive; Second step, systematize the modified prediction thinking and summarize it under the diagnostic framework of specific diseases to ensure its applicability to the diagnosis of the disease. Please return the latest prediction thinking, ensuring that the output is clear and structured."

[0034] After one optimization, the disease diagnosis basis is <patient information, first knowledge point, patient symptoms, latest prediction thinking>. We set a confidence index to measure the level of the disease diagnosis model, where represents the entire training set, represents taking the logarithm of the diagnostic result confidence for the sample .

[0035] Iterative training is carried out on the training set until exceeds the preset threshold , it can be considered that the performance of the disease diagnosis model on the training set already meets the requirements. Through this iterative optimization training method, the disease diagnosis model can gradually improve the diagnostic accuracy on the training set to reach a higher diagnostic level.

[0036] In one embodiment, the similar medical record retrieval model includes an information extraction module, a feature enhancement module, and a temporal optimization module: The hospital medical record library includes multiple hospital patient medical records. The information extraction module extracts patient information, patient symptoms, and disease diagnosis results from the hospital patient medical records; the feature enhancement module is used to enhance the features of patient information, patient symptoms, disease diagnosis results, patient information, patient symptoms, and disease diagnosis results in the hospital patient medical records to better retrieve the specified medical record; the temporal optimization module is used to optimize the features of the patient information, patient symptoms, and disease diagnosis results after feature enhancement according to historical diagnosis data, and optimize the patient information, patient symptoms, and disease diagnosis results according to the temporal data in the historical diagnosis data, so as to retrieve the matching specified medical record.

[0037] In one embodiment, the temporal optimization module is used to optimize the features of the patient information, patient symptoms, and disease diagnosis results after feature enhancement according to historical diagnosis data, including: Encoding and feature enhancing the patient information, patient symptoms, and disease diagnosis results each time in the historical diagnosis data with the patient information, patient symptoms, and disease diagnosis results of this time to obtain corresponding feature vectors , , , where refers to the th feature vector of the medical record, is the sequence length, where, , , are the feature vectors corresponding to the patient information, patient symptoms, and disease diagnosis results of this time. Regarding the medical record sequence in the historical diagnosis data as a time series, input it into the long short-term memory network for processing. The long short-term memory network updates the hidden state through the following formula: ; where, is the th hidden state, LSTM is the long short-term memory network, is the hidden state of the last time step of the sequence, representing the development trend feature of the entire medical record sequence. Use this development trend feature to optimize the feature vector of this time. The formula is as follows: ; where, is the weight matrix, is the bias, represents vector concatenation, is the optimized feature vector, and the other two features and Similarly, the trend of disease development is also incorporated to obtain the optimized feature vector .

[0038] In one embodiment, the similar medical record retrieval model is used to retrieve the specified medical record according to the similarity between the patient information, patient symptoms, disease diagnosis results after feature enhancement and the patient information, patient symptoms and disease diagnosis results in the hospital patient medical record after feature optimization, including: Calculate the fusion similarity between the patient information, patient symptoms, disease diagnosis results after feature optimization and the patient information, patient symptoms and disease diagnosis results in the hospital patient medical record after feature enhancement : + ; is the representation vector of patient information, is the representation vector of patient symptoms, is the representation vector of disease diagnosis results; is the representation vector of patient information in the hospital patient medical record, is the representation vector of patient symptoms in the hospital patient medical record, is the representation vector of disease diagnosis results in the hospital patient medical record; is the weight of patient information , is the weight of patient symptoms , is the weight of disease diagnosis results ; is a similarity function, a function used to calculate the similarity between two feature vectors; retrieve the specified medical record according to the specified fusion similarity .

[0039] As Figure 3 shown, the details of the adaptive reflection and reasoning module are as follows: After the disease analysis module 130 makes a disease diagnosis, the system needs to propose a treatment recommendation for the patient. In order to make this treatment recommendation more rigorous and in line with the patient's disease development, the similar medical record retrieval model, disease development analysis model and medical advice model are used in this part.

[0040] The similar medical record retrieval model retrieves the most valuable reference medical records from hospital medical records through dynamic feature fusion technology. Considering that different diseases in the disease diagnosis results have different sensitivities to patient symptoms and patient information, this model uses a trainable neural network model to dynamically fuse the three types of information to obtain similar medical records. Moreover, for the problem of feature sparsity of inactive users (with less information in all aspects, making it difficult to make disease diagnoses and medical suggestions), this model uses a feature enhancement module to enhance the representation of inactive users. Secondly, according to the historical diagnosis data in the system, the three features of patient information, patient symptoms, and disease diagnosis results in the system are optimized. The historical diagnosis data is time-series data, and the three features of the most recent patient information, patient symptoms, and disease diagnosis results are optimized according to the time-series data, so as to better retrieve similar medical records in the hospital patient medical records.

[0041] The specific steps are as follows: Step 1: Organize the three features (patient information, patient symptoms, disease diagnosis results) to obtain a representation in the form of key-value pairs. The large model encodes each attribute in the features to obtain an initial encoded vector. That is, the patient information is represented as , the patient symptoms are represented as , and the disease diagnosis results are represented as .

[0042] Step 2: Feed the initial encoded vector of each feature into the feature enhancement module. During the training process, a certain proportion of each feature is randomly masked, that is, a part of the features are randomly selected and set to zero to simulate the situation of less active users. After random masking, it is fed into the feature enhancement module. The feature enhancement module uses an autoencoder to generate a new feature vector through dimensionality reduction and reconstruction. The encoder part encodes the input features through dimensionality reduction, and the decoder part restores the hidden representation to a reconstructed version of the input features. The used formulas are: ; ; Among them, is the input feature vector (such as patient information , patient symptoms , disease diagnosis results ), is the weight matrix, is the bias vector, is the activation function, is the hidden representation after encoding; is the weight matrix of the decoder, is the bias vector of the decoder, is the activation function, is the reconstructed feature, that is, the feature after feature enhancement. In the following, the reconstructed feature is represented as patient information , patient symptoms , disease diagnosis result .

[0043] The feature enhancement module can capture the potential complex non - linear relationships under each feature. Compared with the single - layer model, this method can enable the three features to be better and more comprehensively represented, especially in the scenario of inactive users, which can greatly improve the retrieval accuracy.

[0044] Step 3: In order to better apply the similar medical record retrieval model in the system described in the present invention and better adapt to the dynamic development of the patient's condition, it is necessary to continue training the model with the time - series optimization module. The following are the detailed design and implementation steps of this module: Using the encoding method and feature enhancement method mentioned above, obtain the encoding vectors of the three features of each medical record in a series of medical records , , , where refers to the encoding of the th medical record, is the sequence length. Regarding the medical record sequence as a time series, input it into the long short - term memory network for processing to capture the development trend of the condition. The long short - term memory network updates the hidden state through the following formula: ; where, is the hidden state at time , is the hidden state of the last time step of the sequence, representing the development trend characteristics of the entire medical record sequence. Use this development trend characteristic to optimize the encoding vector in the last medical record. The formula is as follows: ; where, is the weight matrix, is the bias, represents vector concatenation, is the optimized feature vector. The other two features and are similarly fused with the development trend of the condition to obtain the optimized feature vectors . Use the optimized feature vector to perform subsequent dynamic weight and similarity calculations to obtain similar cases.

[0045] Here, the time - series optimization module has also completed training. When applied in the system, the time - series optimization module can capture the development trend of the patient's condition, and through the application of the feature enhancement mechanism and dynamic weight mechanism, the retrieval accuracy of similar medical records can be significantly improved.

[0046] Step 4: Considering that different diseases in the disease diagnosis results have different sensitivities to patient symptoms and patient information, in the case of similar medical records, the weights of the three features will also change with the disease. Here, the dynamic weights of the above three features are obtained. , the formula used is as follows: ; where and are both trainable parameters, is the encoding vector of the disease diagnosis result, represents taking the logarithm, is the weight of the three , = 1, 2, 3, respectively representing the indices of the three, and are both intermediate state representations of the weights before normalization, is the weight after normalization, and T is the matrix transpose.

[0047] Step 5: Calculate the fusion similarity between the patient information, patient symptoms, disease diagnosis result after feature optimization and the patient information, patient symptoms, and disease diagnosis result in the hospital patient medical record after feature enhancement , and the calculation formula is as follows: + ; is the representation vector of patient information, is the representation vector of patient symptoms, is the representation vector of the disease diagnosis result; is the representation vector of patient information in the hospital patient medical record, is the representation vector of patient symptoms in the hospital patient medical record, is the representation vector of the disease diagnosis result in the hospital patient medical record; is the weight of patient information , is the weight of patient symptoms , is the weight of the disease diagnosis result ; is the similarity function, a function used to calculate the similarity between two feature vectors; is the fused similarity, which can effectively reflect the similarity degree between medical records. Designate the medical record with the largest similarity as the hospital patient medical record.

[0048] Step 6: To train the model, a training set is required. Construct a training set that includes the medical record sequence data of multiple patients. Each group of sequences consists of the medical records of a patient at different time points, reflecting the change of the condition over time. The sequence lengths are random, with some patients having more medical records and some having fewer, to simulate the diversity of patient visits in the real scenario and improve the generalization ability of the model. The training objective is to retrieve the medical record most similar to the last medical record in the patient's medical record sequence. The model needs to learn from the historical sequences to predict the similar cases that match the last medical record. The mean reciprocal rank is used as the training metric, that is , where is the total number of queries,[[]]ID=5 is the rank of the correct answer of the -th query in the prediction results, and the Adam optimization algorithm is used to update the model parameters. Through the collaborative work of the above modules, the present invention can dynamically adjust the similarity calculation weights according to the disease characteristics and retrieve the most valuable hospital patient medical records.

[0049] The disease development analysis model is based on large model technology and aims to observe and reflect on whether previous treatment suggestions have been effective for patients. If it is found that the treatment effect is poor or very weak, the system will feedback the methods to improve the treatment suggestions to the system for optimizing future treatment suggestions. The workflow of this model is as follows: Step 1: Data collation: Collect the patient's historical disease data, treatment records, and current diagnosis results; Step 2: Reflect on the treatment process: Use the large model to analyze what the patient has done and the change of the condition after receiving medical advice to judge whether the expected treatment goal has been achieved; Step 3: Feedback: Based on the analysis results, generate modification suggestions for the medical advice to ensure that it conforms to the disease development trend of the patient. The prompt for this part is as follows: Prompt = "You are a disease development analysis model, responsible for analyzing the disease development of patients, observing whether the disease development of patients meets the expectations, and proposing improvement methods for the current medical advice. Please perform the following tasks: Step 1, Analyze the past medical advice and the patient's disease development: Analyze the treatment effect and disease development of the patient under the corresponding medical advice according to <each medical advice> and <each patient symptom> and <each disease diagnosis result>. Step 2, Propose improvement suggestions: If the disease development of the patient does not meet the expectations, propose improvement methods for the medical advice. These suggestions should include adjusting the treatment method, adding adjuvant treatment means, etc., so that the improved medical advice is operable and can better conform to the patient's condition".

[0050] ​After obtaining the improved method of similar medical records and medical advice, the system can give more rigorous and targeted medical advice to maximize the improvement of the patient's health condition and achieve good results. The prompt for this part is as follows: Prompt = "You are a medical advice model, responsible for referring to similar medical records and the analysis results of the patient's condition development to provide personalized medical advice. Please perform the following tasks: First step, combine <the current disease diagnosis result> and <patient information, the first knowledge point, patient symptoms, prediction ideas> to give preliminary medical advice, which includes but is not limited to lifestyle adjustment, medication adjustment, recommended follow-up, etc.; Second step, based on the treatment plan of <similar medical records> and the <optimization method for medical advice> obtained from the disease development analysis model, put forward the optimized medical advice to ensure that it meets the individual needs of the patient and the trend of disease development. Please ensure the operability and scientific nature of the advice to improve the treatment effect of the patient."

[0051] The disease diagnosis results and medical advice given by the large model disease diagnosis module and the adaptive reflection reasoning module will be stored in the disease diagnosis database 140 for subsequent use.

[0052] In the embodiment of the present application, the popular science module 120 further includes a knowledge base, a key information extraction model, and a knowledge point extraction model: the knowledge base stores medical knowledge points; the key information extraction model is used to extract the patient's current symptoms, patient problems, and the main content of the video according to patient information, interaction information, historical diagnosis data, and popular science videos; the knowledge point extraction model is used to extract the second knowledge point in the knowledge base according to the patient's current symptoms, patient problems, and the main content of the video.

[0053] The relevant medical knowledge in the knowledge base is systematically organized into independent medical knowledge points. This process includes the analysis and extraction of a large number of medical literature, research papers, and clinical guidelines to ensure that each knowledge point covers the key concepts, symptoms, treatment methods, and the latest research progress of chronic obstructive pulmonary disease. Next, an embedding model is used to convert these text data into high-dimensional vector representations. This representation method can capture the semantic information of the text, making subsequent retrieval more accurate. Finally, these high-dimensional vectors are stored in the knowledge base to ensure that the knowledge base can efficiently support real-time query and retrieval. The medical knowledge points can be knowledge points of chronic obstructive pulmonary disease.

[0054] Send the user's question and background information into the key information extraction model. Use a large model to perform key information extraction through prompt engineering, and extract key topics, entities, and related symptoms and problems from the user input to ensure a clear focus for retrieval. The prompt for this part can be as follows: Prompt = "You are an excellent key information extraction model. You will be given a piece of complex information. Your task is to accurately identify and organize the key information in this information. Your response must include the following three aspects: the patient's current symptoms, the patient's problems, and the main content of the video. The information you need to process includes <patient information>, <interaction information>, <historical diagnosis data>, <science popularization video>". After the knowledge point extraction model obtains the information in three dimensions (the patient's current symptoms, the patient's problems, and the main content of the video) from the key information extraction model in the previous step, it embeds the information in these three dimensions in the same way to obtain three embedding vectors, and performs vector retrieval in the knowledge base using cosine similarity. Calculate the similarity between the medical knowledge points and the input vectors, where represents one of the information in the three dimensions, represents the encoded vector of the medical knowledge points. The maximum similarity of multiple medical knowledge points can be obtained for each dimension of information. All these retrieved second-level knowledge points include medical knowledge points related to the user's condition, the user's interests, and the course content.

[0055] After the above operations, the knowledge point extraction model can obtain the second-level knowledge points that are closest to the user's needs. These second-level knowledge points will continue to be sent into the interaction framework of the digital human model. After referring to these knowledge points, the digital human model can make more rigorous and correct responses.

[0056] After the user enters the system, they can systematically learn science popularization videos in the science popularization module 120. The science popularization videos include knowledge courses on chronic obstructive pulmonary disease and interact with the digital human. The user has two forms of interaction. The first is to pause at any time during the process of watching the content of the science popularization video and ask questions to the digital human. The digital human will provide personalized responses based on the questions raised by the user and the content of the science popularization video. The second is that after the video ends or after the previous interaction is completed, the digital human asks guiding questions to the user to guide the user to further describe their symptoms or to guide the user to watch relevant science popularization videos. In the second form of interaction, for each science popularization video, the digital human will purposefully ask questions to the user to guide the user to answer. On the one hand, this question will combine the video content to obtain the user's actual situation, and on the other hand, after obtaining the user's answer, it will give the next direction (recommend a science popularization video to the user or continue to ask the user questions).

[0057] In one embodiment, the digital human model determines the interaction direction based on patient information, interaction information, historical diagnosis data, and the second knowledge point. The interaction direction includes two interaction directions. The first interaction direction is to recommend a science popularization video at a specified position for the patient to watch, and the second interaction direction is to ask the patient relevant questions.

[0058] After the knowledge point extraction model retrieves the second knowledge point related to the user, the system sends the patient information, interaction information, historical diagnosis data, and the second knowledge point into the digital human interaction framework. The digital human will judge the next interaction direction (recommend a science popularization video to the user or continue to ask the user questions) based on all the information, so as to further obtain the user's condition information or improve the user's understanding of related diseases. In the first interaction direction, the video recommendation expert of the science popularization module 120 recommends a science popularization video at a specified position for the patient to watch. In the second interaction direction, the condition inquiry expert of the science popularization module 120 continues to ask the patient relevant questions. After each user question, the interaction information and the retrieved second knowledge point will be sent to the set digital human interaction framework to guide the user to describe their symptoms in detail or guide the user to further watch the corresponding science popularization video, so as to achieve a good science popularization effect.

[0059] As Figure 4 shown, the details of the digital human interaction framework are as follows: First, in order to make the video recommendation more accurate, it is necessary to pre - do an AI summary for each science popularization video, extract key information such as the diseases, symptoms, treatment methods, etc. described in the video as the video summary for the reference of the video recommendation expert, so that the video recommendation expert can recommend the most suitable video for the user's needs.

[0060] After sorting out the user information, in order to dynamically select the most relevant expert module according to the user's situation, it is necessary to make a further judgment on the user information. If the disease type can be roughly judged based on the user information, the information can be sent to the video recommendation expert. Otherwise, it is necessary to further ask the user about their condition and related symptoms. The user information includes patient information, interaction information, historical diagnosis data, and the second knowledge point. The prompt for this part is as follows: Prompt = "According to the following user information, judge whether the disease type of the user can be roughly determined. If it can, mark it as 'determined' and pass the information to the video recommendation expert; if not, mark it as 'undetermined' and prepare to enter the condition inquiry expert. The information you refer to is <patient information, interaction information, historical diagnosis data, and the second knowledge point>, please provide the judgment result (determined / undetermined):".

[0061] In the medical history inquiry expert, the model generates a series of targeted questions based on the user information. These questions are designed to obtain more detailed information about the user's medical condition. These questions may cover the user's specific symptoms, onset time, changes in the condition, and any relevant lifestyle or environmental factors. In this way, the medical history inquiry expert can understand the user's health status more comprehensively, providing a more accurate basis for subsequent diagnosis and advice. The prompt for this part is: "Based on the user information, generate a series of targeted questions to obtain more detailed information about the user's medical condition. These questions should cover the user's specific symptoms, onset time, changes in the condition, and any relevant lifestyle or environmental factors. The user information you refer to is <patient information, interaction information, historical diagnosis data, and the second knowledge point>".

[0062] In the video recommendation expert, the system uses the previous video summary results of science popularization videos, combines with the user's specific needs and medical condition information, and recommends the most suitable science popularization videos. The video recommendation expert will give priority to selecting those videos that are most relevant to the user's current health status to help the user better understand their medical condition and provide practical health management advice. The prompt for this part is: "Use the following user information <patient information, interaction information, historical diagnosis data, and the second knowledge point> and <video summary of each science popularization video> to recommend the most suitable science popularization videos for the user. The videos should be highly relevant to the user's health status and needs to help the user better understand their medical condition and provide practical health management advice. When recommending videos, give reasons for the recommendation to guide the user to watch the science popularization videos with focus and purpose.

[0063] The system feeds back the responses of the medical history inquiry expert and the video recommendation expert to the user for further interaction. Through the interaction framework, the system digital human model can fully capture the user's medical symptoms and accurately recommend science popularization videos to the user.

[0064] During the digital human interaction process, it is necessary to record the interaction information in detail: These data are completely stored in the "interaction database" to ensure comprehensive recording and analysis. The information for each interaction includes: the user's question (data item recorded as user), the digital human's response (data item recorded as system), relevant interaction details (such as time time, and session ID (sessionId)). The system can distinguish the current interaction from historical interactions. This information storage mechanism not only provides personalized health advice for users but also provides reliable data support for subsequent medical research and diagnosis.

[0065] In one embodiment, the patient symptoms, disease diagnosis results, and medical advice obtained by the disease analysis module 130 are stored in the disease diagnosis database 140. The patient symptoms obtained by the data collation module, the disease diagnosis results obtained by the large model disease diagnosis module, and the medical advice obtained by the adaptive reflection and reasoning module are stored in the disease diagnosis database 140.

[0066] In an embodiment of the present application, the disease development analysis model determines whether the expected treatment goal is achieved based on the disease diagnosis results and the historical diagnosis data in the disease diagnosis database 140; if the expected treatment goal is not achieved, an improvement method is proposed according to the expected treatment goal.

[0067] In an embodiment of the present application, the medical advice model obtains preliminary medical advice based on the disease diagnosis results, patient information, first knowledge points, patient symptoms, and prediction ideas; the medical advice model optimizes the preliminary medical advice according to the medical advice of the specified medical record to obtain medical advice.

[0068] In the system, the function of viewing historical diagnosis data provides a user with an intuitive and convenient interface to help the user comprehensively understand their own health status and disease development trend; the user can freely view their own historical diagnosis data and track the development of their disease.

[0069] The interface provides the user with an intuitive display of the disease changes and supports multi-dimensional filtering. The system defaults to displaying all the user's historical diagnosis data in chronological order. In the form of a timeline, the user can clearly see the date and time of each diagnosis, which is convenient for tracking the disease changes. To enhance the visualization effect, the system provides a variety of chart options, such as line charts, bar charts, and pie charts, to display the user's disease trends, symptom changes, and diagnosis result distributions.

[0070] In the viewing of the historical diagnosis data, the user can click on each diagnosis event to view the detailed diagnosis report and can also view the interaction information during the diagnosis process to timely understand and track the progress of their disease.

[0071] The prediction system for chronic obstructive pulmonary disease of the present invention strengthens the patient's self-management through the popular science module and improves the user's awareness of the disease. Through the disease analysis and diagnosis module, the patient is deeply analyzed and accurately diagnosed, the disease diagnosis results are given, and rigorous and personalized medical advice is provided to help the user conduct effective disease management.

[0072] In addition, the present application also provides a storage medium, on which a computer program is stored. When the computer program is run by a processor, the processor is caused to execute the prediction system for chronic obstructive pulmonary disease according to the embodiments of the present application described above. The storage medium may include, for example, a memory card of a smart phone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0073] Although example embodiments have been described herein with reference to the drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present application thereto. Those of ordinary skill in the art can make various changes and modifications therein without departing from the scope and spirit of the present application. All such changes and modifications are intended to be included within the scope of the present application as claimed in the appended claims.

[0074] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present application.

[0075] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0076] In the specification provided herein, a large number of specific details are set forth. However, it can be understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures, and technologies have not been shown in detail so as not to obscure the understanding of this specification.

[0077] Similarly, it should be understood that, for the purpose of streamlining the present application and facilitating the understanding of one or more of the various inventive aspects, in the description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the method of the present application should not be construed as reflecting an intention that the claimed present application requires more features than those expressly recited in each claim. Rather, as reflected in the corresponding claims, the inventive point lies in that the corresponding technical problem can be solved with features fewer than all the features of a single disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into the detailed description, where each claim itself serves as a separate embodiment of the present application.

[0078] Those skilled in the art will appreciate that, except where features are mutually exclusive, any combination can be used to combine all the features disclosed in this specification (including the accompanying claims, abstract, and drawings), as well as all the processes or units of any method or apparatus so disclosed. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) can be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0079] In addition, those skilled in the art will be able to understand that, although some of the embodiments described herein include certain features included in other embodiments rather than other features, the combination of features of different embodiments means that it is within the scope of the present application and forms different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0080] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) can be used in practice to implement some or all of the functions of some of the modules according to the embodiments of the present application. The present application can also be implemented as a program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, or provided on a carrier signal, or in any other form.

[0081] It should be noted that the above embodiments are illustrative of the present application rather than restrictive thereof, and those skilled in the art can design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present application can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In the unit claims listing several abnormal detection devices of a train traction system, several of these abnormal detection devices of the train traction system may be embodied by the same item of hardware. The use of the words first, second, and third, etc. does not denote any order. These words can be interpreted as names.

[0082] As described above, the above is only a specific implementation manner or an illustration of the specific implementation manner of the present application, and the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. The protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A prediction system for chronic obstructive pulmonary disease, characterized in that, It includes an information collection module, a science popularization module, a disease condition analysis module and a disease condition diagnosis database: The information collection module is used to collect patient information; The science popularization module includes a digital human model, which is used to play science popularization videos and interact with patients through the digital human model to generate interaction information; The disease condition analysis module includes a data sorting module, a large model disease condition diagnosis module and an adaptive reflection and reasoning module; The data sorting module is used to extract the first knowledge points and patient symptoms according to the patient information, science popularization videos and interaction information; The large model disease condition diagnosis module includes a disease condition diagnosis model. The large model disease condition diagnosis module generates a prediction idea, and the disease condition diagnosis model obtains a disease condition diagnosis result according to the patient information, the first knowledge points, the patient symptoms and the prediction idea; The adaptive reflection and reasoning module includes a similar medical record retrieval model, a disease condition development analysis model and a medical advice model; The similar medical record retrieval model is used to retrieve a specified medical record according to the patient information, the patient symptoms and the disease condition diagnosis result; The disease condition development analysis model is used to propose improvement methods according to the disease condition diagnosis result and the historical diagnosis data in the disease condition diagnosis database; The medical advice model gives medical advice by combining the specified medical record and the improvement method.

2. The prediction system for chronic obstructive pulmonary disease according to claim 1, characterized in that, The large model disease condition diagnosis module also includes a reflection model and an optimization model. During the training process, the large model disease condition diagnosis module trains the disease condition diagnosis model through the reflection model and the optimization model: During the training process, the disease condition diagnosis model diagnoses the samples according to the patient information, the first knowledge points and the patient symptoms to obtain an incorrect sample diagnosis result and an incorrect idea; The reflection model analyzes the cause of the error according to the incorrect sample diagnosis result, the corresponding sample and the incorrect idea, and proposes an improvement suggestion for the incorrect idea; The optimization model generates a prediction idea according to the cause of the error and the improvement suggestion for the incorrect idea.

3. The prediction system for chronic obstructive pulmonary disease according to claim 1, wherein The similar medical record retrieval model includes an information extraction module, a feature enhancement module and a time series optimization module: The hospital medical record library includes multiple hospital patient medical records. The information extraction module extracts the patient information, the patient symptoms and the disease condition diagnosis result from the hospital patient medical records; The feature enhancement module is used to enhance the features of the patient information, the patient symptoms, the disease condition diagnosis result, the patient information, the patient symptoms and the disease condition diagnosis result in the hospital patient medical records, so as to better retrieve the specified medical record; The time series optimization module is used to optimize the features of the patient information, the patient symptoms and the disease condition diagnosis result after feature enhancement according to the historical diagnosis data, and optimize the patient information, the patient symptoms and the disease condition diagnosis result according to the time series data in the historical diagnosis data, so as to retrieve the matching specified medical record.

4. The prediction system for chronic obstructive pulmonary disease according to claim 3, wherein The time series optimization module is used to optimize the features of the patient information, the patient symptoms and the disease condition diagnosis result after feature enhancement according to the historical diagnosis data, including: Encode and enhance the features of the patient information, patient symptoms, and disease diagnosis results in each historical diagnosis data, as well as those of the current patient information, patient symptoms, and disease diagnosis results, to obtain the corresponding feature vectors , , , where refers to the feature vector of the th medical record, is the sequence length. Among them, , , are the feature vectors corresponding to the current patient information, patient symptoms, and disease diagnosis results. Treat the medical record sequence in the historical diagnosis data as a time series and input it into a long short-term memory network for processing. The long short-term memory network updates the hidden state through the following formula: ； Among them, is the hidden state at a moment, and LSTM is a long short-term memory network. is the hidden state of the last time step of the sequence, representing the development trend feature of the entire medical record sequence. The feature vector of this time is optimized using this development trend feature, and the formula is as follows: ； Among them, is the weight matrix, is the bias, represents vector concatenation, is the optimized feature vector, and the other two features and similarly, also incorporate the trend of disease development to obtain the optimized feature vector .

5. The prediction system for chronic obstructive pulmonary disease according to claim 4, wherein The similar medical record retrieval model is used to retrieve the specified medical record according to the similarity between the patient information, the patient symptoms and the disease condition diagnosis result after feature optimization and the patient information, the patient symptoms and the disease condition diagnosis result in the hospital patient medical records after feature enhancement, including: Calculate the fusion similarity between the patient information, patient symptoms, and disease diagnosis results after feature optimization and the patient information, patient symptoms, and disease diagnosis results in the hospital patient medical records after feature enhancement : + ； is the representation vector of patient information, is the representation vector of patient symptoms, is the representation vector of the disease diagnosis result; is the representation vector of patient information in the hospital patient medical record, is the representation vector of patient symptoms in the hospital patient medical record, is the representation vector of the disease diagnosis result in the hospital patient medical record; is the patient information weight, is the patient symptom weight, is the disease diagnosis result weight; is a similarity function, a function used to calculate the similarity between two feature vectors; According to the specified fusion similarity , the specified medical record is retrieved.

6. The prediction system for chronic obstructive pulmonary disease according to claim 1, characterized in that, The science popularization module also includes a knowledge base, a key information extraction model and a knowledge point extraction model: The knowledge base stores medical knowledge points; The key information extraction model is used to extract the patient's current symptoms, patient problems, and the main content of the popular science video based on patient information, interaction information, historical diagnosis data, and popular science videos; The knowledge point extraction model is used to extract second knowledge points from the knowledge base based on the patient's current symptoms, patient problems, and the main content of the video; The digital human model determines the interaction direction based on patient information, interaction information, historical diagnosis data, and second knowledge points. The interaction direction includes two interaction directions. The first interaction direction is to recommend that the patient watch a popular science video at a specified location, and the second interaction direction is to ask the patient relevant questions.

7. The prediction system for chronic obstructive pulmonary disease according to claim 1, wherein The patient symptoms, disease diagnosis results, and medical advice obtained by the disease analysis module are stored in the disease diagnosis database.

8. The prediction system for chronic obstructive pulmonary disease according to claim 1, wherein The disease development analysis model determines whether the expected treatment goal is achieved based on the disease diagnosis results and historical diagnosis data in the disease diagnosis database; If the expected treatment goal is not achieved, an improvement method is proposed based on the expected treatment goal.

9. The prediction system for chronic obstructive pulmonary disease according to claim 1, wherein The medical advice model obtains preliminary medical advice based on the disease diagnosis results, patient information, first knowledge points, patient symptoms, and prediction ideas; The medical advice model optimizes the preliminary medical advice according to the medical advice of the specified medical record to obtain medical advice.

10. A storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by a processor, the system according to any one of claims 1-9 is implemented.

Images