Disease data online matching system based on artificial intelligence

Abstract

The invention discloses a disease data online matching system based on artificial intelligence, and aims to efficiently integrate multi-source disease data and accurately match doctor resources. The system comprises a multi-source data acquisition module used for acquiring text, image, inspection and questionnaire data; the multi-modal disease recognition model construction module fuses the multi-source data to perform disease recognition; the disease-symptom-examination knowledge network construction module provides medical knowledge support; the multi-dimensional matching algorithm design module is used for matching doctors by integrating subjects at which the doctors are skilled, resource loads and geographical distance factors, so that hardware-level parallel computing of distance, load and speciality is realized; the diagnosis and treatment scheme generation module generates a scheme according to the matching result; the data encryption transmission module guarantees data security; and the feedback and optimization module continuously optimizes the system performance. The method can improve the disease recognition accuracy and diagnosis and treatment efficiency, reasonably distributes medical resources, and is suitable for the field of Internet medical treatment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a disease data online matching system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of internet medical treatment, the amount of disease data is growing explosively, and the data types are diverse, covering text, image, test and questionnaire data, etc. The traditional medical data processing method is limited by the dispersion and diversity of data, and it is difficult to effectively integrate multi-modal disease data, resulting in the problems of insufficient accuracy and comprehensiveness in disease identification. For example, when only relying on text medical records for disease diagnosis, misdiagnosis or missed diagnosis may occur due to the lack of image data support, affecting the subsequent treatment effect.

[0003] At present, online medical service platforms have obvious defects in doctor resource allocation. Most platforms only allocate doctors to patients according to simple rules such as department classification, without fully considering key factors such as the specific details of the doctor's expertise, current resource load and geographical distance from the patient. This makes patients may match to doctors who lack experience in current disease diagnosis and treatment, have heavy reception tasks or are far away, not only prolonging the patient's medical time, but also reducing the diagnosis and treatment efficiency and quality, causing waste and unreasonable use of medical resources.

[0004] Medical data is highly sensitive, involving important information such as patient personal privacy and life health. In the process of data transmission, if there is a lack of strict and effective security protection mechanism, it is easy to lead to data leakage, causing serious damage to patients. The encryption mechanism of some existing medical data transmission systems is not perfect, and there is a risk of being cracked, which cannot fully guarantee the security of data.

[0005] In addition, the existing medical system generally lacks effective feedback and optimization mechanism, and it is difficult to adjust and improve the disease identification model and doctor matching algorithm in time according to the actual application situation, which cannot adapt to the complex and changing medical needs and data environment changes, especially in actual life there are many special circumstances, such as encountering military personnel and special patient priority or emergency surgery, need to cut off online diagnosis resources immediately, priority processing offline patients. This makes it difficult to continuously improve the accuracy of online disease identification and the rationality of doctor matching of the system, limiting the further improvement of the quality of internet medical service. SUMMARY

[0006] To solve the above problems, the application provides a disease data online matching system based on artificial intelligence, a cross-modal attention fusion mechanism is created to solve the problem of collaborative analysis of multi-source medical data, a dynamic knowledge network enhancement engine is constructed, a multi-dimensional matching algorithm design module is designed, hardware-level parallel computing of distance, load and expertise is realized, a closed-loop feedback mechanism is established, the weight coefficient is dynamically adjusted according to the doctor's rejection rate, the data transmission security reaches the third level of information security protection, and the system is suitable for the scene of hierarchical diagnosis and treatment.

[0007] To achieve the above object, the technical scheme adopted by the application is as follows: a disease data online matching system based on artificial intelligence is provided, which comprises: A multi-source data acquisition module is used to acquire multi-source data of patients, wherein the multi-source data comprises text data, image data, test data and questionnaire data; the text data comprises medical record texts and diagnosis reports; the image data comprises X-ray, CT and MRI image data; the test data comprises blood and urine test indexes; and the questionnaire data comprises symptom evaluation and life habit information filled in by patients; A multi-modal disease recognition model construction module is used to construct a multi-modal disease recognition model, extract data features in the text data, image data, test data and questionnaire data collected by the multi-source data acquisition module, and realize recognition of diseases; A knowledge network construction module is used to construct a disease-symptom-examination knowledge network, wherein the knowledge network comprises the association relationship among diseases, symptoms and examination items, and each association relationship is given a weight, and the weight is used to represent the closeness of the association; A multi-dimensional matching algorithm design module is used to construct a multi-dimensional matching algorithm according to the subject expertise of doctors registered in the system, resource load and geographical distance; the multi-dimensional matching algorithm calculates the matching degree score of the disease recognition result and the doctor according to a preset weight formula, arranges the matching degree scores in descending order, and sends the multi-source data to the doctor with the highest matching degree score.

[0008] Preferably, the text data acquisition tool in the multi-source data acquisition module uses natural language processing technology to acquire and arrange the text data; the image data acquisition tool is connected with the hospital imaging department equipment and performs image preprocessing; the test data acquisition tool obtains test index data from a laboratory information system; and the questionnaire data acquisition tool collects questionnaire information through an online questionnaire platform and performs validity verification.

[0009] Preferably, the multi-modal disease recognition model construction module adopts deep learning technology, wherein a convolutional neural network is used to process image data, a long short-term memory network is used to process time sequence information in text data, a multi-layer perception machine is used to process test data and questionnaire data, and feature fusion networks are used to deeply fuse the features of each modal data.

[0010] As preferred, in the multi-dimensional matching algorithm design module, the doctor's specialty is profiled by analyzing the doctor's historical diagnosis and treatment data and professional certification information; the resource load is calculated according to the current number of doctor's reception and scheduling; the geographical distance is measured based on the latitude and longitude information of the patient and the doctor's location, and the multi-dimensional matching algorithm quantitatively matches the disease recognition result and the doctor's resource according to the preset weight formula; The multi-dimensional matching algorithm design module includes a distance factor calculation submodule, a disease expertise recursive submodule, and a dynamic weight controller, wherein the distance factor calculation submodule is used to convert real-time road condition time into spatial equivalent distance during peak traffic periods; the disease expertise recursive submodule is used to calculate the maximum similarity of neighbor diseases within a relevant range in the knowledge network when the user's current disease is not in the doctor's expertise set.

[0011] More preferably, the weight formula is as follows: ; Wherein, α, β and γ are adjustable weight coefficients, d is the real-time road distance, L is the doctor's resource load rate, wherein the doctor's resource load rate = current number of cases to be diagnosed / daily reception capacity, E is the disease expertise matching degree, and the dynamic weight controller reduces the value of γ weight when the rejection rate is greater than 15%.

[0012] More preferably, the dynamic weight controller includes a rejection rate counting register, which is used to count the doctor's rejection of matching events according to the actual reception situation, wherein the weight coefficient is normalized by a hardware divider at 0 o'clock every day.

[0013] As preferred, the data encryption transmission module uses AES algorithm to encrypt the disease data, and uses SSL / TLS protocol for secure communication during data transmission to prevent data from being stolen or tampered with.

[0014] As preferred, the feedback and optimization module continuously optimizes the multi-modal disease recognition model and multi-dimensional matching algorithm through machine learning algorithm, and regularly extracts new knowledge from authoritative medical knowledge base and expert experience to update the disease-symptom-examination knowledge network.

[0015] As preferred, it also includes a diagnosis and treatment scheme generation module: for generating a preliminary diagnosis and treatment result according to the matched doctor, the doctor combines the patient's detailed medical data with the online historical diagnosis and treatment experience and disease-symptom-examination knowledge network data, and judges whether to accept the patient; Data encryption transmission module: for using encryption algorithm to encrypt the collected disease data, and using secure communication protocol during data transmission to ensure data security.

[0016] As preferred, a feedback and optimization module is also included: for collecting feedback information of patients and doctors, and optimizing the multi-modal disease recognition model and multi-dimensional matching algorithm based on the feedback information, and updating the disease-symptom-examination knowledge network in real time.

[0017] The beneficial effects of the present application are: Improve disease recognition accuracy: Through the multi-source data acquisition module, comprehensive disease-related data is obtained, including text, image, test and questionnaire data, etc. The multi-modal disease recognition model fuses multiple data modalities for disease recognition, fully excavates the rich information in the data, and can more accurately diagnose diseases compared with single data modalities. At the same time, combined with the disease-symptom-examination knowledge network, the medical knowledge and association relationship therein are utilized to further enhance the recognition ability of the disease, provide more accurate basis for subsequent diagnosis and treatment, reduce the misdiagnosis rate and missed diagnosis rate, and improve the medical diagnosis level.

[0018] Optimize the allocation of physician resources: The multi-dimensional matching algorithm considers the factors of physician expertise, resource load and geographical distance, and accurately matches physician resources. Preferentially, cases are assigned to physicians who are close in distance, have small load and are good at the current disease subject, so that patients can receive more timely and professional diagnosis and treatment services. This optimized allocation improves the utilization efficiency of medical resources, avoids the waste and unreasonable allocation of physician resources, and in the case of emergency, physicians refuse diagnosis and calculate the next highest matching physician through the matching algorithm, reasonably utilize medical resources, and improve the operation efficiency of the entire medical service system. Let patients obtain the most suitable medical resources in the shortest time.

[0019] Ensure data security: The data encryption transmission module uses advanced AES encryption algorithm to encrypt disease data, and uses SSL / TLS secure communication protocol during transmission, effectively preventing data from being stolen or tampered with, and ensuring the safety of patient privacy and medical data. Medical staff and patients can use the system for data transmission and sharing with confidence, enhancing the trust in internet medical care and promoting the healthy development of internet medical care.

[0020] Continuously optimize system performance: The feedback and optimization module collects feedback information of patients and doctors, such as patients' evaluation of diagnosis and treatment effect and doctors' opinion on the accuracy of disease recognition results, etc. Based on these feedback data, machine learning algorithms are used to continuously optimize the multi-modal disease recognition model and multi-dimensional matching algorithm, and update the disease-symptom-examination knowledge network in a timely manner. The system can adapt to the changing medical environment and data characteristics, maintain good performance, provide better and more accurate medical services for patients, and promote the continuous progress and innovative development of internet medical technology. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1It is a kind of disease data online matching system framework based on artificial intelligence. DETAILED DESCRIPTION

[0022] In a certain area with 100,000 residents, a kind of disease data online matching system based on artificial intelligence of the application is deployed, by accessing the HIS system of multiple hospitals in the area, the text medical record data of the patient is automatically collected, including the information of chief complaint, history of present illness, past medical history, allergy history, etc. in outpatient medical record and inpatient medical record;At the same time, connect the imaging department equipment of the hospital, collect X-ray, CT image data, and carry out preliminary gray scale adjustment, noise removal and other pretreatment on the image through image processing plug-in;Obtain the blood test report of the patient from the laboratory information system of the laboratory department, obtain the test data of each biochemical index, blood cell count, etc.;Set up questionnaire module on the WeChat public number and self-service terminal of the hospital, guide the patient to fill in the symptom self-evaluation questionnaire and lifestyle questionnaire before treatment, and collect the relevant questionnaire data.

[0023] Please refer to Figure 1 As shown in the figure, the application provides a kind of disease data online matching system based on artificial intelligence, which comprises: Multi-source data acquisition module: for collecting multi-source data of patients, the multi-source data includes text data, image data, test data and questionnaire data;The text data includes medical record text and diagnosis report;The image data includes X-ray, CT and MRI image data;The test data includes blood and urine test indicators;The questionnaire data includes the symptom evaluation and lifestyle information filled in by the patient; Enter the detailed information of multiple doctors in the regional hospital and multiple experts in the cooperative hospital in the system. The doctor information covers basic information (name, gender, age, title, department, etc.), professional qualifications (education background, graduation school, medical license number, professional certification certificate, etc.), diagnosis experience (working years, disease types and specific case numbers, historical cure rate, etc.), resource load (daily average number of patients, appointment scheduling table, etc.) and geographical location (detailed address coordinates of the hospital). At the same time, update the diagnosis data of the doctor regularly to ensure the accuracy and timeliness of the doctor information.

[0024] A patient, Ms. Zhang, went to the community hospital for cough and fever for a week, after outpatient registration, the system automatically collected her outpatient medical record text data in the community hospital. In addition, Ms. Zhang filled out a disease self-evaluation questionnaire through the mobile medical APP, which further supplemented the situational information when the symptoms occurred.

[0025] In the self-assessment questionnaire, there is also a preference for Western and traditional Chinese medicine. Since both Western and traditional Chinese medicine have their own advantages, the appropriate medical treatment should be chosen based on the specific situation. For acute diseases or serious health problems, Western medicine may be more effective; for chronic diseases and conditioning, traditional Chinese medicine has unique advantages; personal constitution and preference are also important factors. For some acute diseases or serious health problems, Western medicine may be more rapid and direct.

[0026] For chronic diseases such as diabetes, hypertension, and asthma, both traditional Chinese medicine and Western medicine have their own advantages. Traditional Chinese medicine focuses on overall conditioning and uses methods such as traditional Chinese medicine, acupuncture, and massage to improve the body's internal balance; Western medicine focuses more on drug treatment and symptom control. In the management of chronic diseases, a combination of traditional Chinese and Western medicine methods can achieve better results. In this example, Ms. Zhang chose the option of Western medicine. The system selects doctors who are good at Western medicine treatment from the doctors within the system range based on the preference selection. If the patient chooses intelligent recommendation in the questionnaire data survey report, the system will consider the patient's own situation and use a combination of traditional Chinese and Western medicine methods.

[0027] Multi-modal disease recognition model construction module: used for constructing a multi-modal disease recognition model, extracting data features from text data, image data, test data and questionnaire data collected by the multi-source data collection module, and realizing disease recognition; The multi-modal disease recognition model is responsible for fusing multi-source disease data and performing disease recognition, and is a key component of the system. Its construction method is as follows: Feature extraction: Text data: natural language processing techniques such as word embedding and topic modeling are used to extract keywords and topics that reflect the patient's condition. For example, when extracting features from medical records, key information such as symptom descriptions and diagnosis results can be converted into vector representations for subsequent processing.

[0028] Image data: computer vision techniques such as convolutional neural networks are used to extract texture and shape features from images. For X-ray images, lung texture features can be extracted to assist in identifying diseases such as pneumonia; for CT images, shape and size features of brain lesion regions can be extracted for brain disease diagnosis.

[0029] Test data: through data preprocessing and feature selection methods, representative index features in test data are extracted. For example, in blood test data, features such as white blood cell count and hemoglobin level related to disease are extracted.

[0030] Questionnaire data: data encoding and statistical analysis methods are used to extract features related to disease from the questionnaire. For symptom self-assessment questionnaires, symptom options selected by patients are encoded and the frequency of occurrence is counted as features for disease recognition.

[0031] Feature Fusion: Feature Concatenation: Simply concatenate the features extracted from different modalities of data to form a unified feature vector. For example, concatenate the keyword vector extracted from text data, the texture feature vector of image data, the indicator feature vector of examination data, and the symptom option feature vector of questionnaire data to form a multi-modal fusion feature vector.

[0032] Attention Mechanism Fusion: Introduce attention mechanism to weight and fuse features from different modalities. Attention mechanism can automatically learn the importance of different modalities in disease recognition and weight the features accordingly, so that the model pays more attention to features related to the disease.

[0033] Model Training and Optimization: Select appropriate deep learning model: According to the fused feature vector, select appropriate deep learning model for training, such as multilayer perceptron, convolutional neural network, recurrent neural network, etc. For example, for multi-modal fusion feature vector, multilayer perceptron can be used for classification learning to identify different disease types.

[0034] Optimize model parameters: Adjust the hyperparameters of the model, such as learning rate, regularization parameter, etc., and use appropriate optimization algorithms, such as stochastic gradient descent, Adam, etc., to optimize the performance of the model. In the training process, minimize the loss function as the goal, constantly adjust the model parameters, improve the accuracy and generalization ability of the model on the training data.

[0035] Model verification and evaluation: Use the validation dataset to verify and evaluate the trained model, calculate the accuracy, recall rate, F1 value, etc. to evaluate the performance of the model. If the model performance is not good, further adjust the model structure or optimization algorithm, and iterate the optimization of the model.

[0036] Model expansion and update: LoRA fine-tuning: Borrowing from the LoRA method used in the LLMI-CDP model, extend the pre-trained model. By fixing most of the parameters of the original model, introduce auxiliary matrices to replicate the comprehensive fine-tuning of model parameters, gradually integrate visual attributes such as images into pre-trained language models, and enhance the model's understanding and processing capabilities of multi-modal data.

[0037] Continuous learning and updating: With the continuous emergence of new medical data and knowledge, the model needs to be continuously learned and updated to maintain its accuracy and effectiveness. Online learning and other methods can be used to allow the model to continuously receive new data and feedback in actual application, automatically update model parameters and knowledge base, and improve the model's ability to identify new diseases and new treatment plans.

[0038] In the present embodiment, the multi-modal disease recognition model can effectively fuse multi-source disease data such as text, image, test and questionnaire through the above construction method, improve the accuracy and comprehensiveness of disease recognition, and provide reliable basis for subsequent doctor resource matching and diagnosis and treatment scheme generation.

[0039] According to the system data, Ms. Zhang has a history of chronic pharyngitis, and recently has symptoms of coughing and fever due to weather changes. The chest X-ray image taken on the same day shows that the lung texture is thickened, and there is no obvious substantial lesion. The blood test report shows that the white blood cell count is slightly higher than the normal range, and the neutrophil ratio is elevated. Ms. Zhang filled out the symptom questionnaire through the hospital WeChat public number before the visit, stating that the cough is dry cough, accompanied by throat itching, no sputum, fever with a maximum body temperature of 38.5℃, no other underlying diseases, and no recent travel history.

[0040] The multi-modal disease recognition model comprehensively analyzes the collected multi-source data. The natural language processing module analyzes the medical record text and extracts the key symptoms of "cough, fever, and chronic pharyngitis history". The image recognition module extracts features from the X-ray image and judges that it is consistent with the image manifestations of upper respiratory tract infection. The test data module analyzes the blood indicators and suggests the possibility of bacterial infection. The questionnaire data module further determines the symptom characteristics related to upper respiratory tract infection in combination with the patient's self-reported symptoms. After fusing the data of each modality, the multi-modal disease recognition model preliminarily judges that the patient Ms. Zhang has acute upper respiratory tract infection with a tendency of bacterial infection. The recognition result is matched with the disease-symptom-test knowledge network to obtain the recommended examination items (such as throat swab culture) and possible complication information (such as bronchitis and pneumonia) related to the disease, providing more comprehensive medical knowledge support for subsequent diagnosis and treatment.

[0041] Knowledge network construction module: used for constructing a disease-symptom-test knowledge network, the knowledge network comprising an association relationship between diseases, symptoms and test items, and each association relationship being assigned a weight representing the closeness of the association; Multi-dimensional matching algorithm design module: used for constructing a multi-dimensional matching algorithm according to the specialities of doctors registered in the system, resource load and geographical distance; the multi-dimensional matching algorithm calculates the matching score of the disease recognition result and the doctor according to a preset weight formula, ranks the matching scores in descending order, and sends the multi-source data to the doctor with the highest matching score; According to the identified acute upper respiratory tract infection disease type, the system calculates the disease recognition result of Ms. Zhang and the quantitative matching result of the regional doctor resources through the multi-dimensional matching algorithm, which is calculated by the following formula: The weight formula is as follows: ; wherein a, b and g are adjustable weight coefficients, d is the real-time road condition distance, L is the doctor resource load rate, wherein the doctor resource load rate = current number of cases to be diagnosed / daily reception capacity, E is the disease expertise matching degree, and wherein the dynamic weight controller reduces the value of g when the rejection rate is greater than 15%.

[0042] The calculation of the disease expertise matching degree E is divided into two main stages, forming a progressive process from precise matching to intelligent recommendation.

[0043] First stage: direct matching calculation: The system first retrieves the historical reception data of the target doctor from the doctor profile library over a period of time (such as the past three years), extracts all the disease codes diagnosed by the doctor, and forms a set of the doctor's specialized diseases D. At the same time, the system obtains the target disease code c that the current patient needs to be treated.

[0044] The core logic is to determine whether the disease code c exists in the doctor's specialized disease set D: if the current disease code (i.e., the doctor has a history of directly receiving the disease): At this time, the calculation of the matching degree E not only considers whether the doctor has seen the disease, but also further considers the quality of the treatment of the disease, and the matching degree E is directly equal to the historical cure rate of the doctor for the current disease c. For example, a doctor has treated 100 cases of pneumonia, of which 93 cases have been successfully cured, so the doctor's expertise matching degree E for pneumonia is 0.93. This ensures that the system preferentially matches patients to experts who not only have seen the disease, but are also more skilled at treating the disease.

[0045] If the current disease code (i.e., the doctor has no direct history of receiving the disease): the system will not simply consider the matching degree as 0, but will start the second stage: recursive reasoning calculation based on knowledge network.

[0046] Starting from the current disease c, find its k-hop neighbor diseases in the knowledge network (i.e., other diseases associated with disease c through several edges, for example, k is set to 2 or 3), forming an associated disease set C. For each disease in the associated disease set C, the system calculates the semantic similarity between it and the current disease c, and confirms the intersection list of the disease expertise matching degree E and C, and the similarity value corresponding to the disease with the highest similarity to the current disease as the value of the matching degree E.

[0047] Weight coefficient setting: according to the resource allocation strategy of the medical alliance and the emergency degree of the patient's condition, the initial weight coefficients are set as a = 0.3, b = 0.3, and g = 0.4.

[0048] The multi-dimensional matching algorithm design module includes a distance factor calculation submodule, a disease expertise recursive submodule, and a dynamic weight controller. The distance factor calculation submodule is used to convert real-time road condition time into spatial equivalent distance during a traffic peak period. The disease expertise recursive submodule is used to calculate the maximum similarity of neighbor diseases within a relevant range in the knowledge network when the user's current disease is not in the doctor's expertise set.

[0049] The dynamic weight controller includes a rejection rate count register, which is used to count the number of matching events rejected by doctors based on actual consultation conditions. The weight coefficient normalization is performed by a hardware divider at 0 o'clock every day.

[0050] The specific steps are as follows: The real-time geographic coordinates of Ms. Zhang are obtained through the positioning function of the mobile medical APP as (116.3957, 39.9128).

[0051] Doctor information update: Select doctors related to respiratory tract surgery in the medical association from the doctor information database, and obtain their real-time waiting case numbers, daily consultation capacity, etc. For example, Dr. Zhang in the emergency surgery department currently has 5 waiting cases and a daily consultation capacity of 20 cases; Dr. Wang in the community health service center has 2 waiting cases and a daily consultation capacity of 10 cases.

[0052] Road distance calculation: Call the map API and combine real-time traffic information to calculate the real-time road distance between Ms. Zhang and the medical institutions where the doctors are located. For example, the distance between Ms. Zhang and the comprehensive hospital where Dr. Zhang is located is 5 kilometers. Since it is during the peak traffic period, the estimated travel time is 30 minutes, which is converted to a real-time road distance of dZhang = 5 x 1.2 (congestion coefficient) = 6 kilometers; the distance to the community health service center where Dr. Wang is located is 1.5 kilometers, and the road is smooth, so the real-time road distance dWang = 1.5 x 1 = 1.5 kilometers.

[0053] Doctor resource load rate calculation: Dr. Zhang's load rate Lzhang = 5 / 20 = 0.25; Dr. Wang's load rate Lwang = 2 / 10 = 0.2.

[0054] Disease expertise matching degree determination: According to Ms. Zhang's condition keywords and examination results, the system knows from the doctor's professional knowledge graph that Dr. Zhang has rich experience in emergency surgery with high historical cure rate, and the disease expertise matching degree Ezhang = 0.9; Dr. Wang is also an ordinary surgeon, but has relatively insufficient experience in complex diagnosis and treatment, and the disease expertise matching degree Ewang = 0.6.

[0055] Matching score calculation: Dr. Zhang's matching score: S Zhang = 0.3 / 6 + 0.3 x (1-0.25) + 0.4 x 0.9 = 0.05 + 0.225 + 0.36 = 0.635; Dr. Wang's matching score: S Wang = 0.3 / 1.5 + 0.3 x (1-0.2) + 0.4 x 0.6 = 0.2 + 0.24 + 0.24 = 0.68; Dynamic weight adjustment and final matching system continuously monitors the matching rejection rate of the platform. It is found that Dr. Zhang is an emergency department doctor and often encounters sudden emergencies. Possible situations include: before Ms. Zhang arrives at Dr. Zhang's hospital, a car accident victim appears, Dr. Zhang needs to urgently treat the car accident victim, resulting in longer waiting time for Ms. Zhang and possibly worsening the patient's condition. Dr. Zhang's current rejection rate is 18% (exceeding the 15% threshold), triggering the dynamic weight controller. According to the set strategy, reduce the γ weight, the adjusted weight coefficients are α = 0.35, β = 0.35, γ = 0.3. Recalculate the matching score: Dr. Zhang's adjusted matching score: S Zhang' = 0.35 / 6 + 0.35 x (1-0.25) + 0.3 x 0.9 ≈ 0.058 + 0.2625 + 0.27 = 0.5905; At this time, Dr. Zhang's matching score is lower than Dr. Wang's, and the system will prioritize matching Ms. Zhang's case to Dr. Wang. At the same time, the platform sends instructions to the community health service center where Dr. Wang is located to assist in preliminary examination and stabilize the patient's condition, preparing for Ms. Zhang's transfer to the community health service center.

[0056] Diagnosis and treatment plan generation module: used to generate a preliminary diagnosis and treatment result based on the matched doctor, the doctor's online historical diagnosis and treatment experience and disease-symptom-examination knowledge network data, combined with the patient's detailed condition data, and determine whether to accept the patient; After receiving Ms. Zhang's case, Dr. Wang combines the multi-source condition data on the medical alliance platform (including preliminary examination results transmitted from other community health service centers), and formulates a detailed diagnosis and treatment plan: immediately perform examination to clarify the respiratory tract lesion, and consider intervention treatment or drug treatment according to the examination results. At the same time, the system feeds back the diagnosis and treatment plan to the system in real time, so as to assist other doctors in preparing for Ms. Zhang's preoperative preparation and preliminary treatment in the future.

[0057] The system generates a diagnosis and treatment plan based on Dr. Wang's experience in treating such patients (who have a high cure rate with antibiotic combination therapy) and the disease-symptom-examination knowledge network (recommended treatment principles include anti-infection and symptomatic support, with commonly used drugs being penicillin and cephalosporin antibiotics, while also emphasizing rest and adequate hydration). The system recommends that the patient undergo a throat swab culture to identify the pathogenic bacteria, and that the patient be given oral cefaclor capsules at 0.25g each, three times a day, for five consecutive days, and that the patient be given compound licorice mixture to relieve cough and reduce sputum, at 10ml each, three times a day. The patient is advised to rest and drink plenty of water, and to ensure adequate ventilation. Dr. Wang adjusts the plan based on the patient's actual situation and sends it to the patient. After one week of treatment, the patient's symptoms have improved significantly, and the patient reports good treatment results during the follow-up visit. Meanwhile, the system feeds back data from this diagnosis and treatment process (including disease identification accuracy, treatment effectiveness, and patient satisfaction) to the feedback and optimization module for subsequent optimization of the disease identification model and doctor matching algorithm, as well as updating the disease-symptom-examination knowledge network.

[0058] During Dr. Zhang's diagnosis and treatment process, the system continuously tracks her condition and treatment response. A disease manager is assigned to track and manage Dr. Zhang, providing assessment, monitoring, intervention, and health education to help her control her condition, improve her quality of life, and reduce complications. For chronic disease patients, a patient health record is established to collect medical history, physical examination data, and lifestyle information. Regular tracking of disease changes, such as blood pressure and blood sugar, assesses disease risk. Based on the patient's condition, an intervention plan is designed for diet, exercise, and medication. Targeted management services are provided for cardiovascular and diabetes diseases. Patients are guided on proper medication and self-monitoring, and bad habits are corrected. Disease knowledge is popularized to improve patient health literacy. Medical resources are coordinated to assist with follow-up appointments, follow-up, and remote consultations. Long-term tracking and management of discharged patients reduces readmission rates. After the diagnosis and treatment, Dr. Zhang evaluates the medical services, including satisfaction with the doctor's professional level and the convenience of seeking medical care. This feedback data will be used to further optimize the weight coefficients and parameter settings of the doctor resource matching algorithm, improving the accuracy and rationality of future matching. For example, if most patients are satisfied with the initial diagnosis and treatment assistance from community doctors, the weights of alpha and beta can be adjusted to encourage more use of community medical resources for initial treatment without affecting specialist treatment.

[0059] Through this embodiment, it can be seen that the doctor resource matching algorithm based on weight calculation can comprehensively consider multiple factors, realize intelligent matching of patients and doctors, and improve the utilization efficiency of medical resources and the quality of patient medical treatment. In actual application, algorithm parameters and weights can be flexibly adjusted according to characteristics of different medical institutions, patient group needs, medical policies and other factors to adapt to diversified medical scenarios and needs.

[0060] Data encryption transmission module: used for encrypting the collected disease data by using an encryption algorithm, and using a secure communication protocol in the data transmission process to ensure the security of the data; Feedback and optimization module: used for collecting feedback information of patients and doctors, and optimizing the multi-modal disease recognition model and multi-dimensional matching algorithm based on the feedback information, and updating the disease-symptom-examination knowledge network in real time.

[0061] Through the beneficial effects of the above aspects, the present application provides an efficient, accurate, safe and sustainable optimization disease data online push solution for the field of internet medical treatment, which helps to improve the overall medical service quality and efficiency, and has significant application value and broad market prospect.

[0062] The above embodiments only describe the preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements of the technical solutions of the present application made by ordinary engineering technicians in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. An artificial intelligence-based online disease data matching system, characterized by: include: Multi-source data acquisition module: used to collect multi-source data of patients, including text data, image data, test data and questionnaire data; the text data includes medical records and diagnosis reports; the image data includes X-ray, CT and MRI images; the test data includes blood and urine test indicators; the questionnaire data includes symptom assessment and lifestyle information filled out by patients; Multimodal disease recognition model construction module: used to construct a multimodal disease recognition model, extract data features from the text data, image data, test data and questionnaire data collected by the multi-source data acquisition module, and realize disease recognition; Knowledge network construction module: used to construct a disease-symptom-examination knowledge network, which includes the associations between diseases, symptoms and examination items, and assigns a weight to each association, which is used to indicate the closeness of the association; Multi-dimensional matching algorithm design module: used to build a multi-dimensional matching algorithm based on the subjects of doctors registered in the system, resource load and geographical distance; The multi-dimensional matching algorithm calculates the matching scores between the disease identification results and the doctors according to a preset weight formula, arranges the matching scores in descending order, and sends the multi-source data to the doctor with the highest matching score.

2. The artificial intelligence-based online disease data matching system according to claim 1, characterized in that: The text data acquisition tool in the multi-source data acquisition module uses natural language processing technology to collect and organize text data; the image data acquisition tool connects to the hospital imaging equipment and performs image preprocessing; the test data acquisition tool obtains test indicator data from the laboratory information system; the questionnaire data acquisition tool collects questionnaire information through the online questionnaire platform and verifies its validity.

3. The artificial intelligence-based online disease data matching system according to claim 1, characterized in that: The multimodal disease recognition model construction module adopts deep learning technology, in which convolutional neural networks are used to process image data, long short-term memory networks are used to process time series information in text data, and multi-layer perceptrons are used to process test data and questionnaire data, and the features of each modality data are deeply fused through a feature fusion network.

4. The artificial intelligence-based online disease data matching system according to claim 1, characterized in that: In the multi-dimensional matching algorithm design module, the subjects that doctors are good at are profiled by analyzing the doctor's historical diagnosis and treatment data and professional certification information; Resource load is calculated based on the doctor's current number of patients and schedule; geographical distance is measured based on the latitude and longitude information of the patient and doctor's locations, and the multi-dimensional matching algorithm quantitatively matches disease identification results with doctor resources according to a preset weight formula; The multidimensional matching algorithm design module includes a distance factor calculation submodule, a disease expertise recursive submodule and a dynamic weight controller, wherein the distance factor calculation submodule is used to convert real-time traffic time into spatial equivalent distance during peak traffic hours; the disease expertise recursive submodule is used to calculate the maximum similarity of neighboring diseases within a relevant range in the knowledge network when the user's current disease is not in the doctor's expertise set.

5. The artificial intelligence-based online disease data matching system according to claim 4, characterized in that: The weight formula is as follows: ； Among them, α, β and γ are adjustable weight coefficients, d is the real-time road distance, L is the doctor resource load rate, where the doctor resource load rate = the current number of cases to be treated / the average daily reception capacity, and E is the disease specialty matching degree. The dynamic weight controller reduces the γ value weight when the rejection rate is greater than 15%.

6. The artificial intelligence-based online disease data matching system according to claim 4, characterized in that: The dynamic weight controller includes a rejection rate counter register, which is used to count the events in which doctors reject matching according to actual consultation situations, wherein the weight coefficient is normalized by a hardware divider at 0:00 every day.

7. The artificial intelligence-based online disease data matching system according to claim 1, characterized in that: It also includes a diagnosis and treatment plan generation module: based on the matched doctor, the doctor uses online historical diagnosis and treatment experience and disease-symptom-examination knowledge network data, combined with the patient's detailed condition data, to generate preliminary diagnosis and treatment results and determine whether to accept the patient; Data encryption transmission module: used to encrypt the collected disease data using encryption algorithms, and adopt secure communication protocols during data transmission to ensure data security.

8. The artificial intelligence-based online disease data matching system according to claim 1, characterized in that: It also includes a feedback and optimization module: used to collect feedback information from patients and doctors, and optimize the multimodal disease recognition model and multidimensional matching algorithm based on the feedback information, and update the disease-symptom-examination knowledge network in real time.

9. The artificial intelligence-based online disease data matching system according to claim 7, characterized in that: The data encryption transmission module uses the AES algorithm to encrypt the disease data and uses the SSL / TLS protocol for secure communication during data transmission to prevent the data from being stolen or tampered with.

10. The artificial intelligence-based online disease data matching system according to claim 8, characterized in that: The feedback and optimization module continuously optimizes the multimodal disease recognition model and the multidimensional matching algorithm through machine learning algorithms, and regularly extracts new knowledge from authoritative medical knowledge bases and expert experience to update the disease-symptom-examination knowledge network.

Images