Medical system working method and device based on large model and knowledge graph

Abstract

The embodiment of the invention provides a medical system working method and device based on a large model and a knowledge graph. The method is applied to information medical field. The method comprises the following steps: searching contents related to patient chief complaint information from a constructed medical knowledge graph by utilizing a large model, analyzing the patient chief complaint information based on the searched related contents, identifying a patient disease, and giving a corresponding doctor-seeing department; the distance between the hospital and the patient position is smaller than a preset range, and the hospital comprises a doctor-seeing department, and number source data of the target hospital are obtained in real time; according to the number source data and the evaluation factors of the doctors in the doctor-seeing departments in the hospitals, a doctor recommendation result is generated through a large model, and constraint examination is conducted on the doctor recommendation result; and after the constraint examination is qualified, performing multiple rounds of inquiry on the patient according to the patient chief complaint information by utilizing the large model, generating a pre-inquiry report and sending the pre-inquiry report to the patient or a doctor in a recommendation result. Therefore, the doctor recommendation accuracy and the doctor seeing efficiency are improved, and the doctor seeing pressure of a hospital is relieved.

Description

Technical Field

[0001] The present disclosure relates to the field of information medicine, and in particular to a medical system working method and device based on a large model and a knowledge graph. Background Art

[0002] In the traditional hospital treatment model, patients face many challenges. First, due to their lack of medical knowledge, patients often find it difficult to determine which department and doctor they should visit. For example, chest pain may be caused by heart or digestive system diseases, which are difficult for ordinary patients to distinguish. In actual treatment, about 30% of patients choose the wrong department or doctor, resulting in delayed treatment, increased time and economic costs; secondly, hospital medical resources are generally tight. In addition, the uneven allocation of medical resources causes patients to flock to large hospitals regardless of the severity of their condition, further exacerbating the pressure of treatment.

[0003] Therefore, how to accurately recommend appropriate departments and doctors to patients, reduce misdiagnosis and disease delays, and improve medical efficiency is an urgent problem that needs to be solved. Summary of the Invention

[0004] The present disclosure provides a medical system working method, apparatus, equipment and storage medium based on a large model and knowledge graph.

[0005] According to a first aspect of the present disclosure, a method for operating a medical system based on a large model and a knowledge graph is provided. The method comprises: Use the big model to search for content related to the patient's chief complaint from the constructed medical knowledge graph, analyze the patient's chief complaint based on the relevant content found, identify the patient's disease, and provide the corresponding medical department; The hospital that is within a preset range from the patient's location and includes the department being treated is selected as the target hospital, and the number source data of the target hospital is obtained in real time; Based on the number source data and the evaluation factors of each doctor in the visiting department of each hospital, a large model is used to generate doctor recommendation results and a constraint check is performed on the doctor recommendation results; After the constraint examination is passed, the patient is questioned multiple times based on the patient's chief complaint using a large model, and a pre-diagnosis report is generated and sent to the patient or the doctor in the recommended results.

[0006] In some implementations of the first aspect, the medical knowledge graph includes: Medical entities and the relationships between them; wherein the medical entities include clinics, doctors, diseases, disease symptoms, treatment methods, examination items, and drugs; The relationship between the medical entities is expressed in the form of triples.

[0007] In some implementations of the first aspect, generating doctor recommendation results using a large model based on the number source data and the evaluation factors of each doctor in the visiting department of each hospital includes: Based on the multiple recommendation indicators set, combined with the number source data and the evaluation factors of each doctor in the visiting department of each hospital, a large model is used to generate doctor recommendation results; wherein, The recommendation indicators include a reinforcement learning composite reward function, each doctor's historical registration success rate, and the patient's historical preference matching degree.

[0008] In some implementations of the first aspect, the constraint checking includes: The matching between the doctor's qualifications and the disease treatment scope in the recommended results is checked from the medical knowledge graph. If they match, the KL divergence is used to further constrain the recommended results so that the recommended results meet the recommended indicators.

[0009] In some implementations of the first aspect, the method further includes: If the doctor's qualifications in the recommended results do not match the disease treatment scope, the medical knowledge graph will be used to retrieve doctors who meet the qualifications in the treatment department of the target hospital, and the big model will be used to sort the retrieved doctors and regenerate the doctor recommendation results.

[0010] In some implementations of the first aspect, the method further includes: For patients without historical medical records, historical medical records of similar patients were obtained based on the patient's basic information; A large model is used to analyze the historical medical records of similar patient groups, as well as the popularity ranking of departments and doctors corresponding to patient diseases obtained from the medical knowledge graph, to initially generate doctor recommendation results.

[0011] In some implementations of the first aspect, the method further includes: If the constraint check fails, a doctor subgraph that meets the preset conditions is retrieved from the medical knowledge graph, candidate doctors are selected from the doctor subgraph, and the candidate doctors are scored according to the comprehensive scoring function. The doctor with the highest score among all the scores is used as the final recommended doctor.

[0012] According to a second aspect of the present disclosure, a working device for a medical system based on a large model and a knowledge graph is provided. The device includes: The department determination module is used to use the big model to search for content related to the patient's chief complaint information from the constructed medical knowledge graph, analyze the patient's chief complaint information based on the relevant content found, identify the patient's disease, and provide the corresponding treatment department; A module for obtaining number source data in real time is used to select a hospital that is less than a preset range from the patient's location and contains the department being treated as a target hospital, and obtain the number source data of the target hospital in real time; A doctor recommendation result acquisition module is used to generate doctor recommendation results using a large model based on the number source data and the evaluation factors of each doctor in the consulting department of each hospital, and to perform constraint checks on the doctor recommendation results; The pre-diagnosis report generation module is used to conduct multiple rounds of inquiries to the patient based on the patient's chief complaint information using a large model after the patient has passed the constraint examination, and generate a pre-diagnosis report to be sent to the patient or the doctor in the recommended results.

[0013] According to a third aspect of the present disclosure, an electronic device is provided. The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0014] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, where the computer instructions are used to cause a computer to execute the method described above.

[0015] In the present disclosure, a medical system working method and device based on a large model and a knowledge graph are provided. The method is specifically as follows: using a large model to search for content related to the patient's chief complaint information from the constructed medical knowledge graph, and analyzing the patient's chief complaint information based on the relevant content found, identifying the patient's disease, and providing the corresponding treatment department; taking the hospital that is less than a preset range from the patient's location and contains the treatment department as the target hospital, and obtaining the number source data of the target hospital in real time; based on the number source data and the evaluation factors of each doctor in the treatment department in each hospital, using the large model to generate a doctor recommendation result and perform a constraint check on the doctor recommendation result; after the constraint check is qualified, using the large model to conduct multiple rounds of inquiries to the patient based on the patient's chief complaint information, generate a pre-diagnosis report and send it to the patient or the doctor in the recommended result. In this way, the accuracy of doctor recommendations and medical efficiency are improved, and the treatment pressure of tertiary hospitals is reduced.

[0016] It should be understood that the contents described in the Summary of the Invention section are not intended to limit the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which: Figure 1 A flowchart of a medical system working method based on a large model and a knowledge graph provided by an embodiment of the present disclosure is shown; Figure 2 A schematic diagram of doctor recommendation results provided by an embodiment of the present disclosure is shown; Figure 3 A schematic diagram of a multi-round query process based on a large model provided by an embodiment of the present disclosure is shown; Figure 4 A diagram showing a working device structure of a medical system based on a large model and a knowledge graph provided by an embodiment of the present disclosure is shown; Figure 5 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0018] To make the purpose, technical solutions, and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present disclosure.

[0019] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0020] In response to the problems arising from the background technology, the embodiments of the present disclosure provide a medical system working method and device based on a large model and a knowledge graph. The method is specifically as follows: using a large model to search for content related to the patient's chief complaint information from the constructed medical knowledge graph, and analyzing the patient's chief complaint information based on the relevant content found, identifying the patient's disease, and providing the corresponding treatment department; taking the hospital that is less than a preset range from the patient's location and contains the treatment department as the target hospital, and obtaining the number source data of the target hospital in real time; based on the number source data and the evaluation factors of each doctor in the treatment department of each hospital, using the large model to generate a doctor recommendation result and perform a constraint check on the doctor recommendation result; after the constraint check is qualified, using the large model to conduct multiple rounds of inquiries to the patient based on the patient's chief complaint information, generate a pre-diagnosis report and send it to the patient or the doctor in the recommended result. In this way, the accuracy of doctor recommendations and medical efficiency are improved, and the treatment pressure of tertiary hospitals is reduced.

[0021] The following describes in detail the working method and device of the medical system based on the big model and knowledge graph provided by the embodiment of the present disclosure through specific embodiments in combination with the accompanying drawings.

[0022] Figure 1 A flowchart of a medical system working method based on a large model and a knowledge graph provided by an embodiment of the present disclosure is shown. The method 100 includes the following steps: S110, using the big model to search for content related to the patient's chief complaint information from the constructed medical knowledge graph, and based on the relevant content found, analyze the patient's chief complaint information, identify the patient's disease, and provide the corresponding medical department.

[0023] In some embodiments, the big model constructs a dynamic context by retrieving data from the medical knowledge graph in real time, analyzes the patient's intention according to the patient's chief complaint information, and searches the medical knowledge graph for content related to the patient's chief complaint information based on the analyzed intention; for example, if the patient's chief complaint information is fever and stomach pain, then the relevant content found by the big model from the medical knowledge graph may include gastroenteritis causing fever and abdominal pain, influenza causing fever, food poisoning causing abdominal pain, and the treatment department can be selected from internal medicine or emergency department, etc.

[0024] In some embodiments, the medical knowledge graph includes: Medical entities and the relationships between them; wherein the medical entities include clinics, doctors, diseases, disease symptoms, treatment methods, examination items, and drugs; The relationship between the medical entities is expressed in triple form, wherein the triple form is expressed as <entity 1, relationship, entity 2>, for example, <department, doctors included in the department, doctors>, <disease, symptoms of the disease, symptoms of the disease>; Furthermore, Table 1 shows the information contained in the disease entity, Table 2 shows the information contained in the department entity, and Table 3 shows the information contained in the doctor entity: Table 1 Table 2 Table 3 In some embodiments, the big model in S110 is a trained big model. During the training stage of the big model, before the training samples are input into the big model, the training samples need to be cleaned and preprocessed first. The preprocessing includes deleting duplicate values, null values, abnormal data, normalizing or standardizing numerical data, segmenting text fields, and removing stop words to ensure data integrity, improve model generalization capabilities, and adapt to model training needs. The big model is trained according to the training samples so that the big model can find content related to the patient's chief complaint information from the constructed medical knowledge graph through knowledge retrieval and context understanding methods; further, the input of the big model includes the patient's chief complaint information.

[0025] S120, taking a hospital that is less than a preset range from the patient's location and includes the treatment department as a target hospital, and obtaining the number source data of the target hospital in real time.

[0026] In some embodiments, the patient's real-time location, the addresses of hospitals around the patient, and the real-time traffic conditions are taken into consideration, and a hospital that is less than a preset range from the patient's location and includes the treatment department is selected as the target hospital, and doctors in the corresponding department of the target hospital are recommended to the patient. This can reduce the pressure on tertiary hospitals, and patients with mild symptoms can receive treatment faster, reducing delays in the disease.

[0027] In some embodiments, the target hospital's appointment data includes the available appointment time periods, the number of remaining appointments, and emergency appointment situations for each doctor in the department that the patient needs to see in the hospital. Combined with the real-time acquired appointment data, it can reduce information delays, improve patient appointment efficiency, and reduce the time taken for patient appointments.

[0028] S130, based on the number source data and the evaluation factors of each doctor in the consultation department of each hospital, a large model is used to generate a doctor recommendation result and a constraint check is performed on the doctor recommendation result.

[0029] In some embodiments, the evaluation factor of each doctor is determined based on historical patient evaluations of each doctor in the department that patients need to visit in the target hospital. The evaluation factor can be comprehensively determined based on factors such as the doctor's professionalism, recent patient reception volume, and treatment effect. The higher the evaluation factor, the more popular the doctor is. Based on the target hospital's number of appointments data and the evaluation factors of each doctor in the department that patients need to visit in the target hospital, a large model is used to generate doctor recommendation results. Comprehensive consideration of the doctor's evaluation factors can effectively reduce the risk of excellent doctors being buried.

[0030] In some embodiments, generating doctor recommendation results using a large model based on the number source data and the evaluation factors of each doctor in the visiting department of each hospital includes: Based on the multiple recommendation indicators set, combined with the number source data and the evaluation factors of each doctor in the visiting department of each hospital, a large model is used to generate doctor recommendation results; wherein, The recommendation indicators include a reinforcement learning composite reward function, each doctor's historical registration success rate, and the patient's historical preference matching degree.

[0031] For example, there are three doctors in the department that the patient needs to see in target hospital A, namely doctor 1, doctor 2, and doctor 3. There are two doctors in the department that the patient needs to see in target hospital B, namely doctor 4 and doctor 5. The evaluation factors of these five doctors are ranked from high to low as doctor 4, doctor 1, doctor 2, doctor 3, and doctor 5. However, the appointment data of target hospital B shows that doctor 4 has no appointment on that day, doctor 1 has an appointment, but the appointment is tight and doctor 1 has the lowest historical registration success rate. Doctor 2 also has an appointment on that day, and has a high historical registration success rate, which is more in line with the patient's historical preferences. The reinforcement learning compound reward function value is optimal. At this time, the large model will recommend doctor 2 as the doctor result and suggest that the patient go to target hospital A to make an appointment with doctor 2. Among them, the reinforcement learning compound reward function integrates the distance between the doctor's hospital and the patient, the matching degree between the doctor's specialty and the disease to be treated by the patient, etc.

[0032] In some embodiments, the method 100 further includes: For patients without historical medical records, historical medical records of similar patients were obtained based on the patient's basic information; A large model is used to analyze the historical medical records of similar patient groups, as well as the popularity ranking of departments and doctors corresponding to patients' diseases obtained from the medical knowledge graph, to initially generate doctor recommendation results; For example, patient A has no historical medical record in target hospital A. Patient A is of the same age and gender as patient B, who has a medical record in target hospital A. Patient A and patient B are divided into the same group. According to patient B's historical medical record in target hospital A, it can be seen that patient B has seen doctor 6 in the department where patient A needs to see a doctor, and doctor 6 appears frequently in this same group, or doctor 6 is very popular in this same group. In this case, doctor 6 can be recommended to patient A.

[0033] In some embodiments, the doctor recommendation result needs to be associated with at least one entity in the medical knowledge graph, which can improve the compliance of medical treatment; The doctor recommendation results include the recommended target hospital, the department the patient needs to visit and the corresponding proportion, and the doctor ranking; Figure 2 As shown, the patient's main complaint mainly describes heart problems, so the large model gives Figure 2 The doctor recommendation results shown in the figure show that the recommended departments are cardiology, with a recommendation probability of 73.7%, endocrinology, with a recommendation probability of 15.8%, and psychiatry, with a recommendation probability of 10.5%. The doctor recommended by the cardiology department is Dr. Zhang, who belongs to the heart center. The doctor recommended by the endocrinology department is Dr. Xu, who belongs to the comprehensive internal medicine center. The doctor recommended by the psychiatry department is Dr. Ding, who belongs to the comprehensive internal medicine center. This shows that the patient's best choice is Dr. Zhang from the cardiology department.

[0034] In some embodiments, the constraint checking includes: The matching between the doctor's qualifications and the disease treatment scope in the recommended results is checked from the medical knowledge graph. If they match, the KL divergence is used to further constrain the recommended results so that the recommended results meet the recommended indicators.

[0035] In some embodiments, if the qualifications of the doctor in the recommendation results do not match the scope of disease treatment, a qualified doctor in the treatment department of the target hospital is retrieved from the medical knowledge graph, and the retrieved doctors are ranked using a large model to regenerate the doctor recommendation results, that is, the doctor with the highest ranking in the ranking result is the doctor finally recommended to the patient.

[0036] In some embodiments, the KL divergence is used to further perform constraint checking on the recommendation results, including: According to the recommendation method given in method 100 and the reference recommendation strategy, a KL divergence regularization term is constructed; Use KL divergence regularization term to constrain recommendation results; Furthermore, the KL divergence regularization term The calculation formula is as follows: ; in, 、 Respectively represent the recommendation method and reference recommendation strategy given by method 100, 、 Represent the CLIP loss function and KL divergence respectively, are the parameters of the large model, is the regularization coefficient, which represents the weight of the KL divergence. By adjusting the regularization coefficient, the recommendation method given by method 100 can be prevented from over-optimizing short-term rewards (such as user click rate) while ignoring medical safety (such as doctor qualifications).

[0037] In some embodiments, Calculated by the following formula: ; in, 、 、 They represent the advantage function, shear hyperparameter, and the action probability ratio of the new and old recommendation strategies respectively. The advantage function is used to measure the pros and cons of an action relative to the average performance. The shear hyperparameter is used to limit the update amplitude of the recommendation strategy. The action probability ratio of the new and old recommendation strategies is used to indicate the update amplitude of the recommendation strategy. The large model will continuously update the recommendation strategy in the process of giving the final recommendation result. Represents the time step t expectations, Indicates that The value is limited to between; Furthermore, Calculated by the following formula: ; in, 、 denote the actions measured by the advantage function and the average performance of the actions in the corresponding recommendation strategy, 、 Represent the new recommendation strategy and the old recommendation strategy respectively.

[0038] In some embodiments, constraint checking further includes: According to the similarity between the doctor recommendation result given by the method 100 and the doctor recommendation result obtained based on the medical knowledge graph, the doctor recommendation result given by the method 100 is constrained; Specifically, based on this method, if the KL divergence If the threshold is exceeded, it means that the recommendation result given by method 100 deviates too much from the recommendation result obtained based on the medical knowledge graph, and it is necessary to re-search the medical knowledge graph or reduce the ranking weight of the authoritative doctors in the medical knowledge graph to maintain the consistency of the recommendation result given by method 100 and the recommendation result obtained based on the medical knowledge graph. The formula is as follows: ; in, 、 They represent the probability of recommending doctors based on the medical knowledge graph and the probability of recommending doctors given by method 100, respectively. 、 Respectively represent the first i The probability of a doctor and the method 100 give the first i The probability of a doctor.

[0039] In some embodiments, KL divergence can be used to balance patient preferences and basic medical knowledge. The balancing method is as follows: ; in, 、 represent basic medical knowledge and patient preferences, is the weight coefficient, 、 They represent the loss function of DPO and the loss function considering DPO and reference recommendation strategies, respectively. This balancing method can prevent large models from overfitting patient preferences (for example, frequently selecting non-authoritative doctors) and violating medical knowledge.

[0040] In some embodiments, The calculation formula is as follows: ; in, represents the Sigmoid function, represents the scaling factor, which is used to control the alignment strength. 、 represent the doctors selected by the patient and the doctors rejected by the patient, respectively. represents the input of the large model, Represents the dataset M The triples collected in expected value; According to this formula, the big model can automatically deduce user preferences from their behavior without manually setting the reward function. When the doctor's practice information in the medical knowledge graph changes, the big model is updated synchronously through fine-tuning. , thus ensuring that preference learning is always based on the latest medical knowledge graph and realizing the self-evolution function of the model.

[0041] In some embodiments, according to the above formulas, the reinforcement learning compound reward function can be The calculation formula is expressed as: ; in, 、 、 、 They represent the reward function for the patient's successful registration, the reward function for the distance between the hospital and the patient, the reward function for the recommended doctor's qualifications, and the reward function for matching user preferences. 、 、 、 Represent the coefficients corresponding to each reward function; Furthermore, if the patient completes the registration, =+1.0, if the distance between the hospital and the patient is d ,but If the recommended doctor's qualifications are good, then =+0.5, .

[0042] In some embodiments, the generalized advantage estimate Optimize the large model, the optimization formula is as follows: ; in, 、 、 They are the discount factor (usually 0.9-0.99), GAE hyperparameter (usually 0.95), and the distance from the current time step to the future time step. is the time step t The timing differential error at is calculated as follows: ; in, 、 、 They are t The reward value of the moment, t Recommended results at all times S long-term expected returns, t +1 moment recommendation result S long-term expected returns.

[0043] S140, after the constraint examination is passed, the patient is questioned multiple times based on the patient's main complaint information using the large model, and a pre-diagnosis report is generated and sent to the patient or the doctor in the recommended results.

[0044] In some embodiments, the method 100 further includes: If the constraint check fails, a doctor subgraph that meets the preset conditions is retrieved from the medical knowledge graph, candidate doctors are selected from the doctor subgraph, and the candidate doctors are scored according to the comprehensive scoring function. The doctor with the highest score among all the scores is selected as the final recommended doctor; For example, when the constraint check fails, the satisfied domain (the area in which doctor N is proficient) is retrieved from the medical knowledge graph. (Patient's inquiry needs) (Distance between patient and doctor) Doctor subgraph, select candidate doctors from the doctor subgraph, score the candidate doctors according to the comprehensive scoring function, and use the doctor with the highest score as the final recommended doctor; where the comprehensive scoring function The calculation formula is as follows: ; in, 、 、 They are the normalization operation of the evaluation factor of doctor N, the distance between the patient and the doctor, and the number source status of doctor N. 、 、 are the coefficients, .

[0045] In some embodiments, the large model has multiple rounds of querying capabilities, such as Figure 3 As shown in the figure, a patient with user ID 123456 inputs "I have a headache today" into the big model. The big model conducts multiple rounds of inquiries based on the patient's input. The question-and-answer process is as follows: Q: When did the headache start today? Is it accompanied by nausea or blurred vision? Answer: This morning, with nausea. Q: Did your nausea this morning start suddenly or gradually worsen? Was it accompanied by vomiting or other symptoms? Answer: It started suddenly, with vomiting. Q: When did the nausea and vomiting start? Is it continuous or intermittent? Answer: 9:00 a.m., intermittent. Q: What is the interval between nausea and vomiting? Are there any episodes after 9:00 a.m.? From the above dialogue, we can see that the big model is imitating a doctor's understanding of the patient's condition. Based on these inquiries, the big model can generate a detailed pre-diagnosis report for the patient to review and recommend to the recommended doctor for diagnosis, improving treatment efficiency. The pre-diagnosis report includes the patient's chief complaint, medical history, allergy history, family history, and other information.

[0046] In some embodiments, the big model can parse the pictures uploaded by the patient, extract the text information in the pictures, and conduct in-depth analysis of the text information. Combined with the medical knowledge graph, the text information is structured to generate corresponding analysis results. For example, the big model parses the report uploaded by the patient to obtain text content, which covers the parsing process, interpretation of cost details, and analysis of inspection indicators. For example, the parsing process is: carefully read and understand the various information in the bill, and identify key data points, such as cost items, amounts, payment methods, etc.; the interpretation of cost details can be: treatment fee, 9.10 yuan, non-invasive ECG monitoring, 10.00 yuan for 1 hour, intravenous infusion, 1 time, 9.10 yuan, total cost 28.20 yuan, but the total in the bill is 2110.55 yuan, which is obviously inconsistent with the calculation, and there may be information omissions or misunderstandings; the payment method is: medical insurance payment, 10.55 yuan; these analysis results are sent to doctors to help doctors make medical decisions.

[0047] In some embodiments, the patient can also interact with the big model as follows. After the patient selects the disease they want to search for, the big model quickly searches for relevant information about the disease from the medical knowledge graph, including disease introduction, cause, treatment, prevention, what examinations should be done, etc., and integrates these relevant information to generate a disease encyclopedia. For example, the disease encyclopedia may include disease introduction, cause, examination items, disease name, preventive measures, etc. For example, the disease introduction may be: Whooping cough is an acute respiratory infectious disease caused by Bordetella pertussis, characterized by paroxysmal spasmodic cough, accompanied by infectious cockroach-like inspiratory roar at the end of the coughing period, with a long course of disease, which can last for several weeks or even about 3 months, so it is called whooping cough. It is more common in children under 5 years old. The clinical phase of whooping cough has complications such as asphyxia, pneumonia, and encephalopathy, with a high mortality rate. Patients caused by Bordetella pertussis have a high infection rate and male carriers have a paroxysmal spasmodic cough and are most contagious during the coughing period of 2-3 weeks. The course of whooping cough is shorter than the catarrhal stage, and the typical course is divided into 3 Stages: (1) Catarrhal stage, cough appears from the onset of the disease, usually 1-2 weeks; (2) Spasmodic cough stage, usually 2-4 weeks or longer, characterized by paroxysmal spasmodic cough;(3) Recovery period, generally 1-2 weeks, during which the frequency and severity of cough attacks decreases and paroxysmal coughs no longer occur. The peripheral blood leukocyte count generally increases significantly, with lymphocytes being the main type. When diagnosing this disease, attention should be paid to differentiating it from bronchial foreign bodies and hilar lymphadenopathy. In recent years, the incidence of this disease in infants and adults has been increasing. The reasons may be: whooping cough is caused by Bordetella pertussis, which belongs to the genus Bordetella. There are four types of Bordetella in the genus Bordetella. In addition to Bordetella pertussis, there are Bordetella parapertussis, Bordetella bronchiseptica and Bordetella avium. Bordetella avium generally does not cause human disease and only causes infection in birds. Bordetella pertussis is about 1.0-1.5μm long and about 0.3-0.5μm wide. It has a membrane and cannot move. It is Gram-negative, aerobic, non-toxic, and has no flagella. It stains darker at both ends with toluidine blue staining. Bacterial culture requires a large amount (15%-25%) of fresh blood to reproduce well, so the colonies are often isolated using Ao-King medium (ie blood, glycerol, potato). Bordetella pertussis grows slowly. After growing in a suitable environment at 35-37℃ for 3-7 days, a small, opaque colony grows. The initial colony is raised and smooth, and is a smooth (S) type, also known as I Phase IV bacteria have the same morphology, have capsules, strong virulence and antigenicity, and high pathogenicity. If the isolated colonies are continued to be cultured in ordinary culture medium, the colonies will change from smooth type to rough (R) type, which is called phase IV bacteria. They have no capsule, lose virulence and antigenicity, and lose pathogenicity. Phase II and phase III are intermediate transitional types. Bordetella pertussis can produce many toxins, and there are five known toxins: (1) Pertussis exotoxin, a protein in Bordetella pertussis cells, formerly known as leukocytosis or lymphocyte promoting factor (LPE), histamin sensitizing factor (HSF), and insulin activating protein (IAP). Pertussis exotoxin has its non-promoting subunit (52-55S) and subunit (52-55S) as non-toxic protein components, which can bind to the host cell membrane and mediate toxic effects through the enzymatically active subunit S1. S1 can catalyze the separation of part of the ribose from nicotinamide adenine dinucleotide through the activity of adenosine diphosphate-ribosyltransferase, transfer it to the cell membrane, inhibit the binding of guanosine triphosphate to the G protein unit, and lead to cell degeneration. At the same time, it can also promote the increase of lymphocytes, activate glandular islet cells and enhance immune response; (2) Heat-resistant endotoxin (ET) can only be partially destroyed at 100°C for 60 minutes and can only be inactivated at 180°C. This toxin can cause fever and immunity in the body;(3) Heat-labile toxin. This toxin can destroy its toxic effect after heating at 55℃ for 30 minutes. Antibodies to this toxin have no protective effect against Bordetella pertussis infection. (4) Tracheal cytotoxin: It can damage the ciliated epithelial cells of the host respiratory tract, causing them to degenerate and necrotize. (5) Adenylate cyclase toxin: It is an enzyme present on the cell surface of Bordetella pertussis. After entering the blood cells, this enzyme can be activated by calcium protein, catalyzing the production of cyclic adenosine monophosphate, interfering with phagocytosis, and inhibiting the chemotaxis of neutrophils and the bactericidal ability of phagocytes, allowing them to persist in infection. Adenylate cyclase toxin is also a hemolysin that can cause hemolysis. The parietal antigens of pertussis are two blood cell active antigens of the pertussis bacteria, one of which is filamentous hemagglutinin (FHA). Because it comes from the hairs on the surface of the bacteria, it is also called fimbriae antigen. Filamentous hemagglutinin plays a decisive role in the process of Bordetella pertussis adhering to respiratory epithelial cells and is the main cause of disease. Experiments have found that mice immunized with filamentous hemagglutinin can resist the lethal attack of Bordetella pertussis. Therefore, filamentous hemagglutinin is a protective antigen. Another agglutinogen (AGG) is a protein component in the outer membrane and fimbriae of Bordetella pertussis, mainly containing three serotype coagulation factors 1, 2, and 3. AGG-1 is species-specific; AGG-2 and AGG-3 are type-specific. The local epidemic situation can be understood by detecting the type of agglutinogen. It is currently believed that the corresponding antibodies of these two heme membrane antigens are protective antibodies. Bordetella pertussis is divided into seven agglutinogens according to the different antigenicity of the agglutinogens. Type 1 agglutinogen is possessed by all Bordetella pertussis, type 7 agglutinogen is shared by Bordetella (including Bordetella parapertussis and Bordetella bronchiseptica), and type 2-6 Bordetella pertussis is divided into different serotypes based on different combinations. The main purpose of serotype determination is to study the serotype of prevalent strains and select special serotype strains for vaccine production. In addition, Bordetella parapertussis and Bordetella pertussis have no cross-immunity and can also cause epidemics. Bordetella pertussis is very sensitive to external physical and chemical factors. After the bacteria enter the respiratory tract of susceptible people with air droplets, the bacterial filamentous hemoglobin adheres to the surface of the ciliated epithelial cells of the throat to the bronchiolar mucosa. Subsequently, the bacteria multiply locally and produce a variety of toxins such as pertussis exotoxin and adenylate cyclase, which cause ciliary paralysis and cell degeneration of the epithelial cells, reduce their protein synthesis, and cause epithelial cell necrosis and shedding, resulting in systemic reactions. Due to the occurrence of epithelial cell lesions and ciliary paralysis, mucus and necrotic epithelium accumulate in the small bronchi, and the recovery period will also be affected by continuous stimulation of the respiratory nerve input, which is transmitted to the cough center in the cerebral cortex and medulla oblongata, causing reflex spasmodic cough. During the recovery period, crying or other infections may also induce whooping cough-like spasms. Examination items may include: white blood cell differential count, blood routine test, ear, nose and throat swab bacterial culture, peripheral blood white blood cell count and differential test. The disease name may be "pertussis".

[0048] According to an embodiment of the present disclosure, a medical system working method and device based on a large model and a knowledge graph are provided. The method is specifically as follows: using a large model to search for content related to the patient's chief complaint information from the constructed medical knowledge graph, and analyzing the patient's chief complaint information based on the relevant content found, identifying the patient's disease, and providing the corresponding treatment department; taking a hospital that is less than a preset range from the patient's location and contains the treatment department as the target hospital, and obtaining the number source data of the target hospital in real time; based on the number source data and the evaluation factors of each doctor in the treatment department in each hospital, using a large model to generate a doctor recommendation result and perform a constraint check on the doctor recommendation result; after the constraint check is qualified, using the large model to conduct multiple rounds of inquiries to the patient based on the patient's chief complaint information, generate a pre-diagnosis report and send it to the patient or the doctor in the recommended result. In this way, the accuracy of doctor recommendations and medical efficiency are improved, and the treatment pressure of tertiary hospitals is reduced.

[0049] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present disclosure is not limited by the order of the actions described, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present disclosure.

[0050] The above is an introduction to the method embodiment. The following is a further explanation of the solution disclosed in the present disclosure through an apparatus embodiment.

[0051] Figure 4 The following diagram shows a working device structure of a medical system based on a large model and a knowledge graph provided by an embodiment of the present disclosure. The device 400 includes: The department determination module 410 is used to use the big model to search for content related to the patient's chief complaint information from the constructed medical knowledge graph, and analyze the patient's chief complaint information based on the relevant content found, identify the patient's disease, and provide the corresponding treatment department.

[0052] The number source data real-time acquisition module 420 is used to select a hospital that is less than a preset range from the patient's location and includes the treatment department as a target hospital, and to acquire the number source data of the target hospital in real time.

[0053] The doctor recommendation result acquisition module 430 is used to generate doctor recommendation results using a large model based on the number source data and the evaluation factors of each doctor in the consultation department of each hospital, and to perform constraint checks on the doctor recommendation results.

[0054] The pre-diagnosis report generation module 440 is used to conduct multiple rounds of inquiries to the patient based on the patient's main complaint information using a large model after the patient passes the constraint examination, and generate a pre-diagnosis report to be sent to the patient or the doctor in the recommended results.

[0055] In some embodiments, the department determination module 410 is specifically configured to: The medical knowledge graph includes: Medical entities and the relationships between them; wherein the medical entities include clinics, doctors, diseases, disease symptoms, treatment methods, examination items, and drugs; The relationship between the medical entities is expressed in the form of triples.

[0056] In some embodiments, the doctor recommendation result acquisition module 430 is specifically used to: The method of generating doctor recommendation results using a large model based on the number source data and the evaluation factors of each doctor in the visiting department of each hospital includes: Based on the multiple recommendation indicators set, combined with the number source data and the evaluation factors of each doctor in the visiting department of each hospital, a large model is used to generate doctor recommendation results; wherein, The recommendation indicators include a reinforcement learning composite reward function, each doctor's historical registration success rate, and the patient's historical preference matching degree.

[0057] In some embodiments, the doctor recommendation result acquisition module 430 is further configured to: The constraint check includes: The matching between the doctor's qualifications and the disease treatment scope in the recommended results is checked from the medical knowledge graph. If they match, the KL divergence is used to further constrain the recommended results so that the recommended results meet the recommended indicators.

[0058] In some embodiments, the apparatus 400 is further configured to: If the doctor's qualifications in the recommended results do not match the disease treatment scope, the medical knowledge graph will be used to retrieve doctors who meet the qualifications in the treatment department of the target hospital, and the big model will be used to sort the retrieved doctors and regenerate the doctor recommendation results.

[0059] In some embodiments, the apparatus 400 is further configured to: For patients without historical medical records, historical medical records of similar patients were obtained based on the patient's basic information; A large model is used to analyze the historical medical records of similar patient groups, as well as the popularity ranking of departments and doctors corresponding to patient diseases obtained from the medical knowledge graph, to initially generate doctor recommendation results.

[0060] In some embodiments, the apparatus 400 is further configured to: If the constraint check fails, a doctor subgraph that meets the preset conditions is retrieved from the medical knowledge graph, candidate doctors are selected from the doctor subgraph, and the candidate doctors are scored according to the comprehensive scoring function. The doctor with the highest score among all the scores is used as the final recommended doctor.

[0061] It is understandable that Figure 4 Each module / unit in the illustrated device 400 has the function of implementing each step in the method 100 provided in the embodiment of the present disclosure and can achieve its corresponding technical effect. For the sake of brevity, they will not be described in detail here.

[0062] Figure 5 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. Electronic device 500 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 500 may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0063] like Figure 5 As shown, electronic device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. RAM 503 can also store various programs and data required for the operation of electronic device 500. Computing unit 501, ROM 502, and RAM 503 are connected to each other via a bus 504. An I / O interface 505 is also connected to bus 504.

[0064] Multiple components in the electronic device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a magnetic disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the electronic device 500 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0065] Computing unit 501 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 501 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed onto electronic device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by computing unit 501, one or more steps of method 100 described above can be performed. Alternatively, in other embodiments, computing unit 501 can be configured to perform method 100 in any other suitable manner (e.g., via firmware).

[0066] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0067] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0068] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0069] It should be noted that the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute method 100 and achieve the corresponding technical effect achieved by executing the method in the embodiment of the present disclosure. For the sake of concise description, they will not be repeated here.

[0070] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0071] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0072] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0073] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0074] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A medical system working method based on a large model and knowledge graph, characterized in that: include: Use the big model to search for content related to the patient's chief complaint from the constructed medical knowledge graph, analyze the patient's chief complaint based on the relevant content found, identify the patient's disease, and provide the corresponding medical department; The hospital that is within a preset range from the patient's location and includes the department being treated is selected as the target hospital, and the number source data of the target hospital is obtained in real time; Based on the number source data and the evaluation factors of each doctor in the visiting department of each hospital, a large model is used to generate doctor recommendation results and a constraint check is performed on the doctor recommendation results; After the constraint examination is passed, the patient is questioned multiple times based on the patient's chief complaint using a large model, and a pre-diagnosis report is generated and sent to the patient or the doctor in the recommended results.

2. The method according to claim 1, characterized in that The medical knowledge graph includes: Medical entities and the relationships between them; wherein the medical entities include clinics, doctors, diseases, disease symptoms, treatment methods, examination items, and drugs; The relationship between the medical entities is expressed in the form of triples.

3. The method according to claim 1, characterized in that The method of generating doctor recommendation results using a large model based on the number source data and the evaluation factors of each doctor in the visiting department of each hospital includes: Based on the multiple recommendation indicators set, combined with the number source data and the evaluation factors of each doctor in the visiting department of each hospital, a large model is used to generate doctor recommendation results; wherein, The recommendation indicators include a reinforcement learning composite reward function, each doctor's historical registration success rate, and the patient's historical preference matching degree.

4. The method according to claim 3, characterized in that The constraint check includes: The matching between the doctor's qualifications and the disease treatment scope in the recommended results is checked from the medical knowledge graph. If they match, the KL divergence is used to further constrain the recommended results so that the recommended results meet the recommended indicators.

5. The method according to claim 4, characterized in that The method further comprises: If the doctor's qualifications in the recommended results do not match the disease treatment scope, the medical knowledge graph will be used to retrieve doctors who meet the qualifications in the treatment department of the target hospital, and the big model will be used to sort the retrieved doctors and regenerate the doctor recommendation results.

6. The method according to claim 1, characterized in that The method further comprises: For patients without historical medical records, historical medical records of similar patients were obtained based on the patient's basic information; A large model is used to analyze the historical medical records of similar patient groups, as well as the popularity ranking of departments and doctors corresponding to patient diseases obtained from the medical knowledge graph, to initially generate doctor recommendation results.

7. The method according to claim 1, characterized in that The method further comprises: If the constraint check fails, a doctor subgraph that meets the preset conditions is retrieved from the medical knowledge graph, candidate doctors are selected from the doctor subgraph, and the candidate doctors are scored according to the comprehensive scoring function. The doctor with the highest score among all the scores is used as the final recommended doctor.

8. A working device of a medical system based on a large model and a knowledge graph, characterized in that: include: The department determination module is used to use the big model to search for content related to the patient's chief complaint information from the constructed medical knowledge graph, analyze the patient's chief complaint information based on the relevant content found, identify the patient's disease, and provide the corresponding treatment department; A module for obtaining number source data in real time is used to select a hospital that is less than a preset range from the patient's location and contains the department being treated as a target hospital, and obtain the number source data of the target hospital in real time; A doctor recommendation result acquisition module is used to generate doctor recommendation results using a large model based on the number source data and the evaluation factors of each doctor in the consulting department of each hospital, and to perform constraint checks on the doctor recommendation results; The pre-diagnosis report generation module is used to conduct multiple rounds of inquiries to the patient based on the patient's chief complaint information using a large model after the patient has passed the constraint examination, and generate a pre-diagnosis report to be sent to the patient or the doctor in the recommended results.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

Images