Multidisciplinary fusion consultation method and system based on knowledge graph

Abstract

The invention discloses a multidisciplinary fusion consultation method based on a knowledge graph, and relates to the field of fusion consultation, and the method comprises the steps: receiving medical record information and medical resource information of a patient; a patient knowledge graph is constructed, and multiple consultation doctors are matched based on the patient knowledge graph; creating a multidisciplinary consultation conference, and sharing patient medical record information in real time; collecting diagnosis and treatment opinions of all consultation doctors; and generating a target diagnosis and treatment scheme. According to the application, the problems of difficulty in time coordination and inconvenience in doctor-patient convergence in a multidisciplinary fusion consultation process can be effectively solved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of fusion consultation, and in particular to a multidisciplinary fusion consultation method and system based on a knowledge graph. Background Art

[0002] Multidisciplinary consultation refers to the integration of the professional knowledge and experience of experts from different disciplines, and a comprehensive analysis of the patient's condition from multiple dimensions, so as to develop a more scientific, comprehensive and personalized treatment plan. Through multidisciplinary consultation, we can give full play to the advantages of each discipline, avoid the limitations of a single discipline, improve the accuracy of diagnosis and the effectiveness of treatment, and effectively promote modern medicine towards refinement and specialization. However, under the traditional offline consultation model, to achieve on-site centralized consultations by experts from different disciplines, there are not only difficulties in time coordination, but also it is extremely inconvenient for patients in remote locations to travel long distances to medical institutions with multidisciplinary expert resources. Summary of the Invention

[0003] The embodiments of the present application provide a multidisciplinary fusion consultation method and system based on a knowledge graph, which is used to solve the problems of difficulty in time coordination and inconvenience in doctor-patient gathering during the multidisciplinary fusion consultation process.

[0004] To achieve the above objectives, the embodiments of the present application adopt the following technical solutions: In the first aspect, a multidisciplinary fusion consultation method based on a knowledge graph is provided, which includes: Receiving patient medical record information of a target patient and medical resource information of a location where the target patient is located; Build a patient knowledge graph based on the patient's medical record information, and match multiple consulting doctors for the target patient based on the patient knowledge graph; Create a multidisciplinary consultation meeting and grant meeting permissions to all consulting doctors, and share patient medical records with all consulting doctors in real time during the multidisciplinary consultation meeting; After all consulting doctors have completed the consultation based on the patient's medical records, their diagnosis and treatment opinions are collected; After integrating all diagnosis and treatment opinions, an initial diagnosis and treatment plan is generated, and the initial diagnosis and treatment plan is revised based on medical resource information to obtain the target diagnosis and treatment plan.

[0005] Optionally, building a patient knowledge graph based on patient medical record information includes the following steps: Pre-process patient medical record information to obtain standard medical record information; Identify all information entities in standard medical record information based on the pre-trained entity recognition model; Conduct context analysis on all information entities based on standard medical record information, and determine the entity attributes of all information entities and the association relationships between all information entities based on the context analysis results; Construct a patient knowledge graph for the target patient based on all information entities, entity attributes, and association relationships.

[0006] Optionally, matching multiple consulting doctors for a target patient based on the patient knowledge graph includes the following steps: Complete structural matching and semantic matching between the patient knowledge graph and the pre-built medical knowledge graph, and assign multiple related departments to the target patient based on the structural matching results and semantic matching results; Get the doctor information of all related doctors in all related departments; For any associated doctor of any associated department, pre-process the doctor information of the associated doctor; Using the keyword extraction algorithm to extract all diagnosis and treatment keywords from the pre-processed doctor information, and integrating all diagnosis and treatment keywords into a diagnosis and treatment keyword set; Vectorize the diagnosis and treatment keyword set to obtain the diagnosis and treatment key vector; The similarity formula is used to calculate the feature similarity between the diagnosis and treatment key vector and the information entity of the patient knowledge graph; If the feature similarity is greater than or equal to a preset first similarity threshold, the associated doctor will be used as the consulting doctor corresponding to the target patient.

[0007] Optionally, completing structural matching and semantic matching between the patient knowledge graph and the pre-built medical knowledge graph, and assigning multiple associated departments to the target patient based on the structural matching results and the semantic matching results includes the following steps: Divide the patient knowledge graph into several patient subgraphs based on the subgraph partitioning rules; For any patient subgraph, the subgraph entities in the patient subgraph are mapped to the pre-built medical knowledge graph, and multiple medical subgraphs are extracted from the medical knowledge graph based on the mapping results; Complete the patient subgraph based on the medical knowledge graph to obtain the standard patient subgraph; Count the number of nodes and edges in the standard patient subgraph and the entire medical subgraph, and calculate the structural similarity between the standard patient subgraph and the entire medical subgraph based on the statistical results of the number of nodes and edges; Calculate the node similarity and edge similarity between the standard patient subgraph and all medical subgraphs, and combine the node similarity and edge similarity to determine the semantic similarity between the standard patient subgraph and all medical subgraphs; The structural similarity and semantic similarity are integrated to obtain the subgraph similarity between the standard patient subgraph and the entire medical subgraph; The medical subgraphs whose subgraph similarity is greater than a preset second similarity threshold are used as associated subgraphs of the standard patient subgraph; Multiple related departments are traced back based on the positions of all related subgraphs in the medical knowledge graph.

[0008] Optionally, sharing patient medical records with all consulting physicians in real time during a multidisciplinary consultation meeting includes the following steps: Divide patient medical record information into patient text information and patient image information according to information attributes; Encrypting the patient's text information using a text encryption algorithm to obtain encrypted text information; dividing the patient image information into a plurality of patient image blocks; For any patient image block, the image compression algorithm is used to perform lossless compression of the patient image block to obtain multiple compressed image sub-bands; Share encrypted text information with all consulting physicians in real time during multidisciplinary consultations, and progressively share all compressed image subbands in real time.

[0009] Optionally, using an image compression algorithm to perform lossless compression of the patient image block to obtain multiple compressed image subbands includes the following steps: Extract the true pixel value of the patient image block; Predicting pixel prediction values of compressed image blocks using a preset prediction function; Combine the true pixel value and the predicted pixel value to construct a pixel difference matrix; The pixel difference matrix is input into the IWT encoder, and the pixel difference matrix is transformed and decomposed by the IWT encoder to obtain multiple image sub-bands; The RLE coding algorithm is used to complete the coding compression of all image sub-bands to obtain multiple compressed image sub-bands.

[0010] Optionally, progressive real-time sharing of all compressed image sub-bands includes the following steps: Analyze the frequency characteristics of all compressed image sub-bands and determine the transmission priority of all compressed image sub-bands based on the frequency characteristics; All compressed image subbands are gradually shared to the multidisciplinary consultation conference according to transmission priority.

[0011] Optionally, after integrating all diagnosis and treatment opinions, an initial diagnosis and treatment plan is generated, and the initial diagnosis and treatment plan is modified according to the medical resource information to obtain the target diagnosis and treatment plan, which includes the following steps: Integrate all diagnosis and treatment opinions to generate an initial diagnosis and treatment plan; Determine resource shortage information in the target patient's location by combining the initial diagnosis and treatment plan with medical resource information; Generate resource secondment plans based on resource shortage information; Use the resource loan plan to improve the initial diagnosis and treatment plan and obtain the target diagnosis and treatment plan.

[0012] In a second aspect, the present application provides a machine-readable storage medium having stored thereon instructions for enabling a machine to execute a method for multidisciplinary fusion consultation based on a knowledge graph according to any one of the items in the first aspect.

[0013] In a third aspect, the present application provides a multidisciplinary fusion consultation system based on a knowledge graph, including: a memory configured to store instructions; and A processor is configured to call instructions from a memory and, when executing the instructions, is capable of implementing the method for multidisciplinary fusion consultation based on a knowledge graph according to any one of the second aspects.

[0014] Through the above technical solution, a patient knowledge graph is constructed through the received patient medical record information, and the patient knowledge graph is used to match multiple suitable consulting doctors for the target patient, which can not only avoid the limitations of a single discipline perspective and ensure the comprehensiveness of the diagnosis and treatment process, but also improve the accuracy of diagnosis and avoid waste of medical resources. After matching multiple suitable consulting doctors, a multidisciplinary consultation meeting is created. The consulting doctors share patient medical record information in real time through the multidisciplinary consultation meeting and discuss the diagnosis and treatment plan together, aiming to provide the best diagnosis and treatment plan for the target patient and ensure that the target patient can recover as soon as possible. In summary, this application integrates medical resources in different regions through a multidisciplinary fusion consultation model, effectively breaks down disciplinary barriers, improves diagnosis and treatment efficiency, and uses online consultation to enable patients in remote and economically underdeveloped areas to enjoy excellent medical resources. While improving the utilization rate of medical resources, it is also conducive to medical progress in remote and economically underdeveloped areas.

[0015] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a multidisciplinary fusion consultation method based on a knowledge graph provided in an embodiment of the present application; Figure 2 An example diagram of a patient knowledge graph provided in an embodiment of the present application; Figure 3 This is an example diagram of a one-dimensional difference sequence and a two-dimensional matrix structure provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] To make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation methods described herein are only used to illustrate and explain the embodiments of the present application and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0018] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back, etc.), such directional indications are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0019] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the fact that they can be implemented by ordinary technicians in this field. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0020] Figure 1 The following schematically shows a flow chart of a method for multidisciplinary fusion consultation based on a knowledge graph according to an embodiment of the present application. Figure 1 As shown, the embodiment of the present application provides a method for multidisciplinary fusion consultation based on knowledge graph, which may include the following steps: S101. Receive the patient medical record information of the target patient and the medical resource information of the target patient's location.

[0021] In this embodiment, the patient's medical record information includes the patient's basic information, patient's medical information, patient's medication information, patient's examination information and patient's symptom information, etc. The patient's basic information includes the patient's ID, name, gender, age, etc., such as patient A (ID: 001, gender: male, age: 41). The patient's medical information includes the patient's medical time, the patient's registration department, the patient's chief complaint, etc., such as patient B's medical record (2023-10-01, Department of Respiratory Medicine, chief complaint: persistent cough). The patient's medication information includes the name of the drug, the dosage of the drug, etc., such as amoxicillin, three times a day, two tablets at a time. The patient's examination information includes the examination name, examination time and examination results, etc., such as 2023-6-02, chest CT, thickened lung texture.

[0022] Medical resource information includes the target patient's local medical institution information, medical staff information, medical equipment information, etc. Medical institution information includes the name of the medical institution, the nature of the medical institution (such as a Grade 3 hospital), etc. Medical staff information includes the number of medical staff, the composition of medical staff (such as chief physicians, specialist nurses), the scope of practice of medical staff (such as surgery), etc. Medical equipment information includes the name of the equipment (such as 1.5T MRI, 64-row high-end CT), the number of equipment, etc.

[0023] S102. Construct a patient knowledge graph based on the patient's medical record information, and match multiple consulting doctors for the target patient based on the patient knowledge graph.

[0024] In this embodiment, a pre-trained entity recognition model is used to identify all information entities in a patient's medical records, such as disease names, medication names, and examination names. Natural language processing technology is then used to perform contextual analysis on the patient's medical records, obtaining the entity attributes of all information entities and the relationships between them. A patient knowledge graph for the target patient is constructed by combining all information entities, entity attributes, and relationships. Natural language processing (NLP) is a key technology in the field of artificial intelligence, enabling computers to understand and process human language. By simulating human language understanding and analysis capabilities, it enables tasks such as human-computer interaction, information extraction, and semantic analysis. The core of NLP technology is to convert natural language into a computer-readable form, then utilize various algorithms and models for semantic understanding, information extraction, and text generation. After constructing the patient knowledge graph for the target patient, appropriate related departments, such as neurology or gastroenterology, are matched to the target patient based on the structural and linguistic similarity between the patient knowledge graph and a pre-built medical knowledge graph. Only by matching the target patient with a suitable related department can we subsequently accurately match a consulting physician suitable for the target patient based on the related department, avoiding the waste of medical resources due to improper matching of consulting physicians.

[0025] After matching the appropriate associated departments, the doctor information of all associated doctors in the associated departments is obtained. In addition, since different hospitals may specialize in different types of diseases, in order to fully utilize the medical resources of each hospital and also to improve the disease cure rate of the target patient, the associated departments may be departments of different hospitals, and the associated doctors may also be doctors in different hospitals. Then, based on the doctor information of each associated doctor, the feature similarity between each associated doctor and the target patient is calculated. That is, the similarity between the disease type that each associated doctor specializes in and the disease type suffered by the target patient is calculated. A high feature similarity indicates that the associated doctor is proficient in treating the target patient's disease. Therefore, the associated doctor can be selected as the target patient's consulting physician to participate in the subsequent multidisciplinary consultation meeting.

[0026] S103. Create a multidisciplinary consultation meeting and grant meeting permissions to all consulting doctors, and share patient medical records information with all consulting doctors in real time during the multidisciplinary consultation meeting.

[0027] In this embodiment, after completing the consulting physician screening step, a multidisciplinary consultation meeting is immediately created for online consultation. Meeting permissions are assigned to all consulting physicians and the target patient's local attending physician. After the meeting permissions are assigned, the consulting physicians can directly access the meeting through their mobile devices, and the target patient's attending physician can also access the meeting through their PC. Furthermore, the attending physician can transmit the target patient's medical records to all consulting physicians through the multidisciplinary consultation meeting, enabling patient record information sharing. Specifically, patient medical records are first divided into patient text information and patient imaging information based on information attributes, such as the information storage format. This is due to the different transmission requirements of patient text information and patient imaging information. Since patient text information may contain the target patient's private information, it must be encrypted during transmission to prevent confidentiality leaks. Patient imaging information is an important basis for consulting physicians to discuss treatment plans, so it is necessary to ensure that the patient images received by each consulting physician are high-definition and lossless images. In addition, patient imaging information is often larger in size than patient text information, so the transmission methods of the two also need to be adaptively adjusted.

[0028] First, the patient's text information is encrypted using a text encryption algorithm. Commonly used text encryption algorithms include AES, DES, and ECC. To improve transmission efficiency and image clarity, the patient images are losslessly compressed and shared using a progressive transmission method. Specifically, the patient images are evenly divided into multiple image blocks of a fixed size (e.g., 16×16 pixels). Each block is then losslessly compressed using the DPCM, IWT, and RLE encoding algorithms, generating multiple compressed image subbands. Finally, during the multidisciplinary consultation, the encrypted text information is shared in real time with all consulting physicians, while all compressed image subbands are shared progressively in real time. Progressive real-time sharing involves initially transmitting a small amount of data to quickly reconstruct a low-resolution image. Image quality is then continuously improved through data overlay, resulting in high-definition, lossless patient images shared with all consulting physicians. This improves transmission speed while ensuring image clarity. This will improve the efficiency of subsequent consulting doctors in discussing treatment plans and the accuracy of the discussion results.

[0029] S104. After all consulting doctors have completed the consultation based on the patient's medical records, collect the diagnosis and treatment opinions of all consulting doctors.

[0030] In this embodiment, the consulting physicians participating in the multidisciplinary consultation will, after fully understanding the target patient's condition, discuss with other participating physicians and provide a final diagnosis and treatment opinion. After the consulting physicians complete their written consultation opinion, they must sign it with the CA for archiving. Furthermore, once all consulting physicians have completed their consultation opinions, the target patient's attending physician will summarize the consulting physicians' opinions and complete a consultation summary.

[0031] S105. After integrating all diagnosis and treatment opinions, an initial diagnosis and treatment plan is generated, and the initial diagnosis and treatment plan is modified according to the medical resource information to obtain the target diagnosis and treatment plan.

[0032] In this embodiment, the consulting physicians' opinions are integrated to produce a preliminary initial treatment plan. This plan includes information such as the target patient's required examinations, treatments, medications, and treatment duration. Based on the initial treatment plan, the medical resources required for the target patient's subsequent diagnosis and treatment are determined, including the type and quantity of medical equipment, the type and quantity of medications, and the number of specialists. This information is then compared with the medical resource information for the target patient's location to determine if any medical resources are missing. The missing resources are then listed one by one to obtain resource shortage information. For example, there may be a shortage of two pulmonary function testers, one fully automatic biochemical analyzer, 56 gefitinib tablets, and two cerebral infarction specialists. Based on this resource shortage information, a resource loan plan is then generated to loan the missing medical equipment, medications, and other medical resources from neighboring areas. By integrating the resource loan plan with the initial treatment plan, the target treatment plan is obtained.

[0033] In one embodiment, constructing a patient knowledge graph based on patient medical record information includes the following steps: Pre-process patient medical record information to obtain standard medical record information; Identify all information entities in standard medical record information based on the pre-trained entity recognition model; Conduct context analysis on all information entities based on standard medical record information, and determine the entity attributes of all information entities and the association relationships between all information entities based on the context analysis results; Construct a patient knowledge graph for the target patient based on all information entities, entity attributes, and association relationships.

[0034] In this embodiment, the patient's medical records are first preprocessed to obtain standardized preprocessed medical records. The preprocessing process is a key step in ensuring data quality and includes data cleaning and format conversion. Data cleaning aims to remove noise and address missing and outliers. For missing values, different processing strategies can be adopted depending on the specific situation, such as mean filling, median filling, or machine learning-based prediction filling. For example, for continuous data such as a patient's blood pressure, if missing values occur, the average blood pressure at adjacent time points can be calculated to fill the missing values. When data is missing frequently or difficult to fill using simple predictions, such as missing results from a patient's examination (e.g., missing chest CT scan results), an exception notification is sent to the target patient's hospital, prompting them to supplement or revise the patient's medical record information. Format conversion is another important step in preprocessing, aiming to standardize data from different sources into a standard format for subsequent processing. This includes standardizing date formats, converting units of measurement, and standardizing coding standards. For example, different systems may use different disease coding standards (e.g., ICD-9 and ICD-10), requiring a mapping relationship for conversion. When performing format conversion, special attention should be paid to maintaining the semantic consistency of the data to avoid information loss or errors caused by the conversion.

[0035] Next, a pre-trained entity recognition model is used to identify all entities in the standard medical record information, such as disease names, drug names, and test names. Entity recognition models include rule-based models, statistical learning models, and deep learning-based models. Deep learning-based models, for example, include the BiLSTM-CRF model (Bidirectional Long Short-Term Memory Network-Conditional Random Field model), the BERT model (Bidirectional Transformer model), and the ID-CNN (Iteratively Dilated Convolutional Neural Network) model. Taking the BERT model as an example, its core innovation lies in its use of a bidirectional Transformer encoder, which simultaneously learns contextual information on both the left and right sides of a word through a masked language model (MLM) task. For example, in understanding the semantics of the sentence "bank," the BERT model can disambiguate the meaning by integrating contextual information (e.g., "river bank" or "financial institution"). Furthermore, since the BERT model is intended for medical use, to improve its entity recognition efficiency, it can be retrained using pre-trained models from this field (such as BioBERT and ClinicalBERT). Specifically, desensitized historical patient medical records are obtained from shared databases across multiple hospitals. The records are then cleaned to remove irrelevant symbols (such as punctuation, special characters) and redundant whitespace. After word segmentation, the cleaned and word-segmented records are annotated using standard data tools such as BRAT or Prodigy. Finally, the completed records serve as the model training set. The model training set is then fed into a medical version of the BERT model (such as BioBERT or ClinicalBERT) for training. The precision, recall, and F1-score are calculated during training, and metrics such as the model learning rate and training batch size are continuously adjusted until the maximum number of training cycles is reached or the model performance meets the target (e.g., the recall rate exceeds a preset threshold). Use the BERT tokenizer to segment standard medical record information into multiple subwords. For example, ["patients developed stomach pain after taking aspirin"] can be segmented into ["patient", "person", "service", "use", "A", "department", "pi", "lin", "after", "out", "appear", "stomach", "pain"]. After adding start and delimiters to the segmented standard medical record information, the information is input into the trained medical version of the BERT model to obtain the predicted label for each subword. After merging the same predicted labels, the complete information entity is obtained, such as "patient", "aspirin", and "stomach pain".

[0036] Then, we use syntactic analysis of the sentences containing information entities in the standard medical record information to identify the subject-predicate, verb-object and other structures of the sentence, thereby identifying the contextual relationship of the information entity in the standard medical record information. We can also define a context window to extract the words around the key entity to capture its contextual relationship in the tent production information. The entity attributes of the information entity and the association relationship between all information entities can be obtained through the contextual relationship. The association relationship refers to the relationship between key entities. For example, by performing a context analysis on "The patient received ceftriaxone treatment due to pneumonia", we can obtain the predicate: receiving treatment, subject: patient, object: ceftriaxone, cause: pneumonia, so that we can get the association relationship between the information entities "patient" and "ceftriaxone" is treatment. Entity attributes refer to the attributes of the information entity, such as troponin T0.5ng / mL (normal <0.1), the entity attribute of the information entity "troponin T" is "0.5ng / mL". Finally, according to the association relationship between information entities, they are connected, and the information entities are connected with their corresponding entity attributes to construct a patient knowledge graph. Figure 2 , the initial knowledge graph can be visualized using visualization tools such as Draw.io and Neo4j Bloom.

[0037] In one embodiment, matching multiple consulting doctors for a target patient based on the patient knowledge graph includes the following steps: Complete structural matching and semantic matching between the patient knowledge graph and the pre-built medical knowledge graph, and assign multiple related departments to the target patient based on the structural matching results and semantic matching results; Get the doctor information of all related doctors in all related departments; For any associated doctor of any associated department, pre-process the doctor information of the associated doctor; Using the keyword extraction algorithm to extract all diagnosis and treatment keywords from the pre-processed doctor information, and integrating all diagnosis and treatment keywords into a diagnosis and treatment keyword set; Vectorize the diagnosis and treatment keyword set to obtain the diagnosis and treatment key vector; The similarity formula is used to calculate the feature similarity between the diagnosis and treatment key vector and the information entity of the patient knowledge graph; If the feature similarity is greater than or equal to a preset first similarity threshold, the associated doctor will be used as the consulting doctor corresponding to the target patient.

[0038] In this embodiment, the medical knowledge graph is jointly constructed by multiple hospitals. It integrates medical knowledge such as disease diagnosis standards, symptom lists, treatment plans and department classification information, and clearly constructs the relationship between diseases and symptoms, diseases and departments, symptoms and departments, etc., which can be used for disease diagnosis and department recommendations. In order to make the department matching results more accurate, it is necessary to first divide the patient knowledge graph into several patient subgraphs. The subgraph entities in the patient subgraph are mapped to the pre-constructed medical knowledge graph. The subgraph entities in the patient subgraph can be directly used as keywords, and the keywords can be used to perform keyword searches in the medical knowledge graph, marking the associated entities corresponding to the patient subgraph in the medical knowledge graph, and extracting several medical subgraphs from the medical knowledge graph based on the marked associated entities. Then, the number of nodes and the number of edges of the standard patient subgraph and all medical subgraphs are counted, as well as the number of nodes and the number of edges that are exactly the same between the standard patient subgraph and any medical subgraph, which are used to calculate the structural similarity between the standard patient subgraph and each medical subgraph. Next, all subgraph entities and the relationships between them in the standard patient subgraph are integrated into a patient node set and a patient edge set, respectively. Furthermore, the associated entities and the relationships between them in each medical subgraph are integrated into a medical node set and a medical edge set, respectively. This is used to calculate the semantic similarity between the standard patient subgraph and each medical subgraph. The structural and semantic similarities between the standard patient subgraph and each medical subgraph are combined to obtain subgraph similarity, and associated subgraphs are selected based on this subgraph similarity. The locations of all associated subgraphs are then marked in the medical knowledge graph. Starting from each graph node in the associated subgraph, the medical knowledge graph is traversed until any graph node reaches a department node (e.g., psychology or internal medicine) in the medical knowledge graph for the first time. Traversal then ceases, and the department node that the graph node first reaches is output as the associated department.

[0039] Next, we screened several hospitals participating in the multidisciplinary consultation model for specialized treatments for diseases in related departments. Specifically, we calculated the average cure rate for all diseases within the same related department at each hospital, and selected the hospitals with the highest average cure rates as related hospitals. We then collected information on all doctors in the corresponding related departments from the hospital databases of each related hospital. For example, if Hospital A ranked first in the average cure rate for all diseases within its related department B, we would select Hospital A as a related hospital, and retrieve information on all doctors in its related department B from its database.

[0040] To extract the disease types that each associated doctor is good at treating, it is necessary to preprocess the doctor information of each associated doctor first, and then use keyword extraction algorithms to extract all the diagnosis and treatment keywords in the doctor information. The preprocessing steps mainly include text cleaning, term unification, and word segmentation. Text cleaning means removing irrelevant content in the doctor information, such as punctuation marks, numbers, and common stop words like "of" and "is". Term unification means using a pre-constructed medical dictionary for unified expression. For example, "myocardial infarction" and "heart attack" are combined into the same term. Word segmentation aims to use a word segmentation tool (such as jieba) and the medical dictionary to segment terms such as diseases, symptoms, and treatment techniques, and at the same time mark disease-related terms as disease keywords. The medical dictionary covers a wide range of medical knowledge, including medical terms, disease names, symptom descriptions, etc. Commonly used keyword extraction algorithms include the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm, the TextRank (graph ranking) algorithm, and the LDA (Latent Dirichlet Allocation) algorithm, etc. Taking the TF-IDF algorithm as an example, the TF-IDF algorithm is a weighting technique used for information retrieval and text mining. TF-IDF can be used to evaluate the importance of a word for a document set or a single document in a corpus. The importance of a word increases proportionally with the number of times it appears in a document, but at the same time decreases inversely with the frequency of its appearance in the corpus. Specifically, first calculate the frequency TF of any disease keyword in the medical information, that is, TF = the number of occurrences of a disease word / the total number of words in the doctor information; count the frequency IDF of the disease keyword in all doctor information, that is, IDF = log(total number of doctor information / (number of doctor information containing the disease keyword + 1)). Multiplying TF by IDF can obtain the word frequency of the disease keyword. Calculate the word frequencies of all disease keywords in the same way, and take the disease keywords with word frequencies greater than the preset threshold as the diagnosis and treatment keywords of the associated doctor. Integrate all the diagnosis and treatment keywords of the associated doctor to obtain the diagnosis and treatment keyword set of the associated doctor.

[0041] Using pre-trained word embedding models (such as BioBERT, ClinicalBERT, or Word2Vec), each diagnosis and treatment keyword in the diagnosis and treatment keyword set (e.g., "diabetes" and "coronary artery stenting") is mapped into a high-dimensional vector (e.g., 300 dimensions). This yields multiple keyword vectors, for example, "diabetes" → vector [0.23, -0.45, 0.7, ..., 0.12]. Based on the frequency of each diagnosis and treatment keyword, all keyword vectors associated with the doctor are weighted averaged to obtain the doctor's diagnosis and treatment key vector. Next, using the same word embedding model as the associated doctor, the information entities corresponding to all disease nodes (e.g., coronary heart disease and heart disease) in the patient knowledge graph are converted into multiple information vectors. The average vector of all information vectors is calculated and used as the patient key vector for the target patient. A similarity formula is used to calculate the feature similarity between the diagnosis and treatment key vector and the patient key vector. If the feature similarity is greater than or equal to a preset first similarity threshold, the associated doctor is designated as the consulting physician for the target patient. Using the same method, we calculate the feature similarity between other associated doctors and the target patient, and then filter all the target patient's consulting doctors based on feature similarity. Common similarity formulas include cosine similarity and Euclidean distance. By matching the target patient with the appropriate consulting doctor, we can significantly increase the target patient's recovery probability, improve diagnosis and treatment efficiency, and reduce the waste of medical resources.

[0042] In one embodiment, completing structural matching and semantic matching between a patient knowledge graph and a pre-built medical knowledge graph, and assigning multiple associated departments to a target patient based on the structural matching results and the semantic matching results includes the following steps: Divide the patient knowledge graph into several patient subgraphs based on the subgraph partitioning rules; For any patient subgraph, the subgraph entities in the patient subgraph are mapped to the pre-built medical knowledge graph, and multiple medical subgraphs are extracted from the medical knowledge graph based on the mapping results; Complete the patient subgraph based on the medical knowledge graph to obtain the standard patient subgraph; Count the number of nodes and edges in the standard patient subgraph and the entire medical subgraph, and calculate the structural similarity between the standard patient subgraph and the entire medical subgraph based on the statistical results of the number of nodes and edges; Calculate the node similarity and edge similarity between the standard patient subgraph and all medical subgraphs, and combine the node similarity and edge similarity to determine the semantic similarity between the standard patient subgraph and all medical subgraphs; The structural similarity and semantic similarity are integrated to obtain the subgraph similarity between the standard patient subgraph and the entire medical subgraph; The medical subgraphs whose subgraph similarity is greater than a preset second similarity threshold are used as associated subgraphs of the standard patient subgraph; Multiple related departments are traced back based on the positions of all related subgraphs in the medical knowledge graph.

[0043] In this embodiment, a variety of subgraph partitioning rules are available. For example, subgraph partitioning can be performed based on the patient's chief complaint. For example, if the patient's chief complaint is "chest pain," then the "chest pain" information entity is used as the core, and all directly connected information entities (symptoms, examinations, medications) are extracted to form a patient subgraph. Furthermore, subgraphs can be separated based on the target patient's body system (such as cardiovascular, respiratory, or digestive). For example, if the patient suffers from stomach problems and migraines, then "stomach problems" and "migraines" are used as the core, and directly connected information entities are extracted to form a patient subgraph. Subgraph partitioning is crucial for subsequently matching appropriate related departments. This is because some patients suffer from multiple, relatively independent diseases. For example, if a patient suffers from lupus erythematosus and a left leg fracture, directly matching the patient knowledge graph with the medical knowledge graph would result in extremely low similarities due to the near-insignificant relationship between the two diseases, making it difficult to match appropriate related departments based on this similarity. Therefore, it is necessary to partition the patient knowledge graph into several patient subgraphs.

[0044] The medical knowledge graph, jointly constructed by multiple hospitals, integrates medical knowledge such as disease diagnosis criteria, symptom lists, treatment plans, and department classification information. It clearly establishes relationships between diseases and symptoms, diseases and departments, and symptoms and departments, enabling disease diagnosis and department recommendations. For any patient subgraph, the subgraph entities in the patient subgraph are mapped to a pre-built medical knowledge graph. Subgraph entities in the patient subgraph can be directly used as keywords for keyword searches within the medical knowledge graph. Associated entities in the medical knowledge graph corresponding to the patient subgraph are marked. Based on the marked associated entities, several medical subgraphs are extracted from the medical knowledge graph. Entities between associated entities and entities directly connected to associated entities can be directly extracted to obtain the medical subgraph. The patient subgraph is then completed based on the medical knowledge graph. For example, if the "diabetes" and "abnormal renal function" information entities exist in the patient subgraph, a "diabetic nephropathy" relationship edge is automatically added based on the complication rules in the medical knowledge graph. This step is intended to improve the accuracy of the subsequently calculated subgraph similarity, thereby improving the accuracy of department matching results.

[0045] Next, count the number of nodes and edges in the standard patient subgraph and all medical subgraphs, obtaining the total number of patient nodes, the total number of patient edges, the total number of first medical nodes, the total number of first medical edges, the total number of second medical nodes, the total number of second medical edges, and so on. The number of nodes refers to the number of entities in the standard patient subgraph and the medical subgraphs. The number of nodes in the standard patient subgraph is the total number of subgraph entities, and the number of edges is the total number of relationships between entities in the standard patient subgraph and the medical subgraphs. The number of identical nodes and edges between the standard patient subgraph and any medical subgraph is counted, obtaining the number of first node overlaps, the number of first edge overlaps, the number of second node overlaps, the number of second edge overlaps, and so on. The terms "first" and "second" are used to distinguish different medical subgraphs. The node overlap ratio and edge overlap ratio between the two are calculated based on the number of nodes and edges of the two, and the sum of the node overlap ratio and edge overlap ratio between the patient subgraph and the medical subgraph is taken as the structural similarity between the two. For example, the total number of patient nodes in the standard patient subgraph is 5, and the total number of patient edges is 4. The total number of medical nodes in a certain medical subgraph is 16, and the total number of medical edges is 15. The number of node overlaps between the two is 4, and the number of edge overlaps is 4. Then the node overlap ratio between the two is 4 / 21, and the edge overlap ratio is 4 / 19. The structural similarity between the standard patient subgraph and the medical subgraph is 4 / 21+4 / 19≈0.401.

[0046] Next, all subgraph entities and the relationships between them in the standard patient subgraph are integrated into patient node sets and patient edge sets. For example, the patient node set is [fever, rash, chickenpox, ...], and the patient edge set is [diagnosis, treatment, cause, ...]. Simultaneously, the associated entities and the relationships between them in each medical subgraph are integrated into medical node sets and medical edge sets. Pretrained graph embedding models (such as TransE and Node2Vec) can then be used to convert the patient node set, patient edge set, medical node set, and medical edge set into patient node vectors, patient edge vectors, medical node vectors, and medical edge vectors, respectively. Similarity formulas are then used to calculate the node similarity between patient node vectors and medical node vectors, and the edge similarity between patient edge vectors and medical edge vectors. Common similarity formulas include cosine similarity and Euclidean distance. Next, weights are assigned to node similarity and edge similarity. Since node similarity better reflects the similarity between the patient subgraph and the medical subgraph than edge similarity, a higher weight is assigned to node similarity. For example, a weight of 0.85 is assigned to node similarity, while a weight of 0.15 is assigned to edge similarity. This is because, due to diagnosis and treatment, equal associations frequently appear in knowledge graphs in the medical field. This means that a high edge similarity between the patient subgraph and the medical subgraph may affect the subgraph similarity between the two, thereby affecting the accuracy of department matching. Therefore, a lower weight is assigned to edge similarity. The weighted node and edge similarities are then weighted and fused to obtain the semantic similarity between the standard patient subgraph and each medical subgraph.

[0047] Next, the structural similarity and semantic similarity are fused to obtain the subgraph similarity between the standard patient subgraph and each medical subgraph. Subgraph similarity can be obtained by simply adding the structural similarity and semantic similarity. Alternatively, weights can be assigned to the structural similarity and semantic similarity and then performing a weighted fusion to obtain weighted fused subgraph similarity.

[0048] If the subgraph similarity between a certain medical subgraph and the standard patient subgraph is greater than a preset similarity threshold, the medical subgraph is used as an associated subgraph of the standard patient subgraph. Then, the positions of all associated subgraphs are marked in the medical knowledge graph, and each graph node in the associated subgraph is used as a starting point to traverse along the medical knowledge graph until any graph node reaches the department node (such as psychology department, internal medicine department) in the medical knowledge graph for the first time, then the traversal is stopped, and the department node that the graph node reaches for the first time is output as the associated department. If there are multiple graph nodes that all reach different department nodes for the first time, all department nodes are output as associated departments. This is because the consultation mode in this application is a multidisciplinary fusion consultation, so doctors from different departments are required to consult. In addition, if there is a department node in the associated subgraph, the department node in the associated subgraph is directly output as the associated department.

[0049] In one embodiment, sharing patient medical records with all consulting physicians in real time during a multidisciplinary consultation meeting includes the following steps: Divide patient medical record information into patient text information and patient image information according to information attributes; Encrypting the patient's text information using a text encryption algorithm to obtain encrypted text information; dividing the patient image information into a plurality of patient image blocks; For any patient image block, the image compression algorithm is used to perform lossless compression of the patient image block to obtain multiple compressed image sub-bands; Share encrypted text information with all consulting physicians in real time during multidisciplinary consultations, and progressively share all compressed image subbands in real time.

[0050] In this embodiment, the patient medical record information is first divided into patient text information and patient image information based on information attributes, such as information storage format. This is because the transmission requirements of patient text information and patient image information are different. Since patient text information may contain the target patient's private information, such as the patient's ID number and the patient's mental illness, it needs to be encrypted during transmission to prevent leakage. Patient image information is an important basis for consulting doctors to discuss diagnosis and treatment plans, so it is necessary to ensure that the patient image information received by each consulting doctor is high-definition and lossless. In addition, patient image information often has a larger data volume than patient text information, so the transmission methods of the two also need to be adaptively adjusted.

[0051] Specifically, the patient text information is first encrypted using a text encryption algorithm to obtain encrypted text information. Commonly used text encryption algorithms include AES encryption algorithm, DES encryption algorithm, ECC encryption algorithm, etc. Taking the AES encryption algorithm as an example, its basic steps include first using PKCS#7 and other filling methods to fill the patient text information to an integer multiple of 128 bits (16 bytes); then dividing the filled patient text information into multiple 128-bit patient text blocks; generating multiple round keys through an extension algorithm based on the original key (128 / 192 / 256 bits). Commonly used extension algorithms include The encryption process involves performing a bitwise exclusive-or (XOR) operation on each patient text block with the first round key as the initial encryption step. Multiple rounds of encryption are then performed on the patient text block that has completed the initial encryption step, including byte substitution, row shifting, column obfuscation, and round key addition. Finally, the column obfuscation step is performed on the patient text block that has completed the multiple rounds of encryption. This involves performing the following steps: byte substitution → row shifting → round key addition to complete the final encryption, resulting in an encrypted text block. All encrypted text blocks are then sequentially combined to form the final ciphertext, or encrypted text information. To improve subsequent compression efficiency, the patient image information is first evenly divided into multiple patient image blocks of a fixed size (e.g., 16×16 pixels). Each patient image block is then losslessly compressed using the DPCM, IWT, and RLE encoding algorithms, respectively, to produce multiple compressed image subbands.

[0052] The DPCM coding algorithm is one of many data compression techniques. It exploits the temporal correlation of data to reduce required storage space or transmission bandwidth, and is primarily used for compressing speech and image signals. Because patient imaging information is highly spatially correlated, direct entropy coding is inefficient for image compression. Therefore, the DPCM coding algorithm must first be used to decorrelate the image and reduce the spatial correlation between pixels before subsequent compression steps. The Integer Wavelet Transform (IWT) is a wavelet transform method based on integer arithmetic. It represents a signal as a discrete wavelet transform (DWT) sequence of integers and is suitable for analyzing and processing discrete-time signals. Its basic concept is to decompose the original signal into wavelet coefficients at multiple scales and frequency bands. Run-Length Encoding (RLE) is a lossless algorithm based on the compression of continuously repeated data. It simplifies storage by counting the number of occurrences (run length) and the value (run value) of the continuously repeated data. For example, the string "AAAABBBCCDAA" can be compressed to [(4,A),(3,B),(2,C),(1,D),(2,A)]. Finally, during the multidisciplinary consultation, encrypted text information is shared in real time with all consulting physicians, while all compressed image subbands are shared using progressive real-time sharing. Progressive real-time sharing involves using a progressive transmission method during the transmission process, initially transmitting a small amount of data to quickly reconstruct a low-resolution version of the image. This is then used to continuously improve image quality through data overlay, resulting in high-definition, lossless patient images shared with all consulting physicians. This improves transmission speed while ensuring image clarity. This improves the efficiency and accuracy of subsequent treatment plan discussions among consulting physicians.

[0053] In one embodiment, performing lossless compression of a patient image block using an image compression algorithm to obtain a plurality of compressed image subbands includes the following steps: Extract the true pixel value of the patient image block; Predicting pixel prediction values of compressed image blocks using a preset prediction function; Combine the true pixel value and the predicted pixel value to construct a pixel difference matrix; The pixel difference matrix is input into the IWT encoder, and the pixel difference matrix is transformed and decomposed by the IWT encoder to obtain multiple image sub-bands; The RLE coding algorithm is used to complete the coding compression of all image sub-bands to obtain multiple compressed image sub-bands.

[0054] In this embodiment, each patient image block is first converted into a pixel difference matrix using the DPCM (Differential Pulse Code Modulation) encoding algorithm. DPCM is one of many data compression techniques that leverages the temporal correlation of data to reduce required storage space or transmission bandwidth. It is primarily used for compressing voice and image signals. Because patient image information has a high degree of spatial correlation, direct image compression via entropy coding is inefficient. Therefore, the DPCM encoding algorithm is first used to decorrelate the image and reduce the spatial correlation between pixels before performing subsequent compression steps. Specifically, the true pixel values of the patient image block are first extracted. The true pixel value refers to the color or brightness information contained in each pixel in the patient image block. For example, pixel values in an image are typically represented as integers between 0 and 255, with 0 representing black, 255 representing white, and intermediate values representing different shades of gray. A predetermined prediction function is then used to predict the pixel predicted value of the compressed image block. The prediction function is typically a linear regression function, for example, directly using the pixel mean of all adjacent pixels as the pixel predicted value. In addition to retaining some pixels of the patient image block (such as the pixels in the upper left corner) without prediction, the same method is used to predict the pixel prediction values of all pixels in the patient image block except for some pixels. After calculating the pixel difference between the true value of each pixel and the predicted value of the pixel at the same position, a one-dimensional difference sequence is obtained. Then, in order to further concentrate energy through IWT, the one-dimensional difference sequence is restored to the two-dimensional matrix structure of the original image (patient image block) to obtain a pixel difference matrix. Specifically, refer to Figure 3 First, obtain the size of the patient image block, such as 16×16 pixels, and then fill the one-dimensional sequence into the two-dimensional matrix structure from left to right and from top to bottom according to the size of the patient image block to obtain the pixel difference matrix, where is the predicted value of the nth pixel, 、 etc. are all in the order of pixel prediction values. In addition, to ensure lossless compression, this application skips the quantization step in the DPCM algorithm and directly performs the subsequent encoding step on the pixel difference matrix. The pixel difference matrix is input into the IWT encoder, and the IWT encoder completes the transformation and decomposition processing of the pixel difference matrix to further remove redundancy.

[0055] The integer wavelet transform (IWT) is a wavelet transform method based on integer arithmetic. It is a discrete wavelet transform method that represents a signal as a sequence of integers. It is suitable for analyzing and processing discrete-time signals. Its basic concept is to decompose the original signal into wavelet coefficients at multiple scales and frequency bands. The key to the integer wavelet transform is how to implement the discrete wavelet transform so that the wavelet coefficients are all integers. Common methods are to use integer filters and integer downsampling. Integer filters smooth and extract details from the original signal, and then perform integer downsampling to obtain a signal at a smaller scale. The integer wavelet transform can use a multi-level decomposition approach to decompose the signal into different scales and frequency bands. The integer wavelet transform has wide applications in image and video compression, signal processing, pattern recognition, and other fields. Compared with the floating-point wavelet transform, the integer wavelet transform has the advantages of fast computational speed, simple implementation, and small storage space, making it more suitable for transmitting patient imaging information. Specifically, wavelet basis functions (such as Haar and (5,3) filters) that are suitable for patient imaging information are selected to capture texture, edge, and structural features at different scales. The pixel difference matrix is then filtered alternately in the row and column directions using wavelet basis functions to generate four subbands: the low-frequency approximation subband (LL), the horizontal detail subband (LH), the vertical detail subband (HL), and the diagonal detail subband (HH). The low-frequency approximation subband is then subjected to several wavelet transforms to further decompose, yielding higher-level low-frequency and high-frequency components: low-frequency subband LL2 and high-frequency subbands LH2, HL2, and HH2. This continues until a predetermined number of transforms (e.g., three) is reached, resulting in multiple image subbands. The RLE encoding algorithm is then used to compress all image subbands. Because an image often contains many blocks of the same color, RLE only stores the color of one block and the number of pixels corresponding to that color. Its core concept is to represent consecutive identical values (called "runs") as a combination of (value, number of repetitions). Therefore, RLE (run-length encoding) is simple to implement and is suitable for processing patient images with a small number of grayscale levels and strong data correlation, such as those with patient images. Specifically, traverse each image subband and record the current pixel value and number of repetitions in each image subband. For example, if the next pixel value is the same as the current pixel value, the number of repetitions is increased by 1. When the next pixel value is different from the current pixel value, output the current pixel value and its number of repetitions, reset the run length, and complete the compression until the last run length is recorded to obtain multiple compressed image subbands.

[0056] In one embodiment, the progressive real-time sharing of all compressed image sub-bands includes the following steps: Analyze the frequency characteristics of all compressed image sub-bands and determine the transmission priority of all compressed image sub-bands based on the frequency characteristics; All compressed image subbands are gradually shared to the multidisciplinary consultation conference according to transmission priority.

[0057] In this embodiment, the frequency characteristics of each compressed image subband can be determined directly based on the decomposition results of the IWT encoder. For example, the frequency characteristics of compressed image subbands corresponding to LH and HL are intermediate frequency subbands, while the frequency characteristics of compressed image subbands corresponding to LH2 and HL2 are high frequency subbands. Transmission priority refers to the transmission order of all compressed image subbands. During progressive transmission, according to the transmission priority, the low-frequency subband is transmitted first to display the primary patient image information, followed by the intermediate and high frequency subbands, gradually supplementing the patient image information details.

[0058] Modern medical imaging technology provides a large number of high-resolution images containing detailed anatomical details, playing a vital role in guiding diagnosis, clinical trials, and case discussions. However, a single high-resolution patient image can range from a few megabytes to tens or even hundreds of megabytes. For example, a computed radiograph (CR) or digital radiograph (DR) image is at least a dozen megabytes in size, while a mammography (MG) image can be tens of megabytes. During multidisciplinary consultations, it is crucial to share the target patient's images with all participating physicians in the shortest possible time, while also ensuring high-definition, lossless images. Therefore, lossless compression is first performed to reduce the image size without sacrificing resolution, thereby improving subsequent transmission efficiency. Subsequently, progressive transmission is used, initially transmitting a small amount of data to quickly reconstruct a low-resolution version of the image. Image quality is then continuously improved through data overlay, ensuring that the patient images shared with all participating physicians are high-definition, lossless images. This improves transmission speed while ensuring image clarity. This will improve the efficiency of subsequent consulting doctors in discussing treatment plans and the accuracy of the discussion results.

[0059] In one embodiment, the steps of generating an initial diagnosis and treatment plan after integrating all diagnosis and treatment opinions, and revising the initial diagnosis and treatment plan based on medical resource information to obtain a target diagnosis and treatment plan include the following steps: Integrate all diagnosis and treatment opinions to generate an initial diagnosis and treatment plan; Determine resource shortage information in the target patient's location by combining the initial diagnosis and treatment plan with medical resource information; Generate resource secondment plans based on resource shortage information; Use the resource loan plan to improve the initial diagnosis and treatment plan and obtain the target diagnosis and treatment plan.

[0060] In this embodiment, the consulting physicians participating in the multidisciplinary consultation meeting will, after fully understanding the target patient's condition, discuss with other participating physicians and provide a final diagnosis and treatment opinion. At the same time, after the consulting physicians complete the written consultation opinion, it must be archived with a signature from the CA. A simple integration of the consulting physicians' diagnosis and treatment opinions can be used to preliminarily obtain an initial diagnosis and treatment plan. Based on the initial diagnosis and treatment plan, the medical resources required for the subsequent diagnosis and treatment of the target patient are determined, including the type of medical equipment, the number of equipment, the type of medication, the number of medications, the number of specialists, etc. This is then compared with the medical resource information in the target patient's location to determine whether there are any missing medical resources. The missing medical resources are then listed one by one to obtain resource shortage information. For example, there may be a shortage of pulmonary function testers (two), a fully automatic biochemical analyzer (one), gefitinib (56 tablets), and two cerebral infarction specialists. A resource loan plan is then generated based on the resource shortage information to loan the missing medical equipment, medications, and other medical resources from neighboring areas where the target patient is located. The plan includes the following steps: 1.1. Collect resource usage information of all neighboring hospitals in the area near the target patient's location; 1.2. Analyze all neighboring hospitals for resource redundancy based on resource usage information, and mark neighboring hospitals with resource redundancy as supporting hospitals; 1.3. Calculate the travel time between all supporting hospitals and the target patient's location; 1.4. Take the traffic time information and resource redundancy information as constraints, and combine the constraints and resource loss information to generate a resource secondment plan.

[0061] Specifically, resource usage information is first collected for all neighboring hospitals in the area surrounding the target patient's location. This resource usage information refers to the use of medical resources required by the target patient, including information such as medical equipment idle time (e.g., average daily idle time), medication inventory, and physician schedule information. Resource usage information is then used to determine whether there is resource redundancy in all neighboring areas. If the idle time of medical equipment in a particular area exceeds a preset threshold, it is considered redundant, and the neighboring hospitals with redundant equipment are marked as supporting hospitals. Drugs with inventory levels exceeding a preset threshold are then marked as redundant, and the neighboring hospitals with redundant drugs are marked as supporting hospitals. The inventory threshold is determined based on the drug's historical average daily consumption and procurement time. It is necessary to ensure that the drug inventory level is sufficient to meet the needs of local patients before the procurement time arrives. Finally, the physician's weekly available hours are determined based on the physician's schedule information. If the physician's weekly available hours exceed the preset threshold, the physician is marked as available, and the neighboring hospitals where the available physician resides are marked as supporting hospitals. Physician schedule information includes information such as the physician's shift type, duty dates, and working hours. If no supporting hospital has all the necessary redundant equipment, redundant medications, and available physicians, the travel time between all supporting hospitals and the target patient's location is calculated. Specifically, the travel time between the target hospital where the target patient is located and all neighboring hospitals in all nearby areas is calculated. Shorter travel times are more beneficial for the target patient's treatment, so travel time information is considered. Each supporting hospital is assigned a support score based on its travel time information, the number of redundant equipment, the number of redundant medications, and the weekly available hours of its available physicians. For example, a support score can be assigned to a supporting hospital with a travel time less than or equal to a first threshold, 0.8 points to a supporting hospital with a travel time between the first threshold and a second threshold, and 0.6 points to a supporting hospital with a travel time greater than the second threshold, where the first threshold is less than the second threshold. Similarly, each supporting hospital is assigned a support score based on the number of redundant equipment, the number of redundant medications, and the weekly available hours of its available physicians, resulting in a total support score for each supporting hospital. A resource secondment plan is generated based on the supporting hospital's total support score.For example, the redundant resources of the supporting hospital ranked first in total support score can be used as the initial allocation resources. If the initial allocation resources can make up for the resource shortage in the target patient's location, a resource loan plan is directly generated based on the initial allocation resources. If they cannot make up for the resource shortage in the target patient's location, the redundant resources of the supporting hospital ranked second in total support score are integrated with the initial allocation resources to obtain the second allocation resources. If the second allocation resources can make up for the resource shortage in the target patient's location, a resource loan plan is directly generated based on the second allocation resources. Otherwise, the redundant resources of the supporting hospital ranked third in total support score are continuously integrated with the second allocation resources until the integrated allocation resources can make up for the resource shortage in the target patient's location. Finally, the resource loan plan and the initial diagnosis and treatment plan are integrated to obtain the target diagnosis and treatment plan.

[0062] An embodiment of the present application also discloses a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the method of multidisciplinary fusion consultation based on the knowledge graph in the above embodiment.

[0063] The present application also discloses a multidisciplinary fusion consultation system based on a knowledge graph, including: a memory configured to store instructions; and The processor is configured to call instructions from the memory and implement the method of multidisciplinary fusion consultation based on knowledge graph according to the above embodiment when executing the instructions.

[0064] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0065] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD) or flash memory card (FC) equipped on the computer device. In addition, the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output. This application does not impose any restrictions on this.

[0066] An embodiment of the present application also provides a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the above-mentioned method of multidisciplinary fusion consultation based on knowledge graph.

[0067] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0069] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0071] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0072] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0073] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0074] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0075] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included within the scope of the claims of the present application.

Claims

1. A multidisciplinary fusion consultation method based on knowledge graph, characterized by: The method comprises the following steps: Receiving patient medical record information of a target patient and medical resource information of a location where the target patient is located; Build a patient knowledge graph based on the patient's medical record information, and match multiple consulting doctors for the target patient based on the patient knowledge graph; Create a multidisciplinary consultation meeting and grant meeting permissions to all consulting doctors, and share patient medical records with all consulting doctors in real time during the multidisciplinary consultation meeting; After all consulting doctors have completed the consultation based on the patient's medical records, their diagnosis and treatment opinions are collected; After integrating all diagnosis and treatment opinions, an initial diagnosis and treatment plan is generated, and the initial diagnosis and treatment plan is revised based on medical resource information to obtain the target diagnosis and treatment plan.

2. The method according to claim 1, characterized in that The construction of a patient knowledge graph based on patient medical record information includes the following steps: Pre-process patient medical record information to obtain standard medical record information; Identify all information entities in standard medical record information based on the pre-trained entity recognition model; Conduct context analysis on all information entities based on standard medical record information, and determine the entity attributes of all information entities and the association relationships between all information entities based on the context analysis results; Construct a patient knowledge graph for the target patient based on all information entities, entity attributes, and association relationships.

3. The method according to claim 1, characterized in that Matching multiple consulting doctors for a target patient based on the patient knowledge graph includes the following steps: Complete structural matching and semantic matching between the patient knowledge graph and the pre-built medical knowledge graph, and assign multiple related departments to the target patient based on the structural matching results and semantic matching results; Get the doctor information of all related doctors in all related departments; For any associated doctor of any associated department, pre-process the doctor information of the associated doctor; Using the keyword extraction algorithm to extract all diagnosis and treatment keywords from the pre-processed doctor information, and integrating all diagnosis and treatment keywords into a diagnosis and treatment keyword set; Vectorize the diagnosis and treatment keyword set to obtain the diagnosis and treatment key vector; The similarity formula is used to calculate the feature similarity between the diagnosis and treatment key vector and the information entity of the patient knowledge graph; If the feature similarity is greater than or equal to a preset first similarity threshold, the associated doctor will be used as the consulting doctor corresponding to the target patient.

4. The method according to claim 3, characterized in that The steps of completing the structural matching and semantic matching between the patient knowledge graph and the pre-built medical knowledge graph, and assigning multiple associated departments to the target patient based on the structural matching results and the semantic matching results include the following: Divide the patient knowledge graph into several patient subgraphs based on the subgraph partitioning rules; For any patient subgraph, the subgraph entities in the patient subgraph are mapped to the pre-built medical knowledge graph, and multiple medical subgraphs are extracted from the medical knowledge graph based on the mapping results; Complete the patient subgraph based on the medical knowledge graph to obtain the standard patient subgraph; Count the number of nodes and edges in the standard patient subgraph and the entire medical subgraph, and calculate the structural similarity between the standard patient subgraph and the entire medical subgraph based on the statistical results of the number of nodes and edges; Calculate the node similarity and edge similarity between the standard patient subgraph and all medical subgraphs, and combine the node similarity and edge similarity to determine the semantic similarity between the standard patient subgraph and all medical subgraphs; The structural similarity and semantic similarity are integrated to obtain the subgraph similarity between the standard patient subgraph and the entire medical subgraph; The medical subgraphs whose subgraph similarity is greater than a preset second similarity threshold are used as associated subgraphs of the standard patient subgraph; Multiple related departments are traced back based on the positions of all related subgraphs in the medical knowledge graph.

5. The method according to claim 1, wherein Sharing the patient's medical history information with all consulting doctors in real time during a multidisciplinary consultation meeting includes the following steps: Divide patient medical record information into patient text information and patient image information according to information attributes; Encrypting the patient's text information using a text encryption algorithm to obtain encrypted text information; dividing the patient image information into a plurality of patient image blocks; For any patient image block, the image compression algorithm is used to perform lossless compression of the patient image block to obtain multiple compressed image sub-bands; Share encrypted text information with all consulting physicians in real time during multidisciplinary consultations, and progressively share all compressed image subbands in real time.

6. The method according to claim 5, characterized in that The method of using an image compression algorithm to perform lossless compression of a patient image block to obtain a plurality of compressed image sub-bands comprises the following steps: Extract the true pixel value of the patient image block; Predicting pixel prediction values of compressed image blocks using a preset prediction function; Combine the true pixel value and the predicted pixel value to construct a pixel difference matrix; The pixel difference matrix is input into the IWT encoder, and the pixel difference matrix is transformed and decomposed by the IWT encoder to obtain multiple image sub-bands; The RLE coding algorithm is used to complete the coding compression of all image sub-bands to obtain multiple compressed image sub-bands.

7. The method according to claim 5, characterized in that The progressive real-time sharing of all compressed image sub-bands comprises the following steps: Analyze the frequency characteristics of all compressed image sub-bands and determine the transmission priority of all compressed image sub-bands based on the frequency characteristics; All compressed image subbands are gradually shared to the multidisciplinary consultation conference according to transmission priority.

8. The method according to claim 1, characterized in that The steps of generating an initial diagnosis and treatment plan after integrating all diagnosis and treatment opinions, and revising the initial diagnosis and treatment plan based on medical resource information to obtain a target diagnosis and treatment plan include the following: Integrate all diagnosis and treatment opinions to generate an initial diagnosis and treatment plan; Determine resource shortage information in the target patient's location by combining the initial diagnosis and treatment plan with medical resource information; Generate resource secondment plans based on resource shortage information; Use the resource loan plan to improve the initial diagnosis and treatment plan and obtain the target diagnosis and treatment plan.

9. A machine-readable storage medium, characterized in that The machine-readable storage medium stores instructions for enabling a machine to execute the method for multidisciplinary fusion consultation based on knowledge graph according to any one of claims 1 to 8.

10. A multidisciplinary fusion consultation system based on knowledge graph, characterized by: include: a memory configured to store instructions; as well as A processor is configured to call instructions from a memory and to implement the method of multidisciplinary fusion consultation based on knowledge graph according to any one of claims 1 to 8 when executing the instructions.

Images