Auxiliary diagnosis method and device based on generative model implemented by computer program

Abstract

Provided is an auxiliary diagnosis method and device based on a generative model implemented by a computer program. The auxiliary diagnosis method includes: constructing a patient-centered knowledge graph based on medical record information in a medical record library; extracting relevant information of the patient to be queried; retrieving a first entity matching the relevant information of the patient to be queried in the knowledge graph; extracting a subgraph from the knowledge graph, the subgraph consisting of a second entity as a first-order neighbor of the first entity, the first-order neighbors of the second entity, and the relationship between them; converting the subgraph into a prompt word; and activating a generative model according to the prompt word to generate an auxiliary diagnosis prediction result.

Description

Technical Field

[0001] The present disclosure relates to the field of smart medical care, and more specifically, to an auxiliary diagnosis method based on a retrieval-enhanced generative model implemented by a computer program, and in particular to a technology for analyzing and predicting clinical data by building a patient-centric knowledge graph combined with a generative model. Background Art

[0002] Generative models represented by large language models can efficiently process patient diagnosis and treatment information contained in medical electronic medical records, generate corresponding clinical prediction results, and assist doctors in clinical diagnosis. They have now become a new paradigm in the field of artificial intelligence medical research. Retrieval enhancement technology can guide the generative model to retrieve information related to the patient to be predicted from a large external database, and then use the retrieved information to guide the generation of clinical prediction results, thereby alleviating the hallucination problem commonly found in generative models and improving the accuracy of the model's output of clinical prediction results.

[0003] The current generative model retrieval enhancement technology in the medical field is mainly implemented through vector similarity calculation. During the retrieval process, the data is loaded, segmented, and embedded in vectors to build an index, and then the relevant results are recalled through vector retrieval. The generation process uses prompt words based on retrieval result enhancement to activate the large language model to generate clinical outcome predictions.

[0004] Existing retrieval enhancement methods mainly rely on distance or similarity in vector space to measure semantic similarity. However, this method may not accurately reflect the true semantic similarity, and noise and outliers in the vector space may interfere with the search results, resulting in a certain degree of randomness or uncertainty in the generation process, which limits the application of this technology in the medical field, which requires high safety and reliability.

[0005] Due to the existence of the above problems, there is an urgent need to design a retrieval enhancement generation method for clinical tasks to more comprehensively retrieve the key information contained in the external database, guide the generation process of the large language model, and verify the output prediction results to enhance the application prospects of this technology in the clinical field. Summary of the invention

[0006] The embodiments of the present disclosure provide an auxiliary diagnosis method and device based on a generative model implemented by a computer program, thereby effectively making up for the deficiencies of existing auxiliary diagnosis methods in clinical tasks.

[0007] An embodiment of the present disclosure provides an auxiliary diagnosis method based on a generative model implemented by a computer program, the auxiliary diagnosis method comprising: constructing a patient-centric knowledge graph based on medical record information in a medical record library; extracting relevant information of the patient to be queried; retrieving a first entity that matches the relevant information of the patient to be queried in the knowledge graph; extracting a subgraph from the knowledge graph, the subgraph consisting of a second entity that is a first-order neighbor of the first entity, the first-order neighbors of the second entity, and the relationship between them; converting the subgraph into a prompt word; and activating the generative model according to the prompt word to generate an auxiliary diagnosis prediction result.

[0008] Optionally, the step of constructing a patient-centric knowledge graph may include: extracting the patient's identity, physiological characteristics, symptoms, disease types, diagnosis and treatment methods, test results, and medication records included in the medical record information as entities; performing relationship discovery on the extracted entities to form triples, wherein the triples have the form of a head entity, a relationship, and a tail entity; and organizing the triples in the form of a graph to form a patient-centric knowledge graph.

[0009] Optionally, the auxiliary diagnosis method may also include: integrating the auxiliary diagnosis prediction results into the knowledge graph to obtain candidate triples corresponding to the auxiliary diagnosis prediction results; performing confidence evaluation on the candidate triples through the knowledge graph representation learning method to obtain the confidence used to characterize the consistency between the candidate triples and the knowledge graph; and outputting the auxiliary diagnosis prediction results whose confidence reaches a preset target as prediction results.

[0010] Optionally, the step of performing confidence assessment on candidate triples through a knowledge graph representation learning method may include: learning the knowledge graph using the knowledge graph representation learning method to obtain embedded representations of entities and relationships in the knowledge graph; obtaining candidate embedded representations by matching the entities and relationships of the candidate triples with the embedded representations; and performing confidence assessment based on the candidate embedded representations.

[0011] Optionally, the relevant information of the patient to be queried includes at least symptoms and physiological characteristics of the patient to be queried.

[0012] An embodiment of the present disclosure provides an auxiliary diagnosis device based on a generative model implemented by a computer program, the auxiliary diagnosis device comprising: a knowledge graph construction unit, configured to construct a patient-centric knowledge graph based on medical record information in a medical record library; a query information acquisition unit, configured to extract relevant information of a patient to be queried; a retrieval unit, configured to retrieve a first entity that matches the relevant information of the patient to be queried in the knowledge graph; a subgraph extraction unit, configured to extract a subgraph from the knowledge graph, the subgraph consisting of a second entity that is a first-order neighbor of the first entity, the first-order neighbors of the second entity, and the relationship between them; a prompt word generation unit, configured to convert the subgraph into a prompt word; and a prediction unit, configured to activate the generative model according to the prompt word to generate an auxiliary diagnosis prediction result.

[0013] Optionally, the knowledge graph construction unit is configured to perform the following steps: extracting the patient's identity, physiological characteristics, symptoms, disease types, diagnosis and treatment methods, test results and medication records included in the medical record information as entities; performing relationship discovery on the extracted entities to form triples in the form of head entity, relationship, and tail entity; and organizing the triples in the form of a graph to form a patient-centric knowledge graph.

[0014] Optionally, the auxiliary diagnosis device may further include: a consistency verification unit. The consistency verification unit is configured to perform the following steps: integrating the auxiliary diagnosis prediction result into the knowledge graph to obtain a candidate triple corresponding to the auxiliary diagnosis prediction result; performing confidence evaluation on the candidate triple by a knowledge graph representation learning method to obtain a confidence for characterizing the consistency between the candidate triple and the knowledge graph; and outputting the auxiliary diagnosis prediction result whose confidence reaches a preset target as a prediction result.

[0015] An embodiment of the present disclosure provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the auxiliary diagnosis method as described above is implemented.

[0016] An embodiment of the present disclosure provides a computing device, which includes: a processor; and a memory storing a computer program, and when the computer program is executed by the processor, the auxiliary diagnosis method as described above is implemented.

[0017] According to one or more embodiments of the embodiments of the present disclosure, the auxiliary diagnosis method and device based on the retrieval enhanced generation model of the present disclosure can effectively integrate various types of information in the electronic medical record by constructing a patient-centric knowledge graph, and integrate the patient's symptoms, physiological characteristics and historical diagnosis and treatment data to make the auxiliary diagnosis prediction results more accurate. In addition, by combining the generation model to generate auxiliary diagnosis prediction results and performing consistency verification on them, the auxiliary diagnosis method and device of the present disclosure can ensure the rationality and scientificity of the diagnosis, and improve the efficiency of the diagnosis so as to formulate a diagnosis and treatment plan that is more suitable for the individual patient's situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects and features of the embodiments of the present disclosure will become more apparent through the following description in conjunction with the accompanying drawings showing the embodiments, in which:

[0019] Figure 1 is a flowchart showing an auxiliary diagnosis method implemented by a computer program according to an embodiment of the present disclosure;

[0020] Figure 2 is a flowchart illustrating construction of a patient-centric knowledge graph according to an embodiment of the present disclosure;

[0021] Figure 3 is a flow chart showing consistency checking of auxiliary diagnosis prediction results according to an embodiment of the present disclosure;

[0022] Figure 4 is a block diagram showing an auxiliary diagnosis device implemented by a computer program according to an embodiment of the present disclosure; and

[0023] Figure 5 is a block diagram illustrating a computing device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0024] The following specific embodiments are provided to help the reader obtain a comprehensive understanding of the methods, devices and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications and equivalents of the methods, devices and / or systems described herein will be clear. For example, the order of operations described herein is only an example and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art may be omitted.

[0025] The features described herein can be implemented in different forms and should not be construed as being limited to the examples described herein. Rather, the examples described herein have been provided to illustrate only some of the many possible ways to implement the methods, devices, and / or systems described herein, which will be clear after understanding the disclosure of the present application.

[0026] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more.

[0027] Although terms such as "first", "second", and "third" may be used herein to describe various members, components, regions, layers, or portions, these members, components, regions, layers, or portions should not be limited by these terms. Instead, these terms are only used to distinguish one member, component, region, layer, or portion from another member, component, region, layer, or portion. Therefore, without departing from the teachings of the examples described herein, the first member, first component, first region, first layer, or first portion referred to in the examples may also be referred to as the second member, second component, second region, second layer, or second portion.

[0028] In the specification, when an element (such as a layer, a region, or a substrate) is described as being “on”, “connected to”, or “coupled to” another element, the element may be directly “on”, “connected to”, or “coupled to” another element, or one or more other elements may be present therebetween. Conversely, when an element is described as being “directly on”, “directly connected to”, or “directly coupled to” another element, there may be no other elements present therebetween.

[0029] The terms used herein are only used to describe various examples and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. The terms "comprise", "include" and "have" indicate the presence of the described features, quantities, operations, components, elements and / or combinations thereof, but do not exclude the presence or addition of one or more other features, quantities, operations, components, elements and / or combinations thereof.

[0030] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which the present disclosure belongs after understanding the present disclosure. Unless explicitly defined as such herein, terms (such as those defined in general dictionaries) should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and should not be interpreted in an idealized or overly formal manner.

[0031] Furthermore, in the description of examples, when it is considered that a detailed description of a well-known related structure or function would cause vague interpretation of the present disclosure, such a detailed description will be omitted.

[0032] Figure 1 is a flow chart illustrating an auxiliary diagnosis method according to an embodiment of the present disclosure. Figure 2 is a flowchart illustrating construction of a patient-centric knowledge graph according to an embodiment of the present disclosure.

[0033] Reference Figure 1 In step S101, a patient-centric knowledge graph is constructed based on the medical record information in the medical record library. According to an embodiment, the medical record information in the case library may include the following: patient identification, physiological characteristics, symptoms, disease types, diagnosis and treatment methods, examination results, and medication records.

[0034] In an embodiment, the patient's identity may include the patient's name, age, gender, ID number, hospitalization number, and other basic information used to uniquely identify the patient. Physiological characteristics may include the patient's weight, height, blood type, heart rate, blood pressure, and other physiological parameters closely related to the patient's current health status. Symptoms may record the patient's main complaint and the symptoms diagnosed and observed by the doctor, for example, pain, fever, fatigue, and other information. Disease types may include the patient's medical history, chronic diseases (such as diabetes, hypertension), and current diagnosed disease information. This part of information helps to better assess the patient's comprehensive condition. Diagnostic and treatment methods may include the patient's previous and current treatment methods, such as surgery, drug therapy, physical rehabilitation, and the like. Examination results may include various laboratory tests and imaging test results, for example, blood tests, urine tests, CT scans, MRIs, and the like, providing detailed data on the patient's condition. Medication records may include the drugs used by the patient, including drug name, dosage, course of treatment, and drug allergy history, so as to avoid drug interactions or allergy risks when formulating new treatment plans.

[0035] Reference Figure 2 , the steps to build a patient-centric knowledge graph can include:

[0036] In step S201, the patient's identity, physiological characteristics, symptoms, disease type, diagnosis and treatment methods, examination results and medication records included in the medical record information are extracted as entities.

[0037] In step S202, relationship discovery is performed on the extracted entities to form triples, and the triples have the form of (head entity, relationship, tail entity).

[0038] In step S203, the triples are organized in the form of a graph to form a patient-centric knowledge graph.

[0039] For the convenience of description, only an example of triples extracted from the medical record database is given as an example:

[0040] (Patient A, suffering from, coronary heart disease); (Patient A, symptoms, chest pain); (Chest pain, examination, electrocardiogram); (Patient A, medication, drug C); (Patient A, weight, 70 kg); (Patient A, height, 175 cm); (Patient A, blood type, type O); (Patient A, heart rate, 80 beats / minute); (Patient A, age, 55 years old);

[0041] (Patient B, weight, 85 kg); (Patient B, height, 165 cm); (Patient B, age, 65 years): (Patient B, disease, hypertension stage 2); (Patient B, disease, diabetes).

[0042] The entities and relationships extracted from the medical records of these triples structuredly express the patient's medical history, symptoms, diagnosis and treatment methods, etc. In the knowledge graph, multiple similar triples are connected through different relationships to form a comprehensive information network centered on the patient. This can better support auxiliary diagnosis and help identify the patient's health status and possible treatment options.

[0043] Return to reference Figure 1 , in step S102, relevant information of the patient to be queried is extracted.

[0044] In an embodiment, the relevant information of the patient to be queried includes at least the symptoms and physiological characteristics of the patient to be queried. The symptoms and physiological characteristics of the patient to be queried have the same or similar meanings as the symptoms and physiological characteristics of the medical record information in the case library described above, and redundant descriptions are omitted here.

[0045] In step S103, the first entity matching the relevant information of the patient to be queried is retrieved in the knowledge graph.

[0046] Assuming that a patient-centric knowledge graph has been constructed with reference to step S101, if the relevant information of patient E to be queried includes symptom "chest pain" and age "55 years old", in step S103, it is necessary to retrieve the first entity matching this information in the knowledge graph.

[0047] Assume that the knowledge graph has been built, which contains many entities and relations, such as "patient", "disease", "symptom", "drug", "physiological characteristics", etc. Each entity is connected to other entities through relations (such as "suffering from", "showing", "taking medicine", etc.).

[0048] The goal of this search is to find entities related to the patient's symptom "chest pain" and age "55 years old". These entities may be diseases, examination items, possible treatments, etc. Specifically, it is necessary to find entities that have some relationship with the symptom "chest pain" in the knowledge graph. Nodes (entities) related to the symptom "chest pain" can be located in the knowledge graph. In addition, the age "55 years old" is also a filtering condition. The system will retrieve disease entities whose age is between 50 and 60 years old and whose symptoms include "chest pain".

[0049] As an example, the first entity that can be found in the search results may include a class of entities related to the symptom of "chest pain", such as "severe chest pain", "intermittent chest pain" or "chest pain with dyspnea", etc. These entities may describe different types and manifestations of chest pain, and may include non-standard symptom expressions that have the same substantive meaning.

[0050] In step S104, a subgraph is extracted from the knowledge graph, where the subgraph consists of a second entity that is a first-order neighbor of the first entity, the first-order neighbors of the second entity, and the relationship between them.

[0051] Subsequently, when extracting a subgraph consisting of a second entity as a first-order neighbor of the first entity, the first-order neighbors of the second entity, and the relationship between them from the knowledge graph, take the first entity as "chest pain" as an example: in the example where the first entity is the symptom "chest pain", the second entity is the patient connected to "chest pain". For example, "patient A", a patient with chest pain symptoms. The first-order neighbors of the second entity include further neighborhood expansion around "patient A" to extract all information directly related to "patient A". For example, the first-order neighbors of the second entity may include the physiological characteristics, symptoms, disease types, diagnosis and treatment methods, examination results, and medication records of "patient A". That is, the query result may be (patient A, age, 55 years old); (patient A, weight, 70 kg); (patient A, exhibits, fatigue); (patient A, suffers from, coronary heart disease); (patient A, medication, drug C); (patient A, examination, electrocardiogram); (patient A, surgery, heart valve reconstruction).

[0052] A subgraph of the knowledge graph is constructed through the second entity ("patient A") corresponding to the first entity ("chest pain" symptom), the first-order neighbors of the second entity, and the relationship between the first entity and its first-order neighbors, which contains entities and relationships closely related to the patient and his condition.

[0053] In step S105, the subgraph is converted into a prompt word. The prompt word is used to generate the input for model activation. The prompt word can be implemented in JSON format or in the form of natural language. The specific implementation format can be determined according to the requirements of the generated model.

[0054] These entities in the subgraph will be used to construct subsequent prompt words to help predict possible auxiliary diagnosis prediction results. In this way, entities directly related to the current symptoms and physiological characteristics of the patient to be queried can be effectively retrieved in the knowledge graph, helping the subsequent diagnosis process to become more systematic and accurate.

[0055] In step S106, the generative model is activated according to the prompt word to generate an auxiliary diagnosis prediction result. The prediction result is based on a comprehensive assessment of the patient's past medical history, current symptoms, and related entity information in the neighborhood expansion.

[0056] The generative model can be a language generation model based on deep learning, such as GPT3, GPT4 or similar pre-trained language models. The model can understand and generate text with medical logic through large-scale training of medical data. When the model is activated, the prompt word is used as input, and the model combines all the contextual information in the prompt word to generate possible diagnosis results. The generative model can process structured information (such as JSON format) and unstructured natural language information, and generate reasonable and accurate diagnosis prediction results by comprehensively analyzing various data.

[0057] Figure 3 is a flow chart showing a consistency check of auxiliary diagnosis prediction results according to an embodiment of the present disclosure.

[0058] The auxiliary diagnosis prediction method also includes step S107: using a knowledge graph representation learning method to perform consistency verification on the auxiliary diagnosis prediction results.

[0059] In step S301, the auxiliary diagnosis prediction result is integrated into the knowledge graph to obtain candidate triples corresponding to the auxiliary diagnosis prediction result.

[0060] In an embodiment, if the prediction result is to recommend that patient E exhibit symptom X take drug C, then a candidate triple (patient E, exhibits, symptom X), (patient E, medication, drug C) may be generated.

[0061] In step S302, a knowledge graph representation learning method is used to perform confidence evaluation on the candidate triples to obtain a confidence level for characterizing the consistency between the candidate triples and the knowledge graph.

[0062] In detail, in step S3021, the knowledge graph is learned using a knowledge graph representation learning method to obtain embedded representations of entities and relationships in the knowledge graph;

[0063] In an embodiment, the knowledge graph representation learning method may be a TransE method. In TransE, each entity and relationship in the knowledge graph is converted into an embedded representation (i.e., a vector representation). If the triple (h, r, t) is valid, then in the vector space, the vector h+r should be close to t, where h represents the head entity, r represents the relationship, and t represents the tail entity.

[0064] Assume that the knowledge graph contains entities such as symptom X, disease Y, and drug C, as well as relations such as "manifestation (r1)" and "drug use (r2)". After learning the knowledge graph using TransE, these entities and relations are represented as embedded representations. For example, for the sake of example, assume that after training, the following embedded representations can be obtained (the values ​​are only examples): symptom X = [0.4, 0.3, 0.5]; disease Y = [0.1, -0.2, 0.4]; drug C = [0.5, 0.1, 0.9]; (r1) = [0.1, -0.1, 0.2]; (r2) = [-0.2, 0.3, 0.1].

[0065] In step S3022, candidate embedding representations are obtained by matching entities and relations of the candidate triples with the embedding representations.

[0066] Referring to the example given in step S301, assuming that patient E has an embedded representation E = [0.2, 0.0, 0.6] in the knowledge graph, for the candidate triples (patient E, exhibiting, symptom X) and (patient E, medication, drug C), the corresponding candidate embedding representations are:

[0067] Candidate embedding representation 1: ([0.2, 0.0, 0.6], [0.1, -0.1, 0.2], [0.4, 0.3, 0.5]);

[0068] Candidate embedding representation 2: ([0.2, 0.0, 0.6], [-0.2, 0.3, 0.1], [0.5, 0.1, 0.9]).

[0069] In step S3023, a confidence assessment is performed based on the candidate embedding representation.

[0070] In an embodiment, the distance d = || (h + r) - t || is calculated, and the distance between the two candidate embedding representations is and . Set the distance threshold δ=0.6. When d≤δ, the triple is considered to have high confidence. Candidate embedding representation 1: d≈0.510≤0.6, high confidence. Candidate triple 2: d≈0.574≤0.6, high confidence.

[0071] In one or more embodiments according to the present disclosure, the most suitable representation learning method can be selected for confidence assessment according to the characteristics of the knowledge graph and application requirements. For example, TransE, TransH, TransR, ComplEx, RotatE, etc. It should be noted that different knowledge graph representation learning methods have different confidence assessment methods. Generally speaking, these methods will generate a scoring function or distance metric to measure the rationality of the triples. According to the characteristics of the specific method, select an appropriate confidence calculation method and set a reasonable threshold to ensure the reliability of the evaluation results.

[0072] In step S303, the auxiliary diagnosis prediction result whose confidence reaches the preset target is output as the prediction result.

[0073] The reliability of the diagnosis results is determined by calculating the confidence scores of the candidate triples. Only results with confidence scores that meet the preset standards will be used as the final output, ensuring the scientificity and rationality of all auxiliary diagnosis prediction results.

[0074] In the above reference Figures 1 to 3 In the described embodiment, an auxiliary diagnosis method based on a patient-centered knowledge graph and a generative model is proposed. The method constructs a knowledge graph containing information such as patient identity, physiological characteristics, symptoms, diseases, diagnostic and treatment methods, test results, and medication records, and uses retrieval enhancement technology to retrieve entities related to the patient to be queried, extract subgraphs and convert them into prompt words, and activate the generative model to generate auxiliary diagnosis prediction results. In particular, the prediction results are integrated into the knowledge graph to generate candidate triples, and the knowledge graph representation learning method (such as TransE) is used to perform confidence assessment. Only results whose confidence reaches the preset standard are output. This technical solution makes full use of the structured information of the knowledge graph and the reasoning ability of the generative model, and improves the efficiency and accuracy of information retrieval and matching through retrieval enhancement, thereby improving the accuracy and reliability of auxiliary diagnosis, and providing effective auxiliary diagnosis prediction results for clinical diagnosis.

[0075] It should be pointed out that all steps of the method are automatically executed by a computer program on a computing device. Specifically, the processes of constructing a patient-centered knowledge graph from the medical record library, extracting relevant information of the patient to be queried, retrieving matching entities in the knowledge graph, extracting subgraphs, converting subgraphs into prompt words, activating the generative model to generate auxiliary diagnosis prediction results, and performing consistency verification on the prediction results are all implemented through computer program instructions. The method does not require human intervention and relies entirely on computer automated processing, which improves the efficiency, accuracy and reliability of auxiliary diagnosis.

[0076] Figure 4is a block diagram illustrating an auxiliary diagnosis device according to an embodiment of the present disclosure.

[0077] Reference Figure 4 According to the embodiment, the auxiliary diagnosis device 400 includes a knowledge graph construction unit 401, a query information acquisition unit 402, a retrieval unit 403, a subgraph extraction unit 404, a prompt word generation unit 405, a prediction unit 406 and a consistency verification unit 407.

[0078] The knowledge graph construction unit 401 is configured to construct a patient-centric knowledge graph based on the medical record information in the medical record library. Figure 1 and Figure 2 Step S101 described.

[0079] The query information acquisition unit 402 is configured to extract relevant information of the patient to be queried, and is configured to perform a reference Figure 1 Step S102 described.

[0080] The retrieval unit 403 is configured to retrieve a first entity matching the relevant information of the patient to be queried in the knowledge graph, and is configured to perform a reference Figure 1 Step S103 described.

[0081] The subgraph extraction unit 404 is configured to extract a subgraph from the knowledge graph, the subgraph consisting of a second entity as a first-order neighbor of the first entity, the first-order neighbors of the second entity, and the relationship between them, and is configured to perform a reference Figure 1 Step S104 described.

[0082] The prompt word generating unit 405 is configured to convert the sub-graph into prompt words, and is configured to perform reference Figure 1 Step S105 described.

[0083] The prediction unit 406 is configured to activate the generation model according to the prompt word to generate an auxiliary diagnosis prediction result, and is configured to perform a reference Figure 1 Step S106 described.

[0084] The consistency check unit 407 is configured to perform the following steps: integrate the auxiliary diagnosis prediction result into the knowledge graph to obtain a candidate triple corresponding to the auxiliary diagnosis prediction result; perform confidence evaluation on the candidate triple by the knowledge graph representation learning method to obtain a confidence for characterizing the consistency between the candidate triple and the knowledge graph; and output the auxiliary diagnosis prediction result whose confidence reaches a preset target as a prediction result, and is configured to perform reference Figure 1 and Figure 3 Step S107 described.

[0085] Because already referred to Figures 1 to 3 These steps have been described and unnecessary repetition is omitted here.

[0086] Figure 5 is a block diagram illustrating a computing device according to an embodiment of the present disclosure.

[0087] Reference Figure 5 According to an embodiment of the present disclosure, the computing device 500 may include a processor 510 and a memory 520. The processor 510 may include (but not limited to) a central processing unit (CPU), a digital signal processor (DSP), a microcomputer, a field programmable gate array (FPGA), a system on a chip (SoC), a microprocessor, an application-specific integrated circuit (ASIC), etc. The memory 520 may store a computer program to be executed by the processor 510. The memory 520 includes a high-speed random access memory and / or a non-volatile computer-readable storage medium. When the processor 510 executes the computer program stored in the memory 520, the auxiliary diagnosis method as described above may be implemented.

[0088] The auxiliary diagnosis method according to the embodiment of the present disclosure can be written as a computer program and stored on a computer-readable storage medium. When the computer program is executed by a processor, the auxiliary diagnosis method as described above can be implemented. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer program and any associated data, data files and data structures in a non-transitory manner and provide the computer program and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. In one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0089] Although some embodiments of the present disclosure have been shown and described, it will be appreciated by those skilled in the art that modifications may be made to the embodiments without departing from the principles and spirit of the present disclosure, the scope of which is defined by the claims and their equivalents.

Claims

1. An auxiliary diagnosis method based on a generative model implemented by a computer program, characterized in that: The auxiliary diagnosis method comprises: Build a patient-centric knowledge graph based on the medical record information in the medical record database; Extract relevant information of the patient to be queried; Retrieving a first entity matching the relevant information of the patient to be queried in the knowledge graph; Extracting a subgraph from the knowledge graph, the subgraph consisting of a second entity that is a first-order neighbor of the first entity, a first-order neighbor of the second entity, and a relationship between them; Converting the sub-image into a prompt word; and The generative model is activated according to the prompt word to generate auxiliary diagnosis prediction results.

2. The auxiliary diagnosis method according to claim 1, characterized in that: The construction of a patient-centered knowledge graph includes: Extracting the patient's identity, physiological characteristics, symptoms, disease type, diagnosis and treatment methods, examination results and medication records included in the medical record information as entities; Performing relationship discovery on the extracted entities to form triples, wherein the triples have a form of a head entity, a relationship, and a tail entity; and The triples are organized in the form of a graph to form the patient-centered knowledge graph.

3. The auxiliary diagnosis method according to claim 1, characterized in that: The auxiliary diagnosis method also includes: Integrating the auxiliary diagnosis prediction result into the knowledge graph to obtain a candidate triple corresponding to the auxiliary diagnosis prediction result; Performing confidence evaluation on the candidate triples by a knowledge graph representation learning method to obtain a confidence for characterizing the consistency between the candidate triples and the knowledge graph; and The auxiliary diagnosis prediction result whose confidence reaches the preset target is output as the prediction result.

4. The method according to claim 3, characterized in that The confidence evaluation of the candidate triples by the knowledge graph representation learning method includes: Using the knowledge graph representation learning method to learn the knowledge graph to obtain embedded representations of entities and relationships in the knowledge graph; Obtaining a candidate embedding representation by matching entities and relations of the candidate triples with the embedding representation; and A confidence assessment is performed based on the candidate embedding representation.

5. The method according to claim 1, characterized in that The relevant information of the patient to be queried includes at least the symptoms and physiological characteristics of the patient to be queried.

6. An auxiliary diagnosis device based on a generative model implemented by a computer program, characterized in that: The auxiliary diagnosis device comprises: A knowledge graph construction unit, configured to construct a patient-centric knowledge graph based on medical record information in a medical record database; A query information acquisition unit, configured to extract relevant information of the patient to be queried; A retrieval unit, configured to retrieve a first entity matching the relevant information of the patient to be queried in the knowledge graph; a subgraph extraction unit configured to extract a subgraph from the knowledge graph, wherein the subgraph is composed of a second entity that is a first-order neighbor of the first entity, a first-order neighbor of the second entity, and a relationship between them; a prompt word generating unit, configured to convert the sub-graph into a prompt word; and The prediction unit is configured to activate the generation model according to the prompt word to generate an auxiliary diagnosis prediction result.

7. The auxiliary diagnosis device according to claim 6, characterized in that: The knowledge graph construction unit is configured to perform the following steps: Extracting the patient's identity, physiological characteristics, symptoms, disease type, diagnosis and treatment methods, examination results and medication records included in the medical record information as entities; Performing relationship discovery on the extracted entities to form a triple in the form of a head entity, a relationship, and a tail entity; as well as The triples are organized in the form of a graph to form the patient-centered knowledge graph.

8. The auxiliary diagnosis device according to claim 6, characterized in that: The auxiliary diagnosis device further comprises: a consistency checking unit, Wherein, the consistency checking unit is configured to perform the following steps: Integrating the auxiliary diagnosis prediction result into the knowledge graph to obtain a candidate triple corresponding to the auxiliary diagnosis prediction result; Performing confidence evaluation on the candidate triples by a knowledge graph representation learning method to obtain a confidence for characterizing the consistency between the candidate triples and the knowledge graph; and The auxiliary diagnosis prediction result whose confidence reaches the preset target is output as the prediction result.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the auxiliary diagnosis method according to any one of claims 1 to 5 is implemented.

10. A computing device, characterized in that: The computing device comprises: Processor; and A memory storing a computer program, which, when executed by a processor, implements the auxiliary diagnosis method according to any one of claims 1 to 5.

Images