Diagnosis and treatment abstract generation method and device, electronic equipment and storage medium
By generating a large model based on disease knowledge graphs and abstracts, content related to disease diagnosis and treatment is screened and extracted from medical record text, the problem of poor efficiency and quality of diagnosis and treatment summary generation in the prior art is solved, and more efficient and accurate diagnosis and treatment summary generation is achieved.
Patent Information
- Application Number
- CN202411843312.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-13
- Publication Date
- 2025-04-15
AI Technical Summary
In the prior art, the efficiency and quality of diagnostic and treatment abstract generation are poor, the rules and methods rely on expert maintenance, the data preparation cost of machine learning methods is high and the generalization is poor, the knowledge graph method maintenance cost is high and is greatly affected by the quality of the graph, and the large-model method lacks professional knowledge and is difficult to deal with complex medical records.
The method of generating a large model based on the disease knowledge graph and trained abstract is adopted. First, the content related to the disease diagnosis information is screened out from the medical record text, and the preliminary diagnosis and treatment summary is obtained, and then the content related to the disease diagnosis and treatment is extracted to generate the medical record diagnosis and treatment summary.
By leveraging the domain knowledge provided by the knowledge graph and the language understanding ability of the big model, a more comprehensive and accurate diagnosis and treatment summary is generated, which improves the efficiency and accuracy of diagnosis and treatment summary generation.
Smart Images
Figure CN119833109A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a diagnosis and treatment summary generation method, device, electronic device and storage medium. Background Art
[0002] With the rapid advancement of information technology, the electronic medical record system has become an indispensable part of modern medicine, which records the patient's entire diagnosis and treatment process in detail in digital form. Based on the content of the full medical record, the complete diagnosis and treatment process of a specific disease is automatically extracted and a diagnosis summary is generated, which can facilitate doctors to quickly grasp the history and treatment overview of a patient's specific condition in an emergency, and will also greatly promote the in-depth development of medical research.
[0003] In the related technologies, there are several methods for generating diagnosis and treatment summaries: extracting diagnosis and treatment summaries based on pre-defined rules, generating diagnosis and treatment summaries using machine learning methods, extracting diagnosis and treatment summaries using knowledge graphs, and generating diagnosis and treatment summaries using large models.
[0004] However, the rule-based method relies on continuous updating and maintenance by experts, the machine learning method has high data preparation costs and poor generalization, the knowledge graph method has high maintenance costs and is greatly affected by the graph quality, and the large model method lacks professional domain knowledge and has difficulty processing complex medical records and input length limitations, resulting in poor efficiency and quality in the generation of diagnosis and treatment summaries. Summary of the invention
[0005] The present invention provides a diagnosis and treatment summary generation method, device, electronic device and storage medium, which are used to solve the defects of poor efficiency and quality of diagnosis and treatment summary generation in the prior art.
[0006] The present invention provides a method for generating a diagnosis and treatment summary, comprising: Determine the disease diagnosis information for which the diagnosis and treatment summary is to be generated; Based on the disease knowledge graph, the content related to the disease diagnosis information is filtered out from the medical record text to obtain a preliminary diagnosis and treatment summary; Based on the trained summary generation model, the content related to disease diagnosis and treatment in the preliminary diagnosis and treatment summary is extracted to obtain the medical record diagnosis and treatment summary.
[0007] According to the diagnosis and treatment summary generation method provided by the present invention, based on the disease knowledge graph, the content related to the disease diagnosis information is screened out from the medical record text to obtain a preliminary diagnosis and treatment summary, including: Preprocessing the medical record text, and dividing the preprocessed medical record text into segments; Extracting knowledge entities related to the disease diagnosis information from the disease knowledge graph; The knowledge entity is matched with each fragment in the medical record text, and the preliminary diagnosis and treatment summary is determined based on each matched fragment.
[0008] According to the diagnosis and treatment summary generation method provided by the present invention, the summary generation macro model based on the training is used to extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary to obtain the medical record diagnosis and treatment summary, including: Generate a summary input text based on the summary prompt information and the preliminary diagnosis and treatment summary, wherein the summary prompt information includes task description information and instructions for generating a medical record diagnosis and treatment summary; The summary input text is input into the summary generation model, and the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary is extracted to obtain the medical record diagnosis and treatment summary output by the summary generation model.
[0009] According to the method for generating a diagnosis and treatment summary provided by the present invention, after obtaining the medical record diagnosis and treatment summary, the method further includes: Based on the disease knowledge graph, content related to the disease diagnosis information is screened out from the diagnosis and treatment project text to obtain preliminary diagnosis and treatment projects; Based on the diagnosis and treatment item verification model and the medical record diagnosis and treatment summary, the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information is verified to obtain the final diagnosis and treatment items; Based on the medical record diagnosis and treatment summary and the final diagnosis and treatment items, a comprehensive diagnosis and treatment summary is obtained.
[0010] According to the diagnosis and treatment summary generation method provided by the present invention, the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information is verified based on the diagnosis and treatment item verification large model and the medical record diagnosis and treatment summary to obtain the final diagnosis and treatment items, including: Generate a verification input text based on the diagnosis and treatment item verification instruction, the disease diagnosis information, the medical record diagnosis and treatment summary and the preliminary diagnosis and treatment items; Inputting the verification input text into the diagnosis and treatment item verification model, verifying the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information, and obtaining the verification result output by the diagnosis and treatment item verification model; Based on the verification result, the final diagnosis and treatment item is determined.
[0011] According to the diagnosis and treatment summary generation method provided by the present invention, the step of determining the disease knowledge graph includes: Extract keywords and candidate keyword pairs from disease-related texts based on statistical methods; Identify the entity relationship between the candidate keyword pairs, obtain the word pairs to be verified based on the identification results, and verify the relationship between the word pairs to be verified based on the trained relationship verification model to obtain the keyword pairs; The disease knowledge graph is constructed with the keywords as nodes and the relationships between the keyword pairs as edges.
[0012] According to the diagnosis and treatment summary generation method provided by the present invention, the statistical method extracts keywords and candidate keyword pairs from disease-related texts, including: Performing word frequency statistics on the segmented words in the text, and filtering the segmented words based on the word frequency statistics results to obtain keywords; The candidate keyword pairs are screened based on the co-occurrence frequencies of each pair of keywords in the text.
[0013] The present invention also provides a diagnosis and treatment summary generating device, comprising: A determination unit, used to determine the disease diagnosis information for which the diagnosis and treatment summary is to be generated; A screening unit, for screening out content related to the disease diagnosis information from the medical record text based on the disease knowledge graph, and obtaining a preliminary diagnosis and treatment summary; The extraction unit is used to generate a large model based on the trained summary, extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary, and obtain the medical record diagnosis and treatment summary.
[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for generating a diagnosis and treatment summary as described above is implemented.
[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for generating a diagnosis and treatment summary as described in any one of the above is implemented.
[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for generating a diagnosis and treatment summary as described above is implemented.
[0017] The diagnosis and treatment summary generation method, device, electronic device and storage medium provided by the present invention obtain a preliminary diagnosis and treatment summary by filtering out content related to disease diagnosis information from the medical record text based on the disease knowledge graph; then extracting content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary based on the trained summary generation model to obtain the medical record diagnosis and treatment summary. The domain knowledge provided by the knowledge graph can provide additional information for the summary generation model, helping the summary generation model to generate a more comprehensive and accurate summary; in addition, the preliminary diagnosis and treatment summary generated using the knowledge graph only includes content related to the current disease diagnosis, which reduces the processing scope and data volume of the summary generation model compared to the original medical record text, thereby improving the efficiency and accuracy of diagnosis and treatment summary generation. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0019] Figure 1 This is one of the flow charts of the diagnosis and treatment summary generation method provided by the present invention.
[0020] Figure 2 This is a training data sample of the summary generation large model provided by the present invention.
[0021] Figure 3 It is a schematic diagram of the text processing process of diagnosis and treatment items provided by the present invention.
[0022] Figure 4 This is a training data sample for the large model for verifying diagnosis and treatment items provided by the present invention.
[0023] Figure 5 This is a training data sample of the relationship verification model provided by the present invention.
[0024] Figure 6 This is the second flow chart of the diagnosis and treatment summary generation method provided by the present invention.
[0025] Figure 7 It is a flowchart of the knowledge graph construction method provided by the present invention.
[0026] Figure 8 It is a structural schematic diagram of the diagnosis and treatment summary generating device provided by the present invention.
[0027] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0029] Electronic medical records have not only significantly improved the accuracy of clinical decision-making and the efficiency of medical services, but also provided a rich data resource for scientific research in the medical field. However, the length of electronic medical records varies depending on the details of diagnosis and treatment covered and the complexity of the patient's condition, ranging from brief outpatient records to detailed and complex hospitalization histories. Especially for patients who have experienced long-term hospitalization and multiple department transfers, their medical records are often numerous pages, which poses a challenge to doctors' rapid review and analysis. This limitation is particularly prominent in emergency medical decision-making scenarios.
[0030] In addition, deep learning technology is increasingly used in medical text research, but due to the model's ability to process input length, direct input of the entire medical record is often difficult to achieve. More importantly, the medical record content usually includes the diagnosis and treatment process of multiple diseases of the patient, and the diagnosis and treatment overview in the discharge record section often cannot fully reflect the details of all diagnoses. Therefore, relying solely on this part of information as a basis for diagnosis may not be sufficient.
[0031] In view of this, it is particularly important to develop a technology that can automatically extract the complete diagnosis and treatment process of a specific disease based on the entire medical record content and generate a diagnosis summary. This technology will not only facilitate doctors to quickly grasp the history and treatment overview of a patient's specific condition in an emergency, but will also greatly promote the in-depth development of medical research and improve the accuracy and reliability of research results.
[0032] In the related art, there are several methods for generating diagnosis and treatment summaries: 1) Use rule-based methods to extract diagnosis and treatment summaries.
[0033] The rule-based approach requires medical experts to pre-define corresponding rules for each disease, including disease-related keywords such as symptoms, examinations, tests, and treatments, and use these rules to match information related to the current diagnosis.
[0034] Although this method is simple and fast to implement and can extract key information very accurately, it requires experts to continuously update and maintain the rules, which is time-consuming and labor-intensive, especially for complex disease conditions and diverse medical record formats.
[0035] 2) Use machine learning methods to generate diagnosis and treatment summaries.
[0036] By training machine learning models, such as support vector machines, decision trees, random forests, etc., diagnosis-related information in medical records can be automatically identified. Such methods usually require a large amount of high-quality labeled data for training, and extract diagnosis and treatment summaries by learning text features, which has high data preparation costs. When the distribution of training data is significantly different from the data distribution in actual applications, it will affect performance and generalization is poor. The internal mechanism of the model is difficult to understand, and it is difficult to explain the specific decision-making process.
[0037] 3) Use knowledge graphs to extract diagnosis and treatment summaries.
[0038] This method first needs to identify key entities such as diseases, symptoms, drugs, and test results in the medical records, and then match the identified entities with entities in the knowledge graph. It also uses the relationships in the knowledge graph to reason, verify the extracted entities and relationships, and score the entities and relationships based on their importance in the graph (such as centrality, frequency, etc.), select the entities and relationships with the highest scores, and generate summaries.
[0039] The quality of the summary generated by this method is highly dependent on the completeness and accuracy of the knowledge graph. If there are errors or missing information in the knowledge graph, the quality of the summary will be greatly affected. Therefore, the graph needs to be evaluated and updated regularly, which has high maintenance costs.
[0040] 4) Use large models to generate diagnosis and treatment summaries.
[0041] To use a large model to generate diagnosis and treatment summaries based on medical records, it is usually necessary to design appropriate prompts, that is, to directly input the task instructions and input text, namely the electronic medical record, into the large language model, and use the large model's strong language understanding and generalization capabilities to automatically generate diagnosis and treatment summaries.
[0042] Large models perform well in general fields, but may lack necessary domain knowledge in specialized fields such as medical care. In addition, inpatient medical records are usually complex, with descriptions of multiple diseases mixed together. Large models have difficulty distinguishing the diagnosis and treatment information of different diseases, resulting in incomplete or inaccurate summaries. In addition, due to the limitation of the input length of large models, it is difficult to directly input the entire medical record text for summary generation.
[0043] In response to the above problems, an embodiment of the present invention proposes a method for generating a diagnosis and treatment summary. In this method, first, based on the disease knowledge graph, the content related to the disease diagnosis information is screened out from the medical record text to obtain a preliminary diagnosis and treatment summary; then, based on the trained summary generation model, the content related to the disease diagnosis and treatment in the preliminary diagnosis and treatment summary is extracted to obtain the medical record diagnosis and treatment summary. The domain knowledge provided by the knowledge graph can provide additional information for the summary generation model, helping the summary generation model to generate a more comprehensive and accurate summary; in addition, the preliminary diagnosis and treatment summary generated by the knowledge graph only includes content related to the current disease diagnosis, which reduces the processing scope and data volume of the summary generation model compared to the original medical record text, thereby improving the efficiency and accuracy of the diagnosis and treatment summary generation.
[0044] The embodiments of the present invention can be applied to scenarios where a diagnosis and treatment summary needs to be generated. The execution subject of the method can be an electronic device such as a terminal device, a computer, a server, a server cluster, or a specially designed diagnosis and treatment summary generating device, or a diagnosis and treatment summary generating device set in the electronic device, and the diagnosis and treatment summary generating device can be implemented by software, hardware, or a combination of the two.
[0045] Figure 1 is one of the flow charts of the method for generating diagnosis and treatment summaries provided by the present invention, such as Figure 1 As shown, the method comprises the following steps: Step 110, determining the disease diagnosis information for which the diagnosis and treatment summary is to be generated.
[0046] Specifically, disease diagnosis information refers to the conclusion drawn by the doctor after systematic analysis and judgment of the patient's disease. Disease diagnosis information may include, for example, the name of the disease diagnosis, the severity of the disease, the scope of the disease, etc. Considering that electronic medical records are usually based on patients, an electronic medical record may include diagnosis information for multiple different diseases, such as pneumonia, elevated myocardial enzymes, common bile duct stones, and other disease diagnosis information.
[0047] The disease diagnosis information for which the diagnosis and treatment summary is to be generated here refers to the disease diagnosis information for which the diagnosis and treatment summary needs to be generated. For example, for Zhang San's pneumonia diagnosis, a diagnosis and treatment summary for pneumonia is generated based on Zhang San's electronic medical record.
[0048] A diagnosis and treatment summary refers to a short text that organizes, summarizes and generalizes key diagnosis and treatment information during a patient's medical treatment process, mainly including disease diagnosis and treatment.
[0049] Step 120, based on the disease knowledge graph, filter out the content related to the disease diagnosis information from the medical record text to obtain a preliminary diagnosis and treatment summary.
[0050] Specifically, the disease knowledge graph is a network structure that contains disease-related entities (such as disease names, symptoms, examination items, drugs, etc.) and their relationships. The relationships between entities can include the relationship between disease and symptoms, the relationship between disease and examination items, the relationship between disease and drugs, etc. The construction of the knowledge graph can be based on existing medical databases, literature, clinical guidelines and other resources, and extracted and integrated through natural language processing, data mining and other technologies.
[0051] The patient's medical record text can be obtained from the electronic medical record system, and the medical record text may include the chief complaint, current medical history, past history, physical examination, auxiliary examination results, etc.
[0052] For the generation of preliminary diagnosis and treatment summaries, the content related to disease diagnosis information can be screened out from the medical record text based on the disease knowledge graph. Named entity recognition can be performed on the medical record text based on the entity library in the disease knowledge graph to extract entities related to disease diagnosis information, such as disease names, symptoms, examination items, etc. Then, based on the relationship template in the disease knowledge graph, the relationship between the entities in the medical record text is extracted, such as the relationship between disease and symptoms, the relationship between disease and examination items, etc. The medical record text is then screened, and the preliminary diagnosis and treatment summary obtained includes symptoms, signs, examination items, diagnostic conclusions, etc. that are directly related to disease diagnosis. At the same time, content that is not related to disease diagnosis information can also be excluded, such as the patient's personal information, historical diseases that are not related to disease diagnosis, etc.
[0053] In other embodiments, knowledge entities related to disease diagnosis information may be first extracted from the disease knowledge graph, and then the knowledge entities may be matched with the medical record text to obtain a preliminary diagnosis and treatment summary.
[0054] Step 130, based on the trained summary generation large model, extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary to obtain the medical record diagnosis and treatment summary.
[0055] Specifically, considering that the preliminary diagnosis and treatment summary generated by the disease knowledge graph is still not short enough and is still too long for doctors to quickly browse or conduct subsequent analysis, a large language model can be used to further compress the preliminary diagnosis and treatment summary based on the preliminary diagnosis and treatment summary.
[0056] Here, a small amount of annotated data can be pre-constructed to fine-tune the instructions of the large language model, and a trained summary generation model can be obtained. The summary generation model trained in this way can extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary, so as to compress the length of the preliminary diagnosis and treatment summary and obtain the medical record diagnosis and treatment summary.
[0057] According to the method provided by the embodiment of the present invention, the domain knowledge provided by the knowledge graph can provide additional information for the summary generation model, thereby helping the summary generation model to generate a more comprehensive and accurate summary. In addition, the preliminary diagnosis and treatment summary generated by the knowledge graph only includes content related to the current disease diagnosis. Compared with the original medical record text, the processing scope and data volume of the summary generation model are reduced, thereby improving the efficiency and accuracy of the diagnosis and treatment summary generation.
[0058] Based on the above embodiment, based on the disease knowledge graph, the content related to the disease diagnosis information is screened out from the medical record text to obtain a preliminary diagnosis and treatment summary, that is, step 120 specifically includes: Step 121, preprocessing the medical record text, and dividing the preprocessed medical record text into segments; Step 122, extracting knowledge entities related to disease diagnosis information from the disease knowledge graph; Step 123, matching the knowledge entity with each segment in the medical record text, and determining a preliminary diagnosis and treatment summary based on each matched segment.
[0059] Specifically, the preprocessing of medical record text mainly includes the following aspects: ① Removing information in the medical record text that is not related to the diagnosis and treatment of the disease, such as the patient's name, medical staff's name, hospital name, etc.; ② Converting the fields represented by codes in the medical record text into the corresponding meanings of the codes. For example, the gender field in the medical record is represented by "1" and "2", which need to be converted into their corresponding meanings "male" and "female" respectively.
[0060] Each field of the medical record text is then divided into smaller segments by periods, which serve as the smallest unit for generating a preliminary diagnosis and treatment summary.
[0061] In step 122, the disease diagnosis name diag of the summary is first generated as needed, all nodes related to the disease diagnosis name diag are queried in the knowledge graph, and the node names and relationship labels in the query results are extracted and sorted into a keyword list, that is, the knowledge entities related to the disease diagnosis information are obtained.
[0062] These keywords are used as matching items to match each fragment in the medical record text. All matched fragments are preliminary diagnosis and treatment summaries. The results are saved, and the medical record chapters and field information corresponding to the fragments are recorded.
[0063] The method provided by the embodiment of the present invention can further improve the accuracy of generating preliminary diagnosis and treatment summaries by first extracting knowledge entities related to disease diagnosis information from the disease knowledge graph, then matching the knowledge entities with each fragment in the medical record text, and determining a preliminary diagnosis and treatment summary based on the matched fragments.
[0064] Based on any of the above embodiments, based on the trained summary generation model, the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary is extracted to obtain the medical record diagnosis and treatment summary, that is, step 130 specifically includes: Step 131, generating a summary input text based on the summary prompt information and the preliminary diagnosis and treatment summary, wherein the summary prompt information includes task description information and instructions for generating a medical record diagnosis and treatment summary; Step 132, input the summary input text into the summary generation model, extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary, and obtain the medical record diagnosis and treatment summary output by the summary generation model.
[0065] Specifically, the realization of medical record diagnosis and treatment summaries can be achieved through the trained summary generation model. The summary generation model is used to extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary, so as to further summarize and simplify the text of the preliminary diagnosis and treatment summary.
[0066] Here, the summary prompt information and the preliminary diagnosis and treatment summary can be integrated to generate a summary input text. The summary prompt information includes task description information and instructions for generating a medical record diagnosis and treatment summary. The task description information is used to describe the task content of the summary generation model. The task description information may include role descriptions, task requirements, input and output format requirements, etc. The instructions are used to instruct the summary generation model to generate a medical record diagnosis and treatment summary.
[0067] The summary input text is then input into the summary generation model to extract the content related to disease diagnosis and treatment in the preliminary diagnosis and treatment summary, and obtain the medical record diagnosis and treatment summary output by the summary generation model.
[0068] The large model for summary generation is fine-tuned based on the large language model. If only a few shot method is used, the output format of the large model is difficult to control, which affects the analysis and evaluation of the output results. Therefore, it is necessary to construct a small amount of labeled data to fine-tune the task of extracting content related to disease diagnosis and treatment in the preliminary diagnosis and treatment summary.
[0069] Figure 2 is a training data sample of the summary generation model provided by the present invention, such as Figure 2 As shown, the training data includes four parts: role description, task description, input text (which may include disease diagnosis information and preliminary diagnosis and treatment summary), and output.
[0070] Based on any of the above embodiments, after obtaining the medical record diagnosis and treatment summary in step 130, the method further includes: Step 140, based on the disease knowledge graph, filter out the content related to the disease diagnosis information from the diagnosis and treatment project text to obtain preliminary diagnosis and treatment projects; Step 150, based on the diagnosis and treatment item verification model and the medical record diagnosis and treatment summary, the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information is verified to obtain the final diagnosis and treatment items; Step 160, obtaining a comprehensive diagnosis and treatment summary based on the medical record diagnosis and treatment summary and the final diagnosis and treatment items.
[0071] Specifically, considering that patients will undergo some examinations and tests during the medical treatment process, which are usually not reflected in the medical record text, in order to further enrich and improve the diagnosis and treatment summary, text extraction can be performed on the patient's diagnosis and treatment item data, thereby supplementing the medical record diagnosis and treatment summary with some diagnosis and treatment item information, that is, integrating the medical record diagnosis and treatment summary and the final diagnosis and treatment items to obtain a comprehensive diagnosis and treatment summary.
[0072] In step 140, the diagnosis and treatment project data is first preprocessed to obtain the diagnosis and treatment project text. The diagnosis and treatment project data mainly include the patient's checklist, test list, doctor's order, settlement list, etc. These contents are usually in a structured form and need to be converted into a coherent natural language form before the language model is processed. The main thing is to remove the fields that are not related to the diagnosis and treatment, and adjust or splice the order of the fields according to the semantics. Taking the test list as an example, Figure 3 It is a schematic diagram of the text processing process of the diagnosis and treatment items provided by the present invention, such as Figure 3 As shown, for the "three thyroid function" test results, they are converted into a coherent natural language form.
[0073] After obtaining the diagnosis and treatment project text, based on the disease knowledge graph, filter out the content related to the disease diagnosis information from the diagnosis and treatment project text to obtain the preliminary diagnosis and treatment project. The screening process here is the same as the screening process of the preliminary diagnosis and treatment summary. Use the knowledge entities related to the disease diagnosis information extracted from the disease knowledge graph as matching items to match each segment of the diagnosis and treatment project text. All the matched segments are the preliminary diagnosis and treatment project extraction results, and the results are saved.
[0074] Considering that the standard for extracting diagnosis and treatment items based on the knowledge graph is relatively loose, all diagnosis and treatment items related to the current diagnosis are extracted, but often a diagnosis and treatment item is related to multiple diagnoses, and judging only by the name of the diagnosis and treatment item is not accurate enough. In order to further improve the accuracy of the extracted diagnosis and treatment items and their relevance to the current diagnosis, it is necessary to further verify the initially extracted diagnosis and treatment items using the diagnosis and treatment item verification model in combination with the current diagnosis and the diagnosis and treatment summary generated by the medical record.
[0075] Here, the large model for diagnosis and treatment project verification is obtained after fine-tuning the instructions based on the large language model, so that the large model for diagnosis and treatment project verification can better understand this task and follow the specified output format. The preliminary diagnosis and treatment items and medical record diagnosis and treatment summaries are input into the verification large model. The model will analyze the information in the preliminary diagnosis and treatment items and the medical record diagnosis and treatment summaries and evaluate the correlation between them. According to the results of the model output, the preliminary diagnosis and treatment items are screened and adjusted to ensure that the final diagnosis and treatment items are highly consistent and reasonable with the information in the medical record diagnosis and treatment summaries.
[0076] According to the results of the correlation check, a final list of diagnosis and treatment items is generated. This list only contains diagnosis and treatment items that are highly relevant and reasonable to the medical record diagnosis and treatment summary.
[0077] Then, the medical record diagnosis and treatment summary and the final diagnosis and treatment items are integrated to form a comprehensive diagnosis and treatment summary for the disease, which includes the patient's entire diagnosis and treatment process and the final diagnosis and treatment items.
[0078] The method provided by the embodiment of the present invention improves the accuracy of extraction of diagnosis and treatment items by utilizing the medical record diagnosis and treatment summary to assist the large model in extracting relevant diagnosis and treatment items, and uses the medical record diagnosis and treatment summary and the extracted related diagnosis and treatment items together as the summary of the current diagnosis, making the summary more complete.
[0079] Based on any of the above embodiments, in step 150, based on the diagnosis and treatment item verification model and the medical record diagnosis and treatment summary, the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information is verified to obtain the final diagnosis and treatment items, including: Step 151, generating a verification input text based on the diagnosis and treatment item verification instruction, disease diagnosis information, medical record diagnosis and treatment summary and preliminary diagnosis and treatment items; Step 152, input the verification input text into the diagnosis and treatment item verification model, verify the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information, and obtain the verification result output by the diagnosis and treatment item verification model; Step 153, determine the final diagnosis and treatment items based on the verification results.
[0080] Specifically, the diagnosis and treatment project verification instruction is used to instruct the large language model to generate a course design plan instruction. The instruction here is in the form of natural language text, for example, it can be "Please determine whether the diagnosis and treatment project is for the disease based on the input disease and medical record diagnosis summary?".
[0081] Disease diagnosis information can be the name of the disease diagnosis. The medical record diagnosis and treatment summary is the diagnosis and treatment summary generated based on the medical record text. The preliminary diagnosis and treatment items are obtained by matching the diagnosis and treatment item text through the disease knowledge graph.
[0082] On this basis, based on the diagnosis and treatment item verification instructions, disease diagnosis information, medical record diagnosis and treatment summary and preliminary diagnosis and treatment items, the verification input text is generated. For example, the diagnosis and treatment item verification instructions, disease diagnosis information, medical record diagnosis and treatment summary and preliminary diagnosis and treatment items can be spliced to obtain the verification input text; or the diagnosis and treatment item verification instructions, disease diagnosis information, medical record diagnosis and treatment summary and preliminary diagnosis and treatment items can be filled into a pre-set input text template to obtain the verification input text, which is not specifically limited in the embodiment of the present invention.
[0083] After obtaining the verification input text, the verification input text can be input into the diagnosis and treatment item verification model, and the diagnosis and treatment item verification model verifies the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information to obtain the verification result.
[0084] The verification result here may include that the preliminary diagnosis and treatment items are related to the disease diagnosis information, that is, the preliminary diagnosis and treatment items are targeted at the disease diagnosis information; or it may include that they are not related, that is, the preliminary diagnosis and treatment items are not targeted at the disease diagnosis information. At the same time, the diagnosis and treatment item verification model can also output the reasons for being related or not related. Figure 4 is a training data sample of the diagnosis and treatment project verification model provided by the present invention, such as Figure 4 As shown in the figure, the data construction method of thinking chain can enable the large model to better understand the reasoning process of the task and improve the performance of the large model on the task.
[0085] The final diagnosis and treatment items can then be determined based on the verification results. It is understandable that the final diagnosis and treatment items are all diagnosis and treatment items related to the disease diagnosis information.
[0086] The method provided by the embodiment of the present invention constructs a verification input text based on diagnosis and treatment item verification instructions, disease diagnosis information, medical record diagnosis and treatment summaries and preliminary diagnosis and treatment items, so that the big model can better understand the reasoning process of the task of verifying the correlation between preliminary diagnosis and treatment items and disease diagnosis information, thereby improving the performance of the big model on this task.
[0087] Based on any of the above embodiments, the step of determining the disease knowledge graph includes: Step 210, extracting keywords and candidate keyword pairs from disease-related texts based on a statistical method; Step 220, identifying entity relationships between candidate keyword pairs, obtaining to-be-verified keyword pairs based on the identification results, and verifying the relationships between the to-be-verified keyword pairs based on the trained relationship verification model to obtain keyword pairs; Step 230, constructing a disease knowledge graph with keywords as nodes and relationships between keyword pairs as edges.
[0088] Specifically, since the disease knowledge graph constructed in the embodiment of the present invention is mainly used for the extraction of diagnosis and treatment basis in medical records, it is hoped that the graph covers as many diseases as possible. First, collect texts related to the disease. When collecting data, select a variety of data sources covering disease-related information, such as ① authoritative databases: medical literature databases, medical books, clinical diagnosis and treatment guidelines, medical terms (such as ICD-10, SNOMED CT); ② clinical data: electronic medical records, diagnosis and treatment projects (usually refers to a series of examinations, treatment measures and medical services for disease diagnosis and treatment during the medical process), management and charging of diagnosis and treatment projects (such as medical advice, settlement lists, etc.); ③ medical websites, etc. These data sources should be authoritative, reliable and extensive to ensure that the collected disease information is comprehensive and accurate.
[0089] All collected text data is cleaned to remove some noise and irrelevant information in the text, such as special characters, advertising content, etc. At the same time, the text is normalized, such as unifying upper and lower case, removing duplicate content, removing stop words, etc., to improve the accuracy of subsequent processing.
[0090] Then, based on statistical methods, keywords and candidate keyword pairs are extracted from disease-related texts, including: Perform word frequency statistics on the words in the text, and filter the words based on the word frequency statistics to obtain keywords; Based on the co-occurrence frequency of each pair of keywords in the text, candidate keyword pairs are screened.
[0091] The cleaned text data is segmented to split the continuous text into individual words or phrases. Existing Chinese word segmentation tools, such as jieba and HanLP, can be used to make appropriate adjustments and optimizations based on the characteristics of the medical field. Using a deep learning model (such as BiLSTM+CRF) to label the part of speech of each word will help with subsequent keyword extraction.
[0092] First, we perform word frequency statistics on the text data after word segmentation and build a word frequency dictionary to record the number of times each word appears in the text.
[0093] Based on the results of word frequency statistics, high-frequency keywords are selected as nodes of the disease knowledge graph. A word frequency threshold can be set to identify words with a frequency higher than the threshold as candidate keywords. Combined with knowledge and experience in the medical field, keywords are further screened and optimized to remove some meaningless or irrelevant words to medicine. Based on part-of-speech tagging, possible keywords such as nouns and verbs are filtered out. Combined with medical field terminology (such as SNOMED CT, ICD-10), keywords that meet the medical field are filtered out.
[0094] Calculate the co-occurrence frequency of each pair of keywords in the text in the previous step and generate a co-occurrence matrix. Set the co-occurrence frequency threshold and filter out high-frequency co-occurring word pairs as candidate keyword pairs.
[0095] In step 220, the relationship between entities is identified from the candidate keyword pairs, and the relationship between them is determined by analyzing the context of the word pairs, such as disease-symptom relationship, disease-examination relationship, disease-treatment relationship, disease-part relationship, etc. A neural network model (such as BERT) can be used to automatically learn and extract complex relationships between entities.
[0096] The recognition result here can be the confidence of the model prediction result. To improve the accuracy of entity relationships, when the confidence of the model prediction result is less than the threshold When the confidence of the model prediction result is greater than the threshold When , the candidate keyword pair and relationship are retained; the confidence of the model prediction result is arrive The words in between can be used as word pairs to be verified.
[0097] Then, based on the trained large-scale relationship verification model, the relationship between the pairs of words to be verified is verified to obtain the keyword pairs. The large-scale relationship verification model is obtained by fine-tuning the instructions based on the large-scale language model. Figure 5 is a training data sample of the relationship verification model provided by the present invention, such as Figure 5 As shown in the figure, the instruction includes role description, task description, demonstration examples, and sample word pairs to be verified. The labeled data is constructed into demonstration examples, each sample includes input and corresponding correct output. The sample word pairs to be verified are the confidence levels of the relationship predictions in the previous step. arrive The word pairs between .
[0098] In step 230, a disease knowledge graph is constructed with keywords as nodes and relationships between keyword pairs as edges. The selected keywords are used as nodes of the disease knowledge graph. Each node can be assigned a unique identifier and its attribute information such as name and word frequency can be recorded. According to the relationship between keyword pairs, edges are established between related nodes and the edge types are marked. A suitable graph database (such as Neo4j, ArangoDB) is selected to store graph data.
[0099] In addition, to solve the problem that the disease knowledge graph needs to be regularly evaluated and updated, the following measures can be taken: ① Introduce a user feedback mechanism: Encourage medical professionals to provide feedback and point out errors and deficiencies in the knowledge graph. Establish an online platform to allow doctors, researchers and patients to participate in the construction of the knowledge graph and form a good interaction and feedback loop; ② Regular tracking and automatic updates: On the one hand, medical experts regularly track the latest medical research, guideline updates, etc. to ensure the timeliness and accuracy of the information; on the other hand, develop an automatic update system, use data mining and other technologies to automatically obtain the latest medical information, and integrate it into the disease knowledge graph.
[0100] The method provided by the embodiment of the present invention is a knowledge graph construction method based on statistics + large model verification, which makes the constructed knowledge graph more accurate.
[0101] Based on any of the above embodiments, Figure 6 This is the second flow chart of the method for generating a diagnosis and treatment summary provided by the present invention. Figure 6 As shown, a method for generating a diagnosis and treatment summary is provided, comprising: S1, build disease knowledge graph. Figure 7 It is a schematic diagram of the process of the knowledge graph construction method provided by the present invention, such as Figure 7 As shown, constructing a disease knowledge graph specifically includes: obtaining text related to the disease, performing word frequency statistics on the segmented words in the text, and filtering the segmented words to obtain keywords based on the word frequency statistics results; based on the co-occurrence frequency of each pair of keywords in the text, filtering candidate keyword pairs.
[0102] Identify the entity relationship between candidate keyword pairs, obtain the word pairs to be verified based on the recognition results, and verify the relationship between the word pairs to be verified based on the trained relationship verification model to obtain keyword pairs; construct a disease knowledge graph with keywords as nodes and the relationship between keyword pairs as edges.
[0103] S2, determining the disease diagnosis information for which the diagnosis and treatment summary is to be generated.
[0104] S3, based on the disease knowledge graph, screen out the content related to disease diagnosis information from the medical record text to obtain a preliminary diagnosis and treatment summary. Specifically, it includes: preprocessing the medical record text and dividing the preprocessed medical record text into segments; extracting knowledge entities related to disease diagnosis information from the disease knowledge graph; matching the knowledge entities with each segment in the medical record text, and determining the preliminary diagnosis and treatment summary based on the matched segments.
[0105] S4, based on the trained summary generation model, extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary to obtain the medical record diagnosis and treatment summary. Specifically, it includes: generating summary input text based on summary prompt information and preliminary diagnosis and treatment summary, the summary prompt information includes task description information and instructions for generating medical record diagnosis and treatment summary; inputting the summary input text into the summary generation model, extracting the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary, and obtaining the medical record diagnosis and treatment summary output by the summary generation model.
[0106] S5, based on the disease knowledge graph, filters out the content related to disease diagnosis information from the diagnosis and treatment project text to obtain preliminary diagnosis and treatment projects.
[0107] S6, based on the diagnosis and treatment project verification model and the medical record diagnosis and treatment summary, the correlation between the preliminary diagnosis and treatment project and the disease diagnosis information is verified to obtain the final diagnosis and treatment project. Specifically, it includes: generating a verification input text based on the diagnosis and treatment project verification instructions, disease diagnosis information, medical record diagnosis and treatment summary and preliminary diagnosis and treatment project; inputting the verification input text into the diagnosis and treatment project verification model, verifying the correlation between the preliminary diagnosis and treatment project and the disease diagnosis information, and obtaining the verification result output by the diagnosis and treatment project verification model; and determining the final diagnosis and treatment project based on the verification result.
[0108] S7, based on the medical record diagnosis and treatment summary and the final diagnosis and treatment items, obtain the comprehensive diagnosis and treatment summary.
[0109] The method provided by the embodiment of the present invention has a higher accuracy rate in constructing a knowledge graph based on the knowledge graph construction method of statistics + big model verification; the method of using the knowledge graph to generate a preliminary diagnosis and treatment summary and further compressing it using the big model can use the domain knowledge provided by the knowledge graph to provide additional information for the summary generation big model, and help the summary generation big model to generate a more comprehensive and accurate summary; use the medical record-based medical record diagnosis and treatment summary to assist the diagnosis and treatment item verification big model to extract relevant diagnosis and treatment items, improve the accuracy of diagnosis and treatment item extraction, and use the medical record diagnosis and treatment summary and the extracted related diagnosis and treatment items together as the comprehensive diagnosis and treatment summary of the current diagnosis, so that the summary is more complete.
[0110] The following is a description of the diagnosis and treatment summary generating device provided by the present invention. The diagnosis and treatment summary generating device described below and the diagnosis and treatment summary generating method described above can be referenced to each other.
[0111] Based on any of the above embodiments, Figure 8 is a schematic diagram of the structure of the diagnosis and treatment summary generating device provided by the present invention, such as Figure 8 As shown, the diagnosis and treatment summary generating device includes: A determination unit 810 is used to determine the disease diagnosis information for which the diagnosis and treatment summary is to be generated; A screening unit 820 is used to screen out content related to the disease diagnosis information from the medical record text based on the disease knowledge graph to obtain a preliminary diagnosis and treatment summary; The extraction unit 830 is used to generate a large model based on the trained summary, extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary, and obtain the medical record diagnosis and treatment summary.
[0112] The device provided by the embodiment of the present invention firstly filters out the content related to the disease diagnosis information from the medical record text based on the disease knowledge graph to obtain a preliminary diagnosis and treatment summary; then, based on the trained summary generation model, extracts the content related to the disease diagnosis and treatment in the preliminary diagnosis and treatment summary to obtain the medical record diagnosis and treatment summary. The domain knowledge provided by the knowledge graph can provide additional information for the summary generation model, helping the summary generation model to generate a more comprehensive and accurate summary; in addition, the preliminary diagnosis and treatment summary generated by the knowledge graph only includes the content related to the current disease diagnosis, which reduces the processing scope and data volume of the summary generation model compared to the original medical record text, thereby improving the efficiency and accuracy of the diagnosis and treatment summary generation.
[0113] Based on the above embodiment, the screening unit is specifically used for: Preprocessing the medical record text, and dividing the preprocessed medical record text into segments; Extracting knowledge entities related to the disease diagnosis information from the disease knowledge graph; The knowledge entity is matched with each fragment in the medical record text, and the preliminary diagnosis and treatment summary is determined based on each matched fragment.
[0114] Based on the above embodiment, the extraction unit is specifically used for: Generate a summary input text based on the summary prompt information and the preliminary diagnosis and treatment summary, wherein the summary prompt information includes task description information and instructions for generating a medical record diagnosis and treatment summary; The summary input text is input into the summary generation model, and the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary is extracted to obtain the medical record diagnosis and treatment summary output by the summary generation model.
[0115] Based on the above embodiment, the diagnosis and treatment summary generating device further includes a synthesis unit, which is used to: Based on the disease knowledge graph, content related to the disease diagnosis information is screened out from the diagnosis and treatment project text to obtain preliminary diagnosis and treatment projects; Based on the diagnosis and treatment item verification model and the medical record diagnosis and treatment summary, the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information is verified to obtain the final diagnosis and treatment items; Based on the medical record diagnosis and treatment summary and the final diagnosis and treatment items, a comprehensive diagnosis and treatment summary is obtained.
[0116] Based on the above embodiment, the integration unit is specifically used for: Generate a verification input text based on the diagnosis and treatment item verification instruction, the disease diagnosis information, the medical record diagnosis and treatment summary and the preliminary diagnosis and treatment items; Inputting the verification input text into the diagnosis and treatment item verification model, verifying the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information, and obtaining the verification result output by the diagnosis and treatment item verification model; Based on the verification result, the final diagnosis and treatment item is determined.
[0117] Based on the above embodiment, the diagnosis and treatment summary generating device further includes a graph construction unit, which is used to: Extract keywords and candidate keyword pairs from disease-related texts based on statistical methods; Identify the entity relationship between the candidate keyword pairs, obtain the word pairs to be verified based on the identification results, and verify the relationship between the word pairs to be verified based on the trained relationship verification model to obtain the keyword pairs; The disease knowledge graph is constructed with the keywords as nodes and the relationships between the keyword pairs as edges.
[0118] Based on the above embodiment, the graph construction unit is specifically used for: Performing word frequency statistics on the segmented words in the text, and filtering the segmented words based on the word frequency statistics results to obtain keywords; The candidate keyword pairs are screened based on the co-occurrence frequencies of each pair of keywords in the text.
[0119] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9 As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930 and a communication bus 940, wherein the processor 910, the communication interface 920 and the memory 930 communicate with each other through the communication bus 940. The processor 910 may call the logic instructions in the memory 930 to execute the diagnosis and treatment summary generation method, which includes: determining the disease diagnosis information for which the diagnosis and treatment summary is to be generated; based on the disease knowledge graph, filtering out the content related to the disease diagnosis information from the medical record text to obtain a preliminary diagnosis and treatment summary; based on the trained summary generation model, extracting the content related to the disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary to obtain the medical record diagnosis and treatment summary.
[0120] In addition, the logic instructions in the above-mentioned memory 930 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0121] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the diagnosis and treatment summary generation method provided by the above-mentioned methods, and the method includes: determining the disease diagnosis information for which the diagnosis and treatment summary is to be generated; based on the disease knowledge graph, filtering out content related to the disease diagnosis information from the medical record text to obtain a preliminary diagnosis and treatment summary; based on the trained summary generation large model, extracting content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary to obtain a medical record diagnosis and treatment summary.
[0122] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the diagnosis and treatment summary generation method provided by the above-mentioned methods, the method comprising: determining the disease diagnosis information for which the diagnosis and treatment summary is to be generated; based on the disease knowledge graph, filtering out content related to the disease diagnosis information from the medical record text to obtain a preliminary diagnosis and treatment summary; based on a trained summary generation large model, extracting content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary to obtain a medical record diagnosis and treatment summary.
[0123] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0124] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for generating a diagnosis and treatment summary, characterized in that: include: Determine the disease diagnosis information for which the diagnosis and treatment summary is to be generated; Based on the disease knowledge graph, the content related to the disease diagnosis information is filtered out from the medical record text to obtain a preliminary diagnosis and treatment summary; Based on the trained summary generation model, the content related to disease diagnosis and treatment in the preliminary diagnosis and treatment summary is extracted to obtain the medical record diagnosis and treatment summary.
2. The method for generating a diagnosis and treatment summary according to claim 1, characterized in that: Based on the disease knowledge graph, the content related to the disease diagnosis information is screened out from the medical record text to obtain a preliminary diagnosis and treatment summary, including: Preprocessing the medical record text, and dividing the preprocessed medical record text into segments; Extracting knowledge entities related to the disease diagnosis information from the disease knowledge graph; The knowledge entity is matched with each fragment in the medical record text, and the preliminary diagnosis and treatment summary is determined based on each matched fragment.
3. The method for generating a diagnosis and treatment summary according to claim 2, characterized in that: The large model for generating summary based on the training is used to extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary to obtain the medical record diagnosis and treatment summary, including: Generate a summary input text based on the summary prompt information and the preliminary diagnosis and treatment summary, wherein the summary prompt information includes task description information and instructions for generating a medical record diagnosis and treatment summary; The summary input text is input into the summary generation model, and the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary is extracted to obtain the medical record diagnosis and treatment summary output by the summary generation model.
4. The method for generating a diagnosis and treatment summary according to claim 1, characterized in that: After obtaining the medical record diagnosis and treatment summary, the method further includes: Based on the disease knowledge graph, content related to the disease diagnosis information is screened out from the diagnosis and treatment project text to obtain preliminary diagnosis and treatment projects; Based on the diagnosis and treatment item verification model and the medical record diagnosis and treatment summary, the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information is verified to obtain the final diagnosis and treatment items; Based on the medical record diagnosis and treatment summary and the final diagnosis and treatment items, a comprehensive diagnosis and treatment summary is obtained.
5. The method for generating a diagnosis and treatment summary according to claim 4, characterized in that: The verification of the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information based on the diagnosis and treatment item verification big model and the medical record diagnosis and treatment summary to obtain the final diagnosis and treatment items includes: Generate a verification input text based on the diagnosis and treatment item verification instruction, the disease diagnosis information, the medical record diagnosis and treatment summary and the preliminary diagnosis and treatment items; Inputting the verification input text into the diagnosis and treatment item verification model, verifying the correlation between the preliminary diagnosis and treatment items and the disease diagnosis information, and obtaining the verification result output by the diagnosis and treatment item verification model; Based on the verification result, the final diagnosis and treatment item is determined.
6. The method for generating a diagnosis and treatment summary according to any one of claims 1 to 5, characterized in that: The steps of determining the disease knowledge graph include: Extract keywords and candidate keyword pairs from disease-related texts based on statistical methods; Identify the entity relationship between the candidate keyword pairs, obtain the word pairs to be verified based on the identification results, and verify the relationship between the word pairs to be verified based on the trained relationship verification model to obtain the keyword pairs; The disease knowledge graph is constructed with the keywords as nodes and the relationships between the keyword pairs as edges.
7. The method for generating a diagnosis and treatment summary according to claim 6, characterized in that: The statistical method extracts keywords and candidate keyword pairs from disease-related texts, including: Performing word frequency statistics on the segmented words in the text, and filtering the segmented words based on the word frequency statistics results to obtain keywords; The candidate keyword pairs are screened based on the co-occurrence frequencies of each pair of keywords in the text.
8. A diagnosis and treatment summary generating device, characterized in that: include: A determination unit, used to determine the disease diagnosis information for which the diagnosis and treatment summary is to be generated; A screening unit, for screening out content related to the disease diagnosis information from the medical record text based on the disease knowledge graph, and obtaining a preliminary diagnosis and treatment summary; The extraction unit is used to generate a large model based on the trained summary, extract the content related to disease diagnosis and disease treatment in the preliminary diagnosis and treatment summary, and obtain the medical record diagnosis and treatment summary.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method for generating a diagnosis and treatment summary according to any one of claims 1 to 7 is implemented.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating a diagnosis and treatment summary as claimed in any one of claims 1 to 7 is implemented.