Glioma diagnosis and treatment information analysis method and system based on large model

Abstract

The invention relates to the technical field of intelligent medical treatment, in particular to a glioma diagnosis and treatment information analysis method and system based on a large model, and the method comprises the following steps: obtaining disease information, carrying out preprocessing and structured processing, and generating a structural medical record report; the structural medical record report is subjected to preliminary diagnosis through a doctor expert Agent, and a preliminary diagnosis suggestion is generated; the preliminary diagnosis suggestions are concluded and summarized through an MDT Leader Agent, and final diagnosis suggestions are generated; carrying out keyword extraction on the final diagnosis suggestion by a network search Agent, and carrying out related clinical test information retrieval on the keyword in a world clinical test registration database to obtain retrieval information; and outputting the final diagnosis suggestion and the retrieval information. According to the application, standardization and individualization of diagnosis and treatment can be ensured, and the efficiency of glioma disease diagnosis is improved.

Description

Technical Field

[0001] This application belongs to the field of intelligent medical technology, and particularly relates to a glioma diagnosis and treatment information analysis method and system based on a large model. Background Art

[0002] Glioma is a tumor originating from glial cells in the brain, accounting for 40% - 60% of all primary central nervous system tumors, and is well-known as the most common primary intracranial tumor in adults. Glioma is a chronic and refractory serious disease, with a complex diagnosis and treatment process, numerous factors affecting diagnosis and treatment decisions, and rapid progress in diagnosis and treatment methods; it requires experts with many years of clinical experience to master and apply it proficiently in the diagnosis and treatment process of patients; at the same time, since the diagnosis and treatment process involves many related professional fields, multi-disciplinary diagnosis and treatment are required to put forward standardized and individualized diagnosis and treatment opinions and plans. In view of the above problems, in the current diagnosis and treatment process, there will be difficulties in clinical expert decision-making, flawed decision results, or even misdiagnosis and mistreatment due to reasons such as incomplete clinical information, fragmented information, inaccurate information, limited time to obtain information, information involving multiple specialties, and complex interaction between information.

[0003] In addition, with the high incidence of glioma in modern society, a situation has formed where medical resources are relatively scarce, resulting in relatively low diagnosis and treatment efficiency, and poor standardization and individuation of diagnosis and treatment plans. Especially in the initial diagnosis of glioma and the determination of treatment plans, on the one hand, patients and their families run around, looking for acquaintances and relationships, spending a lot of time, energy, and cost; on the other hand, because patients cannot provide systematic and standardized examination materials, experts cannot give opinions on the diagnosis and treatment of their own specialties within a limited time; moreover, the diagnosis and treatment of cancer involve multiple disciplines, and a multi-disciplinary expert team is required to provide standardized and individualized plans. Summary of the Invention

[0004] In view of one or more of the problems existing in the prior art, this application proposes a glioma diagnosis and treatment information analysis method based on a large model, including the following steps: Obtain disease information, perform preprocessing and structured processing on the disease information to generate a structured medical record report; The structured medical record report is initially diagnosed by a doctor expert Agent to generate an initial diagnosis suggestion; The initial diagnosis suggestions are summarized by an MDT Leader Agent to generate a final diagnosis suggestion; Extract the keywords of the final diagnosis by a network search Agent, and retrieve relevant clinical trial information in the World Clinical Trials Registration Database to obtain retrieval information; Output the final diagnosis suggestion and the retrieval information.

[0005] Preferably, the disease information includes text information and picture information; the text information is clinical information related to suspected glioma; the text information includes the written description or report of the chief complaint, current medical history, past medical history, physical examination, tests, and examinations; the picture information is imaging pictures or pathological pictures.

[0006] Preferably, preprocessing the disease information includes: Translating the picture information into a written description; Performing noise reduction processing and missing value filling processing on the written descriptions of the text information and picture information.

[0007] Preferably, the preprocessed text information is structurally processed by the resident large model Agent to generate a structured medical record report.

[0008] Preferably, the large model includes an open-source large model and a closed-source large model.

[0009] Preferably, the doctor expert Agent is constructed based on glioma diagnosis and treatment specifications or guidelines.

[0010] Preferably, the doctor expert Agent includes a neurology doctor Agent, a radiology doctor Agent, a pathology doctor Agent, a neurosurgery doctor Agent, a radiotherapy and chemotherapy doctor Agent, and an oncology doctor Agent.

[0011] Preferably, the final diagnosis recommendation includes differential diagnosis and treatment plan for glioma.

[0012] Preferably, the network search Agent is associated with the World Clinical Trials Registration Database through an API.

[0013] Preferably, the retrieved information is registered clinical trial information related to glioma diagnosis and treatment.

[0014] The second aspect of the present application provides a glioma diagnosis and treatment information analysis system based on a large model, including: A disease information processing module, configured to process disease-related information input by a user to generate case content in a reporting form; A disease diagnosis module, configured to assign the reported medical record content to relevant doctor expert Agents for preliminary diagnosis, give preliminary diagnosis recommendations, and summarize the results of the preliminary diagnosis to obtain final diagnosis recommendations; A network search module, configured to extract keywords of the final diagnosis recommendation and call the API of the World Clinical Trials Registration Database to retrieve relevant clinical trial information, and obtain registered clinical trial information related to glioma diagnosis and treatment; Output module, which is used to output the final diagnosis suggestions obtained by the disease diagnosis module and the clinical trial information obtained by the network retrieval module.

[0015] Preferably, the disease information processing module includes: Preprocessing module, which is used to process the disease-related information to obtain text information; Resident doctor large model Agent, which is used to structurally process the preprocessed text information and generate medical record content in the form of a report.

[0016] Preferably, the disease diagnosis module includes: Doctor expert Agent, which is used to diagnose the medical record content in the form of a report to obtain preliminary diagnosis suggestions; MDT Leader Agent, which is used to analyze and summarize the preliminary diagnosis results to obtain the final diagnosis suggestions.

[0017] Preferably, the doctor expert Agent includes a neurology doctor Agent, a radiologist Agent, a pathologist Agent, a neurosurgeon Agent, a radiotherapy and chemotherapy doctor Agent, and an oncology doctor Agent.

[0018] The beneficial effects of this application are: The glioma diagnosis and treatment information analysis method and system based on a large model in this application comprehensively analyzes and judges the condition and diagnosis information of glioma patients. For the condition of the patients, with the assistance of the system, multiple experts conduct a joint consultation. After systematic analysis and induction, the analysis conclusions (diagnosis and recommended treatment plans) are sorted out and output; and keywords are extracted from the conclusions through this information analysis system and searched in the world clinical trial registration database, and the registered clinical trial information related to glioma diagnosis and treatment related to the condition is summarized. Finally, the analysis conclusions and the searched clinical trial information are output for medical staff to refer to for the treatment of the disease, which is convenient for experts to conduct a joint consultation and propose individualized diagnosis and treatment plans. Through this application, the standardization and individualization of diagnosis and treatment can be ensured, and the efficiency of glioma disease diagnosis can be improved. Brief Description of the Drawings

[0019] The drawings are used to provide a further understanding of this application, and constitute a part of the specification. Together with the embodiments of this application, they are used to explain this application and do not constitute a limitation to this application. In the drawings: Figure 1 is a schematic flowchart of the glioma diagnosis and treatment information analysis method based on a large model disclosed in the embodiment of this application. Detailed Embodiments

[0020] In the following text, only some exemplary embodiments are briefly described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present application. Therefore, the accompanying drawings and description are considered to be exemplary in nature rather than restrictive.

[0021] The following disclosure provides many different embodiments or examples for implementing the present application. Of course, they are merely examples and are not intended to limit the present application. The preferred embodiments of the present application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for illustrating and explaining the present application and are not used to limit the present application.

[0022] Figure 1 is a schematic flow chart of a glioma diagnosis and treatment information analysis method based on a large model disclosed in an embodiment of the present application.

[0023] As Figure 1 shown, the glioma diagnosis and treatment information analysis method based on a large model provided by the present application includes the following steps: S1: Obtain disease information, perform preprocessing and structuring to generate a structured medical record report.

[0024] In some embodiments, the disease information includes text information and picture information. The text information is clinical information related to suspected glioma, and the text information includes written descriptions or reports of the chief complaint, current medical history, past medical history, physical examination, tests, and examinations. Written descriptions, such as the narrative of the current medical history and past medical history, records in the medical record book, etc.; reports, such as physical examination reports, cerebrospinal fluid test results, etc. The picture information is imaging pictures or pathological pictures, such as head CT examinations and head MR magnetic resonance examinations, etc.

[0025] In some embodiments, when the user inputs the picture information, the picture is translated into a written description of the picture.

[0026] In some embodiments, corresponding processing is performed according to the type of disease information (text or picture) input by the user. For text information, preprocessing is performed, including deleting long spaces, special characters, and special symbols, and filtering network prohibited sensitive words, and then converting it into a pure text compatible with Chinese and English encoded in UTF-8.

[0027] In some embodiments, preprocessing the obtained disease information includes: removing noise data, filling in missing values, etc., to ensure the quality and usability of the data.

[0028] In some embodiments, assume that the original disease information is represented as , where represents the Eigenvalues, which may contain noise and missing values.

[0029] Denoising: Filtering algorithms such as low-pass filters can be used. Assuming the filtering function is , then the denoised data is:

[0030] Filling missing values: For missing values, methods such as mean filling can be adopted. Assuming the missing value is represented by , its filling formula is:

[0031] where is the mean of feature .

[0032] Structurally process the preprocessed data and convert it into a format suitable for subsequent processing. This structural processing refers to making large segments of text structured by extracting key points, aligning paragraphs, and sorting out logic, etc., so as to improve readability and legibility.

[0033] In some embodiments, the text information is converted into structured tables or JSON data to facilitate the extraction of key information.

[0034] In some embodiments, given a template structured table or JSON file, use a large model to convert the original disease information into structured data. Convert the processed data into a structured form. Assuming the structured conversion function is , then the structured data is:

[0035] In some embodiments, the above preprocessed text information is structurally processed by a large model Agent for resident doctors to generate a structured medical record report. In a specific example, the preprocessed text is summarized and extracted by the resident doctor Agent, and a structured medical record report is generated according to a preset template. The template covers general patient information (gender, age, body temperature, heart rate, respiratory rate, weight), chief complaint, current medical history, past medical history and family history, physical examination, tests, examination results, other information, and preliminary diagnosis.

[0036] The large model Agent is an autonomous intelligent agent based on a large language model (LLM). It can make complex decisions and execute tasks by understanding and generating natural language. This intelligent agent has a certain degree of autonomy and interaction ability, can process and understand a large amount of text information, and through pre-training and task adaptation, can achieve the solution and operation of specific domain problems.

[0037] The large model described in this application is based on a generative language model and includes open-source and closed-source large models.

[0038] In the field of artificial intelligence, a large model refers to a deep neural network with over 1 billion parameters. They can process massive amounts of data and complete various complex tasks, such as natural language processing, computer vision, speech recognition, etc. A generative large model is a generative model based on a large corpus, referring to those large-scale neural network models that can generate, understand, and reason about natural language in an end-to-end manner. By training on a large amount of text data, it can perform a wide range of tasks, including text summarization, translation, sentiment analysis, and so on. A large model refers to a deep neural network with over 1 billion parameters that can process massive amounts of data and complete various complex tasks. With the continuous improvement of computer hardware performance and the continuous optimization of deep learning algorithms, the development of large models has become increasingly rapid. The parameter scale of large models continues to expand, and the training time is also getting longer, but the performance has also improved accordingly. Large models usually rely on deep learning architectures, such as the Transformer, which enables impressive capabilities in various natural language processing tasks. The large models adopted can include but are not limited to ChatGPT, GPT-4, ERNIE, Lingyi Bot, etc. The generative language large model in this embodiment is a model in natural language processing, and its main task is to generate natural language text; this kind of model can learn and understand the internal structure and rules of language, and then generate text that conforms to grammar and semantic rules based on this knowledge and the given context or prompt, such as the open-source large model deepseek and the closed-source large model GPT-4. The embodiments of this application do not make specific limitations.

[0039] Through the large model Agent of resident doctors, the preprocessed text information is adjusted, sorted, and improved to generate medical record content in a reportable form. The form of this medical record content can be various. In a specific embodiment, the reportable form of medical record content includes: basic information, current medical history, past medical history, physical examination, auxiliary examinations, diagnosis, treatment and progress, conclusion, discussion, references, etc.

[0040] S2: The structured medical record report is initially diagnosed by the doctor expert Agent to generate initial diagnosis suggestions.

[0041] The above doctor expert Agent is constructed based on glioma diagnosis and treatment specifications or guidelines. The doctor expert Agent is based on a large language model, such as deepseek or gpt-4, etc. According to the input medical record content, it uses existing medical knowledge and experience to generate initial diagnosis suggestions. This step is equivalent to simulating the process of a resident doctor making a preliminary judgment on the patient's condition.

[0042] In some embodiments, the doctor expert Agents include a neurology doctor Agent, a radiology doctor Agent, a pathology doctor Agent, a neurosurgery doctor Agent, a radiotherapy and chemotherapy doctor Agent, and an oncology doctor Agent.

[0043] In some embodiments, the structured medical record report described in step S1 is assigned to the above-mentioned multiple doctor expert Agents for preliminary diagnosis, and then preliminary diagnosis suggestions are formed. In a specific example, the doctor expert Agent is based on a large language model, such as DEEPSEEK, etc., and its input is the structured medical record report , and the output is the preliminary diagnosis suggestion .

[0044] The forward propagation process of the neural network can be expressed as:

[0045] where represents the neural network model, is the set of model parameters.

[0046] S3: The preliminary diagnosis suggestions are summarized by the MDT Leader Agent to generate the final diagnosis suggestions.

[0047] In the field of glioma diagnosis and treatment, MDT is the abbreviation of "Multidisciplinary Team". MDT is a medical collaboration model aimed at providing comprehensive, standardized, and individualized diagnosis and treatment plans for patients by integrating the professional knowledge and skills of different disciplines.

[0048] The role of the MDT Leader Agent is to summarize multiple preliminary diagnosis suggestions, comprehensively consider the opinions of different doctor expert Agents, and generate the final diagnosis suggestions. This step is similar to the formation of a unified diagnosis opinion after discussions among experts in a multidisciplinary team (MDT) meeting.

[0049] Among them, the MDT Leader Agent not only has the function of summarizing the preliminary diagnosis suggestions given by each doctor expert Agent to obtain the final diagnosis suggestions, but also has the function of calling external tools, such as APIs and databases, etc.

[0050] In some embodiments, the doctor expert Agents include a neurology doctor Agent, a radiology doctor Agent, a pathology doctor Agent, a neurosurgery doctor Agent, a radiotherapy and chemotherapy doctor Agent, and an oncology doctor Agent.

[0051] In some embodiments, it is assumed that the MDT Leader Agent summarizes and collates the preliminary diagnostic suggestions, and the set of preliminary diagnostic suggestions is . In the summarization process, methods such as weighted average can be used. Assuming the weight is , the weight is obtained based on the summary of long-term real-world glioma MDT discussions. For example, the weight given to the imaging department Agent (doctor) in image interpretation is 0.99, and the weight given to the pathology Agent in pathological picture analysis is 0.99. Then the formula for the final diagnostic suggestion is:

[0052] where, , and .

[0053] In some embodiments, the final diagnostic suggestion includes differential diagnosis related to glioma and recommended treatment plans.

[0054] S4: The final diagnostic suggestion is extracted for keywords by the network search Agent, and the keywords are used to retrieve relevant clinical trial information in the "World Clinical Trials Registration Database" to obtain the retrieval information.

[0055] In a specific embodiment, it is assumed that the set of keywords extracted from the final diagnostic suggestion is , and relevant clinical trial information is retrieved in the World Clinical Trials Registration Database. The retrieval process can be expressed as:

[0056] where, represents all clinical trial information in the World Clinical Trials Registration Database, and is the set of relevant clinical trial information retrieved.

[0057] S5: Output the final diagnostic suggestion and the retrieval information.

[0058] The finally output conclusion and the searched clinical trial information can be used as a reference for medical staff for disease treatment, facilitate expert consultation, propose individualized diagnostic and treatment plans, ensure the standardization and individualization of diagnosis and treatment, and improve the efficiency of glioma disease diagnosis.

[0059] In some embodiments, the finally output result is the final diagnostic suggestion and the set of retrieved clinical trial information , that is:

[0060] The second aspect of this application provides a glioma diagnosis and treatment information analysis system based on large models, including: A disease information processing module, configured to process the disease-related information input by the user and generate a case content in a report form; A disease diagnosis module, configured to assign the case content of the report to relevant doctor expert agents for preliminary diagnosis, give preliminary diagnosis suggestions, and summarize the results of the preliminary diagnosis to obtain a final diagnosis suggestion; A network search module, configured to extract keywords from the final diagnosis suggestion and call the API of the World Clinical Trials Registration Database to retrieve relevant clinical trial information, and obtain registered clinical trial information related to glioma diagnosis and treatment; An output module, configured to output the final diagnosis suggestion obtained by the disease diagnosis module and the clinical trial information obtained by the network search module.

[0061] In some embodiments, the disease information processing module includes: A preprocessing module, configured to process the disease-related information to obtain text information; A resident large model agent, configured to structurally process the preprocessed text information to generate a case content in a report form.

[0062] In some embodiments, the disease information includes text information and picture information; the text information is clinical information related to suspected glioma, including the chief complaint, current medical history, past medical history, physical examination, laboratory tests, written descriptions or reports of examinations, etc.; the picture information is imaging pictures or pathological pictures. Specifically, when implemented, the user first fills in personal information (i.e., the user's basic information) when registering for this system, and then the user inputs the disease information into the system by manually entering text or uploading pictures, and the picture information is preprocessed into text information by the background.

[0063] In some embodiments, the resident large model agent structurally processes the text information obtained after the above preprocessing to generate a case content in a report form. The large model is established based on a generative language model, including open-source large models and closed-source large models. The text information obtained after preprocessing is processed by the open-source large model and the closed-source large model for content adjustment, sorting, and improvement to generate a case content in a reportable form. The case content in a reportable form includes: basic information, current medical history, past medical history, physical examination, auxiliary examinations, diagnosis, treatment and progress, conclusions, discussions, references, etc.

[0064] In some embodiments, the disease diagnosis module includes: Doctor expert agents are used to diagnose the medical record content in the form of reports and obtain preliminary diagnosis suggestions. The doctor expert agents include a neurology doctor agent, a radiology doctor agent, a pathology doctor agent, a neurosurgery doctor agent, a radiotherapy and chemotherapy doctor agent, and an oncology doctor agent; The MDT Leader Agent is used to analyze and summarize the preliminary diagnosis results to obtain a final diagnosis.

[0065] This application comprehensively analyzes and judges the disease information of glioma patients through an analysis system. For the patient's condition, with the assistance of the system, after systematic analysis and induction, the analysis conclusions (diagnosis suggestions and recommended treatment plans) are sorted out and output; and keywords are extracted from the diagnosis suggestions through this analysis system and searched in the "World Clinical Trials Registration Database", and the registered clinical trial information related to glioma diagnosis and treatment related to the condition is summarized, and finally the analysis conclusions and the searched clinical trial information are output. This output result can provide reference for medical staff, facilitate expert consultations, and put forward individualized diagnosis and treatment plans. Through this application, the standardization and individualization of diagnosis and treatment can be ensured, and the diagnosis efficiency of glioma diseases can be improved.

[0066] The above are the preferred embodiments of this application and are not used to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. A method for analyzing glioma diagnosis and treatment information based on a large model, characterized in that: The steps include: Acquire disease information, pre-process and structure the disease information, and generate a structured medical record report; The structured medical record report is preliminarily diagnosed by a doctor expert agent to generate a preliminary diagnosis suggestion; The preliminary diagnosis suggestions are summarized and aggregated by the MDT Leader Agent to generate final diagnosis suggestions; Extracting the keywords of the final diagnosis through a network search agent, and searching the relevant clinical trial information in the World Clinical Trial Registration Database with the keywords to obtain search information; The final diagnosis suggestion and the search information are output.

2. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 1, characterized in that: The disease information includes text information and picture information; the text information is clinical information related to suspected glioma; the text information includes text descriptions or reports of the chief complaint, current medical history, past history, physical examination, tests, and examinations; the picture information is imaging pictures or pathological pictures.

3. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 2, characterized in that: Preprocessing the disease information includes: Translating the image information into text description; The text descriptions of text information and image information are subjected to noise reduction and missing value filling.

4. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 1, characterized in that: The preprocessed text information is structured through the resident physician model agent to generate a structured medical record report.

5. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 1, characterized in that: The big model includes an open source big model and a closed source big model.

6. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 1, characterized in that: The doctor expert Agent is constructed based on glioma diagnosis and treatment standards or guidelines.

7. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 6, characterized in that: The doctor expert Agents include neurologist Agents, radiologist Agents, pathologist Agents, neurosurgeon Agents, radiochemotherapy doctor Agents and oncologist Agents.

8. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 1, characterized in that: The final diagnostic recommendations include differential diagnosis and treatment options related to glioma.

9. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 1, characterized in that: The network search agent is associated with the world clinical trial registration database through an API.

10. The method for analyzing glioma diagnosis and treatment information based on a large model according to claim 1, characterized in that: The retrieval information is registered clinical trial information related to glioma diagnosis and treatment.

11. A glioma diagnosis and treatment information analysis system based on a large model, characterized in that: include: The disease information processing module is used to process the disease-related information input by the user and generate case content in the form of a report; The disease diagnosis module is used to assign the reported medical records to relevant doctor expert agents for preliminary diagnosis, give preliminary diagnosis suggestions, and summarize the results of the preliminary diagnosis to obtain final diagnosis suggestions; A network search module, used to extract keywords of the final diagnosis recommendation, and call the API of the World Clinical Trial Registry Database to search for relevant clinical trial information, so as to obtain registered clinical trial information related to glioma diagnosis and treatment; The output module is used to output the final diagnosis recommendations obtained by the disease diagnosis module and the clinical trial information obtained by the network retrieval module.

12. The large model-based glioma diagnosis and treatment information analysis system according to claim 11, characterized in that: The disease information processing module comprises: A preprocessing module, used for processing the disease-related information to obtain text information; The resident physician model agent is used to structure the pre-processed text information and generate medical record content in the form of a report.

13. The large model-based glioma diagnosis and treatment information analysis system according to claim 12, characterized in that: The disease diagnosis module comprises: Doctor expert Agent, used to diagnose the medical record content in the form of report and obtain preliminary diagnosis Cut off suggestions; The MDT Leader Agent is used to analyze and summarize the preliminary diagnosis results to obtain final diagnosis recommendations.

14. The large model-based glioma diagnosis and treatment information analysis system according to claim 13, characterized in that: The doctor expert Agents include neurologist Agents, radiologist Agents, pathologist Agents, neurosurgeon Agents, radiochemotherapy doctor Agents and oncologist Agents.

Images