Large language model system for automatic clinical guide updating and customization

Abstract

A large language model system for automatic clinical guide updating and customization belongs to the technical field of medical information management, and comprises a data collection and processing module, a natural language (NLP) processing module, a model training and updating module, a guide updating and customization module, a knowledge base and intelligent recommendation module, and a user interface and interaction module. The method can quickly and accurately analyze, update and customize clinical guidelines and provide personalized treatment suggestions, and has the advantages that wide medical literatures and clinical data can be integrated and processed, automatic decision support is carried out in combination with expert knowledge, latest medical research results are quickly applied to clinical practice, and the method has a good application prospect. Therefore, doctors and medical teams can be helped to make treatment decisions more effectively, and medical quality and patient nursing are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical information management, and particularly relates to a large language model system for automated clinical guideline update and customization. Background Art

[0002] The existing guideline update and customization process includes: First, the team needs to collect the latest medical research results, clinical trial data, and relevant literature to understand the current evidence status. Exclude low-quality research and data, and conduct strict data analysis and evaluation to obtain reliable evidence support. Deploy a panel of experts to review and update the guidelines. Under the guidance of the expert panel, determine the update plan and scope of work for the guidelines. Determine the purpose, scope, and important topics of the update to ensure the currency and practicality of the guidelines. Conduct a systematic review and evaluation to understand the strengths, weaknesses, and potential problems of the current guidelines. The systematic review may include an assessment of the structure, content, and methods of the guidelines, as well as an assessment of the applicability, operability, and effectiveness of the guidelines. Update the content of the guidelines based on the new evidence and expert opinions. This may involve modifying diagnostic criteria, treatment regimens, prevention strategies, etc., and making revisions according to the latest evidence and expert opinions. Ensure the accuracy, coordination, and consistency of the guidelines. Submit the updated guideline content to the expert panel for evaluation and discussion. The expert panel will evaluate the practicality, feasibility, and operability of the guidelines and provide feedback and suggestions. Submit the updated guideline content to other relevant field experts and stakeholders for review and validation. Accept the review opinions and suggestions on the guidelines and make corresponding revisions and adjustments. If necessary, customize the guidelines according to specific clinical practice needs and patient characteristics. Personalize the general guidelines according to different clinical environments, resource limitations, and patient individual characteristics. Then, submit the updated guidelines to the guideline review committee or a similar institution in the medical institution. After review and inspection, the committee will decide whether to approve and publish the updated guidelines. After the guidelines are updated, effective implementation and dissemination are required to ensure that medical professionals and institutions can fully understand and apply the new guideline content. This may include training, guideline promotion activities, knowledge dissemination, and updating educational materials. Throughout the process, transparent communication, cooperation, and professional management are very crucial to ensure that the updated and customized clinical guidelines can provide the latest and optimal guidance for clinical practice. The time required for the update and customization of clinical guidelines can vary due to various factors, including the scale of the guidelines, update scope, availability of the expert panel, complexity of the review process, etc. Generally, it usually takes several months to several years from the start of the update to the final release of the updated guidelines. Summary of the Invention

[0003] The objective of the present invention is to provide a large language model system for automating the update and customization of clinical guidelines to solve existing problems.

[0004] A large language model system for automating the update and customization of clinical guidelines, comprising a data collection and processing module, a natural language processing module, a model training and updating module, a guideline update and customization module, a knowledge base and intelligent recommendation module, and a user interface and interaction module;

[0005] The said data collection and processing module is responsible for collecting the latest medical research data, clinical trial results and relevant literature, as well as patient data and treatment records related to clinical guidelines. These data will be cleaned, sorted and converted into a format acceptable to the model;

[0006] The said natural language processing module uses advanced natural language processing technologies to process and analyze text data;

[0007] The said model training and updating module uses machine learning and deep learning technologies to train a large language model. During the training process, relevant evidence and data, as well as the knowledge and experience of clinical experts, will be used to guide the learning of the model;

[0008] The said guideline update and customization module, after the training is completed, will automatically update and customize the clinical guidelines. Based on the analysis of new evidence and expert guidance, the model will automatically revise and update the guideline content;

[0009] The said knowledge base and intelligent recommendation module maintains a rich medical knowledge base, including the latest research results, clinical practice guidelines, expert opinions, etc. Based on the clinical situation and characteristics input by the user, the model can intelligently recommend personalized treatment plans and guidelines to assist doctors in making decisions;

[0010] The said user interface and interaction module provides a user-friendly interface, enabling doctors and medical professionals to conveniently access, search and customize clinical guidelines. Users can view the updated guideline content, receive recommended suggestions, and interact with the model to obtain more explanations and answers.

[0011] Preferably, the said natural language processing module uses techniques such as word embedding, entity recognition and relation extraction to structure medical literature and guidelines, and mine the key information therein.

[0012] Preferably, the said model training and updating module adopts a Transformer-based model to update and customize clinical guidelines.

[0013] Preferably, after the training is completed, the guideline update and customization module will automatically update and customize the clinical guidelines. Based on the analysis of new evidence and expert guidance, the model will automatically revise and update the guideline content, including diagnostic criteria, treatment plans, prevention strategies, etc.

[0014] Preferably, the knowledge base and intelligent recommendation module maintains a rich medical knowledge base, including the latest research results, clinical practice guidelines, expert opinions, etc.

[0015] The working process and principle of the present invention:

[0016] 1. Core algorithms and frameworks

[0017] Transformer model: This algorithm adopts the Transformer architecture as the basis of the large language model. Transformer is a powerful sequence modeling technology suitable for processing long text sequences and capturing context relationships. It consists of multiple self-attention layers and feed-forward neural network layers for encoding text sequences and generating corresponding outputs.

[0018] Pre-training and fine-tuning: The system first conducts large-scale pre-training, learning using a corpus based on a large amount of medical literature and clinical data. In this way, the model can learn the knowledge and language patterns in the medical field. After pre-training, the model can be fine-tuned using specialized domain data with clinical guideline update and customization requirements for optimization.

[0019] Natural language processing technology: Integrate various natural language processing technologies in the model, such as entity recognition, syntactic analysis, relation extraction, etc., to process and analyze clinical text data. These technologies can extract key information from clinical guidelines and perform structured and semantic understanding.

[0020] Evidence analysis and integration: The model will learn how to analyze the latest medical research results, clinical trial data, and relevant literature. It can extract effective evidence from a large number of literatures, evaluate its quality and credibility, and integrate it into the clinical guideline update process.

[0021] Personalized recommendation: The model can provide personalized treatment recommendations based on the characteristics and clinical conditions of patients. It can utilize information such as the patient's medical history and laboratory results, combined with the latest guideline content and expert opinions, to generate customized treatment plans suitable for each patient.

[0022] Model evaluation and optimization: Apply model evaluation metrics in the system, such as the consistency, accuracy, and interpretability of the generated suggestions with the actual clinical guidelines. Through monitoring and feedback mechanisms, continuously optimize the performance and results of the model.

[0023] 2. Data processing and optimization

[0024] Data collection and cleaning: First, it is necessary to determine the data sources to be used for updating and customization, including medical literature, clinical trial data, patient medical records, etc. These data may have different formats and structures. Therefore, data cleaning and preprocessing are required, including removing noise, standardizing, and unifying data formats, etc.

[0025] Data integration and unification: Integrate and unify data from different data sources to ensure data consistency and integrity. This may involve converting data into a unified standard format, performing entity alignment and association, and resolving data inconsistencies and conflicts.

[0026] Data annotation and commenting: For the needs of model training and optimization, key data may need to be annotated and commented to help the model understand and learn knowledge and semantics in the medical field. For example, perform entity recognition (such as diseases, drugs, diagnoses, etc.) and relationship extraction on text data.

[0027] Feature selection and extraction: According to the goals of clinical guideline update and customization, perform feature selection and extraction on the data to capture key information. This may include text features, statistical features, time series features, etc.

[0028] Model training and optimization: Use the cleaned, integrated, and annotated data for model training and optimization. This may include the processes of pre-training and fine-tuning, and use a large amount of medical literature and clinical data for training to learn medical semantics and patterns.

[0029] Data validation and evaluation: Evaluate and validate the model by using a validation dataset to check whether the performance and results of the model meet expectations. This helps to discover potential problems and areas for improvement, and perform further optimization and adjustment.

[0030] Iteration and improvement: Based on the evaluation results and user feedback, perform iteration and improvement on the data processing and the model. This may involve adjusting feature selection, model architecture, hyperparameters, etc. to optimize the performance and effectiveness of the system.

[0031] 3. Model fine-tuning and application specialization

[0032] Data preparation: Prepare a specific domain dataset for model fine-tuning and specialization, which includes medical literature, patient data, laboratory results, etc. related to clinical guidelines. Ensure the quality, integrity, and credibility of the data source of the dataset.

[0033] Data preprocessing: Perform preprocessing on the dataset, including data cleaning, standardization, splitting into training set, validation set, and test set, etc. This helps to ensure data consistency and usability.

[0034] Model Architecture Selection: Select a model architecture suitable for clinical guideline update and customization tasks. The structure of large language models is predefined, but certain adjustments or expansions may be required according to the needs of specific tasks.

[0035] Parameter Initialization: Initialize the model using the parameters of the pre-trained model. This helps to retain the pre-trained model's understanding of language and domain during the fine-tuning process.

[0036] Fine-tuning Process: Fine-tune the model by using a domain-specific dataset. This involves iterative training on specific tasks, such as sequence classification, text generation, etc. During the fine-tuning process, parameters such as the learning rate, batch size, and number of training steps can be adjusted.

[0037] Model Evaluation: Evaluate the fine-tuned model using the validation set and test set. Evaluation metrics can include accuracy, recall, F1-score, etc. This helps to check the performance and effectiveness of the model for clinical guideline update and customization tasks.

[0038] Model Specialization: Specialize the model for specific requirements and needs in clinical guideline update and customization tasks. This includes focusing on specific clinical domains, diseases, or treatment regimens during the fine-tuning process. By specializing the model, the accuracy and adaptability of the model in specific domains or tasks can be improved.

[0039] Domain Knowledge Injection: During the fine-tuning and specialization processes, the domain knowledge and experience of experts can be injected into the model. This can be achieved through expert guidance, introduction of rules, or based on prior knowledge. This helps the model to better understand the specific domain requirements and constraints in clinical guidelines.

[0040] Model Validation and Adjustment: After model specialization, validate and evaluate it to ensure its effectiveness and performance in clinical guideline update and customization tasks. According to the validation results, further adjustment and optimization of the model may be required to continuously improve the model's performance.

[0041] Go-live Deployment and Practical Application: According to the validation and evaluation results, deploy the specialized model to the actual clinical environment. This means integrating the model into a user-friendly system or application for medical professionals to use. In this way, doctors and medical teams can obtain personalized clinical guidelines and treatment recommendations for specific clinical situations through this system.

[0042] 4. Real-time Information Processing and Response

[0043] Data Flow and Reception: Ensure the ability to receive real-time data streams from different sources (such as medical devices, sensors, etc.). This may involve establishing appropriate data interfaces and communication protocols.

[0044] Data processing and preprocessing: Real-time data may need to be preprocessed, such as data cleaning, denoising, interpolation, etc., to ensure the quality and consistency of the data.

[0045] Trigger mechanism: Define an appropriate trigger mechanism to determine when real-time information processing and response are required. This can be a preset rule, threshold, or trigger for a specific event.

[0046] Real-time information processing: Design and implement data processing and analysis algorithms to extract useful information from real-time data. This may involve techniques such as data mining, machine learning, deep learning, etc.

[0047] Decision support and response system: Based on the results of real-time information analysis, build a decision support and response system. This system can generate corresponding responses according to predefined rules, logics, or models.

[0048] Real-time response: According to the output of the decision support and response system, make corresponding real-time responses. This can be sending alerts to doctors, providing suggestions, or taking automated control actions.

[0049] Scalability and performance optimization: Ensure that the real-time information processing and response system has high scalability and performance, and can process a large amount of real-time data and provide fast responses.

[0050] Continuous monitoring and improvement: Continuously monitor and evaluate the real-time information processing and response system. Collect data such as user feedback and performance metrics for improvement and optimization.

[0051] Security and privacy protection: In the process of implementing real-time information processing and response, it is crucial to ensure the security and privacy protection of data. Take necessary security measures, such as data encryption, access control, authentication, and auditing, etc., to protect the confidentiality and integrity of data. At the same time, comply with relevant laws, regulations, and privacy regulations to ensure the personal privacy of patients and users is protected. Interactive and context understanding capabilities

[0052] 5. System integration and interface design

[0053] Determine requirements and functions: Clearly define the requirements and functions of the system, including different components, systems, and third-party services that need to be integrated. This helps to define the scope and objectives of the integration.

[0054] Interface specification and standard formulation: Formulate the specifications and standards for interface design to ensure compatibility and interoperability between various components and systems. This may include data formats, protocols, API design, etc.

[0055] System Architecture Design: Design the overall system architecture based on requirements and functions, and determine the relationships and interaction methods between different components and systems. This helps to determine the location and functions of the interfaces.

[0056] Interface Design and Documentation: Design and define the functions, data exchange methods, input and output parameters, etc. of each interface according to interface specifications and standards. Write detailed interface documentation so that developers can understand and use the interfaces.

[0057] Development and Testing: Develop and integrate components and systems according to the interface design and documentation. Conduct appropriate testing and verification during the integration process to ensure the correctness and stability of the interfaces.

[0058] Error Handling and Exception Handling: Consider error and exception handling mechanisms in the interface design. Define appropriate error codes, error messages, and error handling processes so that error situations can be identified and handled during interface calls.

[0059] Monitoring and Logging: Design monitoring and logging mechanisms for the interfaces to track the usage, performance, and problems of the interfaces. This helps to detect and investigate potential problems in a timely manner and make appropriate optimizations and improvements.

[0060] Gradual Integration and Iterative Optimization: Adopt a method of gradual integration to gradually integrate each component and system into the final system. Continuously iterate and optimize during the integration process, and make adjustments and improvements according to the actual situation.

[0061] Documentation and Training: Write detailed system integration documentation to record the system architecture, interface design, and usage instructions. Provide training and support for developers, administrators, and users to ensure that they can use and manage the system correctly.

[0062] 6. Scalability and Maintainability

[0063] Modular Design: Divide the system into modules or components, and each module is responsible for a specific function. This modular design can make the system easier to understand, maintain, and expand. Ensure that the interfaces between modules are simple and clear, and follow the single responsibility principle.

[0064] Use Design Principles and Patterns: Apply design principles and design patterns, such as the SOLID principles, the dependency inversion principle, the factory pattern, etc., to maintain the maintainability and scalability of the code. These principles and patterns provide some guidance to help design the relationships and interaction methods between modules.

[0065] Use appropriate programming languages and frameworks: Selecting appropriate programming languages and frameworks can support modular and scalable development approaches. Some modern programming languages and frameworks provide various tools and mechanisms that enable developers to more easily build scalable and maintainable systems.

[0066] Reasonable code structure and naming conventions: Establish clear code structures and naming conventions to make the code easy to read, understand, and maintain. Write code in a consistent style, use meaningful names, provide appropriate comments and documentation so that other developers can quickly understand and modify the code.

[0067] Testing and automation: Write appropriate unit tests, integration tests, and system tests to ensure the correctness and stability of the code. Use automated testing tools and processes to reduce the workload of manual testing. This can help discover problems, fix bugs faster, and maintain the quality and reliability of the system.

[0068] Exception handling and logging: Reasonably handle exceptional situations and record error and exception information. Incorporate appropriate logging mechanisms into the system to facilitate tracking and investigating problems. This helps identify and solve potential issues and perform maintenance and improvements in a timely manner.

[0069] Documentation and knowledge management: Write clear documentation, including system architecture, design decisions, interface specifications, etc. Establish a good knowledge management mechanism to record and transfer experiences and lessons during the development process. This helps with knowledge sharing and technology transfer among team members and improves the maintainability of the system.

[0070] Continuous integration and deployment: Adopt continuous integration and continuous deployment practices to ensure the fast and reliable release of code updates and feature expansions. Use automated build, test, and deployment processes to reduce human errors and improve delivery efficiency.

[0071] Advantages of the present invention:

[0072] The present invention can quickly and accurately analyze, update, and customize clinical guidelines, as well as provide personalized treatment recommendations. Its advantage lies in being able to integrate and process a wide range of medical literature and clinical data, combine expert knowledge for automated decision support, and quickly apply the latest medical research results to clinical practice. This can help doctors and medical teams make more effective treatment decisions, improving the quality of medical care and patient care. Brief description of the drawings

[0073] Figure 1 It is a flowchart of an embodiment of the present invention. Detailed implementation manners

[0074] Please refer to Figure 1 shown, which is an embodiment of the present invention.

[0075] A large language model system for automated clinical guideline update and customization, including a data collection and processing module, a natural language processing module, a model training and update module, a guideline update and customization module, a knowledge base and intelligent recommendation module, and a user interface and interaction module;

[0076] The data collection and processing module is responsible for collecting the latest medical research data, clinical trial results and relevant literature, as well as patient data and treatment records related to clinical guidelines. These data will be cleaned, sorted and converted into a format acceptable to the model;

[0077] The natural language processing module uses advanced natural language processing techniques to process and analyze text data;

[0078] The model training and update module uses machine learning and deep learning techniques to train a large language model. During the training process, relevant evidence and data, as well as the knowledge and experience of clinical experts, will be used to guide the learning of the model;

[0079] The guideline update and customization module, after the training is completed, will automatically update and customize the clinical guidelines. Based on the analysis of new evidence and expert guidance, the model will automatically revise and update the guideline content;

[0080] The knowledge base and intelligent recommendation module maintains a rich medical knowledge base, including the latest research results, clinical practice guidelines, expert opinions, etc. Based on the clinical situation and characteristics input by the user, the model can intelligently recommend personalized treatment plans and guidelines to help doctors make decisions;

[0081] The user interface and interaction module provides a user-friendly interface, enabling doctors and medical professionals to conveniently access, search and personalize the configuration of clinical guidelines. Users can view the updated guideline content, receive recommended suggestions, and interact with the model to obtain more explanations and answers.

[0082] The natural language processing module uses techniques such as word embedding, entity recognition and relation extraction to structure medical literature and guidelines, and mine the key information therein.

[0083] The model training and update module adopts a Transformer-based model to update and customize clinical guidelines.

[0084] The guideline update and customization module, after the training is completed, will automatically update and customize the clinical guidelines. Based on the analysis of new evidence and expert guidance, the model will automatically revise and update the guideline content, including diagnostic criteria, treatment plans, prevention strategies, etc.

[0085] The described knowledge base and intelligent recommendation module maintain a rich medical knowledge base, including the latest research results, clinical practice guidelines, expert opinions, etc.

[0086] The flowchart shows the main components and algorithms of this large language model system and indicates their functions and roles in the process of clinical guideline update and customization. This framework can automatically extract knowledge and information from a large amount of medical literature and clinical data, providing accurate and personalized clinical guidelines and treatment recommendations for medical professionals.

[0087] This embodiment can automatically track the latest published clinical research results and guideline updates of authoritative institutions globally and integrate them into the clinical practice of local hospitals.

[0088] Taking the update of the international guideline for the treatment of new drugs for cardiovascular diseases as an example, the following is a specific example of the operation of this system:

[0089] Data collection and processing module:

[0090] The system automatically crawls new published articles, research reports and guideline updates from medical databases such as PubMed and Cochrane Library and the official websites of internationally renowned medical institutions every day.

[0091] The data is cleaned, de-duplicated and structured, and key information such as evidence level, recommendation level, indications, contraindications, treatment methods, etc. is extracted.

[0092] Natural language processing module:

[0093] Using NLP technology to parse unstructured medical texts, such as full-text PDF documents, to identify and extract key information, ensuring that the model can understand complex medical terms and contexts.

[0094] Model training and update module:

[0095] For each clinical field, this large language model continuously receives and learns new scientific research data and guideline content, and conducts fine-tuning and iterative training regularly to quickly absorb and digest the latest knowledge.

[0096] Guideline update and customization module:

[0097] When a relevant guideline update for the treatment of new drugs for cardiovascular diseases is detected, the system analyzes and compares the differences between the new and old versions, and automatically generates a revised local or overall guideline.

[0098] At the same time, combining factors such as the regional epidemiological characteristics, hospital resources and doctor preferences in this area, the system can customize clinical practice guidelines that meet the local actual situation.

[0099] Knowledge base and intelligent recommendation module:

[0100] The updated knowledge is integrated into the system's knowledge base to form a comprehensive and real-time clinical decision support system.

[0101] When doctors input patient condition information in the electronic medical record system, CliniGuides AI intelligently recommends the most appropriate diagnosis and treatment plan based on the latest guideline recommendations.

[0102] User interface and interaction module:

[0103] After doctors log in to the system, during the diagnosis and treatment process, the system will pop up reminders and suggestions based on the latest guidelines at appropriate time points to help doctors make decisions.

[0104] Doctors can also query the latest guideline summaries or detailed information on specific diseases through the search function and directly view, download, or print them on the system interface.

[0105] In summary, after receiving the update of the international guidelines for the new drug treatment of cardiovascular diseases, this large language model system quickly completes data integration, knowledge understanding, localization adaptation, and finally presents it to doctors through the user interface to assist them in timely adjusting the diagnosis and treatment strategies and improving the quality and efficiency of medical services.

Claims

1. A large language model system for automating clinical guideline updates and customization, characterized in that: It includes a data collection and processing module, a natural language processing module, a model training and updating module, a guideline updating and customization module, a knowledge base and intelligent recommendation module, and a user interface and interaction module; The data collection and processing module is responsible for collecting the latest medical research data, clinical trial results and relevant literature, as well as patient data and treatment records related to clinical guidelines. These data will be cleaned, sorted and converted into a format acceptable to the model; The natural language processing module uses advanced natural language processing technologies to process and analyze text data; The model training and updating module uses machine learning and deep learning technologies to train a large language model. During the training process, relevant evidence and data, as well as the knowledge and experience of clinical experts, will be used to guide the learning of the model; The guideline updating and customization module, after the training is completed, will automatically update and customize the clinical guidelines. Based on the analysis of new evidence and expert guidance, the model will automatically revise and update the guideline content; The knowledge base and intelligent recommendation module maintains a rich medical knowledge base, including the latest research results, clinical practice guidelines, expert opinions, etc. Based on the clinical situation and characteristics input by the user, the model can intelligently recommend personalized treatment plans and guidelines to help doctors make decisions; The user interface and interaction module provides a user-friendly interface, enabling doctors and medical professionals to conveniently access, search and personalize the configuration of clinical guidelines. Users can view the updated guideline content, receive recommended suggestions, and interact with the model to obtain more explanations and answers.

2. The large language model system for automated clinical guideline update and customization according to claim 1, characterized in that: The natural language processing module uses techniques such as word embedding, entity recognition and relationship extraction to structure medical literature and guidelines, and mine the key information in them.

3. The large language model system for automated clinical guideline update and customization according to claim 1, characterized in that: The model training and updating module adopts a Transformer-based model to update and customize clinical guidelines.

4. A large language model system for automated clinical guideline update and customization according to claim 1, characterized in that: The guideline updating and customization module, after the training is completed, will automatically update and customize the clinical guidelines. Based on the analysis of new evidence and expert guidance, the model will automatically revise and update the guideline content, including diagnostic criteria, treatment plans, prevention strategies, etc.

5. A large language model system for automated clinical guideline update and customization according to claim 1, characterized in that: The knowledge base and intelligent recommendation module maintains a rich medical knowledge base, including the latest research results, clinical practice guidelines, expert opinions, etc.

Images