Teaching case library construction system and method based on AI large model cross-department adaptation

Abstract

The invention discloses a teaching case library construction system and method based on AI large model cross-department adaptation, and the system comprises a public teaching case library which is used for storing case metadata and providing a unified public teaching case library; the input module is used for carrying out analysis and structured processing on the uploaded patient medical record; the output module is used for generating a general teaching version according to disease type input received in the human-computer interface; the optimization module is used for performing dynamic quality optimization, regularly triggering an AI large model to perform quality evaluation through using frequency and like data, automatically correcting description ambiguity, supplementing a latest guide basis or generating a high-score alternative version; and the security module is used for providing a data isolation strategy and avoiding sensitive information in cases. According to the invention, scattered non-standardized cases are converted into a dynamically evolved teaching knowledge network, three problems of islanding of teaching resources, low efficiency of adaptation and uncontrollable quality are solved, and intelligent infrastructure support is provided for clinical education of interns.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information, in particular to a teaching case library construction system and method based on AI large model cross-department adaptation. BACKGROUND

[0002] The existing medical intern teaching system currently has the following problems based on the actual problems encountered in teaching case scenarios: 1. Resource islandization, teaching cases may be scattered in various departments, and teaching teachers lack tools for unified management and sharing, making it difficult to call high-quality teaching case resources for cross-department collaborative teaching; 2. Before the widespread application of AI technology, data resources are difficult to use, original case data (such as medical record texts, examination reports) contain a large amount of free description and unstructured data, which are difficult to directly use for systematic teaching and efficient retrieval, and resources need to be invested for collation for a certain teaching event; 3. The teaching value, standardization, and completeness of teaching cases depend on the personal experience and investment of teaching teachers, and there is a lack of unified quality evaluation and optimization mechanism.

[0003] For example, the cardiology department needs to prepare a hypertension teaching case, and the teaching assistant or teaching teacher needs to spend several hours to collate the patient's complex medical record text (complaints, present illness history, past history, medication history, etc.), extract teaching highlights (such as refractory hypertension differentiation), and make a teaching plan. This case is only used in this department, and it is difficult to ensure the structure of the case. When the whole department wants to find similar cases for teaching, it needs to be collected and collated again, and it cannot be directly reused or generated to adapt to the whole department scenario, which is low in efficiency and depends on personal experience.

[0004] Based on the above situation, the present application intends to solve the problems encountered in clinical teaching by building a teaching case library to achieve the goals of teaching resource integration, improving the efficiency of department teaching staff, and increasing teaching content output. SUMMARY

[0005] The purpose of the present application is to provide a teaching case library construction system and method based on AI large model cross-department adaptation to solve the problems raised in the background art.

[0006] To achieve the above application purpose, one aspect of the present application provides a teaching case library construction system based on AI large model cross-department adaptation, which comprises a public teaching case library, an input module, an output module, an optimization module, and a security module, wherein:

[0007] The public teaching case library is used to store case metadata and provide a unified public teaching case library;

[0008] The input module is used to analyze and structure the uploaded patient medical record;

[0009] The output module generates a whole department teaching version according to the disease received in the human-computer interface;

[0010] The optimization module is used for dynamic quality optimization. By using frequency and like data, the AI large model is triggered regularly to evaluate the quality of metadata in the public teaching case library, and automatically corrects the description ambiguity, supplements the latest guideline basis, or generates a high-score alternative version.

[0011] The security module is used to provide data isolation strategy to avoid sensitive information in the case.

[0012] Further, the public teaching case library stores case metadata through MySQL.

[0013] Further, the input module calls the Spring AI module to connect to the QWAI large model API, and performs structured processing on the input case to generate structured JSON data containing teaching dimensions, and stores it persistently in MySQL. The structured processing includes:

[0014] Extracting key fields, including chief complaint, history of present illness, past medical history, medication history, and physical examination;

[0015] Labeling the logical relationship between illness and drug adjustment;

[0016] Splitting the differential diagnosis process to generate a differential diagnosis tree;

[0017] Supplementing teaching labels.

[0018] Further, the unstructured data uploaded in the input module is received and temporarily stored by the backend SpringCloud microservice.

[0019] Further, the output module performs multidimensional retrieval on the public teaching case library through Elasticsearch, and performs semantic reconstruction of the case through the QWAI large model, real-time rendering of structured data, generation of customized cases suitable for general teaching scenarios, and guidance of clinical thinking training, case discussion, and department examination in the clinical teaching process.

[0020] Further, in the optimization module, Redis is used to record the number of case calls, user triggered like quantity, and regularly scan the quality score TOP10% cases. The scoring mechanism calls the large model for authoritative review and automatically corrects outdated descriptions.

[0021] Further, the optimization module provides a low-score case mechanism, which sets the priority of low-score cases lower than that of high-call frequency and high-like quantity cases.

[0022] Further, in the security module, a caching strategy is provided, and Redis is used to cache structured data of popular cases to improve response speed when called.

[0023] Further, a disaster recovery mechanism is provided in the security module to perform daily incremental backup of MySQL and Elasticsearch to the hospital cloud database, ensuring zero loss of teaching resources.

[0024] Another aspect of the application provides a teaching case library construction method based on AI large model cross-department adaptation, comprising the following steps:

[0025] Step S1, analyzing and structuring the uploaded patient medical records;

[0026] Step S2, generating a general teaching version according to the disease received in the human-computer interface;

[0027] Step S3, regularly triggering AI large model for quality evaluation by using frequency and like data, automatically correcting description ambiguity, supplementing the latest guideline basis or generating high-score alternative version;

[0028] Step S4, providing data isolation strategy to avoid sensitive information in the case.

[0029] Compared with the prior art, the system and method have the following advantages:

[0030] 1. Standardization of teaching resources and knowledge accumulation: using SpringAI to call the Qianwen large model to realize the structured extraction of the key elements of teaching cases, eliminating manual sorting errors, and accumulating hospital teaching content resources through continuous uploading of teaching cases by departments.

[0031] 2. Improve the supply side ability of teaching cases: use Elasticsearch to support multi-field combination free description retrieval (disease + complication + teaching scene), return matching results in milliseconds, and large model generates on demand to solve department difference problems, such as "hypertension + primary diagnosis" in general practice, the system automatically weakens specialized examination and strengthens hierarchical referral logic, case adaptation is reduced, and accurate standard teaching cases are returned.

[0032] 3. Self-optimization and iteration of teaching cases: SpringAI calls the Qianwen large model to build a case quality quantization model through usage frequency and like data, and high-call cases automatically trigger large model review to supplement missing diagnosis basis.

[0033] 4. System performance and scalability guarantee: SpringCloud microservice architecture supports thousands of concurrent requests per second, Redis caches popular case data, Elasticsearch indexes cases, and supports fuzzy query and semantic retrieval.

[0034] 5. Finally realize the upgrading of clinical teaching efficiency, improve the efficiency of teaching teachers, and verify the effectiveness of teaching cases for interns. The reuse rate of cross-department cases is improved, and the homogenization of clinical thinking is promoted. The present application takes "large model structure + intelligent generation + closed-loop optimization" as the core, and converts the scattered non-standard cases into a dynamic evolution of teaching knowledge network, solves the three problems of teaching resource islandization, low efficiency and uncontrollable quality, and provides intelligent infrastructure support for clinical education for interns. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The system structure diagram of the teaching case library construction system based on the AI large model cross-department adaptation is shown. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0037] As Figure 1 The system structure diagram of the present application is shown. The present application adopts a hierarchical architecture to realize intelligent management of the teaching case library, and achieves the goals described in the technical effects. The present application includes an input module, an output module, an optimization module, and a security module, and a public teaching case library, wherein:

[0038] The input module is used for analyzing and structuring the uploaded patient medical records; provides a department collaborative uploading case entry, integrates an intern management system, and uploads the original case (such as a hypertension case of a cardiology department) through the teaching management system uploading case function (Vue front end) of the teaching teacher of the rotating department, that is, even if the unstructured data such as text and pictures are uploaded, the backend SpringCloud microservice receives and temporarily stores them.

[0039] The output module generates a general teaching version according to the disease received in the human-computer interface. The structured reconstruction of the case function of the large model is continuously optimized. The original case is analyzed in depth by calling the Qianwen large model through SpringAI. The clinical elements such as chief complaint, medical history, physical examination, auxiliary examination results, diagnosis conclusion, etc. are automatically identified. The teaching value is strengthened. The teaching labels such as differential diagnosis logic, treatment scheme basis, typical misdiagnosis pitfalls, etc. are supplemented. The structured JSON data containing the teaching dimension is generated and stored persistently in MySQL. The cross-department intelligent teaching case generation entrance is built. The user (such as the general medical department) inputs the demand (disease "hypertension" + requirement "primary difficulty"). The system searches for matching cases in the public teaching case library through Elasticsearch and calls the large model again. The system automatically feeds back the teaching cases generated according to the requirements of the general hospital department.

[0040] The optimization module is used for dynamic quality optimization. According to the number of times the reconstructed teaching cases are called and the user's like data, the quality evaluation of the large model is triggered regularly. The description ambiguity is automatically corrected. The latest guideline basis is supplemented or the high-score alternative version is generated. The "use-feedback-optimization" closed loop is formed.

[0041] The security module is used to provide data isolation strategy. The Elasticsearch index is filtered according to the department authority. The sensitive information in the case is avoided. The generation of the case is mainly for teaching purpose. Therefore, the sensitive information will be replaced in a virtual way. The cache strategy is provided. The structured data of the popular case is cached in Redis. The response speed is provided when calling. The disaster recovery mechanism is established. MySQL and Elasticsearch are backed up to the hospital cloud database every day. The zero loss of teaching resources is ensured.

[0042] The public teaching case library constructs the hospital-level teaching case sharing platform. The case metadata is stored in MySQL. The multi-dimensional retrieval is supported by Elasticsearch. The high-frequency access data is cached in Redis. The unified public teaching case library is formed.

[0043] Taking the uploading of the hypertension case by the cardiology department and the calling of the adapted version by the general medical department as an example, the implementation mode is expanded and described:

[0044] Step S1, the cardiology teaching teacher uploads the original medical record of the patient with primary hypertension to the front end of the practice management system. The case contains picture examination reports and free input text medical history records. The Spring Cloud file service receives the file, parses the text content, calls the Spring AI module to connect to the large model API, and sends prompt words for structured cases. Structured processing includes: (1) extracting key fields such as chief complaint, present illness history, past history, medication history, and physical examination; (2) labeling the logical relationship between blood pressure fluctuations and drug adjustment; (3) splitting the differential diagnosis process to generate a differential diagnosis tree; (4) adding teaching labels such as "refractory hypertension mechanism" and "community management blind area". The large model outputs JSON structured data based on the prompt words, including standardized fields, differential diagnosis trees (nested JSON arrays), and teaching label lists. After verification, the data is stored in the MySQL table, and the case text is indexed to Elasticsearch (index fields: disease, complications, teaching labels). This step accumulates teaching cases for the public teaching case library.

[0045] Step S2, the general medicine teaching teacher inputs the requirements on the practice management system interface, inputs the disease as hypertension in the dialog box, and requires "first diagnosis scene" and "simplified specialist terminology" as two prompt words. The retrieval service sends a Boolean query to Elasticsearch, and the public teaching case library intelligently matches the original hypertension cases uploaded by the cardiology department. The large model reconstructs the case based on the general medical requirements, and the front end renders the structured data based on the requirements in real time to generate a general medical teaching version that weakens specialist examinations such as renin detection. The front end renders the reconstructed structured data in real time, and finally outputs a customized case from the cardiology case that is suitable for the general medical teaching scene. The guide can be directly used in the clinical thinking training, case discussion, and out-of-department examination in the clinical teaching process.

[0046] Step S3, dynamic quality optimization. Redis records the number of calls for this case in real time, and the number of likes triggered by the user. According to the algorithm, the quality score TOP10% cases are scanned by a weekly scheduled task, and the scoring mechanism calls the large model for authoritative review (for example, comparing the updated drug recommendations in the 2024 version of the "China Hypertension Prevention and Treatment Guidelines") to automatically correct outdated descriptions. In addition, a low-score case mechanism is designed, which sets the priority of low-score cases lower than high-call frequency and high-liked cases when the end user needs to generate a case.

[0047] Step S4, provide security and performance guarantee, carry out data isolation strategy. Elasticsearch index is filtered according to department authority, sensitive information in the case is avoided, and the generation of the case is mainly for teaching purpose, so sensitive information will be replaced in a virtual way. Cache strategy is provided, structured data of popular cases is cached in Redis, response speed is provided when calling. Disaster recovery mechanism is established, MySQL and Elasticsearch are backed up to the hospital cloud database daily, and zero loss of teaching resources is ensured.

[0048] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1.A system for constructing a teaching case library based on AI large model cross-department adaptation, characterized in that, The application relates to a medical teaching case system, which comprises a public teaching case library, an input module, an output module, an optimization module and a security module, wherein: The public teaching case library is used for storing case metadata and providing a unified public teaching case library; The input module is used for analyzing and structuring uploaded patient medical records; The output module generates a general medical teaching version according to the disease received in the human-computer interface; The optimization module is used for dynamic quality optimization, and the AI large model is triggered regularly to evaluate the quality of the metadata in the public teaching case library by using frequency and like data, to automatically correct description ambiguity, supplement the latest guideline basis or generate a high-score alternative version; The security module is used for providing a data isolation strategy to avoid sensitive information in the case. 2.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, The public teaching case library stores case metadata through MySQL. 3.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, The input module calls the Spring AI module to connect to the QWAI large model API, performs structured processing on the input case, generates structured JSON data containing teaching dimensions, and stores the data in MySQL, wherein the structured processing comprises: Extracting key fields, including chief complaint, present illness history, past medical history, medication history and physical examination; Labeling the logical relationship between illness and drug adjustment; Splitting the differential diagnosis process to generate a differential diagnosis tree; Supplementing teaching labels. 4.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, Unstructured data uploaded in the input module is received and temporarily stored by the backend SpringCloud microservice. 5.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, The output module performs multidimensional retrieval on the public teaching case library through Elasticsearch, performs semantic reconstruction of the case through the QWAI large model, and generates customized cases suitable for general medical teaching scenarios to guide the clinical thinking training, case discussion and department examination of doctors in the clinical teaching process. 6.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, In the optimization module, Redis is used to record the number of case calls and the number of user triggered likes, and the quality score TOP10% cases are scanned regularly, the scoring mechanism calls the large model for authoritative review, and the outdated description is automatically corrected. 7.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, In the optimization module, a low-score case mechanism is provided, and the priority of the low-score case is set to be lower than that of the high-call frequency and high-like number case. 8.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, In the security module, a cache strategy is provided, and Redis is used to cache the structured data of popular cases to improve response speed during calling. 9.The AI-based large model cross-department adapted teaching case library construction system of claim 1, wherein, In the security module, a disaster recovery mechanism is provided, and MySQL and Elasticsearch are backed up to the hospital cloud database every day to ensure zero loss of teaching resources. 10.A method for constructing a teaching case library based on AI large model cross-department adaptation, characterized in that, The application further relates to a medical teaching case system, which comprises the following steps: Step S1, analyzing and structuring uploaded patient medical records; Step S2, generating a general medical teaching version according to the disease received in the human-computer interface; Step S3, triggering the AI large model to evaluate the quality of the metadata in the public teaching case library by using frequency and like data regularly, to automatically correct description ambiguity, supplement the latest guideline basis or generate a high-score alternative version; Step S4, providing a data isolation strategy to avoid sensitive information in the case.

Images