Teaching system for medical image report
By building a knowledge graph system and integrating the knowledge points of medical imaging reports, the problems of widespread distribution of lesions and scattered knowledge points were solved, a systematic teaching effect was achieved, and students' understanding and skill mastery were improved.
Patent Information
- Application Number
- CN202510748738.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-06
- Publication Date
- 2025-09-16
AI Technical Summary
In the teaching of medical imaging reports, lesions are widely distributed and highly flexible, which makes learning difficult and knowledge points scattered, making it difficult to form a systematic knowledge system.
Build a knowledge graph system, including a structured processing module, a knowledge graph construction module, a knowledge base construction module, a user question retrieval module and an answer content generation module. Use the RAG model to generate accurate answer content and integrate knowledge points through the knowledge graph association framework.
Systematically organizing knowledge points enhances students' understanding of lesion characteristics and disease diagnosis, forms a systematic knowledge system, provides detailed and comprehensive answers, and helps students master relevant skills.
Smart Images
Figure CN120654797A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to medical imaging report teaching, and in particular to a teaching system for medical imaging reports. Background Art
[0002] Currently, teaching medical imaging reports faces numerous complex and challenging situations. Lesions are widely distributed across various parts and organs of the human body, encompassing virtually every area from the head to the limbs, from the body surface to the internal organs. This widespread distribution of lesions complicates teaching, requiring students to master a vast array of lesion characteristics in diverse locations, increasing the learning difficulty and memorization burden.
[0003] The flexibility of lesions also presents a significant challenge. The same disease can manifest itself in diverse ways in different individuals, or even within the same patient at different stages of disease progression. For example, certain tumors can change in morphology, size, and relationship to surrounding tissues between early and late stages. This requires students to not only be familiar with typical lesion manifestations but also to develop the ability to accurately assess these varying conditions, further increasing the complexity of instruction.
[0004] Furthermore, the knowledge involved in teaching medical imaging reports is highly fragmented, ranging from the basic theories of imaging, the principles and methods of various examination techniques, to the imaging characteristics of different diseases, diagnostic evidence, and key points for differential diagnosis, to how to write accurate and standardized reports. These fragmented knowledge points often have complex connections, which can easily confuse students during the learning process and make it difficult for them to form a systematic knowledge system. Summary of the Invention
[0005] (1) Technical problems solved
[0006] In response to the above-mentioned shortcomings of the existing technology, the present invention provides a teaching system for medical imaging reports, which can effectively overcome the defects of the existing technology in that the teaching effect of medical imaging reports is poor and it is difficult to help students form a systematic knowledge system.
[0007] (2) Technical solution
[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:
[0009] A teaching system for medical imaging reports, comprising a structured processing module, a knowledge graph construction module, a knowledge base construction module, a knowledge graph association module, a user question retrieval module, and an answer content generation module;
[0010] The structured processing module describes the medical imaging report in a hierarchical manner and associates the feature entities. The feature entity attributes store relevant interpretation information and related feature vectors.
[0011] The knowledge graph construction module builds the knowledge graph based on the relationships between feature entities and the attributes of feature entities;
[0012] The knowledge base construction module builds the disease library, image feature library and report feature library based on the knowledge graph;
[0013] The knowledge graph association module dynamically obtains medical imaging reports. Experts then associate the medical imaging reports with the knowledge graph to build a knowledge graph association framework. The medical imaging reports are then automatically associated with the knowledge graph based on the disease database, image feature database, and report feature database to complete and fill in the knowledge graph association framework and obtain a knowledge graph association framework.
[0014] The user question retrieval module uses the retriever of the RAG model to convert user input into a search query and retrieves related knowledge graph fragments based on the knowledge graph association framework;
[0015] The answer content generation module splices the user input with the associated knowledge graph fragment and inputs it into the generator of the RAG model to obtain the answer content that matches the user input.
[0016] Preferably, the structured processing module performs hierarchical feature description on the medical imaging report, including:
[0017] 1) First-level features:
[0018] a. Mainly describes the categories of phenomena observed in medical imaging examinations, including lung markings and hilum;
[0019] b. Mainly describe the major categories of quality control observed in medical imaging examinations, including body position and foreign bodies;
[0020] 2) Secondary features:
[0021] a. Mainly describe the attributes of the major categories, including shape characteristics, size characteristics and change characteristics;
[0022] b. Details of questions that mainly describe the first-level characteristics of the major categories, including those that are too broad or too narrow;
[0023] 3) Level 3 features:
[0024] Mainly describes the details of secondary features, including location details and size details.
[0025] Preferably, the knowledge graph construction module constructs a knowledge graph based on the relationships between feature entities and the attributes of feature entities, including:
[0026] Build a knowledge graph based on the relationships and attributes between feature entities, store the knowledge graph in the Neo4j database, and connect medical images with corresponding feature entities;
[0027] Create an external server that uses the Faiss vector retrieval library to provide retrieval services, and update the Faiss vector retrieval library synchronously when the knowledge graph is updated;
[0028] Among them, the relationships between feature entities include strong correlation, weak correlation, and relationship attributes.
[0029] Preferably, the knowledge base construction module constructs a disease library, an image feature library and a report feature library based on the knowledge graph, including:
[0030] Experts build a disease database based on ICD-11 disease categories and related detailed descriptions;
[0031] Experts classify the detailed features of medical images corresponding to specific diseases and build an image feature library.
[0032] Preferably, the knowledge graph association module dynamically obtains medical imaging reports, and experts associate the medical imaging reports with the knowledge graph to build a knowledge graph association basic framework, including:
[0033] Experts split the medical imaging reports according to the characteristic entities in the knowledge graph, clarify the relationship between the characteristic entities and the disease, as well as the information and content that the characteristic entities need to focus on, so as to achieve the association between the medical imaging reports and the knowledge graph and build the basic framework for knowledge graph association.
[0034] Preferably, the knowledge graph association module dynamically obtains medical imaging reports, and automatically associates the medical imaging reports with the knowledge graph according to the disease library, image feature library, and report feature library, so as to improve and fill in the knowledge graph association basic framework, thereby obtaining the knowledge graph association framework, including:
[0035] Experts establish the relationship between medical imaging reports and diseases based on the disease library, image feature library and report feature library, use the pre-trained first feature extraction model to generate text content associated with the feature entities in the knowledge graph, and then use the pre-trained second feature extraction model to extract features from medical images, and add relevant labels to the text content and extracted features to improve and fill in the knowledge graph association basic framework and obtain the knowledge graph association framework.
[0036] Preferably, the user question retrieval module converts user input into a retrieval query using a retriever of the RAG model, and retrieves associated knowledge graph fragments based on the knowledge graph association framework, including:
[0037] Convert user input into search queries and retrieve related knowledge graph fragments through sparse retrieval and dense retrieval based on the knowledge graph association framework;
[0038] Sort the retrieved related knowledge graph fragments by relevance, and select a preset number of top-ranked related knowledge graph fragments as input to the generator of the RAG model;
[0039] Among them, user input includes user questions and medical imaging reports to be analyzed, and retrieval queries include keywords, semantic parsing or vectorization;
[0040] Sparse retrieval includes keyword matching, and dense retrieval includes vector similarity matching.
[0041] Preferably, the answer content generation module splices the user input with the associated knowledge graph fragment and inputs it into the generator of the RAG model to obtain the answer content that matches the user input, including:
[0042] Splice user input with related knowledge graph fragments to form a complete context;
[0043] The context is input into the generator of the pre-trained RAG model, and the generator of the pre-trained RAG model uses the integrated context, comprehensive retrieval content and its own knowledge to generate coherent, accurate answer content that matches the user input and contains relevant knowledge points.
[0044] (3) Beneficial effects
[0045] Compared with the prior art, the teaching system for medical imaging reports provided by the present invention has the following beneficial effects:
[0046] 1) By constructing a knowledge graph, we can systematically summarize and organize numerous and scattered knowledge points and various diseases, clearly sort out the inherent logical relationships between different knowledge points, and closely link diseases with related imaging manifestations, pathophysiological mechanisms, and other information to form an organic whole. In the knowledge graph, a disease can be used as a node, and the related typical lesion manifestations, predilection sites, examination methods, etc. can be used as different branches connecting the node. Students can use intuitive graphical displays to more comprehensively understand the overall picture of the disease and related knowledge;
[0047] 2) Combining RAG retrieval technology can further improve the completeness and comprehensiveness of the answer content. RAG retrieval technology can accurately and quickly retrieve relevant information from a large amount of knowledge resources when students ask questions, and integrate it to generate high-quality answer content. In the teaching of medical imaging reports, whether students have questions about a specific lesion or are confused about the diagnostic ideas of a certain disease, they can get detailed and comprehensive answers with the help of RAG retrieval technology. It can not only provide accurate descriptions of knowledge points, but also combine actual cases and the latest research results to make students' understanding of the problem more in-depth and comprehensive, thereby helping students better master the relevant knowledge and skills of medical imaging reports and lay a solid foundation for future clinical practice. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.
[0049] Figure 1 Schematic diagram of the system of the present invention. DETAILED DESCRIPTION
[0050] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.
[0051] A teaching system for medical imaging reports, e.g. Figure 1 As shown, it includes a structured processing module, a knowledge graph construction module, a knowledge base construction module, a knowledge graph association module, a user question retrieval module and an answer content generation module;
[0052] The structured processing module describes the medical imaging report in a hierarchical manner and associates the feature entities. The feature entity attributes store relevant interpretation information and related feature vectors.
[0053] The knowledge graph construction module builds the knowledge graph based on the relationships between feature entities and the attributes of feature entities;
[0054] The knowledge base construction module builds the disease library, image feature library and report feature library based on the knowledge graph;
[0055] The knowledge graph association module dynamically obtains medical imaging reports. Experts then associate the medical imaging reports with the knowledge graph to build a knowledge graph association framework. The medical imaging reports are then automatically associated with the knowledge graph based on the disease database, image feature database, and report feature database to complete and fill in the knowledge graph association framework and obtain a knowledge graph association framework.
[0056] The user question retrieval module uses the retriever of the RAG model to convert user input into a search query and retrieves related knowledge graph fragments based on the knowledge graph association framework;
[0057] The answer content generation module splices the user input with the associated knowledge graph fragment and inputs it into the generator of the RAG model to obtain the answer content that matches the user input.
[0058] ① The structured processing module provides hierarchical feature descriptions of medical imaging reports, including:
[0059] 1) First-level features:
[0060] a. Mainly describes the categories of phenomena observed in medical imaging examinations (such as X-ray, CT, MRI, etc.), including lung markings and hilum;
[0061] b. Mainly describe the major categories of quality control observed in medical imaging examinations (such as X-ray, CT, MRI, etc.), including body position and foreign matter;
[0062] 2) Secondary features:
[0063] a. Mainly describe the attributes of the major categories, including shape characteristics, size characteristics and change characteristics;
[0064] b. Details of questions that mainly describe the first-level characteristics of the major categories, including those that are too broad or too narrow;
[0065] 3) Level 3 features:
[0066] Mainly describes the details of secondary features, including location details and size details.
[0067] ②The knowledge graph construction module constructs the knowledge graph based on the relationship between feature entities and the attributes of feature entities, including:
[0068] Build a knowledge graph based on the relationships and attributes between feature entities, store the knowledge graph in the Neo4j database, and connect medical images with corresponding feature entities;
[0069] Create an external server that uses the Faiss vector retrieval library to provide retrieval services, and update the Faiss vector retrieval library synchronously when the knowledge graph is updated;
[0070] Among them, the relationships between feature entities include strong correlation, weak correlation, and relationship attributes (such as correlation: 0.8, mutual exclusivity: 0.9, etc.).
[0071] ③ The knowledge base construction module constructs a disease library, image feature library, and report feature library based on the knowledge graph, including:
[0072] Experts build a disease database based on ICD-11 disease categories and related detailed descriptions;
[0073] Experts classify the detailed features of medical images corresponding to specific diseases (such as primary features, secondary features, tertiary features, etc.) and build an image feature library.
[0074] ④ The knowledge graph association module dynamically obtains medical imaging reports, and experts associate the medical imaging reports with the knowledge graph to build a basic knowledge graph association framework, including:
[0075] Experts split the medical imaging reports according to the characteristic entities in the knowledge graph, clarify the relationship between the characteristic entities and the disease, as well as the information and content that the characteristic entities need to focus on, so as to achieve the association between the medical imaging reports and the knowledge graph and build the basic framework for knowledge graph association.
[0076] The knowledge graph association module dynamically obtains medical imaging reports and automatically associates them with the knowledge graph based on the disease library, image feature library, and report feature library to improve and fill in the knowledge graph association basic framework, thus obtaining the knowledge graph association framework, which includes:
[0077] Experts establish the relationship between medical imaging reports and diseases based on the disease library, image feature library and report feature library, use the pre-trained first feature extraction model to generate text content associated with the feature entities in the knowledge graph, and then use the pre-trained second feature extraction model to extract features from medical images, and add relevant labels to the text content and extracted features to improve and fill in the knowledge graph association basic framework and obtain the knowledge graph association framework.
[0078] ⑤ The user question retrieval module uses the retriever of the RAG model to convert user input into a search query, and retrieves related knowledge graph fragments based on the knowledge graph association framework, including:
[0079] Convert user input into search queries and retrieve related knowledge graph fragments through sparse retrieval and dense retrieval based on the knowledge graph association framework;
[0080] Sort the retrieved related knowledge graph fragments by relevance, and select a preset number of top-ranked related knowledge graph fragments as input to the generator of the RAG model;
[0081] User input includes user questions and medical imaging reports to be analyzed, and search queries include keywords, semantic parsing, or vectorization (e.g., encoding into vectors through the BERT model).
[0082] Sparse search includes keyword matching (such as the BM25 algorithm), and dense search includes vector similarity matching (such as using the Faiss vector retrieval library, Pinecone vector retrieval library, etc.).
[0083] ⑥ The answer content generation module splices the user input with the associated knowledge graph fragment and inputs it into the generator of the RAG model to obtain the answer content that matches the user input, including:
[0084] Splice user input with related knowledge graph fragments to form a complete context;
[0085] Input the context into the generator of the pre-trained RAG model (such as BERT, T5, GPT, etc.), and use the pre-trained RAG model generator to generate coherent, accurate, and relevant knowledge-matched answer content based on the integrated context, comprehensive retrieval content, and its own knowledge.
[0086] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A teaching system for medical imaging reports, characterized by: It includes structured processing module, knowledge graph construction module, knowledge base construction module, knowledge graph association module, user question retrieval module and answer content generation module; The structured processing module describes the medical imaging report in a hierarchical manner and associates the feature entities. The feature entity attributes store relevant interpretation information and related feature vectors. The knowledge graph construction module builds the knowledge graph based on the relationships between feature entities and the attributes of feature entities; The knowledge base construction module builds the disease library, image feature library and report feature library based on the knowledge graph; The knowledge graph association module dynamically obtains medical imaging reports. Experts then associate the medical imaging reports with the knowledge graph to build a knowledge graph association framework. The medical imaging reports are then automatically associated with the knowledge graph based on the disease database, image feature database, and report feature database to complete and fill in the knowledge graph association framework and obtain a knowledge graph association framework. The user question retrieval module uses the retriever of the RAG model to convert user input into a search query and retrieves related knowledge graph fragments based on the knowledge graph association framework; The answer content generation module splices the user input with the associated knowledge graph fragment and inputs it into the generator of the RAG model to obtain the answer content that matches the user input.
2. The teaching system for medical image reporting according to claim 1, characterized in that: The structured processing module performs hierarchical feature description on the medical imaging report, including: 1) First-level features: a. Mainly describes the categories of phenomena observed in medical imaging examinations, including lung markings and hilum; b. Mainly describe the major categories of quality control observed in medical imaging examinations, including body position and foreign bodies; 2) Secondary features: a. Mainly describe the attributes of the major categories, including shape characteristics, size characteristics and change characteristics; b. Details of questions that mainly describe the first-level characteristics of the major categories, including those that are too broad or too narrow; 3) Level 3 features: Mainly describes the details of secondary features, including location details and size details.
3. The teaching system for medical image reporting according to claim 2, characterized in that: The knowledge graph construction module constructs a knowledge graph based on the relationships between feature entities and the attributes of feature entities, including: Build a knowledge graph based on the relationships and attributes between feature entities, store the knowledge graph in the Neo4j database, and connect medical images with corresponding feature entities; Create an external server that uses the Faiss vector retrieval library to provide retrieval services, and update the Faiss vector retrieval library synchronously when the knowledge graph is updated; Among them, the relationships between feature entities include strong correlation, weak correlation, and relationship attributes.
4. The teaching system for medical image reporting according to claim 3, characterized in that: The knowledge base construction module constructs a disease library, an image feature library, and a report feature library based on the knowledge graph, including: Experts build a disease database based on ICD-11 disease categories and related detailed descriptions; Experts classify the detailed features of medical images corresponding to specific diseases and build an image feature library.
5. The teaching system for medical image reporting according to claim 4, characterized in that: The knowledge graph association module dynamically obtains medical imaging reports, and experts associate the medical imaging reports with the knowledge graph to build a basic knowledge graph association framework, including: Experts split the medical imaging reports according to the characteristic entities in the knowledge graph, clarify the relationship between the characteristic entities and the disease, as well as the information and content that the characteristic entities need to focus on, so as to achieve the association between the medical imaging reports and the knowledge graph and build the basic framework for knowledge graph association.
6. The teaching system for medical imaging reports according to claim 5, characterized in that: The knowledge graph association module dynamically obtains medical imaging reports and automatically associates the medical imaging reports with the knowledge graph based on the disease library, image feature library, and report feature library to improve and fill the knowledge graph association basic framework, thereby obtaining the knowledge graph association framework, including: Experts establish the relationship between medical imaging reports and diseases based on the disease library, image feature library and report feature library, use the pre-trained first feature extraction model to generate text content associated with the feature entities in the knowledge graph, and then use the pre-trained second feature extraction model to extract features from medical images, and add relevant labels to the text content and extracted features to improve and fill in the knowledge graph association basic framework and obtain the knowledge graph association framework.
7. The teaching system for medical imaging reports according to claim 6, characterized in that: The user question retrieval module converts user input into a search query using the retriever of the RAG model and retrieves related knowledge graph fragments based on the knowledge graph association framework, including: Convert user input into search queries and retrieve related knowledge graph fragments through sparse retrieval and dense retrieval based on the knowledge graph association framework; Sort the retrieved related knowledge graph fragments by relevance, and select a preset number of top-ranked related knowledge graph fragments as input to the generator of the RAG model; Among them, user input includes user questions and medical imaging reports to be analyzed, and retrieval queries include keywords, semantic parsing or vectorization; Sparse retrieval includes keyword matching, and dense retrieval includes vector similarity matching.
8. The teaching system for medical image reporting according to claim 7, characterized in that: The answer content generation module splices the user input with the associated knowledge graph fragment and inputs it into the generator of the RAG model to obtain the answer content that matches the user input, including: Splice user input with related knowledge graph fragments to form a complete context; The context is input into the generator of the pre-trained RAG model, and the generator of the pre-trained RAG model uses the integrated context, comprehensive retrieval content and its own knowledge to generate coherent, accurate answer content that matches the user input and contains relevant knowledge points.