Ultrasonic report interpretation method and system based on retrieval enhancement generation

Abstract

The invention provides an ultrasonic report interpretation method and system based on retrieval enhancement generation, and belongs to the technical field of report interpretation, and the method comprises the steps: S1, receiving a to-be-interpreted ultrasonic report, and extracting key information from the ultrasonic report; s2, retrieving and recalling target ultrasonic medical knowledge feature vectors matched with the key information from a pre-constructed medical knowledge vector library according to the key information; and S3, inputting the ultrasonic report and the recalled target ultrasonic medical knowledge feature vector into the large language model as a model pre-prompt so as to obtain an interpretation result for the ultrasonic report. The method has the advantages that relevant vectors are retrieved from a medical knowledge vector library through a retrieval enhancement generation mechanism and serve as input pre-prompt words of a large language model, automatic interpretation of an ultrasonic report is achieved, and interpretation efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of report interpretation, and in particular to an ultrasound report interpretation method and system based on retrieval enhanced generation. Background Art

[0002] Ultrasound is a commonly used medical imaging technique used to evaluate the structure and function of internal organs. However, interpreting ultrasound reports is often a challenging task for non-medical professionals. Patients often have difficulty understanding medical terminology and complex image data, resulting in insufficient understanding of their own health conditions.

[0003] At present, the interpretation of the text content of ultrasound reports mainly relies on manual analysis, traditional natural language processing (NLP) algorithms, and knowledge graph-based systems. Manual analysis requires patients to obtain the report and make an appointment with a doctor for face-to-face explanation and consultation. The process is time-consuming and inconvenient, and it mainly relies on the doctor's professional knowledge and experience. The interpretations of different doctors may vary greatly, and the accuracy and consistency of the interpretations are difficult to guarantee.

[0004] Traditional natural language processing algorithms usually include text preprocessing, keyword extraction, sentence parsing and basic semantic analysis. They use predefined rules and limited machine learning models to interpret reports. They can only provide basic, static explanations, and are unable to provide deeper and more complex explanations. They also have limited ability to handle complex semantics and contextual associations.

[0005] The knowledge graph-based system uses the medical knowledge graph to parse and interpret the text content of the ultrasound report. The knowledge graph stores and associates medical knowledge in a structured way, which requires a lot of professional knowledge and data, and the construction process is complex and time-consuming. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides an ultrasound report interpretation method based on retrieval enhancement generation; on the other hand, it also provides an ultrasound report interpretation system based on retrieval enhancement generation.

[0007] The technical problem solved by the present invention can be achieved by adopting the following technical solutions:

[0008] A first aspect of the present invention is to provide an ultrasound report interpretation method based on retrieval enhancement generation, comprising:

[0009] Step S1, receiving an ultrasound report to be interpreted, and extracting key information from the ultrasound report;

[0010] Step S2, retrieving and recalling target ultrasound medical knowledge features matching the key information from a pre-built medical knowledge base according to the key information;

[0011] Step S3, inputting the ultrasound report and the recalled target ultrasound medical knowledge features into the large language model as model pre-prompts to obtain an interpretation result for the ultrasound report.

[0012] Preferably, the medical knowledge base is a vector database.

[0013] Preferably, the steps of constructing the medical knowledge base include:

[0014] Step S201, obtaining an ultrasound medical report;

[0015] Step S202, parsing the ultrasound medical report and extracting text information from the parsing result;

[0016] Step S203, cleaning, deduplication and format conversion of the text information to obtain a preprocessed text;

[0017] Step S204, analyzing the preprocessed text, including table conversion, cross-paragraph merging, optical character recognition, and converting the analysis result into an ultrasonic medical knowledge feature vector;

[0018] Step S205: storing the ultrasonic medical knowledge feature vector and the corresponding index item in the medical knowledge base.

[0019] Preferably, it also includes:

[0020] Step S206: regularly updating the medical knowledge base.

[0021] Preferably, the medical knowledge base stores the ultrasonic medical knowledge vector and the corresponding index item;

[0022] The step S2 comprises:

[0023] Step S211, performing vector conversion on the key information to obtain vectorized key information;

[0024] Step S212, retrieving and recalling target index items matching the vectorized key information from the medical knowledge base according to the vectorized key information;

[0025] Step S213, calculating the cosine similarity between the vectorized key information and each of the ultrasound medical knowledge vectors in the set of ultrasound medical knowledge vectors corresponding to the target index item;

[0026] Step S214: determining a corresponding ultrasonic medical knowledge feature vector according to the cosine similarity as the target ultrasonic medical knowledge feature.

[0027] A second aspect of the present invention is to provide an ultrasound report interpretation system based on retrieval enhancement generation, comprising:

[0028] An extraction module, used for receiving an ultrasound report to be interpreted and extracting key information from the ultrasound report;

[0029] A retrieval enhancement generation module, connected to the extraction module, for retrieving and recalling target ultrasound medical knowledge features matching the key information from a pre-built medical knowledge base according to the key information;

[0030] The report interpretation module is connected to the retrieval enhancement generation module and is used to input the ultrasound report and the recalled target ultrasound medical knowledge features into the large language model as model pre-prompts to obtain the interpretation result for the ultrasound report.

[0031] Preferably, the medical knowledge base is a vector database.

[0032] Preferably, it also includes:

[0033] A data collection module for obtaining ultrasound medical reports;

[0034] A parsing module, connected to the data collection module, for parsing the ultrasound medical report and extracting text information from the parsing result;

[0035] A preprocessing module, connected to the parsing module, for cleaning, deduplicating and formatting the text information to obtain a preprocessed text;

[0036] An analysis and vector conversion module, connected to the preprocessing module, is used to analyze the preprocessed text, including table conversion, cross-paragraph merging, optical character recognition, and convert the analysis result into an ultrasonic medical knowledge feature vector;

[0037] A knowledge base construction module is connected to the analysis and vector conversion module and is used to store the ultrasonic medical knowledge feature vector and the corresponding index item in the medical knowledge base.

[0038] Preferably, it also includes:

[0039] The knowledge base updating module is used to regularly update the medical knowledge base.

[0040] Preferably, the medical knowledge base stores the ultrasonic medical knowledge vector and the corresponding index item;

[0041] The retrieval enhancement generation module comprises:

[0042] A vector conversion unit, used for performing vector conversion on the key information to obtain vectorized key information;

[0043] a matching unit, connected to the vector conversion unit, for retrieving and recalling target index items matching the vectorized key information from the medical knowledge base according to the vectorized key information;

[0044] a similarity calculation unit, connected to the matching unit, and used to calculate the cosine similarity between the vectorized key information and each of the ultrasound medical knowledge vectors in the set of ultrasound medical knowledge vectors corresponding to the target index item;

[0045] A target vector determination unit is connected to the similarity calculation unit and is used to determine a corresponding ultrasonic medical knowledge feature vector according to the cosine similarity as the target ultrasonic medical knowledge feature.

[0046] The advantages or beneficial effects of the technical solution of the present invention are:

[0047] The present invention retrieves relevant vectors from a medical knowledge base through a retrieval enhancement generation mechanism and uses them as input pre-prompt words for a large language model, thereby achieving automated interpretation of ultrasound reports and improving interpretation efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of a flow chart of an ultrasound report interpretation method based on retrieval enhancement generation in a preferred embodiment of the present invention;

[0049] Figure 2 A schematic diagram of the process of constructing a medical knowledge base in a preferred embodiment of the present invention;

[0050] Figure 3 A schematic diagram of a process flow of search enhancement generation in a preferred embodiment of the present invention;

[0051] Figure 4 It is a structural block diagram of an ultrasound report interpretation system based on retrieval enhancement generation in a preferred embodiment of the present invention;

[0052] Figure 5 This is a structural block diagram of an ultrasound report interpretation system in a preferred embodiment of the present invention;

[0053] Figure 6 A structural block diagram of a search enhancement generation module in a preferred embodiment of the present invention;

[0054] Figure 7 The figure is a schematic diagram of the overall process of ultrasound report interpretation in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 creative work are within the scope of protection of the present invention.

[0056] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0057] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0058] See also Figures 1 to 7 In a preferred embodiment of the present invention, based on the above problems existing in the prior art, a method for interpreting ultrasound reports based on retrieval-augmented generation (RAG) is provided, comprising:

[0059] Step S1, receiving an ultrasound report to be interpreted, and extracting key information from the ultrasound report;

[0060] Step S2, retrieving and recalling target ultrasound medical knowledge features matching the key information from a pre-built medical knowledge base according to the key information;

[0061] Step S3, inputting the ultrasound report and the recalled target ultrasound medical knowledge features into the large language model as model pre-prompts to obtain the interpretation result for the ultrasound report.

[0062] Specifically, in this embodiment, the ultrasound report A to be interpreted is usually uploaded electronically or manually input into the system by the patient. After receiving the ultrasound report A, the system uses natural language processing technology, such as named entity recognition, keyword extraction, etc., to extract key information from the ultrasound report A. Among them, the key information includes but is not limited to the patient's basic information, examination site, equipment parameters, examination results, abnormal findings, etc. According to the key information extracted in step S1, vector similarity calculation, such as cosine similarity and other methods, is used to retrieve and recall the target ultrasound medical knowledge features that match the key information from the medical knowledge base based on the retrieval enhancement generation mechanism. The original ultrasound report and the retrieved target ultrasound medical knowledge features are provided as input to the large language model.

[0063] The embodiment of the present invention utilizes retrieval enhancement generation technology to retrieve knowledge before the large language model answers, and then uses the retrieved knowledge as a pre-prompt word to allow the large model's answer to dynamically utilize the retrieved knowledge documents when generating answers, thereby helping the large language model to better understand the content of the ultrasound report, significantly improving the accuracy and efficiency of ultrasound report interpretation, and providing a report interpretation that is easier for patients to understand.

[0064] As a preferred embodiment, the medical knowledge base is a vector database.

[0065] Specifically, relational databases are usually used for existing medical knowledge bases. Relational databases store and organize data through table structures, relational models, etc., and need to rely on precise matching and structured queries (such as SQL). They do not provide sufficient support for complex and fuzzy semantic queries. In the medical field, much knowledge and information may be fuzzy and context-based, and relational databases may not be able to handle such queries.

[0066] In this embodiment, the medical knowledge base does not use a relational database, but a vector database, such as Chroma. In a vector database, data is stored in the form of high-dimensional vectors, which are obtained by converting text information such as medical literature, case reports, and expert interpretations, and are usually used to represent complex data types.

[0067] In the subsequent retrieval process, useful information is extracted and retrieved through high-dimensional vector representation and similarity semantic matching, which can be used to more comprehensively represent and retrieve complex data in the medical field.

[0068] Since vector representations do not rely on specific data structures, vector databases can adapt to new data and query requirements.

[0069] As a preferred embodiment, wherein Figure 2 As shown in Figure 1, the steps for building a medical knowledge base include:

[0070] Step S201, obtaining an ultrasound medical report;

[0071] Step S202, parsing the ultrasound medical report and extracting text information from the parsing result;

[0072] Step S203, cleaning, deduplication and format conversion of the text information to obtain a preprocessed text;

[0073] Step S204, analyzing the preprocessed text, including table conversion, cross-paragraph merging, optical character recognition, and converting the analysis results into ultrasonic medical knowledge feature vectors;

[0074] Step S205: storing the ultrasonic medical knowledge feature vector and the corresponding index item in the medical knowledge base.

[0075] Specifically, in this embodiment, knowledge is expressed through text vectorization, table vectorization and image vectorization to construct a medical knowledge base. The construction process is as follows:

[0076] A large number of ultrasound medical reports are obtained through hospital information systems (HIS), electronic medical record systems (EMR) or other channels. These reports can be stored in various formats, such as word, ppt, excel, txt, md and pdf.

[0077] Parse the report, identify and extract the text information therein, and pre-process the text information, including but not limited to cleaning, deduplication and format conversion; remove irrelevant characters, stop words, repeated information, etc. through cleaning, deduplication to ensure the uniqueness of each text information, and convert the text information into a unified format that can be processed by the system through format conversion.

[0078] Next, the content of the preprocessed text is analyzed through content parsing, such as table conversion, cross-paragraph merging, and optical character recognition (OCR), and keywords are extracted and converted into ultrasound medical knowledge feature vectors;

[0079] Furthermore, the preprocessed text is in the form of a document, and the document can be split according to a preset splitting rule. Figure 7 As shown, the preset splitting rule includes at least one of the following:

[0080] Chapter splitting, slice splitting, semantic splitting, transaction splitting or other splitting rules.

[0081] After splitting the documents, data enhancement services are provided, such as QA generation service, keyword service, and summary service.

[0082] Finally, the processed vector data and the corresponding index items are stored in the medical knowledge base for subsequent query and use.

[0083] In this embodiment, the index items may include a keyword index and a vector index. The keyword index and the vector index can be used to quickly retrieve relevant information and accurately reflect the entire content and semantics of the text.

[0084] The keyword index can be established based on the keywords in the text content. In the medical knowledge base, the keyword index can cover various medical terms, disease names, examination items, etc. The construction of the keyword index includes the following steps:

[0085] First, keywords with representativeness and retrieval value are identified and extracted from the preprocessed text;

[0086] Then, the extracted keywords are standardized, such as removing synonyms and abbreviation conversion, to ensure the consistency and accuracy of the keywords;

[0087] Finally, the standardized keywords are associated with the corresponding text or feature vectors to form a keyword index.

[0088] Vector index is established based on the ultrasonic medical knowledge feature vector of text, table or image content. In the medical knowledge base, vector index is usually established by converting text content into feature vectors and assigning a unique index identifier to each feature vector for fast retrieval.

[0089] As a preferred embodiment, the invention further comprises:

[0090] Step S206: regularly update the medical knowledge base.

[0091] Specifically, in this embodiment, an automated knowledge updating mechanism is adopted, such as using crawler technology to automatically collect the latest medical research results and clinical data, to improve the updating speed and efficiency of the knowledge base, and to ensure the timeliness of knowledge.

[0092] As a preferred embodiment, the medical knowledge base stores ultrasound medical knowledge vectors and corresponding index items;

[0093] like Figure 3 As shown, step S2 includes:

[0094] Step S211, performing vector conversion on the key information to obtain vectorized key information;

[0095] Step S212, searching and recalling target index items matching the vectorized key information from the medical knowledge base according to the vectorized key information;

[0096] Step S213, calculating the cosine similarity of each ultrasound medical knowledge vector in the set of ultrasound medical knowledge vectors corresponding to the vectorized key information and the target index item, and the calculation formula is as follows:

[0097]

[0098] Among them, A and B are two n-dimensional vectors, namely the vector of information in the ultrasound report to be interpreted and the vector in the medical knowledge base; i is the dimension index.

[0099] Step S214: determining a corresponding ultrasonic medical knowledge feature vector according to the cosine similarity as a target ultrasonic medical knowledge feature.

[0100] Specifically, in this embodiment, multiple key information in text form, such as various descriptions in the ultrasound report, diagnostic keywords, etc., are vectorized; then, a fast search is performed based on the index items in the medical knowledge base to obtain a set of index items similar to the target key information vector, and each index item is associated with one or more ultrasound medical knowledge vectors.

[0101] Then, the cosine similarity formula is used to calculate the cosine value of the angle between the target key information vector and each candidate ultrasound medical knowledge vector retrieved based on the index. The value is between -1 and 1, and the larger the value, the more similar the two vectors are.

[0102] All calculated angle cosine values ​​are aggregated and sorted to obtain a similarity score ranking, where each score corresponds to a candidate ultrasound medical knowledge vector.

[0103] From the candidate ultrasound medical knowledge vectors, one or more ultrasound medical knowledge vectors with high similarity scores are selected as the vectors most similar to the vectorized key information, that is, the final target ultrasound medical knowledge features.

[0104] Retrieval-augmented generation is a technique that enhances the accuracy and reliability of generative AI models by obtaining facts from external sources. RAG fills a gap in the way large language models work. RAG provides the model with sources that can be cited, helping the model clarify ambiguities in user queries and reducing the likelihood that the model will guess wrong.

[0105] In this embodiment, in addition to RAG, other models that combine retrieval and generation mechanisms can also be used, such as but not limited to REALM, FiD (Fusion-in-Decoder), etc. According to the performance of different models in processing specific tasks, the appropriate model is selected to further improve the accuracy of the explanation.

[0106] See also Figure 4 The second aspect of the present invention is to provide an ultrasound report interpretation system based on retrieval enhancement generation, comprising:

[0107] Extraction module 1, used for receiving the ultrasound report to be interpreted and extracting key information from the ultrasound report;

[0108] The retrieval enhancement generation module 2 is connected to the extraction module 1, and is used to retrieve and recall target ultrasound medical knowledge features that match the key information from a pre-built medical knowledge base according to the key information;

[0109] The report interpretation module 3 is connected to the retrieval enhancement generation module 2 and is used to input the ultrasound report and the recalled target ultrasound medical knowledge features into the large language model as model pre-prompts to obtain the interpretation result for the ultrasound report.

[0110] Specifically, the system of the embodiment of the present invention uses RAG-based natural language processing to process the text and semantic understanding of ultrasound reports. RAG combines the two mechanisms of retrieval and generation, and can retrieve relevant information from the medical knowledge base. The large language model generates natural language interpretation based on the ultrasound report content and the retrieved content.

[0111] Question answering based on a large language model and a dynamically updated medical knowledge base, combined with retrieval enhancement generation technology, can automatically interpret the text content of ultrasound reports, significantly improve the accuracy and efficiency of interpreting the text content of ultrasound reports, and provide comprehensive and dynamically updated medical knowledge support.

[0112] The medical knowledge base contains the latest medical research results and clinical experience, and keeps the knowledge up to date through regular updates. The medical knowledge base can provide the background knowledge required for interpretation, ensure the timeliness and accuracy of the interpretation content, and provide the latest medical knowledge support.

[0113] As a preferred embodiment, the medical knowledge base 4 is a vector database.

[0114] Specifically, in the vector database, data is stored in the form of high-dimensional vectors. In the subsequent retrieval process, useful information is extracted and retrieved through high-dimensional vector representation and similarity semantic matching, which can be applied to more comprehensive representation and retrieval of complex data in the medical field.

[0115] Since vector representations do not rely on specific data structures, vector databases can adapt to new data and query requirements.

[0116] As a preferred embodiment, wherein Figure 5 As shown, it also includes:

[0117] A data collection module 5, for obtaining ultrasound medical reports;

[0118] The parsing module 6 is connected to the data collection module 5 and is used to parse the ultrasound medical report and extract text information from the parsing result;

[0119] The preprocessing module 7 is connected to the parsing module 6 and is used to clean, remove duplicates and convert the format of the text information to obtain a preprocessed text;

[0120] The analysis and vector conversion module 8 is connected to the preprocessing module 7 and is used to analyze the preprocessed text, including table conversion, cross-paragraph merging, optical character recognition, and convert the analysis results into ultrasonic medical knowledge feature vectors;

[0121] The knowledge base construction module 9 and the connection analysis and vector conversion module 8 are used to store the ultrasonic medical knowledge feature vectors and the corresponding index items in the medical knowledge base.

[0122] As a preferred embodiment, wherein Figure 5 As shown, it also includes:

[0123] The knowledge base updating module 10 is used to regularly update the medical knowledge base.

[0124] As a preferred embodiment, the medical knowledge base stores ultrasound medical knowledge vectors and corresponding index items;

[0125] like Figure 6 As shown, the search enhancement generation module 2 includes:

[0126] A vector conversion unit 21, used for performing vector conversion on the key information to obtain vectorized key information;

[0127] A matching unit 22, connected to the vector conversion unit 21, is used to retrieve and recall target index items matching the vectorized key information from the medical knowledge base according to the vectorized key information;

[0128] A similarity calculation unit 23, connected to the matching unit 22, is used to calculate the cosine similarity of each ultrasound medical knowledge vector in the set of ultrasound medical knowledge vectors corresponding to the vectorized key information and the target index item;

[0129] The target vector determination unit 24 is connected to the similarity calculation unit 23 and is used to determine a corresponding ultrasonic medical knowledge feature vector according to the cosine similarity as a target ultrasonic medical knowledge feature.

[0130] More implementation details of the system and the method have been disclosed and will not be repeated in this embodiment.

[0131] The present invention may also provide a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0132] Among them, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU) and a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or at least one of other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc., which can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present invention.

[0133] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory is used to store a computer program, and the processor may execute the computer program accordingly after receiving an execution instruction.

[0134] The present invention may also provide a computer-readable storage medium having a computer program stored thereon, and the computer program implements the steps of the above method when executed by a processor.

[0135] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or the flowchart, and the combination of boxes in the structure diagram and / or the flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0136] In addition, the functional modules or units in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0137] If the above functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a smart phone, a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0138] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. A method for interpreting ultrasound reports based on retrieval-enhanced generation, characterized in that: include: Step S1, receiving an ultrasound report to be interpreted, and extracting key information from the ultrasound report; Step S2, retrieving and recalling target ultrasound medical knowledge features matching the key information from a pre-built medical knowledge base according to the key information; Step S3, inputting the ultrasound report and the recalled target ultrasound medical knowledge features into the large language model as model pre-prompts to obtain an interpretation result for the ultrasound report.

2. The ultrasound report interpretation method based on retrieval enhancement generation according to claim 1, characterized in that: The medical knowledge base is a vector database.

3. The ultrasound report interpretation method based on retrieval enhancement generation according to claim 1 is characterized in that: The steps of constructing the medical knowledge base include: Step S201, obtaining an ultrasound medical report; Step S202, parsing the ultrasound medical report and extracting text information from the parsing result; Step S203, cleaning, deduplication and format conversion of the text information to obtain a preprocessed text; Step S204, analyzing the preprocessed text, including table conversion, cross-paragraph merging, optical character recognition, and converting the analysis result into an ultrasonic medical knowledge feature vector; Step S205: storing the ultrasonic medical knowledge feature vector and the corresponding index item in the medical knowledge base.

4. The ultrasound report interpretation method based on retrieval enhanced generation according to claim 1, characterized in that: Also includes: Step S206: regularly updating the medical knowledge base.

5. The ultrasound report interpretation method based on retrieval enhancement generation according to claim 1, characterized in that: The medical knowledge base stores the ultrasonic medical knowledge vector and the corresponding index item; The step S2 comprises: Step S211, performing vector conversion on the key information to obtain vectorized key information; Step S212, retrieving and recalling target index items matching the vectorized key information from the medical knowledge base according to the vectorized key information; Step S213, calculating the cosine similarity between the vectorized key information and each of the ultrasound medical knowledge vectors in the set of ultrasound medical knowledge vectors corresponding to the target index item; Step S214: determining a corresponding ultrasonic medical knowledge feature vector according to the cosine similarity as the target ultrasonic medical knowledge feature.

6. An ultrasound report interpretation system based on retrieval enhancement generation, characterized in that: include: An extraction module, used for receiving an ultrasound report to be interpreted and extracting key information from the ultrasound report; A retrieval enhancement generation module, connected to the extraction module, for retrieving and recalling target ultrasound medical knowledge features matching the key information from a pre-built medical knowledge base according to the key information; The report interpretation module is connected to the retrieval enhancement generation module and is used to input the ultrasound report and the recalled target ultrasound medical knowledge features into the large language model as model pre-prompts to obtain the interpretation result for the ultrasound report.

7. The ultrasound report interpretation system based on retrieval enhancement generation according to claim 6, characterized in that: The medical knowledge base is a vector database.

8. The ultrasound report interpretation system based on retrieval enhancement generation according to claim 6, characterized in that: Also includes: A data collection module for obtaining ultrasound medical reports; A parsing module, connected to the data collection module, for parsing the ultrasound medical report and extracting text information from the parsing result; A preprocessing module, connected to the parsing module, for cleaning, deduplicating and formatting the text information to obtain a preprocessed text; An analysis and vector conversion module, connected to the preprocessing module, is used to analyze the preprocessed text, including table conversion, cross-paragraph merging, optical character recognition, and convert the analysis result into an ultrasonic medical knowledge feature vector; A knowledge base construction module is connected to the analysis and vector conversion module and is used to store the ultrasonic medical knowledge feature vector and the corresponding index item in the medical knowledge base.

9. The ultrasound report interpretation system based on retrieval enhancement generation according to claim 6, characterized in that: Also includes: The knowledge base updating module is used to regularly update the medical knowledge base.

10. The ultrasound report interpretation system based on retrieval enhancement generation according to claim 6, characterized in that: The medical knowledge base stores the ultrasonic medical knowledge vector and the corresponding index item; The retrieval enhancement generation module comprises: A vector conversion unit, used for performing vector conversion on the key information to obtain vectorized key information; a matching unit, connected to the vector conversion unit, for retrieving and recalling target index items matching the vectorized key information from the medical knowledge base according to the vectorized key information; a similarity calculation unit, connected to the matching unit, and used to calculate the cosine similarity between the vectorized key information and each of the ultrasound medical knowledge vectors in the set of ultrasound medical knowledge vectors corresponding to the target index item; A target vector determination unit is connected to the similarity calculation unit and is used to determine a corresponding ultrasonic medical knowledge feature vector according to the cosine similarity as the target ultrasonic medical knowledge feature.

Images