AI physical examination report generation method based on large language model and storage medium

Abstract

This invention relates to an AI-based method and storage medium for generating medical examination reports based on a large-scale language model. The method includes: data acquisition: collecting medical examination data from medical examination equipment or systems, the medical examination data including structured and unstructured data; data preprocessing: preprocessing the collected medical examination data to convert unstructured data into structured data; AI model correction: using historical medical examination data as training samples to correct and train the AI ​​model; AI model analysis: analyzing the preprocessed medical examination data using the AI ​​model based on a knowledge graph and AST logical analysis framework to obtain analysis results; report generation: generating a medical examination report based on the analysis results obtained from the AI ​​model analysis using a report template. By automatically analyzing and integrating the report using the AI ​​model, the efficiency of medical examination report generation is improved, while significantly reducing the manual operation costs for doctors and accelerating report delivery.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to an AI-based method for generating physical examination reports and a storage medium based on a large language model. Background Technology

[0002] In the field of medical examinations, the examination report serves as the core carrier for health status assessment and intervention recommendations; its generation quality and efficiency directly impact the patient experience and the effectiveness of health management. Currently, the generation of examination reports in the industry still primarily relies on traditional methods, mainly depending on manual integration and standardized template output. Specifically:

[0003] 1. Physical examination data is stored in different devices or systems such as LIS (Laboratory Information System), PACS (Picture Archiving and Communication System), and electrocardiograph. Doctors need to manually retrieve and summarize the results of various examinations, and then combine their personal experience to write diagnostic conclusions and chief physician recommendations.

[0004] 2. Most mainstream fluid inspection report systems use fixed templates, which can only mechanically list the results of a single inspection and compare them with standard reference values, lacking correlation analysis of multi-dimensional data.

[0005] 3. Historical physical examination data lacks systematic integration. The clinical experience and diagnostic logic of different doctors are scattered in individual cases. There is a knowledge transfer gap when collaborating across departments, and newly hired doctors find it difficult to quickly reuse mature diagnostic ideas.

[0006] The existing medical examination report generation method has the following problems:

[0007] Disadvantage 1: Low generation efficiency. Doctors need to spend a lot of time manually summarizing physical examination data from multiple systems, eliminating duplicate information, and filling in missing data. The average generation time for a single physical examination report exceeds 1 hour, which seriously affects the delivery cycle of physical examination reports.

[0008] Disadvantage 2: Insufficient diagnostic relevance. Traditional templates only support the comparison of a single indicator with the reference value and cannot identify the potential association between multiple positive findings, resulting in the omission of some hidden health risks.

[0009] Disadvantage 3: Poor accuracy of suggestions. The standardized template does not take into account individual differences and hospital specialties. The output of the main examination suggestions is highly general but not specific, requiring doctors to make secondary modifications and adjustments, thus extending the report optimization cycle.

[0010] Disadvantage 4: Low knowledge reuse rate. Diagnostic logic and typical cases in historical physical examination reports are stored in a scattered manner and lack structured integration, making it impossible to form a reusable knowledge system. This leads to repeated analysis of similar physical examination cases, making it difficult to ensure diagnostic consistency. Summary of the Invention

[0011] In view of the above problems, this application provides an AI-based method and storage medium for generating physical examination reports based on a large language model, which solves the problems of low generation efficiency caused by the reliance on manual processing in existing physical examination report generation.

[0012] To achieve the above objectives, the inventors provide an AI-based method for generating medical examination reports based on a large-scale language model, comprising:

[0013] Data collection: Collecting medical examination data from medical examination equipment or medical examination system, including structured data and unstructured data;

[0014] Data preprocessing: The collected physical examination data is preprocessed to transform unstructured data into structured data;

[0015] AI model correction: Using historical physical examination data as training samples to correct and train the AI ​​model;

[0016] AI Model Analysis: The preprocessed physical examination data is analyzed using an AI model based on a knowledge graph and AST logical analysis framework to obtain the analysis results;

[0017] Report generation: Generate a physical examination report based on the analysis results obtained from the AI ​​model analysis using the report template.

[0018] In some embodiments, the data acquisition specifically includes;

[0019] It monitors the completion status of medical examination items in real time using medical examination equipment or systems, and collects medical examination data from the medical examination equipment or systems when the medical examination items are completed.

[0020] In some embodiments, the data preprocessing specifically includes the following steps:

[0021] Use regular expressions and rule engines to remove distracting data from unstructured data;

[0022] Named entity recognition technology is used to automatically extract key entities and transform unstructured data into structured data.

[0023] In some embodiments, the data preprocessing further includes the following steps:

[0024] Numerical verification and range standardization are performed on structured data.

[0025] In some embodiments, the AI ​​model correction specifically includes the following steps:

[0026] Historical physical examination data is used as training samples, which contain the mapping relationship between various examination indicator combinations, analysis conclusions, and chief physician recommendations.

[0027] The AI ​​model is trained by labeling correct cases and optimizing the loss function.

[0028] In some embodiments, the AI ​​model analysis specifically includes:

[0029] The AI ​​model constructs a triplet knowledge graph from the physical examination data. The triplet knowledge graph includes abnormal indicators, potential causes, and main examination recommendations.

[0030] In some embodiments, the AI ​​model analysis specifically includes:

[0031] A triplet knowledge graph is constructed from physical examination data through semantic matching, multi-indicator association reasoning, individual matching analysis, and historical trend comparison.

[0032] In some embodiments, it also includes:

[0033] Interaction optimization: Provide a preview of the physical examination report. When a doctor requests an objection to the contents of the physical examination, the AI ​​model analysis is retried, or feedback is given to the AI ​​model for optimization after receiving the doctor's adjustment.

[0034] In some embodiments, it also includes:

[0035] Knowledge base closed-loop update: Once the generated physical examination report is confirmed to be correct, the report is structured and stored in the physical examination knowledge base.

[0036] Another technical solution is also provided: a storage medium storing a computer program, which, when run by a processor, executes the steps of the AI-based physical examination report generation method based on a large language model as described above.

[0037] Unlike existing technologies, the above-mentioned technical solution acquires the examinee's medical examination data from the examination equipment or system during the examination. This data includes both unstructured and structured data. The collected data is preprocessed to convert unstructured data into structured data. Then, historical examination data is used as training samples to train and correct the AI ​​model. The trained AI model analyzes the examination data based on a knowledge graph and AST logical analysis framework to obtain the analysis results. Finally, a medical examination report is generated based on the examination data and analysis results using a medical examination report template. By automatically collecting data and then automatically analyzing and integrating it with the report using an AI model, the efficiency of medical examination report generation is improved, while significantly reducing the manual operation costs for doctors and accelerating report delivery.

[0038] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0039] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of this application and other related content, and should not be considered as limitations on this application.

[0040] In the accompanying drawings of the instruction manual:

[0041] Figure 1 A flowchart illustrating an AI-based health check report generation method based on a large language model, as described in a specific implementation.

[0042] Figure 2 This is another flowchart illustrating the AI-based physical examination report generation method based on a large language model, as described in the specific implementation.

[0043] Figure 3 A schematic diagram of the structure of the storage medium described in a specific embodiment.

[0044] Figure 4 A flowchart illustrating the overall process of the AI-based health check report technology based on a large language model, as described in a specific implementation;

[0045] Figure 5 The diagram shows the modules of the AI-based health check report technology based on a large language model, as described in the specific implementation.

[0046] The reference numerals used in the above figures are explained as follows:

[0047] 310. Storage medium,

[0048] 320. Processor. Detailed Implementation

[0049] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0050] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0051] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0052] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0053] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0054] Unless otherwise specified, the use of terms such as “comprising,” “including,” “having,” or other similar expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0055] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0056] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0057] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. For those skilled in the art to which this application pertains, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0058] Please see Figure 1 This embodiment provides an AI-based method for generating physical examination reports based on a large language model, including:

[0059] Step S110: Data Acquisition: Collect physical examination data from physical examination equipment or physical examination system, wherein the physical examination data includes structured data and unstructured data;

[0060] Step S120: Data preprocessing: Preprocess the collected physical examination data to convert unstructured data into structured data;

[0061] Step S130: AI Model Correction: Use historical physical examination data as training samples to correct and train the AI ​​model;

[0062] Step S140: AI Model Analysis: The preprocessed physical examination data is analyzed using an AI model based on a knowledge graph and AST logical analysis framework to obtain the analysis results;

[0063] Step S150: Report generation: Generate a physical examination report based on the analysis results obtained from the AI ​​model analysis according to the report template.

[0064] When conducting physical examinations, the system acquires the examinees' data from the examination equipment or system. This data includes both unstructured and structured data. The collected data undergoes preprocessing, converting unstructured data into structured data. Then, historical examination data is used as training samples to train and refine the AI ​​model. The trained AI model then analyzes the examination data using a knowledge graph and AST (Abstract Strategy Framework) logic analysis framework to obtain the analysis results. Finally, a physical examination report is generated based on the data and analysis results using a template. This automatic data collection, followed by automatic analysis and report integration by the AI ​​model, not only improves the efficiency of report generation but also significantly reduces manual operation costs for doctors and accelerates report delivery.

[0065] In some embodiments, the data acquisition specifically includes;

[0066] It monitors the completion status of medical examination items in real time using medical examination equipment or systems, and collects medical examination data from the medical examination equipment or systems when the medical examination items are completed.

[0067] An event-driven mechanism is adopted to monitor the completion status of each examination item in real time, ensuring that the physical examination data is collected in a timely and complete manner, providing comprehensive and up-to-date data support for subsequent preprocessing and model analysis.

[0068] In some embodiments, the data preprocessing specifically includes the following steps:

[0069] Use regular expressions and rule engines to remove distracting data from unstructured data;

[0070] Named entity recognition technology is used to automatically extract key entities and transform unstructured data into structured data.

[0071] For unstructured medical text (such as image descriptions and doctor's notes), we use regular expressions and a medical-specific rule engine to remove interfering information and use named entity recognition (NER) technology to automatically extract key entities (such as indicator names, values, lesion locations, and nodule sizes), transforming unstructured data into structured features.

[0072] In some embodiments, the data preprocessing further includes the following steps:

[0073] Numerical verification and range standardization are performed on structured data.

[0074] For structured data, numerical verification and range standardization are performed to ensure data format uniformity and provide high-quality input for model correction.

[0075] In some embodiments, the AI ​​model correction specifically includes the following steps:

[0076] Historical physical examination data is used as training samples, which contain the mapping relationship between various examination indicator combinations, analysis conclusions, and chief physician recommendations.

[0077] The AI ​​model is trained by labeling correct cases and optimizing the loss function.

[0078] Using historical physical examination reports from hospital examination departments as training samples, the general Base model was fine-tuned. The training samples included a mapping relationship of "combination of various examination indicators - analysis conclusion - chief examiner's recommendation" (e.g., "uric acid 460 umol / L + 26-year-old male - elevated uric acid - low-purine diet + regular follow-up examination"). By labeling correct cases and optimizing the loss function, the model was adapted to the physical examination report generation scenario, improving the accuracy of diagnosis and recommendations.

[0079] In some embodiments, the AI ​​model analysis specifically includes:

[0080] The AI ​​model constructs a triplet knowledge graph from the physical examination data. The triplet knowledge graph includes abnormal indicators, potential causes, and main examination recommendations.

[0081] Construct a ternary knowledge graph of "abnormal indicators - potential causes - chief physician recommendations" (e.g., "elevated cytokeratin 19 fragment → risk of lung lesions → further MRI examination" and "minor tricuspid regurgitation → physiological phenomenon → no special treatment required") to enhance the model's ability to perform correlation analysis on multiple positive findings.

[0082] In some embodiments, the AI ​​model analysis specifically includes:

[0083] A triplet knowledge graph is constructed from physical examination data through semantic matching, multi-indicator association reasoning, individual matching analysis, and historical trend comparison.

[0084] The medical examination data is analyzed through several dimensions, including semantic matching (rapidly matching similar historical examination cases), multi-indicator association reasoning (analyzing causal or synergistic relationships between different positive findings), individual adaptation analysis (adjusting diagnostic weights based on age, gender, and past medical history), and historical trend comparison (analyzing indicator trends if user's historical examination data exists). A ternary knowledge graph is then constructed. Through medical scenario data correction using a general Base model and knowledge graph construction, the model can accurately identify the relationships between multiple positive findings, generate personalized suggestions based on individual characteristics, avoid the generality limitations of standardized templates, reduce the workload of doctors in making secondary revisions, and improve the clinical reference value of medical examination reports.

[0085] Please see Figure 2 In some embodiments, it also includes:

[0086] Step S210: Interaction optimization: Provide a preview of the physical examination report. When a doctor requests an objection to the physical examination, the AI ​​model analysis is retried, or feedback is given to the AI ​​model for optimization after receiving the doctor's adjustment.

[0087] It offers report preview and AI regeneration functions. If doctors disagree with the generated results, they can trigger model reanalysis or manually adjust the data and provide feedback for model optimization. Specifically, it adopts a RAG (Retrieval Enhanced Generation) architecture to ensure that the model prioritizes the use of the latest physical examination cases and clinical guidelines when generating suggestions, guaranteeing the timeliness of knowledge. It provides a visual knowledge graph editing interface, allowing physical examination experts to optimize the relationship between "abnormal indicators - causes - suggestions" through drag-and-drop operations, achieving a deep integration of AI intelligent analysis and expert experience, and continuously improving the accuracy of the knowledge base.

[0088] Please see Figure 2 In some embodiments, it also includes:

[0089] Step S220: Knowledge base closed-loop update: After the generated physical examination report is confirmed to be correct, the physical examination report is processed in a structured manner and stored in the physical examination knowledge base.

[0090] Once the doctor confirms that the AI-generated medical examination report is accurate, the system automatically structures the case (including all examination data, analysis conclusions, and final recommendations) and stores it in the medical examination knowledge base, achieving real-time knowledge accumulation. Simultaneously, the system dynamically optimizes the model using reinforcement learning technology. If the diagnostic accuracy for a certain type of medical examination scenario (such as "elevated uric acid + low hemoglobin in young men") falls below 85%, the system automatically triggers a sample expansion process, collecting more similar cases and initiating a second model fine-tuning, continuously improving the accuracy and adaptability of the generated report, forming a virtuous cycle of "practice-feedback-optimization."

[0091] Please see Figure 3 In another embodiment, a storage medium 310 stores a computer program that is executed by a processor 320 to perform the steps of the AI-based physical examination report generation method based on a large language model as described above.

[0092] When conducting physical examinations, the system acquires the examinees' data from the examination equipment or system. This data includes both unstructured and structured data. The collected data undergoes preprocessing, converting unstructured data into structured data. Then, historical examination data is used as training samples to train and refine the AI ​​model. The trained AI model then analyzes the examination data using a knowledge graph and AST (Abstract Strategy Framework) logic analysis framework to obtain the analysis results. Finally, a physical examination report is generated based on the data and analysis results using a template. This automatic data collection, followed by automatic analysis and report integration by the AI ​​model, not only improves the efficiency of report generation but also significantly reduces manual operation costs for doctors and accelerates report delivery.

[0093] In some embodiments, addressing the shortcomings of existing physical examination report generation technologies, this invention proposes an AI-based physical examination report technology based on a large-scale language model. This technology integrates a large-scale language model with historical hospital physical examination data to construct a closed-loop management system encompassing "data acquisition - preprocessing - model correction - report generation - knowledge accumulation." This solves the pain points of efficiency, accuracy, and knowledge reuse in traditional physical examination report generation, providing an innovative solution for the intelligent upgrading of physical examination services, and possesses significant clinical practice value and industry application prospects.

[0094] (1) Please refer to Figure 4 Overall process description:

[0095] Multi-source data acquisition: Two types of core data are collected. Structured data includes basic information of examinees (name, age, gender, etc.), indicator values ​​of various examination items (such as uric acid 460umol / L, hemoglobin 129g / L), diagnostic conclusions (such as paroxysmal supraventricular tachycardia, thyroid nodules), reference ranges, etc. Unstructured data includes doctors' handwritten examination notes, imaging report descriptions (such as "bilateral pulmonary glass nodules 5mmx6mm"), clinical analysis from historical physical examination reports, cross-departmental consultation opinions, etc. The two types of data work together to provide comprehensive support for AI analysis.

[0096] Data preprocessing, as the foundation for accurate model analysis, comprises three core steps. First, data cleaning is performed, using regular expressions to remove outliers, garbled characters, and duplicate records (e.g., deduplicating multiple detection results for the same indicator). Second, semantic standardization is conducted, using the BERT word vector model to map similar diagnoses with different expressions (e.g., "mild anemia" and "hemoglobin slightly below reference value") to a unified semantic representation, eliminating ambiguity. Finally, feature extraction is performed, combining medical data analysis tools to automatically identify key features (e.g., abnormal indicator type, deviation of indicator values, number of positive findings, age, gender, and other related factors), providing standardized input for model correction and analysis.

[0097] AI Large-Scale Model Correction and Analysis:

[0098] Base Model Selection: Use a general large-scale language model as the Base Model (such as a general large-scale model base adapted for the medical field) to ensure that the model has basic natural language understanding and logical reasoning capabilities.

[0099] Data correction process: Using historical physical examination reports from the hospital's physical examination department as training samples, the general Base model was fine-tuned. The training samples contained the mapping relationship of "combination of various examination indicators - diagnostic conclusion - chief examiner's recommendation" (e.g., "uric acid 460umol / L + 26-year-old male - elevated uric acid - low-purine diet + regular follow-up examination"). By labeling correct cases and optimizing the loss function, the model was adapted to the physical examination report generation scenario, improving the accuracy of diagnosis and recommendations.

[0100] Knowledge Graph Construction: Construct a ternary knowledge graph of "abnormal indicators - potential causes - chief physician recommendations" (e.g., "elevated cytokeratin 19 fragment → risk of lung lesions → further MRI examination" and "minor tricuspid regurgitation → physiological phenomenon → no special treatment required") to enhance the model's ability to perform correlation analysis on multiple positive findings.

[0101] Analysis dimensions include semantic matching (quickly matching similar historical physical examination cases), multi-indicator association reasoning (analyzing causal or synergistic relationships between different positive findings), individual adaptation analysis (adjusting diagnostic weights based on age, gender, and past medical history), and historical trend comparison (analyzing indicator change trends if user physical examination data from previous years is available).

[0102] Report generation:

[0103] Dynamic integration of conclusions: Automatically classify and summarize the main diagnoses and positive findings, as well as other diagnoses and positive findings, to ensure a clear structure (as in the "Main Diagnosis" and "Other Diagnosis" classification methods in the physical examination report).

[0104] Precise recommendations are generated based on the analysis results after model correction, combined with knowledge graph and individual project characteristics. Targeted main examination recommendations are generated (e.g., for a 26-year-old unmarried man with elevated uric acid, the recommendation is to "avoid high-purine diets, exercise regularly, and have a follow-up uric acid test in 3 months"; for bilateral pulmonary glass nodules, the recommendation is to "have a follow-up chest MRI in 6 months and closely monitor changes in nodule size").

[0105] Interactive features include: report preview and AI regeneration. If doctors disagree with the generated results, they can trigger model reanalysis or manually adjust the results and provide feedback to the model for optimization.

[0106] Knowledge base closed-loop update: Once the doctor confirms that the AI-generated physical examination report is correct, the system automatically structures the case (including various examination data, diagnostic conclusions, and final recommendations) and stores it in the physical examination knowledge base, achieving real-time knowledge accumulation. Simultaneously, the system dynamically optimizes the model using reinforcement learning technology. If the diagnostic accuracy for a certain type of physical examination scenario (such as "elevated uric acid + low hemoglobin in young men") is below 85%, the system automatically triggers a sample expansion process, collecting more similar cases and initiating a second model fine-tuning, continuously improving the accuracy and adaptability of report generation, forming a virtuous cycle of "practice-feedback-optimization".

[0107] (2) Please refer to Figure 5 Module descriptions:

[0108] Data Acquisition Module: Seamlessly integrates with over 15 mainstream medical examination devices and systems, including hospital LIS, PACS, and ECG machines, via API interfaces to achieve unified collection of structured and unstructured data. Employing an event-driven mechanism, it monitors the completion status of each examination item in real time, ensuring timely and complete data collection and providing comprehensive and up-to-date data support for subsequent preprocessing and model analysis.

[0109] Preprocessing module: For unstructured medical text (such as image descriptions and doctor's notes), it removes interfering information through regular expressions and a medical-specific rule engine, and automatically extracts key entities (such as indicator names, values, lesion locations, and nodule sizes) using named entity recognition (NER) technology, transforming unstructured data into structured features; for structured data, it performs numerical verification and range standardization to ensure data format uniformity and provide high-quality input for model correction.

[0110] AI Model Correction and Analysis Engine: Integrates a model fine-tuning sub-engine and a correlation analysis sub-engine. The former is responsible for optimizing the parameters of the general base model based on historical physical examination data, and enabling the model to master the physical examination diagnosis logic through labeled sample training; the latter, based on a knowledge graph and AST logic analysis framework, accurately identifies the correlation between multiple indicators and supports dynamic adjustment of diagnostic weights. It also supports multi-round interactive analysis. If the doctor adds information such as "the user has a family history of gout," the model can update its suggestions in real time (e.g., shortening the uric acid retest cycle to 1 month).

[0111] The report generation module is based on a "standardization + personalization" approach. It includes built-in report templates that conform to the standards of the physical examination department (including fixed modules such as basic information, main diagnosis, other diagnoses, and chief examiner's recommendations), while also supporting hospitals to customize template structures (such as adding specialty-specific recommendation modules). It automatically integrates AI analysis results to generate standardized and logically clear physical examination reports, and provides preview and regeneration access points to adapt to clinical usage scenarios.

[0112] Knowledge base management module: Adopting the RAG (Retrieval Enhanced Generation) architecture, it ensures that the latest physical examination cases and clinical guidelines are prioritized when the model generates suggestions, thus guaranteeing the timeliness of knowledge; It provides a visual knowledge graph editing interface, allowing physical examination experts to optimize the relationship between "abnormal indicators - causes - suggestions" through drag-and-drop and annotation operations, realizing the deep integration of AI intelligent analysis and expert experience, and continuously improving the accuracy of the knowledge base.

[0113] It has the following advantages:

[0114] Advantage 1: Significantly improved report generation efficiency. The platform automatically collects data from multiple systems, automatically analyzes and integrates reports using AI models, reducing the time to generate a single physical examination report from more than 1 hour to less than 10 minutes, greatly reducing the manual operation costs for doctors and speeding up report delivery.

[0115] Advantage 2: The accuracy of diagnosis and recommendations is greatly improved. Through the correction of medical scenario data and the construction of knowledge graphs based on the general Base model, the model can accurately identify the correlation between multiple positive findings, generate personalized recommendations based on individual characteristics, avoid the generality defects of standardized templates, reduce the workload of doctors in making secondary modifications, and improve the clinical reference value of physical examination reports.

[0116] Advantage 3: Optimized knowledge reuse efficiency. The platform stores historical physical examination reports and expert diagnostic experience in a structured knowledge base. Through the RAG architecture, it achieves efficient knowledge reuse. Newly hired doctors can quickly grasp the diagnostic logic of typical cases with the help of AI-generated standardized reports. Seamless knowledge transfer is achieved during cross-departmental collaboration, reducing diagnostic inconsistencies.

[0117] Advantage 4: Effective reduction in medical resource costs. The platform automates the entire process of data collection, preprocessing, and report generation, reducing the cost of manual summarization and analysis. The model continuously improves its autonomous decision-making ability through closed-loop optimization, reducing reliance on senior doctors and shortening the report iteration cycle, thus achieving dual optimization of efficiency and cost control in physical examination services.

[0118] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A method for generating AI-powered health checkup reports based on a large-scale language model, characterized in that, include: Data collection: Collecting medical examination data from medical examination equipment or medical examination system, including structured data and unstructured data; Data preprocessing: The collected physical examination data is preprocessed to transform unstructured data into structured data; AI model correction: Using historical physical examination data as training samples to correct and train the AI ​​model; AI Model Analysis: The preprocessed physical examination data is analyzed using an AI model based on a knowledge graph and AST logical analysis framework to obtain the analysis results; Report generation: Generate a physical examination report based on the analysis results obtained from the AI ​​model analysis using the report template.

2. The AI-based physical examination report generation method based on a large language model according to claim 1, characterized in that, The data collection specifically includes; It monitors the completion status of medical examination items in real time using medical examination equipment or systems, and collects medical examination data from the medical examination equipment or systems when the medical examination items are completed.

3. The AI-based physical examination report generation method based on a large language model according to claim 1, characterized in that, The data preprocessing specifically includes the following steps: Use regular expressions and rule engines to remove distracting data from unstructured data; Named entity recognition technology is used to automatically extract key entities and transform unstructured data into structured data.

4. The AI-based physical examination report generation method based on a large language model according to claim 1, characterized in that, The data preprocessing also includes the following steps: Numerical verification and range standardization are performed on structured data.

5. The AI-based physical examination report generation method based on a large language model according to claim 1, characterized in that, The AI ​​model correction specifically includes the following steps: Historical physical examination data is used as training samples, which contain the mapping relationship between various examination indicator combinations, analysis conclusions, and chief physician recommendations. The AI ​​model is trained by labeling correct cases and optimizing the loss function.

6. The AI-based physical examination report generation method based on a large language model according to claim 1, characterized in that, The AI ​​model analysis specifically includes: The AI ​​model constructs a triplet knowledge graph from the physical examination data. The triplet knowledge graph includes abnormal indicators, potential causes, and main examination recommendations.

7. The AI-based physical examination report generation method based on a large language model according to claim 6, characterized in that, The AI ​​model analysis specifically includes: A triplet knowledge graph is constructed from physical examination data through semantic matching, multi-indicator association reasoning, individual matching analysis, and historical trend comparison.

8. The AI-based physical examination report generation method based on a large language model according to claim 1, characterized in that, Also includes: Interaction optimization: Provide a preview of the physical examination report. When a doctor requests an objection to the contents of the physical examination, the AI ​​model analysis is retried, or feedback is given to the AI ​​model for optimization after receiving the doctor's adjustment.

9. The AI-based physical examination report generation method based on a large language model according to claim 1, characterized in that, Also includes: Knowledge base closed-loop update: Once the generated physical examination report is confirmed to be correct, the report is structured and stored in the physical examination knowledge base.

10. A storage medium storing a computer program, characterized in that, The computer program is executed by the processor to perform the steps of the AI-based physical examination report generation method based on a large language model as described in any one of claims 1-9.

Images