Method for improving site error correction capability of textual reports using anatomic navigation maps
By using an anatomical navigation library and multi-source information fusion, the problem of LLM's inability to detect site description errors in imaging reports was solved, achieving efficient and accurate error correction of imaging reports and improving report quality and readability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WEST CHINA HOSPITAL SICHUAN UNIV
- Filing Date
- 2026-04-17
- Publication Date
- 2026-07-14
Smart Images
Figure CN122392778A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of medical informatics and artificial intelligence-assisted diagnosis, specifically to a method for improving the site-specific error correction capability of text reports using anatomical navigation maps. Background Technology
[0002] Since the second half of 2024, using Large Language Models (LLMs) for text report quality control has become a hot technology in image quality control. However, LLM-based text quality control has a significant drawback: up to two-thirds of the errors in reports cannot be detected through simple text analysis. These errors include differences in the ability to analyze image features and whether clinical data (such as the purpose of the application and medical history) were referenced. Most of these problems require gradually improving doctors' diagnostic skills or using prospective quality control tools like structured reports. However, one common and easily correctable type of error is the misdescription of the lesion location. For example, the left and right sides of a patient's lesion might be reversed, or the word "femur" might be misspelled as "humerus." These errors are difficult to detect using LLM analysis alone. Summary of the Invention
[0003] This invention provides a method for improving the ability to correct site errors in text reports using anatomical navigation maps. By fusing and comparing multiple sources, including the doctor's intuitive spatial judgment, artificial intelligence's semantic understanding of the report text, and historical clinical information, the method can efficiently and accurately detect and alert to potential anatomical site description errors in imaging reports, thereby significantly improving report quality.
[0004] This invention is achieved through the following technical solution:
[0005] A method for improving the site error correction capability of text reports using anatomical navigation maps includes:
[0006] Design an anatomical navigation library, which contains schematic diagrams of anatomical structures for different medical examination scenarios;
[0007] Based on the patient's examination information, the appropriate target anatomical navigation map is automatically matched and retrieved from the anatomical navigation map library;
[0008] The system acquires and processes anatomical data from at least two independent information sources, and presents the processing results in the form of visual annotations at the corresponding anatomical locations on the target anatomical navigation map. The two independent information sources include a first information source and a second information source. The first information source is acquired based on the doctor's interpretation of the examination images. The second information source is acquired based on artificial intelligence analysis of the draft text of the image report generated from the current examination.
[0009] For visual annotations generated by at least two independent information sources and presented on the target anatomical navigation map, perform an anatomical location consistency comparison; when the consistency comparison result is inconsistent, output the corresponding prompt information.
[0010] As an optimization, the patient examination information includes at least one of the following: examination equipment type, requested examination site, and clinical application description.
[0011] As an optimization, the first information source is specifically: the first visual annotation generated by the doctor directly selecting or circling the anatomical location on the target anatomical navigation map through interactive operation after browsing the examination images.
[0012] As an optimization, the second information source is specifically:
[0013] The image report text draft is analyzed in real time using a large language model (LLM) to automatically identify the names of abnormal anatomical sites described in the image report text draft, and map the names of abnormal anatomical sites to the corresponding anatomical locations in the target anatomical navigation map, thereby generating the obtained second visual annotation.
[0014] As an optimization, the independent information source also includes a third information source; the third information source is obtained based on the text analysis of the patient's historical examination reports and / or current examination request forms by an artificial intelligence model, and is used to generate third visual annotations for pre-prompt on the target anatomical navigation map, wherein the presentation of the third visual annotations is earlier than the presentation of the first visual annotations based on the doctor's interpretation of the images.
[0015] As an optimization, the anatomical structure diagrams in the anatomical navigation library are classified and organized according to at least one of the following dimensions: anatomical system dimension, image view dimension, disease specialty dimension, and examination equipment type dimension.
[0016] As an optimization, the specific logic for performing anatomical region consistency comparison is as follows:
[0017] Determine whether the anatomical region indicated by the first visual annotation is covered by the anatomical region indicated by the second visual annotation; if it is not covered, it is determined to be inconsistent.
[0018] As an optimization, the corresponding prompt message is output in at least one of the following ways:
[0019] Inconsistent visual annotations on the target anatomical navigation map are highlighted, flashed, or have their borders enhanced.
[0020] A dialog box containing details of the discrepancies will pop up in the report editing interface;
[0021] Generate and record system audit logs containing descriptions of the differences.
[0022] As an optimization, the method also includes: after the doctor completes the report writing and confirms all prompts, the target anatomical navigation map with final visual annotations is integrated with the text report content and output as a complete examination report document.
[0023] As an optimization, the method also includes:
[0024] Based on the doctor's actions in the examination image browsing interface, the display content of the target anatomical navigation map is dynamically adjusted so that the anatomical location indicated by the navigation map corresponds anatomically to the section or part of the currently viewed examination image.
[0025] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0026] It fundamentally solves the pain point of insufficient ability to identify errors in the description of parts such as "left-right errors" and "incorrect names of anatomical structures" when relying solely on LLM for text quality control.
[0027] This invention integrates doctors' image cognition, semantic information of report text, and historical clinical data to form a three-dimensional, complementary verification network with high fault tolerance and reliability.
[0028] Using the method of this invention, doctors can complete the annotation simply by clicking on an intuitive diagram, which hardly increases the burden of report writing.
[0029] The method of this invention enables real-time comparison and alerts during the report generation process, eliminating errors in their infancy, which is far more efficient than post-event sampling.
[0030] The method of this invention makes the location of lesions clear at a glance in the final output graphic report, improving the readability of the report and the efficiency of clinical communication.
[0031] Dynamic matching of anatomical navigation maps and intelligent analysis of LLM enable the system to adapt to various examination scenarios, from routine plain films to complex MRI. Attached Figure Description
[0032] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:
[0033] Figure 1 This is a flowchart illustrating a method for improving the site error correction capability of text reports using anatomical navigation maps, as described in this invention.
[0034] Figure 2 A schematic diagram of a cranial psychiatric imaging navigation system;
[0035] Figure 3 A schematic diagram of a navigation system for cerebrovascular cognitive impairment;
[0036] Figure 4 This is a schematic diagram illustrating the positional relationship between the navigation map and the MRI image when browsing MRI images. Detailed Implementation
[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are only for explaining this invention and are not intended to limit this invention.
[0038] Before providing a detailed description of this invention, it should be noted that the anatomical navigation map involved in this invention is based on existing technologies, such as the "System and Method for Writing Structured Reports Based on Graphical Graphics" disclosed in patent document CN106874682A. This document teaches how to use structural diagrams of body parts (such as coronary arteries and prostate) as graphical templates and link them with a structured report list for editing, thereby visualizing the report content.
[0039] This invention represents a significant innovation and functional expansion based on the aforementioned graphical interaction concept. Its core objective has shifted from "aiding writing and visualization" to "real-time intelligent quality control and error correction." To achieve this objective, this invention primarily incorporates the following fundamental improvements:
[0040] 1. A multidimensional anatomical navigation map library with intelligent matching capability was constructed;
[0041] 2. A multi-source information annotation and automated comparison mechanism has been introduced;
[0042] 3. Real-time intervention and prompts based on comparison have been implemented.
[0043] Therefore, this invention does not simply apply the graphical template, but fundamentally expands its functionality and transforms its purpose, specifically in the following ways:
[0044] 1. Different purposes: The core purpose of patent CN106874682A is to assist in report writing and visualization, and improve the intuitiveness of the report. The core purpose of this invention, however, is to perform real-time quality control and error correction, using multi-source information comparison to detect and prevent report errors.
[0045] 2. Functional Evolution:
[0046] From static templates to dynamic intelligent matching: This invention establishes a multi-dimensional anatomical navigation map library, which can automatically match and call the most suitable navigation map based on examination information (equipment, location, medical history), instead of using fixed or manually selected schematic diagrams.
[0047] From one-way editing to multi-source annotation and intelligent comparison: This invention introduces at least two independent information sources (doctor interaction and LLM text analysis) to be annotated on the navigation map, and innovatively designs an automated consistency comparison logic, which is a core quality control step not covered by the CN106874682A patent.
[0048] From results presentation to process intervention: This invention performs real-time comparison during report generation and proactively outputs prompts (such as flashing or pop-ups) when inconsistencies are found, thus achieving process quality control rather than simply adding static figures after report generation.
[0049] 3. Technological Integration: This invention creatively combines the semantic analysis capabilities of Large Language Model (LLM) with the spatial representation capabilities of anatomical navigation maps, realizing the automatic mapping from natural language report text to anatomical spatial locations. This is an intelligent upgrade of the original graphical reporting technology.
[0050] Next, the method of the present invention will be described through Example 1. Example 1 provides a method for improving the site error correction capability of text reports using anatomical navigation maps, such as... Figure 1 As shown, it includes:
[0051] S1. Design an anatomical navigation library, which contains schematic diagrams of anatomical structures for different medical examination scenarios.
[0052] The foundation of this invention is a pre-built, comprehensive anatomical navigation library. This library is not a single collection of schematic diagrams, but rather a structured organization based on a multi-dimensional knowledge system.
[0053] Dimension 1: Anatomical System Dimension: The image library contains illustrations of major categories such as the central nervous system (brain, spinal cord), thoracic system (lungs, mediastinum), abdominal and pelvic system (liver, gallbladder, pancreas, spleen, kidneys, pelvic organs), and musculoskeletal system (whole skeleton, joints).
[0054] Dimension Two: Image View Dimension: For 3D images (such as CT, MR), navigation maps are provided in sagittal, coronal, and transverse (axial) planes; for 2D images (such as X-ray, mammography), navigation maps in standard projection orientations such as anteroposterior and lateral views are provided. Figure 2 , 3 As shown.
[0055] Dimension Three: Disease Specialty Dimension: Dedicated navigation maps are designed to address specific diagnostic concerns for particular diseases. For example:
[0056] Cranial psychiatric imaging navigation map: It highlights the divisions of the cerebral cortex (such as the subdivided gyri of the frontal lobe, parietal lobe, temporal lobe, and occipital lobe) and is used for the assessment of mental illnesses.
[0057] Brain cognitive impairment navigation map: It highlights brain regions prone to atrophy (such as the hippocampus, entorhinal cortex, frontal lobe, parietal lobe, etc.) and is used for the assessment of Alzheimer's disease, etc.
[0058] Hip fracture navigation map: Detailed display of structures such as the ilium, ischium, pubis, acetabulum, femoral head, femoral neck, and intertrochanteric region.
[0059] Dimension 4: Equipment Type Inspection: Differentiate the image characteristics and schematic diagram styles of equipment such as CT, MR, X-ray, ultrasound, and PET-CT.
[0060] S2. Based on the patient's examination information, automatically match and retrieve the applicable target anatomical navigation map from the anatomical navigation map library.
[0061] When a doctor opens a test report, the system automatically matches and retrieves the most suitable target anatomical navigation map based on the patient's test information. The matching criteria include:
[0062] 1. Check the equipment type (e.g., MR).
[0063] 2. The area to be examined (e.g., the skull).
[0064] 3. Clinical Application Description: The application text is analyzed using LLM to further refine the selection. For example, if the application description is for assessing brain atrophy, a brain cognitive impairment navigation map is automatically selected; if it is described as a mental and behavioral disorder, a brain mental imaging navigation map is selected. For breast examinations, a dedicated breast cancer navigation map is directly matched.
[0065] The above-mentioned automatic matching is a dynamic and granular process, the core of which lies in intelligently controlling the granularity (i.e., the level of precision) of the anatomical description according to the diagnostic scenario:
[0066] 1. Initial matching: The system first matches the navigation map of the major category (such as the head MR navigation map) based on the three basic information: the type of inspection equipment, the application site, and the inspection site.
[0067] 2. Intelligent Updates Based on Request Forms: The system further analyzes the descriptions of lesion types in clinical request forms using LLM (Limited Learning Model) and updates or refines the navigation map selection accordingly to match more professional diagnostic views.
[0068] Example 1: Cerebral hemorrhage: If the application form indicates cerebral hemorrhage, the system may automatically switch to or highlight the navigation map of the relevant structures in the nine-point classification of cerebral hemorrhage location. Its granularity focuses on the nine types: epidural, subdural, subarachnoid, brain parenchyma, ventricle, nasal cavity / sinus, tympanic cavity / mastoid, subcutaneous, and surgical area, without needing to be refined to the lobes or sulci of the brain parenchyma.
[0069] Example 2: Hip fracture: If it is a hip X-ray examination, the system will bring up a navigation map with fine focus on the key weight-bearing and fracture-prone sites of the hip joint, including structures such as the ilium, pubis, ischium, acetabulum, femoral head, femoral neck, and proximal femur.
[0070] 3. Multi-view support for navigation maps: For 3D images such as CT and MR, the system can simultaneously or on demand provide accompanying navigation maps in sagittal, coronal, and cross-sectional (axial) planes to meet the needs of multi-directional location description. Doctors can switch between these views while the system keeps the annotations synchronized.
[0071] The table below illustrates how equipment type, inspection location, and application description collectively determine the final call navigation map:
[0072] Navigation map type Equipment type Application / Inspection Site Application description analysis trigger conditions Breast navigation map MG (mammography) mammary gland (Direct match, no special analysis required) Cranial Psychiatric Imaging Navigation Map MR Head CT scan (plain / enhanced) The application description contains keywords such as "mental," "emotional," and "abnormal behavior." Navigation map of cranial cognitive impairment MR Head CT scan (plain / enhanced) The application description included keywords such as "atrophy," "dementia," and "memory decline." ... ... ... ...
[0073] S3. Acquire and process anatomical site data from at least two independent information sources, and present the processing results in the form of visual annotations at the corresponding anatomical locations on the target anatomical navigation map; the two independent information sources include a first information source and a second information source, the first information source being acquired based on the doctor's interpretation of the examination images; the second information source being acquired based on artificial intelligence analysis of the draft image report text generated from the current examination.
[0074] After the target anatomical navigation map is presented, the system obtains anatomical site data from at least two independent information sources and transforms it into intuitive visual annotations displayed on the navigation map. Annotations from different sources are distinguished using different colors or graphic styles; for example, blue is used for physician annotations, yellow for LLM analysis reports, and gray for historical information pre-annotations.
[0075] Primary information source: Doctors' interactive annotations.
[0076] Doctors browse DICOM images on a PACS workstation to determine the location of lesions. Subsequently, on an anatomical navigation map on one side of the screen, doctors can directly identify suspicious or confirmed anatomical sites of lesions by clicking with a mouse, tapping on the touchscreen, or circling. This interaction is captured by the system and generates a primary visual annotation (such as a flashing blue dot or area) at the corresponding location. This process rapidly digitizes the doctor's spatial perception.
[0077] It should be noted that doctors have complete autonomy in deciding whether to add annotations. Once annotations are completed, these annotations have intelligent view correlation: even if a doctor manually switches to different navigation map templates midway through the diagnosis (e.g., from a cross-sectional view to a sagittal view), as long as the anatomical site remains the same, the system will automatically map and display the existing first visual annotations in the corresponding positions of the new navigation map view, thus ensuring the continuity of diagnostic thinking.
[0078] Second source of information: LLM analysis annotations in the draft report text.
[0079] Simultaneously or subsequently, while a physician is writing or dictating a draft of the imaging report (e.g., "patchy abnormal signals visible in the left temporal lobe"), a large language model (LLM) in the background performs real-time semantic analysis on this text. The LLM automatically identifies the name of the abnormal anatomical location described in the text (e.g., the left temporal lobe), and then, using a medical knowledge graph, maps this name to its precise location on the target anatomical navigation map, generating a secondary visual annotation (e.g., a yellow highlighted area). This represents the location described in the report text.
[0080] Third information source: Pre-notification annotations of historical and application form information.
[0081] Before the doctor begins writing the current report or even reviewing the current examination images, the system uses LLM to pre-analyze the text content of the patient's historical examination reports and current examination requests, aiming to provide diagnostic background information and warn of potential errors. The third-party visual annotations generated based on this analysis (usually displayed in neutral, suggestive colors such as gray) follow a clear set of intelligent rules:
[0082] Rule 1 (Filtering Normal Descriptions): When analyzing historical reports or application forms, LLM will not generate any pre-labels for anatomical sites described as normal (e.g., no abnormalities found). This avoids the navigation map being interfered with by irrelevant information.
[0083] Rule 2 (Mapping Valid Anomalies): The system will only generate a third visual annotation at the corresponding anatomical location in the map if the LLM identifies an anomaly in the text description of an anatomical location (e.g., nodule, mass, fracture, postoperative absence, etc.) and the location exists in the currently retrieved target anatomical navigation map.
[0084] Example of trigger annotation: If a historical report mentions a ground-glass nodule in the upper lobe of the right lung, the system will generate a gray annotation in the upper lobe region of the right lung on the lung navigation map.
[0085] Example of no annotation: If the application form describes the spleen's shape and size as normal, the system will not annotate the spleen's location.
[0086] The core value of this information source lies in providing crucial background information before doctors perform diagnostic procedures to prevent two common problems:
[0087] Guide the doctor to focus on key areas: Visually show the doctor the location of lesions already recorded in the historical reports, guiding them to focus on reviewing and comparing that area during the current examination.
[0088] Early warning of missing anatomical structures: Crucially, this feature is used to detect and alert patients to removed organs or tissues in advance. For example, when a request form or historical report clearly states "the patient has undergone a right nephrectomy" or "the gallbladder has been removed," LLM identifies the structure as "absent." The system generates a special "absent" marker (e.g., a gray dashed outline, a specific icon, or a label with the word "absent") at the corresponding anatomical location on the navigation map (e.g., the right kidney area, gallbladder area). This core function aims to prevent physicians from incorrectly describing non-existent anatomical structures during subsequent imaging observations and report writing.
[0089] It is important to note that this third visual annotation primarily serves as background information and a safety warning before diagnosis. It is presented earlier than the first visual annotation generated by the doctor based on the current image. In the core real-time verification process of this invention, it is typically not used as a mandatory basis for "consistency comparison" (i.e., it is not directly compared with the doctor's first visual annotation). Its main function is to provide doctors with valuable preliminary references for diagnosis, thereby improving the comprehensiveness of diagnostic preparation and the security of the report.
[0090] After the system completes multi-source annotation, it enters the automated comparison phase. It's important to clarify that the core consistency comparison primarily involves the first visual annotation (doctor's judgment) and the second visual annotation (report text description). The comparison logic is as follows:
[0091] 1. Determine if the doctor has made annotations: The system first checks whether there are any first visual annotations actively generated by the doctor.
[0092] If not present, the process is simplified. The system only displays the second visual annotations as automatic analysis results on the navigation map for doctors to refer to when reviewing reports; it does not perform comparisons or trigger any mandatory reminders.
[0093] If it exists, proceed to the next comparison step.
[0094] 2. Determine the relationship between annotations: The system compares the set of anatomical parts indicated by the first visual annotation with the set of anatomical parts indicated by the second visual annotation.
[0095] If the area labeled by the first visual marker is completely covered by the area labeled by the second visual marker (i.e., the latter's description range is equal to or greater than the former's), then it is considered consistent. The system will not trigger a warning.
[0096] If the area marked by the first visual label is not covered by the area marked by the second visual label, it is considered inconsistent. The system will immediately trigger the defined prompt message (such as flashing the label in red on the navigation map, highlighting it with a bold red border, or popping up a prompt dialog box).
[0097] S4. Perform an anatomical location consistency comparison on the visual annotations generated by at least two independent information sources and presented on the target anatomical navigation map. When the consistency comparison result is inconsistent, output the corresponding prompt information.
[0098] The system automatically compares the consistency of the anatomical locations represented by the first visual annotation (doctor's judgment) and the second visual annotation (report text description).
[0099] Comparison logic: The system determines whether the set of anatomical parts indicated by the first visual annotation is covered by the set of anatomical parts indicated by the second visual annotation.
[0100] Judgment rules:
[0101] The system executes the following differentiated management process based on whether the doctor has actively annotated the data and the relationship between different annotations:
[0102] Scenario 1: Doctor has not annotated. The system only displays annotations automatically generated by the second information source (draft LLM analysis report) for the doctor's reference. In this scenario, the system does not perform consistency comparisons or trigger any mandatory reminders.
[0103] Scenario 2: The doctor has already labeled the data, and the labeling is consistent with that in the LLM analysis.
[0104] Consistency definition: This refers to the complete overlap between the anatomical locations marked by the physician and those analyzed by the LLM from the report text. For example, if the physician marks the left frontal lobe, while the report text describes an abnormality in the left cerebral hemisphere, the latter's area covers the former, thus it is considered consistent. In this scenario, the system does not output any alerts. The navigation map can display a lightweight marker indicating consistent handwriting.
[0105] Scenario 3: The doctor has labeled the data, but the labeling is inconsistent with that of the LLM analysis.
[0106] Inconsistency definition: This refers to a situation where a doctor has labeled an anatomical location, but the LLM analysis of the complete report text fails to identify that location or its superior location. For example, the doctor labeled the right cerebellar hemisphere, but the entire report text does not mention related terms such as cerebellum, posterior fossa, or cerebellum. In this scenario, the system determines it as inconsistent and immediately triggers the following prompts (such as flashing, pop-up windows, etc.) to force the doctor to review the report:
[0107] Navigation map highlighting: On the anatomical navigation map, inconsistent doctor annotations (first visual annotations) are highlighted with red flashing and red thick borders.
[0108] Pop-up reminder: A non-blocking prompt box will appear in the report editing interface, with content such as: "Please note: You labeled 'right cerebellar hemisphere' on the image, but this area is not mentioned in the report description. Please check."
[0109] Log recording: Record this inconsistency event in the quality control audit log for subsequent statistical analysis.
[0110] Report integration and output:
[0111] After all verifications and modifications are completed, the doctor confirms the report. The system integrates the final version of the target anatomical navigation map, which includes all confirmed visual annotations (which can be uniformly colored), as an appendix with the standard text report content, generating a final examination report document (such as PDF format) that is rich in both text and illustrations. This report is not only accurate in its text, but also visually illustrates the location of the lesion through diagrams, greatly facilitating understanding for both clinicians and patients.
[0112] To enhance user experience and efficiency, this invention may also include advanced interactive features:
[0113] View synchronization: As doctors scroll through different image slices on the PACS interface (e.g., from the top of the skull to the base), the focus or highlighted area on the target anatomical navigation map dynamically adjusts accordingly, indicating in real time the position of the anatomical plane corresponding to the current slice within the overall anatomical structure. For example, when browsing an MRI image at the lateral ventricle level, the area corresponding to the lateral ventricle on the navigation map will be highlighted. This achieves real-time anatomical correspondence between the image and the diagram, helping doctors quickly locate the image, such as... Figure 4 As shown.
[0114] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for improving the site-specific error correction capability of text reports using anatomical navigation maps, characterized in that, include: Design an anatomical navigation library, which contains schematic diagrams of anatomical structures for different medical examination scenarios; Based on the patient's examination information, the appropriate target anatomical navigation map is automatically matched and retrieved from the anatomical navigation map library; The system acquires and processes anatomical data from at least two independent information sources, and presents the processing results in the form of visual annotations at the corresponding anatomical locations on the target anatomical navigation map. The two independent information sources include a first information source and a second information source. The first information source is acquired based on the doctor's interpretation of the examination images. The second information source is acquired based on artificial intelligence analysis of the draft text of the image report generated from the current examination. For visual annotations generated by at least two independent information sources and presented on the target anatomical navigation map, perform an anatomical location consistency comparison; when the consistency comparison result is inconsistent, output the corresponding prompt information.
2. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The patient examination information includes at least one of the following: type of examination equipment, site of examination requested, and clinical application description.
3. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The first information source is specifically: the corresponding first visual annotation generated by the doctor directly selecting or circling the anatomical location on the target anatomical navigation map through interactive operation after browsing the examination images.
4. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The second information source is specifically: The image report text draft is analyzed in real time using a large language model (LLM) to automatically identify the names of abnormal anatomical sites described in the image report text draft, and map the names of abnormal anatomical sites to the corresponding anatomical locations in the target anatomical navigation map, thereby generating the obtained second visual annotation.
5. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The independent information source also includes a third information source; the third information source is obtained based on the text analysis of the patient's historical examination reports and / or current examination request forms by an artificial intelligence model, and is used to generate third visual annotations for pre-prompt on the target anatomical navigation map; wherein, the presentation of the third visual annotations is earlier than the presentation of the first visual annotations based on the doctor's interpretation of the images.
6. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The anatomical navigation map library is classified and organized according to at least one of the following dimensions: anatomical system dimension, image view dimension, disease specialty dimension, and examination equipment type dimension.
7. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 3 or 4, characterized in that, The specific logic for performing anatomical site consistency comparison is as follows: Determine whether the anatomical region indicated by the first visual annotation is covered by the anatomical region indicated by the second visual annotation; if it is not covered, it is determined to be inconsistent.
8. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The corresponding prompt message can be output in at least one of the following ways: Inconsistent visual annotations on the target anatomical navigation map are highlighted, flashed, or have their borders enhanced. A dialog box containing details of the discrepancies will pop up in the report editing interface; Generate and record system audit logs containing descriptions of the differences.
9. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The method also includes: after the doctor completes the report writing and confirms all prompts, integrating the target anatomical navigation map with final visual annotations with the text report content, and outputting a complete examination report document.
10. The method for improving the site error correction capability of text reports using anatomical navigation maps according to claim 1, characterized in that, The method also includes: Based on the doctor's actions in the examination image browsing interface, the display content of the target anatomical navigation map is dynamically adjusted so that the anatomical location indicated by the navigation map corresponds anatomically to the section or part of the currently viewed examination image.