Method for automatically scanning and interpreting optometry report

Abstract

The invention relates to the technical field of optometry report interpretation, in particular to a method for automatically scanning and interpreting an optometry report, which comprises the following steps: denoising and enhancing an input optometry report image so as to improve the definition of subsequent character recognition, converting the image into a black and white binary image, and highlighting a character area so as to accurately extract text information. According to the method, character information in a report is accurately recognized in combination with OCR of deep learning, data are cleaned through term mapping and regular expressions, and format standardization is achieved; according to semantic analysis, key parameters are extracted based on a medical knowledge graph, abnormal data are automatically marked through logic verification, a template is filled with structured data in report generation, an intuitive visual report is generated, the report is pushed to a patient in real time through a webpage or an applet, the processing efficiency is improved by multiple times, the error rate is greatly reduced, and the user experience is improved. The patient can check the interpretation result at any time and obtain personalized suggestions, and the diagnosis and treatment efficiency and the patient participation degree are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of optometry report interpretation, and in particular to a method for automatically scanning and interpreting optometry reports. Background Art

[0002] An optometrists' report is a professional document that records a patient's vision status after an eye examination. It typically includes key parameters such as uncorrected visual acuity, corrected visual acuity, spherical power, cylindrical power, and axial length, as well as auxiliary data such as axial length. Its purpose is to help doctors accurately assess a patient's refractive status and develop personalized correction plans. Optometry reports come in a variety of formats, including paper forms, electronic records, or scanned images, and are a crucial basis for ophthalmic diagnosis, treatment, and vision health management.

[0003] However, in general, the traditional method of scanning and interpreting optometry reports relies mainly on manual operations: doctors or technicians need to manually enter the contents of the paper report into the electronic system one by one, or use basic scanning equipment to generate electronic images, and then visually verify the data and organize and archive it. This method has significant disadvantages: First, manual data entry is time-consuming and labor-intensive, and data errors can easily occur due to fatigue or negligence; second, paper reports are easily affected by environmental factors, and the quality of scanned images varies, further reducing the OCR recognition rate; third, patients find it difficult to understand the complex professional terms in the report and need to rely on doctors to explain repeatedly, which makes communication inefficient. In addition, paper reports are not easy to preserve and share for a long time, which is not conducive to data comparison and analysis during follow-up visits.

[0004] In summary, it is necessary to propose a method for automatically scanning and interpreting optometry reports to solve the above problems. Summary of the Invention

[0005] The object of the present invention is to provide a method for automatically scanning and interpreting an optometry report to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention proposes a method for automatically scanning and interpreting an optometry report. The method is implemented based on a system for automatically scanning and interpreting an optometry report, and includes the following steps:

[0008] S1. De-noise and enhance the input optometry report image to improve the clarity of subsequent text recognition. Convert the image into a black and white binary image, highlighting the text area for accurate text extraction.

[0009] S2. Use deep learning technology to identify text in images, extract patient information and optometry data, clean the text and unify the data format, and eliminate redundant characters and symbol errors;

[0010] S3. Extract key ophthalmic parameters and verify their logical rationality, mark abnormal data, and generate a structured report based on the verified data according to the template, including a summary and abnormality prompts;

[0011] S4. Convert structured data into a visual interface to facilitate user interaction and intuitive understanding. Finally, record the eye examination results in the user's online profile. After removing privacy-sensitive data, the data is used as a large language model to output glasses fitting guidance and eye protection suggestions to provide a dataset.

[0012] Preferably, the implementation steps of step S1 are:

[0013] S1.1. Receive the image of the eye test report uploaded by the user and check whether the image resolution meets the minimum requirements. If the resolution is insufficient, the user is prompted to re-upload the image. Gaussian filtering is performed on the image, and a weighted average of the pixels is calculated using a sliding window to eliminate noise introduced by scanning or photographing while preserving the sharpness of text edges. For images with uneven lighting, histogram equalization is used to adjust the brightness distribution to ensure consistent contrast across the text area. An unsharp masking algorithm is used to enhance text edges and highlight small characters in the eye test report to avoid blurring that could lead to subsequent recognition errors.

[0014] S1.2. Convert the color image to grayscale and use weighted averaging to reduce computational complexity while preserving the brightness difference between the text and the background.

[0015] S1.3. Use the Otsu algorithm to dynamically calculate the optimal threshold to distinguish text from background;

[0016] For example, for reports with light backgrounds, the threshold is automatically increased to avoid text breakage; for reports with dark backgrounds, the threshold is lowered to prevent text sticking;

[0017] S1.4. Perform a closing operation on the binarized image to fill in the small gaps between the strokes of the characters to ensure that the characters are complete.

[0018] S1.5. Detect text region boundaries in images, automatically crop irrelevant background, and correct tilted or distorted text lines through affine transformation.

[0019] Preferably, the implementation steps of step S2 are:

[0020] S2.1. Use projection analysis to segment the binary image into independent text lines to avoid overlapping text. Use a CRNN model to recognize characters line by line. Combined with specialized medical dictionaries (e.g., "OD / OS" and "spherical / cylindrical"), correct misidentification results (e.g., misclassifying "0" as "O").

[0021] S2.2. Categorize the recognition results by fields (e.g., patient name, spherical diopter, axial length) and output them as key-value pairs (e.g., `{"right eye spherical diopter":"-3.25D"}`) for subsequent processing.

[0022] S2.3. Use regular expressions to remove irrelevant symbols (such as """") or garbled characters (such as "&^%") from the OCR results, retaining only numbers, letters, and technical terms.

[0023] S2.4. Based on a pre-defined terminology mapping table (e.g., "OD → right eye," "OS → left eye," "SPH → spherical lens"), convert non-standard expressions to a unified format, eliminate dialect or abbreviation differences, force unit conversion to a unified standard (e.g., "D" to "degree," "mm" to "millimeter"), and format numerical ranges (e.g., "-3.25D" to "myopia 325 degrees").

[0024] S2.5. Verify whether any required fields are missing (such as patient ID, test date). If missing, mark them as "to be completed" and notify the user to re-upload.

[0025] Preferably, the implementation steps of step S3 are:

[0026] S3.1. Using rule engines to extract key parameters from text:

[0027] Naked eye visual acuity: Match "VA:0.8" or "Naked eye visual acuity:1.0" mode;

[0028] Axial length: matches the field "AL:24.3mm" or "Axial length:23.5";

[0029] S3.2. Check the correlation between parameters:

[0030] If "cylindrical power" exists but "axial" is 0°, mark it as "data contradiction";

[0031] If the "corrected visual acuity" is better than the "naked visual acuity" but the correction method is not marked, it will be marked as "manual review required";

[0032] S3.3. Set a range. If the value exceeds the limit, an alarm will be triggered, indicating that there may be a measurement or recognition error.

[0033] S3.4. Select the preset template based on the report type and fill in the corresponding fields with the verified data (e.g., "Corrected Vision" in the "Diagnosis Result" column).

[0034] S3.5. Insert eye-catching labels (such as red icon), with detailed explanation (e.g., “right eye cylindrical power exceeds the common range, reexamination is recommended”);

[0035] S3.6. Supports the generation of reports in JSON, PDF or HTML format to meet the needs of hospital system connection, patient archiving or printing.

[0036] Preferably, the implementation steps of step S4 are:

[0037] S4.1. Convert structured data into line charts (visual acuity trend) and scatter plots (binocular diopter comparison), and render them in real time using SVG or Canvas.

[0038] S4.2. When a user clicks an abnormal data point, a floating window pops up showing the original image fragment, the recognition result, and review suggestions. The user can also jump to a historical report for comparison with one click.

[0039] S4.3. Automatically adjust the layout for different terminals (mobile phones, tablets, computers) to ensure that charts and text are clear and readable in mini-programs and web pages.

[0040] Preferably, the system for automatically scanning and interpreting optometry reports includes an image preprocessing module, a text recognition module, a semantic parsing module, and a report generation module;

[0041] The image preprocessing module is used for image enhancement and image standardization;

[0042] The text recognition module is used for OCR recognition and data cleaning;

[0043] The semantic parsing module is used for key information extraction and data verification;

[0044] The report generation module is used for result structuring and user interface generation.

[0045] Preferably, the image preprocessing module further includes an image enhancement unit and an image standardization unit;

[0046] The image enhancement unit performs a denoising operation on the scanned optometry report image using a Gaussian filtering algorithm to eliminate noise, blur or uneven lighting in the image, thereby improving the clarity of subsequent processing;

[0047] The image standardization unit adopts an adaptive binarization algorithm to perform grayscale conversion and binarization processing on the image, unifies the image format and highlights the text area, ensuring that the image format of the optometry report from different sources is consistent.

[0048] Preferably, the text recognition module further includes an OCR recognition unit and a data cleaning unit;

[0049] The OCR recognition unit performs text recognition on the pre-processed image based on the Tesseract OCR engine combined with a deep learning model, and accurately extracts patient information and optometry data from the optometry report, including visual acuity, corrected visual acuity, spherical power, cylindrical power, axial direction, and axial length from the axial measurement report;

[0050] The data cleaning unit performs redundant character filtering, format error correction and data standardization on the original text output by OCR through regular expression matching and keyword library verification.

[0051] Preferably, the semantic parsing module further includes a key information extraction unit and a data verification unit;

[0052] The key information extraction unit uses rule-based named entity recognition (NER) combined with medical knowledge graphs to extract key parameters such as naked eye vision, corrected visual acuity, diopter, and axial length from the cleaned text;

[0053] The data verification unit verifies the rationality of the extracted parameters through a logic verification algorithm, marks abnormal data and feeds back to the user.

[0054] Preferably, the report generation module further comprises a result structuring unit and a user interface generation unit;

[0055] The result structuring unit uses a template engine to automatically fill in the parsed optometry data according to a preset template, and generates a structured report including data summary and abnormal prompts;

[0056] The user interface generation unit dynamically renders the structured report based on HTML / CSS technology, generates a visual interface, and displays it to the user through a mini program or a web page.

[0057] Compared with the existing technology, the beneficial effects of the present invention are: the method of automatically scanning and interpreting optometry reports based on the present invention uses Gaussian filtering and adaptive binarization technology through image preprocessing to eliminate noise, background interference and uneven lighting problems in the scanned image, ensuring that the text is clear and legible, and combines deep learning OCR to accurately identify text information in the report, and clean data through term mapping and regular expressions to achieve format standardization; semantic analysis extracts key parameters based on the medical knowledge graph, and automatically marks abnormal data through logical verification. Report generation fills structured data into the template to generate an intuitive visual report, and pushes it to the patient in real time through a web page or mini program, which not only increases the processing efficiency several times, but also greatly reduces the error rate. Patients can view the interpretation results and obtain personalized suggestions at any time, significantly improving diagnosis and treatment efficiency and patient participation. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1The method of the present invention for automatically scanning and interpreting an optometry report is shown. DETAILED DESCRIPTION

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0060] In an embodiment, the present invention is implemented based on a system for automatically scanning and interpreting optometry reports. It should be noted that the system for automatically scanning and interpreting optometry reports includes an image preprocessing module, a text recognition module, a semantic parsing module, and a report generation module.

[0061] The image preprocessing module is used for image enhancement and image standardization, the text recognition module is used for OCR recognition and data cleaning, the semantic analysis module is used for key information extraction and data verification, and the report generation module is used for result structuring and user interface generation.

[0062] In this embodiment, it should also be noted that the image preprocessing module further includes an image enhancement unit and an image normalization unit;

[0063] The image enhancement unit uses the Gaussian filtering algorithm to perform denoising on the scanned optometry report image, eliminating noise, blur or uneven lighting in the image and improving the clarity of subsequent processing;

[0064] The image standardization unit uses an adaptive binarization algorithm to perform grayscale conversion and binarization processing on the image, unify the image format and highlight the text area, ensuring that the image format of the optometry report from different sources is consistent.

[0065] In this embodiment, it should also be noted that the text recognition module also includes an OCR recognition unit and a data cleaning unit;

[0066] The OCR recognition unit uses the Tesseract OCR engine combined with a deep learning model to perform text recognition on pre-processed images, accurately extracting patient information and optometry data from optometry reports. The optometry data includes visual acuity, corrected visual acuity, spherical power, cylindrical power, axial length, and axial length from axial measurement reports.

[0067] The data cleaning unit filters redundant characters, corrects format errors, and standardizes data on the original text output by OCR through regular expression matching and keyword library verification.

[0068] In this embodiment, it should also be noted that the semantic parsing module also includes a key information extraction unit and a data verification unit;

[0069] The key information extraction unit uses rule-based named entity recognition (NER) combined with medical knowledge graphs to extract key parameters such as naked eye vision, corrected visual acuity, diopter, and axial length from the cleaned text;

[0070] The data verification unit verifies the rationality of the extracted parameters through a logical verification algorithm, marks abnormal data and feeds back to the user.

[0071] In this embodiment, it should also be noted that the report generation module also includes a result structuring unit and a user interface generation unit;

[0072] The result structured unit uses a template engine to automatically fill in the parsed optometry data according to the preset template and generate a structured report containing data summary and abnormal prompts;

[0073] The user interface generation unit dynamically renders structured reports based on HTML / CSS technology, generates a visual interface, and displays it to users through mini-programs or web pages;

[0074] In practical applications, see Figure 1 A method for automatically scanning and interpreting an optometry report based on the above system specifically includes the following steps:

[0075] S1. De-noise and enhance the input optometry report image to improve the clarity of subsequent text recognition. Convert the image into a black and white binary image, highlighting the text area for accurate text extraction.

[0076] S1.1. Receive the image of the eye test report uploaded by the user and check whether the image resolution meets the minimum requirements. If the resolution is insufficient, the user is prompted to re-upload the image. Gaussian filtering is performed on the image, and a weighted average of the pixels is calculated using a sliding window to eliminate noise introduced by scanning or photographing while preserving the sharpness of text edges. For images with uneven lighting, histogram equalization is used to adjust the brightness distribution to ensure consistent contrast across the text area. An unsharp masking algorithm is used to enhance text edges and highlight small characters in the eye test report to avoid blurring that could lead to subsequent recognition errors.

[0077] S1.2. Convert the color image to grayscale and use weighted averaging to reduce computational complexity while preserving the brightness difference between the text and the background.

[0078] S1.3. Use the Otsu algorithm to dynamically calculate the optimal threshold to distinguish text from background;

[0079] For example, for reports with light backgrounds, the threshold is automatically increased to avoid text breakage; for reports with dark backgrounds, the threshold is lowered to prevent text sticking;

[0080] S1.4. Perform a closing operation on the binarized image to fill in the small gaps between the strokes of the characters to ensure that the characters are complete.

[0081] S1.5. Detect text region boundaries in images, automatically crop irrelevant background, and correct tilted or distorted text lines using affine transformation.

[0082] S2. Use deep learning technology to identify text in images, extract patient information and optometry data, clean the text and unify the data format, and eliminate redundant characters and symbol errors;

[0083] S2.1. Use projection analysis to segment the binary image into independent text lines to avoid overlapping text. Use a CRNN model to recognize characters line by line. Combined with specialized medical dictionaries (e.g., "OD / OS" and "spherical / cylindrical"), correct misidentification results (e.g., misclassifying "0" as "O").

[0084] S2.2. Categorize the recognition results by fields (e.g., patient name, spherical diopter, axial length) and output them as key-value pairs (e.g., `{"right eye spherical diopter":"-3.25D"}`) for subsequent processing.

[0085] S2.3. Use regular expressions to remove irrelevant symbols (such as """") or garbled characters (such as "&^%") from the OCR results, retaining only numbers, letters, and technical terms.

[0086] S2.4. Based on a pre-defined terminology mapping table (e.g., "OD → right eye," "OS → left eye," "SPH → spherical lens"), convert non-standard expressions to a unified format, eliminate dialect or abbreviation differences, force unit conversion to a unified standard (e.g., "D" to "degree," "mm" to "millimeter"), and format numerical ranges (e.g., "-3.25D" to "myopia 325 degrees").

[0087] S2.5. Verify that any required fields are missing (e.g., patient ID, test date). If missing, mark them as "pending" and notify the user to re-upload.

[0088] S3. Extract key ophthalmic parameters and verify their logical rationality, mark abnormal data, and generate a structured report based on the verified data according to the template, including a summary and abnormality prompts;

[0089] S3.1. Using rule engines to extract key parameters from text:

[0090] Naked eye visual acuity: Match "VA:0.8" or "Naked eye visual acuity:1.0" mode;

[0091] Axial length: matches the field "AL:24.3mm" or "Axial length:23.5";

[0092] S3.2. Check the correlation between parameters:

[0093] If "cylindrical power" exists but "axial" is 0°, mark it as "data contradiction";

[0094] If the "corrected visual acuity" is better than the "naked visual acuity" but the correction method is not marked, it will be marked as "manual review required";

[0095] S3.3. Set a range. If the value exceeds the limit, an alarm will be triggered, indicating that there may be a measurement or recognition error.

[0096] S3.4. Select the preset template based on the report type and fill in the corresponding fields with the verified data (e.g., "Corrected Vision" in the "Diagnosis Result" column).

[0097] S3.5. Insert eye-catching labels (such as red icon), with detailed explanation (e.g., “right eye cylindrical power exceeds the common range, reexamination is recommended”);

[0098] S3.6. Supports the generation of reports in JSON, PDF or HTML format to meet the needs of hospital system connection, patient archiving or printing.

[0099] S4. Convert structured data into a visual interface to facilitate user interaction and intuitive understanding. Finally, record the eye examination results in the user's online profile. After removing privacy-sensitive data, the data is used as a large language model to output glasses fitting guidance and eye care recommendations to provide a dataset.

[0100] S4.1. Convert structured data into line charts (visual acuity trend) and scatter plots (binocular diopter comparison), and render them in real time using SVG or Canvas.

[0101] S4.2. When a user clicks an abnormal data point, a floating window pops up showing the original image fragment, the recognition result, and review suggestions. The user can also jump to a historical report for comparison with one click.

[0102] S4.3. Automatically adjust the layout for different terminals (mobile phones, tablets, computers) to ensure that charts and text are clear and readable in mini-programs and web pages.

[0103] In this embodiment, it should also be noted that in this step, DeepSeek-R1 is selected as the language output model, pseudonymization or aggregation is used for desensitization, a method for fine-tuning the training output results is performed, and an output filtering mechanism is established to prevent the generation of unconventional medical advice.

[0104] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatically scanning and interpreting an optometry report, wherein the method is implemented based on a system for automatically scanning and interpreting an optometry report, and is characterized in that: The following steps are involved: S1. De-noise and enhance the input optometry report image to improve the clarity of subsequent text recognition. Convert the image into a black and white binary image, highlighting the text area for accurate text extraction. S2. Use deep learning technology to identify text in images, extract patient information and optometry data, clean the text and unify the data format, and eliminate redundant characters and symbol errors; S3. Extract key ophthalmic parameters and verify their logical rationality, mark abnormal data, and generate a structured report based on the verified data according to the template, including a summary and abnormality prompts; S4. Convert structured data into a visual interface to facilitate user interaction and intuitive understanding. Finally, record the eye examination results in the user's online profile. After removing privacy-sensitive data, the data is used as a large language model to output glasses fitting guidance and eye protection suggestions to provide a dataset.

2. The method for automatically scanning and interpreting an optometry report according to claim 1, characterized in that: The implementation steps of step S1 are: S1.

1. Receive an image of the eye test report uploaded by the user and check whether the image resolution meets the minimum requirements. If the resolution is insufficient, the user is prompted to re-upload the image. Gaussian filtering is performed on the image, and a weighted average of the pixels is calculated using a sliding window to eliminate noise introduced by scanning or photographing while preserving the sharpness of text edges. For images with uneven lighting, histogram equalization is used to adjust the brightness distribution to ensure consistent contrast across the text area. An unsharp masking algorithm is used to enhance text edges and highlight small characters in the eye test report to avoid blurring that could lead to subsequent recognition errors. S1.

2. Convert the color image to grayscale and use weighted averaging to reduce computational complexity while preserving the brightness difference between the text and the background. S1.

3. Use the Otsu algorithm to dynamically calculate the optimal threshold to distinguish text from background; For example, for reports with light backgrounds, the threshold is automatically increased to avoid text breakage; for reports with dark backgrounds, the threshold is lowered to prevent text sticking; S1.

4. Perform a closing operation on the binarized image to fill in the small gaps between the strokes of the characters to ensure that the characters are complete. S1.

5. Detect text region boundaries in images, automatically crop irrelevant background, and correct tilted or distorted text lines through affine transformation.

3. The method for automatically scanning and interpreting an optometry report according to claim 2, characterized in that: The implementation steps of step S2 are: S2.

1. Using projection analysis, we segmented the binary image into independent text lines to avoid overlapping text. We then used a CRNN model to recognize characters line by line, and combined it with a medical dictionary to correct misrecognition results. S2.

2. Categorize the recognition results by field and output them in key-value pair format for easy subsequent processing; S2.

3. Use regular expressions to remove irrelevant symbols or garbled characters from the OCR results, retaining only numbers, letters, and technical terms. S2.

4. Based on a pre-defined terminology mapping table, convert non-standard expressions to a unified format, eliminate dialect or abbreviation differences, force conversion of units to a unified standard, and format numerical ranges; S2.

5. Verify whether any required fields are missing. If missing, mark them as "to be completed" and notify the user to re-upload.

4. The method for automatically scanning and interpreting an optometry report according to claim 3, wherein: The implementation steps of step S3 are: S3.

1. Extract key parameters from text using a rule engine; S3.

2. Check the correlation between parameters; S3.

3. Set a range. If the value exceeds the limit, an alarm will be triggered, indicating that there may be a measurement or recognition error. S3.

4. Select a preset template based on the report type and fill in the corresponding fields with the verified data. S3.

5. Insert a clear label in the structured report and attach detailed instructions; S3.

6. Supports generation of reports in JSON, PDF or HTML format to meet the needs of hospital system connection, patient archiving or printing.

5. The method for automatically scanning and interpreting an optometry report according to claim 4, characterized in that: The implementation steps of step S4 are: S4.

1. Convert structured data into line charts and scatter plots, and render them in real time using SVG or Canvas. S4.

2. When a user clicks an abnormal data point, a floating window pops up showing the original image fragment, the recognition result, and review suggestions. It also supports one-click jump to historical report comparison. S4.

3. Automatically adjust the layout for different terminals to ensure that charts and text are clear and readable in mini-programs and web pages.

6. The method for automatically scanning and interpreting an optometry report according to claim 5, characterized in that: The system for automatically scanning and interpreting optometry reports includes an image preprocessing module, a text recognition module, a semantic parsing module, and a report generation module; The image preprocessing module is used for image enhancement and image standardization; The text recognition module is used for OCR recognition and data cleaning; The semantic parsing module is used for key information extraction and data verification; The report generation module is used for result structuring and user interface generation.

7. The method for automatically scanning and interpreting an optometry report according to claim 6, characterized in that: The image preprocessing module also includes an image enhancement unit and an image standardization unit; The image enhancement unit performs a denoising operation on the scanned optometry report image using a Gaussian filtering algorithm to eliminate noise, blur or uneven lighting in the image, thereby improving the clarity of subsequent processing; The image standardization unit adopts an adaptive binarization algorithm to perform grayscale conversion and binarization processing on the image, unifies the image format and highlights the text area, ensuring that the image format of the optometry report from different sources is consistent.

8. The method for automatically scanning and interpreting an optometry report according to claim 7, characterized in that: The text recognition module also includes an OCR recognition unit and a data cleaning unit; The OCR recognition unit performs text recognition on the pre-processed image based on the Tesseract OCR engine combined with a deep learning model, and accurately extracts patient information and optometry data from the optometry report, including visual acuity, corrected visual acuity, spherical power, cylindrical power, axial direction, and axial length from the axial measurement report; The data cleaning unit performs redundant character filtering, format error correction and data standardization on the original text output by OCR through regular expression matching and keyword library verification.

9. The method for automatically scanning and interpreting an optometry report according to claim 8, characterized in that: The semantic parsing module also includes a key information extraction unit and a data verification unit; The key information extraction unit uses rule-based named entity recognition (NER) combined with medical knowledge graphs to extract key parameters such as naked eye vision, corrected visual acuity, diopter, and axial length from the cleaned text; The data verification unit verifies the rationality of the extracted parameters through a logic verification algorithm, marks abnormal data and feeds back to the user.

10. The method for automatically scanning and interpreting an optometry report according to claim 9, characterized in that: The report generation module also includes a result structuring unit and a user interface generation unit; The result structuring unit uses a template engine to automatically fill in the parsed optometry data according to a preset template, and generates a structured report including data summary and abnormal prompts; The user interface generation unit dynamically renders the structured report based on HTML / CSS technology, generates a visual interface, and displays it to the user through a mini program or a web page.

Images