A method for automatically parsing physical examination report OCR results based on a large model

Abstract

The present invention discloses a method for automatically parsing the OCR results of a physical examination report based on a large model. The method disclosed in the present invention utilizes deep learning and natural language processing technology to efficiently and accurately parse the text information in the physical examination report, so as to realize structured processing and intelligent analysis of the physical examination report data. The present invention can not only improve the parsing efficiency and accuracy of the physical examination report and reduce the occupation of medical resources, but also provide users with more intuitive and easy-to-understand health information, and meet the demand for automated processing of physical examination reports in the field of medical informationization. After adopting the technical solution disclosed in the present invention, it is possible to improve the level of medical informationization, optimize the processing flow of physical examination reports, and provide technical support for medical data analysis and clinical decision-making.

Description

Technical Field

[0001] The invention relates to an automatic parsing method for optical character recognition (OCR) results of a physical examination report based on a large model, and belongs to the technical fields of artificial intelligence, natural language processing and medical health. Background Art

[0002] Today, as health awareness grows, regular physical examinations have become an important part of public health management. As a key document to record personal health status, physical examination reports usually contain the results of various examination items such as general examination, blood analysis, urine test, X-ray imaging, electrocardiogram, B-ultrasound, etc. These results can fully reflect the individual's health status and potential health problems.

[0003] The format and content of physical examination reports vary from medical institution to medical institution, which leads to challenges in information integration and automated processing of physical examination reports. Traditional physical examination report processing relies on manual interpretation and data entry, which is not only time-consuming and error-prone, but also difficult to adapt to the needs of large-scale data processing.

[0004] With the development of optical character recognition (OCR) technology, the digital conversion of physical examination reports has become possible. However, when processing physical examination reports, existing OCR technology can only achieve basic text recognition and simple rule matching analysis, and lacks the ability to deeply analyze and structure the recognition results. For example, patent document CN 118053166 A proposes a method for physical examination report detection, by matching the detection name of each detection item in the physical examination report information after OCR recognition with the corresponding preset physical examination indicator name, so as to parse out the relevant information. This method relies on a preset list of physical examination indicator names, which may not cover all possible physical examination items, and has poor flexibility and adaptability. Moreover, only using name matching loses the context information of the report, which may be easy to make wrong judgments.

[0005] In addition, the medical terms and professional data contained in the physical examination report are difficult for non-professionals to understand, making the automatic parsing of the physical examination report a technical problem that needs to be solved urgently by technicians in this field. How to effectively structure the physical examination report, extract key information, and present it to the user in an easy-to-understand way is the key to improving the efficiency of physical examination report processing and user experience.

[0006] Although OCR technology plays an important role in the digital conversion of physical examination reports, it has significant limitations in automatic parsing and structured processing. Specifically, the existing technology fails to provide effective solutions to the following key technical issues:

[0007] 1. Accuracy of text recognition: Medical terms and professional data in physical examination reports require high accuracy of OCR technology. Existing OCR technology often misidentifies or misses text when recognizing complex formats and professional terms, resulting in inaccurate subsequent analysis.

[0008] 2. Lack of structured analysis: Even if OCR technology can recognize text, existing methods fail to effectively convert the recognition results into structured data. This means that a large amount of text information still needs to be manually sorted and classified, and true automated processing cannot be achieved.

[0009] 3. Difficulty in information integration: The formats of physical examination reports from different medical institutions vary, which makes it impossible to effectively integrate and compare information even between different physical examination cycles of the same user.

[0010] 4. Insufficient user experience: Due to the lack of in-depth understanding and intelligent analysis of the contents of the physical examination report, it is difficult for users to quickly grasp their health status and changing trends when faced with unstructured text information.

[0011] 5. Waste of medical resources: The process of manually reviewing and correcting OCR results takes up a lot of medical resources, including time and human resources, which is particularly prominent in the context of tight medical resources. Summary of the invention

[0012] The technical problems to be solved by the present invention are: how to use a large model to accurately identify and parse medical terms and professional data in physical examination reports, how to convert unstructured OCR results into structured data that is easy to understand and analyze, and how to provide users with an intuitive and accurate physical examination report parsing result.

[0013] In order to solve the above technical problems, the technical solution of the present invention is to disclose a method for automatically parsing the OCR results of a physical examination report based on a large model, which is characterized by comprising the following steps:

[0014] Step 1: Obtain the physical examination report and perform preprocessing: Obtain the physical examination report image and perform preprocessing to improve the accuracy of OCR recognition;

[0015] Step 2, OCR recognition: Use OCR technology to convert the text information in the physical examination report image into machine-readable text data, and extract text attribute features, layout features, and table features;

[0016] Step 3: Large model analysis:

[0017] Step 301, sending text data and its related text attribute features, layout features, table features, etc. into the macro model;

[0018] Step 302: The large model understands the text data and uses text attribute features, layout features, and table features for reasoning to preliminarily split the physical examination report to obtain text blocks including patient information, department unit examination results, and health advice;

[0019] Step 303, identifying and dividing the text block content obtained in step 302 again to obtain a text information block;

[0020] Step 304: extract detailed information from the final text information block;

[0021] Step 4: Structured output display: convert the parsed key information into a structured data format;

[0022] Step 5: Manually review and correct the final parsed structured data;

[0023] Step 6: Save the manually reviewed and corrected data and the corresponding physical examination report data, process and construct them into large model fine-tuning data for iterative optimization of the large model.

[0024] Preferably, in step 1, the preprocessing step includes denoising, contrast enhancement and size normalization.

[0025] Preferably, the denoising adopts median filtering, Gaussian filtering or bilateral filtering, the contrast enhancement adopts histogram equalization, adaptive histogram equalization or local contrast enhancement, and the size is standardized to 1.0 times, 1.2 times or 1.5 times the original size.

[0026] The method disclosed in the present invention utilizes deep learning and natural language processing technology to efficiently and accurately parse the text information in the physical examination report to achieve structured processing and intelligent analysis of the physical examination report data. The present invention can not only improve the parsing efficiency and accuracy of the physical examination report and reduce the occupation of medical resources, but also provide users with more intuitive and easy-to-understand health information, and meet the needs of the medical information field for automated processing of physical examination reports. After adopting the technical solution disclosed in the present invention, the level of medical informationization can be improved, the processing flow of the physical examination report can be optimized, and technical support can be provided for medical data analysis and clinical decision-making.

[0027] Compared with the prior art solutions, the present invention has the following beneficial effects:

[0028] 1. Improve parsing accuracy: By using a large model to disassemble and extract text information from a physical examination report, the present invention can more accurately identify and parse key information in the physical examination report. The contextual understanding and reasoning capabilities of the large model reduce matching errors caused by OCR recognition errors and improve the accuracy of information extraction.

[0029] 2. Enhanced flexibility and adaptability: The present invention does not rely on a preset list of physical examination indicator names, but adapts to the physical examination report formats and terminology of different medical institutions through the learning ability of the large model, thereby enhancing the flexibility and adaptability of the method.

[0030] 3. Improve accuracy: The present invention reduces the time of data processing through automated processes, making the analysis of physical examination reports faster and improving the overall processing efficiency.

[0031] 4. Continuous learning and optimization: The large model can continuously optimize the parsing algorithm through continuous training and learning, adapt to the new physical examination report format and content, and maintain the advancement of technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0034] The method disclosed in the present invention for automatically parsing the OCR results of a physical examination report based on a large model realizes automatic parsing and structural processing of the OCR results of the physical examination report by introducing natural language processing technology based on a large model, thereby providing a new solution for the automatic processing of the physical examination report and promoting the development of medical informationization. The method specifically includes the following steps:

[0035] Step 1, obtaining a physical examination report and performing preprocessing: obtaining a physical examination report image and performing preprocessing to improve the accuracy of OCR recognition. In the embodiment of the present invention, the preprocessing step includes denoising, contrast enhancement and size standardization.

[0036] Step 2, OCR recognition: Use OCR technology to convert the text information in the physical examination report image into machine-readable text data, and extract: text attribute features, including but not limited to font size and type; layout features, including but not limited to text coordinates and line spacing; table features, including but not limited to cells, etc.

[0037] Step 3: Large model analysis:

[0038] Step 301, sending text data and its related text attribute features, layout features, table features and other data into the macro model;

[0039] Step 302: The large model understands the text data and uses data such as text attribute features, layout features, and table features for reasoning to preliminarily split the physical examination report into text blocks such as patient information, department unit examination results, and health advice.

[0040] Step 303: Identify and divide the patient information, department unit and other text block contents obtained above again. For example, the department unit can be further divided into separate text information blocks for each department.

[0041] Step 304: extract detailed information from the final text information block.

[0042] The semantic understanding, reasoning and calculation capabilities of the big model as well as knowledge of document formats are used. Specifically, the big model needs to read and understand the text content extracted by OCR, and make inferences based on coordinate positions, line spacing, and paragraph spacing to ensure that it is not serial. Based on knowledge of document formats and the current input text features, the hierarchical relationships of titles, content, paragraphs, etc. are inferred. In order to avoid the problems of hallucinations and knowledge updates of the big model, the relevant knowledge base is mounted at the same time and enhanced using RAG technology to improve the accuracy of the final disassembly and extraction of the model.

[0043] Step 4: Structured output display: Convert the parsed key information into a structured data format, such as JSON or XML, for easy viewing, correction, and display.

[0044] Step 5: Manually review and correct the final parsed structured data.

[0045] Step 6: Save the manually reviewed and corrected data and the corresponding physical examination report data, process and construct them into large model fine-tuning data for iterative optimization of the large model.

[0046] The following are several embodiments of the present invention, showing process conditions represented by different parameters or parameter ranges:

[0047] Example 1

[0048] In the preprocessing step of step 1, median filtering is used for denoising, histogram equalization is used for contrast enhancement, and the size is normalized to 1.2 times the original size.

[0049] In the OCR recognition step of step 2, a specific OCR engine A is used for text recognition, and the recognition accuracy rate reaches 95%.

[0050] In the large model parsing module of step 3, the pre-trained model A is used, and the accuracy of key information extraction is 98%.

[0051] Example 2

[0052] In the preprocessing step of step 1, Gaussian filtering is used for denoising, adaptive histogram equalization is used for contrast enhancement, and the size is normalized to 1.5 times the original size.

[0053] In the OCR recognition step of step 2, a specific OCR engine B is used for text recognition, and the recognition accuracy rate reaches 97%.

[0054] In the large model analysis of step 3, the pre-trained model B is used, and the accuracy of key information extraction is 97%.

[0055] Example 3

[0056] In the preprocessing step of step 1, bilateral filtering is used for denoising, local contrast enhancement is used for contrast enhancement, and the size is normalized to 1.0 times the original size.

[0057] In the OCR recognition step of step 2, a specific OCR engine C is used for text recognition, and the recognition accuracy rate reaches 96%.

[0058] In the large model analysis of step 3, the pre-trained model C is used, and the accuracy of key information extraction is 96%.

[0059] In the above technical solution, some structures, devices, and method steps are replaceable. For example, the OCR recognition module can be replaced by other OCR engines, and the large model parsing module can be replaced by different pre-trained models, as well as various fine-tuned and enhanced models.

Claims

1. A method for automatically parsing the OCR results of a physical examination report based on a large model, characterized in that: The following steps are involved: Step 1: Obtain the physical examination report and perform preprocessing: Obtain the physical examination report image and perform preprocessing to improve the accuracy of OCR recognition; Step 2, OCR recognition: Use OCR technology to convert the text information in the physical examination report image into machine-readable text data, and extract text attribute features, layout features, and table features; Step 3: Large model analysis: Step 301, sending text data and its related text attribute features, layout features, table features, etc. into the macro model; Step 302: The large model understands the text data and uses text attribute features, layout features, and table features for reasoning to preliminarily split the physical examination report to obtain text blocks including patient information, department unit examination results, and health advice; Step 303, identifying and dividing the text block content obtained in step 302 again to obtain a text information block; Step 304: extract detailed information from the final text information block; Step 4: Structured output display: convert the parsed key information into a structured data format; Step 5: Manually review and correct the final parsed structured data; Step 6: Save the manually reviewed and corrected data and the corresponding physical examination report data, process and construct them into large model fine-tuning data for iterative optimization of the large model.

2. The method for automatically parsing the OCR results of a physical examination report based on a large model as described in claim 1, characterized in that: In step 1, the preprocessing steps include denoising, contrast enhancement and size normalization.

3. The method for automatically parsing the OCR results of a physical examination report based on a large model as described in claim 2, characterized in that: The denoising adopts median filtering, Gaussian filtering or bilateral filtering, the contrast enhancement adopts histogram equalization, adaptive histogram equalization or local contrast enhancement, and the size is standardized to 1.0 times, 1.2 times or 1.5 times the original size.

Images