Deep learning-based digestive endoscopy super minimally invasive surgery report structuring method

Abstract

The invention provides a digestive endoscopy super minimally invasive surgery report structuring method based on deep learning. Belongs to the field of medical information and comprises the steps that in the initial stage, historical digestive endoscopy super minimally invasive surgery medical report data are collected, an initial and extensible template library is constructed based on the data, and the template library comprises common disease information and lesion part information; dynamically updating, namely regularly analyzing new medical report data by utilizing a natural language processing technology, extracting new disease or diseased region information from the new medical report data, and automatically adding the new information into the template library to realize dynamic updating of the template library; the template can widely cover disease information and lesion information which can be collected, and correct matching and structuring of the information are achieved.

Description

Technical Field

[0001] The present invention relates to the field of medical information, and in particular to a method for structuring reports of super-minimally invasive gastrointestinal endoscopy surgery based on deep learning. Background Art

[0002] With the continuous development of artificial intelligence technology, the application of deep learning in the medical field will become more and more extensive. As an important part of this, the structured method of digestive endoscopy super minimally invasive surgery report will be combined with other medical technologies and equipment to jointly promote the development of intelligent medicine.

[0003] There are some challenges and difficulties in the actual application of structuring digestive endoscopy super minimally invasive surgery reports. The structured template may not cover all possible disease information and lesion site information, resulting in some information not being correctly matched and structured. In addition, when encountering new diseases or lesion sites, the template needs to be updated, increasing maintenance costs. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method for structuring reports of super-minimally invasive gastrointestinal endoscopy surgery based on deep learning. The method uses natural language processing technology to regularly analyze new medical report data, extract new disease or lesion site information therefrom, and automatically add the new information to the template library to realize dynamic updating of the template library.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, the method includes the following steps: Initial stage: Collect historical digestive endoscopy super-minimally invasive surgery medical report data, and build an initial, expandable template library based on these data, which contains common disease information and lesion site information; Dynamic update: Utilize natural language processing technology to regularly analyze new medical report data, extract new disease or lesion site information, and automatically add this new information to the template library to achieve dynamic update of the template library.

[0006] Preferably, the specific steps of dynamically updating the template library include: Data preprocessing: Perform preprocessing operations such as cleaning, word segmentation, and part-of-speech tagging on new medical report data; Entity recognition: Use named entity recognition technology to identify new disease or lesion information from pre-processed medical report data; Template update: Add the newly identified disease or lesion site information to the template library and update the corresponding template structure and content.

[0007] Preferably, the method further comprises a multi-template matching algorithm for improving the accuracy of the medical report structuring, the algorithm comprising: Candidate template selection: A set of candidate templates are selected from the template library based on the keywords or features in the medical report; Similarity calculation: The text similarity algorithm is used to calculate the similarity between the medical report and each candidate template; Sorting and selection: Sort the candidate templates according to the similarity calculation results, and select the template with the highest similarity as the best matching template for structured processing.

[0008] Preferably, the multi-template matching algorithm also includes fuzzy matching technology for processing information in the medical report that is not completely consistent with the template, and the specific steps include: Feature extraction: extract key features from medical reports and templates; Feature matching: Try to establish a matching relationship between features in medical reports and templates, and allow a certain degree of difference; Matching degree calculation: Calculate the matching degree of each feature matching pair, and summarize the matching degrees of the feature matching pairs to obtain the overall matching degree between the report and the template; Matching result selection: Select the best matching template for structural processing based on the overall matching degree.

[0009] Preferably, the method further comprises a manual verification and feedback mechanism for improving the accuracy of the structured results and the quality of the template library, the mechanism comprising: Manual verification: Manual review of the structured medical reports to check the accuracy of the structured results; Feedback collection: Collect user feedback on structured results, including error pointing, modification suggestions, etc. Real-time correction: The templates in the template library are corrected in real time based on user feedback to optimize the effect of structured processing.

[0010] Preferably, the method further includes real-time monitoring and optimization of the template library performance, and the specific steps include: Performance evaluation: Regularly evaluate the performance of the template library, including matching success rate and accuracy indicators; Problem identification: Analyze the performance evaluation results and identify the problems and deficiencies in the template library; Optimization measures: Take corresponding optimization measures for the identified problems and deficiencies, including adjusting the template structure and optimizing the matching algorithm; Iterative update: Reapply the optimized template library to the structured processing of medical reports and continuously monitor its performance.

[0011] Preferably, the natural language processing technology includes text segmentation, part-of-speech tagging, named entity recognition, and relationship extraction, and these technologies are comprehensively applied to information extraction and structured processing of medical reports.

[0012] Preferably, the method also includes using machine learning technology to continuously optimize and improve natural language processing technology and template matching algorithms to improve the accuracy and efficiency of structured processing of medical reports.

[0013] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for structuring reports of super minimally invasive gastrointestinal endoscopy surgery based on deep learning. The method utilizes natural language processing technology to regularly analyze new medical report data, extract new disease or lesion site information therefrom, and automatically add the new information to a template library to achieve dynamic updating of the template library; the template can more widely cover the disease information and lesion site information that can be collected, so that the information can be correctly matched and structured.

[0014] The present invention improves the efficiency and accuracy of information transmission, reduces information comprehension bias, promotes clinical decision-making effectiveness, facilitates information traceability and reuse, and promotes the development of medical intelligence. DETAILED DESCRIPTION

[0015] The present invention is described in further detail below.

[0016] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be used for other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0017] Those skilled in the art should understand that, in the disclosure of the present invention, the terms "upper", "lower", "left", "right", "front", "back", "bottom", "inside", "outside" and the like to indicate directions or positions are only for the convenience of describing the simplified description of the present invention, and do not indicate or imply that the device or component referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, the above terms should not be understood as limiting the present invention.

[0018] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity. Example

[0019] A method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, the method includes the following steps: Initial stage: Collect historical digestive endoscopy super-minimally invasive surgery medical report data, and build an initial, expandable template library based on these data, which contains common disease information and lesion site information; Dynamic update: Utilize natural language processing technology to regularly analyze new medical report data, extract new disease or lesion site information, and automatically add this new information to the template library to achieve dynamic update of the template library.

[0020] In the implementation of the present application, an expandable template library is first established, in which historical digestive endoscopy super-minimally invasive surgery medical report data is collected, including common disease information and lesion site information, that is, the initial template library.

[0021] Specific: Data collection: First, a certain amount of historical digestive endoscopy super-minimally invasive surgery report data needs to be collected; the data can come from different hospitals, departments or doctors to ensure the diversity and representativeness of the data; Data preprocessing: preprocess the collected report data, including removing irrelevant information (such as headers and footers, dates, etc.), text cleaning (such as removing spaces, line breaks, punctuation marks, etc.), text sentence segmentation, etc.; Analyze data: Analyze the pre-processed report data in detail to identify common disease information and lesion site information; Define the template structure: Based on the analysis results, define the structure of the template. The template usually includes a key-value pair relationship, where the key represents the disease or lesion site, and the value represents the related description or diagnosis result; Construct the initial template: Based on the defined template structure, manually construct the initial template library. The initial template library should contain common diseases and lesion site information, as well as corresponding descriptions and diagnosis results.

[0022] In the implementation mode of the present application, natural language processing technology is used to regularly analyze new medical report data, extract new disease or lesion site information, and automatically add this new information to the template library, so as to automatically update the initial template library.

[0023] Specific: Utilize natural language processing technology: Use natural language processing tools to segment the report text and split the text into individual words or phrases; perform part-of-speech tagging on the segmented text and assign a part-of-speech label (such as noun, verb, adjective, etc.) to each word or phrase; use named entity recognition technology to identify entities such as diseases and lesion sites in the report and use them as candidate keys for template matching; use relationship extraction technology to identify the relationship between entities, such as the correspondence between diseases and lesion sites, to provide additional information for template matching.

[0024] Analyze new reports: Perform the same preprocessing steps as above on the newly collected report data; apply natural language processing techniques such as word segmentation, part-of-speech tagging, named entity recognition, and relation extraction to the preprocessed new reports.

[0025] Automatically extract new information: Based on the results of named entity recognition, extract entities such as diseases and lesion sites in new reports as candidate keys; based on the results of relationship extraction, extract descriptions or diagnostic results related to candidate keys as candidate values; verify and screen the extracted candidate keys and candidate values ​​through rules or machine learning models to ensure the accuracy and reliability of the extracted information.

[0026] Update the template library: match the extracted new information with the existing template library to find out whether there is a similar key-value pair relationship; if no matching template is found, add the newly extracted key-value pair relationship as a new template to the template library. If a matching template is found, but the new information contains a more detailed or accurate description or diagnostic result, the value in the existing template can be updated; in order to track the change history of the template library, implement a version control strategy, and record the changed content and timestamp each time the template library is updated so that the change can be backtracked or rolled back when necessary.

[0027] In an embodiment of the present application, the method further includes a multi-template matching algorithm for improving the accuracy of the medical report structuring, and the algorithm includes: Candidate template selection: Templates are classified according to dimensions such as disease category and lesion site, and a corresponding index structure is established for effective organization to enable rapid retrieval and matching. When a new report needs to be structured, a group of candidate templates are selected from the template library based on keywords or features in the medical report; the candidate templates will be related to the disease or lesion site in the report.

[0028] Similarity calculation: For each candidate template, calculate the similarity between it and the report. The similarity calculation can be based on a text similarity algorithm, such as cosine similarity, Jaccard similarity, etc. Specifically, the report and template can be converted into vector representations (such as bag-of-words model, TF-IDF, etc.), and then calculate the similarity between these vectors.

[0029] Sorting and selection: Based on the similarity calculation results, the candidate templates are sorted from high to low according to the similarity. The template with the highest similarity is most likely to be the template that best matches the report. Finally, the template with the highest similarity is selected as the final matching result, and the report is structured based on the template.

[0030] The multi-template matching algorithm also includes fuzzy matching technology, which is used to process information in medical reports that is not completely consistent with the template. The specific steps include: Feature extraction: For fuzzy matching, key features are first extracted from reports and templates; these features may include disease names, lesion site names, symptom descriptions, etc. These features can be extracted through natural language processing technologies such as named entity recognition and part-of-speech tagging; Feature matching: After extracting key features, a matching relationship between features is established between the report and the template. Since fuzzy matching allows a certain degree of difference, a fuzzy matching rule or threshold is defined; for example, a certain degree of difference in spelling of the disease name or the lesion site name is allowed (such as abbreviations, synonyms, etc.); Matching degree calculation: For each feature matching pair, calculate its matching degree. The matching degree calculation can be based on factors such as feature similarity and weight. Specifically, a weight can be assigned to each feature, and the matching degree is calculated based on the feature similarity and weight. The matching degrees of all feature matching pairs are summarized to obtain the overall matching degree between the report and the template. The higher the overall matching degree, the higher the matching degree between the report and the template. Matching result selection: Based on the calculation results of the overall matching degree, the template with the highest matching degree is selected as the final matching result, and the report is structured based on the template. If the matching degrees of multiple templates are similar, any one of them can be selected or further manual verification can be performed.

[0031] In the embodiment of the present application, the method further includes a manual verification and feedback mechanism for improving the accuracy of the structured results and the quality of the template library, the mechanism comprising: Manual verification: Manual review of the structured medical reports to check the accuracy of the structured results; Feedback collection: Collect user feedback on structured results, including error pointing, modification suggestions, etc. Real-time correction: The templates in the template library are corrected in real time based on user feedback to optimize the effect of structured processing.

[0032] In the implementation manner of the present application, the method further includes real-time monitoring and optimization of the performance of the template library, and the specific steps include: Performance evaluation: Regularly evaluate the performance of the template library, including matching success rate and accuracy indicators; Problem identification: Analyze the performance evaluation results and identify the problems and deficiencies in the template library; Optimization measures: Take corresponding optimization measures for the identified problems and deficiencies, including adjusting the template structure and optimizing the matching algorithm; Iterative update: Reapply the optimized template library to the structured processing of medical reports and continuously monitor its performance.

[0033] This method uses machine learning technology to continuously optimize and improve natural language processing technology and template matching algorithms to improve the accuracy and efficiency of structured processing of medical reports.

[0034] Those skilled in the art should understand that the embodiments of the present invention shown in the above description are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

Claims

1. A structured method for digestive endoscopy super minimally invasive surgery report based on deep learning, Features: The method comprises the following steps, Initial stage: Collect historical digestive endoscopy super-minimally invasive surgery medical report data, and build an initial, expandable template library based on these data, which contains common disease information and lesion site information; Dynamic update: Utilize natural language processing technology to regularly analyze new medical report data, extract new disease or lesion site information, and automatically add this new information to the template library to achieve dynamic update of the template library.

2. According to claim 1, the method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, Features: The specific steps of dynamically updating the template library include: Data preprocessing: Perform preprocessing operations such as cleaning, word segmentation, and part-of-speech tagging on new medical report data; Entity recognition: Use named entity recognition technology to identify new disease or lesion information from pre-processed medical report data; Template update: Add the newly identified disease or lesion site information to the template library and update the corresponding template structure and content.

3. According to claim 2, the method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, Features: The method also includes a multi-template matching algorithm for improving the accuracy of medical report structuring, the algorithm including: Candidate template selection: A set of candidate templates are selected from the template library based on the keywords or features in the medical report; Similarity calculation: The text similarity algorithm is used to calculate the similarity between the medical report and each candidate template; Sorting and selection: Sort the candidate templates according to the similarity calculation results, and select the template with the highest similarity as the best matching template for structured processing.

4. According to claim 3, the method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, Features: The multi-template matching algorithm also includes fuzzy matching technology for processing information in the medical report that is not completely consistent with the template. The specific steps include: Feature extraction: extract key features from medical reports and templates; Feature matching: Try to establish a matching relationship between features in the medical report and the template, and allow a certain degree of difference; Matching degree calculation: Calculate the matching degree of each feature matching pair, and summarize the matching degrees of the feature matching pairs to obtain the overall matching degree between the report and the template; Matching result selection: Select the best matching template for structural processing based on the overall matching degree.

5. According to any one of claims 1 to 4, the method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, Features: The method also includes a manual verification and feedback mechanism to improve the accuracy of the structured results and the quality of the template library, which includes: Manual verification: Manual review of the structured medical reports to check the accuracy of the structured results; Feedback collection: Collect user feedback on structured results, including error pointing, modification suggestions, etc. Real-time correction: The templates in the template library are corrected in real time based on user feedback to optimize the effect of structured processing.

6. According to claim 5, the method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, Features: The method also includes real-time monitoring and optimization of the template library performance, and the specific steps include: Performance evaluation: Regularly evaluate the performance of the template library, including matching success rate and accuracy indicators; Problem identification: Analyze the performance evaluation results and identify the problems and deficiencies in the template library; Optimization measures: Take corresponding optimization measures for the identified problems and deficiencies, including adjusting the template structure and optimizing the matching algorithm; Iterative update: Reapply the optimized template library to the structured processing of medical reports and continuously monitor its performance.

7. The method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning according to claim 6, Features: The natural language processing technology includes text segmentation, part-of-speech tagging, named entity recognition, and relationship extraction, which are comprehensively applied to information extraction and structural processing of medical reports.

8. According to claim 7, the method for structuring digestive endoscopy super minimally invasive surgery reports based on deep learning, Features: The method also includes using machine learning technology to continuously optimize and improve natural language processing technology and template matching algorithms to improve the accuracy and efficiency of structured processing of medical reports.