A method for automatically extracting and analyzing data information from medical examination forms

Abstract

The present application provides a method for automatically extracting and analyzing data information of a medical test form, which relates to the technical field of data extraction and analysis. The text information of the medical test form is semantically segmented through a text recognition framework to obtain descriptor segments of different test information, and the fuzzy entropy of the data text information is determined based on the discrete test data in each descriptor segment; then, the thrown data in the medical test form is extracted based on the text recognition framework after fuzzy entropy optimization, and the subordinate attention matching mechanism of the test information in the medical test form is determined through the description vectors of all descriptor segments; then, a test data analysis model based on a bidirectional long short-term memory neural network is constructed based on a knowledge graph; the test data analysis model performs test analysis on the thrown data based on the subordinate attention matching mechanism to obtain the analysis result of the medical test form. The present application can perform adaptive subordinate matching analysis on the test information in the medical test form to improve the reliability of medical tests.

Description

Technical Field

[0001] The present application relates to the technical field of data extraction and analysis, and more specifically, to a method for automatically extracting and analyzing data information of a medical examination form. Background Art

[0002] Data extraction and analysis technology is to automatically extract valuable information from unstructured or semi-structured data. In the medical field, data extraction and analysis technology usually uses text analysis frameworks, natural language processing technology and deep learning models to identify, extract and analyze data information of medical test reports.

[0003] With the advancement of medical dataization, automatic extraction of data information from medical test orders has become a key technology to improve the efficiency and accuracy of medical services. Traditional methods for extracting information from medical test orders mostly rely on manual entry or rule-driven text matching technology. Automated extraction methods based on natural language processing technology and machine learning models can improve the accuracy and efficiency of data extraction. However, existing natural language processing technology and machine learning models cannot adaptively perform accurate matching and analysis based on the affiliation between test data, resulting in low reliability of medical test data extraction and analysis. Therefore, how to perform adaptive affiliation matching analysis on the test information in medical test orders to improve the reliability of medical tests has become a difficult problem faced by the industry. Summary of the invention

[0004] The present application provides a method for automatically extracting and analyzing data information of a medical test form, which can perform adaptive subordination matching analysis on the test information in the medical test form to improve the reliability of the medical test.

[0005] The present application provides a method for automatically extracting and analyzing data information of a medical test form, the method comprising the following steps:

[0006] Get the data text information of the medical examination form;

[0007] Preprocessing the data text information, and semantically segmenting the preprocessed data text information through a text recognition framework to obtain descriptor segments of different verification information, and then determining the fuzzy entropy of the data text information based on discrete verification data in each descriptor segment;

[0008] Extract the thrown data in the medical examination form based on the text recognition framework optimized by fuzzy entropy, determine the description vector of each description sub-segment, and then determine the subordinate attention matching mechanism of the examination information in the medical examination form through all the description vectors;

[0009] Obtaining a knowledge graph during medical examination, and constructing a test data analysis model based on a bidirectional long short-term memory neural network according to the knowledge graph;

[0010] The inspection data analysis model inspects and analyzes the thrown data based on the subordinate attention matching mechanism, and then obtains the analysis result of the medical inspection form.

[0011] In this embodiment, the data text information of the medical examination form is obtained by scanning the medical examination form.

[0012] In this embodiment, preprocessing the data text information specifically includes:

[0013] Determine the structured inspection data to be extracted from the data text information, determine the data type and processing method of the structured inspection data, and then convert the structured inspection data into a plain text format.

[0014] In this embodiment, semantic segmentation is performed on the preprocessed data text information through a text recognition framework, and the descriptor segments of different verification information obtained specifically include:

[0015] Determine the model structure of the text recognition framework;

[0016] Determine an encoder of a text recognition framework based on the model structure, and perform contextual semantic analysis on the preprocessed data text information through the encoder to obtain a text sequence with a test label;

[0017] Extract description sub-segments with different test information from text sequences with test labels.

[0018] In this embodiment, determining the fuzzy entropy of the data text information according to the discrete test data in each description sub-segment specifically includes:

[0019] Extracting discrete test data from each descriptor segment;

[0020] Standardize all discrete test data to obtain standardized data sequences;

[0021] The fuzzy entropy of the data text information is determined according to a preset similarity threshold and the standardized data sequence.

[0022] In this embodiment, extracting the thrown data in the medical examination form based on the text recognition framework optimized by fuzzy entropy specifically includes:

[0023] For each descriptor segment, the approximate entropy of the descriptor segment is determined according to all the test data in the descriptor segment, and then the approximate entropy of each descriptor segment is obtained;

[0024] Determine the attention allocation result of the text recognition framework through the approximate entropy of each description sub-segment and the fuzzy entropy;

[0025] The thrown data in the medical examination form is extracted based on the text recognition framework optimized based on the attention allocation result.

[0026] In this embodiment, determining the description vector of each description sub-segment specifically includes:

[0027] Extracting semantic features, numerical features and structural features from each descriptor segment;

[0028] Feature fusion and normalization operations are performed on the semantic features, numerical features and structural features of each description sub-segment respectively to obtain a description vector of each description sub-segment.

[0029] In this embodiment, the subordinate attention matching mechanism for determining the inspection information in the medical inspection form through all description vectors specifically includes:

[0030] Performing weighted aggregation on all description vectors to obtain a global description vector of the medical examination form;

[0031] Determining the membership degree of each description sub-segment based on the global description vector and the description vector of each description sub-segment;

[0032] The affiliation attention matching mechanism of the inspection information in the medical inspection form is determined by all the affiliations.

[0033] In this embodiment, experimental data of medical examinations are extracted from a database of medical examinations, and then a knowledge graph of the medical examinations is acquired based on the pre-trained language model through the experimental data.

[0034] In this embodiment, the text recognition framework is an automated extraction program based on natural language processing.

[0035] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects:

[0036] First, the data text information of the medical test form is obtained; secondly, the data text information is preprocessed, and the preprocessed data text information is semantically segmented through a text recognition framework to obtain description sub-segments of different test information, and then the fuzzy entropy of the data text information is determined according to the discrete test data in each description sub-segment; then, based on the text recognition framework optimized by the fuzzy entropy, the thrown data in the medical test form is extracted, the description vector of each description sub-segment is determined, and then the subordinate attention matching mechanism of the test information in the medical test form is determined through all the description vectors; then, the knowledge graph during the medical test is obtained, and a test data analysis model based on a bidirectional long short-term memory neural network is constructed according to the knowledge graph; finally, the test data analysis model performs test analysis on the thrown data based on the subordinate attention matching mechanism, and then obtains the analysis result of the medical test form.

[0037] It can be seen that in the present application, an adaptive membership matching analysis can be performed on the inspection information in the medical inspection form; wherein, through the text recognition framework optimized by fuzzy entropy, it is convenient to extract high-value information from the inspection data, and at the same time remove redundant or irrelevant data, so as to obtain the thrown data in the medical inspection form; then, the importance of each descriptor segment in the entire medical inspection form can be automatically identified through the membership matching mechanism, and the attention can be focused on the data that has an important impact on the inspection results, providing an adaptive learning method for the model, which is convenient for adaptively adjusting the extraction weights of different inspection information, and is conducive to improving the matching and recognition capabilities of the inspection data system; finally, a test data analysis model based on a bidirectional long short-term memory neural network can be constructed through the knowledge graph, and combined with the membership attention matching mechanism, the test data analysis model can automatically analyze the thrown data, which is conducive to improving the intelligence level of data processing and ensuring the reliability of the analysis results.

[0038] In summary, the technical solution adopted in this application can perform adaptive subordination matching analysis on the inspection information in the medical inspection form to improve the reliability of medical inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 is an exemplary flow chart of the method for automatically extracting and analyzing data information of a medical examination form provided in the present application;

[0041] Figure 2is an exemplary flow chart of determining a description sub-segment according to the present application;

[0042] Figure 3 This is an exemplary flowchart for determining the affiliated attention matching mechanism provided in this application. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0044] The embodiment of the present application provides a method for automatically extracting and analyzing data information of a medical test form, the core of which is, first, obtaining the data text information of the medical test form; second, preprocessing the data text information, and semantically segmenting the preprocessed data text information through a text recognition framework to obtain descriptor segments of different test information, and then determining the fuzzy entropy of the data text information based on the discrete test data in each descriptor segment; then, extracting the thrown data in the medical test form based on the text recognition framework optimized by the fuzzy entropy, determining the description vector of each descriptor segment, and then determining the subordinate attention matching mechanism of the test information in the medical test form through all the description vectors; then, obtaining the knowledge graph during medical examination, and constructing a test data analysis model based on a bidirectional long short-term memory neural network based on the knowledge graph; finally, the test data analysis model performs test analysis on the thrown data based on the subordinate attention matching mechanism, and then obtains the analysis result of the medical test form. The above scheme can be used to perform adaptive subordinate matching analysis on the test information in the medical test form to improve the reliability of medical examination.

[0045] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 As shown in FIG. 1 , this figure is an exemplary flow chart of a method for automatically extracting and analyzing data information of a medical test form according to this embodiment of the present application, and the extraction and analysis method includes the following steps:

[0046] In step S1, the data text information of the medical examination form is obtained.

[0047] In specific implementation, the data text information of the medical examination form can be obtained by scanning the medical examination form, wherein the data text information in the medical examination form can be scanned by an optical character recognition device. In other embodiments, other devices can also be used for scanning, which is not limited here.

[0048] It should be noted that the data text information in the present application refers to a collection of text information and data information in a medical test form, and the data text information includes structured test data and discrete test data, wherein structured test data refers to data that can be clearly expressed in a specific format, for example, the structured test data may include: test items, test results, units, reference ranges and other data, discrete test data refers to data that cannot be directly represented by standard numerical data and does not have a fixed numerical range, for example, the discrete test data may include: descriptive data, status labels, test results and other data.

[0049] In step S2, the data text information is preprocessed, and the preprocessed data text information is semantically segmented through a text recognition framework to obtain description sub-segments of different verification information, and then the fuzzy entropy of the data text information is determined based on the discrete verification data in each description sub-segment.

[0050] In this embodiment, the data text information is preprocessed by the following steps:

[0051] Determine the structured inspection data to be extracted from the data text information, determine the data type and processing method of the structured inspection data, and then convert the structured inspection data into a plain text format.

[0052] Among them, the data type of the structured test data may include: numerical data, Boolean data, character string, time, and data cleaning is required for different types of data, such as: missing value filling and outlier processing; in addition, the plain text format (i.e., CSV format) is often used to transmit and store medical data. The advantage of converting the structured test data into CSV format is that it is easy to generate and read, and most programming languages ​​support data processing in CSV format, which is conducive to the automatic extraction of test data.

[0053] Preferably, in this embodiment, reference Figure 2 As shown in FIG. 1 , this figure is an exemplary flow chart of determining a description sub-segment in an embodiment of the present application. In this embodiment, the pre-processed data text information is semantically segmented by a text recognition framework, and the description sub-segments of different verification information can be obtained by the following steps:

[0054] First, in step S21, the model structure of the text recognition framework is determined;

[0055] Then, in step S22, an encoder of a text recognition framework is determined based on the model structure, and the encoder performs contextual semantic analysis on the preprocessed data text information to obtain a text sequence with a verification label;

[0056] Finally, in step S23, description sub-segments of different verification information are extracted from the text sequence with the verification label.

[0057] It should be noted that, in this embodiment, the text recognition framework is an automated extraction program based on natural language processing, which can be written through a visual detection algorithm; in addition, the descriptive sub-segments of different test information in this application include test items, test results, units, reference ranges and other data of different structured test data. Through the above steps, the various test information in the medical test form can be accurately captured, providing clear structured data for subsequent data analysis and decision-making. The extracted descriptive sub-segments can be directly used for machine learning analysis and statistical analysis, which is conducive to the subsequent automated analysis of the test results.

[0058] In specific implementation, first, the language representation model in the prior art can be used as the model structure of the text recognition framework to facilitate the identification and analysis of key test information in the text, thereby improving the accuracy and effectiveness of text recognition; then, the encoder in the language representation model can be used as the encoder of the text recognition framework, and the encoder in the language representation model can convert the input text data into a vector representation with semantic information. The encoder in the language representation model will model the context of each word, thereby generating an analysis result containing the word context relationship, and the analysis result includes a test label. Different types of test data can be classified through the test label. The test label may include: blood sugar value and white blood cell count, etc., thereby obtaining a text sequence with a test label, wherein context semantic analysis is beneficial to flexible data processing; finally, a set of data with the same test label in the text sequence with a test label is used as a description sub-segment of the test information corresponding to the test label.

[0059] In this embodiment, the fuzzy entropy of the data text information is determined according to the discrete test data in each descriptor segment, which can be specifically implemented by the following steps:

[0060] Extracting discrete test data from each descriptor segment;

[0061] Standardize all discrete test data to obtain standardized data sequences;

[0062] The fuzzy entropy of the data text information is determined according to a preset similarity threshold and the standardized data sequence.

[0063] In specific implementation, a natural language processing algorithm can be used to extract discrete data from text; then, all discrete test data are standardized using Z-Score standardization, and the sequence of the standardized data is used as a standardized data sequence. Extracting discrete test data from each descriptor segment is helpful for auxiliary diagnosis and identification of test results. In the process of analyzing medical test orders, processing discrete test data facilitates revealing the rules of test results, thereby supporting automated analysis and intelligent diagnosis.

[0064] In addition, in a specific implementation, the fuzzy entropy of the data text information is determined based on a preset similarity threshold and the standardized data sequence, wherein the similarity threshold is a parameter used to measure the data similarity between different descriptor segments, and can be preset by the variance of all data in the standardized data sequence, and the similarity threshold is used as the end point of the vector, and each data in the standardized data sequence is used as the starting point of the vector to generate multiple vectors, and the Euclidean distance between all vectors is described by entropy characteristics, and the obtained result is used as the fuzzy entropy of the data text information.

[0065] It should be noted that in the present application, the fuzzy entropy of the data text information is an indicator to measure the complexity of the test data in the medical test form. In the medical test form, by calculating the fuzzy entropy, the uncertainty of the data can be evaluated, which helps to judge the validity or abnormality of the test information. The similarity threshold can be used to measure the similarity of the data between different description sub-segments. In addition, the entropy feature description of the Euclidean distance between all vectors refers to using the fuzzy entropy calculation formula to calculate the entropy value according to the similarity of the data. The similarity can be measured by the Euclidean distance between the vectors. The core of the fuzzy entropy calculation is to analyze the similarity of the standardized data sequence to infer the complexity of the data, that is, the larger the fuzzy entropy, the higher the complexity of the test data in the medical test form, and the smaller the fuzzy entropy, the lower the complexity of the test data in the medical test form.

[0066] In step S3, based on the fuzzy entropy optimized text recognition framework, the thrown data in the medical examination form is extracted, the description vector of each description sub-segment is determined, and then the subordinate attention matching mechanism of the examination information in the medical examination form is determined through all the description vectors.

[0067] In this embodiment, the extraction of the thrown data in the medical examination form based on the text recognition framework after fuzzy entropy optimization can be specifically implemented by the following steps:

[0068] For each descriptor segment, the approximate entropy of the descriptor segment is determined according to all the test data in the descriptor segment, and then the approximate entropy of each descriptor segment is obtained;

[0069] Determine the attention allocation result of the text recognition framework through the approximate entropy of each description sub-segment and the fuzzy entropy;

[0070] The thrown data in the medical examination form is extracted based on the text recognition framework optimized based on the attention allocation result.

[0071] In the specific implementation, for each descriptor segment, all the verification data in the descriptor segment are substituted into the calculation formula of the approximate entropy for calculation, and the calculation result is used as the approximate entropy of the descriptor segment, thereby obtaining the approximate entropy of each descriptor segment, wherein the approximate entropy reflects the complexity of the verification data in the descriptor segment, and the verification weights of different descriptor segments can be allocated based on the complexity.

[0072] In addition, in a specific implementation, for each descriptor segment, the comprehensive entropy value of the descriptor segment is determined by the approximate entropy of the descriptor segment and the fuzzy entropy, and the determination of the comprehensive entropy value can be implemented by using the following formula:

[0073] E i =α·F i +β·A i

[0074] Among them, E i is the comprehensive entropy value of the description sub-segment, F i is the fuzzy entropy of the data text information, A i is the approximate entropy of the description sub-segment, α and β are coefficients for adjusting the weights of fuzzy entropy and approximate entropy, which can be determined by the ratio of the approximate entropy of the description sub-segment to the standard deviation of the fuzzy entropy, i is the serial number of the description sub-segment, and the description sub-segment can be represented as the i-th description sub-segment during the calculation process. The comprehensive entropy value of each description sub-segment can be obtained by the above formula; then, all the comprehensive entropy values ​​are normalized, and the normalized comprehensive entropy value corresponding to the description sub-segment is used as the attention of the corresponding description sub-segment, thereby obtaining the attention of all description sub-segments, and the attention of all description sub-segments is used as the attention allocation result of the text recognition framework.

[0075] In a specific implementation, the attention allocation result can be used as the attention mechanism of the encoder in the text recognition framework for optimization, and then the thrown data in the medical test form can be extracted through the text recognition framework after the optimized attention mechanism.

[0076] It should be noted that the thrown data in this application refers to the key inspection information with abnormal characteristics in the medical inspection form. The abnormal characteristics are manifested as detection values ​​that exceed the reference range. By extracting the thrown data, the key content that needs to be processed as priority in the medical inspection form can be quickly located, reducing manual screening time.

[0077] In this embodiment, the description vector of each description sub-segment may be determined by the following steps:

[0078] Extracting semantic features, numerical features and structural features from each descriptor segment;

[0079] Feature fusion and normalization operations are performed on the semantic features, numerical features and structural features of each description sub-segment respectively to obtain a description vector of each description sub-segment.

[0080] In the specific implementation, a natural language processing model is initialized, and the semantic features, numerical features and structural features in each descriptor segment are extracted through the natural language processing model based on statistical analysis technology. The statistical analysis technology can adopt the dominator tree in the prior art, which is conducive to traversing text data. Other existing statistical analysis technologies can also be used in other embodiments, which are not limited here.

[0081] In addition, in the specific implementation, the semantic features, numerical features and structural features of each descriptor segment are fused respectively through the convolutional neural network to obtain the feature fusion vector of each descriptor segment, and then all the feature fusion vectors are normalized, and then the normalized fusion features corresponding to the descriptor segment are used as the description vector of the corresponding descriptor segment to obtain the description vectors of all descriptor segments.

[0082] It should be noted that the feature fusion vector in this embodiment refers to a vector that comprehensively represents the semantic features, numerical features and structural features of the description sub-segment. The description vector is a standardized vector used to represent the semantic, numerical and structural features of the description sub-segment, and can be used to compare the similarity between different description sub-segments.

[0083] Preferably, in this embodiment, reference Figure 3 As shown, this figure is an exemplary flow chart of determining the subordinate attention matching mechanism in an embodiment of the present application. In this embodiment, the subordinate attention matching mechanism for determining the inspection information in the medical inspection form through all description vectors can be implemented by the following steps:

[0084] First, in step S31, all description vectors are weighted and aggregated to obtain a global description vector of the medical inspection form;

[0085] Then, in step S32, the degree of membership of each description sub-segment is determined based on the global description vector and the description vector of each description sub-segment.

[0086] Finally, in step S33, the affiliation attention matching mechanism of the inspection information in the medical inspection form is determined through all the affiliations.

[0087] In the specific implementation, first, all description vectors are weightedly aggregated using a linear aggregation algorithm based on weighted average, and the vector obtained by weighted aggregation is used as the global description vector of the medical test order, wherein the global description vector contains the characteristics of all description sub-segments and reflects the global characteristics in the medical test order; then, the cosine similarity between the description vector of each description sub-segment and the global description vector is calculated respectively, and then the cosine similarity corresponding to the description sub-segment is substituted into the membership calculation formula, and the calculation result is used as the membership of the corresponding description sub-segment, wherein the membership calculated by cosine similarity is conducive to quantifying the degree of association of each description sub-segment with the medical test order; finally, an attention matrix is ​​constructed by a deep learning optimizer (such as Adam) using all memberships as construction parameters, and the constructed attention matrix is ​​globally optimized by Transformer, and then the globally optimized attention matrix is ​​used as the membership attention matching mechanism of the inspection information in the medical test order.

[0088] It should be noted that the calculation steps of this embodiment can be implemented based on the NumPy library, which is convenient for efficient processing of complex data, thereby improving the generalization ability of the calculation process. In other embodiments, other existing libraries can also be used for calculation, which is not limited here; in addition, it should be noted that the affiliation attention matching mechanism in this application is an attention mechanism based on affiliation allocation. The affiliation attention matching mechanism is determined based on the affiliation of the description sub-segment in the medical test form, which is convenient for dynamically allocating attention weights of different test information. Among them, the membership reflects the importance of the description sub-segment in the overall data, and the attention mechanism weights the information according to the membership, thereby optimizing the extraction and matching process of the test information.

[0089] In step S4, a knowledge graph of medical examination is obtained, and a test data analysis model based on a bidirectional long short-term memory neural network is constructed according to the knowledge graph.

[0090] In this embodiment, experimental data of medical examinations are extracted from a database of medical examinations, and then a knowledge graph of the medical examinations is obtained through the experimental data based on a pre-trained language model. It should be noted that entities (such as diseases, symptoms, test items, etc.) and relationships (such as "disease-symptom" relationships, "disease-test item" relationships) of the test information can be extracted from the experimental data through relationship extraction combined with entity recognition technology, and the connections between the entities are constructed. By using a pre-trained language model to establish a knowledge graph, a structured knowledge base is provided for subsequent test data analysis. The pre-trained language model is conducive to understanding the complex terms and semantics in medical texts, thereby improving the quality and efficiency of knowledge graph construction.

[0091] In this embodiment, the construction of a test data analysis model based on a bidirectional long short-term memory neural network according to the knowledge graph can be implemented by the following steps:

[0092] Initialize a data analysis model based on a bidirectional long short-term memory neural network;

[0093] The data analysis model is trained, evaluated, optimized and tested, wherein the data analysis model is optimized using the knowledge graph, and a test data analysis model is obtained after testing.

[0094] In specific implementation, the BERT pre-trained language model can be used to map entities extracted from the knowledge graph (such as diseases, symptoms, test items, etc.) to the vector space, which is conducive to the neural network's understanding of the relationship between the test information, and the Adam optimizer can be used to optimize the data analysis model through all vectors in the vector space. When the loss value is the smallest, the data analysis model at this time is used as the test data analysis model.

[0095] It should be noted that in the present application, the bidirectional long short-term memory neural network (i.e., Bi-LSTM) captures the bidirectional dependency of the context by simultaneously learning the forward and reverse sequence information, which can enhance the modeling of the previous and next information in the time series data. Bi-LSTM includes multiple LSTM units, each of which contains a forget gate, an input gate, an output gate, and a memory unit, which is conducive to solving complex dependency problems. Through the memory and forgetting mechanism of Bi-LSTM, entities and relationships in the test data of the medical test form can be converted into numerical vectors that can be processed by the neural network, and the semantic information can be retained.

[0096] In step S5, the inspection data analysis model inspects and analyzes the thrown data based on the subordinate attention matching mechanism, and then obtains the analysis result of the medical inspection form.

[0097] In this embodiment, the inspection data analysis model performs inspection and analysis on the thrown data based on the subordinate attention matching mechanism, and then obtains the analysis result of the medical inspection form. This process can be divided into several specific steps, which are described in detail as follows:

[0098] It should be noted that the subordinate attention matching mechanism is a mechanism that guides the model to pay attention to the importance of data through weight distribution. Through the inspection data analysis model based on the subordinate attention matching mechanism, it is possible to effectively extract key valid information from the medical inspection form, and conduct in-depth analysis of the inspection data based on the knowledge graph and contextual relationships. Among them, the subordinate attention matching mechanism plays a role of weighted allocation of attention in the data analysis process of the medical inspection form, so as to reduce the interference of invalid data on the analysis results, so that the inspection data analysis model can provide reliable analysis results for the medical inspection form through adaptive learning.

[0099] It can be seen that in the present application, an adaptive membership matching analysis can be performed on the inspection information in the medical inspection form; wherein, through the text recognition framework optimized by fuzzy entropy, it is convenient to extract high-value information from the inspection data, and at the same time remove redundant or irrelevant data, so as to obtain the thrown data in the medical inspection form; then, the importance of each descriptor segment in the entire medical inspection form can be automatically identified through the membership matching mechanism, and the attention can be focused on the data that has an important impact on the inspection results, providing an adaptive learning method for the model, which is convenient for adaptively adjusting the extraction weights of different inspection information, and is conducive to improving the matching and recognition capabilities of the inspection data system; finally, a test data analysis model based on a bidirectional long short-term memory neural network can be constructed through the knowledge graph, and combined with the membership attention matching mechanism, the test data analysis model can automatically analyze the thrown data, which is conducive to improving the intelligence level of data processing and ensuring the reliability of the analysis results.

[0100] In summary, the technical solution adopted in this application can perform adaptive subordination matching analysis on the inspection information in the medical inspection form to improve the reliability of medical inspections.

[0101] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0103] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. A method for automatically extracting and analyzing data information of a medical examination form, characterized in that: The extraction and analysis method comprises the following steps: Get the data text information of the medical examination form; Preprocessing the data text information, and semantically segmenting the preprocessed data text information through a text recognition framework to obtain descriptor segments of different verification information, and then determining the fuzzy entropy of the data text information based on discrete verification data in each descriptor segment; Extract the thrown data in the medical examination form based on the text recognition framework optimized by fuzzy entropy, determine the description vector of each description sub-segment, and then determine the subordinate attention matching mechanism of the examination information in the medical examination form through all the description vectors; Obtaining a knowledge graph during medical examination, and constructing a test data analysis model based on a bidirectional long short-term memory neural network according to the knowledge graph; The inspection data analysis model inspects and analyzes the thrown data based on the subordinate attention matching mechanism, and then obtains the analysis result of the medical inspection form.

2. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: The data text information of the medical examination form is obtained by scanning the medical examination form.

3. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: Preprocessing the data text information specifically includes: Determine the structured inspection data to be extracted from the data text information, determine the data type and processing method of the structured inspection data, and then convert the structured inspection data into a plain text format.

4. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: The preprocessed data text information is semantically segmented through the text recognition framework to obtain the descriptive sub-segments of different inspection information, including: Determine the model structure of the text recognition framework; Determine an encoder of a text recognition framework based on the model structure, and perform contextual semantic analysis on the preprocessed data text information through the encoder to obtain a text sequence with a test label; Extract description sub-segments with different test information from text sequences with test labels.

5. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: Determining the fuzzy entropy of the data text information according to the discrete test data in each descriptor segment specifically includes: Extracting discrete test data from each descriptor segment; Standardize all discrete test data to obtain standardized data sequences; The fuzzy entropy of the data text information is determined according to a preset similarity threshold and the standardized data sequence.

6. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: Extracting the thrown data in the medical examination form based on the text recognition framework optimized by fuzzy entropy specifically includes: For each descriptor segment, the approximate entropy of the descriptor segment is determined according to all the test data in the descriptor segment, and then the approximate entropy of each descriptor segment is obtained; Determine the attention allocation result of the text recognition framework through the approximate entropy of each description sub-segment and the fuzzy entropy; The thrown data in the medical examination form is extracted based on the text recognition framework optimized based on the attention allocation result.

7. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: Determining the description vector of each description sub-segment specifically includes: Extracting semantic features, numerical features and structural features from each descriptor segment; Feature fusion and normalization operations are performed on the semantic features, numerical features and structural features of each description sub-segment respectively to obtain a description vector of each description sub-segment.

8. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: The affiliated attention matching mechanism for determining the inspection information in the medical inspection form through all description vectors specifically includes: Performing weighted aggregation on all description vectors to obtain a global description vector of the medical examination form; Determining the membership degree of each description sub-segment based on the global description vector and the description vector of each description sub-segment; The affiliation attention matching mechanism of the inspection information in the medical inspection form is determined by all the affiliations.

9. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: The experimental data of the medical examination are extracted from the database of the medical examination, and then the knowledge graph of the medical examination is obtained based on the pre-trained language model through the experimental data.

10. The method for automatically extracting and analyzing data information of a medical examination form according to claim 1, characterized in that: The text recognition framework is an automated extraction program based on natural language processing.

Images