A medical unstructured data recognition method and system

Abstract

The application relates to a medical unstructured data identification method and system, and relates to the technical field of data identification. The application comprises the following steps: acquiring unstructured data; extracting corresponding evidence of the unstructured data; splicing the evidence to form a scene evidence chain; identifying the unstructured data according to the scene evidence chain and each preset expert credible domain to obtain a data type to which the unstructured data belongs; updating a preset default priority order according to the corresponding evidence to obtain an acceptance priority order; identifying the unstructured data according to the acceptance priority order to obtain an acceptance priority; and selecting a data type corresponding to an expert credible domain with the highest acceptance priority level as a result data type and outputting the result data type. The application has the effects of improving the identification accuracy of unstructured data and expanding the effective range of data identification.

Description

Technical Field

[0001] This invention relates to the field of data recognition technology, and in particular to a method and system for recognizing unstructured medical data. Background Technology

[0002] Hospital systems such as HIS, LIS, PACS, and electronic medical records continuously generate unstructured data, including DICOM images, medical record texts, examination reports, and test data. This data flows frequently across business processes, and the same patient or examination number may appear in multiple images, report texts, and interface fields simultaneously. Therefore, how to manage multi-system, multi-modal sensitive data in hospitals and identify data according to regulations is an important issue that urgently needs to be addressed.

[0003] In related technologies, the method for identifying unstructured medical data usually involves actively scanning the core medical system based on a single model to obtain unstructured data, and then identifying the unstructured data according to file format or keywords.

[0004] Regarding the aforementioned technologies, existing identification methods can only identify unstructured data according to file format or keywords, making it difficult to connect with specific medical scenarios. Therefore, the accuracy of identification is low and the scope is limited, with coarse grading granularity, making it difficult to meet the data identification standards in the medical field. Summary of the Invention

[0005] To improve the accuracy and scope of identification of unstructured data, this invention provides a method and system for identifying medical unstructured data.

[0006] In a first aspect, the present invention provides a method for identifying unstructured medical data, employing the following technical solution: A method for identifying unstructured medical data, comprising: Step 1: Obtain unstructured data; Step 2: Extract evidence corresponding to unstructured data, including source evidence, structural evidence, content evidence, circulation evidence, and ledger evidence; Step 3: Piece together the evidence to form a chain of evidence for the scene; Step 4: Identify the unstructured data based on the scene evidence chain and each preset expert trust domain to obtain the data type to which the unstructured data belongs. The expert trust domain includes the structured expert trust domain, the image expert trust domain, the text expert trust domain, and the strategy expert trust domain. Step 5: Update the preset default priority order based on the corresponding evidence to obtain the acceptance priority order; Step 6: Identify unstructured data according to acceptance priority to obtain acceptance priority; Step 7: Select the data type corresponding to the expert trust domain with the highest acceptance priority as the result data type for output.

[0007] By adopting the above technical solution, evidence is extracted from unstructured data to form an evidence chain, and data type identification is completed based on the scenario evidence chain and expert trust domain. This avoids the identification mismatch problem that occurs when identifying according to file format or keywords, and improves the identification accuracy of unstructured data.

[0008] Optionally, methods for updating the default priority order based on evidence to obtain the accepted priority order include: Step 50: Obtain the specific evidence categories contained in the evidence, where the specific evidence categories refer to the categories of evidence that need to be identified in particular and given higher priority for acceptance; Step 51: Find the corresponding default priority based on the expert's trusted domain; Step 500: When a specific evidence category contains a preset enhancement category, find the corresponding expert confidence domain based on the enhancement category and define the expert confidence domain as the pre-enhanced confidence domain; Step 501: Increase the default priority corresponding to the pre-increased trust domain according to the preset increase rules and output it as the acceptance priority; Step 502: When a specific evidence category contains a preset highest-level category, the corresponding default priority is used as the preset highest acceptance priority and output.

[0009] By adopting the above technical solution, the default priority and acceptance priority were adjusted based on specific evidence categories, thereby achieving the goal of incorporating specific medical business and scenarios into the identification process of unstructured data, improving the reliability of expert trust domain identification, and expanding the effective scope of data identification.

[0010] Optionally, an update method with a default priority is also included, which includes: Step 510: Randomly select an expert trust domain and define that expert trust domain as the current expert trust domain; Step 511: Obtain the recognition quantity and recognition error quantity corresponding to the current expert trust domain within a preset recognition period; Step 512: Calculate the error rate of the current expert trust domain based on the number of recognitions and the number of recognition errors; Step 5120: When the error rate is greater than the preset error threshold, obtain the default priority of the current expert trust domain in the identification, and define the default priority as the error priority; Step 5121: Reduce the error priority according to the preset reduction rules and use it as the new default priority output for the expert's trusted domain.

[0011] By adopting the above technical solution, the default priority of the expert trust domain is adjusted based on the error rate of data identification, thus avoiding the problem of erroneous identification continuing to affect the identification of the expert trust domain and improving the effectiveness of unstructured data identification.

[0012] Optional, also includes: Step 5122: When the error rate is less than the preset error threshold, obtain the current expert trust domain of one recognition period as the highest priority number of times the highest priority level is adopted; Step 5123: When the number of times the highest priority is exceeded is greater than the preset selection threshold, obtain the most recent historical priority of the current expert trust domain; Step 5124: Reduce the recent historical priority according to the reduction rules and output it as the new default priority of the current expert trust domain.

[0013] By adopting the above technical solution, the default priority is adjusted by obtaining the highest priority number of times, thus maintaining the balance of recognition capabilities among expert trust domains and solving the problem of a single expert trust domain monopolizing the recognition results.

[0014] Optionally, the method for selecting expert trust domains in the step of identifying unstructured data based on the scenario evidence chain and various preset expert trust domains includes: Step 40: Obtain historical unstructured data and the historical evidence corresponding to the historical unstructured data; Step 41: Based on historical evidence configuration, form an evidence feature library corresponding to the expert trust domain; Step 42: Calculate the matching score between the scene evidence chain and the corresponding expert credibility domain based on the evidence feature library; Step 420: When the matching score is less than the preset admission threshold, the expert trust domain corresponding to the matching score is defined as the low-score expert trust domain. Step 421: After removing the low-scoring expert trust domain from the expert trust domain, proceed with step 5 and subsequent steps; Step 422: When the matching score is greater than the admission threshold, proceed directly to step 5 and subsequent steps.

[0015] By adopting the above technical solution, and through the acquisition of historical evidence and the calculation of matching scores, the determination of valid expert trust domains is realized, avoiding the waste of resources in data identification by invalid expert trust domains and improving the efficiency of data identification.

[0016] Optionally, the analysis method for calculating the matching score between the scene's evidence chain and the corresponding expert credibility domain also includes: Step 423: Calculate the difference between the matching scores of adjacent expert trust domains after removing the low-scoring expert trust domains according to the default priority order, and define the difference as the adjacent difference. Step 4230: If all adjacent differences are greater than the preset minimum threshold, the default priority order is reordered according to the descending order of the corresponding matching scores, and the default priority order is updated and output. Step 4231: If there is an adjacent difference less than the minimum threshold, output according to the default priority order.

[0017] By adopting the above technical solution, the default priority was reordered through the calculation of adjacent differences, which improved the correlation between the expert trust domain and the scene evidence chain and ensured the effectiveness of the expert trust domain in data identification.

[0018] Optional, also includes: Step 4210: If there exists an adjacent difference less than the minimum threshold, then define the two expert confidence domains corresponding to the difference as the minimum difference expert confidence domain, and obtain the historical evidence type and total number of recognitions of the minimum difference expert confidence domain within a recognition period; Step 4211: Count the number of times each type of historical evidence was identified; Step 4212: Select the historical evidence type with the highest number of recognitions and calculate the recognition frequency of the minimal difference expert confidence region; Step 42120: If the recognition frequency is greater than the preset high recognition threshold, the default priority corresponding to the minimal difference expert confidence domain is increased according to the increase rule and then output as the new default priority. Step 42121: If the recognition frequency is less than the high recognition threshold, output according to the default priority order.

[0019] By adopting the above technical solution, the default priority of the expert trust domain is adjusted by calculating the recognition frequency, which improves the recognition efficiency of the expert trust domain for a certain data type and achieves the effect of improving the accuracy of data recognition.

[0020] Optionally, specific methods for extracting evidence from unstructured data include: Step 20: If at least one piece of evidence is missing, then the missing piece of evidence is defined as missing evidence; Step 21: Retrieve historical evidence of the same type as the missing evidence as supplementary evidence; Step 22: Based on the supplementary evidence and the original non-missing evidence, form a supplementary evidence chain, update the supplementary evidence chain to the scene evidence chain, and continue to execute Step 4 and subsequent operations.

[0021] By adopting the above technical solution, missing evidence was supplemented by retrieving historical evidence, ensuring the continuity of the evidence chain in the scene and avoiding the problem that the expert's credibility domain would be unable to identify unstructured data due to missing evidence.

[0022] Optional optimization methods for prioritizing acceptance include: Step 60: When there are at least two expert trust domains with the same level of acceptance priority, sort the acceptance priorities with the same level according to the industry data classification list and update the acceptance priorities. Step 61: Based on the acceptance priority after sequence sorting, select the data type corresponding to the expert trust domain with the highest acceptance priority level as the result data type for output.

[0023] By adopting the above technical solution, the acceptance priority of expert trust domains is sorted by industry data classification list, which solves the problem of conflicting identification results of multiple expert trust domains due to the same acceptance priority, and improves the accuracy of data identification.

[0024] Secondly, this invention provides a medical unstructured data identification system, which adopts the following technical solution: A medical unstructured data identification system, comprising: The acquisition module is used to acquire unstructured data, specific evidence categories, historical unstructured data, historical evidence, types of historical evidence, recognition count, number of recognition errors, number of highest priority recognitions, and total number of recognitions. A memory for storing a program for a medical unstructured data identification method as described above; The processor loads and executes programs from memory.

[0025] By adopting the above technical solution, the acquisition and calculation of the data required for the identification of unstructured data based on expert trust domains are realized, achieving the goal of accurate identification of unstructured data in medical scenarios through expert trust domains.

[0026] In summary, the present invention has at least one of the following beneficial technical effects: 1. By extracting evidence from unstructured data to form a chain of evidence, and based on the scenario evidence chain and expert trust domain, the data type identification was completed, avoiding the identification mismatch problem that occurs when identifying according to file format or keywords, and improving the identification accuracy of unstructured data; 2. Based on specific evidence categories, the default priority and acceptance priority were adjusted, achieving the goal of incorporating specific medical services and scenarios into the identification process of unstructured data, thus expanding the effective scope of data identification; 3. It realizes the acquisition and calculation of the data required for the identification of unstructured data based on expert trust domain, and achieves the goal of accurate identification of unstructured data in medical scenarios through expert trust domain. Attached Figure Description

[0027] Figure 1 This is a flowchart of a method for identifying unstructured medical data.

[0028] Figure 2 This is a flowchart of a method for updating the default priority order based on evidence to obtain the accepted priority order.

[0029] Figure 3 This is a flowchart of the update method with the default priority.

[0030] Figure 4 This is a schematic diagram of a medical unstructured data recognition system. Detailed Implementation

[0031] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0032] This invention discloses a method for identifying unstructured medical data. (Refer to...) Figure 1 A method for identifying unstructured medical data includes: Step 1: Obtain unstructured data.

[0033] Unstructured data is a collective term for data generated during the operation of a medical system, such as DICOM images, medical record texts, examination reports, and test data.

[0034] Unstructured data is acquired by collecting data through the mirror port of the authorized medical system's switch.

[0035] Step 2: Extract evidence corresponding to unstructured data.

[0036] Evidence refers to the source evidence, structural evidence, content evidence, flow evidence, and ledger evidence of unstructured data. Source evidence includes information such as image gateways, inspection interfaces, and research export interfaces; structural evidence includes DICOM tags, JSON / XML key-value pairs, and interface fields; flow evidence includes image uploading, inspection data synchronization, and structured file export; content evidence includes inspection serial numbers, equipment numbers, and inspection order numbers; and ledger evidence includes asset registration, interface authorization, and unique asset identifiers.

[0037] Unstructured data, including traffic messages and ledger records, is broken down to extract entity text, serial numbers, diagnoses, and patient identifiers. This information is then extracted to obtain individual pieces of evidence. For example, label text is extracted from DICOM images to form structured evidence, and the report source is extracted from examination reports to form source evidence.

[0038] Step 3: Piece together the evidence to form a chain of evidence for the scene.

[0039] A scenario-based evidence chain is a complete and ordered combination of evidence corresponding to unstructured data, arranged according to the sequence of medical procedures.

[0040] Multiple individual pieces of evidence are combined in the order of medical procedures to form a chain of evidence for a given scenario. The order of medical procedures refers to the sequence of processes a patient goes through when receiving medical services. For example, the order in which a patient initiates an examination request, the equipment acquires images, the report is generated, the physician writes the electronic medical record, and the patient reviews the record within the hospital constitutes the order of medical procedures.

[0041] Step 4: Identify the unstructured data based on the scenario evidence chain and each preset expert trust domain to determine the data type to which the unstructured data belongs.

[0042] Expert trust domains are multiple domains specifically designed for handling images, structured fields, natural language text, and compliance policies in various scenarios. The recognition result of an expert trust domain has higher priority only when the scenario's evidence chain matches the corresponding expert trust domain. Expert trust domains include structured expert trust domains, image expert trust domains, text expert trust domains, and policy expert trust domains. Specifically, structured expert trust domains are suitable for parsing scenarios such as DICOM tags and JSON / XML key-value pairs; image expert trust domains are suitable for recognizing scanned documents such as DICOM images and PDF reports; text expert trust domains are suitable for recognizing text related to medical record numbers, patient identification numbers, surgery numbers, and drug numbers; and policy expert trust domains are suitable for recognizing scenarios such as hierarchical compliance, authorization boundary determination, and in-hospital asset authorization.

[0043] Data type refers to a specific type of unstructured data. Standard evidence samples are used to pre-define the data type of each expert's trust domain and store them in memory. Each expert's trust domain, based on its own business scenario, independently outputs a data type judgment result for the unstructured data within the evidence chain. For example, for a digital text document, the structured expert trust domain identifies the data type as a DICOM image source, while the text expert trust domain identifies the data type as a hospital admission number within a medical record.

[0044] Step 5: Update the preset default priority order based on the corresponding evidence to obtain the acceptance priority order.

[0045] Default priority refers to the pre-configured ranking of the results of data type identification by each expert's trusted domain. This is pre-set in the program by personnel in the field based on their own experience and industry data grading rules.

[0046] The acceptance priority order is the ranking of the levels corresponding to the acceptance priority. The acceptance priority is the level of the results of each expert's confidence domain in identifying the data type after adjusting the default priority based on the evidence in the scenario's evidence chain.

[0047] Different expert trust domains identify unstructured data and obtain different data types; according to the scenario evidence chain, these data types correspond to different levels, and these levels are sorted in descending order to replace the default priority and thus update the acceptance priority.

[0048] Step 6: Identify unstructured data according to acceptance priority to obtain acceptance priority.

[0049] Different levels of acceptance priority are obtained by assigning ascending numbers to the acceptance priority order corresponding to the data types of unstructured data.

[0050] Step 7: Select the data type corresponding to the expert trust domain with the highest acceptance priority as the result data type for output.

[0051] The result data type refers to the data type of the unstructured data that ultimately serves as the output. During the selection process, the highest-level expert trust domain is determined based on the order of trust levels. Then, the data type identified by that expert trust domain is searched to obtain the result data type.

[0052] Taking the identification of medical record text data as an example, the default priority ranking of the expert trust domains is set in descending order of level: Structured Expert Trust Domain Level 1, Text Expert Trust Domain Level 2, Image Expert Trust Domain Level 3, and Strategy Expert Trust Domain Level 4. After extracting evidence corresponding to the medical record text data to form a scenario evidence chain, the expert trust domains identify the data type: the Structured Expert Trust Domain identifies it as the image sequence number, the Text Expert Trust Domain identifies it as the medical record number in the electronic medical record export scenario, and the Strategy Expert Trust Domain identifies it as the invoice number or settlement number. The updated acceptance priorities are ranked from high to low as Text Expert Trust Domain Level 1, Structured Expert Trust Domain Level 2, Strategy Expert Trust Domain Level 3, and Image Expert Trust Domain Level 4. The data type "medical record number" identified by the highest-level Text Expert Trust Domain is selected as the output data type.

[0053] Reference Figure 2 Methods for updating the default priority order based on evidence to obtain the accepted priority order include: Step 50: Obtain the specific categories of evidence contained within the evidence.

[0054] Specific evidence category refers to the specific category of evidence that needs to be identified and prioritized for acceptance. Different pieces of evidence correspond to unstructured data with different specific data content. The specific evidence category is obtained based on the specific data content. For example, if specific image exchange data appears in the source evidence, then the specific evidence category of image exchange source evidence is selected.

[0055] Step 51: Find the corresponding default priority based on the expert trust domain.

[0056] The search is performed using the default priority corresponding to each expert trust domain that is pre-set in memory.

[0057] Step 500: When a specific evidence category contains a preset enhancement category, find the corresponding expert credibility domain based on the enhancement category and define the expert credibility domain as a pre-enhanced credibility domain.

[0058] The "upgrade category" refers to the specific evidence category that needs to be upgraded from the default priority level. It is pre-defined by people in the field based on their own experience and industry data classification rules.

[0059] When searching for the corresponding expert trust domain, the search is performed according to the category to which the specific evidence belongs, as shown in step 4.

[0060] Step 501: Increase the default priority corresponding to the pre-increased trust domain according to the preset increase rules and output it as the acceptance priority.

[0061] The enhancement rule is a rule that raises the default priority corresponding to the expert's trust domain by one level as the acceptance priority. This is pre-programmed in memory by personnel in the field based on their experience and industry data grading rules.

[0062] If the specific evidence category includes enhanced categories such as image exchange source evidence and DICOM tag structure evidence, then according to steps 50 to 501, the default priorities corresponding to the structured expert trust domain and the image expert trust domain are each increased by one level and output as the acceptance priority. When the specific evidence category includes the enhanced category of medical record export and transfer evidence, then the default priority of the text expert trust domain is increased by one level and output as the acceptance priority.

[0063] Step 502: When a specific evidence category contains a preset highest-level category, the corresponding default priority is used as the preset highest acceptance priority and output.

[0064] The highest-level category refers to the specific evidence category whose default priority needs to be raised to the highest level. The preset method is similar to that of raising the category. The highest acceptance priority refers to the highest level of acceptance priority, which is preset procedurally by those in the field based on their own experience.

[0065] If the highest-level category of evidence is outpatient settlement source evidence, the default priority corresponding to the strategy expert credibility domain will be used as the highest acceptance priority and output.

[0066] Reference Figure 3 The default priority update methods include: Step 510: Randomly select one of the expert trust domains and define that expert trust domain as the current expert trust domain; Step 511: Obtain the recognition quantity and recognition error quantity corresponding to the current expert trust domain within a preset recognition period.

[0067] The identification cycle refers to the period between two unstructured data identifications selected according to a specific time interval. Within an identification cycle, there are multiple completed data identification processes. The identification cycle is pre-programmed by personnel in the field based on their own experience.

[0068] The recognition count refers to the total amount of unstructured data recognized by the current expert trusted domain within a recognition period. The recognition error count is the number of recognition errors that occurred in the current expert trusted domain during the recognition of unstructured data within a recognition period. The recognition count and recognition error count are obtained by collecting past unstructured data recognition data through a mirror port of the medical system.

[0069] Step 512: Calculate the error rate of the current expert trust domain based on the number of recognitions and the number of recognition errors.

[0070] The error rate is the ratio of the number of identification errors to the total number of identifications made during the data recognition process within the current expert trust domain. The formula for calculating the error rate is as follows: ,in v e For error rate, N s For recognition volume, N e To identify the number of errors.

[0071] Step 5120: When the error rate is greater than the preset error threshold, obtain the default priority of the current expert trust domain in the identification and define the default priority as the error priority.

[0072] The error threshold is the critical value for determining that the error rate of the expert trust domain identification data is too high. The error threshold is obtained by pre-testing the expert trust domain with standard evidence samples and is pre-set in memory.

[0073] The method for obtaining the default priority is as described in step 51, and will not be repeated here.

[0074] Step 5121: Reduce the error priority according to the preset reduction rules and use it as the new default priority output for the expert's trusted domain.

[0075] The rule of reducing the default priority is to lower the default priority by one level and set it as the new default priority. The default method is similar to step 501.

[0076] If the error rate is greater than the error threshold, it means that the current expert trust domain has a relatively high error rate in data identification. In this case, the default priority corresponding to the current expert trust domain will be lowered by one level.

[0077] This also includes: Step 5122: When the error rate is less than the preset error threshold, obtain the highest priority count corresponding to the current expert trust domain within a recognition period.

[0078] The highest priority count refers to the number of times the current expert's trusted domain has the highest acceptance priority level within an identification period. The method of obtaining this information is similar to step 511.

[0079] Step 5123: When the number of times the highest priority is exceeded is greater than the preset selection threshold, obtain the most recent historical priority of the current expert trust domain.

[0080] The selection threshold is the critical value at which the highest priority corresponding to the expert trust domain is judged too many times. The selection threshold is obtained by pre-identifying and testing the expert trust domain through standard evidence samples and is pre-set in memory.

[0081] Recent historical priority refers to the most recent recognition priority in the current expert trusted domain recognition history, which is the most recent recognition time in the current data recognition process. The acquisition process is similar to step 51.

[0082] Step 5124: Reduce the recent historical priority according to the reduction rules and output it as the new default priority of the current expert trust domain.

[0083] If the highest priority number of times exceeds the selection threshold, the default priority of the current expert trust domain in the next data identification will be reduced, thus avoiding a single expert trust domain from monopolizing the identification results.

[0084] The method for selecting expert trust domains in the step of identifying unstructured data based on the scenario evidence chain and various preset expert trust domains includes: Step 40: Obtain historical unstructured data and the historical evidence corresponding to the historical unstructured data.

[0085] Historical unstructured data refers to unstructured data that has been identified through expert trust domain analysis. Historical evidence is evidence corresponding to historical unstructured data. The methods for obtaining historical unstructured data and historical evidence are similar to those in step 1, and will not be repeated here.

[0086] Step 41: Based on historical evidence configuration, form an evidence feature library corresponding to the expert trust domain.

[0087] An evidence feature library refers to a collection of evidence types used to quantify matching scores, bound to each category of expert trust domains. This library is configured and formed by statistically analyzing and classifying historical evidence. Evidence type refers to the category to which the evidence corresponding to unstructured data belongs.

[0088] Step 42: Calculate the matching score between the scene evidence chain and the corresponding expert credibility domain based on the evidence feature library.

[0089] The matching score refers to the degree of matching between the scene's chain of evidence and the corresponding expert confidence domain. The formula for calculating the matching score is as follows: ,in Score To match scores, f x For each type of evidence in the scene's chain of evidence, f x satisfy , w x For each type of evidence, assign a matching weight to the corresponding expert credibility domain, and w x satisfy .

[0090] Step 420: When the matching score is less than the preset admission threshold, the expert trust domain corresponding to the matching score is defined as the low-score expert trust domain.

[0091] The admission threshold is the critical value at which the matching degree between the scenario evidence chain and the corresponding expert trust domain is low. It is pre-set in the program based on the matching results of the scenario evidence chain and the corresponding expert trust domain in the past.

[0092] Taking the text expert trust domain as an example, assuming the admission threshold is 60, the matching weights for source evidence, structural evidence, content evidence, circulation evidence, and ledger evidence are 25, 15, 35, 15, and 10, respectively. Since DICOM tags and asset registrations were not extracted from the scene evidence chain, structural evidence and ledger evidence are absent. Furthermore, the extracted image gateway and evidence feature library do not correspond, resulting in a matching score of [insert score here]. .

[0093] Step 421: After removing the low-scoring expert trust domain from the expert trust domain, proceed with step 5 and subsequent steps.

[0094] If the matching score is less than the admission threshold, it indicates a low degree of matching between the scene's evidence chain and the corresponding expert trust domain. Continuing to use this expert trust domain for data identification may lead to identification errors. In this case, the low-scoring expert trust domain is removed from the expert trust domain set before proceeding to step 5 and subsequent steps. The process of removing the low-scoring expert trust domain involves removing it from the set of expert trust domains participating in the current round of unstructured data identification. This removal process only takes effect in the current round of data identification.

[0095] Step 422: When the matching score is greater than the admission threshold, proceed directly to step 5 and subsequent steps.

[0096] If the matching score is greater than the admission threshold, it means that the matching degree between the scene evidence chain and the corresponding expert trust domain meets the recognition requirements. At this time, step 5 and subsequent steps are executed directly.

[0097] The analysis method for calculating the matching score between the scene's evidence chain and the corresponding expert credibility domain also includes: Step 423: Calculate the difference between the matching scores of adjacent expert confidence domains after removing the low-scoring expert confidence domains according to the default priority order, and define the difference as the adjacent difference.

[0098] When calculating the difference between adjacent values, the formula is as follows: , where Δ i Score represents the difference between adjacent values. i The Score is the matching score corresponding to the expert's trust domain. i+1 This is the matching score for the next adjacent expert trust domain according to the default priority order.

[0099] Step 4230: If all adjacent differences are greater than the preset minimum threshold, the default priority order is reordered according to the descending order of the corresponding matching scores, and the default priority order is updated for output.

[0100] The minimum threshold represents the critical value at which the matching scores of adjacent expert confidence domains are similar. It is pre-set in the program by the personnel in that field based on their own experience and past matching scores.

[0101] If all adjacent differences are greater than the preset minimum threshold, it means that the matching scores corresponding to all adjacent expert trust domains are not similar. At this time, the default priorities corresponding to the expert trust domains are reordered in descending order of matching scores. The reordered default priority order is the updated default priority order.

[0102] Step 4231: If there is an adjacent difference less than the minimum threshold, output according to the default priority order.

[0103] If there is an adjacent difference less than the minimum threshold, it indicates that there are cases where the matching scores corresponding to adjacent expert trust domains are similar. In this case, the corresponding expert trust domains are sorted according to the default priority order.

[0104] This also includes: Step 4210: If there exists an adjacent difference less than the minimum threshold, then define the two expert confidence domains corresponding to the difference as the minimum difference expert confidence domain, and obtain the historical evidence type and total number of recognitions of the minimum difference expert confidence domain within a recognition period.

[0105] Historical evidence type refers to the type of evidence corresponding to unstructured data identified by the minimal difference expert trust domain within an identification period. Total identification count is the number of times unstructured data is identified by the minimal difference expert trust domain within an identification period. The acquisition method is similar to step 511.

[0106] Step 4211: Count the number of times each type of historical evidence is identified.

[0107] The number of recognitions refers to the number of times a specific type of historical evidence is recognized within a recognition period by the expert confidence domain with minimal difference. Statistics are generated by summarizing the recognition counts for each type of historical evidence from memory.

[0108] Step 4212: Select the historical evidence type with the highest number of recognitions and calculate the recognition frequency of the minimal difference expert confidence region.

[0109] The number of times each type of historical evidence is identified is filtered to select the type of historical evidence with the highest number of identifications.

[0110] Recognition frequency refers to the ratio of the most frequently recognized historical evidence type in the expert confidence domain to the total number of recognitions. It is calculated by the ratio of the number of recognitions corresponding to the most frequently recognized historical evidence type to the total number of recognitions.

[0111] Step 42120: If the recognition frequency is greater than the preset high recognition threshold, the default priority corresponding to the minimal difference expert confidence domain is increased according to the increase rule and then output as the new default priority.

[0112] The high recognition threshold is a critical value used to determine the highest recognition frequency of the historical evidence type with the most recognitions by the expert credibility domain with the smallest difference. The expert credibility domain is pre-tested using standard evidence samples to obtain the recognition frequency of each evidence type, which is then pre-set in memory by the domain personnel based on their own experience.

[0113] If the recognition frequency is greater than the high recognition threshold, it indicates that the expert trust domain with the smallest difference has a stronger ability to recognize the historical evidence type with the most recognition frequency. In this case, the default priority level corresponding to the expert trust domain should be increased.

[0114] Step 42121: If the recognition frequency is less than the high recognition threshold, output according to the default priority order.

[0115] If the recognition frequency is less than the high recognition threshold, it indicates that the minimal difference expert confidence domain has poor recognition ability for the historical evidence type with the most recognition frequency. In this case, the corresponding minimal difference expert confidence domains are sorted according to the default priority order.

[0116] The specific methods for extracting evidence from unstructured data include: Step 20: If at least one piece of evidence is missing, then the missing piece of evidence is defined as missing evidence.

[0117] Step 21: Retrieve historical evidence of the same type as the missing evidence as supplementary evidence.

[0118] Supplementary evidence refers to historical evidence of the same type used to supplement missing evidence.

[0119] When retrieving historical evidence, the type of evidence corresponding to the missing evidence is searched, and historical evidence of the same type is determined based on the type of evidence.

[0120] Step 22: Based on the supplementary evidence and the original non-missing evidence, form a supplementary evidence chain, update the supplementary evidence chain to the scene evidence chain, and continue to execute Step 4 and subsequent operations.

[0121] The method for piecing together a complete chain of evidence is similar to step 3, and will not be elaborated here. The missing evidence in the original chain of scene evidence is replaced to update the scene evidence chain.

[0122] Among them, the optimization methods for acceptance priority include: Step 60: When there are at least two expert trust domains with the same level of acceptance priority, sort the acceptance priorities with the same level according to the industry data classification list and update the acceptance priorities.

[0123] The industry data grading list is a grading list that is pre-stored in memory according to data classification and grading rules and the level of medical data.

[0124] When at least two expert trust domains have the same level of acceptance priority, the original acceptance priorities with the same level are reordered through the industry data classification list, and the priority of the reordered expert trust domain is updated as the new acceptance priority.

[0125] Step 61: Based on the acceptance priority after sequence sorting, output the data type corresponding to the expert trust domain with the highest acceptance priority level as the result data type.

[0126] After sorting the acceptance priorities of the same level according to the industry data classification list, the acceptance priorities of all expert trust domains are sorted in descending order and there are no identical levels. At this time, the data type corresponding to the expert trust domain with the highest acceptance priority level is used as the result data type for output.

[0127] Based on the same inventive concept, embodiments of the present invention provide a medical unstructured data identification system.

[0128] Reference Figure 4 A medical unstructured data identification system, comprising: The acquisition module is used to acquire unstructured data, specific evidence categories, historical unstructured data, historical evidence, types of historical evidence, recognition count, number of recognition errors, number of highest priority recognitions, and total number of recognitions. A memory for storing a program for a medical unstructured data identification method as described above.

[0129] The processor loads and executes programs from memory.

[0130] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0131] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying unstructured medical data, characterized in that, include: Step 1: Obtain unstructured data; Step 2: Extract evidence corresponding to unstructured data, including source evidence, structural evidence, content evidence, circulation evidence, and ledger evidence; Step 3: Piece together the evidence to form a chain of evidence for the scene; Step 4: Identify the unstructured data based on the scene evidence chain and each preset expert trust domain to obtain the data type to which the unstructured data belongs. The expert trust domain includes the structured expert trust domain, the image expert trust domain, the text expert trust domain, and the strategy expert trust domain. Step 5: Update the preset default priority order based on the corresponding evidence to obtain the acceptance priority order; Step 6: Identify unstructured data according to acceptance priority to obtain acceptance priority; Step 7: Select the data type corresponding to the expert trust domain with the highest acceptance priority as the result data type for output.

2. The method for identifying unstructured medical data according to claim 1, characterized in that, Methods for updating the default priority order based on evidence to obtain the accepted priority order include: Step 50: Obtain the specific evidence categories contained in the evidence, where the specific evidence categories refer to the categories of evidence that need to be identified in particular and given higher priority for acceptance; Step 51: Find the corresponding default priority based on the expert's trusted domain; Step 500: When a specific evidence category contains a preset enhancement category, find the corresponding expert confidence domain based on the enhancement category and define the expert confidence domain as the pre-enhanced confidence domain; Step 501: Increase the default priority corresponding to the pre-increased trust domain according to the preset increase rules and output it as the acceptance priority; Step 502: When a specific evidence category contains a preset highest-level category, the corresponding default priority is used as the preset highest acceptance priority and output.

3. The method for identifying unstructured medical data according to claim 2, characterized in that, It also includes an update method with default priority, which includes: Step 510: Randomly select an expert trust domain and define that expert trust domain as the current expert trust domain; Step 511: Obtain the recognition quantity and recognition error quantity corresponding to the current expert trust domain within a preset recognition period; Step 512: Calculate the error rate of the current expert trust domain based on the number of recognitions and the number of recognition errors; Step 5120: When the error rate is greater than the preset error threshold, obtain the default priority of the current expert trust domain in the identification, and define the default priority as the error priority; Step 5121: Reduce the error priority according to the preset reduction rules and use it as the new default priority output for the expert's trusted domain.

4. The method for identifying unstructured medical data according to claim 3, characterized in that, Also includes: Step 5122: When the error rate is less than the preset error threshold, obtain the current expert trust domain of one recognition period as the highest priority number of times the highest priority level is adopted; Step 5123: When the number of times the highest priority is exceeded is greater than the preset selection threshold, obtain the most recent historical priority of the current expert trust domain; Step 5124: Reduce the recent historical priority according to the reduction rules and output it as the new default priority of the current expert trust domain.

5. The method for identifying unstructured medical data according to claim 1, characterized in that, The method for selecting expert trust domains in the step of identifying unstructured data based on the scenario evidence chain and various preset expert trust domains includes: Step 40: Obtain historical unstructured data and the historical evidence corresponding to the historical unstructured data; Step 41: Based on historical evidence configuration, form an evidence feature library corresponding to the expert trust domain; Step 42: Calculate the matching score between the scene evidence chain and the corresponding expert credibility domain based on the evidence feature library; Step 420: When the matching score is less than the preset admission threshold, the expert trust domain corresponding to the matching score is defined as the low-score expert trust domain. Step 421: After removing the low-scoring expert trust domain from the expert trust domain, proceed with step 5 and subsequent steps; Step 422: When the matching score is greater than the admission threshold, proceed directly to step 5 and subsequent steps.

6. The method for identifying unstructured medical data according to claim 5, characterized in that, The analytical methods for calculating the matching scores between the scene's chain of evidence and the corresponding expert credibility domain also include: Step 423: Calculate the difference between the matching scores of adjacent expert trust domains after removing the low-scoring expert trust domains according to the default priority order, and define the difference as the adjacent difference. Step 4230: If all adjacent differences are greater than the preset minimum threshold, the default priority order is reordered according to the descending order of the corresponding matching scores, and the default priority order is updated and output. Step 4231: If there is an adjacent difference less than the minimum threshold, output according to the default priority order.

7. The method for identifying unstructured medical data according to claim 6, characterized in that, Also includes: Step 4210: If there exists an adjacent difference less than the minimum threshold, then define the two expert confidence domains corresponding to the difference as the minimum difference expert confidence domain, and obtain the historical evidence type and total number of recognitions of the minimum difference expert confidence domain within a recognition period; Step 4211: Count the number of times each type of historical evidence was identified; Step 4212: Select the historical evidence type with the highest number of recognitions and calculate the recognition frequency of the minimal difference expert confidence region; Step 42120: If the recognition frequency is greater than the preset high recognition threshold, the default priority corresponding to the minimal difference expert confidence domain is increased according to the increase rule and then output as the new default priority. Step 42121: If the recognition frequency is less than the high recognition threshold, output according to the default priority order.

8. The method for identifying unstructured medical data according to claim 1, characterized in that, Specific methods for extracting evidence from unstructured data include: Step 20: If at least one piece of evidence is missing, then the missing piece of evidence is defined as missing evidence; Step 21: Retrieve historical evidence of the same type as the missing evidence as supplementary evidence; Step 22: Based on the supplementary evidence and the original non-missing evidence, form a supplementary evidence chain, update the supplementary evidence chain to the scene evidence chain, and continue to execute Step 4 and subsequent operations.

9. A method for identifying unstructured medical data according to claim 1, characterized in that, Optimization methods for prioritizing acceptance include: Step 60: When there are at least two expert trust domains with the same level of acceptance priority, sort the acceptance priorities with the same level according to the industry data classification list and update the acceptance priorities. Step 61: Based on the acceptance priority after sequence sorting, select the data type corresponding to the expert trust domain with the highest acceptance priority level as the result data type for output.

10. A medical unstructured data identification system, characterized in that, include: The acquisition module is used to acquire unstructured data, specific evidence categories, historical unstructured data, historical evidence, types of historical evidence, recognition count, number of recognition errors, number of highest priority recognitions, and total number of recognitions. A memory for storing a program for a medical unstructured data identification method as described in any one of claims 1 to 9; The processor loads and executes programs from memory.

Images