A method, system, device and computer-readable medium for analyzing detailed items

Abstract

The present invention relates to the field of natural language processing, and specifically provides a method for parsing detailed items of medical records of special disease texts, which has the following steps: S1, first obtain the text data related to the special disease that needs to be parsed, and perform hierarchical parsing; S2, disassemble the obtained text data in the form of event streams; S3, according to the information that needs to be parsed, select the content category event stream containing the information in a targeted manner, and further perform detailed structural parsing of various indicators. Compared with the prior art, the present invention processes the data step by step, first divides the event stream by time node and content category, and then performs process parsing of specific detailed items. Compared with directly processing large sections of text, the accuracy of the parsing results is improved, and it is easier to perform error troubleshooting, method optimization and system maintenance.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing, and specifically provides a detailed item parsing method, system, device and computer-readable medium. Background Art

[0002] In the process of building specialized diseases, the electronic medical record data involved are mostly large pieces of unstructured text data, which contain pathological characteristics, immunohistochemistry, genetic testing and many other contents. Doctors or researchers often need to quickly and accurately obtain valuable information from them. Therefore, they need to read, split and perform structured processing on such data to obtain the corresponding detailed indicator data.

[0003] With the rise of natural language processing technology, technologies such as named entity recognition and entity relationship extraction have been successively applied to electronic medical record parsing. However, due to the high complexity and diversity of medical record content, especially the factors involving multiple time nodes and non-uniform standardization of data content in different regions, it is difficult for existing technologies to accurately extract correct parsing results, and there is a lack of a complete set of standardized parsing processes and methods. Summary of the invention

[0004] The present invention aims at the deficiencies of the above-mentioned prior art and provides a method for parsing detailed items of medical records of specific diseases with strong practicality.

[0005] A further technical task of the present invention is to provide a reasonably designed, safe and applicable system for parsing detailed items in medical records of specific diseases.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] A method for parsing detailed items of medical records of specific diseases, comprising the following steps:

[0008] S1. First, obtain the text data related to the specific disease to be analyzed and perform hierarchical analysis;

[0009] S2, disassemble the acquired text data in the form of event stream;

[0010] S3. According to the information that needs to be analyzed, the content category event flow containing the information is selected in a targeted manner, and further detailed structured analysis of each indicator is performed.

[0011] Furthermore, in step S2, it further includes:

[0012] S201, splitting according to text content;

[0013] S202, splitting sentences according to symbols and roughly segmenting the text;

[0014] S203, splitting the event stream according to time nodes.

[0015] Furthermore, in step S201, keyword matching is used to match keywords from the text, and the text content is segmented in the sentences at the corresponding positions to distinguish each content in the data, and each part of the content is labeled with a category name, and the name of the category to which each text segment belongs is stored in a corresponding column field;

[0016] In step S202, the sentences are segmented according to the symbols to form a plurality of complete sentences;

[0017] In step S203, if there is a time mark in the current sentence, it is an event stream. If there is no time mark in the current sentence, it is added to the previous sentence with a time mark, and the time is obtained based on the standard time type first;

[0018] After each statement is split, the medical record content at a certain time node is obtained. Then, the event stream is sorted according to the word order of the original text, and the serial number corresponding to each event stream is saved in the corresponding column field.

[0019] Furthermore, in step S3, the following steps are further included:

[0020] S301, analysis of pathological characteristics;

[0021] S302, analysis of treatment plan;

[0022] S303, detailed analysis of surgical information;

[0023] S304. Analysis of radiotherapy plan.

[0024] Further, in step S301, the pathological characteristic analysis first selects the category as event flow, then extracts the event flow with pathological content according to the relevant keywords, and analyzes the extracted event popular item fields;

[0025] According to the different characteristics of different information items, the symbols are first segmented to extract the small sentences containing the target data from the event stream, and then the appropriate method is selected to extract the content of each field;

[0026] In step S302, each event stream is first split into small sentences according to punctuation marks, and then the event stream is segmented again according to time. Each small sentence segmented corresponds to only one time node, and then each field item is analyzed in detail.

[0027] Further, in step S303, according to data analysis, the surgical information parsing first selects the event stream of the category, and then divides the event stream into small sentences according to time; according to the surgical dictionary table, the small sentences with the surgical name are extracted as surgical descriptions, and detailed analysis is performed in the surgical descriptions;

[0028] In step S304, there is generally only one radiotherapy plan in a data stream. Therefore, in the analysis of the radiotherapy plan, the data stream is not further split into sentences. Keywords with "radiotherapy" but not radiotherapy plans are replaced, and then each field is analyzed in detail.

[0029] A detailed item parsing system for medical records of special diseases includes a pathological characteristics parsing module, a treatment plan parsing module, a surgical information parsing module and a radiotherapy plan parsing module. The data is first parsed in the pathological characteristics parsing module, then the treatment parsing module is entered for data treatment parsing, the surgical information parsing module is entered for detailed item parsing of specific contents, and finally the radiotherapy plan parsing module is entered for parsing.

[0030] Furthermore, in the pathological characteristics analysis module, the pathological characteristics-related content data in the event stream storage table is obtained, and the analyzed pathological examination date, pathological classification, tumor location, metastasis location, TNM stage, immunohistochemistry details, genetic testing details, and lymph node status details are structured and displayed in a table format;

[0031] The parsed drug name, treatment plan name, treatment cycle, plan start time, medication start time, medication end time, medication efficacy evaluation time, medication efficacy evaluation results, medication efficacy evaluation measures, plan category, and medication route are displayed in a structured manner in a table;

[0032] Create a surgery dictionary, which can be a specific surgery name or a surgery name of particular interest. Use the dictionary to match the event stream, select the event stream data containing the specific surgery name, and present the parsed surgery name, surgery time, and anesthesia method in a structured manner in a table.

[0033] The parsed radiotherapy technology name, radiotherapy start time, radiotherapy efficacy evaluation time, radiotherapy efficacy evaluation results, and radiotherapy efficacy evaluation measures are displayed in a structured manner in a table.

[0034] A device for parsing detailed items of medical records of special disease texts first performs hierarchical analysis on the text data related to the special disease that needs to be analyzed, and then disassembles the acquired text data in the form of event streams. According to the information that needs to be analyzed, the event stream of the content category containing the information is selected in a targeted manner, and further detailed structured analysis of various indicators is performed.

[0035] Compared with the prior art, the method and system for parsing detailed items of medical records of specific diseases in the present invention have the following outstanding beneficial effects:

[0036] The present invention processes the data in steps, first dividing the event flow by time nodes and content categories, and then performing process analysis on specific details. Compared with directly processing large sections of text, the accuracy of the analysis results is improved, and it is easier to perform error troubleshooting, method optimization and system maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Attached Figure 1 It is a flowchart of a method for parsing detailed items of medical records of specific diseases;

[0039] Attached Figure 2 It is a schematic diagram of the pathological characteristics analysis process of a method for analyzing detailed items in medical records of a specific disease text;

[0040] Attached Figure 3 It is a schematic diagram of the treatment plan analysis process in the detailed analysis method of medical records of specific diseases;

[0041] Attached Figure 4 It is a schematic diagram of the process of parsing surgical information in a method for parsing medical records in a specific disease text;

[0042] Attached Figure 5 It is a schematic diagram of the radiotherapy plan analysis process in the detailed analysis method of medical records of specific diseases;

[0043] Attached Figure 6 It is a schematic diagram of a system for parsing detailed medical records for specific diseases. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] A best embodiment is given below:

[0046] like Figure 1-5 As shown, a method for parsing detailed items of medical records of a specific disease in this embodiment has the following steps:

[0047] S1. First, obtain the text data related to the specific disease to be analyzed and perform hierarchical analysis;

[0048] First, obtain the text data related to the specific disease that needs to be analyzed, including discharge records-pathological characteristics, admission records, and diagnostic data. The information that needs to be structured includes pathological characteristics, treatment plans, surgical information, radiotherapy information, etc. In-depth analysis of data characteristics involves a large number of time nodes, each of which has different corresponding content and different characteristics, so layered analysis is required.

[0049] S2, disassemble the acquired text data in the form of event stream;

[0050] Further including:

[0051] S201, split by text content:

[0052] By using the keyword matching method, the keywords such as "epidemiological situation", "current medical history", "general condition", "general examination", "medical history summary", "physical examination", "auxiliary examination", "physical examination", and "diagnostic basis" are matched from the text, and the text content is segmented in the sentences at the corresponding positions to distinguish the events contained in each item in the data, and each part of the content is labeled with the category name, and the name of the category to which each paragraph of text belongs is stored in the corresponding column field.

[0053] S202, split the sentence by symbol and roughly segment the text:

[0054] By using the keyword matching method, the keywords such as "epidemiological situation", "current medical history", "general condition", "general examination", "medical history summary", "physical examination", "auxiliary examination", "physical examination", and "diagnostic basis" are matched from the text, and the text content is segmented in the sentences at the corresponding positions to distinguish the events contained in each item in the data, and each part of the content is labeled with the category name, and the name of the category to which each paragraph of text belongs is stored in the corresponding column field.

[0055] S203, splitting the event stream according to time nodes:

[0056] First, split the event stream according to the time node. If there is a time mark in the current sentence, it is an event stream. If there is no time mark in the current sentence, it is added to the previous sentence with a time mark, and the time is obtained based on the standard time type. After each sentence is split, the medical record content of a certain time node is obtained. Secondly, sort the event stream according to the word order of the original text, and save the serial number corresponding to each event stream in the corresponding column field. Here, it is split according to a complete sentence, so there may be multiple time points in an event stream.

[0057] Time nodes: clear time and relative time

[0058] Explicit time refers to: various forms of year-month-day, year-month-day, year-month-month, year; various forms of month-day, month;

[0059] Relative time refers to: (nearly|nearly) (year|month|month|week|week) ago, before admission, after discharge, outpatient clinic, after surgery, after chemotherapy and radiotherapy, etc., at (today|today|now|patient for the purpose, after discharge|after returning home|outpatient clinic for the purpose|after admission|for further outpatient clinic, etc. relative time points;

[0060] At the same time, the time is processed in a standardized format, and the words such as this year are supplemented by taking the year of admission;

[0061] The text content is the content before the first time and the second time.

[0062] So far, all event streams with content annotations and time nodes are obtained. S3. According to the information to be analyzed, the content category event stream containing the information is selected in a targeted manner, and the detailed structural analysis of each indicator is further performed:

[0063] Further includes:

[0064] S301. Analysis of pathological characteristics:

[0065] According to data analysis, the analysis of pathological characteristics first selects the event stream with the category ("epidemiological situation", "medical history summary", "auxiliary examination"), and then extracts the event stream with pathological content based on relevant keywords such as "pathology", "genetic testing", and "immunohistochemistry", and further analyzes the extracted event epidemic item fields.

[0066] According to the different characteristics of different information items, we first segment the symbols and extract the small sentences containing the target data from the event stream. Then we select the appropriate method (keyword matching method or regular expression matching method) to extract the content of each field and structure it to achieve the target requirements.

[0067] ① Pathological examination date: regular expression matching method, take the first time node appearing in the event flow and give priority to the specific time.

[0068] ② Pathological classification: Based on relevant medical knowledge, the keyword matching method is used to match the original words of pathological classification and perform normalization.

[0069] ③ Tumor location: Based on relevant medical knowledge, the keyword matching method is used to match the original words of the tumor location and perform normalization.

[0070] ④ Transfer position: First, extract the small sentences with transfer-related information, then use the keyword matching method to match the original words of the transfer position, and then perform normalization processing.

[0071] ⑤TNM staging: regular expression matching method, unified format, and structured processing.

[0072] ⑥Immunohistochemistry: First, based on the knowledge related to immunohistochemistry, the keyword matching method is used to extract small sentences with immune-related content. Then, the method combining keywords and regular expressions is used to extract the target detailed content, and then normalized and structured.

[0073] ⑦Gene detection: First, based on the relevant knowledge of gene detection, use the keyword matching method to extract small sentences with gene-related content, and then use the combination of keywords and regular expressions to extract the target details, and perform normalization and structure processing.

[0074] ⑧Lymph node status: First, based on the knowledge related to lymph node detection, the keyword matching method is used to extract small sentences with lymph node-related content. Then, the method combining keywords and regular expressions is used to extract the target detailed content, and normalize and structure it.

[0075] S302, Treatment plan analysis:

[0076] First, split each event stream into small sentences according to punctuation marks, and then split the event stream again by time. Each small sentence corresponds to only one time node.

[0077] Detailed analysis of each field, including:

[0078] Drug name: Build a drug dictionary based on the statistics of related specialty drugs, and match based on the dictionary.

[0079] Treatment plan name: Design a regular expression based on the treatment plan abbreviation for regular matching.

[0080] Treatment cycle: Use "cycle" as the keyword for regular matching.

[0081] Project start time: the time when the first small sentence is split.

[0082] Medication start time: When the drug name can be parsed in the small sentence, the medication time will be parsed. The medication start time is the same as the plan start time.

[0083] End time of medication: When a sentence contains "from a certain time point to a certain time point", the first time is the start time of medication and the start time of the plan, and the second time is the end time of medication; when a sentence contains "the last chemotherapy was at a certain time", the time is the end time of medication.

[0084] Medication efficacy evaluation (including medication efficacy evaluation time and medication efficacy evaluation results): analysis method combining keywords and regular expressions.

[0085] Medication efficacy evaluation measures: Keyword regularization extraction.

[0086] Solution category: keyword regular extraction.

[0087] Medication route: keyword regular extraction.

[0088] Only the sentences from which the treatment plan name, plan category, and drug name can be parsed are parsed and stored in the database.

[0089] S303, Detailed analysis of surgical information:

[0090] According to data analysis, surgical information parsing first selects event streams with categories ("epidemiological situation", "medical history summary", "auxiliary examination"), and then divides the event streams into small sentences according to time.

[0091] According to the surgery dictionary table, the small sentence with the surgery name is extracted as 'surgery description'.

[0092] Detailed analysis in the surgical description:

[0093] Operation name: Matched according to the operation dictionary table.

[0094] Operation time: Use regular expression to match the first time in a description as the operation time.

[0095] Anesthesia method: keyword matching "local anesthesia", "general anesthesia", "endobronchial anesthesia", etc.

[0096] S304, Radiotherapy plan analysis:

[0097] There is usually only one radiotherapy plan in a data stream, so the data stream is not further split into sentences during radiotherapy plan analysis.

[0098] Replace keywords that contain "radiotherapy" but are not radiotherapy plans, such as "radiotherapy department", "no radiotherapy", etc.

[0099] Detailed analysis of each field, including:

[0100] Radiotherapy technology name: dictionary matching extraction for the entire event stream;

[0101] Radiotherapy start time: regular analysis for the entire time stream;

[0102] Radiotherapy efficacy evaluation time: divided into sentences by [,,.;], when there are keywords 'after radiotherapy|end of radiotherapy' in the sentence, the time in the sentence is the efficacy evaluation time.

[0103] Radiotherapy efficacy evaluation results: The evaluation results are extracted from the dictionary from the position of the radiotherapy technology name to the position where the first period appears (if there is no period, then to the end).

[0104] Radiotherapy efficacy evaluation measures: Based on the radiotherapy technique name position to the end, the dictionary extracts radiotherapy efficacy evaluation measures.

[0105] like Figure 6 As shown, a detailed item parsing system for medical records of special diseases includes a pathological characteristics parsing module, a treatment plan parsing module, a surgical information parsing module and a radiotherapy plan parsing module. The data is first parsed in the pathological characteristics parsing module, and then the treatment parsing module is entered to perform data treatment analysis. After entering the surgical information parsing, the detailed item analysis of the specific content is performed, and finally the radiotherapy plan parsing module is entered for analysis.

[0106] In the pathological characteristics analysis module, the pathological characteristics-related content data in the event stream storage table is obtained, and the analyzed pathological examination date, pathological classification, tumor location, metastasis location, TNM stage, immunohistochemistry details, genetic testing details, and lymph node status details are structured and displayed in a table format;

[0107] The parsed drug name, treatment plan name, treatment cycle, plan start time, medication start time, medication end time, medication efficacy evaluation time, medication efficacy evaluation results, medication efficacy evaluation measures, plan category, and medication route are displayed in a structured manner in a table;

[0108] Create a surgery dictionary, which can be a specific surgery name or a surgery name of particular interest. Use the dictionary to match the event stream, select the event stream data containing the specific surgery name, and present the parsed surgery name, surgery time, and anesthesia method in a structured manner in a table.

[0109] The parsed radiotherapy technology name, radiotherapy start time, radiotherapy efficacy evaluation time, radiotherapy efficacy evaluation results, and radiotherapy efficacy evaluation measures are displayed in a structured manner in a table.

[0110] A device for parsing detailed items in medical records of specific disease texts, comprising: at least one memory and at least one processor; the at least one memory is used to store a machine-readable program; the at least one processor is used to call the machine-readable program to execute a method for parsing detailed items in medical records of specific disease texts claimed.

[0111] A computer-readable medium stores computer instructions, which, when executed by a processor, cause the processor to execute a method for parsing detailed items in medical records of specialized disease texts.

[0112] The above-mentioned specific implementations are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the above-mentioned specific implementations. Any appropriate changes or substitutions made to the detailed analysis methods, systems, devices and computer-readable media claims that conform to the present invention and are made by any ordinary technician in the technical field shall fall within the patent protection scope of the present invention.

[0113] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for parsing detailed items of medical records of specific diseases, characterized in that: The steps are as follows: S1. First, obtain the text data related to the specific disease to be analyzed and perform hierarchical analysis; S2, disassemble the acquired text data in the form of event stream; Further including: S201, splitting according to text content; Use keyword matching to match keywords from the text, and segment the text content in the sentences at the corresponding positions to distinguish each item in the data, and label each part of the content with the category name, and store the name of the category to which each text belongs in the corresponding column field; S202, splitting sentences according to symbols and roughly segmenting the text; Segment sentences according to symbols to form multiple complete sentences; S203, splitting the event stream according to time nodes; If there is a time mark in the current sentence, it is an event stream. If there is no time mark in the current sentence, it is added to the previous sentence with a time mark, and the time is obtained based on the standard time type first; After each sentence is split, the medical record content at a certain time node is obtained. Then, the event stream is sorted according to the word order of the original text, and the sequence number corresponding to each event stream is saved in a corresponding column field; S3. According to the information that needs to be analyzed, the content category event flow containing the information is selected in a targeted manner, and further detailed structural analysis of various indicators is performed; Further including: S301, analysis of pathological characteristics; Pathological characteristics analysis first selects the category as event flow, then extracts the event flow with pathological content based on relevant keywords, and analyzes the extracted event popular item fields; According to the different characteristics of different information items, the symbols are first segmented to extract the small sentences containing the target data from the event stream, and then the appropriate method is selected to extract the content of each field; S302, analysis of treatment plan; First, split each event stream into small sentences according to punctuation marks, and then split the event stream again by time. Each small sentence corresponds to only one time node, and then each field is analyzed in detail; S303, detailed analysis of surgical information; According to data analysis, surgical information parsing first selects the event stream of category , and then divides the event stream into small sentences according to time; according to the surgical dictionary table, extracts small sentences with surgical names as surgical descriptions, and performs detailed analysis in the surgical descriptions; S304, radiotherapy plan analysis; There is usually only one radiotherapy plan in a data stream. Therefore, in the analysis of radiotherapy plans, the data stream is not further split into sentences. Keywords with "radiotherapy" but not radiotherapy plans are replaced, and then each field is analyzed in detail.

2. A system for parsing detailed medical records of specific diseases, characterized in that: It includes pathological characteristics analysis module, treatment plan analysis module, surgical information analysis module and radiotherapy plan analysis module; The pathological characteristics analysis module is used to first obtain the text data related to the specific disease to be analyzed and perform hierarchical analysis; The treatment plan parsing module is used to disassemble the acquired text data in the form of event stream; include: (1) Split by text content; Use keyword matching to match keywords from the text, and segment the text content in the sentences at the corresponding positions to distinguish each item in the data, and label each part of the content with the category name, and store the name of the category to which each text belongs in the corresponding column field; (2) Break sentences into words according to symbols and roughly segment the text; Segment sentences according to symbols to form multiple complete sentences; (3) Split the event stream by time node; If there is a time mark in the current sentence, it is an event stream. If there is no time mark in the current sentence, it is added to the previous sentence with a time mark, and the time is obtained based on the standard time type first; After each sentence is split, the medical record content at a certain time node is obtained. Then, the event stream is sorted according to the word order of the original text, and the sequence number corresponding to each event stream is saved in a corresponding column field; The surgical information analysis module is used to select the content category event stream containing the information according to the information to be analyzed, and further perform detailed structural analysis of various indicators; include: (a) Analysis of pathological characteristics; Pathological characteristics analysis first selects the category as event flow, then extracts the event flow with pathological content based on relevant keywords, and analyzes the extracted event popular item fields; According to the different characteristics of different information items, the symbols are first segmented to extract the small sentences containing the target data from the event stream, and then the appropriate method is selected to extract the content of each field; (b) Analysis of treatment options; First, split each event stream into small sentences according to punctuation marks, and then split the event stream again by time. Each small sentence corresponds to only one time node, and then each field is analyzed in detail; (c) Detailed analysis of surgical information; According to data analysis, surgical information parsing first selects the event stream of category , and then divides the event stream into small sentences according to time; according to the surgical dictionary table, extracts small sentences with surgical names as surgical descriptions, and performs detailed analysis in the surgical descriptions; The radiotherapy plan analysis module is used in a data stream in which there is generally only one radiotherapy plan. Therefore, in the radiotherapy plan analysis, when the data stream is not further split into sentences, the keywords containing "radiotherapy" but not radiotherapy plans are replaced, and then each field is analyzed in detail.

3. A detailed analysis system for medical records of specific diseases according to claim 2, characterized in that: In the pathological characteristics analysis module, the pathological characteristics-related content data in the event stream storage table is obtained, and the analyzed pathological examination date, pathological classification, tumor location, metastasis location, TNM stage, immunohistochemistry details, genetic testing details, and lymph node status details are structured and displayed in a table format; The parsed drug name, treatment plan name, treatment cycle, plan start time, medication start time, medication end time, medication efficacy evaluation time, medication efficacy evaluation results, medication efficacy evaluation measures, plan category, and medication route are displayed in a structured manner in a table; Create a surgery dictionary, which can be a specific surgery name or a surgery name of particular interest. Use the dictionary to match the event stream, select the event stream data containing the specific surgery name, and present the parsed surgery name, surgery time, and anesthesia method in a structured manner in a table. The parsed radiotherapy technology name, radiotherapy start time, radiotherapy efficacy evaluation time, radiotherapy efficacy evaluation results, and radiotherapy efficacy evaluation measures are displayed in a structured manner in a table.

4. A device for parsing detailed items of medical records of specific diseases, characterized in that: include: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to call the machine-readable program to execute the method of claim 1.

5. A computer-readable medium, characterized in that The computer readable medium stores computer instructions, which, when executed by a processor, cause the processor to perform the method of claim 1 .

Images