Method, device, equipment and storage medium for generating disease name code list

Abstract

The present invention relates to the field of digital medicine, and discloses a method, device, equipment and storage medium for generating a disease name matching list, which is used to accurately segment and match the disease names to be matched by using an improved matching model, thereby improving the matching accuracy and recall rate of the disease names by the matching model. The method of the present invention comprises: obtaining an initial data set, the initial data set comprising a plurality of original medical sentences to be matched; preprocessing the initial data set to obtain medical data of multiple categories; calling a preset entity recognition model to perform entity recognition on the medical data of the multiple categories to obtain a disease name recognition result; calling a preset matching model to match the disease name recognition result with a preset standard data set to obtain a matching result; generating a matching list based on the matching result, the matching list comprising an original phrase column, a standard phrase column and a data encoding column.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, device, equipment and storage medium for generating a disease name code list. Background Art

[0002] Medical clinical data standardization is a basic capability to support various scenarios such as medical alliances, regional data intercommunication and mutual recognition, and medical insurance cross-regional settlement. Before the establishment of a unified standard catalog, different provinces, cities, regions, and hospitals did not adopt unified standards for the coding and names of diagnosis, surgery, diseases, and drugs when establishing information systems, resulting in medical information not being available nationwide.

[0003] At present, in the process of reforming medical payment methods, it is also necessary to unify the coding of historical data. Otherwise, it will be difficult to adapt to the requirements of medical insurance in the new era. The current disease names are generated in lists through manual coding, which is inefficient. Summary of the invention

[0004] The present invention provides a method, device, equipment and storage medium for generating a disease name matching list, which is used to accurately segment and match the disease names to be matched using an improved matching model, thereby improving the matching accuracy and recall rate of the disease names by the matching model.

[0005] A first aspect of an embodiment of the present invention provides a method for generating a disease name coding list, comprising: obtaining an initial data set, the initial data set comprising a plurality of original medical statements that need to be coded; preprocessing the initial data set to obtain medical data of multiple categories; calling a recognition neural network in a preset coding model to perform entity recognition on the medical data of the multiple categories to obtain a disease name recognition result; calling a distributed full-text retrieval engine and a twin network in the coding model to code the disease name recognition result with a preset standard data set to obtain a coding result; generating a coding list based on the coding result, the coding list comprising an original phrase column, a standard phrase column and a data encoding column.

[0006] Optionally, in a first implementation method of the first aspect of the embodiment of the present invention, the initial data set is preprocessed to obtain multiple categories of medical data, including: formatting and text error correction of the initial data set to obtain replaced data text; filtering out non-medical data and duplicate data in the replaced data text to obtain filtered data; completing, modifying or deleting missing data and abnormal data in the filtered data to obtain adjusted data; integrating the adjusted data according to preset department categories to obtain multiple categories of medical data.

[0007] Optionally, in a second implementation method of the first aspect of the embodiment of the present invention, the adjusted data is integrated according to preset department categories to obtain multiple categories of medical data, including: obtaining a keyword set corresponding to each department category in the preset department categories to obtain multiple department keyword sets, each department keyword set corresponds to a department category; matching the adjusted data with the multiple department keyword sets in sequence to obtain multiple matching result pairs, the matching result pairs including matched medical statements and matched department categories; integrating the adjusted data according to the matched department categories in each matching result pair to obtain multiple categories of medical data, each category of medical data including multiple medical-related statements belonging to this category.

[0008] Optionally, in a third implementation method of the first aspect of the embodiment of the present invention, the calling of the recognition neural network in the preset coding model to perform entity recognition on the multiple categories of medical data to obtain disease name recognition results includes: converting the multiple categories of medical data into multiple word vector sequences; annotating the multiple word vector sequences to obtain multiple annotated vector sequences; calling the recognition neural network from the preset coding model to perform entity recognition on the multiple annotated vector sequences to obtain entity recognition results, the entity recognition results including multiple medical entity feature vectors and the part of speech corresponding to each medical entity; filtering out multiple disease names from the multiple medical entity feature vectors of the entity recognition results to obtain disease name recognition results, the disease name recognition results including multiple disease names with noun attributes.

[0009] Optionally, in a fourth implementation manner of the first aspect of the embodiment of the present invention, the recognition neural network is called from a preset pairing model to perform entity recognition on the multiple labeled vector sequences to obtain an entity recognition result, and the entity recognition result includes multiple medical entity feature vectors and the part of speech corresponding to each medical entity, including: calling the BiLSTM layer of the recognition neural network in the preset pairing model to perform feature extraction on the multiple labeled sequences to obtain multiple predicted label sequences, wherein each predicted label in the predicted label sequence corresponds to each word in the labeled sequence; calling the CRF layer of the recognition neural network to perform contextual semantic constraint adjustment on the multiple predicted label sequences to obtain an optimal label sequence; determining the part of speech corresponding to each medical entity feature vector in the optimal label sequence to obtain an optimal part of speech sequence, and determining the optimal label sequence and the optimal part of speech sequence as the entity recognition result.

[0010] Optionally, in a fifth implementation manner of the first aspect of the embodiment of the present invention, the distributed full-text retrieval engine and the twin network in the matching model are called to match the disease name recognition result with a preset standard data set to obtain a matching result, including: full-word matching of each disease name feature vector in the disease name recognition result with a preset standard data set to obtain a full-word matching result; if the full-word matching result is not empty, the matched multiple words are determined as a candidate matching set; if the full-word matching result is empty, the distributed full-text retrieval engine Elasticsearch is used to recall related words to obtain a recall candidate set, and the recall candidate set includes multiple recalled words; the similarity between the recall candidate set and the disease name recognition result is calculated to obtain a similarity sequence arranged from large to small, and a preset number of words at the front of the similarity sequence are determined as a candidate matching set; the twin network in the matching model is called to match the candidate matching set with the preset standard disease name to generate a matching result, and the matching result includes multiple phrase pairs, each phrase pair includes the original disease name and the corresponding standard disease name.

[0011] Optionally, in a sixth implementation manner of the first aspect of the embodiment of the present invention, a matching list is generated based on the matching result, and the matching list includes an original phrase column, a standard phrase column and a data code column, including: determining multiple target original disease names based on multiple phrase pairs in the matching result, and importing the multiple target original disease names into the original phrase column; determining multiple target standard disease names based on multiple phrase pairs in the matching result, and importing the multiple target standard disease names into the standard phrase column; searching for at least one candidate code in a preset coding standard directory based on each target standard disease name; determining the candidate code with the latest update time among the at least one candidate code as the standard disease code corresponding to each target standard disease name, obtaining multiple standard disease codes, and importing the multiple standard disease codes into the data code column; arranging the original phrase column, the standard phrase column and the data code column in sequence to generate a matching list.

[0012] A second aspect of an embodiment of the present invention provides a device for generating a disease name matching list, comprising: an acquisition module for acquiring an initial data set, wherein the initial data set includes multiple original medical statements that need to be matched; a preprocessing module for preprocessing the initial data set to obtain multiple categories of medical data; an identification module for calling a recognition neural network in a preset matching model to perform entity recognition on the multiple categories of medical data to obtain a disease name recognition result; a matching module for calling a distributed full-text retrieval engine and a twin network in the matching model to match the disease name recognition result with a preset standard data set to obtain a matching result; a generation module for generating a matching list based on the matching result, wherein the matching list includes an original phrase column, a standard phrase column, and a data encoding column.

[0013] Optionally, in a first implementation method of the second aspect of the embodiment of the present invention, the preprocessing module includes: a data correction unit, used to perform format adjustment and text correction on the initial data set to obtain a replaced data text; a data screening unit, used to screen out non-medical data and duplicate data in the replaced data text to obtain screened data; a data adjustment unit, used to complete, modify or delete missing data and abnormal data in the screened data to obtain adjusted data; a data integration unit, used to integrate the adjusted data according to preset department categories to obtain medical data of multiple categories.

[0014] Optionally, in a second implementation method of the second aspect of the embodiment of the present invention, the data integration unit is specifically used to: obtain a keyword set corresponding to each department category in the preset department categories to obtain multiple department keyword sets, each department keyword set corresponds to a department category; match the adjusted data with the multiple department keyword sets in sequence to obtain multiple matching result pairs, the matching result pairs include matched medical statements and matched department categories; integrate the adjusted data according to the matched department categories in each matching result pair to obtain multiple categories of medical data, each category of medical data includes multiple medical-related statements belonging to this category.

[0015] Optionally, in a third implementation of the second aspect of the embodiment of the present invention, the recognition module includes: a sequence conversion unit, used to convert the multiple categories of medical data into multiple word vector sequences; a sequence labeling unit, used to label the multiple word vector sequences to obtain multiple labeled vector sequences; an entity recognition unit, used to call a recognition neural network from a preset coding model to perform entity recognition on the multiple labeled vector sequences to obtain an entity recognition result, wherein the entity recognition result includes multiple medical entity feature vectors and the part of speech corresponding to each medical entity; a name screening unit, used to screen out multiple disease names from the multiple medical entity feature vectors of the entity recognition result to obtain a disease name recognition result, wherein the disease name recognition result includes multiple disease names with noun attributes.

[0016] Optionally, in a fourth implementation manner of the second aspect of the embodiment of the present invention, the entity recognition unit is specifically used to: call the BiLSTM layer of the recognition neural network in the preset coding model to perform feature extraction on the multiple labeled sequences to obtain multiple predicted label sequences, wherein each predicted label in the predicted label sequence corresponds to each word in the labeled sequence; call the CRF layer of the recognition neural network to perform contextual semantic constraint adjustment on the multiple predicted label sequences to obtain an optimal label sequence; determine the part of speech corresponding to each medical entity feature vector in the optimal label sequence to obtain an optimal part of speech sequence, and determine the optimal label sequence and the optimal part of speech sequence as the entity recognition result.

[0017] Optionally, in a fifth implementation manner of the second aspect of the embodiment of the present invention, the matching module is specifically used to: perform full-word matching on each disease name feature vector in the disease name recognition result with a preset standard data set to obtain a full-word matching result; if the full-word matching result is not empty, determine the matched multiple words as a candidate matching set; if the full-word matching result is empty, use the distributed full-text retrieval engine Elasticsearch to recall related words to obtain a recall candidate set, and the recall candidate set includes multiple recalled words; calculate the similarity between the recall candidate set and the disease name recognition result to obtain a similarity sequence arranged from large to small, and determine a preset number of words at the front of the similarity sequence as a candidate matching set; call the twin network in the matching model to match the candidate matching set with the preset standard disease name to generate a matching result, and the matching result includes multiple phrase pairs, each phrase pair includes the original disease name and the corresponding standard disease name.

[0018] Optionally, in a sixth implementation manner of the second aspect of the embodiment of the present invention, the generation module is specifically used to: determine multiple target original disease names based on multiple phrase pairs in the matching results, and import the multiple target original disease names into the original phrase column; determine multiple target standard disease names based on multiple phrase pairs in the matching results, and import the multiple target standard disease names into the standard phrase column; search for at least one candidate code in a preset coding standard directory based on each target standard disease name; determine the candidate code with the latest update time among the at least one candidate code as the standard disease code corresponding to each target standard disease name, obtain multiple standard disease codes, and import the multiple standard disease codes into the data code column; arrange the original phrase column, the standard phrase column and the data code column in sequence to generate a matching list.

[0019] The third aspect of an embodiment of the present invention provides a device for generating a disease name code pairing list, a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the device for generating a disease name code pairing list executes the above-mentioned method for generating a disease name code pairing list.

[0020] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores instructions. When the instructions are executed by a processor, the steps of the method for generating a disease name code list described in any of the above embodiments are implemented.

[0021] In the technical solution provided by the embodiment of the present invention, an initial data set is obtained, and the initial data set includes multiple original medical sentences that need to be matched; the initial data set is preprocessed to obtain multiple categories of medical data; the recognition neural network in the preset matching model is called to perform entity recognition on the multiple categories of medical data to obtain the disease name recognition result; the distributed full-text retrieval engine and the twin network in the matching model are called to match the disease name recognition result with the preset standard data set to obtain the matching result; the matching list is generated based on the matching result, and the matching list includes the original phrase column, the standard phrase column and the data encoding column. In the embodiment of the present invention, the recognition neural network in the improved matching model is used to accurately segment the disease name to be matched, and then the distributed full-text retrieval engine in the improved matching model is used to recall, and finally the twin network in the improved matching model is used to match, thereby improving the matching accuracy and recall rate of the improved matching model for the disease name. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of an embodiment of a method for generating a disease name code list in an embodiment of the present invention;

[0023] Figure 2 Schematic diagram of another embodiment of the method for generating a disease name code list in an embodiment of the present invention;

[0024] Figure 3 It is a schematic diagram of an embodiment of a device for generating a disease name code list in an embodiment of the present invention;

[0025] Figure 4 It is a schematic diagram of another embodiment of a device for generating a disease name code list in an embodiment of the present invention;

[0026] Figure 5 The figure is a schematic diagram of an embodiment of a device for generating a disease name code list in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The present invention provides a method, device, equipment and storage medium for generating a disease name matching list, which is used to use a recognition neural network in an improved matching model to accurately segment the disease names to be matched, then use a distributed full-text retrieval engine in the improved matching model to recall them, and finally match them through a twin network in the improved matching model, thereby improving the matching accuracy and recall rate of the improved matching model for the disease names.

[0028] In order to enable those skilled in the art to better understand the solutions of the present invention, the embodiments of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0029] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0030] The server can be a stand-alone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), as well as big data and artificial intelligence platforms.

[0031] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0032] See also Figure 1 , a flowchart of a method for generating a disease name code list provided by an embodiment of the present invention, specifically comprising:

[0033] 101. Obtain an initial data set, where the initial data set includes multiple original medical statements that need to be coded.

[0034] The original medical statement is an unprocessed medical statement, which includes normal data and dirty data. Dirty data refers to missing data, abnormal data, duplicate data or unrecognizable medical data.

[0035] For example, the normal range of blood sugar is 3.9 to 6.1 mmol. If the blood sugar is too high or too low, the order of magnitude of the blood sugar value will not change. If the order of magnitude of the blood sugar value exceeds 100 while the unit remains unchanged, it can be considered as abnormal data.

[0036] It is understandable that the execution subject of the present invention may be a device for generating a disease name code list, or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0037] 102. Preprocess the initial data set to obtain medical data of multiple categories.

[0038] Among them, the preprocessing process mainly includes data cleaning and data integration. Specifically, data cleaning mainly includes screening out data irrelevant to the subject, deleting duplicate data, filling in missing data, and adjusting abnormal data.

[0039] For example, if the initial data set includes three fields such as gender, physical examination date, and ID number, the ID number is an irrelevant field for diabetes detection and needs to be deleted; the gender field can be mapped to 0 and 1 for male and female respectively. If the gender field corresponding to gynecological diseases is male, the data is abnormal data.

[0040] 103. Call the recognition neural network in the preset coding model to perform entity recognition on multiple categories of medical data to obtain disease name recognition results.

[0041] It should be noted that the current medical coding models are mostly based on dictionaries, and only words in a custom dictionary can be recognized. The preset coding model (improved coding model) in this embodiment solves this drawback. The recognition neural network of the preset coding model in this embodiment is based on contextual semantics for recognition, and can consider contextual semantics to determine the word segmentation scenario, thereby improving the accuracy of the entity recognition results. For example, if "psoriasis" is not in the dictionary, the word segmentation tool cannot recognize it. If the recognition neural network of the coding model is used instead, it can be recognized as long as there is enough training data. The traditional coding model does not have the entity recognition function, and in this embodiment, the preamble of the coding model used for data coding is modified, and the recognition neural network is added to increase the entity recognition function.

[0042] 104. Call the distributed full-text search engine and the twin network in the matching model to match the disease name recognition result with the preset standard data set to obtain the matching result.

[0043] Specifically, the server calls the distributed full-text retrieval engine and twin network in the improved matching model to match the disease name recognition results with the preset standard data set to obtain the matching results.

[0044] It should be noted that the improved matching model can preprocess the disease name recognition results, and then perform matching with the preset standard data set based on the preprocessed data. The similarity between each recognized original disease name and each word in the standard data set can be calculated, and the similarities are arranged in descending order according to the similarity. The disease name at the head of the sequence is determined as the matching standard disease name, the matching original disease name and the standard disease name are determined as a phrase pair, and the generated multiple phrase pairs are used as matching results.

[0045] 105. Generate a matching list based on the matching result, where the matching list includes an original phrase column, a standard phrase column, and a data encoding column.

[0046] Specifically, the server puts the original disease names in multiple phrase pairs into the original phrase column, puts the corresponding standard disease names into the standard phrase column, and then searches for the corresponding standard disease codes in the preset coding standards based on the standard phrase column, wherein different original disease names may correspond to the same standard disease name, and the disease code corresponding to the standard disease name is unique.

[0047] In an embodiment of the present invention, the recognition neural network in the improved coding model is used to accurately segment the disease names to be coded, and then the distributed full-text retrieval engine in the improved coding model is used to recall them. Finally, the twin network in the improved coding model is used to perform coding, thereby improving the coding accuracy and recall rate of the improved coding model for disease names.

[0048] See also Figure 2 , another flow chart of the method for generating a disease name code list provided by an embodiment of the present invention, specifically comprising:

[0049] 201. Obtain an initial data set, where the initial data set includes a plurality of original medical statements that need to be coded.

[0050] The original medical statement is an unprocessed medical statement, which includes normal data and dirty data. Dirty data refers to missing data, abnormal data, duplicate data or unrecognizable medical data.

[0051] For example, the normal range of blood sugar is 3.9 to 6.1 mmol. If the blood sugar is too high or too low, the order of magnitude of the blood sugar value will not change. If the order of magnitude of the blood sugar value exceeds 100 while the unit remains unchanged, it can be considered as abnormal data.

[0052] It is understandable that the execution subject of the present invention may be a device for generating a disease name code list, or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0053] 202. Preprocess the initial data set to obtain medical data of multiple categories.

[0054] The server preprocesses the initial data set to obtain multiple categories of medical data. The preprocessing process mainly includes data cleaning and data integration. Specifically, data cleaning mainly includes filtering out data irrelevant to the subject (non-medical data), deleting duplicate data, filling in missing data, and adjusting abnormal data.

[0055] For example, if the initial data set includes three fields such as gender, physical examination date, and ID number, the ID number is an irrelevant field for diabetes detection and needs to be deleted; the gender field can be mapped to 0 and 1 for male and female respectively. If the gender field corresponding to gynecological diseases is male, the data is abnormal data.

[0056] Optionally, the server preprocesses the initial data set to obtain multiple categories of medical data, including:

[0057] 1. The server adjusts the format and corrects the text of the initial data set to obtain the replaced data text;

[0058] Among them, the initial data set is spelled based on the preset dictionary, and the misspelled words are replaced. The replaced data text is obtained. For example, if "Fusemide" appears in the data set, according to the preset standard dictionary, it is found that there is no word "Fusemide", and homophonic or similar words are searched, and "Fusemide" is retrieved, then "Fusemide" is replaced with "Fusemide". Similarly, if "Lasemide" appears and the word cannot be retrieved, after an approximate search, it is found that "Torsemide" is the most likely, and "Torsemide" is replaced with "Lasemide".

[0059] Among them, format adjustment may include removing spaces, converting full-width characters to half-width characters, or useless characters such as modal particles and punctuation marks. It may also perform rule processing on the following features, for example, dividing sentences in a paragraph according to semantic completeness. It may also perform other processing, which is not limited here.

[0060] 2. The server filters out non-medical data and duplicate data in the replaced data text to obtain the filtered data;

[0061] After format adjustment and text error correction, the data still needs to delete non-medical data that does not involve medical knowledge and select duplicate data.

[0062] 3. The server completes, modifies or deletes the missing data and abnormal data in the filtered data to obtain the adjusted data;

[0063] Among them, missing data can be deleted or supplemented. For example, if other attributes corresponding to the missing values ​​are more important or there are more attributes, the missing values ​​need to be supplemented, for example, using the same type of mean interpolation or multiple interpolation. The details are not repeated here. Abnormal data can be deleted or modified. The details are not repeated here.

[0064] 4. The server integrates the adjusted data according to the preset department categories to obtain medical data of multiple categories.

[0065] Specifically, it includes: the server obtains a keyword set corresponding to each department category in the preset department category to obtain multiple department keyword sets, and each department keyword set corresponds to a department category; the server matches the adjusted data with the multiple department keyword sets in sequence to obtain multiple matching result pairs, and the matching result pairs include matched medical statements and matched department categories; the server integrates the adjusted data according to the matched department categories in each matching result pair to obtain multiple categories of medical data, and each category of medical data includes multiple medical-related statements belonging to this category.

[0066] For example, the keyword set for internal medicine may include heart disease, stomach disease, stones, etc. Each department covers a wide range of diseases, and 3 to 5 keywords can be selected for each disease and put into the keyword set for internal medicine. For another example, the keyword set for surgery may involve orthopedics, cardiothoracic surgery, urology, etc., specifically including keywords such as pneumothorax, lung removal, stone surgery, and cerebral hemorrhage.

[0067] 203. Convert multiple categories of medical data into multiple word vector sequences.

[0068] The server converts multiple categories of medical data into multiple word vector sequences, wherein each category of medical data corresponds to a word vector sequence.

[0069] 204. Label the multiple word vector sequences to obtain multiple labeled vector sequences.

[0070] The server labels multiple word vector sequences to obtain multiple labeled vector sequences.

[0071] It should be noted that this embodiment uses the BIO mode for annotation, for example, lesion entity: B-Jibing, I-Jibing, anatomical entity: B-Jiepou, I-Jiepou, procedure entity: B-Shushi, I-Shushi. Among them, B: begin, means beginning; I: inside, means middle; O: other, means other. BIO annotation: In general, each element is marked as "BX", "IX" or "O". For example, "BX" means that the segment where this element is located belongs to type X and this element is at the beginning of this segment, "IX" means that the segment where this element is located belongs to type X and this element is in the middle of this segment, and "O" means that it does not belong to any type.

[0072] 205. Calling a recognition neural network from a preset coding model to perform entity recognition on multiple labeled vector sequences to obtain entity recognition results, wherein the entity recognition results include multiple medical entity feature vectors and the part of speech corresponding to each medical entity.

[0073] Specifically, it includes: 1. The server calls the BiLSTM layer in the recognition neural network from the preset matching model to extract features from multiple labeled sequences to obtain multiple predicted label sequences, wherein each predicted label in the predicted label sequence corresponds to each word in the labeled sequence;

[0074] For example, for “wrist insect bite”, we first get the word vector {wrist, body, insect, bite, injury}. After passing through the BiLSTM layer, for “wrist”, the probability matrix corresponding to each category is [1.5, 0.9, 0.1, 0.08, 0.05], the corresponding scores for “body” are [0.2, 0.4, 0.1, 0.11, 0.05], the corresponding scores for “insect” are [0.09, 0.02, 0.03, 0.08, 0.1], the corresponding scores for “bite” are [0.003, 0.002, 0.2, 0.07, 0.05], and the corresponding scores for “injury” are [0.12, 0.2, 0.1, 0.065, 0.5].

[0075] The recognition neural network is trained in advance, and the training process includes: first labeling the expected: defining the entity type to be recognized, and then labeling it using the BIO mode.

[0076] For example, suppose there are N possible tag sequence combinations, and the score of the i-th tag sequence is P i , then the total score of all N possible tag sequence combinations is:

[0077] P total =P1+P2+P3+…+P N =e S1 +e S2 +e S3 +…+e SN , where e Si represents the score of the ith label sequence. Obviously, among all possible label sequence combinations, there must be one sequence that is the true label sequence, and the remaining label sequence combinations are wrong. In the entire recognition neural network training process, the loss function of the CRF layer only needs two scores: one is the true label sequence score, and the other is the total score of all possible label sequence combinations. Let the true label sequence score account for the highest proportion of the total score, that is, to obtain the optimal recognition neural network. For the score of the true label sequence, e Si , directly calculate S according to the emission score and transition score i Then:

[0078] S i =EmissionScore+TransitionScore.

[0079] 2. The server calls the CRF layer in the recognition neural network to adjust the contextual semantic constraints of multiple predicted label sequences to obtain the optimal label sequence;

[0080] The server uses the above predicted label sequence as the input of the CRF layer, performs contextual semantic constraints in the CRF layer, and selects the label sequence with the highest prediction score as the best label sequence. Among them, the contextual semantic constraints can be semantically completed through named entity recognition. For example, for the text "endovascular treatment of symptomatic atherosclerotic vertebral artery stenosis at the beginning", multiple entities are obtained after initial recognition, such as "symptomatic atherosclerosis", "vertebral artery", and "stenosis", but the recognized phrases cannot retain the complete semantics, so named entity recognition is used to supplement them, and the optimal label sequence output in the end is "symptomatic atherosclerotic vertebral artery stenosis at the beginning".

[0081] 3. The server determines the part of speech corresponding to each medical entity feature vector in the optimal label sequence, obtains the optimal part of speech sequence, and determines the optimal label sequence and the optimal part of speech sequence as the entity recognition result.

[0082] 206. Filter out multiple disease names from multiple medical entity feature vectors of entity recognition results to obtain disease name recognition results, where the disease name recognition results include disease names with multiple noun attributes.

[0083] The server selects multiple disease names from multiple medical entity feature vectors of the entity recognition results to obtain disease name recognition results, which include multiple disease names with noun attributes. For example, if it is related to the disease, the corresponding part-of-speech identifier may be nhd, if it is related to the drug, the corresponding part-of-speech identifier may be nhm; if it is related to health-related nouns such as medical diseases, the corresponding part-of-speech identifier may be nh. For another example, the noun identifier is n, the adjective identifier is a; the quantifier identifier is mq, and the biological name identifier is nb. The details are not repeated here.

[0084] It should be noted that the current medical coding models are mostly based on dictionaries, and only words in a custom dictionary can be recognized. The preset coding model (improved coding model) in this embodiment solves this drawback. The recognition neural network of the preset coding model in this embodiment is based on contextual semantics for recognition, and can consider contextual semantics to determine the word segmentation scenario, thereby improving the accuracy of the entity recognition results. For example, if "psoriasis" is not in the dictionary, the word segmentation tool cannot recognize it. If the recognition neural network of the coding model is used instead, it can be recognized as long as there is enough training data. The traditional coding model does not have the entity recognition function, and in this embodiment, the preamble of the coding model used for data coding is modified, and the recognition neural network is added to increase the entity recognition function.

[0085] 207. Call the distributed full-text search engine and the twin network in the matching model to match the disease name recognition result with the preset standard data set to obtain the matching result.

[0086] Specifically, the server performs full-word matching on each disease name feature vector in the disease name recognition result with a preset standard data set to obtain a full-word matching result; if the full-word matching result is not empty, the server determines the matched multiple words as a candidate pairing set; if the full-word matching result is empty, the server uses the distributed full-text retrieval engine Elasticsearch to recall related words to obtain a recall candidate set, which includes multiple recalled words; the server calculates the similarity between the recall candidate set and the disease name recognition result to obtain a similarity sequence arranged from large to small, and determines a preset number of words at the front of the similarity sequence as a candidate pairing set; the server calls the twin network in the matching model to match the candidate pairing set with the preset standard disease name to generate a matching result, which includes multiple phrase pairs, each phrase pair including the original disease name and the corresponding standard disease name.

[0087] It should be noted that the preset coding model can preprocess the disease name recognition results, and then perform coding matching with the preset standard data set based on the preprocessed data. The similarity between each recognized original disease name and each word in the standard data set can be calculated, and the similarities are arranged in descending order according to the similarity. The disease name at the head of the sequence is determined as the matching standard disease name, the matching original disease name and the standard disease name are determined as a phrase pair, and the generated multiple phrase pairs are used as coding results.

[0088] 208. Generate a matching list based on the matching result, where the matching list includes an original phrase column, a standard phrase column, and a data encoding column.

[0089] Specifically, the server determines multiple target original disease names based on multiple phrase pairs in the matching results, and imports the multiple target original disease names into the original phrase column; the server determines multiple target standard disease names based on multiple phrase pairs in the matching results, and imports the multiple target standard disease names into the standard phrase column; the server searches for at least one candidate code in the preset coding standard directory based on each target standard disease name; the server determines the candidate code with the latest update time among at least one candidate code as the standard disease code corresponding to each target standard disease name, obtains multiple standard disease codes, and imports the multiple standard disease codes into the data code column; the server arranges the original phrase column, the standard phrase column, and the data code column in sequence to generate a matching list.

[0090] It should be noted that different original disease names can correspond to the same standard disease name, and the disease code corresponding to the standard disease name is unique. For example, the original disease names identified may include: heart failure, heart failure, heart failure, etc., and the standard disease name corresponding to the above disease name is "heart failure", and the disease code corresponding to the standard disease name "heart failure" is "I50.902". For another example, the standard disease name corresponding to the names of "cerebral infarction" and "stroke" is "cerebral infarction", and the disease code corresponding to "cerebral infarction" is "I63.902".

[0091] In an embodiment of the present invention, the recognition neural network in the improved coding model is used to accurately segment the disease names to be coded, and then the distributed full-text retrieval engine in the improved coding model is used to recall them. Finally, the twin network in the improved coding model is used to perform coding, thereby improving the coding accuracy and recall rate of the improved coding model for disease names.

[0092] The above describes the method for generating the disease name code list in the embodiment of the present invention. The following describes the device for generating the disease name code list in the embodiment of the present invention. Figure 3 , an embodiment of a device for generating a disease name code list in an embodiment of the present invention includes:

[0093] An acquisition module 301 is used to acquire an initial data set, wherein the initial data set includes a plurality of original medical statements that need to be coded;

[0094] A preprocessing module 302 is used to preprocess the initial data set to obtain multiple categories of medical data;

[0095] The recognition module 303 is used to call the recognition neural network in the preset coding model to perform entity recognition on the medical data of the multiple categories to obtain the disease name recognition result;

[0096] The matching module 304 is used to call the distributed full-text search engine and the twin network in the matching model to match the disease name recognition result with a preset standard data set to obtain a matching result;

[0097] The generating module 305 is used to generate a matching list based on the matching result, wherein the matching list includes an original phrase list, a standard phrase list and a data encoding list.

[0098] In an embodiment of the present invention, the recognition neural network in the improved coding model is used to accurately segment the disease names to be coded, and then the distributed full-text retrieval engine in the improved coding model is used to recall them. Finally, the twin network in the improved coding model is used to perform coding, thereby improving the coding accuracy and recall rate of the improved coding model for disease names.

[0099] See also Figure 4 Another embodiment of the device for generating a disease name code list in the embodiment of the present invention includes:

[0100] An acquisition module 301 is used to acquire an initial data set, wherein the initial data set includes a plurality of original medical statements that need to be coded;

[0101] A preprocessing module 302 is used to preprocess the initial data set to obtain multiple categories of medical data;

[0102] The recognition module 303 is used to call the recognition neural network in the preset coding model to perform entity recognition on the medical data of the multiple categories to obtain the disease name recognition result;

[0103] The matching module 304 is used to call the distributed full-text search engine and the twin network in the matching model to match the disease name recognition result with a preset standard data set to obtain a matching result;

[0104] The generating module 305 is used to generate a matching list based on the matching result, wherein the matching list includes an original phrase list, a standard phrase list and a data encoding list.

[0105] Optionally, the preprocessing module 302 includes:

[0106] A data error correction unit 3021 is used to perform format adjustment and text error correction on the initial data set to obtain a replaced data text;

[0107] A data screening unit 3022 is used to screen out non-medical data and duplicate data in the replaced data text to obtain screened data;

[0108] The data adjustment unit 3023 is used to complete, modify or delete the missing data and abnormal data in the screened data to obtain adjusted data;

[0109] The data integration unit 3024 is used to integrate the adjusted data according to preset department categories to obtain medical data of multiple categories.

[0110] Optionally, the data integration unit 3024 is specifically used for:

[0111] Obtain a keyword set corresponding to each department category in the preset department categories to obtain multiple department keyword sets, each department keyword set corresponding to a department category;

[0112] Matching the adjusted data with the multiple department keyword sets in sequence to obtain multiple matching result pairs, wherein the matching result pairs include matched medical statements and matched department categories;

[0113] The adjusted data are integrated according to the matched department categories in each matching result pair to obtain multiple categories of medical data, each category of medical data including multiple medical-related statements belonging to this category.

[0114] Optionally, the identification module 303 includes:

[0115] A sequence conversion unit 3031, used for converting the medical data of the multiple categories into multiple word vector sequences;

[0116] A sequence tagging unit 3032 is used to tag the multiple word vector sequences to obtain multiple tagged vector sequences;

[0117] The entity recognition unit 3033 is used to call the recognition neural network from the preset coding model to perform entity recognition on the multiple annotated vector sequences to obtain entity recognition results, wherein the entity recognition results include multiple medical entity feature vectors and the part of speech corresponding to each medical entity;

[0118] The name screening unit 3034 is used to screen out multiple disease names from the multiple medical entity feature vectors of the entity recognition result to obtain a disease name recognition result, wherein the disease name recognition result includes multiple disease names with noun attributes.

[0119] Optionally, the entity identification unit 3033 is specifically used for:

[0120] Calling the BiLSTM layer in the recognition neural network in the preset pairing model to perform feature extraction on the multiple annotated sequences to obtain multiple prediction label sequences, wherein each prediction label in the prediction label sequence corresponds to each word in the annotated sequence;

[0121] Calling the CRF layer of the recognition neural network to perform context semantic constraint adjustment on the multiple predicted label sequences to obtain an optimal label sequence;

[0122] The part of speech corresponding to each medical entity feature vector in the optimal label sequence is determined to obtain an optimal part of speech sequence, and the optimal label sequence and the optimal part of speech sequence are determined as entity recognition results.

[0123] Optionally, the code matching module 304 is specifically used for:

[0124] Performing full-word matching on each disease name feature vector in the disease name recognition result with a preset standard data set to obtain a full-word matching result;

[0125] If the whole word matching result is not empty, the matched multiple words are determined as a candidate code set;

[0126] If the full-word matching result is empty, then use the distributed full-text search engine Elasticsearch to recall related words to obtain a recall candidate set, which includes multiple recalled words;

[0127] Calculating the similarity between the recall candidate set and the disease name recognition result to obtain a similarity sequence arranged from large to small, and determining a preset number of words at the front of the similarity sequence as a candidate code set;

[0128] The twin network in the pairing model is called to pair the candidate pairing set with the preset standard disease name to generate a pairing result, wherein the pairing result includes multiple phrase pairs, each phrase pair includes the original disease name and the corresponding standard disease name.

[0129] Optionally, the generating module 305 is specifically used for:

[0130] Determine a plurality of target original disease names based on a plurality of phrase pairs in the coding result, and import the plurality of target original disease names into an original phrase column;

[0131] Determine a plurality of target standard disease names based on a plurality of phrase pairs in the matching result, and import the plurality of target standard disease names into a standard phrase column;

[0132] Searching for at least one candidate code in a preset coding standard catalog based on each target standard disease name;

[0133] Determine the candidate code with the latest update time among the at least one candidate code as the standard disease code corresponding to each target standard disease name, obtain multiple standard disease codes, and import the multiple standard disease codes into the data code column;

[0134] The original phrase sequence, the standard phrase sequence and the data encoding sequence are arranged in sequence to generate a pairing code list.

[0135] In an embodiment of the present invention, the recognition neural network in the improved coding model is used to accurately segment the disease names to be coded, and then the distributed full-text retrieval engine in the improved coding model is used to recall them. Finally, the twin network in the improved coding model is used to perform coding, thereby improving the coding accuracy and recall rate of the improved coding model for disease names.

[0136] above Figure 3 to Figure 4The device for generating the disease name code list in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The device for generating the disease name code list in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0137] Figure 5 1 is a schematic diagram of a device for generating a disease name code list provided by an embodiment of the present invention. The device 500 for generating a disease name code list may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 510 (for example, one or more processors) and a memory 520, and one or more storage media 530 (for example, one or more mass storage devices) storing application programs 533 or data 532. Among them, the memory 520 and the storage medium 530 may be temporary storage or permanent storage. The program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the device 500 for generating a disease name code list. Furthermore, the processor 510 may be configured to communicate with the storage medium 530, and execute a series of instruction operations in the storage medium 530 on the device 500 for generating a disease name code list.

[0138] The device 500 for generating the disease name code list may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and / or one or more operating systems 531, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 5 The structure of the device for generating the disease name code list shown does not constitute a limitation on the device for generating the disease name code list, and may include more or less components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0139] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for generating a disease name code list.

[0140] Furthermore, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0141] The blockchain referred to in this invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0143] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0144] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a disease name code list, characterized in that: include: Acquire an initial data set, wherein the initial data set includes a plurality of original medical statements that need to be coded; Preprocessing the initial data set to obtain medical data of multiple department categories, wherein the preprocessing includes data cleaning and data integration, and the medical data of each department category includes multiple medical-related statements belonging to the department category; Calling the recognition neural network in the preset coding model to perform entity recognition on the medical data of the multiple department categories to obtain disease name recognition results; Calling the distributed full-text search engine and the twin network in the matching model to match the disease name recognition result with a preset standard data set to obtain a matching result; Generate a matching list based on the matching result, wherein the matching list includes an original phrase column, a standard phrase column and a data encoding column; The calling of the recognition neural network in the preset coding model performs entity recognition on the medical data of the multiple department categories to obtain the disease name recognition results, including: Convert the medical data of the multiple department categories into multiple word vector sequences; Annotating the plurality of word vector sequences to obtain a plurality of annotated vector sequences; Calling a recognition neural network from a preset pairing model to perform entity recognition on the multiple annotated vector sequences to obtain entity recognition results, wherein the entity recognition results include multiple medical entity feature vectors and a part of speech corresponding to each medical entity; Filtering out multiple disease names from multiple medical entity feature vectors of the entity recognition result to obtain a disease name recognition result, wherein the disease name recognition result includes multiple disease names with noun attributes; The calling of the distributed full-text search engine and the twin network in the matching model matches the disease name recognition result with a preset standard data set to obtain a matching result, including: Performing full-word matching on each disease name feature vector in the disease name recognition result with a preset standard data set to obtain a full-word matching result; If the whole word matching result is not empty, the matched multiple words are determined as a candidate code set; If the full-word matching result is empty, then use the distributed full-text search engine Elasticsearch to recall related words to obtain a recall candidate set, which includes multiple recalled words; Calculating the similarity between the recall candidate set and the disease name recognition result to obtain a similarity sequence arranged from large to small, and determining a preset number of words at the front of the similarity sequence as a candidate code set; The twin network in the pairing model is called to pair the candidate pairing set with the preset standard disease name to generate a pairing result, wherein the pairing result includes multiple phrase pairs, each phrase pair includes the original disease name and the corresponding standard disease name.

2. The method for generating a disease name code list according to claim 1, characterized in that: The initial data set is preprocessed to obtain medical data of multiple department categories, including: Performing format adjustment and text error correction on the initial data set to obtain a replaced data text; Screening out non-medical data and duplicate data in the replaced data text to obtain screened data; Complete, modify or delete the missing data and abnormal data in the screened data to obtain adjusted data; The adjusted data are integrated according to preset department categories to obtain medical data of multiple department categories.

3. The method for generating a disease name code list according to claim 2, characterized in that: The adjusted data is integrated according to the preset department categories to obtain medical data of multiple department categories, including: Obtain a keyword set corresponding to each department category in the preset department categories to obtain multiple department keyword sets, each department keyword set corresponding to a department category; Matching the adjusted data with the multiple department keyword sets in sequence to obtain multiple matching result pairs, wherein the matching result pairs include matched medical statements and matched department categories; The adjusted data are integrated according to the matched department categories in each matching result pair to obtain medical data of multiple department categories.

4. The method for generating a disease name code list according to claim 1, characterized in that: The recognition neural network is called from the preset pairing model to perform entity recognition on the multiple annotated vector sequences to obtain entity recognition results, wherein the entity recognition results include multiple medical entity feature vectors and the part of speech corresponding to each medical entity, including: Calling the BiLSTM layer of the recognition neural network in the preset pairing model to perform feature extraction on the multiple annotated sequences to obtain multiple prediction label sequences, wherein each prediction label in the prediction label sequence corresponds to each word in the annotated sequence; Calling the CRF layer of the recognition neural network to perform context semantic constraint adjustment on the multiple predicted label sequences to obtain an optimal label sequence; The part of speech corresponding to each medical entity feature vector in the optimal label sequence is determined to obtain an optimal part of speech sequence, and the optimal label sequence and the optimal part of speech sequence are determined as entity recognition results.

5. The method for generating a disease name code list according to any one of claims 1 to 4, characterized in that: The generating a matching list based on the matching result, wherein the matching list includes an original phrase column, a standard phrase column and a data encoding column, including: Determine multiple target original disease names based on multiple phrase pairs in the matching result, and import the multiple target original disease names into the original phrase column; Determine a plurality of target standard disease names based on a plurality of phrase pairs in the matching result, and import the plurality of target standard disease names into a standard phrase column; Searching for at least one candidate code in a preset coding standard catalog based on each target standard disease name; Determine the candidate code with the latest update time among the at least one candidate code as the standard disease code corresponding to each target standard disease name, obtain multiple standard disease codes, and import the multiple standard disease codes into the data code column; The original phrase sequence, the standard phrase sequence and the data encoding sequence are arranged in sequence to generate a pairing code list.

6. A device for generating a disease name code list, characterized in that: include: An acquisition module, used for acquiring an initial data set, wherein the initial data set includes a plurality of original medical statements that need to be coded; A preprocessing module, used for preprocessing the initial data set to obtain medical data of multiple department categories, wherein the medical data of each department category includes multiple medical-related sentences belonging to this department category; A recognition module, used to call a recognition neural network in a preset coding model to perform entity recognition on the medical data of the multiple department categories to obtain a disease name recognition result; A matching module, used for calling the distributed full-text search engine and the twin network in the matching model to match the disease name recognition result with a preset standard data set to obtain a matching result; A generating module, configured to generate a matching list based on the matching result, wherein the matching list includes an original phrase list, a standard phrase list and a data encoding list; The identification module comprises: A sequence conversion unit, used for converting the medical data of the multiple department categories into multiple word vector sequences; A sequence tagging unit, used to tag the multiple word vector sequences to obtain multiple labeled vector sequences; An entity recognition unit, used to call a recognition neural network from a preset pairing model to perform entity recognition on the multiple annotated vector sequences to obtain an entity recognition result, wherein the entity recognition result includes multiple medical entity feature vectors and a part of speech corresponding to each medical entity; A name screening unit, used to screen out a plurality of disease names from a plurality of medical entity feature vectors of the entity recognition result to obtain a disease name recognition result, wherein the disease name recognition result includes a plurality of disease names with noun attributes; The code matching module is specifically used for: Performing full-word matching on each disease name feature vector in the disease name recognition result with a preset standard data set to obtain a full-word matching result; If the whole word matching result is not empty, the matched multiple words are determined as a candidate code set; If the full-word matching result is empty, then use the distributed full-text search engine Elasticsearch to recall related words to obtain a recall candidate set, which includes multiple recalled words; Calculating the similarity between the recall candidate set and the disease name recognition result to obtain a similarity sequence arranged from large to small, and determining a preset number of words at the front of the similarity sequence as a candidate code set; The twin network in the pairing model is called to pair the candidate pairing set with the preset standard disease name to generate a pairing result, wherein the pairing result includes multiple phrase pairs, each phrase pair includes the original disease name and the corresponding standard disease name.

7. The device for generating a disease name code list according to claim 6, characterized in that: The preprocessing modules include: A data error correction unit, used for performing format adjustment and text error correction on the initial data set to obtain a replaced data text; A data screening unit, used for screening out non-medical data and duplicate data in the replaced data text to obtain screened data; A data adjustment unit is used to complete, modify or delete missing data and abnormal data in the screened data to obtain adjusted data; The data integration unit is used to integrate the adjusted data according to preset department categories to obtain medical data of multiple department categories.

8. The device for generating a disease name code list according to claim 7, characterized in that: The data integration unit is specifically used for: obtaining a keyword set corresponding to each department category in the preset department categories to obtain multiple department keyword sets, each department keyword set corresponds to a department category; matching the adjusted data with the multiple department keyword sets in sequence to obtain multiple matching result pairs, the matching result pairs including matched medical statements and matched department categories; integrating the adjusted data according to the matched department categories in each matching result pair to obtain medical data of multiple department categories.

9. A device for generating a disease name code list, characterized in that: The device for generating the disease name code list includes: a memory and at least one processor, the memory stores instructions, and the memory and the at least one processor are interconnected via a line; The at least one processor calls the instructions in the memory to enable the device for generating a disease name code list to execute the method for generating a disease name code list as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed by a processor, implement the method for generating a disease name code list as described in any one of claims 1 to 5.

Images