Disease name coding method, device, equipment and storage medium

Abstract

The embodiment of the present invention discloses a method for encoding a disease name, the method comprising: obtaining a disease name to be encoded, determining a corresponding first vector phrase matrix; inputting the first vector phrase matrix into a medical entity recognition model, determining at least one corresponding disease label, logically reconstructing at least one disease label, and obtaining a reconstructed disease name, wherein the disease label includes one or more of a part, a symptom, and a cause; determining a second vector phrase matrix corresponding to the reconstructed disease name, inputting the second vector phrase matrix into a disease diagnosis category coding prediction model, and obtaining a plurality of pending category codes; determining a plurality of candidate disease names according to the plurality of pending category codes; calculating the similarity between the disease name to be encoded and each candidate disease name, determining a target disease name according to the similarity, and using the detailed code corresponding to the target disease name as the target code. The present invention can improve the accuracy and efficiency of Chinese disease name coding.

Description

Technical Field

[0001] The present invention relates to the fields of computer technology and medical data technology, and in particular to a disease name encoding method, device, computer equipment and computer-readable storage medium. Background Art

[0002] The International Classification of Diseases (ICD) is an international unified disease classification method developed by the WHO. It classifies diseases according to their etiology, pathology, clinical manifestations, and anatomical location, making them an orderly combination and expressing them in a coding system. ICD standardizes and formats disease names. This is the application basis of clinical information systems such as medical informatization and hospital information management. The standardization of Chinese medical terminology lags far behind the standardization of Western medical terminology. The standardization process of medical terminology in my country began with the introduction and translation of a large number of international standards. The Chinese version of ICD-9-CM3 was designated by the National Health and Family Planning Commission as the recommended standard for clinical surgical operation terminology.

[0003] However, due to the wide range of diseases, various types and complex naming, there are great differences in the use of disease codes by domestic medical institutions in disease reporting, medical record statistics management and medical insurance reimbursement. Each of them has made localized modifications to the international standard ICD codes provided by WHO, resulting in multi-source heterogeneity, missing and difficult to unify codes. Summary of the invention

[0004] Based on this, it is necessary to propose a disease name encoding method, device, computer equipment and computer-readable storage medium to address the above problems.

[0005] In the first part of the present invention, a method for encoding a disease name is provided, the method comprising:

[0006] Obtaining a disease name to be encoded, and determining a first vector phrase matrix corresponding to the disease name to be encoded;

[0007] Inputting the first vector phrase matrix into a preset medical entity recognition model, determining at least one disease label corresponding to the disease name to be encoded, and logically reconstructing the at least one disease label to obtain a reconstructed disease name, wherein the disease label includes one or more of a part, a symptom, and a cause of disease;

[0008] Determine a second vector phrase matrix corresponding to the reconstructed disease name, input the second vector phrase matrix into a preset disease diagnosis category code prediction model, and obtain a plurality of to-be-determined category codes;

[0009] Determining multiple candidate disease names according to the multiple pending category codes;

[0010] The similarity between the disease name to be coded and each candidate disease name is calculated, and a target disease name is determined from multiple candidate disease names according to the similarity, and the detailed code corresponding to the target disease name is used as the target code.

[0011] Optionally, the step of determining a first vector phrase matrix corresponding to the disease name to be encoded further includes:

[0012] Decomposing the disease name to be encoded to obtain multiple Chinese characters after decomposition;

[0013] The vector of each Chinese character is determined according to a preset Chinese character vector library, and a first vector phrase matrix is ​​formed based on the name of the disease to be encoded.

[0014] Optionally, the method further includes:

[0015] Obtain medical data containing disease names and annotate the disease labels in each piece of medical data;

[0016] Performing Chinese character segmentation processing on the annotated medical data to obtain multiple Chinese characters, determining the vector of each Chinese character according to a preset Chinese character vector library, and constructing a third vector phrase matrix corresponding to the medical data;

[0017] Based on the Bi-LSTM training model, classification training is performed on the medical data annotated with disease labels to construct the preset medical entity recognition model.

[0018] Optionally, the step of logically reconstructing the at least one disease label to obtain a reconstructed disease name further includes:

[0019] According to a preset label sorting, logically reconstructing the at least one disease label to obtain a reconstructed disease name;

[0020] The step of determining a second vector phrase matrix corresponding to the reconstructed disease name also includes:

[0021] Decomposing the reconstructed disease name to obtain multiple decomposed Chinese characters;

[0022] The vector of each Chinese character is determined according to a preset Chinese character vector library, and a second vector phrase matrix is ​​formed based on the reconstructed disease name.

[0023] Optionally, the step of inputting the second vector phrase matrix into a preset disease diagnosis category code prediction model to obtain a plurality of pending category codes further includes:

[0024] The preset disease diagnosis category coding prediction model includes a plurality of Bi-LSTM models trained according to different parameters, and the Bi-LSTM model includes an LSTM layer, a fully connected layer, and an output layer;

[0025] The category codes are predicted according to multiple Bi-LSTM models to obtain multiple category codes and corresponding probability values;

[0026] N pending category codes are determined according to the probability values, where the probability values ​​of the N pending category codes are greater than the probability values ​​of other category codes, wherein N is equal to 3.

[0027] Optionally, the step of determining a plurality of candidate disease names according to the plurality of pending category codes further includes:

[0028] In a preset disease code database, multiple candidate disease names corresponding to the multiple pending category codes are determined; wherein the codes of the candidate disease names correspond to the upper-level category codes or lower-level detailed codes of the multiple pending category codes.

[0029] Optionally, the step of calculating the similarity between the disease name to be encoded and each candidate disease name further includes:

[0030] According to the preset medical entity word semantic similarity calculation model, the similarity between the disease name to be encoded and each candidate disease name is calculated;

[0031] Among them, according to the formula

[0032]

[0033] Calculate the similarity sim(T1, T2);

[0034] Among them, T1 represents the name of the disease to be coded, T2 represents the candidate disease name, and T1 is composed of Chinese characters A1, A2, ..., A m T2 is composed of Chinese characters B1, B2, ..., B n w represents a Chinese character, idf(w) represents the inverse text frequency of Chinese character w, that is, the total number of disease names in the preset disease coding database divided by the number of disease names containing Chinese character w; maxSim(w, T2) is the sum of the Chinese character w in T1 and the Chinese characters B1, B2, ..., B2 in T2. n The highest similarity between them, that is, maxSim(w, T2) = max(sim(w, B i )), i is 1, 2, 3, ..., n, maxSim(w, T1) is the sum of the Chinese character w in T2 and the Chinese characters A1, A2, ..., A in T1. mThe highest semantic similarity, that is, maxSim(w, T1) = max(sim(w, A i )); sim(w, A i ) or sim(w, B i ) is the Chinese character w and the Chinese character A i or B i The cosine similarity is calculated between them.

[0035] In the second part of the present invention, a device for encoding a disease name is provided, the device comprising:

[0036] A disease name disassembly module, used for obtaining a disease name to be encoded and determining a first vector phrase matrix corresponding to the disease name to be encoded;

[0037] a logic reconstruction module, configured to input the first vector phrase matrix into a preset medical entity recognition model, determine at least one disease label corresponding to the disease name to be encoded, and logically reconstruct the at least one disease label to obtain a reconstructed disease name, wherein the disease label includes one or more of a part, a symptom, and a cause of disease;

[0038] A category code prediction module, used to determine a second vector phrase matrix corresponding to the reconstructed disease name, input the second vector phrase matrix into a preset disease diagnosis category code prediction model, and obtain a plurality of pending category codes;

[0039] The similarity calculation module determines multiple candidate disease names based on the multiple pending category codes; calculates the similarity between the disease name to be coded and each candidate disease name, determines the target disease name from the multiple candidate disease names based on the similarity, and uses the detailed code corresponding to the target disease name as the target code.

[0040] In the third part of the present invention, a computer device is provided, which includes a memory and a processor, wherein the memory has executable code, and when the executable code runs on the processor, the disease name encoding method as described above is implemented.

[0041] In the fourth part of the present invention, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the processor executes the disease name encoding method as described above.

[0042] The embodiments of the present invention have the following beneficial effects:

[0043] After adopting the above-mentioned disease name encoding method, device, computer equipment and computer-readable storage medium, when standardized encoding of disease names is required, for the disease name to be encoded, a first vector phrase matrix corresponding to the disease name to be encoded is determined; the first vector phrase matrix is ​​input into a preset medical entity recognition model to determine at least one disease label corresponding to the disease name to be encoded, and the at least one disease label is logically reconstructed to obtain a reconstructed disease name, wherein the disease label includes one or more of location, symptoms, and cause; a second vector phrase matrix corresponding to the reconstructed disease name is determined, and the second vector phrase matrix is ​​input into a preset disease diagnosis category coding prediction model to obtain multiple pending category codes; multiple candidate disease names are determined based on the multiple pending category codes; the similarity between the disease name to be encoded and each candidate disease name is calculated, and a target disease name is determined from the multiple candidate disease names based on the similarity, and the detailed code corresponding to the target disease name is used as the target code.

[0044] That is to say, in the embodiment of the present application, the medical entity recognition model is used to identify the constituent components (location, symptoms, cause) in the disease name, and reconstruct them in a certain order, so that each Chinese name of the disease diagnosis is arranged in a fixed order (location, symptoms, cause, supplementary instructions). In this way, the reconstructed phrase matrix composed of multiple Chinese characters can better represent and express the original input disease name, so as to improve the accuracy of the disease name encoding.

[0045] Using vectors to express Chinese characters can improve the accuracy of similarity calculations compared to general string-based calculations.

[0046] A disease diagnosis category code prediction model was constructed using natural language processing technology, artificial intelligence technology, and deep neural network technology. First, the category code of the input disease name was predicted, and when the candidate category code was obtained, the similarity was calculated between the disease name corresponding to the detailed code corresponding to the candidate category code. This is more efficient and less time-consuming than calculating the similarity between the disease names with all detailed codes. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] 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 only 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.

[0048] in:

[0049] Figure 1A schematic diagram of a flow chart of a method for encoding a disease name in one embodiment;

[0050] Figure 2 A schematic diagram of an implementation method of a disease name encoding method in an embodiment;

[0051] Figure 3 A schematic diagram of the composition of a coding device for a disease name in one embodiment;

[0052] Figure 4 The figure is a schematic diagram of the structure of a computer device for running the above-mentioned disease name encoding method in one embodiment. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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.

[0054] In this embodiment, a disease name coding method is provided, which can automatically perform natural language processing on Chinese disease names from different medical institutions or different sources, and standardize them into universal standard ICD codes, thereby providing solid data standardization support for management decisions on medical quality and medical insurance funds, and enabling ICD disease information to be shared to the greatest extent possible.

[0055] For details, see Figure 1 , Figure 1 A flowchart of the above-mentioned disease name-based coding method is given, including Figure 1 Steps S102-S110 shown:

[0056] Step S102: Obtain the name of the disease to be encoded, and determine the first vector phrase matrix corresponding to the name of the disease to be encoded.

[0057] ICD code, namely the International Classification of Diseases (ICD). Since the national standard of the Classification of Diseases and Codes (ICD) appeared relatively late, provinces and cities made localized modifications to the Classification of Diseases and Codes (ICD) provided by the WHO to form localized Classification of Diseases and Codes (ICD). In addition, there are currently multiple national standard Classification of Diseases and Codes (ICD). This has resulted in numerous versions of the Classification of Diseases and Codes (ICD), and differences between versions, making it difficult to unify. The same disease code (ICD) has multiple different medical disease terms, and the same medical disease term has multiple different names for clinicians with different medical backgrounds and terminology usage habits.

[0058] Therefore, in order to standardize different disease names or disease diagnosis information, in this embodiment, the disease names need to be standardized to obtain their standardized codes and disease names. Here, the names that need to be standardized are the disease names to be coded, and the disease names to be coded are Chinese names, that is, the disease name coding method provided in this embodiment is a standardized method for Chinese disease names, not for other languages.

[0059] In the specific implementation, the disease name to be encoded needs to be disassembled first, and the disassembly here is to disassemble it into Chinese characters one by one, and then multiple Chinese characters after disassembly can be obtained. For the disassembled Chinese characters, it is necessary to further query according to the preset Chinese character vector library to determine the vector of each Chinese character, wherein the preset Chinese character vector library contains the corresponding vector expression form of the Chinese characters used in medical terms. Therefore, here, after the disease name to be encoded is disassembled, the vector corresponding to each Chinese character can be obtained based on the Chinese character vector library.

[0060] Then, a first vector phrase matrix can be further formed according to the original arrangement order of the Chinese characters in the disease name to be encoded, and the first vector phrase matrix is ​​a vector phrase matrix corresponding to the disease name to be encoded.

[0061] Furthermore, in this embodiment, the construction process of the above-mentioned Chinese character vector library is explained.

[0062] First, a large amount of text data in the medical field is collected to build a medical corpus. Text data in the medical field includes but is not limited to the disease classification and codes, surgical operation classification and codes, medication instructions, disease clinical guidelines and other public data in the country and various provinces and cities. All existing medical corpora are segmented by word to make them independent Chinese characters, numbers or special symbols. If English words are encountered, they are not segmented, and a series of English words are regarded as a whole. The word2vec tool is used to train the vectors of Chinese characters in the medical corpus. The training parameters of the word2vec tool are set as follows: Size (word vector dimension): 200; Alpha (learning rate): 0.025; Window (learning window): 5; Min_count (lower limit of occurrence): 5; SG (training model): 1 (skip-gram); Hs (Huffman tree): 0; Negative (negative sampling): 10; Iter (number of training rounds): 50. After the training is completed, the vector expression of Chinese characters is obtained, that is, the construction of the Chinese character vector library is completed.

[0063] Step S104: input the first vector phrase matrix into a preset medical entity recognition model, determine at least one disease label corresponding to the disease name to be encoded, logically reconstruct the at least one disease label, and obtain a reconstructed disease name, wherein the disease label includes one or more of a location, a symptom, and a cause.

[0064] In order to encode the disease name to be encoded, it is first necessary to identify the medical information contained in the disease name to be encoded. Specifically, the first vector phrase matrix needs to be input into a preset medical entity recognition model to determine at least one disease label corresponding to the disease name to be encoded, and the disease label includes one or more of the part, symptom, and cause. In other words, it is necessary to obtain the specific content of the disease label such as the part, symptom, and cause corresponding to the disease according to the first vector phrase matrix, so as to facilitate the subsequent standardization of the disease name based on these contents.

[0065] The medical entity recognition model is a pre-set model that can identify disease labels such as location, symptoms, and causes, for example, a neural network model. In this step, the first vector phrase matrix is ​​input into the preset medical entity recognition model to obtain at least one disease label output by the preset medical entity recognition model. The disease labels here include one or more of location, symptoms, and causes.

[0066] Furthermore, how to construct the preset medical entity recognition model is explained.

[0067] A large amount of text data in the medical field is collected. The text data here refers to medical data containing disease names, and the symptoms, causes, and parts of the text data are manually standardized. Text data in the medical field includes but is not limited to publicly available data such as national and provincial disease classifications and codes, disease clinical guidelines, etc. Then, the disease labels such as parts, symptoms, causes, etc. in each piece of medical data are annotated (for example, manually).

[0068] The annotated medical data is segmented into Chinese characters to make them independent Chinese characters, numbers or special symbols. If English words are encountered, they are not segmented and a series of English words are regarded as a whole to obtain multiple Chinese characters.

[0069] The vector of each Chinese character is determined according to the preset Chinese character vector library to form a corresponding third vector phrase matrix.

[0070] A Bi-LSTM training model is constructed to classify and train medical data with disease labels such as symptoms, causes, and locations (here refers to the third vector phrase matrix). After the training is completed, a trained medical entity recognition model is obtained, which can automatically recognize and label the symptoms, causes, and locations of the input medical terminology Chinese characters (corresponding vector phrase matrix).

[0071] Among them, the training parameters of the Bi-LSTM model are set as follows: Num_lstm: 250; Rate_drop_lstm (number of random disconnections in the LSTM layer): 0.25; Rate_drop_dense (number of random disconnections in the fully connected layer): 0.25; fully connected layer activation function: relu; output layer activation function: softmax; number of output layer categories: 3; Loss (loss calculation method): categorical_crossentropy; optimization method: nadam; Epochs (number of training rounds): 100; Early stopping parameter: 30; Batch_size: 256; Shuffle (random order): True.

[0072] After obtaining at least one disease label corresponding to the first vector phrase matrix corresponding to the disease name to be encoded, it is necessary to further perform logical reconstruction of the name, wherein the at least one disease label is logically reconstructed according to a preset label sorting to obtain a reconstructed disease name. For example, at least one disease label is logically reconstructed in the order of location, symptoms, cause, and supplementary instructions to obtain a reconstructed disease name corresponding to the disease name to be encoded.

[0073] Step S106: Determine a second vector phrase matrix corresponding to the reconstructed disease name, input the second vector phrase matrix into a preset disease diagnosis category code prediction model, and obtain a plurality of to-be-determined category codes.

[0074] For the reconstructed disease name, before encoding, the corresponding second vector phrase matrix needs to be determined. Specifically, the reconstructed disease name is disassembled, that is, disassembled into Chinese characters one by one, and multiple Chinese characters after disassembly are obtained; then the vector of each Chinese character is determined according to the preset Chinese character vector library, and then the second vector phrase matrix is ​​formed according to the order of the Chinese characters in the reconstructed disease name.

[0075] Then, the second vector phrase matrix is ​​input into a preset disease diagnosis category code prediction model to obtain a plurality of pending category codes. The preset disease diagnosis category code prediction model may be a neural network model or other models, for example, a Bi-LSTM model.

[0076] In a specific embodiment, the preset disease diagnosis category coding prediction model includes multiple Bi-LSTM models trained according to different parameters, and the Bi-LSTM model includes an LSTM layer, a fully connected layer, and an output layer. Among them, the training parameters of the Bi-LSTM model are set as follows: Num_lstm: 300; Rate_drop_lstm (number of random disconnections of the LSTM layer): 0.35; Rate_drop_dense (number of random disconnections of the fully connected layer): 0.35; fully connected layer activation function: relu; output layer activation function: softmax; number of output layer categories: 1965; Loss (loss calculation method): categorical_crossentropy; optimization method: nadam; Epochs (number of training rounds): 200; Early stopping parameter: 30; Batch_size: 128; Shuffle (random order): True.

[0077] Different parameters are set to train multiple Bi-LSTM models and combine them into a voting decision layer; thus, the preset disease diagnosis category code prediction model is constructed, and the input disease name can be classified and multiple category codes with the highest probability can be output. Specifically, the category codes are predicted according to multiple Bi-LSTM models to obtain multiple category codes and corresponding probability values; then, N pending category codes are determined according to the probability values, and the probability values ​​of the N pending category codes are greater than the probability values ​​of other category codes, where N is equal to 3.

[0078] Step S108: Determine a plurality of candidate disease names according to the plurality of pending category codes.

[0079] In this step, after determining the multiple pending category codes with the largest probability values, it is necessary to determine the multiple candidate disease names corresponding to the multiple pending category codes in the preset disease code database. Among them, the preset disease code database is the target disease code database after standardization here, which contains multiple levels of category codes and corresponding disease names. Therefore, the candidate disease names here include disease names corresponding to multiple category codes corresponding to the category codes (or detailed codes) of one or more levels or one or more levels above the determined multiple pending category codes. In this way, multiple candidate disease names that are greater in number than the number of pending category codes are determined.

[0080] Step S110: Calculate the similarity between the disease name to be coded and each candidate disease name, determine the target disease name from multiple candidate disease names based on the similarity, and use the detailed code corresponding to the target disease name as the target code.

[0081] In order to determine the final code, it is necessary to calculate the similarity between the disease name to be coded and each candidate disease name, and determine a candidate disease name as the target disease name and the corresponding category code as the target code based on the similarity.

[0082] In a specific embodiment, the similarity between the disease name to be encoded and each candidate disease name is calculated according to a preset medical entity word semantic similarity calculation model. Specifically, according to the formula

[0083]

[0084] The similarity sim(T1, T2) is calculated.

[0085] Among them, T1 represents the name of the disease to be coded, T2 represents the candidate disease name, and T1 is composed of Chinese characters A1, A2, ..., A m T2 is composed of Chinese characters B1, B2, ..., B n w represents a Chinese character, idf(w) represents the inverse text frequency of Chinese character w, that is, the total number of disease names in the preset disease coding database divided by the number of disease names containing Chinese character w; maxSim(w, T2) is the sum of the Chinese character w in T1 and the Chinese characters B1, B2, ..., B2 in T2. n The highest similarity between them, that is, maxSim(w, T2) = max(sim(w, B i )), i is 1, 2, 3, ..., n, maxSim(w, T1) is the sum of the Chinese character w in T2 and the Chinese characters A1, A2, ..., A in T1. mThe highest semantic similarity, that is, maxSim(w, T1) = max(sim(w, A i )); sim(w, A i ) or sim(w, B i ) is the Chinese character w and the Chinese character A i or B i The cosine similarity is calculated between them.

[0086] For further information, see Figure 2 , a schematic diagram of the implementation of a disease name encoding method is given, in which, for the disease name to be encoded, logic reconstruction, category coding prediction, and similarity calculation are required to obtain the target disease name and target coding, completing the standardization process of disease name encoding.

[0087] In another embodiment, see Figure 3 , a schematic diagram of the composition of the coding device for the disease name is given, wherein the coding device for the disease name includes:

[0088] The disease name disassembling module 102 is used to obtain the disease name to be encoded and determine the first vector phrase matrix corresponding to the disease name to be encoded;

[0089] A logic reconstruction module 104 is used to input the first vector phrase matrix into a preset medical entity recognition model, determine at least one disease label corresponding to the disease name to be encoded, and perform logic reconstruction on the at least one disease label to obtain a reconstructed disease name, wherein the disease label includes one or more of a part, a symptom, and a cause of disease;

[0090] A category code prediction module 106 is used to determine a second vector phrase matrix corresponding to the reconstructed disease name, input the second vector phrase matrix into a preset disease diagnosis category code prediction model, and obtain a plurality of pending category codes;

[0091] The similarity calculation module 108 determines multiple candidate disease names based on the multiple pending category codes; calculates the similarity between the disease name to be coded and each candidate disease name, determines the target disease name from the multiple candidate disease names based on the similarity, and uses the detailed code corresponding to the target disease name as the target code.

[0092] In one embodiment, the disease name disassembly module 102 is also used to disassemble the disease name to be encoded to obtain multiple Chinese characters after the disassembly; determine the vector of each Chinese character according to a preset Chinese character vector library, and form a first vector phrase matrix based on the disease name to be encoded.

[0093] In one embodiment, the logic reconstruction module 104 is also used to obtain medical data containing disease names, and to annotate the disease labels in each piece of medical data; to perform Chinese character segmentation processing on the annotated medical data to obtain multiple Chinese characters, to determine the vector of each Chinese character according to a preset Chinese character vector library, and to construct a third vector phrase matrix corresponding to the medical data; to perform classification training on the medical data annotated with disease labels based on the Bi-LSTM training model to construct the preset medical entity recognition model.

[0094] In one embodiment, the logic reconstruction module 104 is further used to perform logic reconstruction on the at least one disease label according to a preset label sorting to obtain a reconstructed disease name.

[0095] In one embodiment, the category code prediction module 106 is further used to decompose the reconstructed disease name to obtain multiple Chinese characters after decomposition; determine the vector of each Chinese character according to a preset Chinese character vector library, and form a second vector phrase matrix based on the reconstructed disease name.

[0096] In one embodiment, the category code prediction module 106 is also used for the preset disease diagnosis category code prediction model to include multiple Bi-LSTM models trained according to different parameters, and the Bi-LSTM model includes an LSTM layer, a fully connected layer, and an output layer; the category code is predicted according to the multiple Bi-LSTM models to obtain multiple category codes and corresponding probability values; N pending category codes are determined according to the probability values, and the probability values ​​of the N pending category codes are greater than the probability values ​​of other category codes, wherein N is equal to 3.

[0097] In one embodiment, the similarity calculation module 108 is also used to determine a plurality of candidate disease names corresponding to the plurality of pending category codes in a preset disease code database; wherein the codes of the candidate disease names correspond to the upper-level category codes or the lower-level detailed codes of the plurality of pending category codes.

[0098] In one embodiment, the similarity calculation module 108 is further used to calculate the similarity between the disease name to be encoded and each candidate disease name according to a preset medical entity word semantic similarity calculation model; wherein, according to the formula

[0099]

[0100] Calculate the similarity sim(T1, T2); wherein T1 represents the name of the disease to be encoded, T2 represents the candidate disease name, T1 is composed of Chinese characters A1, A2, ..., A m T2 is composed of Chinese characters B1, B2, ..., B nw represents a Chinese character, idf(w) represents the inverse text frequency of Chinese character w, that is, the total number of disease names in the preset disease coding database divided by the number of disease names containing Chinese character w; maxSim(w, T2) is the sum of the Chinese character w in T1 and the Chinese characters B1, B2, ..., B2 in T2. n The highest similarity between them, that is, maxSim(w, T2) = max(sim(w, B i )), i is 1, 2, 3, ..., n, maxSim(w, T1) is the sum of the Chinese character w in T2 and the Chinese characters A1, A2, ..., A in T1. m The highest semantic similarity, that is, maxSim(w, T1) = max(sim(w, A i )); sim(w, A i ) or sim(w, B i ) is the Chinese character w and the Chinese character A i or B i The cosine similarity is calculated between them.

[0101] After adopting the above-mentioned disease name encoding method, device, computer equipment and computer-readable storage medium, when standardized encoding of disease names is required, for the disease name to be encoded, a first vector phrase matrix corresponding to the disease name to be encoded is determined; the first vector phrase matrix is ​​input into a preset medical entity recognition model to determine at least one disease label corresponding to the disease name to be encoded, and the at least one disease label is logically reconstructed to obtain a reconstructed disease name, wherein the disease label includes one or more of location, symptoms, and cause; a second vector phrase matrix corresponding to the reconstructed disease name is determined, and the second vector phrase matrix is ​​input into a preset disease diagnosis category coding prediction model to obtain multiple pending category codes; multiple candidate disease names are determined based on the multiple pending category codes; the similarity between the disease name to be encoded and each candidate disease name is calculated, and a target disease name is determined from the multiple candidate disease names based on the similarity, and the detailed code corresponding to the target disease name is used as the target code.

[0102] That is to say, in the embodiment of the present application, the medical entity recognition model is used to identify the constituent components (location, symptoms, cause) in the disease name, and reconstruct them in a certain order, so that each Chinese name of the disease diagnosis is arranged in a fixed order (location, symptoms, cause, supplementary instructions). In this way, the reconstructed phrase matrix composed of multiple Chinese characters can better represent and express the original input disease name, so as to improve the accuracy of the disease name encoding.

[0103] Using vectors to express Chinese characters can improve the accuracy of similarity calculations compared to general string-based calculations.

[0104] A disease diagnosis category code prediction model was constructed using natural language processing technology, artificial intelligence technology, and deep neural network technology. First, the category code of the input disease name was predicted, and when the candidate category code was obtained, the similarity was calculated between the disease name corresponding to the detailed code corresponding to the candidate category code. This is more efficient and less time-consuming than calculating the similarity between the disease names with all detailed codes.

[0105] Figure 4 FIG. 1 shows an internal structure diagram of a computer device that implements the above-mentioned disease name encoding method in an embodiment. The computer device may be a terminal or a server. Figure 4 As shown, the computer device includes a processor, a memory and a network interface connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the above method. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement the above method. Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0106] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0107] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0108] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for encoding a disease name, characterized in that: The method comprises: Obtain the name of the disease to be encoded, and determine a first vector phrase matrix corresponding to the disease name to be encoded; wherein the step of determining the first vector phrase matrix corresponding to the disease name to be encoded further includes: decomposing the disease name to be encoded, and obtaining a plurality of Chinese characters after decomposition; determining the vector of each Chinese character according to a preset Chinese character vector library, and forming a first vector phrase matrix based on the disease name to be encoded; Input the first vector phrase matrix into a preset medical entity recognition model, determine at least one disease label corresponding to the disease name to be encoded, and logically reconstruct the at least one disease label according to the preset label sorting to obtain the reconstructed disease name, wherein the disease label includes one or more of the location, symptom, and cause; wherein the preset medical entity recognition model is constructed in the following manner: obtain medical data containing disease names, and annotate the disease labels in each piece of medical data; perform Chinese character segmentation processing on the annotated medical data to obtain multiple Chinese characters, determine the vector of each Chinese character according to a preset Chinese character vector library, and construct a third vector phrase matrix corresponding to the medical data; perform classification training on the medical data annotated with disease labels based on the Bi-LSTM training model to construct the preset medical entity recognition model; The reconstructed disease name is disassembled to obtain multiple Chinese characters after the disassembly, the vector of each Chinese character is determined according to a preset Chinese character vector library, and a second vector phrase matrix is ​​formed based on the reconstructed disease name, and the second vector phrase matrix is ​​input into a preset disease diagnosis category code prediction model to obtain multiple pending category codes; wherein, the step of inputting the second vector phrase matrix into the preset disease diagnosis category code prediction model to obtain multiple pending category codes also includes: the preset disease diagnosis category code prediction model includes multiple Bi-LSTM models trained according to different parameters, and the Bi-LSTM model includes an LSTM layer, a fully connected layer, and an output layer; the category code is predicted according to the multiple Bi-LSTM models to obtain multiple category codes and corresponding probability values; N pending category codes are determined according to the probability values, and the probability values ​​of the N pending category codes are greater than the probability values ​​of other category codes, wherein N is equal to 3; Determining multiple candidate disease names according to the multiple pending category codes; Calculate the similarity between the disease name to be coded and each candidate disease name, determine the target disease name from multiple candidate disease names according to the similarity, and use the detailed code corresponding to the target disease name as the target code; The step of calculating the similarity between the disease name to be encoded and each candidate disease name further includes: According to the preset medical entity word semantic similarity calculation model, the similarity between the disease name to be encoded and each candidate disease name is calculated; wherein, according to the formula Calculate the similarity sim(T1,T2); where T1 represents the name of the disease to be encoded, T2 represents the candidate disease name, and T1 consists of Chinese characters A1, A2, ..., A m T2 is composed of Chinese characters B1, B2, ..., B n w represents a Chinese character, idf(w) represents the inverse text frequency of Chinese character w, that is, the total number of disease names in the preset disease coding database divided by the number of disease names containing Chinese character w; maxSim(w,T2) is the difference between Chinese character w in T1 and Chinese characters B1, B2, ..., B2 in T2. n The highest similarity between them, that is, maxSim(w,T2)=max(sim(w,B i ), i is 1, 2, 3, ..., n, maxSim(w, T1) is the sum of the Chinese character w in T2 and the Chinese characters A1, A2, ..., A in T1 m The highest semantic similarity, that is, maxSim(w,T1)=max(sim(w,A i )); sim(w,A i ) or sim(w,B i ) is the Chinese character w and the Chinese character A i or B i The cosine similarity is calculated between them.

2. The disease name encoding method according to claim 1, characterized in that: The step of determining a plurality of candidate disease names according to the plurality of pending category codes further includes: In a preset disease code database, multiple candidate disease names corresponding to the multiple pending category codes are determined; wherein the codes of the candidate disease names correspond to the upper-level category codes or lower-level detailed codes of the multiple pending category codes.

3. A disease name encoding device, characterized in that: The device comprises: A disease name disassembly module is used to obtain the disease name to be encoded and determine a first vector phrase matrix corresponding to the disease name to be encoded; wherein the disease name disassembly module is further used to: disassemble the disease name to be encoded to obtain multiple Chinese characters after disassembly; determine the vector of each Chinese character according to a preset Chinese character vector library, and form a first vector phrase matrix based on the disease name to be encoded; A logic reconstruction module, used for inputting the first vector phrase matrix into a preset medical entity recognition model, determining at least one disease label corresponding to the disease name to be encoded, and logically reconstructing the at least one disease label according to a preset label sorting to obtain a reconstructed disease name, wherein the disease label includes one or more of a site, a symptom, and a cause of disease; wherein the preset medical entity recognition model is constructed in the following manner: obtaining medical data containing disease names, and annotating the disease label in each piece of medical data; performing Chinese character segmentation processing on the annotated medical data to obtain a plurality of Chinese characters, determining the vector of each Chinese character according to a preset Chinese character vector library, and constructing a third vector phrase matrix corresponding to the medical data; and performing classification training on the medical data annotated with disease labels based on a Bi-LSTM training model to construct the preset medical entity recognition model; A category code prediction module is used to decompose the reconstructed disease name, obtain multiple Chinese characters after decomposition, determine the vector of each Chinese character according to a preset Chinese character vector library, and form a second vector phrase matrix based on the reconstructed disease name, input the second vector phrase matrix into a preset disease diagnosis category code prediction model, and obtain multiple pending category codes; wherein the category code prediction module is also used for: the preset disease diagnosis category code prediction model includes multiple Bi-LSTM models trained according to different parameters, and the Bi-LSTM model includes an LSTM layer, a fully connected layer, and an output layer; predict the category code according to multiple Bi-LSTM models to obtain multiple category codes and corresponding probability values; determine N pending category codes according to the probability values, and the probability values ​​of the N pending category codes are greater than the probability values ​​of other category codes, wherein N is equal to 3; A similarity calculation module is configured to determine a plurality of candidate disease names according to the plurality of pending category codes; calculate the similarity between the disease name to be coded and each candidate disease name, determine a target disease name from the plurality of candidate disease names according to the similarity, and use the detailed category code corresponding to the target disease name as the target code; The similarity calculation module is also used to calculate the similarity between the disease name to be encoded and each candidate disease name according to a preset medical entity word semantic similarity calculation model; according to the formula Calculate the similarity sim(T1,T2); where T1 represents the name of the disease to be encoded, T2 represents the candidate disease name, and T1 consists of Chinese characters A1, A2, ..., A m T2 is composed of Chinese characters B1, B2, ..., B n w represents a Chinese character, idf(w) represents the inverse text frequency of Chinese character w, that is, the total number of disease names in the preset disease coding database divided by the number of disease names containing Chinese character w; maxSim(w,T2) is the difference between Chinese character w in T1 and Chinese characters B1, B2, ..., B2 in T2. n The highest similarity between them, that is, maxSim(w,T2)=max(sim(w,B i ), i is 1, 2, 3, ..., n, maxSim(w, T1) is the sum of the Chinese character w in T2 and the Chinese characters A1, A2, ..., A in T1 m The highest semantic similarity, that is, maxSim(w,T1)=max(sim(w,A i )); sim(w,A i ) or sim(w,B i ) is the Chinese character w and the Chinese character A i or B i The cosine similarity is calculated between them.

4. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the disease name encoding method as described in any one of claims 1 to 2.

5. A computer device, characterized in that: The computer device includes a memory and a processor, wherein the memory has executable code, and when the executable code runs on the processor, the disease name encoding method as described in any one of claims 1 to 2 is implemented.

Images