Medical record filing method and device

Abstract

The present invention provides a medical record archiving method and device, which relates to the technical field of archive management, including: for each known disease entity, first construct multiple sub-graphs of the known disease entity according to a preset attribute type; then for each sub-graph of each known disease entity, use a target neural network to construct a sub-graph network corresponding to the sub-graph; then for each known disease entity, construct a feature weighted vector corresponding to the knowledge graph of the known disease entity based on the feature sub-vectors corresponding to all the sub-graphs of the known disease entity; finally, based on the feature weighted vectors corresponding to the knowledge graph of each known disease entity, the medical record of the patient to be tested is archived. The present invention can achieve effective archiving of the medical record of the patient to be tested and improve the accuracy of archiving the medical record of the patient to be tested.

Description

Technical Field

[0001] The present invention relates to the technical field of file management, and in particular to a medical record filing method and device. Background Art

[0002] In daily life, the diagnosis of a disease is generally given directly by doctors based on the patient's symptoms and examination reports. Due to the large differences in the professional level of different doctors, there may be misjudgments caused by subjective factors. In order to solve this problem, the existing technology combines medical record file recognition with image processing technology, but in actual application, it has the disadvantage of low accuracy in medical record file filing due to the difficulty in feature extraction. Summary of the invention

[0003] The purpose of the present invention is to provide a medical record filing method and device to alleviate the technical problem of low accuracy of medical record filing due to difficulty in feature extraction in the prior art.

[0004] In a first aspect, the present invention provides a method for archiving medical records, which includes: for each known disease entity, constructing multiple sub-graphs of the known disease entity according to a preset attribute type; for each sub-graph of each known disease entity, using a target neural network to construct a sub-graph network corresponding to the sub-graph; wherein the sub-graph networks corresponding to different sub-graphs are represented by different feature sub-vectors; for each known disease entity, constructing a feature weighted vector corresponding to the knowledge graph of the known disease entity based on the feature sub-vector corresponding to all the sub-graphs of the known disease entity; wherein the knowledge graph is composed of all the sub-graphs of the known disease entity; and archiving the medical records of the patient to be tested based on the feature weighted vector corresponding to the knowledge graph of each known disease entity.

[0005] Furthermore, the target neural network includes a first graph neural network layer, a subgraph pooling layer, and a second graph neural network layer connected in sequence, and using the target neural network to construct a subgraph network corresponding to the subgraph spectrum includes: extracting a rooted subgraph around each node of the subgraph spectrum; applying each of the rooted subgraph to the first graph neural network layer and the subgraph pooling layer, respectively, to obtain a corresponding subgraph representation; applying all of the subgraph representations to the second graph neural network layer to obtain a subgraph network corresponding to the subgraph spectrum.

[0006] Furthermore, the medical record archiving method also includes: for each of the subgraph networks, calculating the feature vector of each node in the subgraph network; for each of the subgraph networks, concatenating the feature vectors of all nodes to obtain a feature subvector corresponding to the subgraph network.

[0007] Furthermore, the calculation of the eigenvector of each node in the subgraph network includes: obtaining an adjacency matrix between nodes in the subgraph network; determining a diagonal matrix based on an attribute type corresponding to the subgraph network; performing data processing based on the adjacency matrix and the diagonal matrix to obtain a processed adjacency matrix and a node degree matrix; and calculating the eigenvector of each node based on the processed adjacency matrix and the node degree matrix.

[0008] Furthermore, the feature weighted vector corresponding to the knowledge graph of each known disease entity is used to archive the medical record of the patient to be tested, including: obtaining the medical record of the patient to be tested and determining the feature vector to be calculated of the medical record; calculating the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and archiving the medical record according to the similarity calculation result.

[0009] Furthermore, determining the feature vector to be calculated of the medical record file includes: performing entity recognition on the medical record file to identify entities of different attribute types; using a target-related model to convert each entity into a corresponding word vector; and weighting all of the word vectors to obtain the feature vector to be calculated of the medical record file.

[0010] Furthermore, there are multiple feature weighted vectors, and similarity calculation is performed on the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and the medical record file is archived according to the similarity calculation results, including: respectively performing similarity calculation on the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity to obtain multiple similarity calculation results; inputting all the similarity calculation results into a preset neural network, and outputting the weights between the feature vector to be calculated and each of the feature weighted vectors; the weights are used to characterize the probability that the name of the unknown disease entity corresponding to the medical record file is consistent with the name of the known disease entity; determining the name of the unknown disease entity corresponding to the medical record file based on all the weights; and archiving the medical record file according to the name of the unknown disease entity.

[0011] Furthermore, the method of determining the name of the unknown disease entity corresponding to the medical record file based on all the weights includes: based on all the weights, sorting the names of all known disease entities using a preset formula; outputting the name to be determined according to the sorting result, and determining the name of the unknown disease entity corresponding to the medical record file based on the name to be determined and the name of the disease entity corresponding to the historical medical record file; wherein the name to be determined is the name of a known disease entity that meets preset conditions.

[0012] Furthermore, before constructing multiple sub-graphs of each known disease entity according to a preset attribute type, the method also includes: acquiring medical knowledge data from a preset medical knowledge base, and constructing a standard disease knowledge graph based on the medical knowledge data; performing disease entity recognition on the standard disease knowledge graph to determine at least one known disease entity.

[0013] In a second aspect, the present invention provides a medical record archiving device, which includes: a first construction unit, used to construct multiple sub-graphs of each known disease entity according to a preset attribute type; a second construction unit, used to construct a sub-graph network corresponding to each sub-graph of each known disease entity using a target neural network; wherein the sub-graph networks corresponding to different sub-graphs are represented by different feature sub-vectors; a third construction unit, used to construct a feature weighted vector corresponding to the knowledge graph of each known disease entity based on the feature sub-vector corresponding to all the sub-graphs of the known disease entity; wherein the knowledge graph is composed of all the sub-graphs of the known disease entity; an archiving unit, used to archive the medical records of the patient to be tested based on the feature weighted vector corresponding to the knowledge graph of each known disease entity.

[0014] The present invention provides a method and device for archiving medical records, including: first, for each known disease entity, construct multiple sub-graphs of the known disease entity according to a preset attribute type; then, for each sub-graph of each known disease entity, use a target neural network to construct a sub-graph network corresponding to the sub-graph; wherein the sub-graph networks corresponding to different sub-graphs are represented by different feature sub-vectors; then, for each known disease entity, construct a feature weighted vector corresponding to the knowledge graph of the known disease entity based on the feature sub-vectors corresponding to all sub-graphs of the known disease entity; wherein the knowledge graph is composed of all sub-graphs of the known disease entity; finally, the medical records of the patient to be tested are archived based on the feature weighted vectors corresponding to the knowledge graph of each known disease entity. The present application can realize targeted multi-network collaborative processing of data by constructing sub-graphs in a targeted manner according to attribute types, constructing multiple sub-graph networks using a target neural network, and constructing feature weighted vectors of the knowledge graph, thereby avoiding the problem of difficulty in feature extraction caused by different data attributes, and then effectively archiving the medical records of the patient to be tested based on the feature weighted vectors, which can improve the accuracy of archiving the medical records of the patient to be tested.

[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a medical record filing method provided by an embodiment of the present invention;

[0019] Figure 2 The first sub-atlas with the disease name of appendicitis;

[0020] Figure 3 The second sub-atlas with the disease name of appendicitis;

[0021] Figure 4 The third sub-atlas has the disease name of appendicitis;

[0022] Figure 5 A schematic diagram of the structure of a medical record filing device provided in an embodiment of the present invention.

[0023] icon:

[0024] 11 - first building unit; 12 - second building unit; 13 - third building unit; 14 - filing unit. DETAILED DESCRIPTION

[0025] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] When a patient seeks medical treatment, the hospital takes a long time to determine the patient's disease based on the patient's symptoms and examination reports. At the same time, there is a possibility of misjudgment due to subjective factors when doctors are seeing patients, and different diseases have different interpretations in different hospitals, which can easily cause ambiguity and lead to patients not receiving timely treatment. Based on this, the purpose of the present invention is to provide a medical record filing method and device, which can eliminate ambiguity caused by different interpretations based on the unified name of known disease entities, and file the medical records of the patient to be tested based on the feature weighted vectors corresponding to the knowledge graph of each known disease entity. The method can reduce misjudgment caused by subjective factors, and can achieve effective filing of the medical records of the patient to be tested, thereby improving the accuracy of filing the medical records of the patient to be tested.

[0027] To facilitate understanding of this embodiment, a medical record archiving method disclosed in an embodiment of the present invention is first described in detail.

[0028] Embodiment 1:

[0029] According to an embodiment of the present invention, an embodiment of a medical record archiving method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Figure 1 A flowchart of a medical record archiving method provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes the following steps S101 to S104, wherein:

[0031] Step S101, for each known disease entity, construct multiple sub-graphs of the known disease entity according to preset attribute types. The attribute type includes at least one of the following: symptom type, sign type and examination result type. When the attribute type includes three types: symptom type, sign type and examination result type, in order to distinguish the sub-graphs, the sub-graph of the known disease entity constructed based on the symptom type is determined as the first sub-graph, the sub-graph of the known disease entity constructed based on the sign type is determined as the second sub-graph, and the sub-graph of the known disease entity constructed based on the examination result type is determined as the third sub-graph. The above-mentioned examination result type can be understood as the data type of the known disease entity in terms of the examination form data.

[0032] Step S102, for each sub-graph of each known disease entity, a sub-graph network corresponding to the sub-graph is constructed using a target neural network. In an embodiment of the present invention, sub-graph networks corresponding to different sub-graphs are represented by different feature sub-vectors. That is, each image representing different attributes of a known disease entity contained in each sub-graph is input into the target neural network to obtain a feature sub-vector representing each attribute feature of a known medical record entity. In order to distinguish the above-mentioned feature sub-vectors, the embodiment of the present invention determines the sub-graph network corresponding to the first sub-graph as the first sub-graph network, the sub-graph network corresponding to the second sub-graph as the second sub-graph network, and the sub-graph network corresponding to the third sub-graph as the third sub-graph network. Further, the embodiment of the present invention can determine the feature sub-vector extracted based on the first sub-graph network as the first feature sub-vector, the feature sub-vector extracted based on the second sub-graph network as the second feature sub-vector, and the feature sub-vector extracted based on the third sub-graph network as the third feature sub-vector. It should be noted that the dimensions of the three feature sub-vectors in the embodiment of the present invention can be consistent, but the embodiment of the present invention does not specifically limit the dimensions of each feature sub-vector.

[0033] The embodiment of the present invention constructs three subgraph networks for a known case entity, and extracts characteristic feature vectors of a known case entity in different attribute types from the three subgraph networks, so as to prepare for the aggregation of characteristic subvectors corresponding to the three subgraphs of the known disease entity in step S103.

[0034] Step S103, for each known disease entity, construct a feature weighted vector corresponding to the knowledge graph of the known disease entity based on the feature sub-vectors corresponding to all sub-graphs of the known disease entity.

[0035] In an embodiment of the present invention, the knowledge graph is composed of all sub-graphs of known disease entities. When the attribute type includes three types: symptom type, sign type, and test result type, the knowledge graph includes a first sub-graph of known disease entities constructed based on the symptom type, a second sub-graph of known disease entities constructed based on the sign type, and a third sub-graph of known disease entities constructed based on the test result type. For each known disease entity, the feature sub-vectors corresponding to the three sub-graphs of the known disease entity are aggregated (i.e., weighted summation) to construct a feature weighted vector corresponding to the knowledge graph of the known disease entity. In order to distinguish the feature weighted vectors corresponding to the knowledge graphs of each known disease entity, the feature weighted vectors corresponding to the knowledge graphs of each known disease entity are respectively recorded as n1, n2, ..., nM.

[0036] Step S104, archiving the medical records of the patient to be tested based on the feature weighted vectors corresponding to the knowledge graphs of each known disease entity.

[0037] A medical record archiving method provided by an embodiment of the present invention is an intelligent archiving method.

[0038] The above-mentioned target neural network can be understood as a nested graph neural network, which can also be called a two-level NestGN in the embodiment of the present invention. The structure and specific application of the network are the guarantee for the embodiment of the present invention to realize the construction of the subgraph network. In an optional embodiment, the target neural network includes a first graph neural network layer, a subgraph pooling layer, and a second graph neural network layer connected in sequence. Step S102 uses the target neural network to construct a subgraph network corresponding to the subgraph spectrum, including the following steps S201 to S203, wherein: Step S201, extract a rooted subgraph around each node of the subgraph spectrum; Step S202, apply each rooted subgraph to the first graph neural network layer and the subgraph pooling layer respectively to obtain the corresponding subgraph representation; Step S203, apply all subgraph representations to the second graph neural network layer to obtain a subgraph network corresponding to the subgraph spectrum.

[0039] Exemplarily, the specific method of using two-level NestGN to construct a subgraph network is as follows: first extract (i.e., copy) a rooted subgraph (height = 1) around each node of the original graph (i.e., the above-mentioned subgraph), then apply an internal NestGN (i.e., the above-mentioned first graph neural network layer) and a subgraph pooling layer to each rooted subgraph to learn a subgraph representation, and then aggregate the intermediate nodes (i.e., the nodes of the extracted rooted subgraph) into a subgraph representation, and the subgraph representation is used as the final representation of the root node in the original graph. It should be noted that a node represents an entity (or called a noun, keyword), and the embodiment of the present invention does not specifically limit the number of nodes in the subgraph. Generally, there will be as many nodes as the number of nouns that can be recognized. For example, the recognized nouns of symptom types such as fever, nausea, vomiting, and right lower abdominal pain are all nodes in the subgraph.

[0040] In an optional embodiment, after obtaining the subgraph network corresponding to the subgraph spectrum, the medical record archiving method further includes the following steps S204 to S205, wherein: Step S204, for each subgraph network, calculate the characteristic vector of each node in the subgraph network; Step S205, for each subgraph network, concatenate the characteristic vectors of all nodes to obtain the characteristic subvector corresponding to the subgraph network. In an optional embodiment, the calculation of the characteristic vector of each node in the subgraph network in step S204 includes the following steps S301 to S304, wherein: Step S301, obtain the adjacency matrix between nodes in the subgraph network; Step S302, determine the diagonal matrix based on the attribute type corresponding to the subgraph network; Step S303, perform data processing based on the adjacency matrix and the diagonal matrix to obtain the processed adjacency matrix and node degree matrix; Step S304, calculate the characteristic vector of each node based on the processed adjacency matrix and node degree matrix.

[0041] Exemplarily, the specific method for obtaining the characteristic subvector representation of each subgraph network is as follows: for each network in the above three subgraph networks, each node in each subgraph network forms a characteristic matrix X, and each node in each subgraph network will form an N*N dimensional adjacency matrix A (where N is the number of nodes in the subgraph, and the number of nodes in different subgraphs can be the same or different), the entity's adjacency matrix is ​​added to the diagonal matrix representing the entity identity (that is, the above attribute type) to obtain a new adjacency matrix A1 (that is, the adjacency matrix after the above processing) and its corresponding node degree matrix D, and the values ​​on the diagonal of D are divided by 1 to obtain the node's aggregation vector D1 (the D1 is obtained by converting the values ​​on the diagonal of D into the inverse), and then the characteristic vector representation of the node is calculated according to the formula D1*A1*D1*X, and finally the characteristic vectors of each node are concatenated to obtain the characteristic subvector.

[0042] In an optional embodiment, step S104, archiving the medical records of the patient to be tested based on the feature weighted vectors corresponding to the knowledge graph of each known disease entity, includes the following steps S401 to S402, wherein: step S401, obtaining the medical records of the patient to be tested, and determining the feature vector to be calculated of the medical records; step S402, performing similarity calculation between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and archiving the medical records according to the similarity calculation result. In an optional embodiment, step S401, determining the feature vector to be calculated of the medical records, includes: steps S501 to S503, wherein step S501, performing entity recognition on the medical records to identify entities of different attribute types; step S502, using the target-related model to convert each entity into a corresponding word vector; step S503, weighting all the word vectors to obtain the feature vector to be calculated of the medical records. In the embodiment of the present invention, the medical record file can be a paper version or a paper version. When the case file is a paper version, it needs to be converted into an electronic version by OCR (Optical Character Recognition). Entity recognition can identify nouns such as disease names, symptoms, signs, body temperature, platelets, indicators, body parts, etc. and the specific content under the noun.

[0043] Exemplarily, the embodiment of the present invention performs entity recognition on the treatment file of the patient to be tested (ie, the above-mentioned medical record file), and uses the word2vec model to obtain the feature vector V (ie, the feature vector to be calculated) of the patient's medical record file.

[0044] In an optional embodiment, there are multiple feature weighted vectors, and step S402 is to calculate the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and archive the medical record file according to the similarity calculation results, including the following steps S601 to S604, wherein: step S601, respectively calculate the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity to obtain multiple similarity calculation results; step S602, input all the similarity calculation results into a preset neural network, and output the weights between the feature vector to be calculated and each feature weighted vector; the weight is used to characterize the probability that the name of the unknown disease entity corresponding to the medical record file is consistent with the name of the known disease entity; step S603, determine the name of the unknown disease entity corresponding to the medical record file according to all the weights; step S604, archive the medical record file according to the name of the unknown disease entity. In an optional embodiment, step S603, determining the name of the unknown disease entity corresponding to the medical record file based on all weights, includes: step S701, based on all weights, sorting the names of all known disease entities using a preset formula; step S702, outputting the name to be determined based on the sorting result, and determining the name of the unknown disease entity corresponding to the medical record file based on the name to be determined and the name of the disease entity corresponding to the historical medical record file; wherein the name to be determined is the name of the known disease entity that meets the preset conditions. The above preset formula refers to the score between each known disease entity in the standard disease knowledge graph output by the neural network and the actual disease suffered by the patient, and then sorting and recommending the disease name according to the score. The formula is as follows:

[0045]

[0046] Among them, M is the first M related importance rel i The sum of ; M is the total number of all known disease entities, rel i Indicates the relative importance between the unknown name of the patient's disease and the name of the i-th known disease entity, for example: 2 means very important, 1 means important, 0 means unimportant, -1 means garbage name, Tim M Indicates the overall importance of this node (namely, the name of the known disease entity) in the entire network. The quantity threshold P is 3. If the number of disease names that meet the requirements P>3, the names of the top three known disease entities with the highest scores are output. The embodiment of the present invention uses the method of intelligently outputting disease names through a neural network as an auxiliary diagnosis method, and combined with the doctor's manual diagnosis, it can determine the name of the disease suffered by the patient.

[0047] In simple terms, we use the feature weighted vectors n1, n2, n3, ..., n of M known disease entities in the standard disease knowledge graph. M The feature vector V (i.e., the feature vector to be calculated) of the patient's medical record is calculated to obtain a similarity matrix (i.e., the similarity calculation result) between the feature vector V used to characterize the patient's disease feature information and the feature weighted vector of each known disease entity in the standard disease knowledge graph, which is input into the neural network. The network outputs the score (i.e., the weight) of each known disease entity in the standard disease knowledge graph that is similar to the patient's actual disease, and then sorts and recommends the disease name according to the score. Further, after sorting the names of all known disease entities, a weight threshold can be set to determine the names above the weight threshold as the names to be determined; if the number of names to be determined is less than the preset threshold, all the names to be determined within the preset threshold are output; if the number of names to be determined is not less than the preset threshold, the preset threshold number of names to be determined are taken for output. For example, assuming that the preset threshold is 3, if the number of names to be determined that are higher than the weight threshold is 2 and less than 3, two names to be determined are output; if the number of names to be determined that are higher than the weight threshold is 5 and greater than 3, the first 3 names to be determined with the largest weight (highest score) are output.

[0048] In an optional embodiment, two different medical records (i.e., the current case file and the historical case file) may be input for the medical treatment of the same patient at different times. If the results of the two outputs are the same (for example, the number of disease names output is one and the names are the same), then it can be determined that the two medical records belong to the same disease; if the result of the second output includes the result of the first output, then it can be determined and confirmed in combination with the doctor's diagnosis that the two medical records belong to the same disease; if the result of the second output does not include the result of the first output, then it is determined that the two medical records belong to different types of diseases, and the medical records are archived.

[0049] In an optional embodiment, in step S101, before constructing multiple sub-graphs of known disease entities according to preset attribute types for each known disease entity, the following steps S105 to S106 are also included: Step S105, obtaining medical knowledge data from a preset medical knowledge base, and constructing a standard disease knowledge graph based on the medical knowledge data; Step S106, performing disease entity recognition on the standard disease knowledge graph to determine at least one known disease entity. In an embodiment of the present invention, the above-mentioned medical knowledge data includes but is not limited to the above-mentioned three types (symptom type, sign type and test result type) and other content. The above-mentioned standard disease knowledge graph is generally represented by a graph structure. In order to more accurately identify all diseases, the embodiment of the present invention may consider the construction of a standard disease knowledge graph and determine that the constructed standard disease knowledge graph is large enough and has enough information, including knowledge points of all diseases. In an embodiment of the present invention, disease entity recognition may refer to the recognition of noun categories, and appendicitis is used as an example for the following explanation. Figure 2 The nouns for the symptoms of patients suffering from appendicitis include: temperature-temperature, temperature level-fever, skin-skin hyperesthesia, abdomen-right lower abdominal pain, sleep-frequent and rapid breathing, stomach-nausea, vomiting. These nouns are used to construct the first sub-graph; Figure 3 Nouns for the physical signs of patients suffering from appendicitis, including: temperature - temperature, muscle - muscle tension, abdomen - rebound pain, intestine - positive signs for diagnosing appendicitis, abdomen - rebound pain at McBurney's point in the right lower abdomen. These nouns are used to construct the second sub-atlas; Figure 4 The nouns for the patient's disease name is appendicitis in the examination form data, including: total bilirubin content 5.2μmol / L, red blood cell count 4.5×10 12 / L, white blood cell count 4.0×10 12 / L and blood glucose content 7.8mol / L, these terms are used to construct the third sub-map.

[0050] In summary, from the whole process, the embodiment of the present invention first performs entity recognition on the medical knowledge base, identifies nouns such as disease names, symptoms, and signs, and then uses the three different types of data, namely symptoms, signs, and examination form data, to construct three subgraph networks of patient symptoms, patient signs, and examination form data using the method of the nested graph neural network NestGN, perform subgraph node representation learning, and then input it into the F-Extractor feature extraction network for feature extraction. The method of using multiple networks to collaboratively process data in the embodiment of the present invention can largely solve the problem of difficult feature extraction caused by different data attributes in real life, and can also analyze the patient's medical record data in a targeted manner to maximize the use of data. From the perspective of network architecture, the embodiment of the present invention uses two networks, one is the nested graph neural network NestGN, and the other is a neural network for training the weights between nodes and networks, wherein the nested graph neural network NestGN is used to obtain the feature representation of the three subgraph networks of known disease entities, in short, to extract the features of each subgraph network; and the neural network is trained to obtain the weights of the nodes and the network, so that the subsequent neural network can realize the output, thereby realizing the disease name recommendation and sorting functions.

[0051] Embodiment 2:

[0052] The embodiment of the present invention provides a medical record file archiving device, which is mainly used to execute the medical record file archiving method provided in the above content of Example 1. The medical record file archiving device provided by the embodiment of the present invention is specifically introduced below.

[0053] Figure 5 The structure diagram of a medical record filing device provided by an embodiment of the present invention is shown in FIG. Figure 5 As shown, the medical record filing device mainly includes: a first construction unit 11, a second construction unit 12, a third construction unit 13 and a filing unit 14, wherein:

[0054] The first construction unit 11 is used to construct multiple sub-graphs of the known disease entities according to preset attribute types for each known disease entity;

[0055] The second construction unit 12 is used to construct a subgraph network corresponding to each subgraph of each known disease entity using a target neural network; wherein the subgraph networks corresponding to different subgraphs are represented by different feature subvectors;

[0056] The third construction unit 13 is used to construct, for each known disease entity, a feature weighted vector corresponding to the knowledge graph of the known disease entity based on the feature subvectors corresponding to all subgraphs of the known disease entity; wherein the knowledge graph is composed of all subgraphs of the known disease entity;

[0057] The archiving unit 14 is used to archive the medical records of the patient to be tested based on the feature weighted vectors corresponding to the knowledge graphs of each known disease entity.

[0058] The embodiment of the present invention can construct a sub-graph in a targeted manner according to the attribute type through the first construction unit 11, can construct multiple sub-graph networks using the target neural network through the second construction unit 12, and can realize targeted multi-network collaborative processing of data by constructing the feature weighted vector of the knowledge graph through the third construction unit 13, thereby avoiding the problem of difficulty in feature extraction caused by different data attributes, and then can use the archiving unit 14 to effectively archive the medical records of the patient to be tested based on the feature weighted vector, thereby improving the accuracy of archiving the medical records of the patient to be tested.

[0059] Optionally, the target neural network includes a first graph neural network layer, a subgraph pooling layer, and a second graph neural network layer connected in sequence, and the second construction unit 12 includes: an extraction module, a first application module, and a second application module, wherein:

[0060] An extraction module, used to extract a rooted subgraph around each node of the subgraph;

[0061] A first application module is used to apply each rooted subgraph to the first graph neural network layer and the subgraph pooling layer respectively to obtain a corresponding subgraph representation;

[0062] The second application module is used to apply all sub-graph representations to the second graph neural network layer to obtain a sub-graph network corresponding to the sub-graph spectrum.

[0063] Optionally, the medical record filing device further includes: a calculation unit and a splicing unit, wherein:

[0064] A calculation unit, used for calculating a feature vector of each node in each subgraph network;

[0065] The concatenation unit is used to concatenate the feature vectors of all nodes in each subgraph network to obtain a feature subvector corresponding to the subgraph network.

[0066] Optionally, the calculation unit includes: an acquisition module, a determination module, a data processing module and a calculation module, wherein:

[0067] The acquisition module is used to obtain the adjacency matrix between nodes in the subgraph network;

[0068] A determination module, used for determining a diagonal matrix based on an attribute type corresponding to the subgraph network;

[0069] A data processing module, used for performing data processing based on the adjacency matrix and the diagonal matrix to obtain a processed adjacency matrix and a node degree matrix;

[0070] The calculation module is used to calculate the feature vector of each node based on the processed adjacency matrix and node degree matrix.

[0071] Optionally, the archiving unit 14 includes: an acquisition and determination module and a similarity calculation module, wherein:

[0072] An acquisition and determination module is used to acquire the medical records of the patient to be tested and determine the feature vector to be calculated of the medical records;

[0073] The similarity calculation module is used to calculate the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and archive the medical record files according to the similarity calculation results.

[0074] Optionally, the acquisition and determination module includes: an entity recognition submodule, a conversion submodule and a weighted processing submodule, wherein:

[0075] The entity recognition submodule is used to perform entity recognition on medical records and identify entities of different attribute types;

[0076] The conversion submodule is used to convert each entity into a corresponding word vector using the target-related model;

[0077] The weighted processing submodule is used to perform weighted processing on all word vectors to obtain the feature vectors to be calculated for the medical record files.

[0078] Optionally, there are multiple feature weighted vectors, and the similarity calculation module includes: a similarity calculation submodule, an input submodule, a determination submodule and a filing submodule, wherein:

[0079] A calculation submodule is used to calculate the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and obtain multiple similarity calculation results;

[0080] The input submodule is used to input all similarity calculation results into a preset neural network and output the weights between the feature vector to be calculated and each feature weighted vector; the weights are used to represent the probability that the name of the unknown disease entity corresponding to the medical record file is consistent with the name of the known disease entity;

[0081] A determination submodule, used for determining the name of the unknown disease entity corresponding to the medical record file according to all weights;

[0082] The archiving submodule is used to archive medical records according to the names of unknown disease entities.

[0083] Optionally, the determination submodule is also used to: sort the names of all known disease entities based on all weights using a preset formula; output the name to be determined according to the sorting result, and determine the name of the unknown disease entity corresponding to the medical record file based on the name to be determined and the name of the disease entity corresponding to the historical medical record file; wherein the name to be determined is the name of a known disease entity that meets preset conditions.

[0084] Optionally, the medical record filing device further comprises: an acquisition construction unit and a disease entity identification unit, wherein:

[0085] An acquisition construction unit is used to acquire medical knowledge data from a preset medical knowledge base and construct a standard disease knowledge graph based on the medical knowledge data;

[0086] The disease entity recognition unit is used to perform disease entity recognition on the standard disease knowledge graph and determine at least one known disease entity.

[0087] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0088] In an optional embodiment, this embodiment further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the steps of the above method embodiment are implemented.

[0089] In an optional embodiment, this embodiment further provides a computer-readable medium having a non-volatile program code executable by a processor, wherein the program code enables the processor to execute the method of the above method embodiment.

[0090] In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0091] In the description of this embodiment, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they should not be understood as limitations on this embodiment.

[0092] Furthermore, the terms “first”, “second”, and “third” are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0093] In the several embodiments provided in this embodiment, it should be understood that the disclosed methods and devices can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0094] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0095] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0096] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that can be executed by a processor. Based on this understanding, the technical solution of this embodiment, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for 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, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0097] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the aforementioned embodiments, those of ordinary skill in the art should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the aforementioned embodiments within the technical scope disclosed by the present invention, or can easily conceive of changes, or make equivalent replacements for some of the technical features therein. Such modifications, changes 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, and should all be covered within the protection scope of the present invention.

Claims

1. A medical record filing method, characterized in that: include: For each known disease entity, construct multiple sub-graphs of the known disease entity according to preset attribute types; For each of the sub-graphs of each known disease entity, a sub-graph network corresponding to the sub-graph is constructed using a target neural network; wherein the sub-graph networks corresponding to different sub-graphs are represented by different feature sub-vectors; For each known disease entity, a feature weighted vector corresponding to the knowledge graph of the known disease entity is constructed based on the feature sub-vector corresponding to all sub-graphs of the known disease entity; wherein the knowledge graph is composed of all sub-graphs of the known disease entity; and the number of the feature weighted vectors is multiple; Archiving the medical records of the patient to be tested based on the feature weighted vector corresponding to the knowledge graph of each known disease entity; The feature weighted vector corresponding to the knowledge graph of each known disease entity is archived for the medical records of the patient to be tested, including: Obtaining the medical records of the patient to be tested, and determining the feature vector to be calculated of the medical records; Performing similarity calculation on the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and filing the medical record file according to the similarity calculation result; The step of calculating the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and archiving the medical record file according to the similarity calculation result includes: Performing similarity calculations on the feature vector to be calculated and the feature weighted vectors corresponding to the knowledge graphs of each known disease entity to obtain multiple similarity calculation results; All the similarity calculation results are input into a preset neural network, and the weights between the feature vector to be calculated and each feature weighted vector are output; the weights are used to represent the probability that the name of the unknown disease entity corresponding to the medical record is consistent with the name of the known disease entity; Determine the name of the unknown disease entity corresponding to the medical record file according to all of the weights; The medical record files are filed according to the name of the unknown disease entity.

2. The medical record filing method according to claim 1, characterized in that: The target neural network includes a first graph neural network layer, a subgraph pooling layer, and a second graph neural network layer connected in sequence, and using the target neural network to construct a subgraph network corresponding to the subgraph spectrum includes: A rooted subgraph is extracted around each node of the subgraph; Apply each of the rooted subgraphs to the first graph neural network layer and the subgraph pooling layer respectively to obtain a corresponding subgraph representation; Apply all of the sub-graph representations to the second graph neural network layer to obtain a sub-graph network corresponding to the sub-graph spectrum.

3. The medical record filing method according to claim 1, characterized in that: Also includes: For each of the subgraph networks, calculating a feature vector of each node in the subgraph network; For each of the subgraph networks, the feature vectors of all nodes are concatenated to obtain a feature subvector corresponding to the subgraph network.

4. The medical record filing method according to claim 3, characterized in that: The calculating the feature vector of each node in the subgraph network comprises: Obtaining an adjacency matrix between nodes in the subgraph network; Determining a diagonal matrix based on the attribute type corresponding to the subgraph network; Performing data processing based on the adjacency matrix and the diagonal matrix to obtain a processed adjacency matrix and a node degree matrix; Based on the processed adjacency matrix and the node degree matrix, a feature vector of each node is calculated.

5. The medical record filing method according to claim 1, characterized in that: The step of determining the feature vector to be calculated of the medical record file includes: Performing entity recognition on the medical record files to identify entities of different attribute types; Use the target-related model to convert each entity into a corresponding word vector; All the word vectors are weighted to obtain the feature vector to be calculated for the medical record file.

6. The medical record filing method according to claim 1, characterized in that: Determining the name of the unknown disease entity corresponding to the medical record file according to all the weights includes: Based on all of the weights, the names of all known disease entities are sorted using a preset formula; The name to be determined is output according to the sorting result, and based on the name to be determined and the name of the disease entity corresponding to the historical medical record file, the name of the unknown disease entity corresponding to the medical record file is determined; wherein the name to be determined is the name of a known disease entity that meets preset conditions.

7. The medical record filing method according to claim 1, characterized in that: Before constructing multiple sub-graphs of each known disease entity according to a preset attribute type, the method further includes: Acquire medical knowledge data from a preset medical knowledge base, and construct a standard disease knowledge graph based on the medical knowledge data; Disease entity recognition is performed on the standard disease knowledge graph to determine at least one known disease entity.

8. A medical record filing device, characterized in that: include: A first construction unit is used to construct, for each known disease entity, a plurality of sub-graphs of the known disease entity according to a preset attribute type; A second construction unit is used for constructing a subgraph network corresponding to each subgraph of each known disease entity using a target neural network; wherein the subgraph networks corresponding to different subgraphs are represented by different feature subvectors; A third construction unit is used to construct, for each known disease entity, a feature weighted vector corresponding to the knowledge graph of the known disease entity based on the feature subvector corresponding to all the subgraphs of the known disease entity; wherein the knowledge graph is composed of all the subgraphs of the known disease entity; and the number of the feature weighted vectors is multiple; A filing unit, used for filing the medical records of the patient to be tested based on the feature weighted vector corresponding to the knowledge graph of each known disease entity; Wherein, the archiving unit includes: an acquisition determination module and a similarity calculation module; An acquisition and determination module is used to acquire the medical records of the patient to be tested and determine the feature vector to be calculated of the medical records; A similarity calculation module, used for calculating the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, and archiving the medical record file according to the similarity calculation result; Wherein, the similarity calculation module includes: a similarity calculation submodule, an input submodule, a determination submodule and an archiving submodule; A calculation submodule, used for respectively calculating the similarity between the feature vector to be calculated and the feature weighted vector corresponding to the knowledge graph of each known disease entity, to obtain multiple similarity calculation results; An input submodule, used to input all the similarity calculation results into a preset neural network, and output the weights between the feature vector to be calculated and each feature weighted vector; the weights are used to represent the probability that the name of the unknown disease entity corresponding to the medical record file is consistent with the name of the known disease entity; A determination submodule, used to determine the name of the unknown disease entity corresponding to the medical record file according to all the weights; The filing submodule is used to file the medical record files according to the name of the unknown disease entity.

Images