CN110299209B - Similar medical record search method, device, device and readable storage medium - Google Patents

Abstract

The present invention provides a similar medical record search method, device, equipment and readable storage medium. By acquiring query medical record data and a plurality of historical medical record data; Graph structure data, wherein the query graph structure data and the historical graph structure data both include the first type of subgraph and the second type of subgraph, and the intermediate nodes and leaf nodes of the second type of subgraph are used to identify the features of the first type of subgraph Obtained; according to the similarity of the root node, the similarity of the first type of sub-graph and the similarity of the second type of sub-graph, the similarity degree of each historical graph structure data and the query graph structure data is obtained; according to the preset selection rules and similarity, determine The search results of similar medical records in the medical record data are inquired, and the inherent and identifiable subgraphs in the medical record data are extracted, and the correlation of the contents of the corresponding subgraphs is measured in the comparison, which improves the accuracy of similar medical records search.

Description

Similar medical record search method, device, device and readable storage medium

technical field

The present invention relates to the technical field of information processing, and in particular, to a method, apparatus, device and readable storage medium for searching similar medical records.

Background technique

In the medical field, the retrieval of similar medical records is of great significance in scientific research and clinical practice. For example, when a patient visits a doctor, the doctor can quickly find the medical records similar to the patient, and can make effective judgments in time through the diagnosis and treatment path and effect of the similar medical records; , it is possible to obtain some reference diagnostic opinions and treatment methods by referring to historical medical records with a certain degree of similarity; or, in clinical research, in some cases, it is necessary to use a certain medical record as a starting point to find more similar medical records. Conduct research discussions.

The current medical record matching retrieval method is usually the retrieval of full text information of medical records. For example, by retrieving fever and dyspnea as keywords, all medical records with the two keywords of fever and dyspnea in the pre-stored medical records can be retrieved.

However, because the corresponding diseases of the same symptoms are quite different, the accuracy of the existing similar medical record retrieval methods is not high.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a similar medical record search method, device, device and readable storage medium, which improve the accuracy and reliability of similar medical record search.

According to a first aspect of the present invention, a method for searching similar medical records is provided, comprising:

Obtain and query medical record data and multiple historical medical record data;

Obtain the query graph structure data corresponding to the query medical record data, and the history graph structure data corresponding to each of the historical medical record data, wherein the query graph structure data and the historical graph structure data both include the first type of subgraph and The second type of subgraph, the middle node of the first type of subgraph is a medical record field category, and the middle node and the leaf node of the second type of subgraph are obtained by performing feature recognition on the first type of subgraph;

Obtain each of the historical graph structural data and the query graph structure according to the similarity of the root node, the similarity of the first type of subgraph and the similarity of the second type of subgraph in each of the historical graph structural data and the query graph structural data the similarity of the data;

According to the preset selection rule and the similarity degree, a similar medical record search result of the query medical record data is determined from the plurality of historical medical record data, wherein the historical graph structure data corresponding to the similar medical record search result has The degree of similarity that satisfies the preset selection rule.

According to a second aspect of the present invention, there is provided a similar medical record search device, comprising:

The medical record acquisition module is used to acquire and query medical record data and multiple historical medical record data;

The graph structure module is used to obtain the query graph structure data corresponding to the query medical record data, and the historical graph structure data corresponding to each of the historical medical record data, wherein the query graph structure data and the historical graph structure data are both Including a first type of subgraph and a second type of subgraph, the middle node of the first type of subgraph is the medical record field category, and the middle node and leaf node of the second type of subgraph are related to the first type of subgraph. obtained by feature identification;

The processing module is configured to obtain each of the historical graph structural data and the similarity of each of the historical graph structural data according to the similarity of the root node, the similarity of the first type of subgraph and the similarity of the second type of subgraph in each of the historical graph structural data and the query graph structural data. the similarity of the query graph structure data;

A selection module, configured to determine a similar medical record search result of the query medical record data in the plurality of historical medical record data according to a preset selection rule and the similarity degree, wherein the history corresponding to the similar medical record search result The graph structure data has the similarity degree satisfying the preset selection rule.

According to a third aspect of the present invention, a device is provided, comprising: a memory, a processor, and a computer program, where the computer program is stored in the memory, and the processor executes the computer program to execute the first aspect of the present invention and a computer program. In the first aspect, various possible designs of the similar medical record searching method are provided.

According to a fourth aspect of the present invention, there is provided a readable storage medium, where a computer program is stored in the readable storage medium, and when the computer program is executed by a processor, is used to implement the first aspect and each of the first aspect of the present invention A possible design of the similar medical record finding method.

The present invention provides a similar medical record search method, device, equipment and readable storage medium, by acquiring query medical record data and a plurality of historical medical record data; Historical graph structure data corresponding to medical record data, wherein both the query graph structural data and the historical graph structural data include a first type of subgraph and a second type of subgraph, and the intermediate node of the first type of subgraph is a medical record Field category, the intermediate nodes and leaf nodes of the second type of subgraph are obtained by feature identification of the first type of subgraph; according to the similarity between the root node in each of the historical graph structure data and the query graph structure data , the similarity of the first type of sub-graph and the similarity of the second type of sub-graph, to obtain the similarity degree of each of the historical graph structure data and the query map structure data; according to the preset selection rule and the similarity degree, in the The similar medical record search results of the query medical record data are determined in a plurality of historical medical record data, so as to extract the inherent subgraphs and identifiable subgraphs in the query medical record data, and compare the corresponding subgraphs in the query medical record data and the historical medical record data medical records. The correlation of the data in the graph is measured, which improves the accuracy of finding similar medical records.

Description of drawings

1 is a schematic flowchart of a method for searching similar medical records provided by an embodiment of the present invention;

2 is a schematic diagram of a query graph structure data provided by an embodiment of the present invention;

3 is a schematic diagram of a first type of subgraph and a second type of subgraph provided by an embodiment of the present invention;

FIG. 4 is a schematic flowchart of an optional embodiment of step S103 in FIG. 1 according to an embodiment of the present invention;

5 is a schematic structural diagram of a similar medical record search device provided by an embodiment of the present invention;

FIG. 6 is a schematic diagram of a hardware structure of a device provided by an embodiment of the present invention.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments It is only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The terms "first", "second" and the like in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It is to be understood that the data so used may be interchanged under appropriate circumstances such that the embodiments of the invention described herein can be practiced in sequences other than those illustrated or described herein.

It should be understood that, in various embodiments of the present invention, the size of the sequence numbers of each process does not mean the sequence of execution, and the execution sequence of each process should be determined by its functions and internal logic, and should not be used in the embodiments of the present invention. Implementation constitutes any limitation.

It should be understood that in the present invention, "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to Those steps or elements that are expressly listed may instead include other steps or elements that are not expressly listed or are inherent to the process, method, product or apparatus.

It should be understood that, in the present invention, "plurality" refers to two or more.

It should be understood that in the present invention, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A, according to A B can be determined. Determining B based on A does not mean determining B based only on A, but also determining B based on A and/or other information. The matching between A and B means that the similarity between A and B is greater than or equal to a preset threshold.

"If" as used herein may be interpreted as "at" or "when" or "in response to determining" or "in response to detecting," depending on the context.

The technical solutions of the present invention will be described in detail below with specific examples. The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

A piece of medical record data usually includes main diagnostic information, fields inherent in multiple medical records, and field contents. The content of these fields is the content filled in by the patient or doctor on the field names in the standard medical records, for example, the field names include: gender, age, department, consultation time, type of diagnosis and treatment, doctor's order, chief complaint, history of present illness, physical examination, and auxiliary examination. In the current medical record matching retrieval method, the full text information of the medical record or the content of these fields is usually retrieved. For example, by searching with the keywords "fever" and "breathing", all medical records with the two keywords of fever and shortness of breath in the pre-stored medical records can be retrieved. It is also possible to specify that the medical records containing the keyword "fever" in the main complaint will be the search results. However, the same chief complaint may correspond to completely different symptoms and diseases, such as fever caused by a cold and fever caused by allergies, and the treatment and diagnosis methods are so different that they do not have the value of comparing similar medical records. Therefore, the accuracy of the existing similar medical record retrieval methods is not high.

In order to solve the problem of low accuracy of the existing similar medical record retrieval methods, the embodiment of the present invention provides a similar medical record retrieval method. Describe the historical graph structure data, the second type of subgraph is obtained by feature identification of the first type of subgraph, and finally according to the similarity of the root node, the similarity of the first type of subgraph and the similarity of the second type of subgraph, determine Querying the similarity between medical record data and historical medical record data improves the accuracy of finding similar medical records.

Referring to FIG. 1 , it is a schematic flowchart of a method for searching similar medical records provided by an embodiment of the present invention. The execution body of the method shown in FIG. 1 may be software and/or hardware devices, such as a server. The method shown in FIG. 1 includes steps S101 to S104, and the details are as follows:

S101, obtaining and querying medical record data and multiple historical medical record data.

For example, the server acquires multiple historical medical record data when receiving the queried medical record data that needs to be searched for similar medical records. The historical medical record data may be pre-stored in the medical record database, or may be acquired from a distributed storage unit. In some embodiments, the medical record data related to the queried medical record data may be used as the historical medical record data. For example, the server obtains the queried medical record data; and then determines the full-text search keyword according to the queried medical record data. The full-text search keywords are, for example, words such as "cold", "cough", and "XX drug allergy" extracted from the query medical record data. Finally, according to the full-text search key, a full-text search of medical records is performed in the medical record database, and a plurality of historical medical record data including the full-text search key are obtained.

S102, acquiring query graph structure data corresponding to the query medical record data, and historical graph structure data corresponding to each of the historical medical record data, wherein both the query graph structure data and the historical graph structure data include the first type of subclasses graph and the second type of subgraph, the middle node of the first type of subgraph is the medical record field category, and the middle node and leaf node of the second type of subgraph are obtained by feature recognition of the first type of subgraph .

For example, using feature engineering to structure medical record data, a graph structure data body is generated for the query medical record data and the historical medical record data respectively, and the query graph structure data and the historical graph structure data are obtained. The historical graph structure data may be structured in the same way as the query graph structure data is obtained, or may be obtained from the outside in advance and stored corresponding to the historical medical record data, which is not limited herein. This embodiment is described by taking the acquisition process of query graph structure data as an example.

Referring to FIG. 2 , it is a schematic diagram of querying graph structure data according to an embodiment of the present invention. In this embodiment, the query graph structure data corresponding to the query medical record data is taken as an example. In FIG. 2 , the arrows point from the child nodes to the respective parent nodes, and the root node is the main diagnosis.

Usually in the query of medical record data, it includes three categories of data: main diagnosis, basic information and basic medical conditions. The primary diagnosis is text-type data. The basic information may include, for example, items to be filled in by the doctor: age and gender. The basic condition may include, for example, items that need to be filled in by the doctor: doctor's order, physical examination, type of diagnosis and treatment, chief complaint, auxiliary examination, department, history of current illness, and visit time.

For example, the main diagnosis of the queried medical record can be used as the description information of the root node. The intermediate nodes under the root node can be divided into two categories. The first type of intermediate nodes can be field categories in the query medical record data, such as basic information and basic medical conditions, see FIG. 2 . The second type of intermediate node may be a preset feature, for example, a feature that is to be identified from the item content filled in by the doctor. As shown in FIG. 2 , the second type of intermediate nodes may include, for example, allergies, operations, medicines, population attributes, symptoms, diseases, signs, tests, and examinations.

For the first type of intermediate node, the medical record field category included in the query medical record may be used as the first type of intermediate node, and the fields included in the field category may be regarded as the first type of leaf nodes of the field category, respectively. The description information corresponding to the description field is used as the description information of the first type of leaf node. For example, if the field category is "Basic Information", then "Basic Information" is used as the first-type intermediate node, and "Age" and "Gender" are respectively used as two first-type leaf nodes under the first-type intermediate node. The leaf nodes of the first type of intermediate node "Basic Conditions" include "Doctor's Order", "Physical Examination", "Diagnosis and Treatment Type", "Main Complaint", "Auxiliary Examination", "Department", "Present Illness", and "Duration of Consultation" "8 first-type leaf nodes, see the graph structure data shown in Figure 2.

For the second type of intermediate node, the preset feature can be used as the second type of intermediate node first. For example, in the example shown in FIG. 2 , the preset feature "symptom" is used as a second-type intermediate node. The second type of intermediate node may be a preset feature that needs to be identified. The server uses the feature to perform natural language understanding NLU recognition on each of the fields and the description information of each field in the query medical record data, and obtains the description information of the feature, or obtains the description information of the feature and the description The addition factor attribute information or the multiplication factor attribute information of the information. For example, according to the description information of fields and fields such as the history of present illness and main complaint, NLU processing is performed on the feature "symptom", and the duration of cough (1 month), fever and fever time (current) are obtained. 24 hours). Then, the description information "cough" and "fever" can be used as two second-type leaf nodes under the second-type intermediate node "symptom". In addition, "yes" may be used as multiplier attribute information of the description information "cough", and "slight degree" and "continue for 1 month" may be used as the addition factor attribute information of the description information "cough". Similarly, "continue for 24 hours" can be used as the plus factor attribute information of the description information "fever". Another example is to perform NLU processing on the feature "population attribute" according to the fields and description information of the fields such as the history of present illness and the chief complaint to obtain the elderly. Then, the description information "elderly people" can be used as the second type of leaf node under the second type of intermediate node "crowd attribute". After determining the description information of the feature, the server may use the description information of the feature as the second type of leaf node of the feature, wherein the second type of leaf node having the attribute information of the addition factor or the attribute information of the multiplication factor has the attribute information of the second type of leaf node. The data type is a complex data type.

Finally, according to the root node, the first type intermediate node, the first type leaf node, the second type intermediate node and the second type leaf node obtained above, the query medical record data can be obtained Corresponding query graph structure data, wherein the root node, the first type intermediate node and the first type leaf node form a first type subgraph; the root node, the second type intermediate node and all The second type of leaf nodes form the second type of subgraph. Referring to FIG. 3 , it is a schematic diagram of a first type of subgraph and a second type of subgraph provided by an embodiment of the present invention. Two first-type subgraphs and two second-type subgraphs are illustrated in FIG. 3 . The root nodes of the first type subgraph and the second type subgraph are the main diagnosis, but the first type intermediate node of the first type subgraph is a field category, such as basic information or basic condition, while the first type of the second type subgraph is a field category. Class II intermediate nodes are features, such as symptoms or signs. The first type of subgraph is divided into leaf nodes, and each first type of subgraph includes a single first type of leaf node. The second type of subgraph is divided by intermediate nodes. A second type of subgraph includes a single second type of intermediate node, but may include multiple second type of leaf nodes.

After obtaining the above query graph structure data, in order to facilitate the subsequent similarity measurement using subgraphs as comparison units, the query graph structure data may be divided into multiple first type subgraphs and multiple second type subgraphs.

S103, according to the similarity of the root node, the similarity of the first type of sub-graph and the similarity of the second type of sub-graph in each of the historical graph structural data and the query graph structural data, obtain each of the historical graph structural data and the query Similarity of graph-structured data.

In the comparison of the similarity of medical records, if the main diagnosis of the added node is completely different, it is usually not a similar medical record. Therefore, the similarity of the root node is used as one of the basis for comparing the similarity between the historical graph structure data and the query graph structure data, which can improve similar medical records. Find the accuracy.

Specifically, it can be to obtain the root node, the first type subgraph and the second type subgraph of the historical graph structure data, and to query the root node, the first type subgraph and the second type subgraph of the graph structure data, and then use two The root node similarity is obtained by comparing the root nodes of the two, the first type subgraph similarity is obtained by comparing the first type subgraphs of the two, and the second type subgraph similarity is obtained by comparing the second type subgraphs of the two.

In order to explain the above step S103 more clearly (according to the similarity of the root node, the similarity of the first type of sub-graph and the similarity of the second type of sub-graph in each of the historical graph structural data and the query graph structural data, obtain each of the historical graphs The degree of similarity between the graph structure data and the query graph structure data) is described below with reference to the accompanying drawings and specific embodiments.

Referring to FIG. 4 , it is a schematic flowchart of an optional embodiment of step S103 in FIG. 1 according to an embodiment of the present invention. The method shown in FIG. 4 includes steps S201 to S204, which are as follows:

S201: Determine a first similarity measure between the historical graph structural data and the query graph structural data according to the similarity of the root node in the historical graph structural data and the query graph structural data.

For example, the root node similarity may be directly used as the first similarity measure between the historical graph structure data and the query graph structure data.

Since the main diagnosis of the root node is usually text-type data, before determining the first similarity metric value, or before step 103 shown in FIG. 1 , the model and the root node can also be determined according to the preset text-type similarity. The description information is determined, and the similarity between each historical graph structure data and the root node in the query graph structure data is determined. The text-type similarity determination model may be, for example, the similarity determination model corresponding to the text-type data shown in Table 1 below.

S202, according to the saliency category of the leaf node, in the first type of subgraph, determine the first type of saliency subgraph and the first type of non-salient subgraph.

The saliency category of the leaf node can be set according to the design application requirements when the actual similarity is matched. Assume that the design application requirements are: "age", "sex", "diagnosis and treatment type", and "diagnosis time" are the screening conditions for similar medical record results. For example, the first type of submap including "age", "sex", "time of consultation", and "type of diagnosis and treatment" can be used as the first type of significant submap. Among them, "gender" and "diagnosis and treatment type" are similar medical records matching filter conditions of Boolean data type. For example, if the similar medical record filter conditions are "gender: male", then the similar medical record search matches the result output of "gender: male". Among them, "age" and "visit time" are the similar medical record screening and matching conditions of numerical data type, and the numerical decay function is used for similar medical record matching screening, for example: the similar medical record screening condition is "age: 20 years old", similar medical records are for " Age" is used for attenuation matching screening. Assuming that other matching items of similar medical record retrieval results are basically the same, the closer "age" is to 20 years old, the higher the degree of similarity. In this embodiment, the first type of subgraphs including "doctor's order", "physical examination", "main complaint", "auxiliary examination", "department", and "history of present illness" can be used as the first type of non-significant subgraphs.

S203: Determine a second similarity measure between the historical graph structural data and the query graph structural data according to the similarity between the historical graph structural data and the first type of significant subgraphs in the query graph structural data.

Both the historical graph structure data and the query graph structure data may include multiple first-type saliency subgraphs. The graphs are compared to obtain the similarity of the first type of significant subgraphs. Then, the server may take the product of the similarity between the historical graph structure data and the first type of significant subgraphs of the query graph structure data as the second similarity between the historical graph structure data and the query graph structure data metric. In the form of a product, the degree of influence of the similarity of each first type of significant subgraphs on the second similarity metric value can be improved. For example, as long as the similarity of any first-type saliency subgraph is 0, that is, the first-type saliency subgraphs of the two graph structure data are completely irrelevant, the direct value will indeed set the second similarity measure as 0.

S204: Determine the historical graph structural data and the query graph structural data according to the similarity of the first type of non-significant subgraph and the second type of subgraph similarity in the historical graph structural data and the query graph structural data The third similarity measure of .

When the similarity of the first type of non-salient sub-graph and the similarity of the second type of sub-graph in the historical graph structure data and the query graph structure data are obtained, a third similarity measure value can be obtained accordingly.

Since the first type of non-salient subgraph and the second type of subgraph usually correspond to the specific conditions of the medical record and do not play a decisive role in the similarity, in the process of determining the third similarity measure, for example, the historical graph structure data can be The sum of similarity with each of the first type of non-significant subgraphs and each of the second type of subgraphs in the query graph structure data is taken as the difference between the historical graph structure data and the query graph structure data. The third similarity measure. It can be understood that the third similarity metric value is embodied as the sum of the similarity of each of the first type of non-salient sub-graphs and the similarity of each of the second type of sub-graphs.

In this embodiment, the above steps S201, S203, and S204 are not limited by the sequence described in FIG. 4. Steps S201, S203, and S204 may be performed in other sequences or simultaneously, which are not limited here.

S205, according to the first similarity metric value, the second similarity metric value and the third similarity metric value of each of the historical graph structural data and the query graph structural data, determine each of the historical graph structural data Similarity to the query graph structured data.

Specifically, the following formula 1 may be used to determine the degree of similarity between the historical graph structure data and the query graph structure data.

Wherein, A=I-M;

R(d,q) represents the similarity between the historical graph structure data d and the query graph structure data q;

Sim is the similarity operator, Sim(v _{d, root} , v _{q, root} ) is the first similarity measure value, v _{d, root} is the root node of historical graph structure data, v _{q, root} is the root of query graph structure data node;

为第二相似度量值，u_i∈M,u_j∈M表示第一类显著子图的叶子节点集合，由此限定了第二相似度量值中的S_d,i为历史图结构数据的第i个第一类显著子图，第二相似度量值中的S_q,为查询图结构数据的第j个第一类显著子图；

is the second similarity metric value, u _i ∈ M, u _j ∈ M represents the leaf node set of the first type of saliency subgraph, which defines S _{d, i} in the second similarity metric value as the first rank of the historical graph structure data. i first-type saliency subgraphs, S _q in the second similarity measure, is the j-th first-type saliency subgraph of the query graph structure data;

为第三相似度量值，I是所有叶子节点的集合，u_i∈A,u_j∈A表示除第一类显著子图的叶子节点集合以外的叶子节点，由此限定了第三相似度量值中的S_d,为历史图结构数据的第i个第一类非显著子图或第二类子图，第三相似度量值中的S_q,为查询图结构数据的第j个第一类非显著子图或第二类子图。

is the third similarity metric value, I is the set of all leaf nodes, u _i ∈ A, u _j ∈ A represents the leaf nodes except the leaf node set of the first type of saliency subgraph, thus defining the third similarity metric value S _{d in} the historical graph structure data is the i-th first-type non-salient sub-graph or second-type sub-graph, and S _q in the third similarity measure is the j-th first-class sub-graph of the query graph structure data Non-significant subplots or subplots of the second type.

In the above formula 1, the root node is, for example, the main diagnosis shown in Figure 2, and the M set corresponds to the subgraph to which the leaf nodes belong. ” belongs to the first category of significant subgraphs; the subgraphs corresponding to set A can be, for example, other subgraphs shown in FIG. Figure, including the second type of submap and the first type of non-significant submap.

S104, according to a preset selection rule and the degree of similarity, determine a similar medical record search result of the query medical record data among the plurality of historical medical record data, wherein the historical graph structure data corresponding to the similar medical record search result , having the similarity degree that satisfies the preset selection rule.

For example, it may be the quantity of historical medical record data in the similar medical record search result output based on the decision maker model control. For example, sort the historical medical record data from high similarity to low degree of similarity, and then use a preset truncation ratio (for example, 50%) to rank several historical medical record data in the top order of the truncation ratio as the search result of similar medical records. . According to the preset number of truncations, the historical medical record data in the front row that satisfy the number of truncations (for example, 5) may be used as the search result of similar medical records. The preset selection rule may be number-based or proportion-based, which is not limited herein.

A method for searching similar medical records provided in this embodiment is obtained by acquiring query medical record data and multiple historical medical record data; acquiring query graph structure data corresponding to the queried medical record data, and historical graph structure data corresponding to each of the historical medical record data , wherein the query graph structure data and the history graph structure data both include a first type of subgraph and a second type of subgraph, the intermediate node of the first type of subgraph is a medical record field category, and the second type of subgraph The intermediate nodes and leaf nodes of the subgraph are obtained by performing feature recognition on the first type of subgraph; according to the similarity of the root node and the similarity of the first type of subgraph in each of the historical graph structure data and the query graph structure data and the similarity degree of the second type of sub-graph, to obtain the similarity degree of each of the historical graph structure data and the query graph structure data; according to the preset selection rule and the similarity degree, determine the degree of similarity among the plurality of historical medical record data The search results of similar medical records of the query medical record data are described, so as to extract the inherent subgraphs and identifiable subgraphs in the query medical record data, and measure the correlation of the data in the corresponding subgraphs in the medical records of the query medical record data and the historical medical record data. , which improves the accuracy of finding similar medical records.

Since the first type of sub-image and the second type of sub-image are formed in different ways and the types of content embodied, the embodiment of the present invention can use two different similarity determination methods to determine the similarity and the second type of sub-image respectively. Second-class subgraph similarity.

Determining the similarity of the first type of subgraphs includes determining the similarity of the first type of salient subgraphs and the similarity of the first type of non-salient subgraphs.

In the above embodiment, it can be understood that in step S203 (according to the similarity between the historical graph structure data and the first type of significant subgraphs in the query graph structural data, determine the historical graph structural data and the query graph. the second similarity measure value of the structural data) and step S204 (according to the similarity of the first type of non-salient subgraph and the similarity of the second type of subgraph in the historical graph structural data and the query graph structural data, determine the Before the third similarity measure between the historical graph structure data and the query graph structure data), a process of calculating the similarity of the first type of salient subgraph and the similarity of the first type of non-salient subgraph may also be included.

For example, in the first type subgraph with the same intermediate node in the historical graph structure data and the query graph structure data, according to the similarity of leaf nodes, the root node and the root node in the first type subgraph The edge preset weight of the leaf node and the similarity of the root node are obtained, and the similarity between the historical graph structure data and the first type of subgraphs in the query graph structure data is obtained. Wherein, the similarity of the first type of subgraph includes the similarity of the first type of salient subgraph and the similarity of the first type of non-salient subgraph, that is, the calculation method of the similarity of the first type of subgraph is The calculation method of the similarity of the first type of salient subgraph and the similarity of the first type of non-salient subgraph.

In some embodiments, the similarity between the historical graph structure data and the first type of subgraphs in the query graph structure data may be obtained by the following formula 2.

Sim(S _d ,S _q )=Sim(u _d ,u _q )*weight(uv)*Sim(v _d ,v _q ) Formula 2

Among them, Sim represents the similarity operator; S _d represents the first type subgraph of historical graph structure data, S _q represents the first type subgraph of query graph structure data; _ud represents the first type subgraph of historical graph structure data The leaf node of S _d , u _q represents the leaf node of the first type subgraph S _q of the query graph structure data; weight(uv) represents the preset weight of the edge between the root node and the leaf node in the first type subgraph; v _d represents the history The root node of the first type subgraph S _d of the graph structure data, v _q represents the root node of the first type subgraph S _q of the query graph structure data.

The first type of subgraph in the above formula 2 can be in the graph structure data shown in Fig. 2, with age, gender, doctor's order, physical examination, type of diagnosis and treatment, chief complaint, auxiliary examination, department, history of present illness and visit time as the leaves. A subgraph of a node. Among them, the subgraphs with age, gender, type of diagnosis and treatment, and treatment time as leaf nodes are the first type of significant subgraphs, and the rest are the first type of non-significant subgraphs. The preset weights of the edges of the root node and the leaf nodes may be preset weights based on the experience of medical experts.

It can be understood that in the above step S204 (according to the similarity of the first type of non-significant subgraph and the similarity of the second type of subgraph in the historical graph structure data and the query graph structure data, determine the historical graph structure data. Before the third similarity metric value of the query graph structure data), the similarity between the historical graph structure data and the second type of subgraphs in the query graph structure data may also be obtained first. Specifically, the preset weights of the edges of the root node and the intermediate nodes in the second type of subgraph and the statistical weights of the edges of the leaf nodes and the intermediate nodes in the second type of subgraph may be obtained first. It should be understood that, in the historical graph structure data, ud _{, i} represent the leaf nodes of the second type of subgraph, cd represents the intermediate node, and v _d represents the root node, then the edge relationship is: ud _{, c d +} c _d v _d =ud _, v _d . The weight value of the edge c _d v _d between the root node and the intermediate node, for different leaf nodes _ud and ud belonging to the same intermediate node c _{d ,} belong to the same type of leaf node, that is, c _d v _d = ud _{d ,} v _d - ud _, c _d is the same as c _d v _d = ud _, v _d - ud _, c _d . Therefore, the weights of the edges ud _and c _d between the intermediate node and the leaf nodes are positively linearly related to the weights of the edges ud _and v _d between the root node and the leaf nodes. For example, the weights of edges ud _{, i} v _d represent the correlation statistics between the main diagnosis and leaf nodes such as symptom A and symptom B, which are medically interpretable. Therefore, the leaf nodes are determined according to the statistical results of the correlation between the root node and the leaf nodes. Statistical weights of edges between nodes and intermediate nodes. For example, the descriptive information of the main diagnosis and features is based on the weighted average of mutual information and chi-square statistics. For example, specifically, at least one leaf node under the middle node in the second type of subgraph may be obtained first; and the mutual information between the root node and the at least one leaf node in the second type of subgraph may be obtained. Then obtain the chi-square statistic value of the root node and the at least one leaf node in the second type of subgraph; finally, the weighted sum of the mutual information value and the chi-square statistic value corresponding to the intermediate node, as the statistical weight of the edge between the leaf node and the intermediate node in the second type of subgraph. Next, in the second type of subgraph with the same intermediate node in the historical graph structure data and the query graph structure data, according to the similarity of leaf nodes in the second type of subgraph, the root node and the The edge preset weight of the intermediate node, the edge statistical weight of the leaf node and the intermediate node, and the similarity of the root node are obtained, and the similarity of the second type of sub-graph in the historical graph structure data and the query graph structure data is obtained. For example, the intermediate node of the historical graph structure data is the second type subgraph of symptoms, and the similarity calculation is performed on the second type subgraph of which the intermediate node of the query graph structure data is the symptom, and the second type subgraph of which the intermediate node is the symptom is obtained. Image similarity.

For example, the following formula 3 may be used to obtain the similarity between the historical graph structure data and the second type of subgraphs in the query graph structure data.

Among them, the second type of subgraph of historical graph structure data is S _d ={ud _,1 ,ud _,2 ,...,ud _,m , _{cd ,v d }} ,{ud _{,1 cd} ,u _{d, 2} c _d ,…,u _d,m c _d ,c _d v _d }), ud _,m is the mth leaf node of the second type subgraph of the historical graph structure data, c _d is the historical graph structure The intermediate node of the second type of subgraph of the data, v _d is the root node of the second type of subgraph of the historical graph structure data, ud _, c _d are the leaf nodes ud _, and the edge of the intermediate node c _d , c _d v _d is the edge between the intermediate node c _d and the root node v _d ; the second subgraph of the query graph structure data is S _q ={u _q,1 ,u _q,2 ,…,u _q,n ,c _q ,v _q },{u _q,1 c _q ,u _q,2 c _q ,…,u _q,n c _q ,c _q v _q }), u _q,n is the second type of subgraph of the query graph structure data The nth leaf node, c _q is the middle node of the second type subgraph of the query graph structure data, v _q is the root node of the second type subgraph of the query graph structure data, u _{q, n} c _q are the leaf nodes u _{q, n} and the edge of the intermediate node c _q , c _q v _q is the edge of the intermediate node c _q and the root node v _q ;

weight(u _{d, ic d} ) represents the edge statistical weight between leaf nodes and intermediate nodes in the second type of subgraph;

weight(c _d v _d ) represents the preset weight of the edge between the root node and the intermediate node in the second type of subgraph;

其中，α、β为常量。operator

Among them, α and β are constants.

Table I

In the above embodiment of determining the similarity of the first type of subgraphs, before acquiring the similarity between the historical graph structure data and the first type of subgraphs in the query graph structure data, the data according to the leaf node may also be included. The steps of determining the similarity of each leaf node of the first type of subgraph by selecting the corresponding method. Specifically, the data types of the leaf nodes of the first type of subgraphs in the historical graph structure data and the query graph structure data may be obtained. Referring to Table 1, there are four optional similarity determination models for data types provided by the embodiment of the present invention. In the similarity determination model corresponding to the complex data type shown in Table 1, the multiplication factor attribute set is the set composed of all multiplication factor attribute information of node x, and the multiplication factor attribute set is the set composed of all multiplication factor attribute information of node x, where Nodes x and y for similarity comparison have one-to-one corresponding multiplication factor attribute information and one-to-one corresponding adding factor attribute information. The data types of the leaf nodes of the first type of subgraphs are usually simple data types, such as numeric data types, Boolean data types, and text data types in Table 1. A target similarity determination model corresponding to the data type is obtained, for example, see FIG. 1 for the corresponding relationship. Then, the model is determined according to the target similarity, and the description information of the leaf nodes in the first type of subgraphs in the historical graph structure data and the query graph structure data, to obtain the historical graph structure data and the query The similarity of each leaf node of the first type of subgraph in the graph structure data. That is, the description information of the leaf nodes in the first type of subgraph is calculated by the target similarity determination model, and the similarity of each leaf node of the first type of subgraph is obtained. The numerical data type is, for example, the description information of fields such as age and medical treatment time. The Boolean data type is, for example, the description information of the gender field (for example, the default value is 1 for male and 0 for female). The text data type is, for example, the description information of fields such as chief complaint and current illness history. By selecting the corresponding target similarity determination model to calculate the similarity of each leaf node, the accuracy of the similarity of the leaf nodes is improved, and the accuracy of the final search structure is further improved.

In the above embodiment of determining the similarity of the second type of subgraph, before acquiring the similarity between the historical graph structure data and the second type of subgraph in the query graph structure data, it may also include selecting the corresponding subgraph according to the data type of the leaf node. The steps of determining the similarity of each leaf node of the second type of subgraph. Specifically, the data types of the leaf nodes of the second type of subgraphs in the historical graph structure data and the query graph structure data may be obtained. The data type of the leaf node of the second type of subgraph may be any of the four data types in Table 1. Among them, the numeric data type, the Boolean data type, and the text data type are simple data types, and the others are complex data types.

If the data type is a simple data type, a target similarity determination model is determined according to the data type corresponding to the leaf node of the simple data type. According to the target similarity determination model and the description information of the leaf nodes of the simple data type, the relationship between the historical graph structure data and the simple data types of the second type of subgraphs in the query graph structure data is obtained. Leaf node similarity. The implementation manner of the similarity calculation of the leaf nodes of the second type of subgraphs of the simple data type is similar to the implementation manner of the similarity calculation of the leaf nodes of the first type of subgraphs, and will not be repeated here.

If the data type is a complex data type, the multiplication factor attribute information and the addition factor attribute information corresponding to the leaf node of the complex data type are acquired. For complex data types, such as the second type of subgraph with "symptom" as the intermediate node in the query graph structure data, the leaf node includes "cough", and the multiplication factor attribute information of "cough" is "1" (indicating that there is), which adds The factor attribute information is, for example, "severe" and "24 hours". Then, the product of the similarity of the attribute information of the multiplication factor corresponding to the leaf node of the complex data type can be weighted with the similarity of the attribute information of the multiplication factor corresponding to the leaf node of the complex data type The product of the sum is used as the similarity of leaf nodes of each complex data type of the second type of subgraph in the historical graph structure data and the query graph structure data. For example, if the attribute information of the multiplication factor corresponding to the historical graph structure data is "1" (indicating that there is), and the attribute information of the addition factor is, for example, "slight" and "1 hour", then the similarity of the Boolean data type is used to determine the multiplication factor of the model. (1,1) Calculate, use the similarity determination model of the text data type to calculate the attribute information of the addition factor (“very serious”, “slight”), and use the similarity determination model of the numerical data type to determine the attribute information of the addition factor ( "24 hours", "1 hour") to calculate. The description information of each leaf node in the second type of subgraph corresponds to the multiplication factor attribute information or the addition factor attribute information, and the corresponding relationship may be preset. When the description information of each leaf node in the second type of subgraph is detected, according to the record of the description information of each leaf node in the second type of subgraph in the preset corresponding relationship, the corresponding information of each leaf node in the second type of subgraph is obtained. Multiply factor attribute information, plus factor attribute information.

In the embodiment of the above method for finding similar medical records, matching and rearranging of similar medical records are realized, and the metric calculation of the relationship between the characteristics of the medical records, the metric calculation of each feature attribute of the medical records, etc. are realized, thereby improving the accuracy of searching for similar medical records. .

Referring to FIG. 5, it is a schematic structural diagram of a similar medical record search device provided by an embodiment of the present invention. The similar medical record search device 50 shown in FIG. 5 includes:

The medical record obtaining module 51 is used for obtaining and querying medical record data and a plurality of historical medical record data.

The graph structuring module 52 is configured to obtain query graph structure data corresponding to the query medical record data, and historical graph structure data corresponding to each of the historical medical record data, wherein the query graph structure data and the historical graph structure data Both include the first type subgraph and the second type subgraph, the middle node of the first type subgraph is the medical record field category, and the middle node and leaf node of the second type subgraph are related to the first type subgraph. The graph is obtained by feature recognition.

The processing module 53 is configured to obtain each of the historical graph structural data according to the similarity of the root node, the similarity of the first type of sub-graph and the similarity of the second type of sub-graph in the historical graph structural data and the query graph structural data Similarity to the query graph structured data.

The selection module 54 is configured to, according to the preset selection rule and the degree of similarity, determine the similar medical record search results of the query medical record data in the plurality of historical medical record data, wherein the similar medical record search results corresponding to the The historical graph structure data has the similarity degree that satisfies the preset selection rule.

The apparatus for searching similar medical records in the embodiment shown in FIG. 5 can correspondingly be used to execute the steps executed by the server in the method embodiment shown in FIG. 1 , and the implementation principles and technical effects thereof are similar, which will not be repeated here.

Optionally, the processing module 53 is configured to determine the first similarity measure between the historical graph structural data and the query graph structural data according to the similarity of the root node in the historical graph structural data and the query graph structural data. value; according to the saliency category of the leaf node, in the first type of subgraph, determine the first type of salient subgraph and the first type of non-salient subgraph; according to the historical graph structure data and the query graph structure data the similarity of the first type of significant subgraphs in Class non-salient sub-graph similarity and second-class sub-graph similarity, determine the third similarity measure between the historical graph structure data and the query graph structural data; according to each of the historical graph structural data and the query graph The first similarity metric value, the second similarity metric value and the third similarity metric value of the structural data determine the degree of similarity between each of the historical graph structural data and the query graph structural data.

Optionally, the processing module 53 is to determine the historical graph structural data and the query graph structural data according to the similarity of the first type of significant subgraphs in the historical graph structural data and the query graph structural data. The second similarity metric value of Before the third similarity metric between the data and the query graph structure data, it is also used to, in the first type subgraph with the same intermediate node of the historical graph structure data and the query graph structure data, according to the The similarity of the leaf nodes in the first type of subgraph, the preset weight of the edge between the root node and the leaf node, and the similarity of the root node are obtained, and the similarity between the historical graph structure data and the first type of subgraph in the query graph structure data is obtained. degree, wherein the similarity of the first type of subgraphs includes the similarity of the first type of salient subgraphs and the similarity of the first type of non-salient subgraphs.

Optionally, the processing module 53 is configured to use the product of the similarity between the historical graph structure data and the first type of significant subgraphs of the query graph structural data as the historical graph structural data and the query graph structure. The second similarity measure for the data.

Optionally, the processing module 53 determines the historical graph according to the similarity of the first type of non-significant subgraphs and the similarity of the second type of subgraphs in the historical graph structural data and the query graph structural data. Before the third similarity metric between the structural data and the query graph structural data, it is also used to obtain the edge statistical weights of leaf nodes and intermediate nodes in the second type of subgraph; between the historical graph structural data and the query In the second type of subgraph with the same intermediate node in the graph structure data, the preset weights are based on the similarity of leaf nodes, the edge between the root node and the intermediate node, and the edge between the leaf node and the intermediate node in the second type of subgraph. The weight and the similarity of the root node are counted, and the similarity between the historical graph structure data and the second type of subgraphs in the query graph structure data is obtained.

Optionally, the processing module 53 is configured to obtain the similarity of the second type of sub-graphs in the historical graph structure data and the query graph structure data with the following formula 3:

Among them, the second type of subgraph of historical graph structure data is S _d ={ud _,1 ,ud _,2 ,...,ud _,m , _{cd ,v d }} ,{ud _{,1 cd} ,u _{d, 2} c _d ,…,u _d,m c _d ,c _d v _d }), ud _,m is the mth leaf node of the second type subgraph of the historical graph structure data, c _d is the historical graph structure The intermediate node of the second type of subgraph of the data, v _d is the root node of the second type of subgraph of the historical graph structure data, ud _{, m} c _d are the leaf nodes ud _, and the edge of the intermediate node c _d , c _d v _d is the edge between the intermediate node c _d and the root node v _d ; the second type of subgraph of the query graph structure data is S _q ={u _q,1 ,u _q,2 ,...,u _q,n ,c _q , v _q },{u _q,1 c _q ,u _q,2 c _q ,…,u _q,n c _q ,c _q v _q }), u _q,n is the second type of subgraph of query graph structure data The nth leaf node of , c _q is the middle node of the second type subgraph of the query graph structure data, v _q is the root node of the second type subgraph of the query graph structure data, u _q, c _q are the leaf nodes u _q, the edge with the intermediate node c _q , c _q v _q is the edge between the intermediate node c _q and the root node v _q ;

weight(u _{d, ic d} ) represents the edge statistical weight between leaf nodes and intermediate nodes in the second type of subgraph;

weight(c _d v _d ) represents the preset weight of the edge between the root node and the intermediate node in the second type of subgraph;

其中，α、β为常量。operator

Among them, α and β are constants.

Optionally, the processing module 53 is configured to obtain at least one leaf node under the intermediate node in the second type of subgraph; obtain mutual information between the root node and the at least one leaf node in the second type of subgraph; Obtain the chi-square statistic value of the root node and the at least one leaf node in the second type of subgraph; take the weighted sum of the mutual information value and the chi-square statistic value corresponding to the intermediate node as the Statistical weights of edges between the leaf nodes and the intermediate nodes in the second type of subgraph.

Optionally, the processing module 53 is used to compare the similarity between the historical graph structure data and the similarity of each of the first type of non-significant subgraphs and the similarity of each of the second type of subgraphs in the query graph structure data. and, as the third similarity measure between the historical graph structure data and the query graph structure data.

Optionally, the processing module 53, in the first type of subgraph having the same intermediate node in the historical graph structure data and the query graph structure data, according to the leaves in the first type of subgraph. Node similarity, the edge preset weight between the root node and the leaf node, and the similarity of the root node, to obtain the similarity between the historical graph structure data and the first type of subgraph in the query graph structure data, wherein the first The similarity of the class subgraphs includes the similarity of the first type of salient subgraphs and the similarity of the first type of non-salient subgraphs, and is also used to obtain the historical graph structure data and the query graph structure data, the data type of the leaf node of the first type of subgraph; obtain the target similarity determination model corresponding to the data type; determine the model according to the target similarity, and the historical graph structure data and the query graph From the description information of the leaf nodes in the first type of subgraph in the structural data, the similarity between the historical graph structure data and each leaf node of the first type of subgraph in the query graph structure data is obtained.

Optionally, the processing module 53, in the second type of subgraph having the same intermediate node in the historical graph structure data and the query graph structure data, according to the leaf of the second type of subgraph. Node similarity, the edge preset weight between the root node and the intermediate node, the edge statistical weight between the leaf node and the intermediate node, and the root node similarity, and obtain the second type of subgraph in the historical graph structure data and the query graph structure data Before the similarity, it is also used to obtain the data type of the leaf node of the second type of subgraph in the historical graph structure data and the query graph structure data; if the data type is a simple data type, according to the simple The data type corresponding to the leaf node of the data type is to determine the target similarity determination model; according to the target similarity determination model, and the description information of the leaf node of the simple data type, the historical graph structure data is obtained Similarity with leaf nodes of each simple data type of the second type of subgraph in the query graph structure data; if the data type is a complex data type, obtain the multiplication factor corresponding to the leaf node of the complex data type attribute information and factor attribute information; the product of the similarity of each of the multiplication factor attribute information corresponding to the leaf node of the complex data type and the attribute information of each addition factor corresponding to the leaf node of the complex data type The product of the weighted sum of the similarities is taken as the similarity of leaf nodes of each complex data type of the second type of subgraph in the historical graph structure data and the query graph structure data.

Optionally, the processing module 53 obtains the similarity of the root node, the similarity of the first type of sub-graph and the similarity of the second type of sub-graph according to the historical graph structure data and the query graph structure data. Before the similarity between the historical graph structural data and the query graph structural data, it is also used to determine the model and the description information of the root node according to the preset textual similarity, and determine the relationship between the historical graph structural data and the Query the similarity of root nodes in graph structure data.

Optionally, the graph structuring module 52 is configured to take the main diagnosis of the queried medical record as the description information of the root node; take the medical record field category included in the queried medical record as the first type of intermediate node, and use the field category The included fields are respectively used as the first-type leaf nodes of the field category; the description information corresponding to the fields is used as the description information of the first-type leaf nodes; the preset features are used as the second-type intermediate nodes, and Perform natural language understanding (NLU) identification on each of the fields and the description information of each of the fields with the feature, and obtain the description information of the feature, or obtain the description information of the feature and the factor-added attribute of the description information. information or multiplication factor attribute information; take the description information of the feature as the second type leaf node of the feature, wherein the data type of the second type leaf node with the addition factor attribute information or multiplication factor attribute information is complex Data type; according to the root node, the first type intermediate node, the first type leaf node, the second type intermediate node and the second type leaf node, obtain the query corresponding to the query medical record data Graph structure data, wherein the root node, the first type intermediate node and the first type leaf node form a first type subgraph; the root node, the second type intermediate node and the second type Class leaf nodes form a second class of subgraphs.

Optionally, the medical record obtaining module 51 is configured to obtain and query medical record data; according to the query medical record data, determine a full-text search keyword; perform a full-text search of medical records in the medical record database according to the full-text search key, and obtain a full-text search containing the full-text search key. Multiple historical medical record data for keywords.

6 is a schematic diagram of a hardware structure of a device provided by an embodiment of the present invention, the device 60 includes: a processor 61, a memory 62, and a computer program; wherein

The memory 62 is used for storing the computer program, and the memory can also be a flash memory (flash). The computer program is, for example, an application program, a functional module, and the like that implement the above-mentioned method.

The processor 61 is configured to execute the computer program stored in the memory, so as to realize each step executed by the server in the above method for searching similar medical records. For details, refer to the relevant descriptions in the foregoing method embodiments.

Optionally, the memory 62 may be independent or integrated with the processor 61 .

When the memory 62 is a device independent of the processor 61, the device may further include:

The bus 63 is used to connect the memory 62 and the processor 61 . The device in FIG. 6 may further include a transmitter (not shown in the figure) for sending out the similar medical record search results of the query medical record data obtained by the processor 61 .

The present invention further provides a readable storage medium, where a computer program is stored in the readable storage medium, and when the computer program is executed by a processor, is used to implement the search methods for similar medical records provided by the above-mentioned various embodiments.

The readable storage medium may be a computer storage medium or a communication medium. Communication media includes any medium that facilitates transfer of a computer program from one place to another. Computer storage media can be any available media that can be accessed by a general purpose or special purpose computer. For example, a readable storage medium is coupled to the processor such that the processor can read information from, and write information to, the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium may be located in application specific integrated circuits (Application Specific Integrated Circuits, ASIC for short). Alternatively, the ASIC may be located in the user equipment. Of course, the processor and the readable storage medium may also exist in the communication device as discrete components. The readable storage medium may be read only memory (ROM), random access memory (RAM), CD-ROMs, magnetic tapes, floppy disks, optical data storage devices, and the like.

The present invention also provides a program product including execution instructions stored in a readable storage medium. At least one processor of the device can read the execution instruction from the readable storage medium, and the execution of the execution instruction by the at least one processor causes the device to implement the similar medical record search method provided by the above-mentioned various embodiments.

In the embodiment of the above device, it should be understood that the processor may be a central processing unit (English: Central Processing Unit, referred to as: CPU), or other general-purpose processors, digital signal processors (English: Digital Signal Processor, referred to as: DSP) ), application specific integrated circuit (English: Application Specific Integrated Circuit, referred to as: ASIC) and so on. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in conjunction with the present invention can be directly embodied as executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: The technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features thereof can be equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention. scope.

Claims (16)

1. a similar medical record search method, is characterized in that, comprises: Obtain and query medical record data and multiple historical medical record data; Obtain the query graph structure data corresponding to the query medical record data, and the history graph structure data corresponding to each of the historical medical record data, wherein the query graph structure data and the historical graph structure data both include the first type of subgraph and The second type of subgraph, the middle node of the first type of subgraph is a medical record field category, and the middle node and the leaf node of the second type of subgraph are obtained by performing feature recognition on the first type of subgraph; Obtain each of the historical graph structural data and the query graph structure according to the similarity of the root node, the similarity of the first type of subgraph and the similarity of the second type of subgraph in each of the historical graph structural data and the query graph structural data The similarity degree of data; the similarity of the first type of subgraph is based on the first type of subgraph with the same intermediate node in the historical graph structure data and the query graph structure data, according to the first type of subgraph The similarity of the middle leaf node, the preset weight of the edge between the root node and the leaf node, and the similarity of the root node are obtained; the similarity of the second type of subgraph is obtained in the relationship between the historical graph structure data and the query graph structure data. In the second type subgraph of the same intermediate node, according to the leaf node similarity of the second type subgraph, the preset weight of the edge between the root node and the intermediate node, the statistical weight of the edge between the leaf node and the intermediate node, and the similarity of the root node obtained; According to the preset selection rule and the similarity degree, a similar medical record search result of the query medical record data is determined from the plurality of historical medical record data, wherein the historical graph structure data corresponding to the similar medical record search result has The degree of similarity that satisfies the preset selection rule. 2. The method according to claim 1, characterized in that, according to the similarity of the root node, the similarity of the first type of subgraph and the similarity of the second type of subgraph in each of the historical graph structure data and the query graph structure data Similarity, obtains the similarity between the historical graph structure data and the query graph structure data, including: determining a first similarity measure between the historical graph structural data and the query graph structural data according to the similarity between the root node in the historical graph structural data and the query graph structural data; According to the saliency category of the leaf node, in the first type of subgraph, determine the first type of saliency subgraph and the first type of non-salient subgraph; determining a second similarity measure between the historical graph structural data and the query graph structural data according to the similarity of the first type of significant subgraphs in the historical graph structural data and the query graph structural data; According to the similarity of the first type of non-salient sub-graph and the similarity of the second type of sub-graph in the historical graph structure data and the query graph structure data, determine the first type of similarity between the historical graph structure data and the query graph structure data Three similarity measures; According to the first similarity metric value, the second similarity metric value and the third similarity metric value of each of the historical graph structural data and the query graph structural data, determine the relationship between each historical graph structural data and the query graph structural data. Describe the similarity of query graph structured data. 3 . The method according to claim 2 , wherein, determining the historical graph structural data according to the similarity of the first type of significant subgraphs in the historical graph structural data and the query graph structural data. 4 . a second similarity measure with the query graph structured data; and determining the historical graph structural data and the query graph structural data according to the similarity of the first type of non-significant subgraph and the second type of subgraph similarity in the historical graph structural data and the query graph structural data Before the third similarity measure of , also includes: In the first type of subgraph with the same intermediate node in the historical graph structure data and the query graph structure data, according to the similarity of leaf nodes, the edge between the root node and the leaf node in the first type of subgraph Preset weight and root node similarity, and obtain the similarity between the historical graph structure data and the first type of subgraph in the query graph structure data, wherein the similarity of the first type of subgraph includes the first type of subgraph. Class saliency subgraph similarity and the first class non-salient subgraph similarity. 4. The method according to claim 2 or 3, wherein the historical graph structure is determined according to the similarity of the first type of significant subgraphs in the historical graph structural data and the query graph structural data A second similarity measure between the data and the query graph structure data, including: The product of the similarity between the historical graph structure data and each first type of significant subgraphs of the query graph structure data is used as a second similarity measure between the historical graph structure data and the query graph structure data. 5. The method according to claim 2, characterized in that, in the first type of non-significant subgraph similarity and the second type of subgraph similarity according to the historical graph structure data and the query graph structure data before determining the third similarity metric between the historical graph structure data and the query graph structure data, further comprising: Obtain the edge statistical weights of leaf nodes and intermediate nodes in the second type of subgraph; In the second type of subgraph with the same intermediate node in the historical graph structure data and the query graph structure data, according to the similarity of leaf nodes, the edge between the root node and the intermediate node in the second type of subgraph Preset weights, edge statistical weights between leaf nodes and intermediate nodes, and root node similarity are obtained, and the similarity between the historical graph structure data and the second type of subgraphs in the query graph structure data is obtained. 6. The method according to claim 5, characterized in that, according to the similarity of leaf nodes in the second type of subgraph, the preset weight of the edge between the root node and the intermediate node, the statistical weight of the edge between the leaf node and the intermediate node, and Root node similarity, obtaining the similarity between the historical graph structure data and the second type of subgraphs in the query graph structure data, including: The similarity between the historical graph structure data and the second type of subgraphs in the query graph structure data is obtained by the following formula:

Among them, the second type of subgraph of the historical graph structure data is S _d =({ud _,1 ,ud _,2 ,...,ud _,m ,cd,vd _} ,{ _{ud ,} 1c _d , ud _{, 2} c _d , ..., ud _{, m} c _d , c _d v _d }), ud _{, m} is the mth leaf node of the second type subgraph of the historical graph structure data, c _d is the middle node of the second type subgraph of the historical graph structure data, v _d is the root node of the second type subgraph of the historical graph structure data, ud _{, m} c _d are the leaf nodes ud _{, m} and the middle node c The edge of _d , cd v _d is the edge between the intermediate node _cd and the root node v _{d ;} the second type of subgraph for query graph structure data is S _q = ({u _{q, 1} , u _{q, 2} , ... , uq _,n , _cq , _vq },{ _{uq ,} 1cq, _{uq , 2cq} ,...,uq _{, ncq} , _cqvq }),uq _,n is the nth leaf node of the second type subgraph of the query graph structure data, c _q is the middle node of the second type subgraph of the query graph structure data, v _q is the root of the second type subgraph of the query graph structure data Node, u _{q, n} c _q is the edge between the leaf node u _{q, n} and the intermediate node c _q , and c _q v _q is the edge between the intermediate node c _q and the root node v _q ; weight(u _{d, ic d} ) represents the edge statistical weight between leaf nodes and intermediate nodes in the second type of subgraph; weight(c _d v _d ) represents the preset weight of the edge between the root node and the intermediate node in the second type of subgraph; operator

Among them, α and β are constants. 7. The method according to claim 5, wherein the obtaining the edge statistical weights of leaf nodes and intermediate nodes in the second type of subgraph comprises: Obtain at least one leaf node under the intermediate node in the second type of subgraph; obtaining the mutual information value of the root node and the at least one leaf node in the second type of subgraph; Obtain the chi-square statistics of the root node and the at least one leaf node in the second type of subgraph; The weighted sum of the mutual information value and the chi-square statistic value corresponding to the intermediate node is used as the edge statistical weight of the leaf node and the intermediate node in the second type of subgraph. 8. The method according to claim 2, 3, 5, 6 or 7, characterized in that, according to the similarity of the first type of non-salient sub-graphs in the historical graph structure data and the query graph structure data and the second type of sub-graph similarity, determine the third similarity measure of the historical graph structure data and the query graph structure data, including: The sum of the similarity of each of the first type of non-significant subgraphs and the similarity of each of the second type of subgraphs in the historical graph structure data and the query graph structure data is taken as the historical graph structure data and A third similarity measure of the query graph structure data. 9 . The method according to claim 3 , wherein, in the first type of subgraph having the same intermediate node in the historical graph structure data and the query graph structure data, according to the first The similarity of leaf nodes in a class of subgraphs, the edge preset weights between the root node and the leaf nodes, and the similarity of root nodes, and the similarity between the historical graph structure data and the first class of subgraphs in the query graph structure data is obtained, Wherein, the similarity of the first type of subgraphs includes the similarity of the first type of salient subgraphs and the similarity of the first type of non-salient subgraphs, and further includes: Obtain the data type of the leaf node of the first type of subgraph in the historical graph structure data and the query graph structure data; obtaining a target similarity determination model corresponding to the data type; According to the target similarity determination model, and the description information of the leaf nodes in the first type subgraph in the history graph structure data and the query graph structure data, the history graph structure data and the query graph are obtained. The similarity of each leaf node of the first type of subgraph in the structural data. 10 . The method according to claim 5 , wherein, in the second type of subgraph having the same intermediate node in the historical graph structure data and the query graph structure data, according to the first The similarity of leaf nodes in the second-class subgraph, the preset edge weight between the root node and the intermediate node, the statistical weight of the edge between the leaf node and the intermediate node, and the similarity of the root node are obtained, and the historical graph structure data and the query graph are obtained. Before the second type of subgraph similarity in the structural data, it also includes: Obtain the data type of the leaf node of the second type of subgraph in the historical graph structure data and the query graph structure data; If the data type is a simple data type, the target similarity determination model is determined according to the data type corresponding to the leaf node of the simple data type; the model is determined according to the target similarity, and the The description information of the leaf node, obtaining the similarity of the leaf node of each simple data type of the second type of subgraph in the historical graph structure data and the query graph structure data; If the data type is a complex data type, obtain the multiplication factor attribute information and the multiplication factor attribute information corresponding to the leaf node of the complex data type; The product of the similarity degrees, and the product of the weighted sum of the similarity degrees of each of the factor-added attribute information corresponding to the leaf nodes of the complex data type, as the historical graph structure data and the query graph structure data. The similarity of leaf nodes of each complex data type of the second type of subgraph. 11. The method according to claim 1, 9 or 10, characterized in that, according to the similarity of the root node, the similarity of the first type of sub-graph and the similarity of the first type of sub-graph according to each of the historical graph structure data and the query graph structure data. The similarity of the two types of subgraphs, before obtaining the similarity between the historical graph structure data and the query graph structure data, further includes: According to the preset text-type similarity determination model and the description information of the root node, the similarity between each historical graph structure data and the root node in the query graph structure data is determined. 12. The method according to claim 1, 9 or 10, wherein the acquiring query graph structure data corresponding to the query medical record data, wherein the query graph structure data and the history graph structure data are both Including a first type of subgraph and a second type of subgraph, the middle node of the first type of subgraph is the medical record field category, and the middle node and leaf node of the second type of subgraph are related to the first type of subgraph. Obtained by feature identification, including: Taking the main diagnosis of the query medical record as the description information of the root node; The medical record field category included in the query medical record is used as the first-type intermediate node, and the fields included in the field category are respectively used as the first-type leaf node of the field category; the description information corresponding to the field is used as the Describe the description information of the first type of leaf node; Taking the preset feature as the second type of intermediate node, and using the feature to perform natural language understanding NLU identification on each of the fields and the description information of each of the fields, to obtain the description information of the feature, or to obtain the The description information of the feature and the attribute information of the addition factor or the attribute information of the multiplication factor of the description information; the description information of the feature is taken as the second type of leaf node of the feature, which has the attribute information of the addition factor or the multiplication factor The data type of the second type of leaf node of the attribute information is a complex data type; According to the root node, the first type intermediate node, the first type leaf node, the second type intermediate node and the second type leaf node, obtain the query graph structure data corresponding to the query medical record data , wherein the root node, the first type intermediate node and the first type leaf node form a first type subgraph; the root node, the second type intermediate node and the second type leaf node A second type of subgraph is formed. 13. The method according to claim 1, wherein the obtaining and querying medical record data and a plurality of historical medical record data comprises: Obtain and query medical record data; Determine full-text search keywords according to the query medical record data; According to the full-text search key, a full-text search of medical records is performed in the medical record database, and a plurality of historical medical record data including the full-text search key are obtained. 14. A similar medical record search device, characterized in that it comprises: The medical record acquisition module is used to acquire and query medical record data and multiple historical medical record data; The graph structure module is used to obtain the query graph structure data corresponding to the query medical record data, and the historical graph structure data corresponding to each of the historical medical record data, wherein the query graph structure data and the historical graph structure data are both Including a first type of subgraph and a second type of subgraph, the middle node of the first type of subgraph is the medical record field category, and the middle node and leaf node of the second type of subgraph are related to the first type of subgraph. obtained by feature identification; The processing module is configured to obtain each of the historical graph structural data and the similarity of each of the historical graph structural data according to the similarity of the root node, the similarity of the first type of subgraph and the similarity of the second type of subgraph in each of the historical graph structural data and the query graph structural data. The similarity degree of the query graph structure data; the first type of subgraph similarity is in the first type of subgraphs with the same intermediate node of the historical graph structure data and the query graph structure data, according to the The similarity of the leaf nodes in the first type of subgraph, the preset weight of the edge between the root node and the leaf node, and the similarity of the root node are obtained; the similarity of the second type of subgraph is obtained from the historical graph structure data and the query In the second type of subgraph with the same intermediate nodes in the graph structure data, according to the leaf node similarity of the second type of subgraph, the preset weight of the edge between the root node and the intermediate node, and the statistical weight of the edge between the leaf node and the intermediate node And the similarity of the root node is obtained; A selection module, configured to determine a similar medical record search result of the query medical record data in the plurality of historical medical record data according to a preset selection rule and the similarity degree, wherein the history corresponding to the similar medical record search result The graph structure data has the similarity degree satisfying the preset selection rule. 15. An electronic device, comprising: a memory, a processor, and a computer program, wherein the computer program is stored in the memory, and the processor executes the computer program to execute any one of claims 1 to 13. Describe the method of finding similar medical records. 16. A readable storage medium, wherein a computer program is stored in the readable storage medium, and when the computer program is executed by a processor, the computer program is used to implement the computer program to execute any one of claims 1 to 13. Describe the method of finding similar medical records.

Images