A system for obtaining target identifiers corresponding to medical record texts

Abstract

The present invention relates to a system for acquiring target identifiers corresponding to medical record texts, the system comprising a sample text list and a sample rule set, a processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the following steps are implemented: according to the initial electronic medical record text, a preset text list corresponding to the initial electronic medical record text is acquired, a first candidate rule set and a first instruction information set are acquired, the first instruction information set is input into a rule engine, a first target identifier set is acquired, a second candidate rule set and a second instruction information set are acquired, the second instruction information set is input into a candidate LLM model, and a second target identifier set is acquired, thereby acquiring a target identifier set corresponding to the initial electronic medical record text. The present invention combines rules with LLM models, improves the efficiency of acquiring target identifiers, and adjusts the weights corresponding to different types of instruction information according to the output results, so that the accuracy of the acquired target identifiers is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of text processing, and in particular to a system for acquiring target identifiers corresponding to medical record texts. Background Art

[0002] With the continuous growth of medical business volume and the continuous development of artificial intelligence technology, the electronicization of medical records has become a trend. How to efficiently and accurately obtain the existing problems in the electronic medical record text has become a hot research direction. With the widespread application of large language models in natural language processing, how to effectively utilize large language models and combine large prediction models with other methods has become an important method to solve problems in electronic medical record texts.

[0003] In the prior art, the method for obtaining the target identifier of the electronic medical record text is: obtaining the electronic medical record text, processing the electronic medical record text into short texts, using the LLM model to obtain keywords in each short text, and determining the identifier corresponding to the electronic medical record text based on the obtained keywords.

[0004] In summary, the method of obtaining the target medical record text has the following problems: it is limited to the LLM model, which makes the efficiency of obtaining the target identifier low; the weights of different output results are not adjusted, which makes the accuracy of the target identifier corresponding to the obtained initial electronic medical record text low. Summary of the invention

[0005] In view of the above technical problems, the technical solution adopted by the present invention is: a system for obtaining target identifiers corresponding to medical record texts, the system comprising: a sample text list A={A1, ..., A i , ..., A n}、The sample rule set B corresponding to A = {B1, ..., B i , ..., B n}, a processor and a memory storing a computer program, wherein B i ={B 1 i , B 2 i}, A i is the i-th sample text, B 1 i A i The corresponding first sample rule list, B 2 i A i The second sample rule list, i=1...n, n is the number of sample texts, when the computer program is executed by the processor, the following steps are implemented:

[0006] S100, according to the initial electronic medical record text, obtain a preset text list E corresponding to the initial electronic medical record text = {E1, ..., Er ,……,E s}, E r is the rth preset text, r=1…s, s is the number of preset texts.

[0007] S200, according to E, obtain the first candidate rule set G corresponding to E = {G1, ..., G r , ..., G s}, G r For E r The corresponding first candidate rule list, where E r With A i When consistent, obtain G r =B 1 i .

[0008] S300, according to G, obtain the first instruction information set M corresponding to E={M1, ..., M r , ..., M s}, where M r For E r The corresponding first instruction information list.

[0009] S400, input M into the rule engine, and obtain a first target identifier set C={C1, ..., C r , ..., C s}, where C r M r The corresponding first target identifier list includes a plurality of first target identifiers, and the first target identifiers are corresponding symbols obtained by the rule engine through screening and matching based on the first instruction information.

[0010] S500, according to E, obtain a second candidate rule set H corresponding to E = {H1, ..., H r , ..., H s}, H r For E r The corresponding second candidate rule list, where E r With A i When they are consistent, obtain H r =B 2 i .

[0011] S600, according to H, obtain the second instruction information set N={N1, ..., N r , ..., N s}, where N r For E r The corresponding second instruction information list.

[0012] S700, input N into the candidate LLM model to obtain a second target identifier set D={D1, ..., D r , ..., D s}, where D r N r A corresponding second target identifier list, wherein the second target identifier list includes a plurality of second target identifiers, and the second target identifier is a symbol of the output of the candidate LLM model obtained based on the second instruction information in the second instruction information list.

[0013] S800, according to C and D, obtaining a target identification set corresponding to the initial electronic medical record text, wherein the target identification set includes a plurality of target identifications.

[0014] Compared with the prior art, the present invention has obvious beneficial effects. By means of the above technical solution, the system for obtaining the target identifier corresponding to the medical record text provided by the present invention can achieve considerable technical progress and practicality, and has wide industrial utilization value, and has at least the following beneficial effects:

[0015] A system for acquiring a target identifier corresponding to a medical record text, the system comprising a preset database, a processor, and a memory storing a computer program, wherein the preset database comprises a sample text list and a sample rule set corresponding to the sample text list, and when the computer program is executed by the processor, the following steps are implemented: according to an initial electronic medical record text, acquiring a preset text list corresponding to the initial electronic medical record text, according to the preset text list, acquiring a first candidate rule set corresponding to the preset text list, according to the first candidate rule set, acquiring a first instruction information set corresponding to the preset text list, inputting the first instruction information set into a rule engine, acquiring a first target identifier set, according to the preset text list , obtain a second candidate rule set corresponding to the preset text list, obtain a second instruction information set corresponding to the preset text list according to the second candidate rule set, input the second instruction information set into the candidate LLM model, obtain a second target identification set, and obtain a target identification set corresponding to the initial electronic medical record text according to the first target identification set and the second target identification set. The present invention obtains corresponding rules based on the knowledge graph, combines the rules with the LLM model, improves the efficiency of obtaining the target identification, obtains the target identification in different ways based on different output results, and adjusts the weights corresponding to different types of instruction information according to the output results, so that the accuracy of the obtained target identification is higher.

[0016] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following specifically cites a preferred embodiment and describes it in detail with the accompanying drawings as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a computer program for executing a system for acquiring a target identifier corresponding to a medical record text provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0020] Example

[0021] This embodiment provides a system for obtaining a target identifier corresponding to a medical record text, the system comprising: a sample text list A={A1, ..., A i , ..., A n}、The sample rule set B corresponding to A = {B1, ..., B i , ..., B n}, a processor and a memory storing a computer program, wherein B i ={B 1 i , B 2 i}, A i is the i-th sample text, B 1 i A i The corresponding first sample rule list, B 2 i A i A second sample rule list, i=1...n, n is the number of sample texts, when the computer program is executed by a processor, the following steps are implemented, such as Figure 1 As shown:

[0022] S100, according to the initial electronic medical record text, obtain a preset text list E corresponding to the initial electronic medical record text = {E1, ..., E r ,……,E s}, E r is the rth preset text, r=1…s, s is the number of preset texts.

[0023] Specifically, the sample text is a text representing disease-related issues obtained based on a sample electronic medical record information set, for example, whether there are long-term medical instructions for medication, whether there are contraindications to statins, and other sample texts.

[0024] Furthermore, the sample electronic medical record information set includes several sample electronic medical record information, and the sample electronic medical record information is information related to the disease obtained from the database. Among them, those skilled in the art know that any public medical database that can obtain medical records can be selected according to actual needs, and all fall within the protection scope of the present invention and will not be repeated here.

[0025] Specifically, the first sample rule list includes several first sample rules, wherein the first sample rule is a basis for selecting an answer corresponding to a sample text obtained based on the candidate knowledge graph.

[0026] Furthermore, the candidate knowledge graph includes a plurality of nodes and a plurality of edges, wherein the nodes represent entities associated with the disease, and the edges represent relationships between nodes, for example, entities such as contraindications of P2Y12 receptor antagonists, urokinase, and thrombolytic therapy, and relationships such as urokinase being a required drug for thrombolytic therapy.

[0027] Specifically, the second sample rule list includes a plurality of second sample rules, wherein the second sample rule is a basis for determining the answer corresponding to the sample text obtained based on knowledge related to the disease.

[0028] Specifically, the initial electronic medical record text includes a number of attribute features and a text corresponding to each attribute feature.

[0029] Furthermore, the attribute characteristics are characteristics obtained from the initial electronic medical record information, such as: admission records, discharge records, discharge diagnoses, medical advice, drugs, treatment plans and other attribute characteristics.

[0030] Furthermore, the initial electronic medical record information is characteristic information associated with a disease corresponding to any object obtained.

[0031] Specifically, the initial electronic medical record text is the text obtained by inputting the initial electronic medical record information into the information extraction model, wherein, those skilled in the art know that the information extraction model can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be repeated here; it can be understood that: each initial electronic medical record text corresponds to an object, and an initial electronic medical record text includes several levels, each level represents an attribute feature, and each level includes the text corresponding to the attribute feature obtained from the initial electronic medical record information.

[0032] Specifically, the preset text is a text that characterizes disease-related issues, for example, whether there are contraindications to statins, whether there are contraindications to P2Y12 receptor antagonists, and other preset texts.

[0033] S200, according to E, obtain the first candidate rule set G corresponding to E = {G1, ..., G r , ..., G s}, G r For E r The corresponding first candidate rule list, where E r With A i When consistent, obtain G r =B 1 i .

[0034] Specifically, the first candidate rule list includes a plurality of first candidate rules, and the first candidate rules are first sample rules obtained from the first sample rule list.

[0035] S300, according to G, obtain the first instruction information set M corresponding to E={M1, ..., M r , ..., M s}, where M r For E r The corresponding first instruction information list.

[0036] Specifically, the first instruction information list includes a plurality of first instruction information, and the first instruction information is instruction information generated based on the first candidate rule.

[0037] Furthermore, the first instruction information includes a first candidate text corresponding to the first candidate rule and a discrimination condition, and the first candidate text is an attribute feature obtained from the initial electronic medical record text based on the first candidate rule and a text corresponding to the attribute feature.

[0038] Furthermore, those skilled in the art are aware that methods for generating instruction information based on rules in the prior art all fall within the protection scope of the present invention and will not be described in detail herein.

[0039] S400, input M into the rule engine, and obtain a first target identifier set C={C1, ..., C r , ..., C s}, where C r M r The corresponding first target identifier list includes a plurality of first target identifiers, and the first target identifiers are corresponding symbols obtained by the rule engine through screening and matching based on the first instruction information.

[0040] Specifically, the rule engine is a component that performs rule matching based on different screening dimensions. Those skilled in the art know that the rule engine can be selected according to actual needs, which falls within the protection scope of the present invention and will not be described in detail here.

[0041] S500, according to E, obtain a second candidate rule set H corresponding to E = {H1, ..., H r , ..., H s}, H r For E r The corresponding second candidate rule list, where E r With A i When they are consistent, obtain H r =B 2 i .

[0042] Specifically, the second candidate rule list includes a plurality of second candidate rules, and the second candidate rules are second sample rules obtained from the second sample rule list.

[0043] S600, according to H, obtain the second instruction information set N={N1, ..., N r , ..., N s}, where N r For E r The corresponding second instruction information list.

[0044] Specifically, the second instruction information list includes a plurality of second instruction information, and the second instruction information is instruction information generated based on the second candidate rule.

[0045] Furthermore, the second instruction information includes a second candidate text and a discrimination condition corresponding to the second candidate rule, and the second candidate text is an attribute feature obtained from the initial electronic medical record text based on the second candidate rule and a text corresponding to the attribute feature. For example, when the second candidate rule is to judge whether there are contraindications to statins such as cardiogenic shock, myocarditis, high-risk hypertension, dyslipidemia combined with hypokalemia, severe liver or kidney function abnormalities, liver or kidney tumors based on the discharge record and discharge diagnosis, the second instruction information is based on the text corresponding to the discharge record and discharge diagnosis in the initial electronic medical record text to judge whether the object corresponding to the initial electronic medical record text has contraindications to statins such as cardiogenic shock, myocarditis, high-risk hypertension, dyslipidemia combined with hypokalemia, severe liver or kidney function abnormalities, liver or kidney tumors, and please answer "yes" or "no".

[0046] S700, input N into the candidate LLM model to obtain a second target identifier set D={D1, ..., D r , ..., D s}, where D r N r A corresponding second target identifier list, wherein the second target identifier list includes a plurality of second target identifiers, and the second target identifier is a symbol of the output of the candidate LLM model obtained based on the second instruction information in the second instruction information list.

[0047] Specifically, the candidate LLM model is the LLM model after parameter adjustment. Those skilled in the art know that the LLM model can be selected according to actual needs, which falls within the protection scope of the present invention and will not be repeated here, for example: ChatGLM2-6B model and other LLM models.

[0048] Furthermore, those skilled in the art know that any method for adjusting the parameters of the LLM model in the prior art falls within the protection scope of the present invention and will not be described in detail herein, such as the parameter adjustment methods such as P-Tuning v2.

[0049] S800, according to C and D, obtaining a target identification set corresponding to the initial electronic medical record text, wherein the target identification set includes a plurality of target identifications.

[0050] In the above, based on the preset text list and sample rule list corresponding to the initial electronic medical record text, wherein the sample rules in the sample rule list are rules obtained based on the candidate knowledge graph, the first target identification set and the second target identification set corresponding to the preset text list are obtained, the first target identification in the first target identification set is an identification obtained based on the first candidate rule and the rule engine, the second target identification in the second target identification set is an identification obtained based on the second candidate rule and the candidate LLM model, the corresponding rules are obtained based on the knowledge graph, the rules are combined with the LLM model, and the efficiency of obtaining the target identification is improved.

[0051] Specifically, in S800, the target identifier is obtained through the following steps:

[0052] S801, when C r is an empty set and D r If it is not an empty set, get D r The second target identifier in is the target identifier corresponding to the initial electronic medical record text.

[0053] S803, when C r is not an empty set and D r When it is an empty set, get C r The first target identifier in is the target identifier corresponding to the initial electronic medical record text.

[0054] S805, when C is not an empty set and D is not an empty set, for C r and D r Processing is performed to obtain a target identifier, wherein S805 also includes the following steps:

[0055] S8051, get C r The corresponding first initial weight list C 0 r ={C 0 r1 , ..., C 0 rx , ..., C 0 rp（r）}, C 0 rx C r The first initial weight corresponding to the xth first target identifier in C, x=1...p(r), p(r) is r The number of first target identifiers in C 0 rx =0.5 / p(r).

[0056] S8053, Get D r The corresponding second initial weight list D 0 r ={D0 r1 , ..., D 0 ry , ..., D 0 rq（r）}, D 0 ry D r The second initial weight corresponding to the yth second target identifier in D is y=1…q(r), where q(r) is r The number of second target identifiers in D 0 rx =0.5 / q(r).

[0057] S8055, according to C 0 r , get C r The corresponding first target weight list U 0 r ={U 0 r1 , ..., U 0 rx , ..., U 0 rp（r）}, where the following conditions are met:

[0058] U 0 rx =C 0 rx ×(g 1 rx / g rx ), g 1 rx C-based 0 rx The number of entities associated with the disease matched by the corresponding first candidate rule, g rx For C-based 0 rx The number of entities associated with the disease in the corresponding first candidate rule.

[0059] S8057, when Z 0 >Z 1 When C r The first target identifier in is the target identifier corresponding to the initial electronic medical record text, where Z 0 =∑ p（r） x=1 U 0 rx , Z 1 =∑ q（r） y=1 D 0 ry .

[0060] S8059, when Z 0 ≤Z 1 , get D r The second target identifier in is the target identifier corresponding to the initial electronic medical record text.

[0061] In the above, a first initial weight list corresponding to the first target identification list and a second initial weight list corresponding to the second target identification list are obtained, and the first initial weight is adjusted based on the output result corresponding to the first candidate rule corresponding to the first target identification list, and the adjusted first initial weight list is compared with the second initial weight list to obtain the target identification corresponding to the initial electronic medical record text, and the target identification is obtained in different ways based on different output results, and the weights corresponding to different types of instruction information are adjusted according to the output results, so that the accuracy of the obtained target identification is higher.

[0062] The present embodiment provides a system for obtaining a target identifier corresponding to a medical record text, the system comprising a preset database, a processor, and a memory storing a computer program, wherein the preset database comprises a sample text list and a sample rule set corresponding to the sample text list, and when the computer program is executed by the processor, the following steps are implemented: according to the initial electronic medical record text, obtaining a preset text list corresponding to the initial electronic medical record text, according to the preset text list, obtaining a first candidate rule set corresponding to the preset text list, according to the first candidate rule set, obtaining a first instruction information set corresponding to the preset text list, inputting the first instruction information set into a rule engine, obtaining a first target identifier set, and according to the preset text list, obtaining a first candidate rule set corresponding to the preset text list. This list obtains the second candidate rule set corresponding to the preset text list, obtains the second instruction information set corresponding to the preset text list according to the second candidate rule set, inputs the second instruction information set into the candidate LLM model, obtains the second target identification set, obtains the target identification set corresponding to the initial electronic medical record text according to the first target identification set and the second target identification set. The present invention obtains the corresponding rules based on the knowledge graph, combines the rules with the LLM model, improves the efficiency of obtaining the target identification, obtains the target identification in different ways based on the different output results, and adjusts the weights corresponding to different types of instruction information according to the output results, so that the accuracy of the obtained target identification is higher.

[0063] Although some specific embodiments of the present invention have been described in detail by way of example, it will be appreciated by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It will also be appreciated by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A system for obtaining target identifiers corresponding to medical record texts, characterized in that: The system comprises: a sample text list A={A1, ..., A i , ..., A n }、The sample rule set B corresponding to A = {B1, ..., B i , ..., B n }, a processor and a memory storing a computer program, wherein B i ={B 1 i , B 2 i }, A i is the i-th sample text, B 1 i A i The corresponding first sample rule list, B 2 i A i The second sample rule list of i=1...n, n is the number of sample texts, the first sample rule list includes a plurality of first sample rules, wherein the first sample rule is a basis for selecting an answer corresponding to the sample text obtained based on the candidate knowledge graph, the candidate knowledge graph includes a plurality of nodes and a plurality of edges, wherein the nodes represent entities associated with the disease, and the edges represent the relationship between nodes, the second sample rule list includes a plurality of second sample rules, wherein the second sample rule is a basis for determining an answer corresponding to the sample text obtained based on knowledge related to the disease, and when the computer program is executed by a processor, the following steps are implemented: S100, according to the initial electronic medical record text, obtain a preset text list E corresponding to the initial electronic medical record text = {E1, ..., E r ,……,E s }, E r is the rth preset text, r=1...s, s is the number of preset texts, and the preset text is a text representing disease-related problems; S200, according to E, obtain the first candidate rule set G corresponding to E = {G1, ..., G r , ..., G s }, G r For E r The corresponding first candidate rule list, where E r With A i When consistent, obtain G r =B 1 i ; S300, according to G, obtain the first instruction information set M corresponding to E={M1, ..., M r , ..., M s }, where M r For E r A corresponding first instruction information list, the first instruction information list includes a plurality of first instruction information, the first instruction information includes a first candidate text corresponding to a first candidate rule and a discrimination condition, the first candidate text is an attribute feature obtained from an initial electronic medical record text based on the first candidate rule and a text corresponding to the attribute feature; S400, input M into the rule engine, and obtain a first target identifier set C={C1, ..., C r , ..., C s }, where C r M r a corresponding first target identifier list, wherein the first target identifier list includes a plurality of first target identifiers, and the first target identifiers are corresponding symbols obtained by the rule engine through screening and matching based on the first instruction information; S500, according to E, obtain a second candidate rule set H corresponding to E = {H1, ..., H r , ..., H s }, H r For E r The corresponding second candidate rule list, where E r With A i When they are consistent, obtain H r =B 2 i ; S600, according to H, obtain the second instruction information set N={N1, ..., N r , ..., N s }, where N r For E r A corresponding second instruction information list, the second instruction information list includes a plurality of second instruction information, the second instruction information includes a second candidate text corresponding to the second candidate rule and a discrimination condition, the second candidate text is an attribute feature obtained from the initial electronic medical record text based on the second candidate rule and a text corresponding to the attribute feature; S700, input N into the candidate LLM model to obtain a second target identifier set D={D1, ..., D r , ..., D s }, where D r N r a corresponding second target identifier list, the second target identifier list including a plurality of second target identifiers, the second target identifier being a symbol of an output of a candidate LLM model obtained based on the second instruction information in the second instruction information list; S800, according to C and D, obtain a target identification set corresponding to the initial electronic medical record text, wherein the target identification set includes a plurality of target identifications, and the target identification is obtained in S800 through the following steps: S801, when C r is an empty set and D r If it is not an empty set, get D r The second target identifier in is the target identifier corresponding to the initial electronic medical record text; S803, when C r is not an empty set and D r When it is an empty set, get C r The first target identifier in is the target identifier corresponding to the initial electronic medical record text; S805, when C is not an empty set and D is not an empty set, for C r and D r Processing is performed to obtain a target identifier, wherein S805 also includes the following steps: S8051, get C r The corresponding first initial weight list C 0 r ={C 0 r1 , ..., C 0 rx , ..., C 0 rp（r） }, C 0 rx C r The first initial weight corresponding to the xth first target identifier in C, x=1...p(r), p(r) is r The number of first target identifiers in C 0 rx = 0.5 / p(r); S8053, Get D r The corresponding second initial weight list D 0 r ={D 0 r1 , ..., D 0 ry , ..., D 0 rq（r） }, D 0 ry D r The second initial weight corresponding to the yth second target identifier in D is y=1…q(r), where q(r) is r The number of second target identifiers in D 0 rx = 0.5 / q(r); S8055, according to C 0 r , get C r The corresponding first target weight list U 0 r ={U 0 r1 , ..., U 0 rx , ..., U 0 rp（r） }, where the following conditions are met: U 0 rx =C 0 rx ×(g 1 rx / g rx ), g 1 rx C-based 0 rx The number of entities associated with the disease matched by the corresponding first candidate rule, g rx For C-based 0 rx the number of entities associated with the disease in the corresponding first candidate rule; S8057, when Z 0 >Z 1 When C r The first target identifier in is the target identifier corresponding to the initial electronic medical record text, where Z 0 =∑ p（r） x=1 U 0 rx , Z 1 =∑ q（r） y=1 D 0 ry ; S8059, when Z 0 ≤Z 1 , get D r The second target identifier in is the target identifier corresponding to the initial electronic medical record text.

2. The system for obtaining target identifiers corresponding to medical record texts according to claim 1, characterized in that: The sample text is text representing disease-related issues obtained based on a sample electronic medical record information set.

3. The system for obtaining target identifiers corresponding to medical record texts according to claim 2, characterized in that: The sample electronic medical record information set includes a plurality of sample electronic medical record information, and the sample electronic medical record information is disease-related information obtained from a database.

4. The system for obtaining target identifiers corresponding to medical record texts according to claim 1, characterized in that: The initial electronic medical record text includes a plurality of attribute features and a text corresponding to each attribute feature.

5. The system for obtaining target identifiers corresponding to medical record texts according to claim 1, characterized in that: The rule engine is a component that performs rule matching based on different screening dimensions.

6. The system for obtaining target identifiers corresponding to medical record texts according to claim 1, characterized in that: The candidate LLM model is the LLM model after parameter adjustment.

Images