Medical record document quality inspection method, training method of medical record quality inspection model and related device

Abstract

The invention discloses a medical record document quality inspection method, a medical record quality inspection model training method and a related device. The method comprises the following steps: acquiring a target medical record document; inputting the target medical record document into a medical record quality inspection model to obtain a quality inspection result of the target medical record document, the quality inspection result comprising a judgment result of whether the target medical record document has an error and a judgment basis; wherein the medical record quality inspection model is obtained by fine tuning based on the pseudo-annotation data and the sample medical record document, and the pseudo-annotation data is obtained based on the sample medical record document and the language model. Through the above mode, the method can use the model to obtain the pseudo-annotation data training, and then uses the pseudo-annotation data to train the medical record quality inspection model, thereby saving the manual annotation cost and improving the model training efficiency while guaranteeing the quality inspection effect of the medical record quality inspection model.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular to a medical record document quality inspection method, a medical record quality inspection model training method, and related devices. Background Art

[0002] Medical records are essential tools for doctors when diagnosing and treating patients. These documents record the patient's medical history, disease progression, medical advice and treatment process, and are crucial to the patient's health management and medical decision-making.

[0003] When writing medical records, doctors often face complex medical terminology and grammatical requirements, making it easy for them to make mistakes, including spelling omissions and errors in diagnoses. To identify errors in medical records, doctors need to thoroughly review the patient's entire medical record and compare it with professional medical knowledge bases to ensure the accuracy and completeness of the diagnosis. This task is not only time-consuming and labor-intensive, but also prone to human error due to the limitations of doctors' knowledge and experience, reducing the efficiency and quality of medical work.

[0004] In recent years, with the rise of language models, they have also been applied in the medical field. However, in order for the model to perform automated quality inspection tasks and achieve good quality inspection results, it is often necessary to build a large amount of training data to train the model. Labeling this data places very high demands on the labelers, who need to have in-depth medical knowledge and a deep understanding of the writing standards of medical documents. In addition, the labeling process also takes a lot of time and energy. Summary of the Invention

[0005] The main technical problem solved by this application is to provide a medical record document quality inspection method, a medical record quality inspection model training method and related devices, which can use the model to obtain pseudo-annotated data training, and then use the pseudo-annotated data to train the medical record quality inspection model. While ensuring the quality inspection effect of the medical record quality inspection model, it saves manual labeling costs and improves model training efficiency.

[0006] In order to solve the above-mentioned technical problems, the first aspect of the present application provides a medical record document quality inspection method, which includes: obtaining a target medical record document; inputting the target medical record document into a medical record quality inspection model to obtain a quality inspection result of the target medical record document, the quality inspection result including a judgment result of whether there is an error in the target medical record document and the basis for the judgment; wherein, the medical record quality inspection model is fine-tuned based on pseudo-annotated data and sample medical record documents, and the pseudo-annotated data is obtained based on sample medical record documents and a language model.

[0007] To solve the above-mentioned technical problems, the second aspect of this application provides a training method for a medical record quality inspection model, which includes: obtaining pseudo-annotated data of sample medical records obtained by a language model at least based on sample medical records; inputting at least the sample medical records into the medical record quality inspection model to obtain quality inspection results for the sample medical records; and adjusting the parameters of the medical record quality inspection model based on the pseudo-annotated data and the quality inspection results.

[0008] In order to solve the above-mentioned technical problems, the third aspect of the present application provides a medical record document quality inspection device, which includes: a first acquisition module and a first quality inspection module; wherein the first acquisition module is used to obtain the target medical record document; the first quality inspection module is used to input the target medical record document into a medical record quality inspection model to obtain the quality inspection result of the target medical record document, and the quality inspection result includes a judgment result of whether there is an error in the target medical record document and the basis for the judgment; wherein the medical record quality inspection model is obtained by fine-tuning based on pseudo-annotated data and sample medical record documents, and the pseudo-annotated data is obtained based on sample medical record documents and a language model.

[0009] In order to solve the above-mentioned technical problems, the fourth aspect of the present application provides a training device for a medical record quality inspection model, which includes: a second acquisition module, a second quality inspection module and a parameter adjustment module; wherein the second acquisition module is used to obtain pseudo-annotated data of sample medical records obtained by the language model at least based on the sample medical records; the second quality inspection module is used to input at least the sample medical records into the medical record quality inspection model to obtain the quality inspection results of the sample medical records; the parameter adjustment module is used to adjust the parameters of the medical record quality inspection model based on the pseudo-annotated data and the quality inspection results.

[0010] To solve the above technical problems, the fifth aspect of the present application provides an electronic device, which includes a processor and a memory, wherein the memory stores program instructions, and the processor is used to execute the program instructions to implement the method provided in the first or second aspect above.

[0011] In order to solve the above technical problems, the sixth aspect of the present application provides a computer-readable storage medium, which stores program instructions, and the program instructions are used to implement the method provided by the first aspect or the second aspect above.

[0012] The beneficial effects of the present application are as follows: Different from the prior art, after obtaining the target medical record document, the present application inputs the target medical record document into the medical record quality inspection model to obtain the quality inspection result of the target medical record document, and the quality inspection result includes the judgment result of whether there is an error in the target medical record document and the basis for the judgment; wherein, the medical record quality inspection model is fine-tuned based on pseudo-annotated data and sample medical record documents, and the pseudo-annotated data is obtained based on sample medical record documents and a language model. Pseudo-annotated data with higher accuracy can be obtained through the language model, which can avoid manual labeling of data and improve efficiency. Reusing the pseudo-annotated data to train the medical record quality inspection model can enable the medical record quality inspection model to have a higher medical record quality inspection capability and ensure the quality inspection effect of the medical record quality inspection model, thereby saving manual labeling costs and improving model training efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of an implementation method of a medical record document quality inspection method provided by this application;

[0014] Figure 2 This is a flow chart of another embodiment of the medical record quality inspection method provided by this application;

[0015] Figure 3 This is a flowchart of an implementation method for training a medical record quality inspection model provided by this application;

[0016] Figure 4 This is a schematic diagram of the framework structure of an embodiment of the medical record document quality inspection device provided by this application;

[0017] Figure 5 This is a schematic diagram of the framework structure of an embodiment of a training device for a medical record quality inspection model provided by this application;

[0018] Figure 6 This is a schematic diagram of the framework structure of an embodiment of an electronic device provided by the present application;

[0019] Figure 7 It is a schematic diagram of the framework structure of an embodiment of the computer-readable storage medium provided in this application. DETAILED DESCRIPTION

[0020] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0021] It should be noted that the terms "first," "second," and so on, used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include at least one of these features.

[0022] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0023] In recent years, large-scale generative language models (such as GPT-3 and GPT-4) have achieved breakthroughs in natural language processing. Based on deep learning techniques and undergoing large-scale pre-training, these models possess outstanding natural language understanding and generation capabilities. They automatically understand the semantics and context of text and, incorporating relevant expertise, generate high-quality text content. This technology has been widely applied in fields such as automatic translation, chatbots, text summarization, and sentiment analysis.

[0024] However, it is still very difficult to make the generative language model achieve better results in the task of automated quality inspection of medical records. The inventors found in long-term research that due to the particularity of the task of automated quality inspection of medical records (such as professional medical knowledge, differences in the format or style of medical records written by different doctors, etc.), it is impossible to use super-large generative models for such private scenario tasks, which will limit the capabilities of the model itself. In order to optimize the effect of automated quality inspection tasks for smaller-sized models, it is often necessary to build a large amount of training data to fine-tune the parameters of the model. Labeling this data requires very high requirements for the labelers, who need to have in-depth medical knowledge and a good understanding of the writing standards of medical documents. In addition, the labeling process also takes a lot of time and energy.

[0025] Traditional medical record quality inspection model optimization approaches primarily employ the following approaches: rule engine quality inspection optimization, instruction engineering optimization, or supervised training data optimization. Rule engine quality inspection optimization relies on predefined rules and templates to inspect medical records. These rules can include spelling checks, term matching, and formatting requirements. Optimizing the effectiveness of this rule engine approach requires professional doctors to manually summarize the relevant quality inspection rules, which consumes significant manpower and resources. Furthermore, the rules lack generalizability and are not universal across heterogeneous scenarios. They also struggle to handle complex grammatical structures and contextual semantics, leading to missed or false positives. Instruction engineering optimization involves designing and adjusting various quality control instructions to achieve better quality control results for large models. However, this approach is limited by the capabilities of the large model itself. When the model is small and lacks sufficient capabilities, simply adjusting instructions alone cannot achieve optimal results for medical document quality control tasks, resulting in significant limitations.

[0026] To solve the above problems, this application obtains pseudo-annotated data based on sample medical records and a language model, fine-tunes the pseudo-annotated data and sample medical records to obtain a medical record quality inspection model, and finally uses the medical record quality inspection model to obtain the quality inspection results of the target medical record. The language model can obtain pseudo-annotated data with high accuracy, which can avoid manual labeling of data and improve efficiency. Using the pseudo-annotated data to train the medical record quality inspection model can enable the medical record quality inspection model to have a higher medical record quality inspection capability and ensure the quality inspection effect of the medical record quality inspection model, thereby saving the cost of manual labeling and improving the efficiency of model training.

[0027] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of a method for quality inspection of medical records provided by this application, which includes:

[0028] S11: Obtain target medical records.

[0029] The method of this embodiment is used to perform quality inspection on target medical records to determine whether they contain errors. In one embodiment, the target medical records include one or more of medical records, diagnostic reports, and surgical records. The target medical records can be directly input by the user or selected from a database storing target medical records. Alternatively, the electronic device can proactively retrieve the target medical records. For example, the electronic device can retrieve a preset number of target medical records at preset intervals for inspection.

[0030] It is understood that the target medical record document can be generated in real time and stored in the database. When quality inspection is required, the target medical record document to be inspected can be directly selected from the database; or the target medical record document generated in real time can be directly inspected.

[0031] S12: Input the target medical record document into the medical record quality inspection model to obtain the quality inspection result of the target medical record document.

[0032] In one embodiment, a medical record quality inspection model can be pre-trained, and then the target medical record document is input into the trained medical record quality inspection model. The medical record quality inspection model can directly output the quality inspection result of the target medical record document.

[0033] Among them, the medical record quality inspection model can be fine-tuned based on pseudo-annotated data and sample medical record documents, and the pseudo-annotated data is obtained based on the sample medical record documents and the language model. Specifically, the language model can be any existing model with excellent natural language understanding capabilities, and the language model can more accurately determine whether there are errors in the sample medical record documents. For example, the sample medical record documents include preoperative diagnosis records and operation records. It can compare whether the contents of the preoperative diagnosis records and the operation records are consistent. If they are inconsistent, it is considered that there are errors in the sample medical record documents. When training the medical record quality inspection model, the sample medical record documents can be input into the medical record quality inspection model. The medical record quality inspection model can output the sample quality inspection results of the sample medical record documents, and the medical record quality inspection model is fine-tuned according to the sample quality inspection results and the pseudo-annotated data. The pseudo-annotated data is obtained according to the output results of the language model after the sample medical record documents are input into the language model.

[0034] In the above method, after obtaining the target medical record document, the target medical record document is input into the medical record quality inspection model to obtain the quality inspection result of the target medical record document. The quality inspection result includes the judgment result of whether the target medical record document contains errors and the basis for the judgment; wherein, the medical record quality inspection model is fine-tuned based on pseudo-annotated data and sample medical records, and the pseudo-annotated data is obtained based on sample medical records and a language model. The language model can obtain pseudo-annotated data with high accuracy, which can avoid manual labeling of data and improve efficiency. Then, using the pseudo-annotated data to train the medical record quality inspection model can make the medical record quality inspection model have a higher medical record quality inspection capability and can ensure the quality inspection effect of the medical record quality inspection model, thereby saving the cost of manual labeling and improving the efficiency of model training.

[0035] In one embodiment, a target medical record document is only input into the medical record quality inspection model once to obtain its corresponding quality inspection result. In other embodiments, to further improve the accuracy of the quality inspection results, the same target medical record document can be input into the medical record quality inspection model multiple times to obtain multiple quality inspection results corresponding to the target medical record document; and the final quality inspection result corresponding to the target medical record document is obtained based on the multiple quality inspection results.

[0036] In one embodiment, each quality inspection result output by the medical record quality inspection model corresponds to a score or confidence level, and the quality inspection result with the highest score or confidence level can be selected as the final quality inspection result.

[0037] In another embodiment, the quality inspection results can be clustered by voting to obtain a final quality inspection result. Specifically, the quality inspection result includes a judgment result and a basis for judging whether the target medical record document contains errors. Voting is performed on multiple judgment results, and the judgment results with the most votes are selected as candidate judgment results. The judgment basis corresponding to each candidate judgment result is clustered to obtain a cluster result. From the judgment basis corresponding to each candidate judgment result, the judgment basis closest to the cluster center is selected as the final judgment basis. The cluster center is determined based on the clustering result. The final judgment basis and the candidate judgment results corresponding to the final judgment basis are used as the final quality inspection result.

[0038] The judgment result can be "yes" or "no"; it can also be "qualified" or "unqualified." Taking the judgment result of "yes" or "no" as an example, the number of "yes" and "no" judgment results are counted respectively, and the judgment result with the most votes is obtained. For example, the judgment result of "yes" has the most votes, and all the judgment results with "yes" are used as candidate judgment results. The judgment basis corresponding to the judgment result of "yes" is clustered to obtain a cluster result. Based on the cluster result, the cluster center can be determined. Then, the distance from the judgment basis corresponding to the judgment result of "yes" to the cluster center is calculated. The judgment basis with the closest distance is used as the final judgment basis. The final judgment basis and the candidate judgment results corresponding to the final judgment basis are used as the final quality inspection result. In this way, the quality inspection result with the highest accuracy can be obtained.

[0039] See also Figure 2 , Figure 2 : is a flow chart of another embodiment of the medical record document quality inspection method provided by this application, the method comprising:

[0040] S21: Obtain target medical record documents.

[0041] S22: Match the target medical record document with professional medical knowledge to obtain target medical knowledge that matches the target medical record document.

[0042] In one embodiment, medical records often contain professional medical knowledge. In order to determine whether the medical knowledge in the medical record is correct, the target medical record can be matched with the professional medical knowledge to obtain target medical knowledge that matches the target medical record.

[0043] Among them, patent medical knowledge can be obtained from a pre-built medical knowledge base. When constructing the medical knowledge base, professional medical books and documents are collected, and then natural language processing technology is used to perform text analysis and information extraction on the collected professional medical books and documents. Through technologies such as entity recognition and relationship extraction, the medical knowledge in professional medical books and documents is converted into a structured medical knowledge base. Finally, medical experts carefully review the generated medical knowledge base to check whether the information in the knowledge base is accurate and complete, and modify and supplement any possible errors or omissions. Knowledge base maintenance is an ongoing process to ensure that the knowledge base is constantly updated as the medical field develops. Combined with natural language processing technology, the knowledge base can be quickly and automatically constructed and updated, and the participation of medical experts ensures the high quality and accuracy of the knowledge base.

[0044] S23: Input the target medical record document and target medical knowledge into the medical record quality inspection model to obtain the quality inspection result of the target medical record document.

[0045] At this time, the medical record quality inspection model can compare the target medical knowledge with the medical knowledge in the target medical record document to obtain a quality inspection result. In this embodiment, the introduction of professional medical knowledge can more accurately identify whether the target medical record document contains errors.

[0046] In this embodiment, the medical record quality inspection model is fine-tuned based on pseudo-annotated data, sample medical records, and sample medical knowledge matched to the sample medical records. The pseudo-annotated data is derived from the sample medical records, sample medical knowledge, and the language model. Specifically, the sample medical records and sample medical knowledge are input into the medical record quality inspection model, which then outputs a sample quality inspection result for the sample medical records. The model is then fine-tuned based on the sample quality inspection result and the pseudo-annotated data. The pseudo-annotated data is derived from the output of the language model after the sample medical records and sample medical knowledge are input into the language model.

[0047] In other embodiments, professional medical knowledge can be stored in a module in a medical record quality inspection model, and the target medical record document can be input into the medical record quality inspection model, and the medical record quality inspection model can match the target medical knowledge that matches the target medical record document, and then the quality inspection results can be obtained based on the target medical record document and the target medical knowledge.

[0048] In one embodiment, pseudo-annotated data can be obtained by first collecting a large number of sample medical records, including documents from various medical fields, such as medical records, diagnostic reports, and surgical records. Quality control instructions are then constructed, transforming the task of quality control of medical records into a judgment task, such as determining whether a medical record contains a missed diagnosis or medication error. The sample medical records are then input into a language model, which then judges the sample medical records according to the quality control instructions, generating pseudo-annotated data. The pseudo-annotated data includes the sample judgment results of whether the sample medical records are incorrect, as well as the basis for the sample judgment.

[0049] In one embodiment, to improve the accuracy of pre-annotated data, a language model is used to repeatedly judge the same sample medical record document according to quality control instructions, resulting in multiple pseudo-annotated data. Based on the multiple pseudo-annotated data, the final pseudo-annotated data corresponding to the sample medical record document is determined. Specifically, voting clustering can be performed on the multiple pseudo-annotated data. The voting clustering operation is consistent with the aforementioned operation, namely, voting is performed on the multiple sample judgment results, and the sample judgment results of the category with the most votes are selected as candidate judgment results. The sample judgment basis corresponding to each candidate judgment result is clustered to obtain a cluster result. From the sample judgment basis corresponding to each candidate judgment result, the sample judgment basis closest to the cluster center is selected as the final judgment basis. The final judgment basis and the candidate judgment results corresponding to the final judgment basis are used as the final pseudo-annotated data.

[0050] In another embodiment, in order to further improve the accuracy of pseudo-annotated data, the sample medical records may be preprocessed by at least one of the following: cleaning the text in the sample medical records, segmenting the text in the sample medical records, and removing noise and / or irrelevant information in the sample medical records.

[0051] After obtaining the pseudo-annotated data of the sample medical records, the pseudo-annotated data and the sample medical records can be used to train the medical record quality inspection model. Specifically, the sample medical records are input into the medical record quality inspection model to obtain the quality inspection results of the sample medical records; based on the pseudo-annotated data and the quality inspection results, the parameters of the medical record quality inspection model are adjusted. The specific training task is as follows: Encode the sample medical records into a text token sequence X = {x1…x n} Input into the medical record quality inspection model, so that the medical record quality inspection model can autoregressively predict the medical record document quality control results x i+1 The training objective can be to maximize the following likelihood function:

[0052]

[0053] In other implementations, sample medical records and the medical knowledge matched to the sample medical records can be input into a medical record quality inspection model to obtain quality inspection results for the sample medical records. The parameters of the medical record quality inspection model can be adjusted based on the pseudo-annotated data and the quality inspection results. The medical knowledge matched to the sample medical records can be obtained from a medical knowledge base.

[0054] See also Figure 3 , Figure 3 This is a flow chart of an embodiment of a training method for a medical record quality inspection model provided by this application, which includes:

[0055] S31: Acquire pseudo-annotated data of a sample medical record document obtained by a language model at least based on the sample medical record document.

[0056] S32: Input at least the sample medical record document into the medical record quality inspection model to obtain a quality inspection result of the sample medical record document.

[0057] S33: Adjust the parameters of the medical record quality inspection model based on the pseudo-annotated data and quality inspection results.

[0058] Please refer to the above description for detailed training steps, which will not be repeated here.

[0059] The following is an example of the input and output of a medical record quality inspection model:

[0060] Input example: Based on the preoperative diagnosis and surgical records of a given patient, analyze the corresponding surgical sites to determine whether the given preoperative diagnosis is consistent with the surgical site in the surgical record. Specific requirements are as follows:

[0061] 1. For visceral surgery, the surgical site should be determined accurately to the specific organ. For example, the surgical site of hepatic vascular intervention is the liver.

[0062] 2. For surgeries related to the limbs and head, it is necessary to distinguish between left and right when determining the surgical site

[0063] The following is the given preoperative diagnosis and surgical records:

[0064] ##Preoperative Diagnosis##

[0065] $Replace preoperative diagnosis$

[0066] ##Operation Record##

[0067] Replace surgical records

[0068] ##

[0069] It is required to output in JSON format. Do not output anything other than JSON.

[0070] Output example: {"Quality Inspection Result":"No","Interpretability of Quality Control Conclusion (i.e., Judgment Basis)":"The preoperative diagnosis was left femoral head necrosis, but the operation steps mentioned in the surgical record were performed in the right lateral decubitus position, so the surgical site is inconsistent with the diagnosis site"}.

[0071] See also Figure 4 , Figure 4 It is a schematic diagram of the framework structure of an embodiment of the medical record document quality inspection device provided in this application.

[0072] The medical record document quality inspection device 40 includes a first acquisition module 41 and a first quality inspection module 42. The first acquisition module 41 is used to obtain the target medical record document; the first quality inspection module 42 is used to input the target medical record document into the medical record quality inspection model to obtain the quality inspection result of the target medical record document. The quality inspection result includes the judgment result of whether there is an error in the target medical record document and the basis for the judgment; wherein, the medical record quality inspection model is obtained by fine-tuning based on pseudo-annotated data and sample medical record documents, and the pseudo-annotated data is obtained based on sample medical record documents and a language model.

[0073] In one embodiment, the target medical record document is input into the medical record quality inspection model to obtain the quality inspection result of the target medical record document, including: inputting the target medical record document into the medical record quality inspection model multiple times to obtain multiple quality inspection results; and selecting the final quality inspection result from the multiple quality inspection results.

[0074] In one embodiment, a final quality inspection result is selected from multiple quality inspection results, including: voting on multiple judgment results, and taking the judgment results of the category with the most votes as candidate judgment results; clustering the judgment basis corresponding to each candidate judgment result to obtain a clustering result; selecting the judgment basis closest to the cluster center from the judgment basis corresponding to each candidate judgment result as the final judgment basis; wherein the cluster center is determined based on the clustering result; and taking the final judgment basis and the candidate judgment results corresponding to the final judgment basis as the final quality inspection result.

[0075] In one embodiment, the medical record document quality inspection device 40 also includes a matching module. After obtaining the target medical record document, it also includes: the matching module matches the target medical record document with professional medical knowledge to obtain the target medical knowledge that matches the target medical record document; the first quality inspection module 42 is also used to input the target medical record document and the target medical knowledge into the medical record quality inspection model to obtain the quality inspection result of the target medical record document; wherein, the medical record quality inspection model is fine-tuned based on pseudo-annotated data, sample medical record documents and sample medical knowledge matching the sample medical record documents, and the pseudo-annotated data is obtained based on the sample medical record documents, sample medical knowledge and the language model.

[0076] In one embodiment, the medical record document quality inspection device 40 further includes a knowledge base construction module, which is used to construct a medical knowledge base, wherein at least one of professional medical knowledge and sample medical knowledge comes from the medical knowledge base.

[0077] In one embodiment, the medical record document quality inspection device 40 also includes a preprocessing module, which is used to perform at least one of the following preprocessing on the sample medical record document: cleaning the text in the sample medical record document, segmenting the text in the sample medical record document, and removing noise and / or irrelevant information in the sample medical record document.

[0078] In one embodiment, the medical record document quality inspection device 40 also includes an instruction construction module and a labeling data determination module. Before the target medical record document is input into the medical record quality inspection model to obtain the quality inspection result of the target medical record document, the instruction construction module is used to construct quality control instructions; the labeling data determination module is used to use the language model to judge the sample medical record document according to the quality control instructions to obtain pseudo-labeling data.

[0079] In one embodiment, a language model is used to judge a sample medical record document according to quality control instructions to obtain pseudo-annotated data, including: using a language model to judge the same sample medical record document multiple times according to quality control instructions to obtain multiple pseudo-annotated data; based on the multiple pseudo-annotated data, determining the final pseudo-annotated data corresponding to the sample medical record document.

[0080] See also Figure 5 , Figure 5 It is a schematic diagram of the framework structure of an embodiment of a training device for a medical record quality inspection model provided in this application.

[0081] The training device 50 for the medical record quality inspection model includes a second acquisition module 51, a second quality inspection module 52 and a parameter adjustment module 53. The second acquisition module 51 is used to obtain pseudo-annotated data of sample medical records obtained by the language model at least based on the sample medical records; the second quality inspection module 52 is used to input at least the sample medical records into the medical record quality inspection model to obtain the quality inspection results of the sample medical records; the parameter adjustment module 53 is used to adjust the parameters of the medical record quality inspection model based on the pseudo-annotated data and the quality inspection results.

[0082] See also Figure 6 , Figure 6 This is a schematic diagram of the framework structure of an embodiment of the electronic device provided in this application.

[0083] The electronic device 60 includes a memory 61 and a processor 62 coupled to each other. The memory 61 stores program instructions, and the processor 62 is configured to execute the program instructions stored in the memory 61 to implement the steps of any of the above-described method implementations. In a specific implementation scenario, the electronic device 60 may include, but is not limited to, a microcomputer and a server. Furthermore, the electronic device 60 may also include a mobile device such as a laptop computer or a tablet computer, which is not limited herein.

[0084] Specifically, the processor 62 is used to control itself and the memory 61 to implement the steps of any of the above-mentioned method implementation methods. The processor 62 can also be called a CPU (Central Processing Unit). The processor 62 may be an integrated circuit chip with signal processing capabilities. The processor 62 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 62 can be implemented by an integrated circuit chip.

[0085] See also Figure 7 , Figure 7 It is a schematic diagram of the framework structure of an embodiment of the computer-readable storage medium provided in this application.

[0086] The computer-readable storage medium 70 stores program instructions 71 , which, when executed by a processor, are used to implement the steps of any of the above-mentioned method implementations.

[0087] The computer-readable storage medium 70 can specifically be a medium that can store computer programs, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or it can also be a server that stores the computer program. The server can send the stored computer program to other devices for execution, or it can also run the stored computer program itself.

[0088] The above description of the various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced with each other and will not be repeated herein for the sake of brevity.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation methods described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0090] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0091] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0092] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0093] If the technical solution of this application involves personal information, the product that applies the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing personal information. If the technical solution of this application involves sensitive personal information, the product that applies the technical solution of this application has obtained the individual's separate consent before processing sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, a clear and prominent sign is set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that they agree to the collection of their personal information; or on the personal information processing device, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload their personal information; among which, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

[0094] The above description is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A medical record document quality inspection method, characterized in that: include: Obtain target medical records; The target medical record document is input into the medical record quality inspection model to obtain the quality inspection result of the target medical record document, and the quality inspection result includes the judgment result of whether the target medical record document has errors and the basis for the judgment; wherein, the medical record quality inspection model is fine-tuned based on pseudo-annotated data and sample medical record documents, and the pseudo-annotated data is obtained based on the sample medical record documents and the language model.

2. The method according to claim 1, characterized in that Inputting the target medical record document into the medical record quality inspection model to obtain the quality inspection result of the target medical record document includes: Inputting the target medical record document into the medical record quality inspection model multiple times to obtain multiple quality inspection results; A final quality inspection result is selected from the multiple quality inspection results.

3. The method according to claim 2, characterized in that The selecting a final quality inspection result from the multiple quality inspection results includes: Voting on the multiple judgment results, and taking the judgment results with the most votes as candidate judgment results; Clustering the judgment basis corresponding to each candidate judgment result to obtain a clustering result; Selecting the judgment basis closest to the cluster center from the judgment basis corresponding to each candidate judgment result as the final judgment basis; wherein the cluster center is determined based on the clustering result; The final judgment basis and the candidate judgment result corresponding to the final judgment basis are used as the final quality inspection result.

4. The method according to claim 1, wherein After obtaining the target medical record document, the method further includes: Matching the target medical record document with professional medical knowledge to obtain target medical knowledge that matches the target medical record document; Inputting the target medical record document into the medical record quality inspection model to obtain the quality inspection result of the target medical record document includes: The target medical record document and the target medical knowledge are input into the medical record quality inspection model to obtain the quality inspection result of the target medical record document; wherein, the medical record quality inspection model is fine-tuned based on pseudo-annotated data, sample medical record documents and sample medical knowledge matching the sample medical record documents, and the pseudo-annotated data is obtained based on the sample medical record documents, the sample medical knowledge and the language model.

5. The method according to claim 4, characterized in that The method further comprises: Building a medical knowledge base, wherein at least one of the professional medical knowledge and the sample medical knowledge comes from the medical knowledge base; And / or, the sample medical record document is preprocessed by at least one of the following: cleaning the text in the sample medical record document, segmenting the text in the sample medical record document, and removing noise and / or irrelevant information in the sample medical record document.

6. The method according to claim 1, characterized in that Before inputting the target medical record document into the medical record quality inspection model to obtain the quality inspection result of the target medical record document, the method further includes: Construct quality control instructions; The language model is used to judge the sample medical record document according to the quality control instruction to obtain the pseudo-annotated data.

7. The method according to claim 6, characterized in that The method of using the language model to judge the sample medical record document according to the quality control instruction to obtain the pseudo-annotated data includes: Using the language model to judge the same sample medical record document multiple times according to the quality control instruction to obtain a plurality of pseudo-annotated data; Based on the plurality of pseudo-annotated data, final pseudo-annotated data corresponding to the sample medical record document is determined.

8. A training method for a medical record quality inspection model, characterized in that: include: Acquire pseudo-annotated data of the sample medical record document obtained by the language model at least based on the sample medical record document; At least inputting the sample medical record document into the medical record quality inspection model to obtain a quality inspection result of the sample medical record document; Based on the pseudo-annotated data and the quality inspection results, the parameters of the medical record quality inspection model are adjusted.

9. A medical record document quality inspection device, characterized in that: include: The first acquisition module is used to acquire the target medical record document; The first quality inspection module is used to input the target medical record document into the medical record quality inspection model to obtain the quality inspection result of the target medical record document, and the quality inspection result includes the judgment result of whether there is an error in the target medical record document and the basis for the judgment; wherein, the medical record quality inspection model is fine-tuned based on pseudo-annotated data and sample medical record documents, and the pseudo-annotated data is obtained based on the sample medical record documents and the language model.

10. A training device for a medical record quality inspection model, characterized in that: include: A second acquisition module is configured to acquire pseudo-annotated data of the sample medical record document obtained by the language model at least based on the sample medical record document; a second quality inspection module, configured to input at least the sample medical record document into the medical record quality inspection model to obtain a quality inspection result of the sample medical record document; A parameter adjustment module is used to adjust the parameters of the medical record quality inspection model based on the pseudo-annotated data and the quality inspection results.

11. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein program instructions are stored in the memory, and the processor is configured to execute the program instructions to implement the method according to any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that The storage medium stores program instructions, and the program instructions are used to implement the method according to any one of claims 1 to 8.

Images