CN117373592A - Medical record diagnosis write omission detection method, device, equipment and readable storage medium - Google Patents

Abstract

The application discloses a medical record diagnosis write-missing detection method, device and equipment and a readable storage medium. In the scheme, after the medical record text to be detected is determined, candidate write omission diagnosis corresponding to the medical record text is determined; then determining a target candidate missed writing diagnosis from the candidate missed writing diagnoses, wherein the target candidate missed writing diagnosis is a candidate missed writing diagnosis which is consistent with written diagnosis semantics in the medical record text, and is determined to be a candidate missed writing diagnosis of symptoms of the written diagnosis; and finally, determining other candidate missed writing diagnoses except the target candidate missed writing diagnosis in the candidate missed writing diagnoses as missed writing diagnoses.

Description

Medical record diagnosis omission detection method, device, equipment and readable storage medium

Technical Field

The present application relates to the technical field of natural language processing, and more specifically, to a method, device, equipment and readable storage medium for detecting omissions in medical record diagnosis.

Background Art

Filling out medical record diagnosis is a crucial link in the medical field. A complete and accurate diagnosis helps reduce the risk of medical disputes and is of great significance to the subsequent analysis of diagnosis and treatment behaviors, optimizing resource allocation, and improving the quality of medical services. Therefore, it is particularly necessary to detect missed diagnoses in medical records.

At present, medical record diagnosis omission detection generally adopts manual detection, that is, doctors or hospital medical record quality inspectors check medical records and mark diagnoses that may be omitted in medical records. However, the actual writing of medical records is very complicated. Manual detection of omissions in medical record diagnoses is a huge workload, with low detection efficiency, and the professional quality of the inspectors has a great impact on the accuracy of the detection results. In order to solve the problems of low detection efficiency and accuracy caused by manual detection of medical record diagnosis omissions, there are some medical record diagnosis omission detection technologies implemented by deep learning models, rule bases or knowledge bases in the prior art. However, these technologies only perform diagnosis omission detection based on the semantic similarity of diagnoses, and in fact, two diagnoses with different semantics may be the same diagnosis, which will lead to low accuracy of medical record diagnosis omission detection results.

Therefore, how to provide a method for detecting omissions in medical record diagnosis to improve the accuracy of the detection results of omissions in medical record diagnosis has become a technical problem that needs to be urgently solved by technical personnel in this field.

Summary of the invention

In view of the above problems, this application proposes a method, device, equipment and readable storage medium for detecting missed medical record diagnosis. The specific solution is as follows:

A method for detecting missed medical record diagnoses, the method comprising:

Determine the medical records to be tested;

Determining a candidate omitted diagnosis corresponding to the medical record text;

Determine a target candidate omitted diagnosis from the candidate omitted diagnoses, the target candidate omitted diagnosis being a candidate omitted diagnosis that is semantically consistent with a written diagnosis in the medical record text, and a candidate omitted diagnosis that is determined to be a symptom of the written diagnosis;

The other candidate omission diagnoses among the candidate omission diagnoses except the target candidate omission diagnosis are determined as omission diagnoses.

Optionally, determining a candidate omitted diagnosis corresponding to the medical record text includes:

Generating prompt information corresponding to the medical record text, wherein the prompt information is used to indicate the candidate omitted diagnosis corresponding to the medical record text;

Calling a generative model, and inputting prompt information corresponding to the medical record text into the generative model;

Based on the response of the generative model, a candidate omitted diagnosis corresponding to the medical record text is determined.

Optionally, the generating of prompt information corresponding to the medical record text includes:

Segmenting the medical record text to obtain multiple medical record text segments;

For each medical record text fragment, prompt information corresponding to the medical record text fragment is generated, and the prompt information is used to indicate the candidate omitted diagnosis corresponding to the medical record text fragment; the combination of prompt information corresponding to each medical record text fragment is the prompt information corresponding to the medical record text.

Optionally, the generative model is obtained by fine-tuning an existing generative model in the medical field; the method of fine-tuning the existing generative model in the medical field includes:

Obtaining a training medical record text, a diagnosis corresponding to the training medical record text, and a label of each diagnosis in the training medical record text, wherein the label of each diagnosis in the training medical record text includes a first label, a second label, and a third label, wherein the first label is used to indicate whether the diagnosis is confirmed, the second label is used to indicate the diagnosis type of the diagnosis, and the third label is used to indicate a recall support segment of the diagnosis;

Based on the training medical record text, the diagnosis corresponding to the training medical record text, and the label of each diagnosis in the training medical record text, a plurality of training corpora are generated, each of the training corpora includes an input text and an expected output text;

The input text is input into an existing generative model in the medical field, and the existing generative model in the medical field is fine-tuned with the goal of making the output of the existing generative model in the medical field approach the expected output text corresponding to the input text.

Optionally, determining a target candidate omission diagnosis from the candidate omission diagnoses includes:

Based on the candidate missing diagnosis and the written diagnosis, determining a plurality of diagnosis pairs, each diagnosis pair including one of the candidate missing diagnosis and one of the written diagnosis;

For each diagnosis pair, determining whether the semantics of two diagnoses in the diagnosis pair are consistent;

If the semantics of the two diagnoses in the diagnosis pair are consistent, determining the candidate omitted diagnosis in the diagnosis pair as the target candidate omitted diagnosis;

If the semantics of the two diagnoses in the diagnosis pair are inconsistent, comparing the diagnosis pair with each diagnosis-symptom pair in a preset diagnosis-symptom pair set;

If the diagnosis pair is consistent with a diagnosis symptom pair in the diagnosis symptom pair set, the candidate omission diagnosis is determined to be a target candidate omission diagnosis.

Optionally, the determining a plurality of diagnosis pairs based on the candidate missed diagnosis and the written diagnosis includes:

Normalizing the candidate omission diagnoses to obtain normalized candidate omission diagnoses;

Performing normalization processing on the written diagnosis to obtain a normalized written diagnosis;

The plurality of diagnosis pairs are determined based on the normalized candidate missed diagnoses and the normalized written diagnoses.

Optionally, the determining a plurality of diagnosis pairs based on the candidate missed diagnosis and the written diagnosis includes:

Combining the candidate missing diagnosis and the written diagnosis in pairs to form candidate diagnosis pairs; each of the candidate diagnosis pairs includes one candidate missing diagnosis and one written diagnosis;

Performing normalization processing on each of the candidate diagnosis pairs to obtain normalized candidate diagnosis pairs;

De-duplication processing is performed on each of the normalized candidate diagnosis pairs to determine the multiple diagnosis pairs.

Optionally, the preset diagnostic symptom pair set is determined in a manner including:

Obtaining a reference medical record text, a candidate omitted diagnosis corresponding to the reference medical record text, and a written diagnosis corresponding to the reference medical record text;

Combining the written diagnoses corresponding to the reference medical record text in pairs to obtain a first diagnosis pair set, combining the candidate omitted diagnoses corresponding to the reference medical record text and the written diagnoses corresponding to the reference medical record text in pairs, and combining the candidate omitted diagnoses corresponding to the reference medical record text in pairs to obtain a second diagnosis pair set;

Calculating the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set;

The diagnosis-symptom pair set is determined from the first diagnosis pair set and the second diagnosis pair set based on the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set.

A device for detecting omissions in medical record diagnosis, the device comprising:

A medical record text determination unit, used to determine the medical record text to be detected;

A candidate omitted diagnosis determining unit, used to determine a candidate omitted diagnosis corresponding to the medical record text;

a target candidate omitted diagnosis determining unit, configured to determine a target candidate omitted diagnosis from the candidate omitted diagnoses, wherein the target candidate omitted diagnosis is a candidate omitted diagnosis that is semantically consistent with a written diagnosis in the medical record text, and a candidate omitted diagnosis that is determined to be a symptom of the written diagnosis;

The omission diagnosis determining unit is configured to determine other candidate omission diagnoses among the candidate omission diagnoses except the target candidate omission diagnosis as omission diagnoses.

Optionally, the candidate omission diagnosis determination unit includes:

A prompt information generating subunit, used for generating prompt information corresponding to the medical record text, wherein the prompt information is used for indicating the candidate omitted diagnosis corresponding to the medical record text;

A generative model calling subunit, used to call the generative model and input the prompt information corresponding to the medical record text into the generative model;

The candidate omitted diagnosis determination subunit is used to determine the candidate omitted diagnosis corresponding to the medical record text based on the response of the generative model.

Optionally, the prompt information generating subunit is specifically used to:

Segmenting the medical record text to obtain multiple medical record text segments;

For each medical record text fragment, prompt information corresponding to the medical record text fragment is generated, and the prompt information is used to indicate the candidate omitted diagnosis corresponding to the medical record text fragment; the combination of prompt information corresponding to each medical record text fragment is the prompt information corresponding to the medical record text.

Optionally, the generative model is obtained by fine-tuning an existing generative model in the medical field; the device includes a fine-tuning unit, and the fine-tuning unit is specifically used to:

Obtaining a training medical record text, a diagnosis corresponding to the training medical record text, and a label of each diagnosis in the training medical record text, wherein the label of each diagnosis in the training medical record text includes a first label, a second label, and a third label, wherein the first label is used to indicate whether the diagnosis is confirmed, the second label is used to indicate the diagnosis type of the diagnosis, and the third label is used to indicate a recall support segment of the diagnosis;

Based on the training medical record text, the diagnosis corresponding to the training medical record text, and the label of each diagnosis in the training medical record text, a plurality of training corpora are generated, each of the training corpora includes an input text and an expected output text;

The input text is input into an existing generative model in the medical field, and the existing generative model in the medical field is fine-tuned with the goal of making the output of the existing generative model in the medical field approach the expected output text corresponding to the input text.

Optionally, the target candidate missing writing diagnosis determination unit includes:

A diagnosis pair determination unit, configured to determine a plurality of diagnosis pairs based on the candidate missing diagnosis and the written diagnosis, each diagnosis pair including one of the candidate missing diagnosis and one of the written diagnosis;

A judging unit, configured to judge, for each diagnosis pair, whether the semantics of two diagnoses in the diagnosis pair are consistent;

A first processing unit, configured to determine a candidate omitted diagnosis in the diagnosis pair as a target candidate omitted diagnosis if the semantics of the two diagnoses in the diagnosis pair are consistent;

A comparing unit, configured to compare the diagnosis pair with each diagnosis-symptom pair in a preset diagnosis-symptom pair set if the semantics of the two diagnoses in the diagnosis pair are inconsistent;

The second processing unit is configured to determine the candidate omission diagnosis as a target candidate omission diagnosis if the diagnosis pair is consistent with a diagnosis symptom pair in the diagnosis symptom pair set.

Optionally, the diagnosis pair determination unit is specifically configured to:

Normalizing the candidate omission diagnoses to obtain normalized candidate omission diagnoses;

Performing normalization processing on the written diagnosis to obtain a normalized written diagnosis;

The plurality of diagnosis pairs are determined based on the normalized candidate missed diagnoses and the normalized written diagnoses.

Optionally, the diagnosis pair determination unit is specifically configured to:

Combining the candidate missing diagnosis and the written diagnosis in pairs to form candidate diagnosis pairs; each of the candidate diagnosis pairs includes one candidate missing diagnosis and one written diagnosis;

Performing normalization processing on each of the candidate diagnosis pairs to obtain normalized candidate diagnosis pairs;

De-duplication processing is performed on each of the normalized candidate diagnosis pairs to determine the multiple diagnosis pairs.

Optionally, the device further comprises a diagnostic symptom pair set determining unit, wherein the diagnostic symptom pair set determining unit is specifically configured to:

Obtaining a reference medical record text, a candidate omitted diagnosis corresponding to the reference medical record text, and a written diagnosis corresponding to the reference medical record text;

Combining the written diagnoses corresponding to the reference medical record text in pairs to obtain a first diagnosis pair set, combining the candidate omitted diagnoses corresponding to the reference medical record text and the written diagnoses corresponding to the reference medical record text in pairs, and combining the candidate omitted diagnoses corresponding to the reference medical record text in pairs to obtain a second diagnosis pair set;

Calculating the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set;

The diagnosis-symptom pair set is determined from the first diagnosis pair set and the second diagnosis pair set based on the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set.

A device for detecting omissions in medical record diagnosis, comprising a memory and a processor;

The memory is used to store programs;

The processor is used to execute the program to implement the various steps of the medical record diagnosis omission detection method as described above.

A readable storage medium stores a computer program, which, when executed by a processor, implements the various steps of the medical record diagnosis omission detection method as described above.

By means of the above technical scheme, the present application discloses a method, device, equipment and readable storage medium for detecting omissions in medical record diagnoses. In this scheme, after determining the medical record text to be detected, first determine the candidate omission diagnoses corresponding to the medical record text; then determine the target candidate omission diagnoses from the candidate omission diagnoses, the target candidate omission diagnoses are candidate omission diagnoses that are semantically consistent with the written diagnoses in the medical record text, and candidate omission diagnoses that are determined to be symptoms of the written diagnoses; finally, determine the other candidate omission diagnoses except the target candidate omission diagnoses in the candidate omission diagnoses as omission diagnoses. This scheme performs omission detection of medical record diagnoses from two aspects: the semantic similarity of diagnoses and the symptom determination of diagnoses. When a candidate omission diagnosis is a symptom of a written diagnosis, even if the two are semantically consistent, it is still possible to avoid the candidate omission diagnosis from being detected as an omission diagnosis, thereby improving the accuracy of the results of omission detection of medical record diagnoses.

BRIEF DESCRIPTION OF THE DRAWINGS

Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

FIG1 is a flow chart of a method for detecting missed medical record diagnosis disclosed in an embodiment of the present application;

FIG2 is a flow chart of a method for determining candidate omitted diagnoses corresponding to medical record texts using a generative large model disclosed in an embodiment of the present application;

FIG3 is a flow chart of a method for fine-tuning an existing generative model in the medical field disclosed in an embodiment of the present application;

FIG4 is a flow chart of a method for determining a target candidate omission diagnosis from the candidate omission diagnoses disclosed in an embodiment of the present application;

FIG5 is a flow chart of a method for determining a preset diagnostic symptom pair set disclosed in an embodiment of the present application;

FIG6 is a schematic diagram of the structure of a device for detecting missed medical record diagnosis disclosed in an embodiment of the present application;

FIG. 7 is a hardware structure block diagram of a medical record diagnosis omission detection device disclosed in an embodiment of the present application.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. 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 creative work are within the scope of protection of this application.

Next, the medical record diagnosis omission detection method provided by the present application is introduced through the following embodiments.

Referring to FIG. 1 , FIG. 1 is a flow chart of a method for detecting missed medical record diagnosis disclosed in an embodiment of the present application. The method may include the following steps:

Step S101: Determine the medical record text to be detected.

In the present application, the medical record text is the text corresponding to the patient's medical record, which can be an electronic medical record entered based on the patient's self-description and the doctor's inquiries, or it can be a text obtained by optical character recognition of a medical record image obtained by scanning or photographing a handwritten paper medical record. This application does not impose any limitations on this.

The medical record text to be tested is the medical record text that needs to be tested for omissions in medical record diagnosis. It can be the original medical record text to be tested, or it can be the medical record text obtained after preprocessing the original medical record text to be tested. The preprocessing includes desensitization and information screening. The desensitization and information screening processing can specifically include deleting fields such as the individual or doctor's name, address, admission time, bed number, differential diagnosis, etc. in the original medical record text to obtain the medical record text to be tested. This application does not impose any restrictions on this.

Step S102: Determine the candidate omitted diagnosis corresponding to the medical record text.

The medical record text contains a large amount of description of the condition and information about the examinations and treatments received by the patient, as well as the conclusive information obtained after the doctor's diagnosis. Among them, the description of the condition and the information about the examinations and treatments received by the patient can also indirectly reflect the diseases that the patient may suffer from, while the conclusive information obtained by the doctor's diagnosis can directly reflect the diseases that the patient suffers from. In this application, the candidate omitted diagnosis corresponding to the medical record text is the disease that the patient may suffer from corresponding to the medical record text. The candidate omitted diagnosis may be directly recorded in the medical record text, or it may not be directly recorded in the medical record text. This application does not impose any restrictions on this.

In the present application, there are many ways to determine the candidate omitted diagnosis corresponding to the medical record text. As an implementable method, the name of the disease can be used as the nomenclature, and the medical record text can be identified by the nomenclature. It is also possible to collect the information of the disease standard or specific medicine in advance, and use the examination item or the name of the drug as the nomenclature, and identify the nomenclature of the medical record text to obtain the examination item or drug name contained in the medical record text, and the corresponding disease standard or specific medicine information, thereby determining the candidate omitted diagnosis corresponding to the medical record text. In addition, the diseases contained in the sample medical record text can also be pre-marked, and the deep learning model can be trained accordingly, so that the deep learning model can learn the mapping relationship between the medical record text and the disease, so that the disease contained in the input medical record text can be output as a candidate omitted diagnosis. In addition, generative models such as ChatGPT model, PaLM (Pathways Language Model) model, Pangu model, Spark cognitive model, etc. have strong natural language understanding capabilities, and generative models can also be used in this application to determine the candidate omitted diagnosis corresponding to the medical record text. In this regard, this application does not make any restrictions.

It should be noted that in the present application, in order to improve the completeness of the candidate omission diagnoses corresponding to the medical record text, the above-mentioned multiple methods can be used to process the medical record text separately to obtain multiple candidate omission diagnosis results, and then the multiple candidate omission diagnosis results are fused to obtain the final candidate omission diagnosis. This application does not limit this.

Step S103: determining a target candidate omitted diagnosis from the candidate omitted diagnoses, wherein the target candidate omitted diagnosis is a candidate omitted diagnosis that is semantically consistent with a written diagnosis in the medical record text, and a candidate omitted diagnosis that is determined to be a symptom of the written diagnosis.

The written diagnosis in the medical record text can be determined by obtaining the diagnosis list field in the medical record text, for example, the discharge diagnosis list field on the first page of the medical record text or the discharge diagnosis list field in the discharge record of the medical record text, and determining the written diagnosis in the medical record text based on the diagnosis list field in the medical record text.

In this application, for each candidate omitted diagnosis, it is necessary to determine whether it is an omitted diagnosis. Generally, if the semantics of two diagnoses are consistent, it can be determined that the two diagnoses are the same diagnosis. Therefore, in this application, when a candidate omitted diagnosis is semantically consistent with a written diagnosis, the candidate omitted diagnosis can be excluded. However, in the actual writing and quality evaluation of medical records, according to the eighth edition of "Clinical Diagnosis" and other textbooks and medical record writing specifications, some clinical symptoms and signs that do not have specificity in the early stage of hospitalization should be excluded if there is a clear diagnosis after in-depth examination and comprehensive analysis. However, "no specific clinical symptoms and signs in the early stage of hospitalization" is likely to be inconsistent with the semantics of "clear diagnosis after in-depth examination and comprehensive analysis". In this case, "no specific clinical symptoms and signs in the early stage of hospitalization" will be determined as an omitted diagnosis. Therefore, in this application, candidate omitted diagnoses that are not omitted diagnoses are excluded from the two aspects of diagnostic semantic similarity and diagnostic symptom determination.

In the present application, the candidate omitted diagnosis can be compared with the written diagnosis in the medical record text based on the diagnostic semantic similarity and the diagnosis symptom judgment, and then the target candidate omitted diagnosis can be determined from the candidate omitted diagnoses. The specific implementation method will be described in detail through the subsequent embodiments and will not be described in detail here.

Step S104: determining other candidate omission diagnoses among the candidate omission diagnoses except the target candidate omission diagnosis as omission diagnoses.

This embodiment discloses a method for detecting omissions in medical record diagnoses. In this solution, after determining the medical record text to be detected, first determine the candidate omission diagnoses corresponding to the medical record text; then determine the target candidate omission diagnosis from the candidate omission diagnoses, the target candidate omission diagnosis is a candidate omission diagnosis that is semantically consistent with the written diagnosis in the medical record text, and a candidate omission diagnosis that is determined to be a symptom of the written diagnosis; finally, determine the other candidate omission diagnoses except the target candidate omission diagnosis in the candidate omission diagnoses as omission diagnoses. This solution performs omission detection of medical record diagnoses from two aspects: diagnostic semantic similarity and diagnostic symptom determination. When a candidate omission diagnosis is a symptom of a written diagnosis, even if the two are semantically consistent, it is still possible to avoid the candidate omission diagnosis from being detected as an omission diagnosis, thereby improving the accuracy of the medical record diagnosis omission detection results.

In the above embodiments, it is mentioned that generative models such as ChatGPT model, PaLM (Pathways Language Model) model, Pangu model, Spark cognitive model, etc. have strong natural language understanding capabilities. In the present application, the generative model can also be used to determine the candidate omission diagnosis corresponding to the medical record text. Next, in another embodiment of the present application, the specific implementation method of using the generative model to determine the candidate omission diagnosis corresponding to the medical record text is described. With reference to Figure 2, Figure 2 is a flow chart of a method for determining the candidate omission diagnosis corresponding to the medical record text using a generative model disclosed in an embodiment of the present application. The method may include the following steps:

Step S201: Generate prompt information corresponding to the medical record text, where the prompt information is used to indicate the candidate omitted diagnosis corresponding to the medical record text.

In the present application, specific questions and preset prompt information templates can be preset. The preset specific questions can be, for example, "What diagnoses are there in the medical record text?", "What diagnoses are included in the above medical record text?", etc. The preset prompt information template can include slots for filling in preset questions and slots for filling in medical record text.

As an implementable method, the medical record text and a preset specific question can be filled into a corresponding slot of a prompt information template to generate prompt information corresponding to the medical record text.

For ease of understanding, assume that the preset prompt information template is "Given medical record text: {input}\{prompt}", where the slot "{input}" is used to fill in the medical record text, and the slot "{prompt}" is used to fill in the preset question. The preset question is "What diagnoses are included in the above medical record text?" The medical record text is "Chief Physician Teng** believed during the ward rounds that the patient had intestinal adhesions, gallbladder stones with atrophic cholecystitis, and common bile duct stones. The patient underwent laparoscopic intestinal adhesion lysis + cholecystectomy + common bile duct incision and lithotomy + T-tube drainage. The operation was successful. The patient's vital signs are stable, there is no bile-like and turbid fluid in the abdominal drainage tube, and there are no signs of peritonitis in the abdomen, indicating that there is no bile leakage or intestinal fistula." The prompt information corresponding to the medical record text can be:

"Given medical record text: Chief Physician Teng** believed during the ward round that the patient had intestinal adhesions, gallbladder stones with atrophic cholecystitis, and common bile duct stones. He underwent laparoscopic intestinal adhesion lysis + cholecystectomy + common bile duct incision and lithotomy + T-tube drainage. The operation was successful. The patient's vital signs are now stable. There is no bile-like or turbid fluid in the abdominal drainage tube. There are no signs of peritonitis in the abdomen, indicating that there is no bile leakage or intestinal fistula.

What diagnoses are included in the above medical record text? ".

Considering that the length of the medical record text may exceed the processable text length threshold of the generative model, as another possible implementation method, the medical record text can be segmented to obtain multiple medical record text segments, and for each medical record text segment, prompt information corresponding to the medical record text segment is generated, and the prompt information is used to indicate the candidate omitted diagnosis corresponding to the medical record text segment. The combination of the prompt information corresponding to each medical record text segment is the prompt information corresponding to the medical record text.

In the present application, there are many ways to segment the medical record text, such as segmentation by length (for example, segmentation every 256 words), segmentation by field (for example, extracting fields such as diagnosis process and ward rounds records separately), segmentation by specific punctuation marks (for example, segmenting the medical record text according to "."), and one or more combinations of these methods. The present application does not impose any limitations on this.

In the present application, for each medical record text fragment, the medical record text fragment and a preset specific question can be filled into a corresponding slot of a prompt information template to generate prompt information corresponding to the medical record text fragment.

For ease of understanding, assume that the medical record text is text, and segment it to obtain multiple medical record text segments text ₁ , text ₂ , ..., text _n , where n is an integer greater than or equal to 1, and the preset prompt information template is "Given medical record text: {input}\{prompt}", where the slot "{input}" is used to fill in the medical record text segment, and the slot "{prompt}" is used to fill in the preset question, the preset question is "What diagnoses are included in the above medical record text?", then the prompt information corresponding to each of the medical record text segments is as follows:

"Given medical record text: text ₁

What diagnoses are included in the above medical record text? "

“Given medical record text: text ₂

What diagnoses are included in the above medical record text? "

…

"Given medical record text: text _n

What diagnoses are included in the above medical record text? "

Step S202: calling a generative model, and inputting prompt information corresponding to the medical record text into the generative model.

In the present application, if the prompt information corresponding to the medical record text includes prompt information corresponding to multiple medical record text fragments, when inputting the prompt information corresponding to each medical record text fragment into the generative model, the prompt information corresponding to one medical record text fragment can be input at a time, or the prompt information corresponding to multiple medical record text fragments can be input at a time. This application does not impose any limitations on this.

It should be noted that the generative model of the present application can be an existing generative model in the medical field. Considering that the existing large generative models in the medical field have poor understanding of medical record texts, in the present application, the training medical record text can also be used to fine-tune the existing generative model in the medical field to obtain a fine-tuned generative model for calling in this step. The fine-tuning method of the existing generative model in the medical field will be described in detail in the following embodiments and will not be described in detail here.

Step S203: Based on the response of the generative model, determine the candidate omitted diagnosis corresponding to the medical record text.

In the present application, the candidate omitted diagnosis corresponding to the medical record text is structured information, such as a list, which contains multiple elements, each element representing a candidate omitted diagnosis. The response of the generative model is generally natural language description information, from which the candidate omitted diagnosis corresponding to the medical record text can be extracted by regular matching and other methods.

For example, if the generative model responds with “the medical record text contains diagnosis ₁ , diagnosis ₂ , ..., diagnosis _n ”, then the candidate omitted diagnoses corresponding to the medical record text may be [diagnosis ₁ , diagnosis ₂ , ..., diagnosis _n ].

In the prior art, the medical record diagnosis omission detection technology implemented by deep learning models requires multiple models to recall candidate omission diagnoses in stages, which is prone to error transmission and requires repeated analysis of the same medical record text, resulting in low detection efficiency. In this embodiment, a large generative model with strong natural language comprehension ability is used to analyze the medical record text, which solves the problem of error transmission caused by staged recall. The generative model can take into account the semantic connection between different diagnoses in the medical record text, making the recalled candidate omission diagnoses more accurate and comprehensive, and can also avoid repeated analysis of the same medical record text, thereby improving detection efficiency.

Next, in another embodiment of the present application, a method for fine-tuning an existing generative model in the medical field is described in detail, with reference to FIG3 , which is a flow chart of a method for fine-tuning an existing generative model in the medical field disclosed in an embodiment of the present application, and the method may include the following steps:

Step S301: Obtain a training medical record text, the diagnosis corresponding to the training medical record text, and the label of each diagnosis in the training medical record text, wherein the label of each diagnosis in the training medical record text includes a first label, a second label, and a third label, wherein the first label is used to indicate whether the diagnosis is confirmed, the second label is used to indicate the diagnosis type of the diagnosis, and the third label is used to indicate a recall support fragment of the diagnosis.

In the present application, the method for determining the medical record text for training can be to first collect original medical record texts from different sources and different departments, pre-process the collected original medical record texts to obtain pre-processed medical record texts, and then segment the pre-processed long medical record texts to obtain training medical record texts. The pre-processing includes desensitization and information screening processing. The desensitization and information screening processing can specifically be to delete the fields such as personal or physician name, address, admission time, bed number, differential diagnosis, etc. in the original medical record text. There are many ways to segment the medical record text, such as segmentation by length (for example, segmentation every 256 words), segmentation by field (for example, extracting fields such as diagnosis process and ward rounds records separately), segmentation by specific punctuation marks (for example, segmenting the medical record text according to "."), and other one or more methods. This application does not impose any restrictions on this.

As an implementable method, the diagnosis corresponding to the training medical record text can be determined by manual processing or by using a deep learning model, a rule base or a knowledge base.

For example, the training medical record text is "Chief Physician Teng** believed during the ward rounds that the patient had intestinal adhesions, gallstones with atrophic cholecystitis, and common bile duct stones. The patient underwent laparoscopic intestinal adhesion lysis + cholecystectomy + common bile duct incision and lithotomy + T-tube drainage. The operation was successful and the patient's vital signs are stable. There is no bile or turbid fluid in the abdominal drainage tube, and there are no signs of peritonitis in the abdomen, indicating that there is no bile leakage or intestinal fistula." The diagnostic entities of the training medical record text can be extracted by entity extraction to obtain the diagnosis "intestinal adhesions, intestinal fistulas, gallstones with atrophic cholecystitis" corresponding to the training medical record text.

The labels of each diagnosis corresponding to the training medical record text can be manually annotated or rule-based. For example, for the diagnosis of "intestinal fistula", the regular expression "(no|denial)[^.,]*occurrence" matches "no bile leakage, intestinal fistula occurrence", so that it can be judged that "intestinal fistula" is not a confirmed diagnosis; and by classifying the rules, such as the rule is a "denial type" rule, it can be judged that "intestinal fistula" is not a confirmed diagnosis because it is denied; the manually annotated supporting fragment is "the patient's vital signs are stable, there is no bile sample and turbid fluid in the abdominal drainage tube, and there is no peritonitis sign in the abdomen, indicating that there is no bile leakage or intestinal fistula occurrence.". Therefore, the diagnosis of "intestinal fistula" has the following labels: "Is it confirmed: unconfirmed; Diagnosis type: denial type diagnosis; Supporting fragment: "The patient's vital signs are stable, there is no bile sample and turbid fluid in the abdominal drainage tube, and there is no peritonitis sign in the abdomen, indicating that there is no bile leakage or intestinal fistula occurrence. ".

Step S302: Based on the training medical record text, the diagnosis corresponding to the training medical record text and the label of each diagnosis in the training medical record text, a plurality of training corpora are generated, each of which includes an input text and an expected output text.

In the present application, multiple training corpora can be generated based on the training medical record text, the diagnosis corresponding to the training medical record text, and the labels of each diagnosis in the training medical record text, combined with a preset template. The training corpus can involve different task types, such as single-negative tasks, interpretation tasks, extraction tasks, etc. The expression form of the training corpus needs to be as rich as possible, for example, the method of manually annotating thought chains can be adopted.

Preset templates include: "{input}\nWhich diagnoses exist in this medical record text?", "Which diagnoses exist in the following medical record text\n{input}?", "{input}\nIs {diagnosis} a confirmed diagnosis in this text?", "{input}\nWhy is {diagnosis} not a confirmed diagnosis in the above medical record text?", etc., where input is the medical record text and diagnosis is a specific diagnosis.

For ease of understanding, the following examples of training corpus are given:

Example 1:

Example 2:

Example 3:

Step S303: input the input text into an existing generative model in the medical field, and fine-tune the existing generative model in the medical field with the goal of making the output of the existing generative model in the medical field approach the expected output text corresponding to the input text.

In this application, the expected output text can be used as the correct answer, the output of the existing generative model in the medical field can be compared with the expected output text, a loss function is designed to measure the difference between the two outputs, and the model parameters are continuously updated through a feedback mechanism until the model converges and fine-tuning is completed.

After fine-tuning the existing generative model in the medical field using the embodiments of the present application, the generative model will have good diagnostic recall capabilities, thus laying the foundation for the accuracy of subsequent test results.

In another embodiment of the present application, a specific implementation method of determining a target candidate omission diagnosis from the candidate omission diagnoses in step S103 is described, with reference to FIG. 4 , which is a flow chart of a method for determining a target candidate omission diagnosis from the candidate omission diagnoses disclosed in an embodiment of the present application. The method may include the following steps:

Step S401: Based on the candidate missing diagnosis and the written diagnosis, a plurality of diagnosis pairs are determined, each diagnosis pair including one of the candidate missing diagnosis and one of the written diagnosis.

As an implementable method, the candidate missing diagnoses and the written diagnoses may be combined in pairs to form diagnosis pairs.

Taking into account that the same diagnosis may be described in different ways, resulting in the possibility that multiple diagnoses actually represent the same diagnosis, in order to avoid repetitive judgments on the same diagnosis, as a feasible implementation method, in the present application, the candidate omitted diagnoses may be first normalized to obtain a normalized candidate omitted diagnoses, and the written diagnoses may be normalized to obtain a normalized written diagnoses, and then multiple diagnosis pairs may be determined based on the normalized candidate omitted diagnoses and the normalized written diagnoses.

As another possible implementation method, the candidate omitted diagnosis and the written diagnosis can be combined in pairs to form candidate diagnosis pairs, each candidate diagnosis pair including one of the candidate omitted diagnosis and one of the written diagnosis, and then each candidate diagnosis pair is normalized to obtain normalized candidate diagnosis pairs, and finally the normalized candidate diagnosis pairs are deduplicated to determine multiple diagnosis pairs.

It should be noted that the normalization processing can be to vectorize the diagnosis text, or to obtain the diagnosis's leading words, location, cause, nature and other key elements through the knowledge graph for normalization.

For ease of understanding, assuming that the candidate missed diagnoses include [diagnosis ₁ , diagnosis ₂ , ..., diagnosis _n ], and the written diagnoses include [diagnosisOrigin ₁ , diagnosisOrigin ₂ , ..., diagnosisOrigin _n ], then an example of a diagnosis pair is [diagnosis ₁ , diagnosisOrigin ₁ ].

Step S402: for each diagnosis pair, determine whether the semantics of the two diagnoses in the diagnosis pair are consistent. If the semantics of the two diagnoses in the diagnosis pair are consistent, execute step S403; if the semantics of the two diagnoses in the diagnosis pair are inconsistent, execute step S404.

In the present application, the semantic similarity between the candidate omitted diagnosis and the written diagnosis in the diagnosis pair may be calculated, thereby determining whether the semantics of the two diagnoses in the diagnosis pair are consistent.

Step S403: determining the candidate omission diagnosis in the diagnosis pair as the target candidate omission diagnosis.

Step S404: Compare the diagnostic pair with each diagnostic symptom pair in a preset diagnostic symptom pair set. If the diagnostic pair is consistent with a diagnostic symptom pair in the diagnostic symptom pair set, determine the candidate omission diagnosis as a target candidate omission diagnosis. If the diagnostic pair is inconsistent with each diagnostic symptom pair in the diagnostic symptom pair set, determine the candidate omission diagnosis as a non-target candidate omission diagnosis.

In the present application, the preset diagnostic symptom pair set includes multiple diagnostic symptom pairs, each diagnostic symptom pair includes two diagnoses, one of which is the symptom corresponding to the other diagnosis. For ease of understanding, assuming that a diagnostic symptom pair is [diagnosis _a , diagnosis _b ], then diagnosis _a is the symptom of diagnosis _b , or, diagnosis _b is the symptom of diagnosis _a .

Among them, the preset diagnostic symptom pair set is determined based on a large amount of medical record data, and the specific determination method will be described in detail through the following embodiments.

In another embodiment of the present application, a method for determining a preset diagnostic symptom pair set is described in detail. Referring to FIG. 5 , FIG. 5 is a flow chart of a method for determining a preset diagnostic symptom pair set disclosed in an embodiment of the present application, and the method may include the following steps:

Step S501: Obtain a reference medical record text, a candidate omitted diagnosis corresponding to the reference medical record text, and a written diagnosis corresponding to the reference medical record text.

The reference medical record text may be determined by first collecting original medical record texts from different sources and different departments, preprocessing the collected original medical record texts, and obtaining the preprocessed medical record texts as reference medical record texts. The preprocessing method may refer to the relevant description in the aforementioned embodiment, which will not be repeated here. The candidate omitted diagnosis corresponding to the reference medical record text may be determined manually or by using the prior art, and this application does not impose any limitation on this. The method for determining the written diagnosis corresponding to the reference medical record text may refer to the relevant description in the aforementioned embodiment, which will not be repeated here.

Step S502: Combining the written diagnoses corresponding to the reference medical record text in pairs to obtain a first diagnosis pair set, combining the candidate omitted diagnoses corresponding to the reference medical record text and the written diagnoses corresponding to the reference medical record text in pairs, and combining the candidate omitted diagnoses corresponding to the reference medical record text in pairs to obtain a second diagnosis pair set.

It should be noted that the first diagnostic pair set includes multiple first diagnostic pairs, and the second diagnostic pair set includes multiple second diagnostic pairs. Each first diagnostic pair includes two written diagnoses, and each second diagnostic pair includes one candidate missing diagnosis and one written diagnosis, or two candidate missing diagnoses.

Step S503: Calculate the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set.

In the present application, the number of times the first diagnosis pair appears in all reference medical record texts and the ratio between the total number of times each diagnosis pair in the first diagnosis pair set appears in all reference medical record texts can be calculated as the co-occurrence probability of the first diagnosis pair.

Step S504: determining a diagnosis-symptom pair set from the first diagnosis pair set and the second diagnosis pair set based on the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set.

The higher the co-occurrence ratio of the diagnosis pair, the greater the co-occurrence probability, the more dense the co-occurrence situation, the more common the existence of the diagnosis pair, the more likely it is that they are two types of diseases that often occur in the same patient, or the relationship between a common disease and complications, and the less likely one disease is to be a symptom of another disease. On the contrary, the lower the co-occurrence ratio, the lower the co-occurrence probability, and the fewer co-occurrence situations, which means that doctors will not write two diagnoses at the same time in actual writing, and the more likely one disease is to be a symptom of another disease. Therefore, in this application, it is necessary to exclude the first diagnosis pair with a higher co-occurrence probability and the second diagnosis pair that is semantically consistent with the first diagnosis pair with a higher co-occurrence probability.

As an implementable method, a first threshold and a second threshold may be preset, the first threshold being greater than the second threshold, and the specific values of the first threshold and the second threshold may be set according to scenario requirements, and this application does not impose any limitation. Then, a target second diagnostic pair may be determined from the second diagnostic pair set, and the target second diagnostic pair is a second diagnostic pair that is semantically consistent with the first diagnostic pair whose co-occurrence probability is greater than the first preset threshold. The first diagnostic pairs in the first diagnostic pair set whose co-occurrence probability is less than the second preset threshold, and the second diagnostic pairs in the second diagnostic pair set except the target second diagnostic pair, are combined to obtain a set of candidate diagnostic symptom pairs.

As an implementable method, the candidate diagnosis symptom pair set can be used as the diagnosis symptom pair set. Considering that the description content of each diagnosis in the candidate diagnosis symptom pair set is not standardized, as another implementable method, the candidate diagnosis symptom pair set can be labeled to determine the diagnosis symptom pair set.

In the present application, manual full labeling can be adopted, or manual sampling labeling can be performed first, and then the diagnosis pairs with relatively consistent types can be classified through diagnosis pair similarity or other clustering methods to achieve batch labeling of diagnosis pairs (for example, the diagnosis pairs consisting of renal function injury, hypocalcemia and hypochloremia can be clustered into one category through clustering, and only one pair is labeled), reducing the labeling workload.

The following is a description of a medical record diagnosis omission detection device disclosed in an embodiment of the present application. The medical record diagnosis omission detection device described below and the medical record diagnosis omission detection method described above can be referenced to each other.

Referring to Figure 6, Figure 6 is a schematic diagram of the structure of a medical record diagnosis omission detection device disclosed in an embodiment of the present application. As shown in Figure 6, the medical record diagnosis omission detection device may include:

A medical record text determination unit 11, used to determine the medical record text to be detected;

A candidate omitted diagnosis determining unit 12, used to determine a candidate omitted diagnosis corresponding to the medical record text;

A target candidate omitted diagnosis determining unit 13 is used to determine a target candidate omitted diagnosis from the candidate omitted diagnoses, wherein the target candidate omitted diagnosis is a candidate omitted diagnosis that is semantically consistent with a written diagnosis in the medical record text, and a candidate omitted diagnosis that is determined to be a symptom of the written diagnosis;

The omission diagnosis determining unit 14 is configured to determine other candidate omission diagnoses among the candidate omission diagnoses except the target candidate omission diagnosis as omission diagnoses.

As an implementable embodiment, the candidate missing writing diagnosis determination unit includes:

A prompt information generating subunit, used for generating prompt information corresponding to the medical record text, wherein the prompt information is used for indicating the candidate omitted diagnosis corresponding to the medical record text;

A generative model calling subunit, used to call the generative model and input the prompt information corresponding to the medical record text into the generative model;

The candidate omitted diagnosis determination subunit is used to determine the candidate omitted diagnosis corresponding to the medical record text based on the response of the generative model.

As an implementable embodiment, the prompt information generating subunit is specifically used for:

Segmenting the medical record text to obtain multiple medical record text segments;

For each medical record text fragment, prompt information corresponding to the medical record text fragment is generated, and the prompt information is used to indicate the candidate omitted diagnosis corresponding to the medical record text fragment; the combination of prompt information corresponding to each medical record text fragment is the prompt information corresponding to the medical record text.

As an implementable embodiment, the generative model is obtained by fine-tuning an existing generative model in the medical field; the device includes a fine-tuning unit, and the fine-tuning unit is specifically used to:

Acquire a training medical record text, a diagnosis corresponding to the training medical record text, and a label of each diagnosis in the training medical record text, wherein the label of each diagnosis in the training medical record text includes a first label, a second label, and a third label, wherein the first label is used to indicate whether the diagnosis is confirmed, the second label is used to indicate the diagnosis type of the diagnosis, and the third label is used to indicate a recall support segment of the diagnosis;

Based on the training medical record text, the diagnosis corresponding to the training medical record text, and the label of each diagnosis in the training medical record text, a plurality of training corpora are generated, each of the training corpora includes an input text and an expected output text;

The input text is input into an existing generative model in the medical field, and the existing generative model in the medical field is fine-tuned with the goal of making the output of the existing generative model in the medical field approach the expected output text corresponding to the input text.

As an implementable embodiment, the target candidate missing writing diagnosis determination unit includes:

A diagnosis pair determination unit, configured to determine a plurality of diagnosis pairs based on the candidate missing diagnosis and the written diagnosis, each diagnosis pair including one of the candidate missing diagnosis and one of the written diagnosis;

A judging unit, configured to judge, for each diagnosis pair, whether the semantics of two diagnoses in the diagnosis pair are consistent;

A first processing unit, configured to determine a candidate omitted diagnosis in the diagnosis pair as a target candidate omitted diagnosis if the semantics of the two diagnoses in the diagnosis pair are consistent;

A comparing unit, configured to compare the diagnosis pair with each diagnosis-symptom pair in a preset diagnosis-symptom pair set if the semantics of the two diagnoses in the diagnosis pair are inconsistent;

The second processing unit is configured to determine the candidate omission diagnosis as a target candidate omission diagnosis if the diagnosis pair is consistent with a diagnosis symptom pair in the diagnosis symptom pair set.

As an implementable embodiment, the diagnosis pair determination unit is specifically used for:

Normalizing the candidate omission diagnoses to obtain normalized candidate omission diagnoses;

Performing normalization processing on the written diagnosis to obtain a normalized written diagnosis;

The plurality of diagnosis pairs are determined based on the normalized candidate missed diagnoses and the normalized written diagnoses.

As an implementable embodiment, the diagnosis pair determination unit is specifically used for:

Combining the candidate missing diagnosis and the written diagnosis in pairs to form candidate diagnosis pairs; each of the candidate diagnosis pairs includes one candidate missing diagnosis and one written diagnosis;

Performing normalization processing on each of the candidate diagnosis pairs to obtain normalized candidate diagnosis pairs;

De-duplication processing is performed on each of the normalized candidate diagnosis pairs to determine the multiple diagnosis pairs.

As an implementable embodiment, the device further includes a diagnostic symptom pair set determining unit, and the diagnostic symptom pair set determining unit is specifically used to:

Obtaining a reference medical record text, a candidate omitted diagnosis corresponding to the reference medical record text, and a written diagnosis corresponding to the reference medical record text;

Combining the written diagnoses corresponding to the reference medical record text in pairs to obtain a first diagnosis pair set, combining the candidate omitted diagnoses corresponding to the reference medical record text and the written diagnoses corresponding to the reference medical record text in pairs, and combining the candidate omitted diagnoses corresponding to the reference medical record text in pairs to obtain a second diagnosis pair set;

Calculating the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set;

The diagnosis-symptom pair set is determined from the first diagnosis pair set and the second diagnosis pair set based on the co-occurrence probability of each of the first diagnosis pairs in the first diagnosis pair set.

Referring to FIG. 7 , FIG. 7 is a hardware structure block diagram of a medical record diagnosis omission detection device provided in an embodiment of the present application. Referring to FIG. 7 , the hardware structure of the medical record diagnosis omission detection device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;

The processor 1 may be a central processing unit CPU, or an application-specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention, etc.;

The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory;

The memory stores a program, and the processor can call the program stored in the memory, and the program is used to:

Determine the medical records to be tested;

Determining a candidate omitted diagnosis corresponding to the medical record text;

Determine a target candidate omitted diagnosis from the candidate omitted diagnoses, the target candidate omitted diagnosis being a candidate omitted diagnosis that is semantically consistent with a written diagnosis in the medical record text, and a candidate omitted diagnosis that is determined to be a symptom of the written diagnosis;

The other candidate omission diagnoses among the candidate omission diagnoses except the target candidate omission diagnosis are determined as omission diagnoses.

Optionally, the detailed functions and extended functions of the program may refer to the above description.

The embodiment of the present application further provides a readable storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

Determine the medical records to be tested;

Determining a candidate omitted diagnosis corresponding to the medical record text;

Determine a target candidate omitted diagnosis from the candidate omitted diagnoses, the target candidate omitted diagnosis being a candidate omitted diagnosis that is semantically consistent with a written diagnosis in the medical record text, and a candidate omitted diagnosis that is determined to be a symptom of the written diagnosis;

The other candidate omission diagnoses among the candidate omission diagnoses except the target candidate omission diagnosis are determined as omission diagnoses.

Optionally, the detailed functions and extended functions of the program may refer to the above description.

Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims (11)

1. A method for detecting missed writing in medical record diagnosis, characterized in that the method includes: Determine the medical record text to be tested; Determine the candidate missed diagnosis corresponding to the medical record text; A target candidate missed diagnosis is determined from the candidate missed diagnosis, the target candidate missed diagnosis is a candidate missed diagnosis that is semantically consistent with a written diagnosis in the medical record text, and is determined to be the written diagnosis. Candidates for diagnosed symptoms omit diagnosis; Among the write-missed diagnosis candidates, other write-missed diagnosis candidates other than the target write-missed diagnosis candidate are determined as write-missed diagnoses. 2. The method according to claim 1, wherein determining the candidate omission diagnosis corresponding to the medical record text includes: Generate prompt information corresponding to the medical record text, where the prompt information is used to indicate the determination of a candidate omission diagnosis corresponding to the medical record text; Call the generative model and input the prompt information corresponding to the medical record text into the generative model; Based on the response of the generative model, a candidate omission diagnosis corresponding to the medical record text is determined. 3. The method according to claim 2, characterized in that generating prompt information corresponding to the medical record text includes: Segment the medical record text to obtain multiple medical record text fragments; For each medical record text fragment, prompt information corresponding to the medical record text fragment is generated, and the prompt information is used to indicate the determination of the candidate omission diagnosis corresponding to the medical record text fragment; the combination of prompt information corresponding to each medical record text fragment is Prompt information corresponding to the medical record text. 4. The method according to claim 2, wherein the generative model is obtained by fine-tuning an existing generative model in the medical field; and the generative model in the existing medical field is fine-tuned. Fine-tuning methods include: Obtain the training medical record text, the diagnosis corresponding to the training medical record text, and the labels of each diagnosis in the training medical record text. The label of each diagnosis in the training medical record text includes a first label, a second label, and a third label. Three labels, the first label is used to indicate whether the diagnosis is confirmed, the second label is used to indicate the diagnosis type of the diagnosis, and the third label is used to indicate the recall support segment of the diagnosis; Generate multiple training corpus based on the training medical record text, the diagnosis corresponding to the training medical record text, and the labels of each diagnosis in the training medical record text, each of the training corpus including input text and expected output text ; The input text is input into an existing generative model in the medical field, with the goal that the output of the existing generative model in the medical field approaches the expected output text corresponding to the input text. Fine-tuning domain generative models. 5. The method according to claim 1, wherein determining a target candidate missed write diagnosis from the candidate missed write diagnoses includes: Based on the candidate missed write diagnosis and the written diagnosis, determine a plurality of diagnosis pairs, each diagnosis pair including one of the candidate missed write diagnosis and one of the written diagnosis; For each diagnosis pair, determine whether the semantics of the two diagnoses in the diagnosis pair are consistent; If the semantics of the two diagnoses in the diagnosis pair are consistent, then determining the candidate missed writing diagnosis in the diagnosis pair to be the target candidate missing writing diagnosis; If the semantics of two diagnoses in the diagnosis pair are inconsistent, compare the diagnosis pair with each diagnostic symptom pair in the preset diagnostic symptom pair set; If the diagnosis pair is consistent with a certain diagnosis symptom pair in the diagnosis symptom pair set, the candidate write-missing diagnosis is determined to be the target candidate write-missing diagnosis. 6. The method of claim 5, wherein determining a plurality of diagnosis pairs based on the candidate missed write diagnosis and the written diagnosis includes: Standardize the candidate missed write diagnosis to obtain the standardized candidate missed write diagnosis; Standardize the written diagnosis to obtain the standardized written diagnosis; The plurality of diagnosis pairs are determined based on the normalized candidate missed write diagnosis and the normalized written diagnosis. 7. The method of claim 5, wherein determining a plurality of diagnosis pairs based on the candidate missed write diagnosis and the written diagnosis includes: The candidate missed write diagnosis and the written diagnosis are combined in pairs to form a candidate diagnosis pair; each candidate diagnosis pair includes one of the candidate missed write diagnosis and one of the written diagnosis; Perform normalization processing on each of the candidate diagnosis pairs to obtain standardized candidate diagnosis pairs; Deduplication processing is performed on each of the normalized candidate diagnosis pairs to determine the plurality of diagnosis pairs. 8. The method according to claim 5, wherein the method for determining the preset diagnostic symptom pair set includes: Obtain the reference medical record text, the candidate missed diagnosis corresponding to the reference medical record text, and the written diagnosis corresponding to the reference medical record text; Perform pairwise combinations of written diagnoses corresponding to the reference medical record text to obtain a first set of diagnosis pairs, perform pairwise combinations of candidate missed diagnoses corresponding to the reference medical record text and written diagnoses corresponding to the reference medical record text, And perform pairwise combinations of the candidate missed diagnoses corresponding to the reference medical record text to obtain a second set of diagnosis pairs; Calculate the co-occurrence probability of each first diagnosis pair in the first diagnosis pair set; Based on the co-occurrence probability of each first diagnosis pair in the first diagnosis pair set, the diagnostic symptom pair set is determined from the first diagnosis pair set and the second diagnosis pair set. 9. A medical record diagnosis missed writing detection device, characterized in that the device includes: The medical record text determination unit is used to determine the medical record text to be detected; A candidate missed writing diagnosis determination unit, used to determine the candidate missing writing diagnosis corresponding to the medical record text; A target candidate missed writing diagnosis determination unit, configured to determine a target candidate missing writing diagnosis from the candidate missing writing diagnosis, and the target candidate missing writing diagnosis is a candidate missing writing diagnosis that is semantically consistent with a written diagnosis in the medical record text. , and, a candidate missed diagnosis determined to be a symptom of said written diagnosis; A missed write diagnosis determination unit configured to determine other missed write diagnosis candidates other than the target candidate missed write diagnosis as missed write diagnoses. 10. A medical record diagnosis missed writing detection device, which is characterized in that it includes a memory and a processor; The memory is used to store programs; The processor is configured to execute the program to implement each step of the medical record diagnosis missed writing detection method according to any one of claims 1 to 8. 11. A readable storage medium with a computer program stored thereon, characterized in that when the computer program is executed by a processor, the medical record diagnosis missed writing detection method according to any one of claims 1 to 8 is implemented. various steps.