Disease diagnosis result evaluation method, device and equipment and storage medium

Abstract

The invention provides a disease diagnosis result evaluation method and device, equipment and a storage medium. The method comprises the steps of obtaining at least one to-be-evaluated disease diagnosis result determined based on a target medical record; for each to-be-evaluated disease diagnosis result, inputting the to-be-evaluated disease diagnosis result and the target medical record into a large language model to obtain at least two diagnosis roles output by the large language model and diagnosis dimensions of each diagnosis role; for each diagnosis role, based on the diagnosis role, the diagnosis dimension corresponding to the diagnosis role, the to-be-evaluated illness state diagnosis result and the target medical record, determining first prompt information, and inputting the first prompt information into a large language model to obtain a diagnosis analysis report corresponding to the diagnosis role output by the large language model; and based on the diagnosis analysis report corresponding to each diagnosis role, determining a target evaluation result of the to-be-evaluated illness state diagnosis result, the target evaluation result being used for representing whether the to-be-evaluated illness state diagnosis result is correct or not. The evaluation accuracy of the illness state diagnosis result can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for evaluating disease diagnosis results. Background Art

[0002] In the modern medical system, diagnostic review is a key link in ensuring the quality and accuracy of medical care. During the clinical diagnostic review process, it is often found that medical records have problems such as multiple diagnoses, which not only affects the patient's treatment effect, but also leads to unnecessary medical expenses.

[0003] At present, in order to solve the problem of multiple-write detection in diagnosis, it is usually necessary to manually pre-define some audit logic and rule templates, and then extract key content from medical records such as the patient's admission records and diagnosis texts, and compare these key content with the audit logic and rule templates to evaluate the diagnosis results and output the final evaluation results.

[0004] However, the above-mentioned method of evaluating the diagnosis results of a disease, which relies on audit logic and rule templates, has poor generalization and is prone to missing information in medical records, resulting in low accuracy of the evaluation results. Summary of the invention

[0005] The present invention provides a method, device, equipment and storage medium for evaluating disease diagnosis results, which are used to solve the defect of low evaluation accuracy of disease diagnosis results in the prior art and achieve the purpose of improving the evaluation accuracy of disease diagnosis results.

[0006] The present invention provides a method for evaluating a disease diagnosis result, comprising: Obtaining at least one diagnosis result of a condition to be evaluated determined based on the target medical record; For each of the to-be-evaluated disease diagnosis results, the to-be-evaluated disease diagnosis result and the target medical record are input into a large language model to obtain at least two diagnostic roles output by the large language model and diagnostic dimensions of each of the diagnostic roles; For each of the diagnostic roles, based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the diagnostic result of the condition to be evaluated and the target medical record, first prompt information is determined, and the first prompt information is input into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model; Based on the diagnostic analysis reports corresponding to the respective diagnostic roles, a target evaluation result of the diagnosis result of the condition to be evaluated is determined, and the target evaluation result is used to indicate whether the diagnosis result of the condition to be evaluated is correct.

[0007] According to a method for evaluating a disease diagnosis result provided by the present invention, determining a target evaluation result of the disease diagnosis result to be evaluated based on the diagnosis analysis report corresponding to each of the diagnostic roles includes: Determine second prompt information based on each of the diagnostic roles, the diagnostic analysis report and the diagnostic dimension corresponding to each of the diagnostic roles; Inputting the second prompt information into the large language model to obtain a summary evaluation result output by the large language model, wherein the summary evaluation result includes a diagnosis analysis report corresponding to all the diagnosis roles; For each of the diagnosis roles, determining voting information of the diagnosis role for the diagnosis result of the condition to be evaluated based on the summary evaluation result, wherein the voting information includes the voting result of the diagnosis role for whether the diagnosis result of the condition to be evaluated is correct; Based on the voting information of each of the diagnostic roles, a target evaluation result of the disease diagnosis result to be evaluated is determined.

[0008] According to a method for evaluating a disease diagnosis result provided by the present invention, the step of determining the voting information of the diagnosis role for the disease diagnosis result to be evaluated based on the summary evaluation result includes: Determining third prompt information based on the summary evaluation result and the diagnosis role; The third prompt information is input into the large language model to obtain the voting information of the diagnostic role for the diagnosis result of the disease to be evaluated output by the large language model.

[0009] According to a method for evaluating a disease diagnosis result provided by the present invention, the voting information also includes a reason for voting; The step of determining the target evaluation result of the condition diagnosis result to be evaluated based on the voting information of each of the diagnosis roles includes: In the case where there are inconsistent voting results among all the voting results, determining fourth prompt information based on the voting results and voting reasons corresponding to each of the diagnostic roles; Inputting the fourth prompt information into the large language model to obtain a summary voting result output by the large language model, wherein the summary voting result includes voting information corresponding to all the diagnosis roles; For each of the diagnosis roles, determining fifth prompt information based on the aggregate voting result and the diagnosis role; Inputting the fifth prompt information into the large language model, obtaining new voting information of the diagnosis role output by the large language model, and repeating the above steps until all new voting results are consistent, or the number of votes reaches a preset number; Based on the final voting information, a target evaluation result of the disease diagnosis result to be evaluated is determined.

[0010] According to a method for evaluating a disease diagnosis result provided by the present invention, the target evaluation result of the disease diagnosis result to be evaluated is determined based on the final voting information, including: Determining sixth prompt information based on the final voting information and the diagnosis result of the condition to be evaluated; The sixth prompt information is input into the large language model to obtain a comprehensive evaluation report output by the large language model, wherein the comprehensive evaluation report includes the target evaluation results and diagnostic opinions, and the diagnostic opinions include the same diagnostic opinions and / or divergent diagnostic opinions of each of the diagnostic roles.

[0011] According to a method for evaluating a disease diagnosis result provided by the present invention, the method further comprises: When all the voting results are consistent, the voting result is determined as the target evaluation result of the disease diagnosis result to be evaluated.

[0012] According to a method for evaluating disease diagnosis results provided by the present invention, when a number of disease diagnosis results to be evaluated greater than a first preset number among all the disease diagnosis results to be evaluated belong to diseases corresponding to the same department, the at least two diagnostic roles include a number of diagnostic roles belonging to the department greater than a second preset number.

[0013] The present invention also provides a device for evaluating disease diagnosis results, comprising: An acquisition module, used to acquire at least one diagnosis result of a condition to be evaluated determined based on a target medical record; An input module, for inputting the diagnosis result of the condition to be evaluated and the target medical record into a large language model for each of the diagnosis results of the condition to be evaluated, to obtain at least two diagnostic roles output by the large language model and diagnostic dimensions of each of the diagnostic roles; A determination module, configured to determine, for each of the diagnosis roles, first prompt information based on the diagnosis role, the diagnosis dimension corresponding to the diagnosis role, the diagnosis result of the condition to be evaluated, and the target medical record; The input module is further used to input the first prompt information into the large language model to obtain a diagnosis analysis report corresponding to the diagnosis role output by the large language model; The determination module is further used to determine a target evaluation result of the diagnosis result of the condition to be evaluated based on the diagnosis analysis report corresponding to each of the diagnostic roles, and the target evaluation result is used to indicate whether the diagnosis result of the condition to be evaluated is correct.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for evaluating the disease diagnosis result as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for evaluating a disease diagnosis result as described in any one of the above.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for evaluating disease diagnosis results.

[0017] The present invention provides a method, apparatus, device and storage medium for evaluating disease diagnosis results. At least one disease diagnosis result to be evaluated is obtained based on a target medical record. For each disease diagnosis result to be evaluated, the disease diagnosis result to be evaluated and the target medical record are input into a large language model to obtain at least two diagnostic roles output by the large language model and the diagnostic dimensions of each diagnostic role. For each diagnostic role, first prompt information is determined based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the disease diagnosis result to be evaluated and the target medical record. The first prompt information is input into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model. Based on the diagnostic analysis reports corresponding to each diagnostic role, a target evaluation result of the disease diagnosis result to be evaluated is determined. The target evaluation result is used to characterize whether the disease diagnosis result to be evaluated is correct. Since the large language model can be used to determine different diagnostic roles and the diagnostic dimensions of each diagnostic role, and then the diagnostic analysis report of each diagnostic role based on the corresponding diagnostic dimensions can be determined based on the large language model, the final target evaluation result can be determined according to the diagnostic analysis reports of different diagnostic roles. The large language model has strong semantic understanding and processing capabilities, as well as the advantage of being able to tap into rich medical knowledge. Therefore, when using the large language model for disease diagnosis review, it is possible to avoid using fixed review logic and rule templates, thereby improving the generalization of the evaluation method. Moreover, by fully tapping into the useful information in the target case through the large language model, it can adapt to complex and changeable medical scenarios and improve the accuracy of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A schematic diagram of a flow chart of a method for evaluating a disease diagnosis result provided in an embodiment of the present invention.

[0020] Figure 2 A flowchart of a method for evaluating disease diagnosis results provided in an embodiment of the present invention.

[0021] Figure 3 A schematic diagram of the structure of a device for evaluating disease diagnosis results provided in an embodiment of the present invention.

[0022] Figure 4 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] With the continuous advancement of medical technology, diagnostic technology plays an increasingly important role in improving the efficiency, accuracy and patient experience of diagnosis and treatment. In the process of clinical diagnosis review, it is often found that there are problems with over-writing of diagnoses in medical records, which not only affects the treatment effect of patients, but also leads to unnecessary medical expenses. Therefore, diagnostic review plays a very important role in the medical system.

[0025] In the prior art, when conducting multiple review of medical diagnosis, medical experts usually need to pre-formulate some review logic and rule templates, and then compare the medical record information such as admission records and diagnosis texts with the review logic and rule templates, and output the evaluation results of the medical diagnosis results after some yes / no logical judgments. However, in the above method, due to the need to rely on specific review logic and rule templates, the evaluation method has poor generalization, and relies on less information in the case, it is easy to miss information in the medical record, and it is difficult to adapt to complex and changeable medical scenarios. In summary, the existing evaluation method of medical diagnosis results will result in low accuracy of the evaluation results.

[0026] In response to the above problems, an embodiment of the present invention provides an evaluation method for disease diagnosis results. In this method, a large language model can be used to review the disease diagnosis results. Since the large language model has strong semantic understanding and processing capabilities, and can tap into the advantages of rich medical knowledge, when using the large language model to review the disease diagnosis, it is possible to avoid using fixed review logic and rule templates, thereby improving the generalization of the evaluation method. Moreover, by fully tapping into the useful information in the case through the large language model, it can adapt to complex and changeable medical scenarios and improve the accuracy of the evaluation results.

[0027] Combine the following Figure 1 and Figure 2 The evaluation method of the disease diagnosis result provided by the embodiment of the present invention is described. The execution subject of the method can be an electronic device such as a terminal device, a computer or a server, or a smart device specially designed for evaluating the disease diagnosis result, or it can be an evaluation device for the disease diagnosis result set in the electronic device or smart device, and the evaluation device for the disease diagnosis result can be implemented by software, hardware or a combination of both. The method can be applied in the medical field, in scenarios where the correctness of the disease diagnosis result is evaluated through a large language model, and in particular, in scenarios where a diagnosis result is evaluated as to whether it is a multiple-write diagnosis result.

[0028] Figure 1 A flow chart of a method for evaluating a disease diagnosis result provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes: Step 101: Obtain at least one diagnosis result of a condition to be evaluated determined based on a target medical record.

[0029] In this step, the target medical record may include personal information, examination results, and diagnosis information. Personal information includes the patient's age, gender, and medical history, examination results include relevant laboratory results and imaging examination reports, and diagnosis information includes disease name, symptom description, and clinical manifestations. By analyzing the target case, at least one diagnosis result of the condition to be evaluated can be determined, such as type 2 diabetes, hypertension, and chronic kidney disease.

[0030] Step 102: For each diagnosis result of the condition to be evaluated, the diagnosis result of the condition to be evaluated and the target medical record are input into the large language model to obtain at least two diagnostic roles and diagnostic dimensions of each diagnostic role output by the large language model.

[0031] In this step, each condition diagnosis result to be evaluated can be evaluated according to the evaluation method of condition diagnosis results provided in the embodiment of the present invention to determine whether it is an overwritten diagnosis result. In the audit task of overwritten diagnosis, since different diagnostic results may exist in different categories of people and different department scenarios, and the conditions for different diagnoses vary greatly, to determine whether the diagnosis result is overwritten, experts from different departments and different professions must judge and discuss separately in different audit dimensions, and how to determine the list of experts to be organized and the content to be discussed based on the diagnosis dynamics also requires professional medical knowledge. Therefore, it is necessary to first determine the role of the host, and then the host analyzes the input personal information, test results and diagnostic information based on the key information in the target medical record, so as to dynamically generate at least two suitable diagnostic roles and the diagnostic dimensions corresponding to each diagnostic role, wherein the diagnostic role can be a virtual diagnostic expert.

[0032] Specifically, through the prompt word engineering, the Large Language Model (LLM) can be guided to play the role of a medical conference host and identify key information, so that it has a professional understanding of the organization of medical conferences. The prompt word examples are as follows: You are the main initiator and host of a medical conference with the theme of "Diagnosis, Multi-Write and Review". You understand the disease category information that multidisciplinary experts are good at, and are proficient in disease diagnosis standards and medical record writing standards. For example, experts in endocrinology, nutrition and other specialties are more proficient in diagnoses such as "diabetes", and professional experts in oncology, radiotherapy and chemotherapy, hematology and other specialties are more proficient in diagnoses such as "acute myeloid leukemia". For example, the diagnosis of "liver cyst" should be reviewed from the dimensions of past medical history, imaging examinations, etc., and "hypokalemia" should be more concerned with "laboratory projects".

[0033] Among them, keywords such as "medical conference", "disease diagnosis knowledge", and "multidisciplinary experts" in the prompt words can guide the large language model to fully activate the internal medical knowledge base, deeply analyze diagnosis and medical record related information, and roughly classify the diagnosis.

[0034] In addition, key information can be extracted from the target case through preset rules or large language models, such as personal information, test results, and diagnostic information.

[0035] For each diagnosis result to be evaluated, a prompt word can be generated based on the diagnosis result to be evaluated and the key information extracted from the target case. Through the prompt word engineering, the large language model is guided to generate diagnostic roles for different departments and those that are good at different audit dimensions. The diagnostic role can be, for example, an expert role.

[0036] For example, assuming that the diagnosis of the condition to be evaluated is type 2 diabetes with complications, the generated prompt words can be: Given the following {diag}, medical record information {data}, in order to review whether there is any duplication of the diagnosis result, you need to generate a list of medical expert roles attending the meeting, such as professional experts in internal medicine, surgery, and imaging, and determine the review dimensions of each expert, such as laboratory indicators, imaging examination matching, medication regimen, etc.

[0037] Enter the patient's diagnosis results (diag) and some medical records (data), including: Diagnosis of the condition to be assessed: Type 2 diabetes with complications; Some medical record information includes: Other diagnostic information: hypertension, malnutrition, cholecystitis, mild pulmonary hypertension, chronic kidney disease; Department: Endocrinology Department; Age: 55 years old; Medical history: 10 years of history of type 2 diabetes, 5 years of history of coronary heart disease, no history of surgery; Laboratory test items: 2022.02.15 full set of biochemistry: Blood sugar value: HbA1c 9.0%, Blood type: ABO blood type AB type, Rh (D) blood type positive, routine stool + occult blood test showed no abnormalities.

[0038] The above information is input into the large language model, and after analysis, the large language model can output the diagnostic role E and the diagnostic dimension AD of each diagnostic role, wherein the diagnostic role E and the diagnostic dimension AD of each diagnostic role are exemplified as follows: Diagnosis role set: {Endocrinologist 1, Endocrinologist 2, Nephrologist 1, ...}; Diagnostic dimension set: {laboratory indicators, medication regimen, past medical history, ...}, among which endocrinologists should analyze from the audit perspectives of laboratory test indicators..., etc., ...

[0039] In summary, through prompt word engineering, the large language model can be guided to generate expert roles in different departments and with expertise in different audit dimensions based on the diagnosis results of the disease to be evaluated and the information extracted from the target cases.

[0040] Among them, the diagnostic roles and diagnostic dimensions of each diagnostic role can be dynamically generated according to the diagnostic results of the disease to be evaluated and the target medical records, so that each diagnostic role can be reviewed from multiple diagnostic dimensions to ensure that no information in the target case is missed, thereby improving the comprehensiveness of the review.

[0041] Step 103: For each diagnostic role, based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the diagnostic result of the condition to be evaluated and the target medical record, determine the first prompt information, and input the first prompt information into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model.

[0042] In this step, taking the diagnosis role as an expert role as an example, for each expert role, the first prompt information can be generated based on the diagnosis result of the condition to be evaluated, the target case and the diagnosis dimension corresponding to the expert role. For example, the generated first prompt information can be: "You are an endocrinologist, and the diagnosis result of the condition to be evaluated is "type 2 diabetes with complications". Given the target medical record {data}, please review it according to the above information, paying special attention to the laboratory indicators, medication regimen, and past medical history. Please provide your detailed review opinions and information that needs further confirmation, and finally give a conclusion of overwriting / non-overwriting.", wherein the target medical record data includes: "...past medical history:...10 years of diabetes history...daily course of disease:...laboratory tests:...HbA1c 9.0%...24-hour urine protein: 3g...UAE: 300μg / min...doctor's orders: metformin,...insulin,...sacubitril valsartan...furosemide injection...charge details:...insulin...".

[0043] After the generated first prompt information is input into the large language model, the large language model analyzes the input information through the professional background corresponding to the expert role, thereby obtaining a diagnostic analysis report corresponding to the expert role. Among them, the diagnostic analysis report corresponding to the expert role may include the evaluation and analysis process, including analysis of each diagnostic dimension, further questions or information that needs to be clarified regarding the analysis results, and the evaluation results. The diagnostic analysis report corresponding to the expert role can be exemplified as follows: Analysis process: Laboratory indicators: ① HbA1c can reflect the average blood sugar level in the past 8 to 12 weeks and is the "gold standard" for evaluating the patient's blood sugar control. The normal reference value detected by the standard detection method is: 4% to 6%. The patient's HbA1c reached 9.0%, which meets the diagnostic criteria for diabetes. ② The patient's 24-hour urine protein reached 3g, indicating that the patient may have kidney disease.

[0044] Medication plan: ① Insulin, a typical medication for diabetes, was found in both the doctor's orders and the billing details. ② Metformin, included in the doctor's orders in the patient's medical records, is a first-line drug for the treatment of type 2 diabetes, especially for overweight patients. The above analysis suggests that the patient has diabetes.

[0045] Past medical history: The patient's other diagnoses included chronic kidney disease, but it was unclear whether it was a complication of diabetes.

[0046] Further questions or information that needs clarification: Through analysis of medical record information, the patient can be diagnosed with diabetes and kidney disease, but whether the kidney disease is a complication of diabetes requires comprehensive consideration of detailed medical history, test information, etc.

[0047] Evaluation result: There may be overwriting or wrong writing.

[0048] After obtaining the diagnostic analysis report corresponding to each expert role, each diagnostic analysis report will be stored in the system.

[0049] Step 104: Based on the diagnostic analysis reports corresponding to each diagnostic role, determine the target evaluation result of the diagnosis result of the condition to be evaluated, and the target evaluation result is used to indicate whether the diagnosis result of the condition to be evaluated is correct.

[0050] In this step, after obtaining the diagnostic analysis report corresponding to each diagnostic role, by analyzing the analysis process, further questions or information that needs to be clarified, and the evaluation results in each diagnostic analysis report, and comprehensively considering each diagnostic analysis report, the target evaluation result for the diagnostic result of the condition to be evaluated can be obtained to determine whether the diagnostic result of the condition to be evaluated is correct, for example, whether it is an overwritten diagnostic result.

[0051] In one possible implementation, the diagnostic analysis report corresponding to each diagnostic role can be input into the large language model, and each diagnostic analysis report can be analyzed by the large language model to obtain the final target evaluation result. In another possible implementation, it is also possible to determine whether the number of identical evaluation results in the diagnostic analysis reports corresponding to all diagnostic roles is greater than a preset number. If so, the above identical evaluation results are used as the final target evaluation results.

[0052] The method for evaluating the diagnosis results of a disease state provided in an embodiment of the present invention obtains at least one diagnosis result of a disease state to be evaluated determined based on a target medical record, and for each diagnosis result of a disease state to be evaluated, inputs the diagnosis result of the disease state to be evaluated and the target medical record into a large language model to obtain at least two diagnostic roles output by the large language model and the diagnostic dimensions of each diagnostic role, and for each diagnostic role, determines first prompt information based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the diagnosis result of the disease state to be evaluated and the target medical record, and inputs the first prompt information into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model, and determines a target evaluation result of the diagnosis result of the disease state to be evaluated based on the diagnostic analysis report corresponding to each diagnostic role, and the target evaluation result is used to characterize whether the diagnosis result of the disease state to be evaluated is correct. Since the large language model can be used to determine different diagnostic roles and the diagnostic dimensions of each diagnostic role, and then the diagnostic analysis report of each diagnostic role based on the corresponding diagnostic dimensions can be determined based on the large language model, the final target evaluation result can be determined according to the diagnostic analysis reports of different diagnostic roles. The large language model has strong semantic understanding and processing capabilities, as well as the advantage of being able to tap into rich medical knowledge. Therefore, when using the large language model for disease diagnosis review, it is possible to avoid using fixed review logic and rule templates, thereby improving the generalization of the evaluation method. Moreover, by fully tapping into the useful information in the target case through the large language model, it can adapt to complex and changeable medical scenarios and improve the accuracy of the evaluation results.

[0053] Exemplarily, on the basis of the above-mentioned embodiments, when determining the target evaluation result of the diagnosis result of the disease to be evaluated based on the diagnosis analysis report corresponding to each diagnostic role, the second prompt information can be determined based on each diagnostic role, the diagnosis analysis report corresponding to each diagnostic role and the diagnostic dimension, and the second prompt information can be input into the large language model to obtain the summary evaluation result output by the large language model, the summary evaluation result includes the diagnosis analysis reports corresponding to all diagnostic roles, and for each diagnostic role, the voting information of the diagnostic role for the diagnosis result of the disease to be evaluated is determined based on the summary evaluation result, the voting information includes the voting result of the diagnostic role on whether the diagnosis result of the disease to be evaluated is correct, and based on the voting information of each diagnostic role, the target evaluation result of the diagnosis result of the disease to be evaluated is determined.

[0054] Specifically, after obtaining the diagnostic analysis report corresponding to each diagnostic role, all diagnostic analysis reports need to be summarized. Therefore, the second prompt information can be determined based on the diagnostic role, the diagnostic analysis report corresponding to each diagnostic role, and the diagnostic dimension. The second prompt information can be, for example: "You are the main initiator and moderator of a medical conference with the theme of "Diagnosis Multi-Writing Review". Please summarize the review opinions of the following {E} expert roles: {report personal}, identify points of agreement and disagreement among the expert roles, and generate preliminary summary evaluation results based on diagnostic dimensions such as {AD}, and mark issues that require further discussion. ” Wherein, E represents at least one diagnostic role generated in the above embodiment, that is, an expert role, report personal It represents the diagnostic analysis report corresponding to each diagnostic role, and AD represents the diagnostic dimension corresponding to each diagnostic role.

[0055] After the second prompt information is generated, the second prompt information is input into the large language model, and each diagnostic analysis report is classified and conflict analyzed according to similarity through the large language model, consensus is summarized, and inconsistent opinions are identified and marked as issues that need further discussion. According to the degree of influence and rationality of the opinions, the conflicts that need to be discussed in depth are sorted, thereby forming a preliminary summary evaluation result. The summary evaluation result includes the diagnostic analysis reports corresponding to all diagnostic roles, that is, the main views and consensus of each diagnostic role, issues that need further clarification, and the overall conclusion on whether the diagnostic results of the condition to be evaluated are overwritten.

[0056] The summary evaluation results output by the large language model can be exemplified as follows: "Endocrinologist 1 believes that it is uncertain whether there is over-writing,... Laboratory indicator dimension: ① HbA1c is the "gold standard" for evaluating the patient's blood sugar control,..., HbA1c reaches 9.0%, and type 2 diabetes is diagnosed. ② The patient's 24-hour urine protein reaches 3g, suggesting that the patient may have kidney disease. ③ UAE..., the patient's indicator reaches 300μg / min, which can be diagnosed with diabetic nephropathy. Medication regimen dimension: ...there are insulin and metformin, etc.,..., there are furosemide injections for kidney disease, etc.... Past medical history:... Issues that need further clarification: ...whether the patient's kidney disease is caused by diabetes... Overall conclusion: there is a tendency for there to be no over-writing."

[0057] After the summary evaluation results are generated, the consensus voting mechanism can be used to form the final target evaluation results by voting by each diagnostic role. Each diagnostic role can determine its own voting information based on the content of the summary evaluation results on which the diagnostic roles have reached consensus, the content on which they disagree with other diagnostic roles, and the issues that need further clarification. The voting information includes the voting results of the diagnostic roles on whether the diagnosis results of the condition to be evaluated are correct.

[0058] Exemplarily, when determining the voting information of the diagnostic role regarding the diagnosis result of the disease to be evaluated based on the summarized evaluation results, the third prompt information can be determined based on the summarized evaluation results and the diagnostic role, and the third prompt information is input into the large language model to obtain the voting result of the diagnostic role output by the large language model regarding whether the diagnosis result of the disease to be evaluated is correct.

[0059] Specifically, for each diagnosis role, a third prompt message may be generated based on the summary evaluation result and the diagnosis role. The third prompt message may be as follows: "Please select the diagnosis role according to the summary evaluation result {Report interim}, vote on the conclusion given. When voting, please integrate the opinions of other expert roles, choose to support or oppose, and explain your reasons. " After the third prompt information is generated, the third prompt information is input into the large language model, and the large language model outputs the voting information of the diagnostic role for the diagnostic result of the condition to be evaluated by considering the opinion of the diagnostic role and the opinions of other diagnostic roles. Among them, an example of the voting information output by the large language model is as follows: "Endocrinologist 1: Support. According to the content provided by the summary evaluation results, I ignored the UAE indicator, which is the gold standard for diabetic nephropathy... Endocrinologist 2: Oppose. According to..., further clarification is needed. Nephrologist: Support. According to..." In the above method, the third prompt information generated based on the summary evaluation results and the diagnostic role can be input into the large language model, and the voting information of each diagnostic role regarding the diagnosis result of the disease to be evaluated can be determined through the large language model. When voting, each diagnostic role will comprehensively consider the opinions and evaluation results of other diagnostic roles, so that the voting information can be more accurate.

[0060] After the voting information of each diagnosis role is determined, the voting results in each voting information may be comprehensively considered to determine whether the target evaluation result is overwritten for the diagnosis result of the condition to be evaluated.

[0061] In this embodiment, after the diagnostic analysis reports corresponding to all diagnostic roles are comprehensively analyzed and the summary evaluation results are obtained, each diagnostic role can vote based on the summary evaluation results, thereby determining the target evaluation results of the disease diagnosis results to be evaluated based on the voting information of each diagnostic role. Since the voting information of all diagnostic roles can be comprehensively considered, the error caused by the evaluation results determined by a single diagnostic role can be reduced, making the target evaluation results more fair and improving the accuracy of the target evaluation results.

[0062] Exemplarily, based on the above embodiments, the above voting information also includes voting reasons. When determining the target evaluation result of the disease diagnosis result to be evaluated based on the voting information of each diagnosis role, it can be achieved in the following manner: In the case that there are inconsistent voting results among all the voting results, fourth prompt information is determined based on the voting results and voting reasons corresponding to each diagnostic role, and the fourth prompt information is input into the large language model to obtain the aggregated voting results output by the large language model. The aggregated voting results include the voting information corresponding to all diagnostic roles. For each diagnostic role, fifth prompt information is determined based on the aggregated voting results and the diagnostic role, and the fifth prompt information is input into the large language model to obtain new voting information of the diagnostic role output by the large language model. The above steps are repeated until all new voting results are consistent, or the number of votes reaches a preset number, and based on the final voting information, the target evaluation result of the diagnosis result of the disease to be evaluated is determined.

[0063] Specifically, after each round of voting, if it is determined that there are inconsistent voting results among all the voting results, a summary voting result, that is, a voting brief, will be generated based on the voting results and voting reasons corresponding to each diagnostic role. The voting brief includes the statistical number of votes in favor and against, based on which the proportion of support and opposition can be analyzed. In addition, it can also include the content currently reached consensus and the content that needs further discussion. For example, after the fourth prompt information determined based on the voting results and voting reasons corresponding to each diagnostic role is input into the large language model, the summary voting result output by the large language model can be obtained. Among them, the summary voting results can be exemplified as follows: Statistics of votes in favor and against: Total votes: 3 Support Tickets: 2 Votes against: 1 Summary of comments: consensus: Laboratory indicator dimensions:…… Medication regimen dimensions: Past medical history:…… Disagreement: ...needs further clarification.

[0064] Furthermore, for each diagnostic role, a fifth prompt information can be generated based on the aggregate voting results and the diagnostic role, and the fifth prompt information can be input into the large language model, and the aggregate voting results are analyzed by the large language model to output new voting information for the diagnostic role. It should be understood that the new voting information corresponding to the diagnostic role output by the large language model is the voting information after the large language model adjusts its own evaluation results based on the content reached consensus in the aggregate voting results and information from other diagnostic roles.

[0065] If there are still inconsistent voting results in all the new voting information, the above process can be repeated until all the new voting results are consistent, or the number of votes reaches a preset number, so as to determine the target evaluation result for the disease diagnosis result to be evaluated based on the final voting information.

[0066] In this embodiment, when there are inconsistent voting results among all the voting results, the final voting results can be determined by iterative voting, so as to continuously adjust the voting information corresponding to each diagnostic role with reference to the voting information of other diagnostic roles, so that each voting information integrates and refers to the voting results of different diagnostic roles, thereby improving the accuracy of the final voting information, thereby further improving the accuracy of the target evaluation results. In addition, the above-mentioned evaluation method of the disease diagnosis results, which uses a large language model to conduct multiple rounds of voting for diagnostic roles to proofread the evaluation results, can solve the problem of error transmission caused by the large number of steps in the task itself when evaluating through deep learning in the prior art, so that the final evaluation result is controllable and the reliability of the target evaluation result is improved.

[0067] Illustratively, based on the above-mentioned embodiment, when determining the target evaluation result of the diagnosis result of the disease to be evaluated based on the final voting information, the sixth prompt information can be determined based on the final voting information and the diagnosis result of the disease to be evaluated, and the sixth prompt information can be input into the large language model to obtain a comprehensive evaluation report output by the large language model. The comprehensive evaluation report includes the target evaluation result and the diagnostic opinion, and the diagnostic opinion includes the same diagnostic opinion and / or divergent diagnostic opinions of each diagnostic role.

[0068] Specifically, if there are still inconsistent voting results in the final voting information after the number of votes reaches the preset number, the sixth prompt information can be generated based on the final voting information and the diagnosis result of the disease to be evaluated. The sixth prompt information can be as follows: Please generate a comprehensive report of multiple reviews of the {diag} diagnosis for this patient based on the voting information {Votei}. The report will include the consensus reached in each diagnostic dimension and the remaining differences, and finally give a recommended review conclusion.

[0069] Among them, Votei is the final voting information, and diag is the diagnosis result of the disease to be evaluated.

[0070] The sixth prompt information is input into the large language model, and the large language model is guided by prompt words to generate a comprehensive evaluation report, which includes target evaluation results and diagnostic opinions. The diagnostic opinions include the same diagnostic opinions and / or divergent diagnostic opinions of each diagnostic role.

[0071] In this embodiment, when the number of votes reaches a preset number, if there are still inconsistencies in the voting results, a comprehensive evaluation report can be output based on the large language model. The comprehensive evaluation report includes not only the target evaluation results, but also the diagnostic opinions, so that the comprehensive evaluation report is more comprehensive, thereby providing a more comprehensive decision-making basis for subsequent diagnosis.

[0072] Exemplarily, when all voting results are consistent, the voting result is determined as the target evaluation result of the disease diagnosis result to be evaluated.

[0073] Specifically, when the voting results of all diagnostic roles are consistent, it means that all diagnostic roles have the same opinion on the evaluation result of the diagnosis result of the disease to be evaluated. Therefore, the final voting result can be directly determined as the target evaluation result of the diagnosis result of the disease to be evaluated, thereby ensuring the consistency of the final target evaluation result.

[0074] Exemplarily, based on the above embodiments, when more than a first preset number of all the diagnosis results for the condition to be evaluated belong to diseases corresponding to the same department, at least two diagnostic roles include more than a second preset number of diagnostic roles belonging to the department.

[0075] Specifically, the diagnostic role can be understood as a virtual diagnostic expert. Diagnostic experts from different departments are good at diagnosing different diseases. Therefore, if among all the patient's condition diagnosis results to be evaluated, there are more than a first preset number of condition diagnosis results to be evaluated that belong to the diseases corresponding to the same department, it means that most of the patient's symptoms belong to the same department. At this time, in order to fully tap the medical knowledge of the diagnostic experts in this department, it can be set that more than a second preset number of diagnostic roles belong to this department, thereby improving the accuracy of the target evaluation results for the condition diagnosis results to be evaluated.

[0076] Figure 2 A flowchart of a method for evaluating a disease diagnosis result provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, in an embodiment of the present invention, a large language model can be used to perform a multi-role-playing mode to diagnose multiple writing reviews or evaluations. The method includes: Step 1: During the meeting preparation stage, the large language model can be used through prompt word engineering to play the role of the medical conference host, receive the diagnosis results of the condition to be evaluated and the target medical record in which it is located. The target case can be understood as the full case text. The diagnostic role is generated based on key information such as the diagnosis list, department, personal information, etc. in the full case text, and the meeting outline, that is, the diagnostic dimensions of each diagnostic role, is determined. Among them, the diagnostic role can be understood as an expert role.

[0077] Step 2: In the expert analysis stage, the large language model can be used to play different expert roles through prompt word engineering. Each expert role conducts in-depth analysis based on the input diagnosis results of the disease to be evaluated and the target case according to the corresponding diagnostic dimension to obtain a diagnostic analysis report containing the key analysis points and conclusions of each diagnostic dimension.

[0078] Step 3: In the summary stage of the diagnostic analysis report, multiple diagnostic analysis reports can be input into the large language model, and the large language model can be used to analyze and extract the expert's analysis ideas and conclusions from the multiple diagnostic analysis reports to generate a summary evaluation result.

[0079] Step 4: During the multi-round voting stage, the summary evaluation results can be input into the large language model, and the voting information corresponding to each expert role can be generated through the prompt word engineering to give the conclusion and reasons of agreement / disagreement. If the voting results of multiple expert roles are inconsistent, the voting information corresponding to each expert role can be input into the large language model, and the voting briefing can be generated after re-summarizing through the host role, and the new voting results of each expert role can be output again through the large language model. Repeat multiple rounds of voting until the voting results of all expert roles are consistent or the specified number of votes is reached.

[0080] Step 5: Based on the final voting information, a comprehensive evaluation report is generated through the large language model. The comprehensive evaluation report includes the conclusion of overwriting / underwriting and analysis of multiple diagnostic dimensions.

[0081] The method for evaluating the diagnosis results of a disease state provided in an embodiment of the present invention obtains at least one diagnosis result of a disease state to be evaluated determined based on a target medical record, and for each diagnosis result of a disease state to be evaluated, inputs the diagnosis result of the disease state to be evaluated and the target medical record into a large language model to obtain at least two diagnostic roles output by the large language model and the diagnostic dimensions of each diagnostic role, and for each diagnostic role, determines first prompt information based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the diagnosis result of the disease state to be evaluated and the target medical record, and inputs the first prompt information into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model, and determines a target evaluation result of the diagnosis result of the disease state to be evaluated based on the diagnostic analysis report corresponding to each diagnostic role, and the target evaluation result is used to characterize whether the diagnosis result of the disease state to be evaluated is correct. Since the large language model can be used to determine different diagnostic roles and the diagnostic dimensions of each diagnostic role, and then the diagnostic analysis report of each diagnostic role based on the corresponding diagnostic dimensions can be determined based on the large language model, the final target evaluation result can be determined according to the diagnostic analysis reports of different diagnostic roles. The large language model has strong semantic understanding and processing capabilities, as well as the advantage of being able to tap into rich medical knowledge. Therefore, when using the large language model for disease diagnosis review, it is possible to avoid using fixed review logic and rule templates, thereby improving the generalization of the evaluation method. Moreover, by fully tapping into the useful information in the target case through the large language model, it can adapt to complex and changeable medical scenarios and improve the accuracy of the evaluation results.

[0082] The following is a description of the disease diagnosis result evaluation device provided by the present invention. The disease diagnosis result evaluation device described below and the disease diagnosis result evaluation method described above can be referenced to each other.

[0083] Figure 3 A schematic diagram of the structure of the device for evaluating the diagnosis result of a disease provided by an embodiment of the present invention, referring to Figure 3 As shown, the evaluation device 300 for disease diagnosis results includes: An acquisition module 11 is used to acquire at least one diagnosis result of a condition to be evaluated determined based on a target medical record; An input module 12 is used for inputting the diagnosis result of the condition to be evaluated and the target medical record into a large language model for each of the diagnosis results of the condition to be evaluated, so as to obtain at least two diagnostic roles output by the large language model and diagnostic dimensions of each of the diagnostic roles; A determination module 13 is used to determine, for each of the diagnostic roles, first prompt information based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the diagnostic result of the condition to be evaluated, and the target medical record; The input module 12 is further used to input the first prompt information into the large language model to obtain a diagnosis analysis report corresponding to the diagnosis role output by the large language model; The determination module 13 is further used to determine a target evaluation result of the diagnosis result of the condition to be evaluated based on the diagnosis analysis report corresponding to each of the diagnosis roles, and the target evaluation result is used to indicate whether the diagnosis result of the condition to be evaluated is correct.

[0084] In an exemplary embodiment, the determining module 13 is specifically configured to: Determine second prompt information based on each of the diagnostic roles, the diagnostic analysis report and the diagnostic dimension corresponding to each of the diagnostic roles; Inputting the second prompt information into the large language model to obtain a summary evaluation result output by the large language model, wherein the summary evaluation result includes a diagnosis analysis report corresponding to all the diagnosis roles; For each of the diagnosis roles, determining voting information of the diagnosis role for the diagnosis result of the condition to be evaluated based on the summary evaluation result, wherein the voting information includes the voting result of the diagnosis role for whether the diagnosis result of the condition to be evaluated is correct; Based on the voting information of each of the diagnostic roles, a target evaluation result of the disease diagnosis result to be evaluated is determined.

[0085] In an exemplary embodiment, the determining module 13 is specifically configured to: Determining third prompt information based on the summary evaluation result and the diagnosis role; The third prompt information is input into the large language model to obtain the voting information of the diagnostic role for the diagnosis result of the disease to be evaluated output by the large language model.

[0086] In an exemplary embodiment, the voting information also includes a reason for voting; The determining module 13 is specifically used for: In the case where there are inconsistent voting results among all the voting results, determining fourth prompt information based on the voting results and voting reasons corresponding to each of the diagnostic roles; Inputting the fourth prompt information into the large language model to obtain a summary voting result output by the large language model, wherein the summary voting result includes voting information corresponding to all the diagnosis roles; For each of the diagnosis roles, determining fifth prompt information based on the aggregated voting result and the diagnosis role; Inputting the fifth prompt information into the large language model to obtain new voting information of the diagnosis role output by the large language model, and repeating the above steps until all new voting results are consistent, or the number of votes reaches a preset number; Based on the final voting information, a target evaluation result of the disease diagnosis result to be evaluated is determined.

[0087] In an exemplary embodiment, the determining module 13 is specifically configured to: Determining sixth prompt information based on the final voting information and the diagnosis result of the condition to be evaluated; The sixth prompt information is input into the large language model to obtain a comprehensive evaluation report output by the large language model, wherein the comprehensive evaluation report includes the target evaluation results and diagnostic opinions, and the diagnostic opinions include the same diagnostic opinions and / or divergent diagnostic opinions of each of the diagnostic roles.

[0088] In an exemplary embodiment, the determination module 13 is further configured to determine the voting result as the target evaluation result of the disease diagnosis result to be evaluated when all the voting results are consistent.

[0089] In an exemplary embodiment, when more than a first preset number of the condition diagnosis results to be evaluated among all the condition diagnosis results to be evaluated belong to diseases corresponding to the same department, the at least two diagnostic roles include more than a second preset number of diagnostic roles belonging to the department.

[0090] The device for evaluating the disease diagnosis result of this embodiment can be used to execute the method of any embodiment in the embodiment for evaluating the disease diagnosis result. Its specific implementation process and technical effects are similar to those in the embodiment for evaluating the disease diagnosis result. For details, please refer to the detailed introduction in the embodiment for evaluating the disease diagnosis result, which will not be repeated here.

[0091] Figure 4 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention, such as Figure 4As shown, the electronic device may include: a processor (processor) 410 , a communication interface (Communications Interface) 420 , a memory (memory) 430 and a communication bus 440 , wherein the processor 410 , the communication interface 420 , and the memory 430 communicate with each other through the communication bus 440 . The processor 410 can call the logic instructions in the memory 430 to execute the evaluation method of the disease diagnosis result, which includes: obtaining at least one disease diagnosis result to be evaluated determined based on the target medical record; for each of the disease diagnosis results to be evaluated, inputting the disease diagnosis result to be evaluated and the target medical record into the large language model to obtain at least two diagnostic roles output by the large language model and the diagnostic dimensions of each diagnostic role; for each of the diagnostic roles, based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the disease diagnosis result to be evaluated and the target medical record, determining a first prompt information, and inputting the first prompt information into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model; based on the diagnostic analysis report corresponding to each of the diagnostic roles, determining a target evaluation result of the disease diagnosis result to be evaluated, and the target evaluation result is used to characterize whether the disease diagnosis result to be evaluated is correct.

[0092] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0093] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the evaluation method of the disease diagnosis results provided by the above methods, which method includes: obtaining at least one disease diagnosis result to be evaluated determined based on the target medical record; for each of the disease diagnosis results to be evaluated, inputting the disease diagnosis result to be evaluated and the target medical record into a large language model to obtain at least two diagnostic roles output by the large language model and the diagnostic dimensions of each diagnostic role; for each of the diagnostic roles, based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the disease diagnosis result to be evaluated and the target medical record, determining a first prompt information, and inputting the first prompt information into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model; based on the diagnostic analysis report corresponding to each of the diagnostic roles, determining a target evaluation result of the disease diagnosis result to be evaluated, and the target evaluation result is used to characterize whether the disease diagnosis result to be evaluated is correct.

[0094] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an evaluation method for executing the disease diagnosis results provided by the above-mentioned methods, the method comprising: obtaining at least one disease diagnosis result to be evaluated determined based on a target medical record; for each of the disease diagnosis results to be evaluated, inputting the disease diagnosis result to be evaluated and the target medical record into a large language model to obtain at least two diagnostic roles output by the large language model and diagnostic dimensions of each of the diagnostic roles; for each of the diagnostic roles, determining a first prompt message based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the disease diagnosis result to be evaluated and the target medical record, and inputting the first prompt message into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model; based on the diagnostic analysis report corresponding to each of the diagnostic roles, determining a target evaluation result of the disease diagnosis result to be evaluated, the target evaluation result being used to characterize whether the disease diagnosis result to be evaluated is correct.

[0095] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0096] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating a disease diagnosis result, characterized in that: include: Obtaining at least one diagnosis result of a condition to be evaluated determined based on the target medical record; For each of the to-be-evaluated disease diagnosis results, the to-be-evaluated disease diagnosis result and the target medical record are input into a large language model to obtain at least two diagnostic roles output by the large language model and diagnostic dimensions of each of the diagnostic roles; For each of the diagnostic roles, based on the diagnostic role, the diagnostic dimension corresponding to the diagnostic role, the diagnostic result of the condition to be evaluated and the target medical record, first prompt information is determined, and the first prompt information is input into the large language model to obtain a diagnostic analysis report corresponding to the diagnostic role output by the large language model; Based on the diagnostic analysis reports corresponding to the respective diagnostic roles, a target evaluation result of the diagnosis result of the condition to be evaluated is determined, and the target evaluation result is used to indicate whether the diagnosis result of the condition to be evaluated is correct.

2. The method for evaluating disease diagnosis results according to claim 1, characterized in that: Determining the target evaluation result of the condition diagnosis result to be evaluated based on the diagnosis analysis report corresponding to each of the diagnosis roles includes: Determine second prompt information based on each of the diagnostic roles, the diagnostic analysis report and the diagnostic dimension corresponding to each of the diagnostic roles; Inputting the second prompt information into the large language model to obtain a summary evaluation result output by the large language model, wherein the summary evaluation result includes a diagnosis analysis report corresponding to all the diagnosis roles; For each of the diagnosis roles, determining voting information of the diagnosis role for the diagnosis result of the condition to be evaluated based on the summary evaluation result, wherein the voting information includes the voting result of the diagnosis role for whether the diagnosis result of the condition to be evaluated is correct; Based on the voting information of each of the diagnostic roles, a target evaluation result of the disease diagnosis result to be evaluated is determined.

3. The method for evaluating disease diagnosis results according to claim 2, characterized in that: The step of determining the voting information of the diagnosis role for the diagnosis result of the condition to be evaluated based on the summary evaluation result includes: Determining third prompt information based on the summary evaluation result and the diagnosis role; The third prompt information is input into the large language model to obtain the voting information of the diagnostic role for the diagnosis result of the disease to be evaluated output by the large language model.

4. The method for evaluating disease diagnosis results according to claim 2, characterized in that: The voting information also includes the reason for voting; The step of determining the target evaluation result of the condition diagnosis result to be evaluated based on the voting information of each of the diagnosis roles includes: In the case where there are inconsistent voting results among all the voting results, determining fourth prompt information based on the voting results and voting reasons corresponding to each of the diagnostic roles; Inputting the fourth prompt information into the large language model to obtain a summary voting result output by the large language model, wherein the summary voting result includes voting information corresponding to all the diagnosis roles; For each of the diagnosis roles, determining fifth prompt information based on the aggregate voting result and the diagnosis role; Inputting the fifth prompt information into the large language model, obtaining new voting information of the diagnosis role output by the large language model, and repeating the above steps until all new voting results are consistent, or the number of votes reaches a preset number; Based on the final voting information, a target evaluation result of the disease diagnosis result to be evaluated is determined.

5. The method for evaluating disease diagnosis results according to claim 4, characterized in that: Determining the target evaluation result of the condition diagnosis result to be evaluated based on the final voting information includes: Determining sixth prompt information based on the final voting information and the diagnosis result of the condition to be evaluated; The sixth prompt information is input into the large language model to obtain a comprehensive evaluation report output by the large language model, wherein the comprehensive evaluation report includes the target evaluation results and diagnostic opinions, and the diagnostic opinions include the same diagnostic opinions and / or divergent diagnostic opinions of each of the diagnostic roles.

6. The method for evaluating disease diagnosis results according to claim 4, characterized in that: The method further comprises: When all the voting results are consistent, the voting result is determined as the target evaluation result of the disease diagnosis result to be evaluated.

7. The method for evaluating disease diagnosis results according to any one of claims 1 to 6, characterized in that: When more than a first preset number of the condition diagnosis results to be evaluated among all the condition diagnosis results to be evaluated belong to diseases corresponding to the same department, the at least two diagnosis roles include more than a second preset number of diagnosis roles belonging to the department.

8. A device for evaluating disease diagnosis results, characterized in that: include: An acquisition module, used to acquire at least one diagnosis result of a condition to be evaluated determined based on a target medical record; An input module, for inputting the diagnosis result of the condition to be evaluated and the target medical record into a large language model for each of the diagnosis results of the condition to be evaluated, to obtain at least two diagnostic roles output by the large language model and diagnostic dimensions of each of the diagnostic roles; A determination module, configured to determine, for each of the diagnosis roles, first prompt information based on the diagnosis role, the diagnosis dimension corresponding to the diagnosis role, the diagnosis result of the condition to be evaluated, and the target medical record; The input module is further used to input the first prompt information into the large language model to obtain a diagnosis analysis report corresponding to the diagnosis role output by the large language model; The determination module is further used to determine a target evaluation result of the diagnosis result of the condition to be evaluated based on the diagnosis analysis report corresponding to each of the diagnostic roles, and the target evaluation result is used to indicate whether the diagnosis result of the condition to be evaluated is correct.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method for evaluating the disease diagnosis result according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for evaluating the disease diagnosis result according to any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for evaluating the disease diagnosis result according to any one of claims 1 to 7 is implemented.

Images