LLM-based auxiliary diagnosis and treatment method, system, equipment and medium

Abstract

The invention relates to the technical field of auxiliary diagnosis and treatment, in particular to an LLM-based auxiliary diagnosis and treatment method, system and device and a medium, and the method comprises the steps: obtaining the illness state self-description of a user, and extracting user semantics through an LLM according to the user self-description; obtaining a semantic extraction result of the disease of the user, performing construction according to the semantic extraction result to obtain medical record information, and performing matching according to the medical record information to obtain an inspection suggestion; an LLM symptom extraction model is constructed, CNN model selection cues are obtained, an applicable CNN image model is selected according to the symptom possibility extracted by the LLM symptom extraction model and the CNN model selection cues, and results of the LLM symptom extraction model and the CNN image model are integrated to match a corresponding LLM question model; obtaining a question asking result corresponding to the examination suggestion, and inputting the question asking result into the LLM disease comprehensive judgment model to obtain disease related information; the first diagnosis and treatment opinion information is obtained through matching according to the medical record information and the disease possibility related information, accurate diagnosis and treatment opinions are given, and the diagnosis and treatment efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of auxiliary diagnosis and treatment technology, and in particular to an auxiliary diagnosis and treatment method, system, device and medium based on LLM. Background Art

[0002] With the development of science and technology, people have higher and higher requirements for the quality of life, and their awareness of the importance of health is also increasing day by day. Going to offline for registration, medical treatment and other processes takes up patients' time, and the corresponding doctors' time to serve each patient is also decreasing. The imbalance between supply and demand of medical resources leads to poor medical experience for patients.

[0003] In order to alleviate the imbalance between supply and demand of medical resources, language models are usually used to build auxiliary diagnosis and treatment systems, and pre-set language logic is used to communicate with patients, so as to complete the patient's interview and achieve auxiliary diagnosis and treatment. However, communicating with patients only through language logic cannot truly interpret the patient's examination results and give accurate diagnosis and treatment opinions. It has certain limitations in overall auxiliary medical care, and the above problems need to be solved. Summary of the invention

[0004] In order to provide accurate diagnosis and treatment advice and improve the efficiency of diagnosis and treatment, this application provides an auxiliary diagnosis and treatment method, system, device and medium based on LLM, which adopts the following technical solutions:

[0005] In a first aspect, the present application provides an auxiliary diagnosis and treatment method based on LLM, comprising:

[0006] Obtain the user's self-description of his or her condition, and extract the user's semantics based on the user's self-description through LLM;

[0007] Obtain the semantic extraction results of the user's self-reported disease, construct medical record information based on the semantic extraction results, and obtain examination suggestions based on the medical record information matching;

[0008] Construct an LLM symptom extraction model, obtain CNN model selection prompt words, select an applicable CNN image model based on the symptom possibility extracted by the LLM symptom extraction model and the CNN model selection prompt words, combine the results of the LLM symptom extraction model and the CNN image model, and match the corresponding LLM questioning model;

[0009] Obtain the follow-up question results corresponding to the examination suggestions, input the follow-up question results into the LLM disease comprehensive judgment model, and obtain relevant information about the disease;

[0010] Based on the medical record information and information related to the possibility of illness, the first diagnosis and treatment opinion information is matched.

[0011] Preferably, auxiliary information of disease symptoms is obtained, and follow-up analysis is performed on medical record information and information related to the possibility of illness based on the auxiliary information of disease symptoms and the LLM follow-up model to obtain follow-up question results.

[0012] Preferably, it also includes:

[0013] Obtain the follow-up question result information corresponding to the follow-up question information, perform a comprehensive diagnosis on the medical record information, disease possibility information and follow-up question results according to the LLM disease comprehensive judgment model, and obtain the second diagnosis and treatment opinion information.

[0014] Preferably, it also includes:

[0015] Acquire disease diagnosis auxiliary information, and the LLM disease comprehensive judgment model performs comprehensive diagnosis on medical history information, disease possibility information and questioning result information according to the disease diagnosis auxiliary information.

[0016] Preferably, the inspection data includes inspection numerical information and inspection image information.

[0017] Preferably, it also includes:

[0018] When the CNN model selects several prompt words, the medical record information is analyzed one by one with the several CNN model selected prompt words according to the LLM symptom extraction model to obtain several analysis sub-results.

[0019] In a second aspect, the present application provides an auxiliary diagnosis and treatment system based on LLM, comprising:

[0020] The condition acquisition module is used to obtain the user's self-description of the condition and extract the user semantics based on the user's self-description through LLM;

[0021] The examination suggestion acquisition module is used to obtain the semantic extraction results of the user's self-description of the disease, construct the medical record information based on the semantic extraction results, and obtain the examination suggestions based on the medical record information matching;

[0022] The model calling module is used to build the LLM symptom extraction model, obtain the CNN model selection prompt words, select the applicable CNN image model according to the symptom possibility extracted by the LLM symptom extraction model and the CNN model selection prompt words, and match the corresponding LLM questioning model by combining the results of the LLM symptom extraction model and the CNN image model;

[0023] The possibility analysis module is used to obtain the follow-up question results corresponding to the examination suggestions, and input the follow-up question results into the LLM disease comprehensive judgment model to obtain relevant information about the disease;

[0024] The preliminary diagnosis module is used to match the first diagnosis and treatment opinion information based on the medical record information and information related to the possibility of illness.

[0025] Preferably, it also includes:

[0026] The follow-up diagnosis module is used to obtain auxiliary information of disease symptoms, conduct follow-up analysis on medical record information and information related to the possibility of illness based on the auxiliary information of disease symptoms and the LLM follow-up model, and obtain the follow-up question results.

[0027] In a third aspect, the present application provides an LLM-based auxiliary diagnosis and treatment device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the LLM-based auxiliary diagnosis and treatment method as described above.

[0028] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the LLM-based auxiliary diagnosis and treatment method as described above when run.

[0029] In summary, compared with the prior art, the technical solution provided by this application has at least the following beneficial effects:

[0030] The present application obtains inquiry information after obtaining user identity information, thereby asking questions to the user and obtaining the user's reply information to the inquiry information. Several reply information are counted and constructed to obtain medical record information reflecting the user's condition. The medical record information is analyzed according to the LLM judgment model and CNN prompt words, and the corresponding suitable CNN question-answering model is matched according to the analysis results. Examination suggestions are obtained based on the matching of medical record information. The examination suggestions instruct the user to have an examination, thereby obtaining examination data corresponding to the examination suggestions. The CNN question-answering model judges the examination data and obtains disease possibility information. The LLM judgment model matches the medical record information obtained through pre-diagnosis with the disease possibility information obtained by the model, and obtains first diagnosis and treatment opinion information reflecting the user's condition, so that the auxiliary diagnosis and treatment system can give accurate diagnosis and treatment opinions, and improve diagnosis and treatment efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of an auxiliary diagnosis and treatment method based on LLM described in an embodiment of the present application.

[0032] Figure 2 It is a module diagram of an LLM-based auxiliary diagnosis and treatment system described in an embodiment of the present application.

[0033] Description of reference numerals:

[0034] 1. Condition acquisition module; 2. Examination suggestion acquisition module; 3. Model calling module; 4. Possibility analysis module; 5. Preliminary diagnosis module. DETAILED DESCRIPTION

[0035] The following combination Figure 1-Figure 2 The present application is described in further detail. The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to be limiting.

[0036] Reference Figure 1 The present application relates to an auxiliary diagnosis and treatment method based on LLM, which specifically includes:

[0037] Step S1: Obtain the user's self-description of his / her condition, and extract user semantics based on the user's self-description through LLM;

[0038] Step S2: obtaining semantic extraction results of the user's self-reported disease, constructing medical record information based on the semantic extraction results, and obtaining examination suggestions based on matching of the medical record information;

[0039] Step S3: construct an LLM symptom extraction model, obtain CNN model selection prompt words, select an applicable CNN image model according to the symptom possibility extracted by the LLM symptom extraction model and the CNN model selection prompt words, combine the results of the LLM symptom extraction model and the CNN image model, and match the corresponding LLM questioning model;

[0040] Step S4: Obtain the follow-up question results corresponding to the examination suggestions, input the follow-up question results into the LLM disease comprehensive judgment model, and obtain disease-related information;

[0041] Step S5: According to the medical record information and the information related to the possibility of illness, the first diagnosis and treatment opinion information is matched.

[0042] Specifically, the present application obtains inquiry information for asking questions to the user after obtaining the user's identity information, obtains the user's reply information to the inquiry information, and collects statistics of several user reply information to construct an initial diagnosis result reflecting the user's condition, that is, medical record information. The medical record information is analyzed according to the LLM judgment model and CNN prompt words, and then the corresponding suitable CNN question-and-answer model is matched according to the analysis results. Examination suggestions are obtained according to the medical record information matching, and the examination suggestions instruct the user to have an examination, so as to obtain the examination data corresponding to the examination suggestions. The CNN question-and-answer model judges the examination data and obtains information on the possibility of illness. The LLM judgment model matches the medical record information obtained through pre-diagnosis with the possibility of illness information obtained by the model to obtain the first diagnosis and treatment opinion information reflecting the user's condition, so that the auxiliary diagnosis and treatment system can give accurate diagnosis and treatment opinions and improve the diagnosis and treatment efficiency.

[0043] Among them, the user identity information is usually set as the user's identification ID or consultation ID in the auxiliary diagnosis and treatment system of this application. The above ID corresponds to the identity of the user. When the user identity information is obtained, the system of the embodiment of this application starts the pre-diagnosis of the user. When the user identity information is confirmed to be correct, the default inquiry information is usually set to "Excuse me, what is wrong with you?" as the first default question, and the user's reply information is obtained according to the preset inquiry information.

[0044] After executing the first sentence of the question, the inquiry information also includes corresponding questions about the chief complaint, current medical history, past history, surgical history, family history, allergy history, personal history, etc., so as to obtain the corresponding user response information. After obtaining the above-mentioned corresponding user response information, the medical record information is constructed according to a certain preset template based on the user response information as the preliminary consultation result.

[0045] As one implementation mode, it also includes:

[0046] When the CNN model selects several prompt words, the medical record information is analyzed one by one with the several CNN model selected prompt words according to the LLM symptom extraction model to obtain several analysis sub-results.

[0047] Specifically, traditional CNN models are mostly binary classification models for specific diseases. The answers given for untrained diseases have low credibility. If model selection is not performed, and the examination data is directly input into all models, it will not only waste computing power, but also easily lead to misdiagnosis when combining the results of multiple models. Therefore, it is necessary to prioritize and call the corresponding model based on the pre-diagnosis results. There are several CNN question-and-answer models in the embodiment of the present application, and each CNN question-and-answer model includes at least one CNN prompt word, that is, a small model calling keyword. After obtaining the medical record information, the CNN prompt words corresponding to several pre-built CNN question-and-answer models are obtained, and the LLM judgment model is used to determine whether the medical record information meets the CNN prompt.

[0048] One implementation method is that each small model, that is, the CNN question-answering model, will correspond to a key information extraction prompt word. LLM will determine whether to call the model through the combination of this prompt, medical record information, and image type. For example, if the CNN model is a model for senile cataracts, corresponding to prompt1: Please determine whether there is content related to "senile, blurred vision, and white fog when looking at things" in the <medical record information>. If so, please answer: Yes, do not generate redundant content and punctuation marks; corresponding to prompt2: Please determine whether there is content related to "cataract surgery" in the <medical record information>. If so, please answer: Yes, do not generate redundant content and punctuation marks; corresponding to prompt3: Please determine the operation time and surgical eye of "cataract surgery" in the <medical record information>. Please answer: Cataract operation time and surgical eye, do not generate redundant content and punctuation marks. Finally, the system determines which pictures will be input into the CNN model for classification based on the results of the three prompts and the image type label.

[0049] Another implementation method is, Prompt 1: Please determine whether there is any content related to "diabetes, blurred vision, fundus hemorrhage" in the <medical record information>, if so, please answer: yes, do not generate redundant content and punctuation marks; Prompt 2: Please determine whether there is any content related to "fundus examination" in the <medical record information>, if so, please answer: yes, do not generate redundant content and punctuation marks; Prompt 3: Please determine the specific results of "fundus examination" in the <medical record information>, please answer: fundus examination results, do not generate redundant content and punctuation marks. Medical record information combined with Prompt 1: The system determines whether there is a description of "diabetes, blurred vision, fundus hemorrhage" in the medical record information based on Prompt 1, if so, it outputs "yes". Medical record information combined with Prompt 2: The system continues to determine whether the medical record mentions "fundus examination" related content, if so, it outputs "yes". Medical record information combined with Prompt 3: The system extracts the specific results of fundus examination and outputs the corresponding description. Finally, the system determines which images will be input into the CNN model of diabetic retinopathy for classification based on the results of the three prompts and the image type labels.

[0050] Another implementation method is, Prompt 1: Please determine whether there is any content related to "lung nodules, dyspnea, and long-term cough" in the <medical record information>. If so, please answer: Yes. Do not generate redundant content and punctuation marks; Prompt 2: Please determine whether there is any content related to "lung CT examination" in the <medical record information>. If so, please answer: Yes. Do not generate redundant content and punctuation marks; Prompt 3: Please determine whether the results of "lung CT examination" in the <medical record information> mention "lung nodules". Please answer: The size and location of the lung nodules. Do not generate redundant content and punctuation marks; Medical record information is combined with Prompt 1: The system first determines whether the medical record mentions symptoms such as "lung nodules, dyspnea, and long-term cough". If so, it outputs "Yes". Medical record information is combined with Prompt 2: The system then determines whether there is a record of "lung CT examination". If so, it outputs "Yes". Medical record information is combined with Prompt 3: The system extracts the size and location of the lung nodules in the CT examination and outputs the corresponding information. Finally, the system determines whether to call the CNN model for lung nodule detection for classification based on the results of the three prompts and the image type, such as lung CT images.

[0051] Another implementation method is, Prompt 1: Please determine whether there is any content related to "stroke, hemiplegia, sudden headache" in the <medical record information>, if yes, please answer: yes, do not generate redundant content and punctuation marks; Prompt 2: Please determine whether there is any content related to "head CT / MRI examination" in the <medical record information>, if yes, please answer: yes, do not generate redundant content and punctuation marks; Prompt 3: Please determine whether the result of "head CT / MRI examination" in the <medical record information> mentions "stroke lesions", please answer: the size and location of the stroke lesions, do not generate redundant content and punctuation marks; medical record information is combined with Prompt 1: the system determines whether there are symptoms such as "stroke, hemiplegia, sudden headache" in the medical record information according to Prompt 1, if yes, then output "yes". Medical record information is combined with Prompt 2: the system then determines whether there is any record of "head CT / MRI examination" in the medical record, if yes, then output "yes". Medical record information is combined with Prompt 3: The system extracts the specific results of the head CT or MRI examination, determines whether there is a stroke lesion, and outputs the size and location of the lesion. Finally, the system determines whether to call the CNN model for stroke detection for classification based on the results of the three prompts and the image type, such as head CT or MRI images.

[0052] Another implementation method is, Prompt 1: Please determine whether there is any content related to "breast lumps, breast calcification, and breast pain" in the <medical record information>. If so, please answer: Yes, and do not generate redundant content and punctuation marks; Prompt 2: Please determine whether there is any content related to "breast ultrasound examination" in the <medical record information>. If so, please answer: Yes, and do not generate redundant content and punctuation marks; Prompt 3: Please determine whether the results of "breast ultrasound examination" in the <medical record information> mention "breast lumps or calcifications". Please answer: Breast ultrasound examination results, and do not generate redundant content and punctuation marks; Medical record information is combined with Prompt 1: The system determines whether the description of "breast lumps, breast calcification, and breast pain" is mentioned in the medical record based on Prompt 1. If so, it outputs "Yes". Medical record information is combined with Prompt 2: The system continues to determine whether the medical record mentions the record of "breast ultrasound examination". If so, it outputs "Yes". Medical record information is combined with Prompt 3: The system extracts the specific results of breast ultrasound examination, determines whether there is a mass or calcification, and outputs the corresponding results. Finally, the system determines whether to call the CNN model for breast tumor screening for classification based on the results of the three prompts and the image type, such as breast ultrasound images.

[0053] Through the above embodiment, key information is extracted according to the prompt, and finally a decision is made whether to call the corresponding CNN model to perform image classification.

[0054] As one implementation manner, the inspection data includes inspection value information and inspection image information.

[0055] Specifically, the embodiment of the present application obtains examination suggestions based on matching of medical record information. After obtaining the examination suggestions, the user performs item examinations according to the examination suggestions. The obtained examination data includes examination values ​​and examination images. The examination data is bound to the user's identity information. The examination data, i.e., the values ​​and images are input into a small CNN model of the corresponding disease to determine the disease category and give the possibility of disease.

[0056] After the initial generation of medical records, the system will recommend corresponding examination items based on the keywords of the symptoms in the medical records. For example, if the eyes are dull and painful for three days, the intraocular pressure test, distorted vision, etc., it is judged that a fundus color photo is needed. The user performs an examination before diagnosis in order to further make an accurate diagnosis and treatment.

[0057] As one implementation mode, it also includes:

[0058] Obtain auxiliary information of disease symptoms, conduct follow-up analysis on medical record information and information related to the possibility of illness based on the auxiliary information of disease symptoms and the LLM follow-up model, and obtain the follow-up question results.

[0059] Specifically, after the user performs an examination and selects the corresponding CNN question-answering model, the system of the embodiment of the present application uses the selected CNN question-answering model to obtain disease possibility information, uses the LLM judgment model to match the medical record information and disease possibility information from the disease symptom knowledge base, obtains the corresponding first diagnosis and treatment opinion information in the disease symptom auxiliary information, preliminarily generates the first diagnosis and treatment opinion information, and generates follow-up questions that require further questioning or secondary inquiry to confirm the condition.

[0060] The types of diseases in the disease symptom knowledge base will correspond to the questions that need to be asked. The system will then compare the questions asked in the previous pre-consultation to see if they have asked all the questions that need to be asked for the disease diagnosis. If not, the system will continue to ask questions. If yes, it will proceed to ask questions for the next disease diagnosis.

[0061] As one implementation mode, it also includes:

[0062] Obtain the follow-up question result information corresponding to the follow-up question information, perform a comprehensive diagnosis on the medical record information, disease possibility information and follow-up question results according to the LLM disease comprehensive judgment model, and obtain the second diagnosis and treatment opinion information.

[0063] Acquire disease diagnosis auxiliary information, and the LLM disease comprehensive judgment model performs comprehensive diagnosis on medical history information, disease possibility information and questioning result information according to the disease diagnosis auxiliary information.

[0064] Specifically, the embodiment of the present application is an LLM assisted by a disease diagnosis knowledge base, which gives users a comprehensive diagnosis and provides treatment advice based on the results obtained from the pre-questioning clinic, the results of the small model judgment, and the results of further questioning. It combines the results of the small model with the pre-questioning results to build a follow-up diagnosis module based on LLM comprehensive analysis, generates questions that need to be further asked, and extracts relevant information from the patient's answers to form auxiliary diagnosis results and treatment opinions. The embodiment of the present application intends to complete the entire process of auxiliary diagnosis and treatment, namely: consultation, examination, questioning, diagnosis, and provide patients and doctors with AI assistance throughout the diagnosis and treatment cycle, which can give accurate diagnosis and treatment advice and improve diagnosis and treatment efficiency.

[0065] Reference Figure 2 , an auxiliary diagnosis and treatment system based on LLM is provided for an embodiment of the present application, the system comprising:

[0066] The condition acquisition module 1 is used to obtain the user's self-description of the condition and extract the user semantics based on the user's self-description through LLM;

[0067] Examination suggestion acquisition module 2 is used to obtain the semantic extraction results of the user's self-description of the disease, construct the medical record information according to the semantic extraction results, and obtain the examination suggestions according to the medical record information matching;

[0068] Model calling module 3 is used to build an LLM symptom extraction model, obtain CNN model selection prompt words, select an applicable CNN image model according to the symptom possibility extracted by the LLM symptom extraction model and the CNN model selection prompt words, and match the corresponding LLM questioning model by combining the results of the LLM symptom extraction model and the CNN image model;

[0069] Possibility analysis module 4 is used to obtain the follow-up question results corresponding to the examination suggestions, and input the follow-up question results into the LLM disease comprehensive judgment model to obtain relevant information about the disease;

[0070] The preliminary diagnosis module 5 is used to match the first diagnosis and treatment opinion information based on the medical record information and the information related to the possibility of illness.

[0071] As one of the implementation modes, the follow-up diagnosis module is used to obtain auxiliary information of disease symptoms, conduct follow-up analysis on medical record information and information related to the possibility of illness based on the auxiliary information of disease symptoms and the LLM follow-up model, and obtain the follow-up question results.

[0072] As one of the implementation modes, the follow-up diagnosis module is also used to obtain follow-up question information corresponding to the follow-up question information, and perform a comprehensive diagnosis on the medical record information, the disease possibility information and the follow-up question information according to the LLM judgment model to obtain the second diagnosis and treatment opinion information.

[0073] As one of the implementation modes, the follow-up diagnosis module is also used to obtain disease diagnosis auxiliary information, and the LLM judgment model performs a comprehensive diagnosis on the medical history information, disease possibility information and follow-up questioning result information according to the disease diagnosis auxiliary information.

[0074] An embodiment of the present application provides an auxiliary diagnosis and treatment device based on LLM, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the auxiliary diagnosis and treatment method based on LLM as described above.

[0075] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the aforementioned LLM-based auxiliary diagnosis and treatment method when running.

[0076] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and product can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0077] In several embodiments provided in this application, it should be understood that the disclosed methods, systems, devices and program products may be implemented in other ways.

[0078] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0079] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application 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 application.

Claims

1. An auxiliary diagnosis and treatment method based on LLM, characterized in that: include: Obtain the user's self-description of his or her condition, and extract the user's semantics based on the user's self-description through LLM; Obtain the semantic extraction results of the user's self-reported disease, construct medical record information based on the semantic extraction results, and obtain examination suggestions based on the medical record information matching; Construct an LLM symptom extraction model, obtain CNN model selection prompt words, select an applicable CNN image model based on the symptom possibility extracted by the LLM symptom extraction model and the CNN model selection prompt words, combine the results of the LLM symptom extraction model and the CNN image model, and match the corresponding LLM questioning model; Obtain the follow-up question results corresponding to the examination suggestions, input the follow-up question results into the LLM disease comprehensive judgment model, and obtain relevant information about the disease; Based on the medical record information and information related to the possibility of illness, the first diagnosis and treatment opinion information is matched.

2. The auxiliary diagnosis and treatment method based on LLM according to claim 1, characterized in that: Also includes: Obtain auxiliary information of disease symptoms, conduct follow-up analysis on medical record information and information related to the possibility of illness based on the auxiliary information of disease symptoms and the LLM follow-up model, and obtain the follow-up question results.

3. The auxiliary diagnosis and treatment method based on LLM according to claim 2, characterized in that: Also includes: Obtain the follow-up question result information corresponding to the follow-up question information, perform a comprehensive diagnosis on the medical record information, disease possibility information and follow-up question results according to the LLM disease comprehensive judgment model, and obtain the second diagnosis and treatment opinion information.

4. The auxiliary diagnosis and treatment method based on LLM according to claim 3, characterized in that: Also includes: Acquire disease diagnosis auxiliary information, and the LLM disease comprehensive judgment model performs comprehensive diagnosis on medical history information, disease possibility information and questioning result information according to the disease diagnosis auxiliary information.

5. The auxiliary diagnosis and treatment method based on LLM according to claim 1, characterized in that: The inspection data includes inspection value information and inspection image information.

6. The auxiliary diagnosis and treatment method based on LLM according to claim 1, characterized in that: Also includes: When the CNN model selects several prompt words, the medical record information is analyzed one by one with the several CNN model selected prompt words according to the LLM symptom extraction model to obtain several analysis sub-results.

7. An auxiliary diagnosis and treatment system based on LLM, characterized in that: include: The condition acquisition module is used to obtain the user's self-description of the condition and extract the user semantics based on the user's self-description through LLM; The examination suggestion acquisition module is used to obtain the semantic extraction results of the user's self-description of the disease, construct the medical record information based on the semantic extraction results, and obtain the examination suggestions based on the medical record information matching; The model calling module is used to build the LLM symptom extraction model, obtain the CNN model selection prompt words, select the applicable CNN image model according to the symptom possibility extracted by the LLM symptom extraction model and the CNN model selection prompt words, and match the corresponding LLM questioning model by combining the results of the LLM symptom extraction model and the CNN image model; The possibility analysis module is used to obtain the follow-up question results corresponding to the examination suggestions, and input the follow-up question results into the LLM disease comprehensive judgment model to obtain relevant information about the disease; The preliminary diagnosis module is used to match the first diagnosis and treatment opinion information based on the medical record information and information related to the possibility of illness.

8. The LLM-based auxiliary diagnosis and treatment system according to claim 7, characterized in that: Also includes: The follow-up diagnosis module is used to obtain auxiliary information of disease symptoms, conduct follow-up analysis on medical record information and information related to the possibility of illness based on the auxiliary information of disease symptoms and the LLM follow-up model, and obtain the follow-up question results.

9. An auxiliary diagnosis and treatment device based on LLM, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the LLM-based auxiliary diagnosis and treatment method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the LLM-based auxiliary diagnosis and treatment method according to any one of claims 1 to 6 when running.

Images