Auxiliary diagnosis model training method, disease symptom collection method and equipment

Abstract

The invention provides a training method of an auxiliary diagnosis model and a disease symptom collection method and device, and relates to the technical field of medical data processing.The method comprises the steps that a plurality of sample data sets of preset diseases are obtained, randomly generating a simulated initial symptom description statement corresponding to each piece of sample data according to the characteristics of multiple symptoms of one patient in each piece of sample data in the sample data set of the preset disease by adopting a patient simulation large model, and carrying out initial diagnosis on each piece of sample data by adopting an initial auxiliary diagnosis model according to the simulated initial symptom description statement and each piece of sample data, and generating a simulated disease symptom corresponding to each piece of sample data, and training the initial auxiliary diagnosis model according to the simulated disease symptom and the preliminary diagnosis information to obtain a target auxiliary diagnosis model. According to the invention, the application prospect of the auxiliary diagnosis model is expanded, so that the disease symptoms of the target patient can be collected more comprehensively and fully.

Description

Technical Field

[0001] The present application relates to the technical field of medical data processing, and in particular to a training method for an auxiliary diagnosis model, a disease symptom collection method and equipment. Background Art

[0002] Medical artificial intelligence is a field that has attracted much attention in recent years. Among them, artificial intelligence-assisted disease diagnosis is a long-awaited goal. The development of large-scale language models has brought unprecedented opportunities for artificial intelligence-based medical auxiliary diagnosis.

[0003] At present, before the big language model used in medical auxiliary diagnosis can provide auxiliary diagnosis for patients, it must ensure that the big language model receives comprehensive information about the patient, including but not limited to personal medical history, examination reports, living habits, etc., so that the model can provide auxiliary diagnosis for the target patient based on these detailed data and the patient's complete information.

[0004] However, in actual diagnosis scenarios, doctors initially know nothing about the patient's condition, and patients are unable to fully describe their condition. If a large language model requires doctors to collect all important information in advance to make a diagnosis, then the auxiliary diagnosis function of the large language model cannot be fully utilized. Summary of the invention

[0005] The purpose of this application is to provide a training method for an auxiliary diagnosis model, a disease symptom collection method and equipment to address the deficiencies in the above-mentioned prior art, thereby expanding the application prospects of the auxiliary diagnosis model and making the collection of disease symptoms of target patients more comprehensive and sufficient.

[0006] To achieve the above purpose, the technical solution adopted in the embodiment of the present application is as follows: In a first aspect, an embodiment of the present application provides a method for training an auxiliary diagnosis model, the method comprising: Acquire a plurality of sample data sets of preset diseases, wherein each sample data in each sample data set of the preset disease includes: characteristics of a plurality of symptoms of a patient, associated medical history characteristics, and preliminary diagnosis information for each preset disease; Using a large patient simulation model, randomly generating a simulated initial symptom description statement corresponding to each sample data according to the characteristics of the multiple symptoms; Using the initial auxiliary diagnosis model, according to the simulated initial symptom description statement and each sample data, generate a simulated disease symptom corresponding to each sample data; The initial auxiliary diagnosis model is trained according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain a target auxiliary diagnosis model.

[0007] In a second aspect, another embodiment of the present application provides a disease symptom collection method, which is applied to a target-assisted diagnosis model obtained by training using any of the training methods described in the first aspect above, wherein the target-assisted diagnosis model includes: a doctor simulation model, a target disease screening model, and a target disease symptom collection model; the disease symptom collection method includes: Obtaining an inputted initial symptom description sentence for a target patient; Using the target disease screening model, generating initial description symptoms and symptoms to be confirmed associated with the initial description symptoms according to the initial symptom description statement; Using the doctor simulation large model, generating a first inquiry question for the symptom to be confirmed according to the symptom to be confirmed; Obtaining the input feedback status of the target patient regarding the symptom to be confirmed; Using the target disease screening model, generating initial disease diagnosis information according to the initial described symptoms and the feedback status of the symptoms to be confirmed; Using the target disease symptom collection model, generating the collected disease symptoms of the target patient according to the preset medical literature records corresponding to the initial disease diagnosis information; The doctor simulation model is used to generate a disease symptom description statement for the target patient based on the target auxiliary disease diagnosis information.

[0008] In a third aspect, another embodiment of the present application provides a training device for an auxiliary diagnosis model, the device comprising: An acquisition module, used for acquiring sample data sets of multiple preset diseases, each sample data in the sample data set of each preset disease includes: characteristics of multiple symptoms of a patient, associated medical history characteristics and preliminary diagnosis information for each preset disease; A first generating module is used to randomly generate a simulated initial symptom description statement corresponding to each sample data by using a patient simulation large model according to the characteristics of the multiple symptoms; A second generating module is used to generate a simulated disease symptom corresponding to each sample data according to the simulated initial symptom description statement and each sample data by using the initial auxiliary diagnosis model; The training module is used to train the initial auxiliary diagnosis model according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain a target auxiliary diagnosis model.

[0009] In the fourth aspect, another embodiment of the present application provides a computer device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, when the computer device is running, the processor and the memory communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the method as described in any one of the first and second aspects above.

[0010] In a fifth aspect, another embodiment of the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in any one of the first and second aspects are executed.

[0011] The beneficial effects of this application are: The present application provides a method for training an auxiliary diagnosis model, a method for collecting disease symptoms, and a device, by obtaining a sample data set of multiple preset diseases, using a large patient simulation model to randomly generate a simulated initial symptom description sentence corresponding to each sample data according to the characteristics of multiple symptoms of a patient in each sample data of the preset disease sample data set, using an initial auxiliary diagnosis model to generate simulated disease symptoms corresponding to each sample data according to the simulated initial symptom description sentence and each sample data, and training the initial auxiliary diagnosis model according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain a target auxiliary diagnosis model. The target auxiliary diagnosis model trained by the auxiliary diagnosis model training method in the present application can more realistically simulate the real conversation between the patient and the doctor when visiting the doctor, and train the information collection ability of the auxiliary diagnosis model, so that the auxiliary diagnosis model does not need to collect the disease-related information of the target patient in advance, and only gradually collects the patient's disease-related information by simulating the real conversation between the patient and the doctor when visiting the doctor, thereby improving the application prospects of the auxiliary diagnosis model and making the disease symptoms of the target patient more comprehensive and sufficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 A flowchart of a method for training an auxiliary diagnosis model provided in an embodiment of the present application; Figure 2 A schematic diagram of a process for determining simulated disease symptoms in a training method for an auxiliary diagnosis model provided in an embodiment of the present application; Figure 3An interactive schematic diagram of an auxiliary diagnosis model provided in an embodiment of the present application; Figure 4 A schematic diagram of a process for determining initial screening disease information in a training method for an auxiliary diagnosis model provided in an embodiment of the present application; Figure 5 A schematic diagram of a process for determining symptoms to be confirmed in a training method for an auxiliary diagnosis model provided in an embodiment of the present application; Figure 6 A schematic diagram of a process for determining initial screening disease information in another auxiliary diagnosis model training method provided in an embodiment of the present application; Figure 7 A schematic diagram of the structure of an auxiliary diagnosis model in an auxiliary diagnosis model training method provided in an embodiment of the present application; Figure 8 A schematic diagram of a process for determining a target disease screening model in a training method for an auxiliary diagnosis model provided in an embodiment of the present application; Fig. 9 A schematic diagram of a process for determining simulated disease symptoms in another method for training an auxiliary diagnosis model provided in an embodiment of the present application; Fig.10 A schematic diagram of the structure of a diagnostic decision-making program in an auxiliary diagnosis model training method provided in an embodiment of the present application; Fig.11 A process for determining a sample data set in a training method for an auxiliary diagnosis model provided in an embodiment of the present application; Fig.12 A flowchart of a disease symptom collection method provided in an embodiment of the present application; Fig.13 A schematic diagram of the structure of a training device for an auxiliary diagnosis model provided in an embodiment of the present application; Fig.14 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0014] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0015] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0016] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0017] At present, the existing auxiliary diagnosis model cannot collect the patient's disease data according to the patient's symptom description statement in the medical scenario, so as to perform auxiliary diagnosis on the target patient according to the collected results. The existing auxiliary diagnosis model needs to obtain the complete information of the target patient before it can realize auxiliary diagnosis of the target patient. However, if the doctor is required to collect all important information in advance to make a diagnosis, the doctor usually already knows what disease the patient has in the process of collecting information, so the auxiliary diagnosis model cannot fully reflect its role. In other words, the application scenarios of the existing auxiliary diagnosis model are relatively limited and cannot fully simulate the actual scenario of the patient's visit. To this end, the present application provides a training method for an auxiliary diagnosis model, by obtaining a sample data set of multiple preset diseases, and using a patient simulation large model to randomly generate a simulated initial symptom description statement corresponding to each sample data according to the characteristics of multiple symptoms, using an initial auxiliary diagnosis model, according to the model initial symptom description statement and each sample data, generate a simulated disease symptom corresponding to each sample data, and train the initial auxiliary diagnosis model according to the simulated disease symptoms and multiple symptoms corresponding to each preset disease to obtain a target auxiliary diagnosis model. The target auxiliary diagnosis model in this application can collect the patient's disease data according to the patient's initial symptom description statement when performing auxiliary diagnosis on the patient based on the full simulation of the doctor's role, so as to obtain the collected disease symptoms of the target patient, improve the utilization efficiency of the auxiliary diagnosis model, and ensure the comprehensiveness and adequacy of the collection of disease symptoms of the target patient. It is worth noting that the auxiliary diagnosis model in this application is only used for auxiliary diagnosis of patients. Specifically, it can be applied to the reception desk of a preset medical institution to triage patients, or to perform auxiliary diagnosis on patients through an application of a preset medical structure. The actual patient diagnosis results need to be judged by the actual doctor based on the patient's actual examination data.

[0018] The training method of the auxiliary diagnosis model provided in the embodiment of the present application is described below with reference to a plurality of drawings. Figure 1 A flowchart of a training method for an auxiliary diagnosis model provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes: Step 101: Obtain sample data sets of multiple preset diseases.

[0019] Each sample data in the sample data set of each preset disease includes: characteristics of multiple symptoms of a patient, related medical history characteristics and preliminary diagnosis information for each preset disease.

[0020] The features of multiple symptoms are binary vectors of multiple symptoms, and the symptoms may be headache, abdominal pain, nausea, vomiting, etc. The associated medical history feature may be a vector of fixed length, and the associated medical history may be medical history, social history, and family history. The preliminary diagnosis information of each preset disease is the patient's diagnosis result, which is used to indicate whether it is the preset disease.

[0021] Optionally, sample data sets of multiple preset diseases can be obtained from a medical data set including multiple preset diseases; or a sample data set of each preset disease can be obtained from the medical data set of each preset disease, and sample data sets of multiple preset diseases can be determined based on the sample data set of each preset disease.

[0022] Step 102: using a large patient simulation model, randomly generate a simulated initial symptom description statement corresponding to each sample data according to the characteristics of multiple symptoms.

[0023] Among them, the patient simulation large model is a patient simulation large model built based on the large language model (LLM), which is used to simulate patients in the training process of the auxiliary diagnosis model. The patient simulation large model can convert symptom features into corresponding symptom description sentences and understand the acquired natural language text to obtain corresponding features. Among them, the symptom description sentence is a natural language text.

[0024] Optionally, a patient simulation model is used to randomly determine a symptom feature from multiple symptom features, and a simulated initial symptom description statement corresponding to each sample data is generated based on the symptom feature. In this application, by randomly determining a symptom feature, it can be ensured that the patient's medical treatment scenario is fully simulated. In the actual medical treatment process, the patient will randomly select a symptom for medical treatment based on his or her multiple symptoms.

[0025] Step 103: using the initial auxiliary diagnosis model, based on the simulated initial symptom description statement and each sample data, generate a simulated disease symptom corresponding to each sample data.

[0026] Among them, the simulated disease symptoms are simulated disease-related symptoms of the target patient obtained by simulating the initial auxiliary diagnosis model, such as abdominal pain, diarrhea, fever and other symptoms, or more detailed disease symptoms with relevant indicator data, which is not limited in this embodiment of the present application.

[0027] Optionally, the initial auxiliary diagnosis model generates corresponding symptom features based on the simulated initial symptom description statement, and generates simulated disease symptoms corresponding to each sample data based on the symptom features and the associated medical history features in each sample data.

[0028] Step 104: train the initial auxiliary diagnosis model according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain a target auxiliary diagnosis model.

[0029] Optionally, according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms, the loss value of the initial auxiliary diagnosis model is calculated, and the initial auxiliary diagnosis model is optimized according to the loss value to obtain the target auxiliary diagnosis model. The specific optimization may be to adjust the structure of the initial auxiliary diagnosis model, adjust the model parameters, etc., which is not limited in the embodiments of the present application.

[0030] Optionally, the patient's predicted disease information can also be determined based on the simulated disease symptoms, the loss value of the initial auxiliary diagnosis model diagnosis can be determined based on the predicted disease information and the preliminary diagnosis information, and the initial auxiliary diagnosis model can be optimized based on the loss value of the initial auxiliary diagnosis model diagnosis to obtain the target auxiliary diagnosis model. Among them, the specific method for obtaining the predicted disease information can be determined by a preset large language model, or by other determination methods, and the embodiments of the present application do not limit this.

[0031] The present application provides a method for training an auxiliary diagnosis model, by obtaining a sample data set of multiple preset diseases, using a large patient simulation model to randomly generate a simulated initial symptom description sentence corresponding to each sample data according to the characteristics of multiple symptoms of a patient in each sample data of the preset disease sample data set, using an initial auxiliary diagnosis model to generate a simulated disease symptom corresponding to each sample data according to the simulated initial symptom description sentence and each sample data, and training the initial auxiliary diagnosis model according to the simulated disease symptoms and preliminary diagnosis information to obtain a target auxiliary diagnosis model. The target auxiliary diagnosis model trained by the training method of the auxiliary diagnosis model in the present application can more realistically simulate the real conversation between the patient and the doctor when visiting the doctor, and train the information collection ability of the auxiliary diagnosis model, so that the auxiliary diagnosis model does not need to collect the disease-related information of the target patient in advance, and only gradually collects the patient's disease-related information by simulating the real conversation between the patient and the doctor when visiting the doctor, thereby improving the application prospects of the auxiliary diagnosis model and making the simulated disease symptoms of the target patient more comprehensive and accurate.

[0032] Based on the above embodiment, the initial auxiliary diagnosis model includes: a large doctor simulation model, an initial disease screening model, and an initial disease symptom collection model. To this end, the present application also provides a process for determining simulated disease symptoms in a training method for an auxiliary diagnosis model. Figure 2 A schematic diagram of a process for determining simulated disease symptoms in a training method for an auxiliary diagnosis model provided in an embodiment of the present application, such as Figure 2As shown, in the above step 103, the initial auxiliary diagnosis model is used to generate simulated disease symptoms corresponding to each sample data according to the simulated initial symptom description statement and each sample data, including: Step 201: Use the doctor simulation large model to generate simulated symptom characteristics corresponding to each sample data based on the simulated initial symptom description statement.

[0033] The doctor simulation big model is a doctor simulation big model built based on the Large Language Model (LLM), which is used to simulate the role of doctors in the process of training the auxiliary diagnosis model. The doctor simulation big model can understand the simulated initial symptom description sentence to obtain the simulated description symptom characteristics. The specific doctor simulation big model can be the GPT-4 Turbo 04-09 model or the Llama3-70B-Instruct model.

[0034] Optionally, a doctor simulation model is used to perform natural language understanding on the simulated initial symptom description sentence, thereby generating simulated description symptom features corresponding to each sample data. The simulated initial symptom description sentence is a simulated initial symptom description sentence randomly generated by the patient simulation model according to the characteristics of multiple symptoms of the patient for each sample data.

[0035] Step 202: using the initial disease screening model, based on the simulated description of symptom characteristics, determine the symptoms to be confirmed corresponding to each sample data from the symptoms of multiple preset diseases.

[0036] Among them, the preset disease screening model is a model for disease screening built based on a neural network model.

[0037] Optionally, the initial disease screening model is used to determine the simulated description symptoms according to the simulated description symptom characteristics, and other characteristics of the symptoms of multiple preset diseases other than the simulated description symptoms are determined from the symptoms of multiple preset diseases. And according to other characteristics of the symptoms of multiple preset diseases other than the simulated description symptoms, the symptoms to be confirmed corresponding to each sample data are determined. The symptoms to be confirmed may be symptoms completely different from the simulated description symptoms, or symptoms that are more detailed descriptions of the simulated description symptoms. For example, the simulated description symptom is abdominal pain, and the symptoms to be confirmed may be headache or abdominal pain for three days.

[0038] Step 203: Use the doctor simulation large model to generate inquiry questions about the symptoms to be confirmed based on the symptoms to be confirmed.

[0039] The number of symptoms to be confirmed may be one or more, and this embodiment of the present application does not impose any limitation on this.

[0040] Optionally, a doctor simulation model is used to understand the symptoms to be confirmed and generate inquiry questions for the symptoms to be confirmed, wherein the inquiry questions are natural language texts.

[0041] Step 204: Using a large patient simulation model, a feedback status of the symptoms to be confirmed is generated according to the inquiry questions and the characteristics of the multiple symptoms.

[0042] Optionally, a patient simulation large model is used to perform natural language understanding on the inquiry question to obtain the unconfirmed symptom corresponding to the inquiry question, and a feedback status of the unconfirmed symptom is generated according to the unconfirmed symptom corresponding to the inquiry question and the characteristics of multiple symptoms of a patient in the sample data set. The feedback status is used to indicate whether the patient has the unconfirmed symptom corresponding to the inquiry question.

[0043] Step 205: using the initial disease screening model, based on the simulated description of symptom characteristics and the feedback status of the symptoms to be confirmed, generate initial screening disease information corresponding to each sample data.

[0044] The initial screening disease information may be one or more, and this embodiment of the present application does not impose any limitation on this.

[0045] Step 206: using the initial disease symptom collection model, based on the initial screening disease information and each sample data, generate a simulated disease symptom corresponding to each sample data.

[0046] For example, Figure 3 An interactive schematic diagram of an auxiliary diagnosis model provided in an embodiment of the present application, such as Figure 3 As shown, the initial auxiliary diagnosis model includes: a doctor simulation model, an initial disease screening model and an initial disease symptom collection model. The doctor simulation model receives the simulated initial symptom description statement of the patient simulation model and generates a simulated description of the symptom characteristics. The initial disease screening model is used to determine the symptoms to be confirmed corresponding to each sample data from the symptoms of multiple preset diseases based on the simulated description of the symptom characteristics. The doctor simulation model is used to generate inquiry questions for the symptoms to be confirmed based on the symptoms to be confirmed. The patient simulation model generates feedback status of the symptoms to be confirmed based on the inquiry questions and the characteristics of multiple symptoms. The initial disease screening model is used to generate the initial screening disease information corresponding to each sample data based on the simulated description of the symptom characteristics and the feedback status of the symptoms to be confirmed. The initial disease symptom collection model is used to generate the simulated disease symptoms corresponding to each sample data based on the initial screening disease information and each sample data.

[0047] In the above step 104, the initial auxiliary diagnosis model is trained according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain a target auxiliary diagnosis model, including: Step 207: train the initial disease screening model in the initial auxiliary diagnosis model according to the initial screening disease information and the preliminary diagnosis information in each sample data to obtain a target disease screening model.

[0048] Optionally, based on the initial screening disease information and the preliminary diagnosis information in each sample data, the loss value of the initial disease screening model obtained by the initial disease screening model in the initial auxiliary diagnosis model is determined, and the initial disease screening model is optimized according to the loss value of the initial disease screening model to obtain the target disease screening model.

[0049] Step 208: Train the initial disease symptom collection model in the initial auxiliary diagnosis model according to the simulated disease symptoms and multiple symptoms corresponding to each preset disease to obtain a target disease symptom collection model.

[0050] Among them, the target assisted diagnosis model includes: doctor simulation large model, target disease screening model and target disease symptom collection model.

[0051] Optionally, the actual symptoms of the simulated disease are determined from multiple symptoms corresponding to each preset disease based on the simulated disease symptoms, the accuracy of the simulated disease symptoms obtained by the initial disease symptom collection model in the initial auxiliary diagnosis model is determined based on the simulated disease symptoms and the actual symptoms of the simulated disease, and the initial disease symptom collection model is supervised by a supervised learning method to obtain a target disease screening model.

[0052] In the embodiment of the present application, the patient simulation large model interacts with the initial auxiliary diagnosis model to simulate the real dialogue between the doctor and the patient, thereby determining the initial screening disease information and simulated disease symptoms based on the simulated initial symptom description statement, and determining the target disease screening model and the target disease symptom collection model based on the initial screening disease information, simulated disease symptoms and each sample data. The disease screening model can be further optimized and customized to improve the ability of the disease screening model to extract information and the accuracy of the initial screening disease information, while improving the comprehensiveness and accuracy of the disease symptom collection model for the simulated disease symptoms.

[0053] Based on the above embodiments, the present application also provides a process for determining initial screening disease information in a training method for an auxiliary diagnosis model. Figure 4 A schematic diagram of a process for determining initial screening disease information in a training method for an auxiliary diagnosis model provided in an embodiment of the present application, such as Figure 4 As shown, in the above step 205, the initial disease screening model is used to generate initial screening disease information corresponding to each sample data according to the feedback status of the simulated description of symptom characteristics and the symptoms to be confirmed, including: Step 401, using the initial disease screening model, based on the simulated description of symptom characteristics and the feedback status of the symptoms to be confirmed, re-determine the new symptoms to be confirmed corresponding to each sample data from the symptoms of multiple preset diseases, and repeat the execution until the new feedback status of the symptoms to be confirmed is obtained.

[0054] For example, an initial disease screening model is used, and according to the simulated description of symptom characteristics and the feedback status of the symptoms to be confirmed, a new symptom to be confirmed corresponding to each sample data is re-determined from the symptoms of multiple preset diseases, and the feedback status of the new state to be confirmed is obtained, and then the next new symptom to be confirmed is re-confirmed according to the simulated description of symptom characteristics, the feedback status of the symptoms to be confirmed, and the feedback status of the new state to be confirmed. In other words, the symptoms to be confirmed in this application are determined based on the simulated description of symptom characteristics and the feedback status set of multiple previously confirmed symptoms to be confirmed, and the above steps are repeated until the feedback status of the new symptom to be confirmed is obtained.

[0055] Step 402: using the initial disease screening model, generating initial screening disease information corresponding to each sample data according to the feedback status of each symptom to be confirmed when the symptom characteristics are simulated and the first preset iteration stop condition is met.

[0056] The first preset iteration stop condition may be performed 9 times, 19 times, or until the initial disease screening model can determine the initial screening disease information, which is not limited in the present embodiment. At this time, there are multiple symptoms to be confirmed, and the initial disease screening model can determine the feedback status of multiple symptoms to be confirmed.

[0057] Optionally, an initial disease screening model is used to determine the initial screening disease information corresponding to each sample data based on the feedback status of each symptom to be confirmed when the symptom characteristics are simulated and the first preset iteration stop condition is met.

[0058] In an embodiment of the present application, the initial disease screening model is trained by simulating the actual consultation process between doctors and patients, thereby improving the question-answering ability of the initial disease screening model, thereby making the questions raised by the initial disease screening model more effective and improving the efficiency of determining the initial screening disease information.

[0059] Based on the above embodiments, the present application also provides a process for determining symptoms to be confirmed in a training method of an auxiliary diagnosis model. Figure 5 A flowchart of determining symptoms to be confirmed in a training method of an auxiliary diagnosis model provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, in the above step 202, the initial disease screening model is used to determine the symptoms to be confirmed corresponding to each sample data from the symptoms of multiple preset diseases according to the simulated description of the symptom characteristics, including: Step 501: construct an initial state space for each sample data according to the characteristics of multiple symptoms and associated medical history characteristics.

[0060] The initial state space includes: multiple states and symptom indication information of each state, and each state includes: a symptom feature and associated medical history features. The associated medical history features are embedded into the initial state space through a text embedding model, and the associated medical history features remain unchanged during the training process of the auxiliary diagnosis model.

[0061] The symptom indication information of each state is represented by a binary triplet. The symptom indication information of a state can be: unknown symptom, confirmed symptom, and denied symptom. Unknown symptom can be represented by [0, 0, 1], confirmed symptom can be represented by [0, 1, 0], and denied symptom can be represented by [1, 0, 0]. In the initial state space, the symptom indication information of all symptoms is unknown.

[0062] For example, when multiple symptoms are characterized by and associated medical history characteristics , then the initial state space ,in, , is the number of symptoms, Each symptom is represented by a triple.

[0063] Step 502: construct an initial operation space according to symptoms of multiple preset diseases.

[0064] The initial operation space includes symptoms of multiple diseases, and the symptoms in the initial operation space are divided into first-layer symptoms and second-layer symptoms. The second-layer symptoms are associated with the first-layer symptoms, and one symptom in the first-layer symptoms is associated with multiple symptoms in the second-layer symptoms. When one symptom in the first-layer symptoms is determined, the symptoms in the second-layer symptoms can be confirmed.

[0065] Optionally, an initial operation space is constructed according to symptoms of a plurality of preset diseases, so that the symptoms to be confirmed are determined according to the symptoms in the initial operation space.

[0066] Step 503: using the initial disease screening model, according to the symptom indication information of the first state simulating and describing the symptom characteristics in the initial state space, determine the symptoms to be confirmed from the initial operation space.

[0067] For example, using the initial disease screening model, if the symptom indication information of the first state simulating the description of the symptom characteristics in the initial state space is a confirmed symptom, the symptom to be confirmed is determined from the initial operation space as other symptoms in the first layer of symptoms or symptoms in the second layer of symptoms corresponding to the simulated description of the symptom. If the symptom indication information of the first state simulating the description of the symptom characteristics in the initial state space is a denied symptom, the symptom to be confirmed is determined from the initial operation space as other symptoms in the first layer of symptoms.

[0068] In the above step 205, the initial disease screening model is used to generate initial screening disease information corresponding to each sample data according to the feedback status of the simulated description of symptom characteristics and the symptoms to be confirmed, including: Step 504: According to the feedback state of the symptom to be confirmed, the symptom indication information of the second state of the symptom to be confirmed in the initial state space is updated to obtain a new state space.

[0069] There may be multiple symptoms to be confirmed, and when there are multiple symptoms to be confirmed, there are also multiple symptom indication information of the corresponding second state. The new state space includes multiple states and new symptom indication information of each state.

[0070] Optionally, according to the feedback state of the symptom to be confirmed, the symptom indication information of the second state of the symptom to be confirmed in the initial state space is updated, and the symptom indication information of the second state of the symptom to be confirmed in the new state space is determined.

[0071] Step 505: using the initial disease screening model, generating initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space.

[0072] Optionally, an initial disease screening model is used to determine the patient's confirmed symptoms, unknown symptoms, and denied symptoms based on the symptom indication information of the second state and the symptom indication information of the first state in the new state space, and initial screening disease information is generated based on the patient's confirmed symptoms.

[0073] In the embodiment of the present application, the symptoms to be confirmed are determined through the initial state space and the initial operation space, so that a new state space is obtained according to the feedback state of the symptoms to be confirmed, and initial screening disease information is generated according to the new state space and the initial disease screening model. The present application promotes the communication efficiency between doctors and patients, makes the information obtained more effective, speeds up the diagnosis process, and improves the diagnosis efficiency.

[0074] Based on the above embodiment, the initial disease screening model includes: an initial strategy model and an initial screening model, and correspondingly, the target disease screening model includes: a target strategy model and a target screening model. To this end, the present application also provides another process for determining the initial screening disease information in the training method of the auxiliary diagnosis model, Figure 6 A schematic diagram of a process for determining initial screening disease information in another auxiliary diagnosis model training method provided in an embodiment of the present application, such as Figure 6 As shown, in the above step 503, the initial disease screening model is used to determine the symptoms to be confirmed from the initial operation space according to the symptom indication information of the first state simulating the description of the symptom characteristics in the initial state space, including: Step 601: adopt the initial strategy model and determine the symptoms to be confirmed from the initial operation space according to the symptom indication information of the first state in the initial state space.

[0075] Among them, the initial strategy model is a strategy model built based on the neural network model.

[0076] Optionally, an initial strategy model is used to determine the probability of each symptom based on the symptom indication information of the first state in the initial state space from the multilayer perceptron, and based on the probability of each symptom, the symptom with the highest probability is determined from the initial operation space as the symptom to be confirmed.

[0077] In the above step 505, the initial disease screening model is used to generate initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space, including: Step 602: using the initial screening model, generating initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space.

[0078] Optionally, an initial screening model is used to make predictions based on the symptom indication information of the second state and the symptom indication information of the first state in the new state space to generate initial screening disease information.

[0079] For example, Figure 7 A schematic diagram of the structure of an auxiliary diagnosis model in an auxiliary diagnosis model training method provided in an embodiment of the present application, such as Figure 7As shown, the auxiliary diagnosis model includes: a doctor simulation model, a target disease screening model and a target disease symptom collection model. The target disease screening model also includes: a target screening model and a target strategy model. Among them, the target strategy model is used to determine the symptoms to be confirmed from the initial operation space according to the symptom indication information of the first state in the initial state space. The target screening model is used to generate initial screening disease information according to the symptom indication information of the second state in the new state space and the symptom indication information of the first state. The target disease symptom collection model is used to generate simulated disease symptoms corresponding to each sample data based on the initial screening disease information and each sample data. The doctor simulation model is used to understand the input natural language and generate corresponding natural language according to the features.

[0080] In the embodiment of the present application, the initial strategy model is used to determine the symptoms to be confirmed from the initial operation space according to the symptom indication information of the first state in the initial state space, and the initial screening model is used to generate initial screening disease information according to the symptom indication information of the second state in the new state space and the symptom indication information of the first state. The present application can ensure the accuracy of the determined symptoms to be confirmed and improve the efficiency of the initial strategy model in determining the symptoms to be confirmed.

[0081] Based on the above embodiments, the present application also provides a process for determining a target disease screening model in a training method for an auxiliary diagnosis model. Figure 8 A flowchart of determining a target disease screening model in a training method for an auxiliary diagnosis model provided in an embodiment of the present application is shown in FIG. Figure 8 As shown, in the above step 207, the initial disease screening model in the initial auxiliary diagnosis model is trained according to the initial screening disease information and the preliminary diagnosis information in each sample data to obtain the target disease screening model, including: Step 801: Calculate the reward parameters of the initial strategy model according to the feedback status of the symptom to be confirmed.

[0082] Optionally, As the reward function of the initial strategy model, when the feedback status of the to-be-confirmed symptom is a confirmed symptom, the reward function of the to-be-confirmed status is Recorded as 1, when the feedback status of the symptom to be confirmed is the denied symptom, the reward function of the state to be confirmed is Recorded as 0. According to the reward function corresponding to all states to be confirmed The proportion of 1 in the value of is used to calculate the reward parameters of the initial policy model.

[0083] Optionally, the present application further includes using an initial strategy model to determine the symptoms to be confirmed from the initial operation space according to the symptom indication information of the first state in the initial state space and the reward function of the first symptom. Specifically, according to a multilayer perceptron from the state space The symptom indication information of the first state in the learning representation obtains the reward function of the first symptom , through two multi-layer perceptrons according to the reward function Generate state probability functions respectively ,in, For symptoms, is the number of symptoms. According to the state probability function The probability of each state determines that the symptom with the highest probability is the symptom to be confirmed.

[0084] Step 802: Train the initial strategy model according to the reward parameter to obtain the target strategy model.

[0085] Optionally, the initial strategy model is subjected to intensive training according to the reward parameter so that the reward parameter reaches a maximum value, thereby obtaining the target strategy model. The maximum value is determined according to specific circumstances, and the embodiment of the present application does not limit this.

[0086] Step 803: Train the initial screening model according to the initial screening disease information and the preliminary diagnosis information to obtain the target screening model.

[0087] For example, Table 1 is a schematic table of the detection results of a target screening model provided in an embodiment of the present application, as shown in Table 1: Table 1 Schematic diagram of the test results of the target screening model

[0088] Among them, according to Table 1, when there is a target strategy model, the accuracy of the initial disease screening information obtained is greater than the accuracy when there is no target strategy model. When the number of symptoms to be confirmed is large, the accuracy of the initial disease screening information obtained is greater than the accuracy when the number of symptoms to be confirmed is small. When there is a lot of sample data, the accuracy of the initial disease screening information obtained is greater than the accuracy when the sample data is small. When the doctor's large language model is different, the accuracy of the corresponding initial disease screening information is also different. The more initial disease screening information the target screening model in this application obtains, the higher the accuracy of the corresponding target screening model.

[0089] In an embodiment of the present application, the initial strategy model is trained according to the reward parameter to obtain the target strategy model, and the initial screening model is trained according to the initial screening disease information and the preliminary diagnosis information to obtain the target screening model. The present application can improve the efficiency and accuracy of the initial strategy model in determining the symptoms to be confirmed, and at the same time can improve the prediction accuracy of the initial screening model.

[0090] Based on the above embodiment, the present application also provides another process for determining simulated disease symptoms in a training method of an auxiliary diagnosis model. Fig. 9 A schematic diagram of a process for determining simulated disease symptoms in another training method for an auxiliary diagnosis model provided in an embodiment of the present application, such as Fig. 9 As shown, in the above step 206, the initial disease symptom collection model is used to generate simulated disease symptoms corresponding to each sample data according to the initial screening disease information and each sample data, including: Step 901: According to the initial screening disease information, a target sample data set corresponding to the initial screening disease information is determined from sample data sets of multiple preset diseases.

[0091] Among them, the sample data set of the preset disease can be a MIMIC-IV (Medical Information Mart for Intensive Care, Version IV) data set, and the MIMIC-IV data set includes sample data sets corresponding to 98 common diseases.

[0092] Optionally, when the initial screening disease information is one, a target sample data set corresponding to the initial screening disease information is obtained from the MIMIC-IV data set, and when the initial screening disease information is multiple, multiple target sample data sets corresponding to the number of initial screening disease information are obtained from the MIMIC-IV data set. Each target sample data set includes positive and negative sample data, and the positive and negative sample data are the same. For example, when the positive sample data includes 160, the negative sample data also includes 160.

[0093] Step 902: According to the initial screening disease information, obtain the disease document record corresponding to the initial screening disease information from the preset medical documents.

[0094] The preset medical literature may be medical literature stored in a public medical database or in a database of a preset medical institution. The preset medical literature includes disease literature records corresponding to multiple diseases, and the disease file records may include: disease overview, disease physiology, differential diagnosis methods, and clinical research, etc., which are not limited in this embodiment of the present application. There may be 120 disease literature records, and the number of positive and negative samples in the medical literature is the same.

[0095] Step 903: using the initial disease symptom collection model, based on disease literature records, generate a diagnostic decision-making procedure corresponding to the initial screening disease information.

[0096] The initial disease symptom collection model can be a model for disease symptom collection based on a large language model. The diagnostic decision program is a specific inquiry flow chart for initial screening of disease information, including multiple nodes. The diagnostic decision program can determine the next node based on the feedback status of the current node. The diagnostic decision program is used to ask questions to the target patient.

[0097] Optionally, an initial disease symptom collection model is adopted and constructed based on disease literature records, thereby generating a diagnostic decision procedure corresponding to the disease literature records.

[0098] Step 904: Use the doctor simulation model to generate a query question for the first node according to the first node in the diagnosis decision-making process.

[0099] The inquiry question of the first node in the diagnosis decision procedure is the first inquiry question generated based on the preset disease literature in the diagnosis decision procedure.

[0100] Optionally, a doctor simulation model is used to generate natural language for the content of the first node in the diagnosis decision-making process, thereby generating an inquiry question for the first node.

[0101] Step 905: Use the patient simulation large model to generate the feedback state of the first node according to the inquiry question of the first node and each sample data in the target sample data set.

[0102] Optionally, a patient simulation large model is used to perform natural language understanding on the inquiry question of the first node, and the feedback state of the first node is determined according to the result of natural language understanding and each sample data in the target sample data set. The feedback state of the first node may include: whether there is a symptom in the inquiry question and specific data of the symptom, and the inquiry question includes at least one symptom.

[0103] Step 906: using the initial disease symptom collection model, according to the feedback state of the first node, determine the next second node of the first node, and generate the query question of the second node.

[0104] Optionally, the initial disease symptom collection model is used to perform natural language understanding on the feedback state of the first node, determine the next second node of the first node according to the natural language understanding result, and generate the inquiry question of the second node. For example, when the patient in the feedback state has the symptom in the inquiry question, the second node is determined to be the yes node corresponding to the first node, otherwise the second node is the no node corresponding to the first node.

[0105] Step 907: Using the patient simulation large model, generate the feedback state of the second node according to the query question of the second node and each sample data in the target sample data set.

[0106] Optionally, a patient simulation large model is used to perform natural language understanding on the inquiry questions of the second node, and the feedback status of the second node is determined based on the results of the natural language understanding and each sample data in the target sample data set. There can be multiple second nodes, and the specific number of second nodes is determined according to the number of nodes in the diagnosis decision process. Correspondingly, the inquiry questions of the second node and the feedback status of the second node are the same as the number of second nodes.

[0107] Step 908: using the initial disease symptom collection model, generate simulated disease symptoms according to the feedback status of each known node in the diagnosis decision program when the second preset iteration stop condition is met.

[0108] The second preset iteration stop condition may be when there is no new second node, or when the acquired data can determine the simulated disease symptoms. The specific preset iteration stop condition may be determined according to the actual usage scenario, and the embodiment of the present application does not limit this. A known node is a node in a feedback state.

[0109] Optionally, the initial disease symptom collection model is used to generate simulated disease symptoms according to the feedback state of each known node in the diagnostic decision-making program when the second preset iteration stop condition is met and the corresponding known node. The present application can also optimize the initial disease symptom collection model according to the simulated disease symptoms to obtain a disease symptom collection model, thereby generating a new diagnostic decision process according to the target differential diagnosis program, thereby improving the accuracy of generating simulated disease symptoms.

[0110] For example, Fig.10 A structural diagram of a diagnostic decision-making procedure in an auxiliary diagnosis model training method provided in an embodiment of the present application, such as Fig.10 As shown, the diagnostic decision procedure includes multiple nodes. According to the feedback state of the current node, the next node of the node is determined to be the second node. According to the feedback state of the second node, the next node of the second node is determined to be the new second node until there is no node under the new second node. Among them, the feedback state can be yes or no. The white node is the corresponding second node when the feedback state is yes, and the black node is the corresponding second node when the feedback state is no.

[0111] For example, Table 2 is a schematic table of detection results of a target disease symptom collection model provided in an embodiment of the present application, as shown in Table 2: Table 2 Detection results of the target disease symptom collection model

[0112] Among them, the success rate in Table 2 is used to indicate the proportion of successful simulation of disease symptoms. Accuracy is used to indicate that accuracy refers to the degree of closeness between simulated disease symptoms and initial diagnostic information. Precision is used to indicate the consistency or repeatability of simulated disease symptoms measured or estimated multiple times. Recall is used to indicate the proportion of correctly identified positive samples to all actual positive samples. The harmonic mean of precision and recall is used to evaluate the model's ability to strike a balance between precision and recall. In this application, the proportion of successful simulation of disease symptoms through the calibrated diagnostic decision-making procedure is the highest, and the accuracy of predictions by different models is also different.

[0113] In the embodiment of the present application, a target sample data set is determined based on a sample set of a preset disease, a disease document record corresponding to the initial screening disease information is obtained from a preset medical document, a diagnostic decision program is generated, and simulated disease symptoms are generated based on the diagnostic decision program and the target sample data set. The application can improve the speed of generating a diagnostic decision program from an initial disease symptom collection model and the accuracy of generating simulated disease symptoms.

[0114] Based on the above embodiments, the present application also includes a process for determining a sample data set in a training method for an auxiliary diagnosis model. Fig.11 The process of determining a sample data set in a training method for an auxiliary diagnosis model provided in an embodiment of the present application is as follows: Fig.11 As shown, in the above step 101, a plurality of sample data sets of preset diseases are obtained, including: Step 1101: Acquire clinical data sets of multiple preset diseases from a preset clinical medicine database.

[0115] The clinical data set of each preset disease includes: multiple groups of clinical data files; the multiple groups of clinical data files correspond to multiple patients who are initially diagnosed with each preset disease. Each clinical file data includes at least: admission records, discharge summaries, laboratory test results, imaging test results, etc.

[0116] The preset clinical medicine database may be the MIMIC-IV (Medical Information Mart for Intensive Care, Version IV) database, which includes 40,000 electronic medical records (EMRs). According to the 400,000 EMRs, 98 common diseases are determined as preset diseases, and the preset diseases do not include injuries caused by accidents and mental illnesses. From the EMRs corresponding to the 98 diseases in the MIMIC-IV database, 40,000 cases are determined to form a clinical data set corresponding to the preset diseases.

[0117] Step 1102: Perform symptom analysis of each preset disease on the structured data in each group of clinical data files to obtain multiple symptom features corresponding to each group of clinical data files.

[0118] The structured data includes: History of Present Illness (HPI) and Chief Complaint (CC). HPI and CC can be determined from the admission records in the clinical data file. HPI is the information of the patient's current visit. Symptom features are represented by binary vectors.

[0119] Optionally, according to a preset symptom check tool, the structured data in each group of clinical data files is analyzed for symptoms of each preset disease to obtain multiple symptoms corresponding to each group of clinical data files, and the multiple symptoms are converted into multiple symptom features through a preset doctor big model, thereby obtaining multiple symptom features corresponding to each group of clinical data files. The preset symptom check tool can be a symptom checker developed by the Mayo Clinic.

[0120] For example, when the preset symptom check tool can be a symptom checker developed by the Mayo Clinic, the symptom checker performs symptom analysis of each preset disease on the HPI and CC in each group of clinical data files, and obtains 717 symptom features corresponding to each group of clinical data files. Among them, the 717 symptom features include 28 first-layer category features and 689 second-layer category features, and the second-layer category features are associated with the first-layer category features. The second-layer category features are used to refine the content of the first-layer category features, and each first-layer category feature corresponds to multiple second-layer category features.

[0121] Step 1103: extract features from the unstructured data in each group of clinical data files to obtain associated medical history features corresponding to each group of clinical data files.

[0122] Among them, unstructured data includes medical history, social history and family history. Medical history is used to represent the patient's past medical history.

[0123] Optionally, the medical history, social history and family history in each group of clinical data files are obtained, and features of the medical history, social history and family history in each group of clinical data files are extracted through a preset doctor model to obtain associated medical history features corresponding to each group of clinical data files.

[0124] Step 1104: Generate a sample data of each preset disease according to the multiple symptom characteristics, the associated medical history characteristics and the preliminary diagnosis information of each preset disease corresponding to each group of clinical data files.

[0125] In the embodiment of the present application, clinical data sets of multiple preset diseases are obtained from a preset clinical medicine database, and multiple symptom features corresponding to each set of clinical data files and associated medical history features corresponding to each set of clinical data files are obtained, thereby generating a sample data of each preset disease. The auxiliary diagnosis model can be accurately trained, thereby improving the diagnostic accuracy of the auxiliary diagnosis model.

[0126] Based on the above embodiments, the present application further provides a disease symptom collection method, which is applied to a target-assisted diagnosis model obtained by the above training method, wherein the target-assisted diagnosis model includes: a doctor simulation model, a target disease screening model, and a target disease symptom collection model; Fig.12 A flow chart of a disease symptom collection method provided in an embodiment of the present application is shown in FIG. Fig.12 As shown, the method includes: Step 1201: Obtain an input initial symptom description statement for a target patient.

[0127] Among them, the initial symptom description sentence is a natural language text.

[0128] Optionally, the target patient can input the initial symptom description statement through the medical institution's guidance equipment, or input the initial symptom description statement through the preset application software interface of the medical structure. This application does not limit the method of inputting the initial symptom description statement, and the specific input method can be text input, voice input, etc.

[0129] Step 1202: Using the target disease screening model, generate initial description symptoms and symptoms to be confirmed associated with the initial description symptoms according to the initial symptom description statement.

[0130] Optionally, a doctor simulation model is used to perform natural language understanding on the initial symptom description statement to generate initial symptom description features. Then, a target disease screening model is used to determine the symptoms to be confirmed corresponding to each sample data from the symptoms of multiple preset diseases based on the initial symptom description features.

[0131] Specifically, the initial state space of each sample data is constructed according to the characteristics of multiple symptoms and the associated medical history characteristics, and the initial operation space is constructed according to the symptoms of multiple preset diseases. The target disease screening model is adopted, and the symptoms to be confirmed are determined from the initial operation space according to the symptom indication information of the first state that initially describes the symptom characteristics in the initial state space.

[0132] Step 1203: Use the doctor simulation large model to generate the first inquiry question of the symptoms to be confirmed based on the symptoms to be confirmed.

[0133] Optionally, a large doctor simulation model is used to perform natural language understanding on the symptoms to be confirmed and generate the first inquiry question of the symptoms to be confirmed.

[0134] Step 1204: Obtain the input feedback status of the target patient regarding the symptom to be confirmed.

[0135] Optionally, the first inquiry question can be displayed on the input interface of the target patient, and the target patient responds according to the first inquiry question. The doctor simulates the large model to understand the target patient's response in natural language, thereby determining the target patient's feedback status for the symptoms to be confirmed. Specifically, the patient's response can be to select from multiple options generated according to the first inquiry question, or to judge the first inquiry question, or to directly input the answer to the first inquiry question, and the embodiment of the present application does not limit this.

[0136] Step 1205: using the target disease screening model, generate initial disease diagnosis information based on the initial description of symptoms and the feedback status of the symptoms to be confirmed.

[0137] Optionally, according to the feedback state of the symptom to be confirmed, the symptom indication information of the second state of the symptom to be confirmed in the initial state space is updated to obtain a new state space. The initial disease screening model is used to generate initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space.

[0138] Step 1206: Using the target disease symptom collection model, generate the collected disease symptoms of the target patient according to the preset medical literature records corresponding to the initial disease diagnosis information.

[0139] Optionally, according to the initial screening disease information, a target sample data set corresponding to the initial screening disease information is determined from a plurality of sample data sets of preset diseases, and according to the initial screening disease information, a preset medical literature record corresponding to the initial screening disease information is obtained from preset medical literature. A target disease symptom collection model is used to generate a diagnostic decision program corresponding to the initial screening disease information according to the disease literature record. A doctor simulation large model is used to generate an inquiry question for the first node according to the first node in the diagnostic decision program, and the feedback state of the first node is obtained. A target disease symptom collection model is used to determine the next second node of the first node according to the feedback state of the first node, generate an inquiry question for the second node, and obtain the feedback state of the second node. A target disease symptom collection model is used to generate the collected disease symptoms of the target patient according to the feedback state of each known node in the diagnostic decision program when the second preset iteration stop condition is met.

[0140] Step 1207: Use the doctor simulation large model to generate a disease-assisted diagnosis description statement for the target patient based on the target-assisted disease diagnosis information.

[0141] Optionally, a doctor simulation large model is used to generate natural language based on the target auxiliary disease diagnosis information, thereby generating a disease auxiliary diagnosis description statement for the target patient.

[0142] For example, this application takes heart failure as an example to illustrate the method for collecting disease symptoms. The specific dialogue process is as follows: Auxiliary diagnostic model: Do you feel shortness of breath with mild to moderate exercise? Target patients: Yes.

[0143] Auxiliary Diagnosis Model: Have you recently experienced shortness of breath? Target Patient: Yes.

[0144] Auxiliary diagnosis model: Do you cough? Target patient: No, I don’t cough.

[0145] Auxiliary diagnosis model: Do you feel nauseous or vomiting? Target patient: No, I don’t feel nauseous or vomiting.

[0146] Auxiliary diagnosis model: Do you have abdominal pain? Target patient: No, I don’t have abdominal pain.

[0147] Auxiliary Diagnosis Model: Do you have diarrhea? Target Patient: No, I don’t have diarrhea.

[0148] Auxiliary diagnosis model: Are your feet or hands swollen? Target patient: Yes, there is slight swelling throughout the day.

[0149] Diagnostic aid model: Can you describe whether your swelling of your feet or hands is accompanied by redness or pain? Target patient: No.

[0150] Diagnostic aid model: Is the swelling around the ankle or foot? Target patient: Yes.

[0151] Auxiliary diagnostic model: You are at risk for heart failure, pulmonary embolism, and pneumonia.

[0152] Auxiliary diagnosis model: Are there risk factors (such as myocardial infarction, arterial hypertension, family history of coronary artery disease (CAD), history of diabetes, history of alcohol abuse, history of chronic kidney disease, history of cardiotoxic chemotherapy, family history of cardiomyopathy (CMP), or history of sudden death)? Target patients: I have a history of hypertension, diabetes, and a family history of CAD.

[0153] Auxiliary Diagnosis Model: Are there any symptoms or signs of heart failure, such as dyspnea, ankle swelling, limb swelling, wheezing, paroxysmal nocturnal dyspnea, decreased exercise tolerance, fatigue, tiredness, or increased recovery time after exercise? Target Patient: Yes, I have symptoms of dyspnea, wheezing, and paroxysmal nocturnal dyspnea; I also have ankle or limb swelling; I have decreased exercise tolerance, resulting in fatigue and tiredness.

[0154] Auxiliary diagnosis model: Is the electrocardiogram (ECG) abnormal (such as atrial fibrillation, Q wave, left ventricular hypertrophy)? Target patients: Yes, ECG abnormal, left ventricular hypertrophy, left bundle branch block.

[0155] Auxiliary diagnostic model: Is the N-terminal pro-B-type natriuretic peptide (NT-proBNP) level elevated (125 pg / mL) or the B-type natriuretic peptide (BNP) level elevated (≥35 pg / mL)? Target patients: Yes, NT-proBNP level is elevated (>35,000).

[0156] Auxiliary Diagnosis Model: Does the echocardiogram show abnormalities, such as dysfunction, low ejection fraction, abnormal chamber size, left ventricular hypertrophy (LVH), or regional wall motion? Target Patients: Yes, the echocardiogram shows abnormalities, such as dysfunction, low ejection fraction (EF), and regional wall motion abnormalities. Auxiliary Diagnosis Model: Is the left ventricular ejection fraction (LVEF) less than or equal to 40%? Target Patients: Yes.

[0157] From the above example, we can see that "You are at risk of heart failure, pulmonary embolism, and pneumonia" is the initial disease diagnosis information, and the information after the initial disease diagnosis information is a description statement of the target patient's disease symptoms.

[0158] Based on the same inventive concept, the embodiments of the present application also provide a training device for an auxiliary diagnosis model corresponding to the training method for the auxiliary diagnosis model. Since the principle of solving the problem by the device in the embodiments of the present application is similar to the training method for the auxiliary diagnosis model in the embodiments of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0159] Fig.13 A schematic diagram of a training device for an auxiliary diagnosis model provided in an embodiment of the present application, the device comprising: An acquisition module 1301 is used to acquire sample data sets of multiple preset diseases, where each sample data in the sample data set of each preset disease includes: characteristics of multiple symptoms of a patient, associated medical history characteristics, and preliminary diagnosis information for each preset disease; The first generating module 1302 is used to randomly generate a simulated initial symptom description statement corresponding to each sample data according to the characteristics of multiple symptoms by using a large patient simulation model; The second generating module 1303 is used to generate a simulated disease symptom corresponding to each sample data according to the simulated initial symptom description statement and each sample data by using the initial auxiliary diagnosis model; The training module 1304 is used to train the initial auxiliary diagnosis model according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain the target auxiliary diagnosis model.

[0160] Optionally, the initial auxiliary diagnosis model includes: a doctor simulation model, an initial disease screening model and an initial disease symptom collection model; the second generation module 1303 is specifically used to: use the doctor simulation model to generate a simulated description symptom feature corresponding to each sample data according to the simulated initial symptom description statement; Using the initial disease screening model, based on the simulated description of symptom characteristics, the symptoms to be confirmed corresponding to each sample data are determined from the symptoms of multiple preset diseases; A doctor simulation model is used to generate inquiry questions about the symptoms to be confirmed based on the symptoms to be confirmed; A large patient simulation model is used to generate feedback status of symptoms to be confirmed based on the questions asked and the characteristics of multiple symptoms; The initial disease screening model is used to generate the initial screening disease information corresponding to each sample data based on the simulated description of symptom characteristics and the feedback status of the symptoms to be confirmed; The initial disease symptom collection model is used to generate simulated disease symptoms corresponding to each sample data based on the initial screening disease information and each sample data; Optionally, the training module 1304 is specifically used to: train the initial disease screening model in the initial auxiliary diagnosis model according to the initial screening disease information and the preliminary diagnosis information in each sample data to obtain a target disease screening model; According to the simulated disease symptoms and multiple symptoms corresponding to each preset disease, the initial disease symptom collection model in the initial auxiliary diagnosis model is trained to obtain the target disease symptom collection model. The target auxiliary diagnosis model includes: a doctor simulation large model, a target disease screening model and a target disease symptom collection model.

[0161] Optionally, the second generating module 1303 is specifically used to: adopt the initial disease screening model, and according to the feedback status of the simulated description of the symptom characteristics and the symptoms to be confirmed, re-determine the new symptoms to be confirmed corresponding to each sample data from the symptoms of multiple preset diseases, and repeat the execution until the feedback status of the new symptoms to be confirmed is obtained; The initial disease screening model is used to generate initial screening disease information corresponding to each sample data based on the feedback status of each symptom to be confirmed when the symptom characteristics are simulated and the first preset iteration stop condition is met.

[0162] Optionally, the second generation module 1303 is specifically used to: construct an initial state space of each sample data according to the characteristics of multiple symptoms and the associated medical history characteristics, the initial state space including: multiple states and symptom indication information of each state, each state including: a symptom characteristic and associated medical history characteristics; Construct an initial operation space based on the symptoms of multiple preset diseases; Using the initial disease screening model, the symptoms to be confirmed are determined from the initial operation space according to the symptom indication information of the first state simulating the description of the symptom characteristics in the initial state space; The initial disease screening model is used to generate initial screening disease information corresponding to each sample data based on the simulated description of symptom characteristics and the feedback status of the symptoms to be confirmed, including: According to the feedback state of the symptom to be confirmed, the symptom indication information of the second state of the symptom to be confirmed in the initial state space is updated to obtain a new state space; The initial disease screening model is adopted to generate initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space.

[0163] Optionally, the initial disease screening model includes: an initial strategy model and an initial screening model, and correspondingly, the target disease screening model includes: a target strategy model and a target screening model; the second generation module 1303 is specifically used to: adopt the initial strategy model, and determine the symptoms to be confirmed from the initial operation space according to the symptom indication information of the first state in the initial state space; The initial screening model is used to generate initial screening disease information based on the symptom indication information of the second state and the symptom indication information of the first state in the new state space.

[0164] Optionally, the training module 1304 is specifically used to: calculate the reward parameter of the initial strategy model according to the feedback state of the symptom to be confirmed; According to the reward parameters, the initial strategy model is trained to obtain the target strategy model; According to the initial screening disease information and preliminary diagnosis information, the initial screening model is trained to obtain the target screening model.

[0165] Optionally, the second generating module 1303 is specifically configured to: determine, according to the initial screening disease information, a target sample data set corresponding to the initial screening disease information from sample data sets of multiple preset diseases; According to the initial screening disease information, obtain the disease literature records corresponding to the initial screening disease information from the preset medical literature; Using the initial disease symptom collection model, a diagnostic decision-making procedure corresponding to the initial screening disease information is generated based on disease literature records; Using the doctor simulation model, the inquiry question of the first node is generated according to the first node in the diagnosis decision process; A patient simulation large model is used to generate a feedback state of the first node according to the inquiry questions of the first node and the sample data in the target sample data set; Using the initial disease symptom collection model, according to the feedback state of the first node, determining the next second node of the first node, and generating the inquiry question of the second node; The patient simulation model is used to generate the feedback state of the second node according to the inquiry questions of the second node and the sample data in the target sample data set; The initial disease symptom collection model is used to generate simulated disease symptoms according to the feedback status of each known node in the diagnosis decision program when the second preset iteration stop condition is met.

[0166] Optionally, the acquisition module 1301 is specifically used to: acquire clinical data sets of multiple preset diseases from a preset clinical medicine database, each clinical data set of the preset disease includes: multiple groups of clinical data files; the multiple groups of clinical data files respectively correspond to multiple patients who are initially diagnosed with each preset disease; Performing symptom analysis on each preset disease on the structured data in each group of clinical data files to obtain multiple symptom features corresponding to each group of clinical data files; Extract features from the unstructured data in each group of clinical data files to obtain the associated medical history features corresponding to each group of clinical data files; A sample data of each preset disease is generated according to a plurality of symptom characteristics, associated medical history characteristics and preliminary diagnosis information of each preset disease corresponding to each group of clinical data files.

[0167] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0168] The present application also provides a computer device, Fig.14 A schematic diagram of the structure of a computer device provided in an embodiment of the present application, such as Fig.14 As shown, the computer device includes: a processor 1401, a memory 1402, and optionally, a bus 1403. The memory 1402 stores machine-readable instructions executable by the processor 1401. When the computer device is running, the processor 1401 communicates with the memory 1402 through the bus 1403. When the machine-readable instructions are executed by the processor 1401, the steps of the auxiliary diagnosis model training method and the disease symptom collection method are executed.

[0169] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned auxiliary diagnosis model training method and disease symptom collection method are executed.

[0170] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the method embodiment, and will not be repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0171] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or part of the technical solution that contributes to the prior art or 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 several 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, RandomAccess Memory), disk or optical disk and other media that can store program code.

[0172] The above are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for training an auxiliary diagnosis model, characterized in that: The method comprises: Acquire a plurality of sample data sets of preset diseases, wherein each sample data in each sample data set of the preset disease includes: characteristics of a plurality of symptoms of a patient, associated medical history characteristics, and preliminary diagnosis information for each preset disease; Using a large patient simulation model, randomly generating a simulated initial symptom description statement corresponding to each sample data according to the characteristics of the multiple symptoms; Using the initial auxiliary diagnosis model, according to the simulated initial symptom description statement and each sample data, generate a simulated disease symptom corresponding to each sample data; The initial auxiliary diagnosis model is trained according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain a target auxiliary diagnosis model.

2. The training method of the auxiliary diagnosis model according to claim 1, characterized in that: The initial auxiliary diagnosis model includes: a doctor simulation model, an initial disease screening model and an initial disease symptom collection model; the initial auxiliary diagnosis model is used to generate a simulated disease symptom corresponding to each sample data according to the simulated initial symptom description statement and each sample data, including: Using the doctor simulation large model, according to the simulated initial symptom description statement, generate the simulated description symptom characteristics corresponding to each sample data; Using the initial disease screening model, according to the symptom characteristics described in the simulation, determine the symptoms to be confirmed corresponding to each sample data from the symptoms of multiple preset diseases; Using the doctor simulation model, generating inquiry questions about the symptoms to be confirmed according to the symptoms to be confirmed; Using the patient simulation macromodel, generating a feedback status of the to-be-confirmed symptom according to the inquiry questions and the characteristics of the multiple symptoms; Using the initial disease screening model, generating initial screening disease information corresponding to each sample data according to the simulated description of symptom characteristics and the feedback status of the symptoms to be confirmed; Using the initial disease symptom collection model, based on the initial screening disease information and each sample data, generate simulated disease symptoms corresponding to each sample data; The training of the initial auxiliary diagnosis model according to the multiple symptoms corresponding to each preset disease and the simulated disease symptoms to obtain a target auxiliary diagnosis model includes: According to the initial screening disease information and the preliminary diagnosis information in each sample data, the initial disease screening model in the initial auxiliary diagnosis model is trained to obtain a target disease screening model; According to the simulated disease symptoms and the multiple symptoms corresponding to each preset disease, the initial disease symptom collection model in the initial auxiliary diagnosis model is trained to obtain a target disease symptom collection model, and the target auxiliary diagnosis model includes: the doctor simulation large model, the target disease screening model and the target disease symptom collection model.

3. The training method of the auxiliary diagnosis model according to claim 2, characterized in that: The initial disease screening model is used to generate initial screening disease information corresponding to each sample data according to the simulated description of symptom characteristics and the feedback status of the to-be-confirmed symptom, including: Using the initial disease screening model, according to the simulated description of the symptom characteristics and the feedback status of the symptom to be confirmed, re-determine the new symptom to be confirmed corresponding to each sample data from the symptoms of the multiple preset diseases, and repeat the execution until the feedback status of the new symptom to be confirmed is obtained; The initial disease screening model is used to generate initial screening disease information corresponding to each sample data according to the feedback status of each symptom to be confirmed when the simulation describes the symptom characteristics and satisfies the first preset iteration stop condition.

4. The training method of the auxiliary diagnosis model according to claim 2, characterized in that: The initial disease screening model is used to determine the to-be-confirmed symptom corresponding to each sample data from the symptoms of a plurality of preset diseases according to the symptom characteristics described in the simulation, including: According to the characteristics of the multiple symptoms and the associated medical history characteristics, constructing the initial state space of each sample data, the initial state space including: multiple states and symptom indication information of each state, each state including: a symptom characteristic and the associated medical history characteristics; constructing an initial operation space according to the symptoms of the plurality of preset diseases; Using the initial disease screening model, according to the symptom indication information of the first state simulating and describing the symptom characteristics in the initial state space, determining the to-be-confirmed symptom from the initial operation space; The initial disease screening model is used to generate initial screening disease information corresponding to each sample data according to the simulated description of symptom characteristics and the feedback status of the to-be-confirmed symptom, including: According to the feedback state of the symptom to be confirmed, updating the symptom indication information of the second state of the symptom to be confirmed in the initial state space to obtain a new state space; The initial disease screening model is used to generate the initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space.

5. The training method of the auxiliary diagnosis model according to claim 4, characterized in that: The initial disease screening model includes: an initial strategy model and an initial screening model. Correspondingly, the target disease screening model includes: a target strategy model and a target screening model. The initial disease screening model is used to determine the to-be-confirmed symptom from the initial operation space according to the symptom indication information of the first state simulating the description of the symptom characteristics in the initial state space, including: Adopting the initial strategy model, and according to the symptom indication information of the first state in the initial state space, determining the to-be-confirmed symptom from the initial operation space; The adopting the initial disease screening model to generate the initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space comprises: The initial screening model is used to generate the initial screening disease information according to the symptom indication information of the second state and the symptom indication information of the first state in the new state space.

6. The training method of the auxiliary diagnosis model according to claim 5, characterized in that: The initial disease screening model in the initial auxiliary diagnosis model is trained according to the initial screening disease information and the preliminary diagnosis information in each sample data to obtain a target disease screening model, including: Calculating the reward parameter of the initial strategy model according to the feedback state of the symptom to be confirmed; Training the initial strategy model according to the reward parameter to obtain the target strategy model; The initial screening model is trained according to the initial screening disease information and the preliminary diagnosis information to obtain the target screening model.

7. The training method of the auxiliary diagnosis model according to claim 2, characterized in that: The adopting the initial disease symptom collection model to generate a simulated disease symptom corresponding to each sample data according to the initial screening disease information and each sample data includes: According to the initial screening disease information, determining a target sample data set corresponding to the initial screening disease information from sample data sets of the plurality of preset diseases; According to the initial screening disease information, obtaining disease literature records corresponding to the initial screening disease information from preset medical literature; Using the initial disease symptom collection model, based on the disease literature records, a diagnostic decision-making procedure corresponding to the initial screening disease information is generated; Using the doctor simulation macro model, according to the first node in the diagnosis decision procedure, generate a query question for the first node; Using the patient simulation macro model, generating a feedback state of the first node according to the query question of the first node and each sample data in the target sample data set; Using the initial disease symptom collection model, according to the feedback state of the first node, determining the next second node of the first node, and generating a query question for the second node; Using the patient simulation macro model, generating a feedback state of the second node according to the query question of the second node and each sample data in the target sample data set; The initial disease symptom collection model is used to generate the simulated disease symptoms according to the feedback status of each known node in the diagnosis decision program when the second preset iteration stop condition is met.

8. The training method of the auxiliary diagnosis model according to claim 1, characterized in that: The step of obtaining sample data sets of multiple preset diseases includes: Acquire clinical data sets of multiple preset diseases from a preset clinical medicine database, each clinical data set of the preset disease comprising: multiple groups of clinical data files; the multiple groups of clinical data files respectively correspond to multiple patients preliminarily diagnosed with each of the preset diseases; Performing symptom analysis on the structured data in each group of clinical data files for each preset disease to obtain a plurality of symptom features corresponding to each group of clinical data files; Extracting features from the unstructured data in each group of clinical data files to obtain associated medical history features corresponding to each group of clinical data files; A sample data of each preset disease is generated according to the multiple symptom characteristics, the associated medical history characteristics and the preliminary diagnosis information of each preset disease corresponding to each group of clinical data files.

9. A method for collecting disease symptoms, characterized in that: The target-assisted diagnosis model obtained by training using the training method described in any one of claims 2 to 8 above, the target-assisted diagnosis model comprising: a large doctor simulation model, a target disease screening model and a target disease symptom collection model; the disease symptom collection method comprising: Obtaining an inputted initial symptom description sentence for a target patient; Using the target disease screening model, generating initial description symptoms and symptoms to be confirmed associated with the initial description symptoms according to the initial symptom description statement; Using the doctor simulation large model, generating a first inquiry question for the symptom to be confirmed according to the symptom to be confirmed; Obtaining the input feedback status of the target patient regarding the symptom to be confirmed; Using the target disease screening model, generating initial disease diagnosis information according to the initial described symptoms and the feedback status of the symptoms to be confirmed; Using the target disease symptom collection model, generating the collected disease symptoms of the target patient according to the preset medical literature records corresponding to the initial disease diagnosis information; The doctor simulation model is used to generate a disease symptom description statement of the target patient based on the collected disease symptoms.

10. A computer device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor executes the machine-readable instructions to perform the steps of any one of the methods according to claims 1 to 9.

Images