Intelligent medical consultation method, system, computer device and storage medium

By strengthening learning of multi-agent models, combining disease prior probability prediction and agent strategies, the interactive consultation process of the intelligent consultation system is realized, solving the problems of high computational complexity and unexplainability in the existing technology, and improving the Diagnostic accuracy and user experience.

CN109192300BActive Publication Date: 2024-11-08BAIDU ONLINE NETWORK TECH (BEIJIBG) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201810941864.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2018-08-17
Publication Date
2024-11-08
Estimated Expiration
2038-08-17

AI Technical Summary

Technical Problem

The existing intelligent consultation methods have high computational complexity and unexplainability in disease diagnosis, especially based on probability graph models and deep learning models, which are difficult to optimize the rationality of consultation logic while ensuring the accuracy of consultation results. User experience.

Method used

A multi-agent model based on reinforcement learning is adopted, and through disease prior probability prediction and agent strategy determination is determined, doctor consultation logic is simulated, questions are asked step by step and interactive consultation is conducted until the disease diagnosis is determined.

Benefits of technology

While ensuring the accuracy of diagnosis, the rationality of the consultation logic and user experience are optimized, and the interactiveness and diagnostic efficiency of the intelligent consultation system are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN109192300B_ABST
    Figure CN109192300B_ABST
Patent Text Reader

Abstract

The present invention discloses an intelligent medical consultation method, system, computer device and storage medium. The intelligent medical consultation method includes: predicting the disease prior probability according to the main complaint of the target object to obtain the disease prior probability for the target object; determining at least one candidate disease based on the sorting of the disease prior probability; determining at least one agent strategy corresponding to at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the output of questions to be asked in the next interaction under the current symptom state; based on each agent strategy, interactive medical consultation is performed according to the interactive selection of the questions to be asked by the target object for the next interaction. The embodiment of the present invention can optimize the rationality of the relevant disease medical consultation logic and user experience while ensuring the accuracy of the medical consultation results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an intelligent medical consultation method, system, computer equipment and computer-readable storage medium. Background Art

[0002] At present, there are two main methods in the field of medical intelligent consultation to achieve intelligent consultation: one is based on the traditional probabilistic graphical model (PGM) and information gain, which can appropriately simulate the consultation process and give a disease diagnosis; the other is a relatively new deep learning model in recent years, which diagnoses the current possible disease based on the symptoms observed in the patient. Commonly used methods based on deep learning models include networks based on DNN (Deep Neural Networks), CNN (Convolution Neural Network), LSTM (Long Short-Term Memory) and their extensions.

[0003] However, the above traditional methods based on probabilistic graphical models (PGM) and information gain often have high computational complexity in the model inference part. At the same time, the disease-symptom transfer matrix of the probabilistic graphical model often requires professional doctors to label. This expert system-like model faces great challenges in disease expansion. In addition, the above intelligent consultation method based on deep learning models can give a relatively high disease prediction accuracy when all observed symptoms are known. However, under normal circumstances, we cannot observe all the symptoms of the patient. Due to its unexplainability, it is difficult for the deep learning-based model to give a reasonable consultation process.

[0004] Therefore, a new approach is needed to make a breakthrough in the field of intelligent medical consultation. Summary of the invention

[0005] The object of the present invention is to solve one of the above-mentioned technical problems at least to a certain extent.

[0006] To this end, the first object of the present invention is to provide an intelligent diagnosis method that can optimize the rationality of the diagnosis logic of related diseases and the user experience while ensuring the accuracy of the diagnosis results.

[0007] The second objective of the present invention is to provide an intelligent medical consultation system.

[0008] A third object of the present invention is to provide a computer device.

[0009] A fourth object of the present invention is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above-mentioned purpose, the intelligent medical consultation method proposed in the first embodiment of the present invention includes:

[0011] Predicting the disease prior probability according to the main complaint of the target object to obtain the disease prior probability for the target object;

[0012] Determining at least one candidate disease based on the ranking of the disease prior probabilities;

[0013] Determine at least one agent strategy associated with at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the question to be asked in the next interaction under the current symptom state;

[0014] Based on each agent strategy, an interactive consultation is performed according to the target object's interactive selection of questions to be asked in the next interaction.

[0015] The intelligent medical consultation method of the embodiment of the present invention predicts the disease prior probability according to the main complaint of the target object, obtains the disease prior probability for the target object, and then determines at least one candidate disease based on the ranking of the disease prior probability, and determines at least one agent strategy corresponding to at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the output of the questions to be asked in the next interaction under the current symptom state, and then, based on each agent strategy, interactive medical consultation is performed according to the interactive selection of the questions to be asked by the target object for the next interaction. That is, according to the patient's main complaint information, and according to the interactive medical consultation answers between the user and the system, and finally according to the generated medical consultation sequence and the user's answer, a preliminary disease diagnosis is given, thereby optimizing the rationality of the relevant disease medical consultation logic and the user experience while ensuring the accuracy of the medical consultation results.

[0016] In addition, the intelligent medical consultation method according to the above embodiment of the present invention may also have the following additional technical features:

[0017] Optionally, the predicting of a priori probability of a disease according to the main complaint of the target object to obtain a priori probability of a disease for the target object includes:

[0018] Based on a combined model of text feature extraction and disease prior probability prediction, text features are extracted from the target object's main complaint, and disease prior probability is predicted based on the extracted text features to obtain the disease prior probability for the target object.

[0019] Optionally, the combined model of text feature extraction and disease prior probability prediction is obtained in advance by the following steps:

[0020] Based on the medical entity database, the patient complaint content samples to be trained are segmented to obtain medical-related words;

[0021] Marking the medical-related words to obtain the categories to which the words belong;

[0022] Obtaining medical text data samples as training data, and using medical conversation data as labels for the training data;

[0023] According to the medical-related words, the categories to which the words belong and the labeled training data, a combined deep network model based on a bidirectional long short-term memory unit and a convolutional neural network is trained to obtain a combined model of text feature extraction and disease prior probability prediction.

[0024] Optionally, the multi-agent model based on reinforcement learning is pre-established by the following steps:

[0025] Obtain multiple consultation information streams between doctors and patients, and obtain disease diagnosis labels corresponding to each of the consultation information streams;

[0026] Establishing a reward function according to the multiple medical inquiry information streams and the disease diagnosis label corresponding to each of the medical inquiry information streams;

[0027] Build a corresponding intelligent agent for all the consultation information flows related to each disease diagnosis label;

[0028] Based on the reward function, reinforcement learning is performed on the agent corresponding to each disease to obtain the optimal strategy for each disease agent;

[0029] The optimal strategy of each disease agent is used as the strategy of each agent to complete the establishment of the multi-agent model based on reinforcement learning.

[0030] Optionally, the interactive consultation based on the strategy of each intelligent agent and according to the interactive selection of the target object for the question to be asked in the next interaction includes:

[0031] S1, providing the target object with questions to be asked for the next interaction based on each of the agent strategies;

[0032] S2, updating the current symptom state in each agent strategy according to the target object's interactive selection of the question to be asked in the next interaction;

[0033] S3, providing the target object with new questions to be asked for the next interaction again according to the updated current symptom status through each agent strategy;

[0034] S4, repeatedly executing steps S2 and S3 until the action probability of the target disease among the at least one candidate disease is greater than or equal to a preset threshold, outputting the entire interactive consultation sequence for the main complaint of the target object, and outputting the target disease with the action probability greater than or equal to the preset threshold as the final predicted disease.

[0035] To achieve the above-mentioned purpose, the intelligent medical consultation system proposed in the second embodiment of the present invention includes:

[0036] A disease priori prediction module is used to predict the disease priori probability according to the main complaint of the target object, and obtain the disease priori probability for the target object;

[0037] A candidate disease determination module, configured to determine at least one candidate disease based on the ranking of the disease prior probabilities;

[0038] An agent strategy determination module, used to determine at least one agent strategy corresponding to at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the output of questions to be asked in the next interaction under the current symptom state;

[0039] The interactive consultation module is used to conduct interactive consultation based on the strategies of each intelligent agent and according to the interactive selection of the target object for the questions to be asked in the next interaction.

[0040] The intelligent medical consultation system of the embodiment of the present invention can predict the disease prior probability according to the main complaint of the target object, obtain the disease prior probability for the target object, and then determine at least one candidate disease based on the ranking of the disease prior probability, and determine at least one agent strategy corresponding to at least one candidate disease from the pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the output of the questions to be asked in the next interaction under the current symptom state, and then, based on each agent strategy, interactive medical consultation is performed according to the interactive selection of the questions to be asked by the target object for the next interaction. That is, according to the patient's main complaint information, and according to the interactive medical consultation answers between the user and the system, and finally according to the generated medical consultation sequence and the user's answer, a preliminary disease diagnosis is given, thereby optimizing the rationality of the relevant disease medical consultation logic and the user experience while ensuring the accuracy of the medical consultation results.

[0041] In addition, the intelligent medical consultation system according to the above embodiment of the present invention may also have the following additional technical features:

[0042] Optionally, the disease a priori prediction module is specifically used for:

[0043] Based on a combined model of text feature extraction and disease prior probability prediction, text features are extracted from the target object's main complaint, and disease prior probability is predicted based on the extracted text features to obtain the disease prior probability for the target object.

[0044] Optionally, the intelligent medical consultation system further includes:

[0045] A combined model building module, used to pre-build a combined model of text feature extraction and disease prior probability prediction;

[0046] Wherein, the combined model building module includes:

[0047] A word segmentation unit is used to perform word segmentation processing on the patient's main complaint content sample to be trained based on the medical entity library to obtain medical-related words;

[0048] A marking unit, used to mark the medical-related words to obtain the category to which the words belong;

[0049] A training data generating unit, used to obtain medical text data samples as training data, and use medical conversation data as labels for the training data;

[0050] A model training unit is used to train a combined deep network model based on a bidirectional long short-term memory unit and a convolutional neural network according to the medical-related words, the categories to which the words belong and the labeled training data, so as to obtain a combined model of text feature extraction and disease prior probability prediction.

[0051] Optionally, the intelligent medical consultation system further includes:

[0052] A multi-agent model building module, used to pre-build the multi-agent model based on reinforcement learning;

[0053] Wherein, the multi-agent model building module includes:

[0054] A sample acquisition unit, used to acquire multiple consultation information streams between doctors and patients, and to acquire a disease diagnosis label corresponding to each of the consultation information streams;

[0055] A reward function establishing unit, used for establishing a reward function according to the multiple medical consultation information streams and the disease diagnosis label corresponding to each of the medical consultation information streams;

[0056] An agent building unit is used to build a corresponding agent for all the consultation information flows related to each disease diagnosis label;

[0057] A reinforcement learning unit is used to perform reinforcement learning on the agent corresponding to each disease based on the reward function, obtain the optimal strategy of each disease agent, and use the optimal strategy of each disease agent as the strategy of each agent to complete the establishment of the multi-agent model based on reinforcement learning.

[0058] Optionally, the interactive consultation module is specifically used for:

[0059] S1, providing the target object with questions to be asked for the next interaction based on each of the agent strategies;

[0060] S2, updating the current symptom state in each agent strategy according to the target object's interactive selection of the question to be asked in the next interaction;

[0061] S3, providing the target object with new questions to be asked for the next interaction again according to the updated current symptom status through each agent strategy;

[0062] S4, repeatedly executing steps S2 and S3 until the action probability of the target disease among the at least one candidate disease is greater than or equal to a preset threshold, outputting the entire interactive consultation sequence for the main complaint of the target object, and outputting the target disease with the action probability greater than or equal to the preset threshold as the final predicted disease.

[0063] To achieve the above-mentioned purpose, the computer device proposed in the third aspect of the embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the intelligent diagnosis method described in the first aspect of the embodiment of the present invention is implemented.

[0064] To achieve the above-mentioned purpose, the fourth aspect of the present invention proposes a non-temporary computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the intelligent diagnosis method described in the second aspect of the present invention is implemented.

[0065] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0067] Figure 1 is a flow chart of an intelligent medical consultation method according to an embodiment of the present invention;

[0068] Figure 2 is a flowchart of establishing a combined model of text feature extraction and disease prior probability prediction according to an embodiment of the present invention;

[0069] Figure 3 is a flowchart of establishing the multi-agent model based on reinforcement learning according to an embodiment of the present invention;

[0070] Figure 4 is a flowchart of an intelligent medical consultation method based on reinforcement learning according to an embodiment of the present invention;

[0071] Figure 5 is a schematic diagram of the structure of an intelligent medical consultation system according to an embodiment of the present invention;

[0072] Figure 6 is a structural schematic diagram of an intelligent medical consultation system according to another embodiment of the present invention;

[0073] Figure 7 is a structural diagram of an intelligent medical consultation system according to yet another embodiment of the present invention;

[0074] Figure 8 A block diagram of an exemplary computer device suitable for implementing embodiments of the present invention is shown. DETAILED DESCRIPTION

[0075] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0076] In the era of information society, every traditional field is impacted by emerging technologies, and machine learning and artificial intelligence technologies have achieved milestone breakthroughs in various fields. In the field of Go, AlphaGo (an artificial intelligence program for Go) based on a reinforcement learning model defeated the world's top human Go players. Go, a field that was once thought to be impossible to subvert, has been broken under the impact of emerging technologies. Intelligent medical consultation was originally considered impossible, but now with the advent of the era of big data and artificial intelligence, more and more large companies and research institutions have begun to enter the field of Internet medical care and intelligent medical consultation. IBM Watson robot is considered to be the first company to enter the field of intelligent medical consultation. At present, major companies have their own products in the field of Internet medical care, but in the field of intelligent medical consultation and diagnosis, especially in terms of how to simulate doctor consultation and the natural interaction between the intelligent medical consultation system and the patient, it is still difficult to be satisfactory.

[0077] Medical intelligent consultation is different from disease prediction. It not only predicts diseases based on known patient symptoms, but also analyzes the patient's main complaint and simulates the doctor asking about possible symptoms. Then, based on the patient's interactive answers, it simulates the consultation again until the correct disease is finally given. At present, it is not difficult for the intelligent consultation system to predict the corresponding disease given the typical symptoms. However, how to make the system, like a doctor, conduct consultations step by step based on the patient's main complaint, simulate the consultation process, and finally get the correct disease diagnosis results is very difficult in the entire intelligent medical field, but it is very important.

[0078] The ideal intelligent consultation method should ensure that the logic of the consultation is reasonable and consistent with that of professional doctors, while avoiding too many consultation rounds. In addition, based on the consultation flow sequence and the patient's interactive questions and answers, the intelligent consultation system can give an accurate disease diagnosis. However, it is often not easy to have these at the same time. For example, the current intelligent consultation system, on the one hand, the current method based on deep learning (CNN, LSTM), when all observed symptoms are known, the model can give a relatively high disease prediction accuracy, but usually, we cannot observe all the symptoms of the patient, and the model based on deep learning is difficult to give a reasonable consultation process due to its unexplainability. On the other hand, the method based on probabilistic graphical model (PGM) and information gain can conduct effective consultation based on the disease-symptom transition probability matrix (not perfect), but the method based on probabilistic graphical model often has a high computational complexity in the model inference part. At the same time, the disease-symptom transfer matrix of the probabilistic graphical model often requires professional doctors to label. This expert system-like model faces great challenges in disease expansion. Therefore, a new method is needed to make a breakthrough in the field of intelligent consultation.

[0079] To this end, the present invention proposes a new method, system, computer device and computer-readable storage medium for medical intelligent consultation based on a large amount of simulated medical consultation flow dialogue data, as well as consultation dialogue data parsed from medical literature and medical related cases. With the breakthrough of AlphaGo in the field of Go, more and more fields use models and methods based on reinforcement learning, especially considering the field of human-computer interaction, where reinforcement learning has its natural advantages. Considering the problem of medical intelligent consultation, the logic of the consultation should simulate the doctor as much as possible, and the interactivity between the patient and the intelligent consultation system should be considered. The reinforcement learning model allows the system to learn the optimal consultation logic and accurately give the diagnosis result. Unlike traditional methods, the present invention will consider the accuracy of the consultation logic and disease diagnosis at the same time, and transform the intelligent consultation into the training of a reinforcement model to optimize the model. The intelligent consultation method proposed in the embodiment of the present invention can be applied to multiple scenarios, including but not limited to: 1) medical intelligent auxiliary consultation; 2) medical intelligent triage and guidance; 3) medical intelligent consultation diagnosis and solution, etc.

[0080] The following describes the intelligent medical consultation method, system, computer device, and computer-readable storage medium according to embodiments of the present invention with reference to the accompanying drawings.

[0081] Figure 1It is a flow chart of an intelligent medical consultation method according to an embodiment of the present invention. It should be noted that the intelligent medical consultation method according to the embodiment of the present invention can be applied to the intelligent medical consultation system according to the embodiment of the present invention, and the intelligent medical consultation method can be configured on a computer device.

[0082] like Figure 1 As shown, the intelligent consultation method may include:

[0083] S110, predicting the disease prior probability according to the main complaint of the target object, and obtaining the disease prior probability for the target object.

[0084] The target object can be understood as a patient. Optionally, in an embodiment of the present invention, the intelligent medical consultation method of the embodiment of the present invention can provide a user with an input interface for the main complaint content, and the user can input the main complaint content of the target object through the input interface by text or voice input, so that the main complaint content of the target object can be obtained through the input interface.

[0085] After obtaining the target subject's chief complaint, that is, obtaining the target subject's self-reported symptoms or (and) signs, nature, and duration, etc., the target subject's chief complaint content can be segmented and feature extracted, and the disease prior probability can be predicted based on the extracted text features to obtain the disease prior probability for the target subject.

[0086] As an example, a pre-established combined model based on text feature extraction and disease prior probability prediction can be used to extract text features from the target object's chief complaint, and the disease prior probability can be predicted based on the extracted text features to obtain the disease prior probability for the target object.

[0087] Optionally, the text feature extraction module in the combined model is used to perform word segmentation on the main complaint content of the target object. Different from the word segmentation of traditional texts, the embodiment of the present invention can establish a special medical entity database for texts in the medical field. Based on the sentence segmentation of the medical entity library, the main complaint content of the target object can be segmented into multiple word segments, and corresponding text features can be extracted based on the multiple word segments. Then, the disease prior probability prediction module in the combined model can be used to predict the disease prior probability based on the extracted text features to obtain the disease prior probability for the target object, thereby achieving disease prediction within a certain range, locking the possible diseases of the target object within a certain range, and narrowing the range for the subsequent reinforcement learning agent strategy output.

[0088] That is to say, the embodiment of the present invention can perform text feature extraction and disease prior prediction on the main complaint content of the target object by utilizing a combination of two deep neural networks, which can achieve disease prediction within a certain range, lock the possible diseases of the target object within a certain range, and narrow the range for the subsequent reinforcement learning agent strategy output. It should be noted that there may be many ways to combine the two deep neural networks. The present invention provides a specific possible combination method, namely, a combination method using a bidirectional long short-term memory unit and a convolutional neural network. A possible implementation method of the combined model of text feature extraction and disease prior probability prediction established by using a combination of a bidirectional long short-term memory unit and a convolutional neural network will be given below.

[0089] As an example of a possible implementation, Figure 2 As shown, the combined model of text feature extraction and disease prior probability prediction can be pre-established by the following steps:

[0090] S210, performing word segmentation processing on the patient's main complaint content sample to be trained based on the medical entity library to obtain words related to medical treatment;

[0091] It can be understood that, unlike the word segmentation of traditional texts, for texts in the medical field, the embodiment of the present invention can establish a special medical entity database. Based on the sentence segmentation of the medical entity database, the patient complaint content samples to be trained can be segmented to obtain corresponding words.

[0092] S220, marking the medical-related words to obtain the categories to which the words belong;

[0093] Optionally, based on the sentence segmentation of the medical entity library, while segmenting the patient's main complaint content samples to be trained into corresponding words, the medical-related words can also be marked with corresponding categories, for example, "vomiting" is marked as "symptom", "pneumonia" is marked as "disease", etc.

[0094] S230, obtaining a medical text data sample as training data, and using the medical conversation data as a label of the training data;

[0095] S240, according to the medical-related words, the categories to which the words belong and the labeled training data, a combined deep network model based on a bidirectional long short-term memory unit and a convolutional neural network is trained to obtain a combined model of text feature extraction and disease prior probability prediction.

[0096] Optionally, after obtaining the medical-related words, the categories to which the words belong, and the labeled training data, the medical-related words and training data can be mapped to vector expressions using word vectorization technology. Then, using the deep learning framework, a combined deep network model based on Bi-LSTM (bidirectional long short-term memory unit) and CNN (convolutional neural network) is built.

[0097] It can be seen that the embodiment of the present invention is based on the combined deep network model of Bi-LSTM and CNN. Compared with the framework of traditional neural networks (such as DNN, RNN), on the one hand, it takes into account the order relationship between words in a sentence, which is more in line with the basic assumption of natural language processing (i.e., word order affects the expression of semantics). On the other hand, the method based on long short-term memory unit (LSTM) effectively solves the problems of gradient explosion and gradient vanishing in traditional recurrent neural networks (RNN), making model training more stable. Since the semantic features of words and sentences are taken into account, the combined model of text feature extraction and disease prior probability prediction obtained based on the above training method can predict diseases within a certain range, lock the possible diseases of patients within a certain range, and narrow the range for the subsequent reinforcement learning agent strategy output.

[0098] S120, determining at least one candidate disease based on the ranking of disease prior probabilities.

[0099] Optionally, after obtaining the disease prior probability for the target object, the obtained disease prior probability can be sorted, and at least one candidate disease can be determined based on the sorting of the disease prior probability. For example, based on the sorting of the disease prior probability, the disease with the probability ranked in the top N can be selected as the at least one candidate disease, where N is a positive integer; for another example, based on the sorting of the disease prior probability, the disease with a probability greater than or equal to a certain threshold can be selected as the at least one candidate disease. In this way, a candidate set for coarse-grained disease prediction can be given within a certain range.

[0100] S130, determining at least one agent strategy associated with at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the questions to be asked in the next interaction under the current symptom state.

[0101] Optionally, after obtaining a candidate set of disease predictions within a certain range, interactive consultation logic is required to gradually determine the disease. In an embodiment of the present invention, a multi-agent model can be established based on a multi-agent reinforcement learning method, and the multi-agent model has multiple agent strategies, each agent strategy corresponding to an interactive consultation logic for a disease. To this end, in this step, after obtaining the at least one candidate disease, at least one agent strategy corresponding to the at least one candidate disease can be determined from the pre-established multi-agent model based on reinforcement learning, so that an interactive consultation on the main complaint of the target object can be implemented based on the at least one agent strategy.

[0102] It should be noted that, in the embodiment of the present invention, the multi-agent model based on reinforcement learning may be pre-established. Figure 3 As shown, the multi-agent model based on reinforcement learning can be pre-established by the following steps:

[0103] S310, obtaining multiple consultation information streams between doctors and patients, and obtaining disease diagnosis labels corresponding to each of the consultation information streams;

[0104] For example, the existing medical consultation dialogue database may include but is not limited to medical books, medical literature, medical reports, case analysis, and may also include simulated patient and doctor consultation data. Since these medical data are carefully compiled by relevant medical experts and have a standardized annotated format, the embodiment of the present invention can extract the doctor and patient consultation information flow from these databases and obtain the final disease diagnosis label of the consultation information flow.

[0105] S320, establishing a reward function according to the multiple medical inquiry information streams and the disease diagnosis label corresponding to each of the medical inquiry information streams;

[0106] Optionally, based on a multi-agent reinforcement learning method, the reward function is set through reinforcement learning and according to the multiple medical consultation information flows and the disease diagnosis label corresponding to each of the medical consultation information flows.

[0107] S330, constructing a corresponding intelligent agent for all the inquiry information flows related to each disease diagnosis label;

[0108] Optionally, a reinforcement learning-based model is established. For the multi-agent-based reinforcement learning model, an agent is constructed for all the consultation flow information related to each disease diagnosis label.

[0109] S340, based on the reward function, performing reinforcement learning on the agent corresponding to each disease to obtain the optimal strategy for each disease agent;

[0110] Optionally, based on the reward function, reinforcement learning is performed on the agent corresponding to each disease, and finally the optimal strategy (i.e., Q * (a,S) function), where S represents a set of known symptoms (s1,s2,…s n ), a represents the optimal action under the current symptom state, a∈S unknown , S unknown Represents a set of symptoms that have not yet been queried.

[0111] S350, taking the optimal strategy of each disease agent as the strategy of each agent, and completing the establishment of the multi-agent model based on reinforcement learning.

[0112] Thus, the establishment of the multi-agent model based on reinforcement learning is realized through the above steps S310 to S350. The embodiment of the present invention optimizes the intelligent consultation logic based on the reinforcement learning of multi-agents, that is, according to the consultation data of the disease doctor, for each disease, an agent is constructed to perform reinforcement learning, and the optimal strategy Q of each agent is obtained. * (a,S) function, thereby completing the establishment of the multi-agent model based on reinforcement learning.

[0113] S140, based on each agent strategy, interactive questioning is performed according to the target object's interactive selection of questions to be asked in the next interaction.

[0114] Optionally, after determining at least one agent strategy corresponding to at least one candidate disease, the output can be performed according to the at least one agent strategy, and the next consultation symptom in each agent strategy (i.e., the question to be asked in the next interaction) can be given. Then, according to the interactive selection of the target object, the current state is updated, and then the agent outputs the next round of consultation symptoms again according to the updated state. This iteration is carried out until the final disease probability meets a certain threshold or reaches a certain interruption mechanism to exit the intelligent consultation. As an example, the interactive consultation can be completed in the following way:

[0115] S1, providing the target object with questions to be asked for the next interaction based on each of the agent strategies;

[0116] S2, updating the current symptom state in each agent strategy according to the target object's interactive selection of the question to be asked in the next interaction;

[0117] S3, providing the target object with new questions to be asked for the next interaction again according to the updated current symptom status through each agent strategy;

[0118] S4, repeatedly executing steps S2 and S3 until the action probability of the target disease among the at least one candidate disease is greater than or equal to a preset threshold, outputting the entire interactive consultation sequence for the main complaint of the target object, and outputting the target disease with the action probability greater than or equal to the preset threshold as the final predicted disease.

[0119] That is to say, after obtaining at least one intelligent agent strategy corresponding to at least one candidate disease, the intelligent agent strategy can be output, and the next round of symptom consultation in each intelligent agent strategy can be given. The target object can choose the next round of symptom consultation, and the selected option is used as the question to be asked in the next interaction to complete a consultation interaction, and the selected question to be asked in the next interaction is used as the updated current state. Then each intelligent agent strategy provides the target object with new questions to be asked in the next interaction based on the new symptom state. In this way, the interaction between the patient and the "doctor" (i.e., the intelligent consultation system) is iteratively completed. Finally, when the action probability of the disease is greater than or equal to the preset threshold, the intelligent consultation ends, and the interactive consultation sequence and the final predicted disease are given to complete the entire disease consultation and prediction process.

[0120] In order to make those skilled in the art understand the present invention more clearly, examples are given below.

[0121] For example, Figure 4 As shown, the main complaint content of the target object can be obtained, and then the main complaint content can be segmented based on the medical entity library to obtain medical-related words, and these medical-related words are vectorized using word vectorization technology to obtain corresponding word vectors, and text feature extraction is performed on the obtained word vectors through a combined model based on text feature extraction and disease prior probability prediction, and disease prior probability prediction is performed based on the extracted text features to obtain the disease prior probability for the target object, thereby giving a candidate set of coarse-grained disease prediction within a certain range. Afterwards, the agent strategies corresponding to the candidate diseases in the candidate set can be determined from the pre-established multi-agent model based on reinforcement learning. Since the multi-agent-based reinforcement learning model of the embodiment of the present invention will learn a strategy for an agent for each disease, and in reality there are many diseases, too many candidate agents may make it impossible to give a final strategy, and output the final selected action a (a∈S unknown ).

[0122] Therefore, the embodiment of the present invention can combine the two methods of disease prior prediction based on combined deep neural network and multi-agent reinforcement learning optimization intelligent consultation logic, and give the final strategy generation. The specific implementation process can be as follows: First, the disease prior probability of the target object's chief complaint content can be predicted by a combined model based on text feature extraction and disease prior probability prediction. After obtaining the prior probability Pprior of the predicted disease, the top k diseases (such as the diseases ranked in the top k in prior probability) can be selected according to pre-set parameters; then, according to the selected top k diseases, the strategy Q of the corresponding reinforcement learning generation agent is associated d (a,S), according to the current state set S, each agent according to Q d The function selects the best action a in the current state d , and the corresponding Q(a d ,S) value q d , then consider q d Normalized to get the corresponding action probability p d , for example, using the softmax function as follows:

[0123]

[0124] In the above formula, j represents a possible action (i.e., a symptom that may be asked about). Here, the exponential values ​​of the possible values ​​of each action j are summed up and used as the denominator to achieve normalization.

[0125] Finally, according to the p vector composed of the top k diseases, the final action a is output according to probability sampling * (a∈S unknown ). The target object responds to the next symptom a given by the system. * , answer, complete a consultation interaction, and change the current state set (S t ->S t+1 ), then each agent according to the new state S t+1 ), calculate Q d (a,S t+1 ), and then iterates the interaction between the patient and the "doctor" (i.e., the intelligent consultation system) in sequence. Finally, when the probability of the disease reaches the specified threshold τ, the intelligent consultation ends, and the interactive consultation sequence and the final predicted disease d are given, completing the entire disease consultation and prediction process.

[0126] Experiments have proved that the intelligent medical consultation method based on reinforcement learning in the embodiment of the present invention takes into account how to simulate the doctor's medical consultation logic during the training process and takes into account the interaction between the patient and the medical consultation system, thereby optimizing the medical consultation experience while ensuring accuracy, making the medical consultation more in line with the doctor's logic.

[0127] The intelligent medical consultation method of the embodiment of the present invention predicts the disease prior probability according to the main complaint of the target object, obtains the disease prior probability for the target object, and then determines at least one candidate disease based on the ranking of the disease prior probability, and determines at least one agent strategy corresponding to at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the output of the questions to be asked in the next interaction under the current symptom state, and then, based on each agent strategy, interactive medical consultation is performed according to the interactive selection of the questions to be asked by the target object for the next interaction. That is, according to the patient's main complaint information, and according to the interactive medical consultation answers between the user and the system, and finally according to the generated medical consultation sequence and the user's answer, a preliminary disease diagnosis is given, thereby optimizing the rationality of the relevant disease medical consultation logic and the user experience while ensuring the accuracy of the medical consultation results.

[0128] Corresponding to the intelligent medical consultation methods provided in the above-mentioned embodiments, an embodiment of the present invention also provides an intelligent medical consultation system. Since the intelligent medical consultation system provided in the embodiment of the present invention corresponds to the intelligent medical consultation methods provided in the above-mentioned embodiments, the implementation methods of the aforementioned intelligent medical consultation methods are also applicable to the intelligent medical consultation system provided in this embodiment and will not be described in detail in this embodiment. Figure 5 FIG. 1 is a schematic diagram of the structure of an intelligent medical consultation system according to an embodiment of the present invention. Figure 5 As shown, the intelligent consultation system 500 may include: a disease priori prediction module 510, a candidate disease determination module 520, an intelligent agent strategy determination module 530 and an interactive consultation module 540.

[0129] Specifically, the disease prior prediction module 510 is used to predict the disease prior probability according to the main complaint of the target object, and obtain the disease prior probability for the target object. As an example, the disease prior prediction module 510 extracts text features from the main complaint of the target object based on a combined model of text feature extraction and disease prior probability prediction, and predicts the disease prior probability according to the extracted text features to obtain the disease prior probability for the target object.

[0130] As an example, Figure 6 As shown, the intelligent medical consultation system 500 may also include: a combined model building module 550, which is used to pre-build the combined model of text feature extraction and disease prior probability prediction. Figure 6As shown, the combined model building module 550 may include: a word segmentation unit 551, a labeling unit 552, a training data generation unit 553 and a model training unit 554. The word segmentation unit 551 is used to perform word segmentation processing on the patient's main complaint content sample to be trained based on the medical entity library to obtain medical-related words; the labeling unit 552 is used to label the medical-related words to obtain the category to which the words belong; the training data generation unit 553 is used to obtain medical text data samples as training data, and use medical conversation data as labels for the training data; the model training unit 554 is used to perform model training on the combined deep network model based on the bidirectional long short-term memory unit and the convolutional neural network according to the medical-related words, the category to which the words belong and the labeled training data, to obtain the combined model of text feature extraction and disease prior probability prediction.

[0131] The candidate disease determination module 520 is used to determine at least one candidate disease based on the ranking of the disease prior probabilities.

[0132] The agent strategy determination module 530 is used to determine at least one agent strategy corresponding to at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the question to be asked in the next interaction under the current symptom state. As an example, Figure 7 As shown, the intelligent medical consultation system 500 may also include: a multi-agent model building module 560, which is used to pre-build the multi-agent model based on reinforcement learning. Figure 7 As shown, the multi-agent model building module 560 may include: a sample acquisition unit 561, a reward function establishment unit 562, an agent construction unit 563 and a reinforcement learning unit 564. Among them, the sample acquisition unit 561 is used to obtain multiple consultation information flows between doctors and patients, and obtain the disease diagnosis label corresponding to each of the consultation information flows; the reward function establishment unit 562 is used to establish a reward function according to the multiple consultation information flows and the disease diagnosis label corresponding to each of the consultation information flows; the agent construction unit 563 is used to construct a corresponding agent for all consultation information flows related to each disease diagnosis label; the reinforcement learning unit 564 is used to perform reinforcement learning on the agent corresponding to each disease based on the reward function, obtain the optimal strategy of each disease agent, and use the optimal strategy of each disease agent as the strategy of each agent to complete the establishment of the multi-agent model based on reinforcement learning.

[0133] The interactive consultation module 540 is used to perform interactive consultation based on the strategy of each agent and the interactive selection of the target object for the question to be asked in the next interaction. As an example, the interactive consultation module 540 can implement interactive consultation in the following manner:

[0134] S1, providing the target object with questions to be asked for the next interaction based on each of the agent strategies;

[0135] S2, updating the current symptom state in each agent strategy according to the target object's interactive selection of the question to be asked in the next interaction;

[0136] S3, providing the target object with new questions to be asked for the next interaction again according to the updated current symptom status through each agent strategy;

[0137] S4, repeatedly executing steps S2 and S3 until the action probability of the target disease among the at least one candidate disease is greater than or equal to a preset threshold, outputting the entire interactive consultation sequence for the main complaint of the target object, and outputting the target disease with the action probability greater than or equal to the preset threshold as the final predicted disease.

[0138] The intelligent medical consultation system of the embodiment of the present invention can predict the disease prior probability according to the main complaint of the target object, obtain the disease prior probability for the target object, and then determine at least one candidate disease based on the ranking of the disease prior probability, and determine at least one agent strategy corresponding to at least one candidate disease from the pre-established multi-agent model based on reinforcement learning, wherein each agent strategy is used to indicate the output of the questions to be asked in the next interaction under the current symptom state, and then, based on each agent strategy, interactive medical consultation is performed according to the interactive selection of the questions to be asked by the target object for the next interaction. That is, according to the patient's main complaint information, and according to the interactive medical consultation answers between the user and the system, and finally according to the generated medical consultation sequence and the user's answer, a preliminary disease diagnosis is given, thereby optimizing the rationality of the relevant disease medical consultation logic and the user experience while ensuring the accuracy of the medical consultation results.

[0139] In order to implement the above embodiments, the present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the intelligent diagnosis method described in any one of the above embodiments of the present invention is implemented.

[0140] In order to implement the above embodiments, the present invention further proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the intelligent medical consultation method described in any of the above embodiments of the present invention is implemented.

[0141] Figure 8 A block diagram of an exemplary computer device suitable for implementing embodiments of the present invention is shown. Figure 8 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0142] like Figure 8 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0143] The bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnection (PCI) bus.

[0144] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0145] The memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 8not shown, usually called a "hard drive"). Although Figure 8 Not shown, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a compact disc read only memory (CD-ROM), a digital versatile disc read only memory (DVD-ROM), or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present application.

[0146] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0147] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed via an input / output (I / O) interface 22. In addition, the computer device 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the computer device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the terminal device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0148] The processing unit 16 executes various functional applications and data processing by running the programs stored in the system memory 28, such as implementing the intelligent medical consultation method mentioned in the above embodiment.

[0149] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0150] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0151] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0152] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0153] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0154] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0155] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0156] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An intelligent medical consultation method, characterized in that: The steps include: Predicting the disease prior probability according to the main complaint of the target object to obtain the disease prior probability for the target object; Determining at least one candidate disease based on the ranking of the disease prior probabilities; At least one agent strategy associated with at least one candidate disease is determined from a pre-established multi-agent model based on reinforcement learning, wherein the multi-agent model is established in the following manner: according to the consultation data in the medical consultation dialogue database, for each disease, an agent is constructed to perform reinforcement learning, and the optimal strategy Q of each agent is obtained * (a,S) function, where S represents the set of known symptoms (s1,s2,…s n ), a represents the optimal action under the current symptom state, a∈S unknown , S unknown represents a set of symptoms that have not been asked about yet; the consultation data in the medical consultation dialogue database includes medical books, medical literature, medical reports, case analysis, and simulated consultation data between patients and doctors; each agent strategy corresponds to the interactive consultation logic of a disease, and each agent strategy is used to indicate the output of questions to be asked in the next interaction under the current symptom status; Based on each of the agent strategies, interactive consultation is performed according to the interactive selection of the target object for the question to be raised in the next interaction; wherein, based on each of the agent strategies, interactive consultation is performed according to the interactive selection of the target object for the question to be raised in the next interaction, including: S1, providing the target object with questions to be asked for the next interaction based on each of the agent strategies; S2, updating the current symptom state in each agent strategy according to the target object's interactive selection of the question to be asked in the next interaction; S3, providing the target object with new questions to be asked for the next interaction again according to the updated current symptom status through each agent strategy; S4, repeatedly executing steps S2 and S3 until the action probability of the target disease in the at least one candidate disease is greater than or equal to a preset threshold, outputting the entire interactive consultation sequence for the main complaint of the target subject, and outputting the target disease whose action probability is greater than or equal to the preset threshold as the final predicted disease; Among them, for each of the agent strategies, the optimal selected action a in the current state is selected according to the optimal strategy function of the agent strategy d and the corresponding Q * (a d ,S) value q d , use the softmax function to * (a d ,S) value q d Normalized to get the corresponding action probability P d ; Wherein, the softmax function is expressed as follows: Where j represents a candidate action or a symptom of a candidate inquiry, and exp represents an exponential function.

2. The intelligent medical consultation method according to claim 1, characterized in that: The predicting of the disease prior probability according to the main complaint of the target object to obtain the disease prior probability for the target object includes: Based on a combined model of text feature extraction and disease prior probability prediction, text features are extracted from the target object's main complaint, and disease prior probability is predicted based on the extracted text features to obtain the disease prior probability for the target object.

3. The intelligent medical consultation method according to claim 2, characterized in that: The combined model of text feature extraction and disease prior probability prediction is obtained in advance through the following steps: Based on the medical entity database, the patient complaint content samples to be trained are segmented to obtain medical-related words; Marking the medical-related words to obtain the categories to which the words belong; Obtaining medical text data samples as training data, and using medical conversation data as labels for the training data; According to the medical-related words, the categories to which the words belong and the labeled training data, a combined deep network model based on a bidirectional long short-term memory unit and a convolutional neural network is trained to obtain a combined model of text feature extraction and disease prior probability prediction.

4. The intelligent medical consultation method according to claim 1, characterized in that: The multi-agent model based on reinforcement learning is pre-established by the following steps: Obtain multiple consultation information streams between doctors and patients, and obtain disease diagnosis labels corresponding to each of the consultation information streams; Establishing a reward function according to the multiple medical inquiry information streams and the disease diagnosis label corresponding to each of the medical inquiry information streams; Build a corresponding intelligent agent for all the consultation information flows related to each disease diagnosis label; Based on the reward function, reinforcement learning is performed on the agent corresponding to each disease to obtain the optimal strategy for each disease agent; The optimal strategy of each disease agent is used as the strategy of each agent to complete the establishment of the multi-agent model based on reinforcement learning.

5. An intelligent medical consultation system, characterized in that: include: A disease priori prediction module is used to predict the disease priori probability according to the main complaint of the target object, and obtain the disease priori probability for the target object; A candidate disease determination module, configured to determine at least one candidate disease based on the ranking of the disease prior probabilities; The agent strategy determination module is used to determine at least one agent strategy corresponding to at least one candidate disease from a pre-established multi-agent model based on reinforcement learning, wherein the multi-agent model is established in the following manner: according to the consultation data in the medical consultation dialogue database, for each disease, an agent is constructed to perform reinforcement learning, and the optimal strategy Q of each agent is obtained * (a,S) function, where S represents the set of known symptoms (s1,s2,…s n ), a represents the optimal action under the current symptom state, a∈S unknown , S unknown represents a set of symptoms that have not been asked about yet; the consultation data in the medical consultation dialogue database includes medical books, medical literature, medical reports, case analysis, and simulated consultation data between patients and doctors; each agent strategy corresponds to the interactive consultation logic of a disease, and each agent strategy is used to indicate the output of questions to be asked in the next interaction under the current symptom status; An interactive consultation module is used to conduct interactive consultation based on the strategy of each agent and according to the interactive selection of the target object for the question to be asked in the next interaction; wherein the interactive consultation module is specifically used to: S1, providing the target object with questions to be asked for the next interaction based on each of the agent strategies; S2, updating the current symptom state in each agent strategy according to the target object's interactive selection of the question to be asked in the next interaction; S3, providing the target object with new questions to be asked for the next interaction again according to the updated current symptom status through each agent strategy; S4, repeatedly executing steps S2 and S3 until the action probability of the target disease in the at least one candidate disease is greater than or equal to a preset threshold, outputting the entire interactive consultation sequence for the main complaint of the target subject, and outputting the target disease whose action probability is greater than or equal to the preset threshold as the final predicted disease; Among them, for each of the agent strategies, the optimal selected action a in the current state is selected according to the optimal strategy function of the agent strategy d and the corresponding Q * (a d ,S) value q d , use the softmax function to * (a d ,S) value q d Normalized to get the corresponding action probability P d ; Wherein, the softmax function is expressed as follows: Where j represents a candidate action or a symptom of a candidate inquiry, and exp represents an exponential function.

6. The intelligent medical consultation system according to claim 5, characterized in that: The disease priori prediction module is specifically used for: Based on a combined model of text feature extraction and disease prior probability prediction, text features are extracted from the target object's main complaint, and disease prior probability is predicted based on the extracted text features to obtain the disease prior probability for the target object.

7. The intelligent medical consultation system according to claim 6, characterized in that: Also includes: A combined model building module, used to pre-build a combined model of text feature extraction and disease prior probability prediction; Wherein, the combined model building module includes: A word segmentation unit is used to perform word segmentation processing on the patient's main complaint content sample to be trained based on the medical entity library to obtain medical-related words; A marking unit, used to mark the medical-related words to obtain the category to which the words belong; A training data generating unit, used to obtain medical text data samples as training data, and use medical conversation data as labels for the training data; A model training unit is used to train a combined deep network model based on a bidirectional long short-term memory unit and a convolutional neural network according to the medical-related words, the categories to which the words belong and the labeled training data, so as to obtain a combined model of text feature extraction and disease prior probability prediction.

8. The intelligent medical consultation system according to claim 5, characterized in that: Also includes: A multi-agent model building module, used to pre-build the multi-agent model based on reinforcement learning; Wherein, the multi-agent model building module includes: A sample acquisition unit, used to acquire multiple consultation information streams between doctors and patients, and to acquire a disease diagnosis label corresponding to each of the consultation information streams; A reward function establishing unit, used for establishing a reward function according to the multiple medical consultation information streams and the disease diagnosis label corresponding to each of the medical consultation information streams; An agent building unit is used to build a corresponding agent for all the consultation information flows related to each disease diagnosis label; A reinforcement learning unit is used to perform reinforcement learning on the agent corresponding to each disease based on the reward function, obtain the optimal strategy of each disease agent, and use the optimal strategy of each disease agent as the strategy of each agent to complete the establishment of the multi-agent model based on reinforcement learning.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the intelligent medical consultation method according to any one of claims 1 to 4 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the intelligent medical consultation method according to any one of claims 1 to 4 is implemented.