A method of medical inquiry and related devices

Abstract

The application discloses a diagnosis method and related equipment thereof. The method comprises the following steps: after an electronic device acquires the illness self-description information of a patient to be diagnosed, the electronic device first determines a symptom set to be used according to the illness self-description information; then, according to the symptom set to be used, a symptom to be inquired is determined, so that after acquiring the inquiry feedback information of the patient to be diagnosed on the symptom to be inquired, the symptom set to be used is updated according to the inquiry feedback information, and the step of "determining the symptom to be inquired according to the symptom set to be used" is continuously executed until the diagnosis result of the patient to be diagnosed is determined according to the symptom set to be used when a first stop condition is reached, so that the diagnosis result can accurately represent the disease that the patient to be diagnosed is more likely to have, and the doctor can refer to the diagnosis result and other examination results to make more accurate disease diagnosis on the patient to be diagnosed, so that automatic diagnosis can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical treatment, in particular to a method for inquiry and a related device thereof. BACKGROUND

[0002] Inquiry is a method for diagnosing diseases by querying the occurrence, development, current symptoms and treatment history of diseases from patients and their informed persons through dialogue.

[0003] At present, the inquiry process can be guided by medical staff, for example, after a patient sees a doctor, the doctor can implement the inquiry process for the patient through dialogue, so that the doctor can refer to the inquiry result and other examination results (for example, urine test results, blood test results, computer tomography results, etc.) of the patient to diagnose the disease of the patient.

[0004] However, the above artificial inquiry process has defects, resulting in poor inquiry effect. SUMMARY

[0005] The main purpose of the embodiments of the present application is to provide a method for inquiry and a related device thereof, which can realize automatic inquiry, thereby effectively improving the inquiry effect.

[0006] The embodiments of the present application provide a method for inquiry, applied to an electronic device, the method comprising: after obtaining disease self-description information of a patient to be diagnosed, determining a symptom set to be used according to the disease self-description information; determining a symptom to be inquired according to the symptom set to be used; after obtaining inquiry feedback information of the patient to be diagnosed for the symptom to be inquired, updating the symptom set to be used according to the inquiry feedback information, and continuing to execute the step of determining the symptom to be inquired according to the symptom set to be used until a first stop condition is reached, and determining an inquiry result of the patient to be diagnosed according to the symptom set to be used.

[0007] The embodiments of the present application also provide an inquiry device, comprising: a first determining unit configured to determine a symptom set to be used according to disease self-description information of a patient to be diagnosed after obtaining the disease self-description information; a second determining unit configured to determine a symptom to be inquired according to the symptom set to be used; and a third determining unit configured to update the symptom set to be used according to inquiry feedback information of the patient to be diagnosed for the symptom to be inquired after obtaining the inquiry feedback information, continue to execute the step of determining the symptom to be inquired according to the symptom set to be used, and determine an inquiry result of the patient to be diagnosed according to the symptom set to be used when a first stop condition is reached.

[0008] The embodiment of the present application further provides a device, characterized in that the device comprises a processor, a memory and a system bus.

[0009] The processor and the memory are connected through the system bus.

[0010] The memory is used for storing one or more programs, wherein the one or more programs comprise instructions, which, when executed by the processor, cause the processor to perform any of the embodiments of the method for inquiring diagnosis provided by the present application.

[0011] The embodiment of the present application further provides a computer readable storage medium, wherein the computer readable storage medium stores instructions, and the instructions, when executed on a terminal device, cause the terminal device to perform any of the embodiments of the method for inquiring diagnosis provided by the present application.

[0012] The embodiment of the present application further provides a computer program product, wherein the computer program product, when executed on a terminal device, causes the terminal device to perform any of the embodiments of the method for inquiring diagnosis provided by the present application.

[0013] Based on the above technical solutions, the present application has the following beneficial effects:

[0014] In the technical solutions provided by the present application, for a patient to be diagnosed with a need for medical treatment, after an electronic device obtains the illness self-description information of the patient to be diagnosed, the electronic device first determines a symptom set to be used (for example, cough, sneezing, etc.) according to the illness self-description information; and then determines a symptom to be inquired (for example, headache) according to the symptom set to be used, so that the symptom to be inquired is different from any symptom in the symptom set to be used, thereby enabling the patient to be diagnosed to confirm whether the patient to be diagnosed has the symptom to be inquired, so as to update the symptom set to be used according to the inquiry feedback information (for example, information that the patient to be diagnosed has a headache or does not have a headache) of the patient to be diagnosed for the symptom to be inquired, and continue to perform the step of determining the symptom to be inquired according to the symptom set to be used until a first stop condition is reached, and then determine the inquiring diagnosis result of the patient to be diagnosed according to the symptom set to be used, so that the inquiring diagnosis result can represent the physical condition (for example, a disease that the patient to be diagnosed is likely to have, and / or various symptoms of the patient to be diagnosed, etc.), so that a doctor can refer to the inquiring diagnosis result and other examination results (for example, urine test results, blood test results, computed tomography results, etc.) to make more accurate disease diagnosis for the patient to be diagnosed.

[0015] It can be seen that the technical solution provided by the embodiment of the present application is realized by an electronic device (for example, a terminal device or a server, etc.), so that the medical personnel do not need to participate in the inquiry process provided by the embodiment of the present application, and thus the automatic inquiry can be realized, thereby the technical problem that the inquiry effect is poor due to the defects in the manual inquiry process can be overcome as much as possible, and the inquiry effect can be effectively improved.

[0016] In addition, the inquiry result of the patient to be diagnosed is determined according to the symptoms carried by the illness description information of the patient to be diagnosed and the symptoms obtained through the multi-round inquiry interaction process, so that the inquiry result of the patient to be diagnosed can more accurately represent the physical condition of the patient to be diagnosed, thereby the doctor can refer to the inquiry result of the patient to be diagnosed to perform more accurate disease diagnosis on the patient to be diagnosed. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiment or the prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A flowchart of an inquiry method provided by the embodiment of the present application;

[0019] Figure 2 An example diagram of an automatic inquiry process provided by the embodiment of the present application;

[0020] Figure 3 A structural schematic diagram of a disease pre-diagnosis model provided by the embodiment of the present application;

[0021] Figure 4 A schematic diagram of the working principle of a node in a GRU provided by the embodiment of the present application;

[0022] Figure 5 A schematic diagram of an automatic inquiry application instance provided by the embodiment of the present application;

[0023] Figure 6 A schematic diagram of an automatic inquiry process provided by the embodiment of the present application;

[0024] Figure 7 A schematic diagram of a discriminative representation data provided by the embodiment of the present application;

[0025] Figure 8 A structural schematic diagram of an inquiry device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0026] The inventor found in the research on the interrogation process that the artificial interrogation process has the following defects: ① Since the interrogation process for each patient needs to be participated by at least one medical staff, it is easy to cause a hospital to consume a large amount of medical resources to participate in the interrogation process when the patient flow is large, which easily leads to the shortage of medical resources of the hospital. ② Since the interrogation process for a patient is usually dominated by medical staff, in order to ensure the effectiveness of the interrogation process, it is usually necessary to ensure that the medical staff participating in the interrogation process has high interrogation experience, which leads to that those medical staffs with insufficient experience cannot participate in the interrogation process, thereby leading to the more shortage of medical resources of the hospital. ③ Since some diseases (for example, XXX pneumonia) of some patients have strong infectivity, the medical staff who face-to-face interrogate these patients are easy to be infected, in order to avoid the expansion of the infection range, it is necessary to isolate these medical staff, which easily leads to the more shortage of medical resources of the hospital.

[0027] Based on the above findings, in order to solve the technical problems shown in the background art, the embodiment of the present application provides an interrogation method applied to an electronic device, which comprises the following steps: after obtaining the illness self-description information of a patient to be diagnosed, first determining a symptom set to be used (for example, cough, sneezing, etc.) according to the illness self-description information; then determining a symptom to be asked (for example, headache) according to the symptom set to be used, so that the symptom to be asked is different from any symptom in the symptom set to be used, so that the patient to be diagnosed can confirm whether he / she has the symptom to be asked, so that after obtaining the inquiry feedback information (for example, information such as having a headache and not having a headache) of the patient to be diagnosed on the symptom to be asked, updating the symptom set to be used according to the inquiry feedback information, and continuing to execute the step of "determining the symptom to be asked according to the symptom set to be used" until the first stop condition is reached, determining the interrogation result of the patient to be diagnosed according to the symptom set to be used, so that the interrogation result can represent the physical condition of the patient to be diagnosed (for example, the disease that is likely to have, various symptoms of the body, etc.), so that the doctor can refer to the interrogation result and other examination results (for example, urine test results, blood test results, computer tomography results, etc.) to make more accurate disease diagnosis on the patient to be diagnosed.

[0028] It can be seen that the technical solution provided in the embodiments of the present application is implemented by an electronic device (for example, a terminal device or a server, etc.), so that the medical personnel do not need to participate in the inquiry process in the inquiry process provided in the embodiments of the present application, so that automatic inquiry can be realized, thereby the technical problem that the inquiry effect is poor due to defects in the artificial inquiry process can be overcome as much as possible, and the inquiry effect can be effectively improved. In addition, the inquiry result of the patient to be diagnosed is determined according to the symptoms carried by the illness self-report information of the patient to be diagnosed and the symptoms obtained through the multi-round inquiry interaction process, so that the inquiry result of the patient to be diagnosed can more accurately represent the physical condition of the patient to be diagnosed, so that the doctor can refer to the inquiry result of the patient to be diagnosed to perform more accurate disease diagnosis on the patient to be diagnosed.

[0029] In addition, the embodiments of the present application do not limit the implementation of the above-mentioned "electronic device", for example, it can be a terminal device or a server and other data processing devices. Among them, the terminal device can be a smart phone, a computer, a personal digital assistant (PDA) or a tablet computer, etc. The server can be a stand-alone server, a cluster server or a cloud server.

[0030] In order 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 described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0031] Method embodiment one

[0032] Referring to Figure 1 , the figure is a flow chart of an inquiry method provided in the embodiments of the present application.

[0033] The inquiry method applied to the electronic device provided in the embodiments of the present application comprises S1-S5:

[0034] S1: After obtaining the illness self-report information of the patient to be diagnosed, the symptom set to be used is determined according to the illness self-report information.

[0035] The "patient to be diagnosed" refers to a person who needs to be diagnosed; and the "patient to be diagnosed" is not limited in the embodiments of the present application, for example, when the diagnosis method provided by the embodiments of the present application is applied to a hospital (for example, a hospital diagnosis system, a hospital diagnosis robot, etc.), the "patient to be diagnosed" can refer to a patient who seeks diagnosis in the hospital. For another example, when the diagnosis method provided by the embodiments of the present application is applied to online diagnosis (for example, an intelligent doctor, etc.), the "patient to be diagnosed" can refer to a user who triggers a diagnosis request on a webpage of the online diagnosis.

[0036] The "disease self-description information" refers to the relevant content described by the patient to be diagnosed or a disease informant (for example, a relative of the patient to be diagnosed, etc.) of the patient to be diagnosed, which is actively described and introduced for the physical condition of the patient to be diagnosed. For example, the "disease self-description information" can include Figure 2 "Doctor, I have a runny nose and cough".

[0037] In addition, the embodiments of the present application do not limit the acquisition method of the "disease self-description information", for example, when the execution subject of the diagnosis method provided by the embodiments of the present application is a terminal device, the "disease self-description information" can be input into the terminal device by the patient to be diagnosed or the disease informant of the patient to be diagnosed through any input device (for example, a keyboard, a pickup, a touchable display screen, a camera, etc.) of the terminal device. For another example, when the execution subject of the diagnosis method provided by the embodiments of the present application is a server, and the server can communicate data with a terminal device, the "disease self-description information" can be first input into the terminal device by the patient to be diagnosed or the disease informant of the patient to be diagnosed; and then the "disease self-description information" is sent to the server by the terminal device, so that the server can acquire the "disease self-description information".

[0038] In addition, the embodiments of the present application do not limit the data type of the "disease self-description information", for example, it can be voice data, text data, image data, etc.

[0039] The "symptom set to be used" is used to record at least one symptom appearing on the body of the patient to be diagnosed; and the "symptom set to be used" can be initialized by using the "disease self-description information". For example, when the "disease self-description information" is Figure 2 "Doctor, I have a runny nose and cough" as shown, the "symptom set to be used" can be initialized as {runny nose, cough}.

[0040] It should be noted that the embodiments of the present application do not limit the implementation manner of "determining the symptom set to be used according to the disease self-description information" in S1, and any method that can extract symptoms from the "disease self-description information" can be used.

[0041] Based on the related content of S1, for the interrogation process of the patient to be diagnosed, the patient to be diagnosed or the patient's condition informant can provide the patient's condition self-reporting information, so that the condition self-reporting information can represent some symptoms appearing on the patient's body; then, the condition self-reporting information is used to initialize the symptom set to be used, so that the symptom set to be used is used to record the symptoms appearing on the patient's body, so that subsequent at least one round of interrogation interaction process can be performed on the patient to be diagnosed or the patient's condition informant based on the symptoms recorded in the symptom set to be used.

[0042] S2: determining a symptom to be inquired according to the symptom set to be used.

[0043] The "symptom to be inquired" refers to a symptom that needs to be confirmed by the patient to be diagnosed or the patient's condition informant through an inquiry method whether it appears on the patient's body. For example, when the "symptom set to be used" is {runny nose, cough}, the "symptom to be inquired" can include fever, cough, and / or sneezing, etc.

[0044] In addition, the embodiments of the present application do not limit the implementation of S2. For example, in one possible implementation, S2 can specifically include S21-S22:

[0045] S21: determining a disease diagnosis result to be used according to the symptom set to be used.

[0046] The "disease diagnosis result to be used" is used to represent a disease that the patient to be diagnosed may have; and the "disease diagnosis result to be used" is determined according to the symptom set to be used. For example, when the "symptom set to be used" is {runny nose, cough}, the disease diagnosis result to be used can include rhinitis and pneumonia.

[0047] The embodiments of the present application do not limit the implementation of S21. For example, in one possible implementation, it can specifically include S211-S212:

[0048] S211: inputting the symptom set to be used into a pre-constructed disease pre-diagnosis model to obtain a first disease pre-diagnosis result output by the disease pre-diagnosis model.

[0049] The "first disease pre-diagnosis result" is used to represent the probability of the patient to be diagnosed having at least one candidate disease (for example, the probability of rhinitis is 0.4; the probability of pneumonia is 0.39; the probability of dermatitis is 0.008; the probability of mastitis is 0.0001; …).

[0050] It should be noted that the above "candidate disease" is used to represent the disease that the doctor needs to consider when diagnosing a patient; and the embodiments of the present application do not limit the acquisition method of "at least one candidate disease", for example, it can be obtained by disease mining from a large amount of medical data (such as medical record data, medical papers, medical book data, medical journals, medical clinical data, etc.).

[0051] The above "disease pre-diagnosis model" is used for disease prediction processing for the input data of the disease pre-diagnosis model; and the embodiments of the present application do not limit the "disease pre-diagnosis model", for example, it can be implemented by using any machine learning model existing or appearing in the future.

[0052] In addition, in order to further improve the effect of interrogation, the embodiments of the present application also provide another possible implementation of the above "disease pre-diagnosis model", in which, as shown in Figure 3 The disease pre-diagnosis model 300 can include a feature vector determination module 301, a pooling layer 302, a splicing module 303, a full connection module 304, and a decision module 305. The pooling layer 302 includes N pooling modules, and the input data of each pooling module includes the output data of the feature vector determination module 301; the input data of the splicing module 303 includes the output data of the N pooling modules; the input data of the full connection module 304 includes the output data of the splicing module 303; and the input data of the decision module 305 includes the output data of the full connection module 304.

[0053] In order to facilitate understanding of the working principle of the disease pre-diagnosis model 300, the determination process of the above "first disease pre-diagnosis result" is taken as an example for description.

[0054] As an example, the process of determining the "first disease pre-diagnosis result" by using the disease pre-diagnosis model 300 can specifically include steps 11-15:

[0055] Step 11: input the set of symptoms to be used into the feature vector determination module 301 to obtain the vector representation result output by the feature vector determination module 301.

[0056] The above "feature vector determination module 301" is used for feature vector extraction processing for the input data of the feature vector determination module 301.

[0057] In addition, the embodiment of the present application is not limited to the implementation of the above-mentioned "characteristic vector determination module 301", for example, it can be implemented by using any existing or future method capable of performing characteristic vector extraction processing on text data. For example, in order to improve the extraction efficiency of the characteristic vector, the "characteristic vector determination module 301" can be implemented by means of a Gate Recurrent Unit (GRU) model; and the implementation process can specifically include: first performing character vectorization processing (for example, word2vec, etc.) on the set of symptoms to be used, to obtain the character vector of the set of symptoms to be used; and then inputting the character vector of the set of symptoms to be used into the GRU model to obtain the vector representation result output by the GRU model.

[0058] It should be noted that the characteristic of the above-mentioned GRU model is that the network structure is simple and the calculation speed is fast. In addition, as shown in Figure 4 , for the GRU model at the current time, it can refer to the output data h (t-1) of the GRU model at the previous time and the character vector x (t) input into the GRU model at the current time to determine the output data of the GRU model at the current time; and the determination process can specifically include: first using two gating mechanisms to mine the features we want, i.e. the update gate (as shown in formula (1)) and the reset gate (as shown in formula (2)), so that the update gate is used to control the degree of state information at the previous time being brought into the current state, and the greater the value of the update gate, the more state information at the previous time is brought in, and the reset gate controls how much information of the previous state is written into the current candidate set h ～t , the smaller the reset gate, the less information of the previous state is written; then using the gating signal r t and formula (3) to calculate the similarity between the features after the reset gate and the features to be updated this time; finally, according to formula (4), the degree of updating the features this time and fusing the features last time is controlled to ensure that the model training is stable and does not produce large fluctuations, and is easy to converge.

[0059] z t =σ(W z ·x (t) +U z ·h (t-1) +b z ) (1)

[0060] r t =σ(W r ·x (t) +U r ·h (t-1) +b r ) (2)

[0061] h～t = tanh(Wh x + U h (t) + U h (r t · h (t-1) ) + b h (3)

[0062] h (t) = h ～t · z t + h (t-1) · (1 - z t ) (4)

[0063] wherein h (t) represents output data of the GRU model at a current time; h (t-1) represents output data of the GRU model at a previous time; x (t) represents an input character vector of the GRU model at the current time; W z , U z represent trainable weights in the update gate, and b z represents a trainable inductive bias in the update gate; W r , U r represent trainable weights in the reset gate, and b r represents a trainable inductive bias in the reset gate; W h , U h represent trainable weights, and b h represents a trainable inductive bias; σ() represents a nonlinear activation function; “·” represents the product of two matrices; tanh() represents a hyperbolic tangent function.

[0064] The vector representation result is used to represent symptom description information carried by the symptom set to be used.

[0065] Step 12: input the vector representation result into the nth pooling module to obtain an nth pooling result output by the nth pooling module. Wherein n is a positive integer, and n≤N.

[0066] The nth pooling module is used to perform pooling processing on input data of the nth pooling module.

[0067] In addition, the working principle of the nth pooling module is different from the working principle of any other pooling module in the N pooling modules except the nth pooling module. For example, if N is 2, the first pooling module can be implemented in a maximum pooling processing mode, and the second pooling module can be implemented in an average pooling processing mode.

[0068] Step 13: input the 1st pooling result to the Nth pooling result into the splicing module 303 to obtain a splicing result output by the splicing module 303.

[0069] The splicing module 303 is configured to perform splicing processing (for example, horizontal splicing processing, etc.) on input data of the splicing module 303.

[0070] Step 14: input the splicing result into the full connection module 304 to obtain a full connection result output by the full connection module 304.

[0071] The full connection module 304 is configured to perform full connection processing on input data of the full connection module 304. The embodiments of the full connection module 304 are not limited, and any existing or future full connection network (for example, a full connection neural network, etc.) can be used for implementation.

[0072] Step 15: input the full connection result into the decision module 305 to obtain a first disease pre-diagnosis result output by the decision module 305.

[0073] The decision module 305 is configured to perform decision processing on input data of the decision module 305. The embodiments of the decision module 305 are not limited, and any existing or future decision network (for example, a softmax function, etc.) can be used for implementation.

[0074] Based on the related content of steps 11 to 15, for the disease pre-diagnosis model 300, the character feature extraction of the to-be-used symptom set can be performed by the feature vector determination module 301 to obtain a vector representation result. The vector representation result is classified by the pooling layer 302, the splicing module 303, the full connection module 304, and the decision module 305 to obtain a first disease pre-diagnosis result, so that the first disease pre-diagnosis result can represent the occurrence probability of various candidate diseases predicted based on the to-be-used symptom set.

[0075] The embodiments of the disease pre-diagnosis model are not limited, and any existing or future model construction method can be used for implementation. For example, in order to improve the prediction effect of the model, the model construction method shown in the following S211 can be used for implementation. Method embodiment two

[0076] Based on the related content of S211, after obtaining the to-be-used symptom set, the disease pre-diagnosis model constructed in advance can be used to perform disease prediction processing on the to-be-used symptom set to obtain and output a first disease pre-diagnosis result, so that the first disease pre-diagnosis result can represent the occurrence probability of various candidate diseases of the patient to be diagnosed.​

[0077] S212: determining the disease diagnosis result to be used according to the first disease prognosis result.

[0078] The embodiments of the present application do not limit the implementation of S212, for example, it can include: determining at least one candidate disease in the first disease prognosis result whose occurrence probability reaches the first probability condition as the disease diagnosis result to be used.

[0079] The "first probability condition" can be pre-set; moreover, the embodiments of the present application do not limit the "first probability condition", for example, the "first probability condition" can be: reaching a pre-set probability threshold (for example, 0.3). For another example, when the first disease prognosis result is used to record at least one candidate disease arranged in descending order of occurrence probability, the "first probability condition" can also be: the arrangement serial number is lower than Q (that is, the top Q candidate diseases are taken). Wherein, Q is a positive integer.

[0080] Based on the related content of S21, after obtaining the symptom set to be used, the disease prediction processing can be performed on the patient to be diagnosed by referring to all symptoms in the symptom set to be used, and the disease diagnosis result to be used is obtained, so that the disease diagnosis result to be used can represent the disease that the patient to be diagnosed may have.

[0081] S22: determining the symptom to be asked according to the symptom set to be used and the disease diagnosis result to be used.

[0082] The embodiments of the present application do not limit the implementation of S22, for example, it can specifically include: randomly selecting at least one symptom from the symptom set corresponding to the disease diagnosis result to be used, and determining the symptom to be asked. Wherein, the "symptom set corresponding to the disease diagnosis result to be used" is used to record the performance symptoms corresponding to all diseases in the disease diagnosis result to be used.

[0083] It should be noted that the performance symptoms corresponding to a disease refer to at least one symptom that a patient with the disease may have. For example, the performance symptoms corresponding to rhinitis can include: no nasal congestion, sneezing, nosebleed, …; the performance symptoms corresponding to pneumonia can include: shortness of breath, no cyanosis, no chills, …; the performance symptoms corresponding to dermatitis can include: itching, rash, skin and mucous membrane congestion, …; the performance symptoms corresponding to mastitis can include: breast pain, chills, tongue with thin white fur, …; in addition, the performance symptoms corresponding to various diseases can be obtained by data analysis (for example, data mining) from a large amount of medical data (for example, medical record data, medical papers, medical book data, medical journals, etc.).

[0084] In addition, in order to further improve the accuracy of the to-be-inquired symptoms, the embodiment of the present application further provides another possible implementation of S22, which can specifically include S221-S222:

[0085] S221: input the to-be-used symptom set and the to-be-used disease diagnosis result into the pre-constructed symptom prediction model to obtain a first symptom prediction result output by the symptom prediction model.

[0086] The above-mentioned "symptom prediction model" is used for symptom prediction processing on the input data of the symptom prediction model; and the embodiment of the present application does not limit the "symptom prediction model", which can be any machine learning model existing or to be developed in the future.

[0087] In addition, in order to further improve the diagnosis effect, the embodiment of the present application further provides another possible implementation of the above-mentioned "symptom prediction model", in which the model architecture of the "symptom prediction model" is the same as that of the above-mentioned "disease prediction model". As can be seen, the above-mentioned "symptom prediction model" can also be implemented by using the model structure shown in the above-mentioned "disease prediction model". Figure 3

[0088] In fact, because the number of symptoms (for example, 119) involved in the above-mentioned "symptom prediction model" is different from the number of diseases (for example, 12) involved in the above-mentioned "disease prediction model", in order to improve the prediction effect of the model, the number of output nodes (for example, 119) of the above-mentioned "symptom prediction model" is different from the number of output nodes (for example, 12) of the above-mentioned "disease prediction model", so that the data dimension (for example, 119) of the output data of the above-mentioned "symptom prediction model" is different from the dimension (for example, 12) of the output data of the above-mentioned "disease prediction model".

[0089] It should be noted that when the data dimension of the output data of the above-mentioned "symptom prediction model" is 119, the output data of the above-mentioned "symptom prediction model" includes the occurrence probability of 118 symptoms and the occurrence probability of stopping inquiring (that is, the occurrence probability of no symptoms to be inquired). Among them, the "occurrence probability of stopping inquiring" is used to indicate whether the to-be-used symptom set of the current round has included enough symptom information.

[0090] The embodiment of the present application does not limit the construction process of the above-mentioned "symptom prediction model", which can be implemented by using any model construction method existing or to be developed in the future. For example, in order to improve the prediction effect of the model, the model construction method shown in the following can be used for implementation. Method embodiment two

[0091] ​​The first symptom prediction result is used to represent the occurrence probability of various symptoms of the patient to be diagnosed (for example, the occurrence probability of fever is 0.3; the occurrence probability of cough is 0.29; the occurrence probability of sneezing is 0.28; the occurrence probability of shortness of breath is 0.02; and so on).

[0092] S222: determining the symptom to be inquired according to the first symptom prediction result.

[0093] The embodiments of the present application do not limit the implementation of S222, for example, it can include: determining the symptom to be inquired according to at least one symptom (for example, fever, etc.) in the first symptom prediction result that meets a second probability condition.

[0094] The second probability condition can be preset, and the embodiments of the present application do not limit the second probability condition, for example, the second probability condition can be that it reaches a preset probability threshold (for example, 0.2). For another example, when the first symptom prediction result is used to record at least one symptom arranged in descending order of occurrence probability, the second probability condition can also be that the arrangement serial number is lower than P (that is, the top P symptoms are taken). Wherein, P is a positive integer. For another example, in order to improve the effectiveness of the symptom to be inquired, the determination process of the symptom to be inquired shown in the following Method embodiment three

[0095] Based on the related content of S22, after obtaining the disease diagnosis result to be used, the disease diagnosis result to be used and the symptom set to be used can be referred to to determine the symptom to be inquired, so that the symptom to be inquired can better represent the symptoms that are more likely to appear on the body of the patient to be diagnosed.

[0096] Based on the related content of S2, after obtaining the symptom set to be used, all symptoms in the symptom set to be used can be referred to to predict the symptoms that are likely to appear on the body of the patient to be diagnosed, and the symptom to be inquired is obtained, so that the subsequent inquiry method can be used to confirm whether the symptom to be inquired appears on the body of the patient to be diagnosed.

[0097] S3: determining whether the first stop condition is met, if yes, executing S5; if no, executing S4.

[0098] ​The first stop condition can be preset, and the embodiment of the present application does not limit the first stop condition. For example, in order to improve the effect of inquiry, when the to-be-inquired symptom is determined according to the first symptom prediction result, and the first symptom prediction result includes the probability of stopping inquiry (that is, the probability of no symptom needing inquiry), the first stop condition can be specifically: the probability of stopping inquiry is higher than the probability of any symptom in the first symptom prediction result (that is, the to-be-inquired symptom determined based on the first symptom prediction result includes the pseudo-symptom of stopping inquiry).

[0099] S4: After obtaining the inquiry feedback information of the to-be-diagnosed patient for the to-be-inquired symptom, updating the to-be-used symptom set according to the inquiry feedback information, and returning to execute S2.

[0100] The inquiry feedback information is used to indicate whether the to-be-inquired symptom appears on the body of the to-be-diagnosed patient, and the embodiment of the present application does not limit the acquisition method of the inquiry feedback information. For example, it can be achieved by means of dialogue interaction process between the to-be-diagnosed patient (or the person who knows the condition of the to-be-diagnosed patient). It can be seen that the acquisition process of the inquiry feedback information can include steps 21-23:

[0101] Step 21: generating symptom inquiry information according to the to-be-inquired symptom.

[0102] The symptom inquiry information is used to inquire from the to-be-diagnosed patient (or the person who knows the condition of the to-be-diagnosed patient) whether the to-be-inquired symptom appears on the body of the to-be-diagnosed patient. For example, when the to-be-inquired symptom is fever, the symptom inquiry information can be "whether fever".

[0103] In addition, the embodiment of the present application does not limit the generation process of the symptom inquiry information. For example, it can specifically include: filling the to-be-inquired symptom into the pre-set symptom inquiry template (for example, "whether (Symptoms to be inserted here) ") to obtain the symptom inquiry information.

[0104] In fact, the symptom inquiry templates corresponding to different symptoms can be different. Based on this, the embodiment of the present application further provides another possible implementation manner of generating the symptom inquiry information, which can specifically include: first searching for a candidate template corresponding to the to-be-inquired symptom from at least one pre-set candidate template, and determining the to-be-used template; and then filling the to-be-inquired symptom into the to-be-used template to obtain the symptom inquiry information.

[0105] Step 22: sending the symptom inquiry information to the to-be-diagnosed patient (or the person who knows the condition of the to-be-diagnosed patient), so that the to-be-diagnosed patient (or the person who knows the condition of the to-be-diagnosed patient) replies to the symptom inquiry information.

[0106] The application embodiments are not limited to the sending manner of the above-mentioned "symptom inquiry information". For example, when the execution subject of the inquiry method provided by the application embodiments is a terminal device, the terminal device can directly display the "symptom inquiry information" according to a preset display manner (for example, a voice playing manner, a text display manner, an image display manner, etc.), so that the patient to be diagnosed (or a person who knows the illness of the patient to be diagnosed) can learn the "symptom inquiry information" from the terminal device. For another example, when the execution subject of the inquiry method provided by the application embodiments is a server, and the server can communicate data with a terminal device, the server can first send the "symptom inquiry information" to the terminal device, so that the terminal device can display the "symptom inquiry information" according to a preset display manner, so that the patient to be diagnosed (or a person who knows the illness of the patient to be diagnosed) can learn the "symptom inquiry information" from the terminal device.

[0107] Step 23: After obtaining the reply content of the patient to be diagnosed (or a person who knows the illness of the patient to be diagnosed) to the symptom inquiry information, the inquiry feedback information of the patient to be diagnosed to the inquiry symptom is determined according to the reply content.

[0108] The above-mentioned "reply content" is obtained by the patient to be diagnosed (or a person who knows the illness of the patient to be diagnosed) replying to the symptom inquiry information, so that the "reply content" can indicate whether the inquiry symptom appears on the patient to be diagnosed, so that the reply content carries the inquiry feedback information of the patient to be diagnosed to the inquiry symptom.

[0109] Actually, for the above-mentioned "reply content", the reply content can completely indicate the symptom (for example, "fever, cough and sneezing") of the patient to be diagnosed, so that the inquiry feedback information can be directly determined according to the reply content; but the reply content can not completely indicate the symptom (for example, "yes") of the patient to be diagnosed, so that the inquiry feedback information can be determined by combining the reply content with the corresponding symptom inquiry information (for example, "is there any discomfort in the throat?"). Figure 2 Figure 2 Figure 2

[0110] ​​​Based on this, the embodiment of the present application further provides another possible implementation of step 23, which can specifically include: after obtaining the reply content of the patient to be diagnosed (or the illness informant of the patient to be diagnosed) to the symptom inquiry information, if it is determined that the reply content meets the preset complete condition, extracting the inquiry feedback information of the patient to be diagnosed to the to-be-inquired symptom from the reply content; if it is determined that the reply content does not meet the preset complete condition, determining the inquiry feedback information of the patient to be diagnosed to the to-be-inquired symptom according to the reply content and the symptom inquiry information.

[0111] It should be noted that the "preset complete condition" can be preset, and the embodiment of the present application does not limit the "preset complete condition", for example, which can specifically include: the reply content includes the core character information (for example, "pharynx", "part", "not", "suitable", "feeling") of the to-be-inquired symptom.

[0112] Based on the above-mentioned related content of steps 21 to 23, after obtaining the to-be-inquired symptom, the symptom inquiry information can be generated according to the to-be-inquired symptom, so that the symptom inquiry information is used to confirm whether the patient to be diagnosed has the to-be-inquired symptom; then the to-be-inquired symptom is sent to the patient to be diagnosed (or the illness informant of the patient to be diagnosed), so that the patient to be diagnosed (or the illness informant of the patient to be diagnosed) can reply to the symptom inquiry information to obtain the reply content, so that the reply content can indicate whether the patient to be diagnosed has the to-be-inquired symptom; finally, the inquiry feedback information of the patient to be diagnosed to the to-be-inquired symptom is determined according to the reply content, so that the inquiry feedback information can indicate whether the patient to be diagnosed has the to-be-inquired symptom.

[0113] The embodiment of the present application does not limit the implementation of "updating the to-be-used symptom set according to the inquiry feedback information" in S4, for example, which can specifically include: first determining a supplementary symptom (for example, no fever, cough, sneezing, etc.) according to the inquiry feedback information, so that the supplementary symptom is used to supplement the description of the physical condition of the patient to be diagnosed; then adding the supplementary symptom to the to-be-used symptom set, so that the to-be-used symptom set can add one symptom, so that the to-be-used symptom set can better represent the physical condition of the patient to be diagnosed.

[0114] Based on the related content of S4, when it is determined that the first stop condition is not reached, it can be determined that the interrogation process for the patient to be diagnosed has not ended, so it can be determined that the patient to be diagnosed (or the patient to be diagnosed The patient's condition informant) still needs to provide other symptom description information (such as no fever, cough, sneezing, etc.) in addition to all symptoms in the symptom set to be used. Therefore, after obtaining the symptoms to be inquired, the inquiry feedback information of the patient to be diagnosed for the symptoms to be inquired can be obtained first, and then the inquiry feedback information is used to update the symptom set to be used, so that the number of symptoms in the updated symptom set to be used is higher than the number of symptoms in the symptom set to be used before updating, so that the updated symptom set to be used can better represent the physical condition of the patient to be diagnosed.

[0115] It should be noted that when S2 includes S21-S22, S4 can specifically include: after obtaining the inquiry feedback information of the patient to be diagnosed for the symptoms to be inquired, updating the symptom set to be used according to the inquiry feedback information, and returning to execute S21.

[0116] S5: Determine the interrogation result of the patient to be diagnosed according to the symptom set to be used.

[0117] The above-mentioned "interrogation result of the patient to be diagnosed" refers to the automatic interrogation result of the patient to be diagnosed, so that the "interrogation result of the patient to be diagnosed" is used to represent the physical condition of the patient to be diagnosed; and the application embodiment does not limit the "interrogation result of the patient to be diagnosed", for example, it can carry the disease that the patient to be diagnosed may have and / or the symptoms that appear on the patient to be diagnosed.

[0118] In addition, the application embodiment does not limit the determination process of the "interrogation result of the patient to be diagnosed", for example, it can specifically include: directly determining the symptom set to be used as the interrogation result of the patient to be diagnosed. For another example, it can specifically include step 31:

[0119] Step 31: first determine the disease information to be used according to the symptom set to be used.

[0120] The above-mentioned "disease information to be used" is used to represent the disease that the patient to be diagnosed may have.

[0121] In addition, the application embodiment does not limit the implementation of step 31, for example, step 31 can be: first matching all symptoms in the symptom set to be used with the characteristic symptoms corresponding to various candidate diseases to obtain the matching degree between the symptom set to be used and various candidate diseases; then collect the U candidate diseases with the highest matching degree to obtain the disease information to be used. For another example, step 31 can be implemented by means of the disease prediction model shown in the following Method embodiment four

[0122] ​Step 32: Determine the consultation results of the patient to be diagnosed based on the disease information to be used.

[0123] The embodiments of this application do not limit the implementation of step 32. For example, step 32 may specifically be: directly determining the disease information to be used as the consultation result of the patient to be diagnosed. Alternatively, step 32 may specifically be: performing aggregate processing on the disease information to be used and all symptoms in the symptom set to be used to obtain the consultation result of the patient to be diagnosed.

[0124] Based on the relevant content of S5 above, it can be seen that when the first stopping condition is met, the consultation process for the patient to be diagnosed has ended. Therefore, the consultation results of the patient to be diagnosed can be determined by referring to the symptom set to be used, so that the consultation results can indicate the physical condition of the patient to be diagnosed (e.g., what symptoms the body has, what diseases it is likely to have, etc.).

[0125] Based on the above-mentioned content of S1 to S5, it can be seen that, for the consultation method provided in this application embodiment, after the electronic device obtains the patient's self-reported medical condition information, the electronic device first determines the set of symptoms to be used (e.g., cough, sneezing, etc.) based on the self-reported medical condition information; then, based on the set of symptoms to be used, it determines the symptom to be inquired about (e.g., headache), so that the symptom to be inquired about is different from any symptom in the set of symptoms to be used, thereby enabling the patient to confirm whether they have the symptom to be inquired about, so as to obtain the patient's feedback information on the symptom to be inquired about (e.g., a slight headache, no headache, etc.). After receiving the information, the symptom set to be used is updated based on the feedback information, and the above-mentioned step of "determining the symptoms to be inquired about based on the symptom set to be used" continues until the first stopping condition is met. At this point, the consultation results of the patient to be diagnosed are determined based on the symptom set to be used, so that the consultation results can indicate the physical condition of the patient to be diagnosed (e.g., the diseases that are likely to be present, and / or the various symptoms that are present), so that the doctor can refer to the consultation results and other examination results (e.g., urine test results, blood test results, computed tomography scan results, etc.) to make a more accurate diagnosis of the disease in the patient to be diagnosed.

[0126] As can be seen, since the technical solution provided in this application embodiment is implemented by electronic devices (e.g., terminal devices or servers), the consultation process provided in this application embodiment (such as...) Figure 5 As shown, no medical staff are required to participate in the consultation, thus enabling automated consultation. This can overcome the technical problem of poor consultation results caused by the defects of the manual consultation process, and effectively improve the consultation results.

[0127] In addition, since the interrogation result of the patient to be diagnosed is determined according to the symptoms carried by the illness self-reporting information of the patient to be diagnosed and the symptoms obtained through the multi-round interrogation interaction process, the interrogation result of the patient to be diagnosed can more accurately represent the physical condition of the patient to be diagnosed, so that the doctor can make more accurate disease diagnosis for the patient to be diagnosed according to the interrogation result of the patient to be diagnosed.

[0128] It should be noted that, Figure 5 In the present application, the "displayed symptoms" refer to the symptoms carried by the illness self-reporting information, and the "implicit symptoms" refer to the symptoms inquired through at least one round of interrogation process.

[0129] Method embodiment two

[0130] In fact, when the above S2 is implemented by means of the disease pre-diagnosis model and the symptom prediction model, the prediction accuracy of the to-be-inquired symptoms can be affected not only by the prediction performance of the symptom prediction model, but also by the prediction performance of the disease pre-diagnosis model. Specifically, the higher the prediction performance of the disease pre-diagnosis model is, the more accurate the to-be-used disease diagnosis result output by the disease pre-diagnosis model is for the to-be-used symptom set, so that the input data of the symptom prediction model is more accurate, and thus the to-be-inquired symptoms output by the symptom prediction model are more accurate. In this way, the next round of input data of the disease pre-diagnosis model determined based on the to-be-inquired symptoms is more accurate, which is conducive to better determining the interrogation result of the patient to be diagnosed.

[0131] It can be seen that the prediction performance of the disease pre-diagnosis model and the prediction performance of the symptom prediction model can affect each other, so in order to further improve the model prediction accuracy, the disease pre-diagnosis model and the symptom prediction model can be jointly constructed. Based on this, the present application embodiment provides a model construction method, which can specifically include steps 41-49:

[0132] Step 41: training the to-be-used model by using the first symptom sample and the actual disease diagnosis result of the first symptom sample, to obtain a to-be-optimized pre-diagnosis model.

[0133] The "first symptom sample" is used to represent the sample data required for training the to-be-used model, and the present application embodiment does not limit the "first symptom sample". For example, the "first symptom sample" can include at least one symptom. In addition, the present application embodiment does not limit the acquisition method of the "first symptom sample", for example, it can be extracted from a medical record sample.

[0134] The "actual disease diagnosis result of the first symptom sample" is used to represent a disease actually diagnosed by a doctor referring to the symptom information carried by the first symptom sample; and the embodiment of the present application does not limit the acquisition method of the "actual disease diagnosis result of the first symptom sample", for example, it can be extracted from a medical record sample. For another example, it can be manually annotated by a doctor.

[0135] The "to-be-used model" is used for disease prediction processing on input data of the to-be-used model; and the model structure of the "to-be-used model" can be consistent with the model structure of the "disease pre-diagnosis model" above, so that the "disease pre-diagnosis model" can be constructed based on the trained to-be-used model subsequently.

[0136] The embodiment of the present application does not limit the training process of the "to-be-used model", for example, any existing or future model training method can be used for implementation.

[0137] The "to-be-optimized pre-diagnosis model" is used to represent a trained to-be-used model.

[0138] Step 42: determining a sample symptom set according to the second symptom sample.

[0139] The "second symptom sample" is used to represent sample data required when training the to-be-optimized pre-diagnosis model and the to-be-optimized symptom model; and the embodiment of the present application does not limit the "second symptom sample", for example, the "second symptom sample" can include at least one symptom. In addition, the embodiment of the present application does not limit the acquisition method of the "second symptom sample", for example, it can be extracted from a medical record sample. Furthermore, the embodiment of the present application does not limit the relationship between the "second symptom sample" and the "first symptom sample" above, which can be the same or different.

[0140] The "sample symptom set" is used to record the symptom information carried by the second symptom sample; and the "sample symptom set" is similar to the "to-be-used symptom set" above.

[0141] Step 43: inputting the sample symptom set into the to-be-optimized pre-diagnosis model to obtain a second disease pre-diagnosis result output by the to-be-optimized pre-diagnosis model.

[0142] The "second disease pre-diagnosis result" is used to represent the occurrence probability of at least one candidate disease determined by referring to the sample symptom set; and the "second disease pre-diagnosis result" is similar to the "first disease pre-diagnosis result" above.

[0143] Step 44: obtaining a second symptom prediction result according to the sample symptom set, the second disease pre-diagnosis result and the to-be-optimized symptom model.

[0144] The "to-be-optimized symptom model" is used for symptom prediction processing on input data of the to-be-optimized symptom model, and the model structure of the "to-be-optimized symptom model" can be consistent with the model structure of the "symptom prediction model" to enable subsequent determination of the "symptom prediction model" based on the trained to-be-optimized symptom model.

[0145] The "second symptom prediction result" is used to represent the occurrence probability of various symptoms determined with reference to the sample symptom set, and the "second symptom prediction result" is similar to the "first symptom prediction result".

[0146] In addition, the embodiments of the present application do not limit the implementation of step 44, for example, it can specifically include: first determining the sample disease diagnosis result according to the second disease pre-diagnosis result; and then inputting the sample disease diagnosis result and the sample symptom set into the to-be-optimized symptom model to obtain the second symptom prediction result output by the to-be-optimized symptom model.

[0147] The "sample disease diagnosis result" is used to represent the possible disease determined with reference to the sample symptom set, and the "sample disease diagnosis result" is similar to the "to-be-used disease diagnosis result".

[0148] Step 45: determining whether a second stop condition is reached, if yes, executing step 49; if no, executing steps 46-48.

[0149] The "second stop condition" can be pre-set, for example, any existing or future model stop condition can be used for implementation. For another example, the second stop condition shown below can be used for implementation.

[0150] Step 46: updating the to-be-optimized pre-diagnosis model according to the second disease pre-diagnosis result and the actual disease diagnosis result of the second symptom sample.

[0151] The "actual disease diagnosis result of the second symptom sample" is used to represent the disease actually diagnosed by a doctor with reference to the symptom information carried by the second symptom sample, and the embodiments of the present application do not limit the acquisition method of the "actual disease diagnosis result of the second symptom sample", for example, it can be extracted from a medical record sample. For another example, it can be manually annotated by a doctor.

[0152] It can be seen that when it is determined that the second stop condition is not reached, it can be determined that the prediction performance of the to-be-optimized pre-diagnosis model is still relatively poor, so the to-be-optimized pre-diagnosis model can be updated according to the difference between the second disease pre-diagnosis result and the actual disease diagnosis result of the second symptom sample, so that the updated to-be-optimized pre-diagnosis model has better prediction performance.

[0153] Step 47: updating the symptom model to be optimized according to the second symptom prediction result and the actual associated symptoms corresponding to the sample symptom set.

[0154] The actual associated symptoms corresponding to the sample symptom set refer to the symptoms having an association relationship with at least one symptom in the sample symptom set. It should be noted that for two symptoms, if both of the two symptoms belong to the performance symptoms corresponding to the same disease, the two symptoms have an association relationship (for example, "no nasal congestion" and "sneezing" both belong to the performance symptoms corresponding to rhinitis, and "no nasal congestion" and "sneezing" have an association relationship); if the two symptoms belong to the performance symptoms corresponding to different diseases respectively, the two symptoms do not have an association relationship (for example, "sneezing" belongs to the performance symptoms corresponding to rhinitis, and "itching" belongs to the performance symptoms corresponding to dermatitis, and "sneezing" and "itching" do not have an association relationship).

[0155] The embodiment of the present application does not limit the determination process of the actual associated symptoms corresponding to the sample symptom set, for example, it can specifically include: first, performing set processing on the performance symptoms corresponding to at least one disease in the actual disease diagnosis result of the second symptom sample to obtain the representation symptom set corresponding to the second symptom sample; and then determining the difference set between the representation symptom set and the sample symptom set as the actual associated symptoms corresponding to the sample symptom set.

[0156] Based on the related content of the above step 47, it can be determined that the prediction performance of the symptom model to be optimized is poor when it is determined that the second stop condition is not reached, and therefore the symptom model to be optimized can be updated according to the difference between the second symptom prediction result and the actual associated symptoms corresponding to the sample symptom set, so that the updated symptom model to be optimized has better prediction performance.

[0157] It should be noted that the embodiment of the present application does not limit the execution order of step 47 and step 46, for example, step 47 and step 46 can be executed in sequence, step 46 and step 47 can be executed in sequence, or step 47 and step 46 can be executed simultaneously.

[0158] Step 48: returning to step 43.

[0159] In the embodiment of the present application, after obtaining the updated pre-diagnosis model to be optimized and the updated symptom model to be optimized, step 43 and the subsequent steps can be executed based on the updated pre-diagnosis model to be optimized and the updated symptom model to be optimized to implement the next round of model training process.

[0160] Step 49: determining the disease pre-diagnosis model and the symptom prediction model according to the pre-diagnosis model to be optimized and the symptom model to be optimized.

[0161] In this embodiment of the application, when the second stopping condition is reached, it can be determined that both the pre-diagnosis model to be optimized and the symptom model to be optimized have good predictive performance. Therefore, the trained pre-diagnosis model to be optimized can be determined as the disease pre-diagnosis model so that the disease pre-diagnosis model has good disease prediction performance, and the trained symptom model to be optimized can be determined as the symptom prediction model so that the symptom prediction model has good symptom prediction performance.

[0162] Based on the relevant content of steps 41 to 49 above, it can be seen that in some cases, a two-stage training process can be used to jointly construct the "disease pre-diagnosis model" and the "symptom prediction model" so that both the constructed "disease pre-diagnosis model" and the "symptom prediction model" have good predictive performance.

[0163] In addition, to further improve the model prediction performance, this application provides another possible implementation of the model construction method. In this implementation, in addition to steps 41 to 49 described above, the model construction method may also include steps 50-52:

[0164] Step 50: Based on the second symptom prediction results, determine the symptoms to be added.

[0165] The “symptoms to be added” mentioned above refer to symptoms that may need to be added to the “sample symptom set”; and these “symptoms to be added” are similar to the “symptoms to be inquired about” mentioned above.

[0166] Step 51: Determine whether the symptom to be added belongs to the actual associated symptom corresponding to the sample symptom set. If yes, proceed to step 52; otherwise, proceed to step 48.

[0167] In this embodiment, after obtaining the symptom to be added, it can be determined whether the symptom to be added belongs to the actual associated symptom corresponding to the sample symptom set. If it does, it can be determined that adding the symptom to be added to the sample symptom set can help improve the prediction performance of the pre-diagnosis model and the symptom model to be optimized. Therefore, the symptom to be added can be added to the sample symptom set first, and then step 43 and subsequent steps can be executed. If it does not belong to the sample symptom set, it can be determined that adding the symptom to be added to the sample symptom set can easily lead to the deterioration of the prediction performance of the pre-diagnosis model and the symptom model to be optimized. Therefore, step 43 and subsequent steps can be executed directly.

[0168] Step 52: Add the symptoms to be added to the sample symptom set, and return to execute step 43.

[0169] It should be noted that when the above "model construction method" includes steps 41-52, the above "second stop condition" can include: the occurrence probability of the "second symptom prediction result" stopping inquiry is higher than the occurrence probability of any symptom in the second symptom prediction result (that is, the "symptom to be inquired" determined based on the second symptom prediction result includes "stop inquiry").

[0170] It should also be noted that the execution time of step 52 is later than the execution time of step 47, and also later than the execution time of step 46, but earlier than the execution time of step 48.

[0171] Based on the above related content of steps 41-52, in some cases, the updating process of the to-be-optimized diagnosis model, the updating process of the to-be-optimized symptom model, and the updating process of the sample symptom set can be used to guide the to-be-optimized diagnosis model and the to-be-optimized symptom model to achieve a better learning process, which is conducive to further improving the model prediction performance.

[0172] Method embodiment three

[0173] In order to further improve the inquiry effect, the embodiment of the application also provides another possible implementation manner of the above S222, which can specifically include steps 61-63:

[0174] Step 61: obtaining at least one candidate symptom according to the first symptom prediction result.

[0175] The above "at least one candidate symptom" is used to represent a symptom with a higher occurrence probability in the first symptom prediction result; and the application embodiment does not limit the determination process of the "at least one candidate symptom", for example, it can include: determining at least one symptom in the first symptom prediction result that meets the third probability condition as the at least one candidate symptom.

[0176] The above "third probability condition" can be pre-set; and the application embodiment does not limit the "third probability condition", for example, the "third probability condition" can be: reaching a preset probability threshold (for example, 0.15). For another example, when the first symptom prediction result is used to record at least one symptom arranged in descending order of occurrence probability, the "third probability condition" can also be: the arrangement serial number is lower than E (that is, the top E symptoms are ranked). Wherein, E is a positive integer.

[0177] Step 62: using a pre-set invalid symptom screening rule to screen at least one candidate symptom to obtain at least one valid symptom.

[0178] The "invalid symptom screening rule" can be preset, and the embodiments of the present application do not limit the "invalid symptom screening rule", for example, it can include at least one of the following rules: (I) the symptoms already existing in the "to-be-used symptom set" are considered as invalid symptoms; (II) the symptoms that have been inquired from the patient to be diagnosed are considered as invalid symptoms; (III) the symptoms whose association degrees with each symptom in the "to-be-used symptom set" all satisfy a preset irrelevant condition are considered as invalid symptoms. It should be noted that the related content of the "association degree" can be referred to the related content in step 623 below.

[0179] The "preset irrelevant condition" can be preset, and the embodiments of the present application do not limit the "preset irrelevant condition", for example, it can be lower than a preset association degree threshold. For another example, the "preset irrelevant condition" can specifically include that the association degree between the to-be-screened symptom and each symptom in the "to-be-used symptom set" indicates that the to-be-screened symptom does not belong to the high-association symptom set corresponding to the to-be-used symptom set. The "to-be-screened symptom" can be any candidate symptom.

[0180] The "high-association symptom set corresponding to the to-be-used symptom set" is used to record the symptoms having a higher association degree with at least one symptom in the "to-be-used symptom set", and the embodiments of the present application do not limit the determination process of the "high-association symptom set corresponding to the to-be-used symptom set", for example, when the "to-be-used symptom set" includes Y symptoms, and the performance symptoms corresponding to at least one candidate disease include at least one alternative symptom, the determination process of the "high-association symptom set corresponding to the to-be-used symptom set" can specifically include: first, sorting the association degrees between each alternative symptom and the yth symptom in the "to-be-used symptom set" from large to small to obtain a symptom association ranking sequence corresponding to the yth symptom; then, collecting the symptoms with a ranking sequence number lower than a preset sequence number threshold (for example, 50) in the symptom association ranking sequence corresponding to the yth symptom to determine the high-association symptom set corresponding to the yth symptom; wherein y is a positive integer, y≤Y, and Y is a positive integer; finally, collecting the union set of the high-association symptom set corresponding to the first symptom, the high-association symptom set corresponding to the second symptom,..., and the high-association symptom set corresponding to the Yth symptom to determine the "high-association symptom set corresponding to the to-be-used symptom set".

[0181] The "valid symptom" is used to represent the candidate symptom that does not satisfy the invalid symptom screening rule.

[0182] The embodiments of the present application do not limit the implementation manner of step 62, for example, when the "invalid symptom screening rule" includes (I)-(III) above, step 62 can specifically include steps 621-624:

[0183] Step 621: If it is determined that the at least one candidate symptom includes at least one first symptom, each first symptom is determined as an invalid symptom. The first symptom belongs to the symptom set to be used.

[0184] The first symptom is used to represent a candidate symptom that is the same as a symptom in the symptom set to be used. It can be seen that the first symptom refers to a candidate symptom that belongs to the symptom set to be used, and the at least one first symptom refers to an intersection between the at least one candidate symptom and the symptom set to be used.

[0185] In addition, the embodiments of the present application do not limit the determination process of the first symptom. For example, when the at least one candidate symptom includes C candidate symptoms, the determination process of the first symptom can specifically include: if it is determined that there is a symptom in the symptom set to be used that is the same as the cth candidate symptom, the cth candidate symptom can be determined as the first symptom; if it is determined that there is no symptom in the symptom set to be used that is the same as the cth candidate symptom, the cth candidate symptom can be determined as not the first symptom. Wherein c is a positive integer, c≤C, and C is a positive integer.

[0186] Step 622: If it is determined that the at least one candidate symptom includes at least one second symptom, each second symptom is determined as an invalid symptom.

[0187] The second symptom is used to represent a symptom that has been asked to the patient to be diagnosed before the current round of inquiry interaction process. It can be seen that the second symptom refers to a candidate symptom that belongs to the historical inquiry symptom set, and the at least one second symptom refers to an intersection between the at least one candidate symptom and the historical inquiry symptom set. The historical inquiry symptom set is used to record symptoms that have been asked to the patient to be diagnosed (that is, symptoms that have been asked to the patient to be diagnosed before the current round of inquiry interaction process).

[0188] In addition, the embodiments of the present application do not limit the determination process of the second symptom. For example, when the at least one candidate symptom includes C candidate symptoms, the determination process of the second symptom can specifically include: if it is determined that there is a symptom in the historical inquiry symptom set that is the same as the cth candidate symptom, the cth candidate symptom can be determined as the second symptom; if it is determined that there is no symptom in the historical inquiry symptom set that is the same as the cth candidate symptom, the cth candidate symptom can be determined as not the second symptom. Wherein c is a positive integer, c≤C, and C is a positive integer.

[0189] Step 623: If it is determined that the at least one candidate symptom includes at least one third symptom, each third symptom is determined as an invalid symptom.

[0190] The "third symptom" is used to represent a candidate symptom which satisfies the preset irrelevant condition with respect to the degree of association between the third symptom and any symptom in the symptom set to be used.

[0191] In addition, the embodiment of the present application does not limit the determination process of the "third symptom". For example, when the "at least one candidate symptom" includes C candidate symptoms, the determination process of the "third symptom" can specifically include: first, calculating the degree of association between the cth candidate symptom and each symptom in the symptom set to be used, to obtain at least one to-be-used association degree corresponding to the cth candidate symptom; and then, judging whether the at least one to-be-used association degree corresponding to the cth candidate symptom satisfies the preset irrelevant condition or not. If yes, it can be determined that the degree of association between the cth candidate symptom and each symptom in the symptom set to be used is relatively low, and therefore the cth candidate symptom can be determined as the third symptom. If no, it can be determined that there is a symptom in the symptom set to be used which has a relatively high degree of association with the cth candidate symptom, and therefore the cth candidate symptom is not the third symptom. Wherein, c is a positive integer, c≤C, and C is a positive integer.

[0192] In addition, the embodiment of the present application does not limit the calculation process of the degree of association between two symptoms. For the convenience of understanding, the following will be described by taking an example.

[0193] As an example, when the symptom set to be used includes a to-be-compared symptom (for example, the cth candidate symptom), the determination process of the degree of association between the third symptom and the to-be-compared symptom can specifically include: step 71-step 72:

[0194] Step 71: If the to-be-compared symptom satisfies the first condition, the point mutual information (PMI) between the third symptom and the to-be-compared symptom is determined as the degree of association between the third symptom and the to-be-compared symptom.

[0195] The "first condition" can be preset, and the embodiment of the present application does not limit the "first condition". For example, it can specifically be: a positive symptom. It should be noted that in the medical field, each symptom has positive and negative, and it specifically includes: if the symptom is present, it is positive; if the symptom is not present, it is negative. For example, fever is a positive symptom, and no fever is a negative symptom.

[0196] The "point mutual information between the third symptom and the to-be-compared symptom" is used to describe the co-occurrence relationship between the third symptom and the to-be-compared symptom in the corpus. Wherein, the "corpus" is used to record the data in the medical field; and the "corpus" can include a large number of medical history data, a large number of medical book data, a large number of medical papers, etc.

[0197] Step 72: If the symptom to be compared satisfies the second condition, a preset correlation degree value is determined as the correlation degree between the third symptom and the symptom to be compared.

[0198] The above-mentioned "second condition" can be preset, and the embodiments of the present application do not limit the "second condition", for example, it can be specifically that: not belonging to a positive symptom (for example, belonging to a negative symptom, or belonging to an irrelevant symptom).

[0199] It should be noted that the "irrelevant symptom" refers to a symptom irrelevant to the disease suffered by the patient to be diagnosed, and the embodiments of the present application do not limit the determination process of the "irrelevant symptom", for example, for a historical inquiry symptom (for example, breast distending pain), if the patient to be diagnosed does not perform feedback within a preset feedback time for the historical inquiry symptom, it can be determined that the patient to be diagnosed considers that the historical inquiry symptom is irrelevant to the disease (for example, suspected respiratory disease) suffered by the patient to be diagnosed, and further determines that the historical inquiry symptom belongs to an irrelevant symptom; if the patient to be diagnosed completes the feedback within the preset feedback time for the historical inquiry symptom, the corresponding positive symptom (for example, breast distending pain) or negative symptom (for example, no breast distending pain) of the historical inquiry symptom can be determined according to the feedback content.

[0200] The above-mentioned "preset correlation degree value" can be preset, and the embodiments of the present application do not limit the "preset correlation degree value", for example, it can be specifically 0. It can be seen that the difference between the character information carried by the non-positive symptom (for example, the negative symptom or the irrelevant symptom) and the character information carried by the possible symptom to be inquired from the patient to be diagnosed is relatively large, so the preset correlation degree value can be directly determined as the correlation degree between the non-positive symptom and the possible symptom, so that the correlation degree can more accurately represent the large difference between the non-positive symptom and the possible symptom.

[0201] It should be noted that the correlation degree between the cth candidate symptom and the yth symptom in the symptom set to be used can be implemented by the above-mentioned steps 71 to 72, only by replacing the "third symptom" in the above-mentioned steps 71 to 72 with the "c th candidate symptom", and replacing the "symptom to be compared" with the yth symptom. Wherein, y is a positive integer, y≤Y, Y is a positive integer; c is a positive integer, c≤C, C is a positive integer.

[0202] Based on the related content of the above-mentioned step 623, after obtaining at least one candidate symptom, whether each candidate symptom satisfies the invalid symptom screening rule shown in (III) above can be determined according to the correlation degree between each candidate symptom and each symptom in the symptom set to be used, so as to determine the invalid symptom in these candidate symptoms by referring to the determination result subsequently.

[0203] Step 624: after determining that there is at least one invalid symptom in the at least one candidate symptom, deleting the at least one invalid symptom from the at least one candidate symptom to obtain at least one valid symptom.

[0204] In an embodiment of the present application, after determining that there is at least one invalid symptom (for example, at least one first symptom, at least one second symptom, and / or at least one third symptom, etc.) in the above-mentioned "at least one candidate symptom", these invalid symptoms can be deleted from the "at least one candidate symptom" to obtain at least one valid symptom, so that the "at least one valid symptom" is used to represent the candidate symptom that does not satisfy the "invalid symptom screening rule" shown in (I)-(III) above.

[0205] Based on the related content of the above step 62, after obtaining the at least one candidate symptom, at least one invalid symptom that satisfies the invalid symptom screening rule can be first screened from the candidate symptoms, and then the invalid symptoms are removed from the candidate symptoms to obtain at least one valid symptom, so that the valid symptoms are used to represent the candidate symptoms that do not satisfy the invalid symptom screening rule, so that the symptoms that need to be asked to the patient to be diagnosed in the current round of inquiry interaction process can be selected from the valid symptoms.

[0206] Step 63: determining the symptom to be inquired from the at least one valid symptom.

[0207] The embodiment of the present application does not limit the implementation of step 63, for example, it can specifically be that one (or, multiple) valid symptom is randomly selected from the at least one valid symptom and determined as the symptom to be inquired. For another example, it can specifically be that the valid symptom with the highest occurrence probability (or multiple valid symptoms with higher occurrence probability) in the at least one valid symptom is determined as the symptom to be inquired.

[0208] Based on the related content of the above steps 61 to 63, for a round of inquiry process for a patient to be diagnosed, after determining the first symptom prediction result by using the symptom set to be used, at least one candidate symptom can be first determined by referring to the first symptom prediction result, then at least one valid symptom can be obtained by removing those symptoms that do not need to be asked to the patient to be diagnosed (for example, invalid symptoms such as symptoms already existing in the "symptom set to be used", symptoms that have been asked, and symptoms irrelevant to all symptoms in the "symptom set to be used") from the candidate symptoms, and finally the symptom to be inquired can be selected from the valid symptoms, so as to effectively avoid asking some invalid symptoms to the patient to be diagnosed in the current round of inquiry interaction process, thereby facilitating to improve the inquiry effect.

[0209] Method embodiment four

[0210] To further improve the diagnosis effect, the embodiment of the present application also provides another possible implementation of S5, which can specifically include S51-S52:

[0211] S51: input the to-be-used symptom set into the pre-constructed disease prediction model to obtain a disease prediction result output by the disease prediction model.

[0212] The "disease prediction model" is used for disease prediction processing on input data of the disease prediction model; and the "disease prediction model" is not limited in the embodiment of the present application, for example, any machine learning model can be implemented.

[0213] The "disease prediction result" is used to represent the occurrence probability of at least one candidate disease (for example, the occurrence probability of rhinitis is 0.4; the occurrence probability of pneumonia is 0.39; the occurrence probability of dermatitis is 0.008; the occurrence probability of mastitis is 0.0001; …) of the patient to be diagnosed.

[0214] S52: determine the diagnosis result of the patient to be diagnosed according to the disease prediction result.

[0215] The embodiment of the present application does not limit the implementation of S52, for example, S52 can specifically include S521-S522:

[0216] S521: determine the to-be-used disease information according to the disease prediction result.

[0217] The embodiment of the present application does not limit the implementation of S521, for example, it can specifically include: determining the candidate disease with the highest occurrence probability (or multiple candidate diseases with higher occurrence probability) in the disease prediction result as the to-be-used disease information.

[0218] S522: determine the diagnosis result of the patient to be diagnosed according to the to-be-used disease information.

[0219] It should be noted that the related content of S522 can be referred to the related content of step 32 above.

[0220] Based on the related content of S51-S52 above, as shown in the following table, Figure 6 when the first stop condition is reached, it can be determined that the diagnosis process for the patient to be diagnosed has ended, so the pre-constructed disease prediction model can be used to perform disease prediction processing on all symptoms (for example, patient self-reported symptoms and symptoms obtained through the diagnosis interaction process) in the to-be-used symptom set, and the disease prediction result is obtained and output; then the diagnosis result of the patient to be diagnosed is determined by referring to the disease prediction result, so that the diagnosis result can more accurately represent the physical condition of the patient to be diagnosed, so that the doctor can make more accurate disease diagnosis on the patient to be diagnosed by referring to the diagnosis result.

[0221] It can be seen that, since the disease prediction model has good disease prediction performance, the disease prediction result obtained by using the disease prediction model to predict all symptoms in the set of symptoms to be used can better represent the probability of the patient suffering from various candidate diseases, so that the diagnosis result determined by referring to the disease prediction result can more accurately represent the physical condition of the patient (for example, which symptoms appear in the body, which symptoms do not appear, which diseases are suspected to have, etc.).

[0222] In addition, in order to further improve the diagnosis effect, the embodiment of the present application also provides another possible implementation of the above S5, which can specifically include steps 81-82:

[0223] Step 81: input the set of symptoms to be used into the mth disease prediction model to obtain the mth disease prediction result output by the mth disease prediction model. Wherein, m is a positive integer, m≤M.

[0224] The mth disease prediction model is used for disease prediction processing on the input data of the mth disease prediction model, and the present application does not limit the mth disease prediction model, for example, any machine learning model can be implemented.

[0225] In addition, in order to further improve the disease prediction accuracy, a plurality of disease prediction models with different disease prediction performances can be used to perform disease prediction processing on all symptoms in the set of symptoms to be used. It can be seen that the disease prediction performance of the mth disease prediction model is different from that of any other disease prediction model in the M disease prediction models except the mth disease prediction model. Specifically, the model structure (and / or construction process) of the mth disease prediction model is different from that of any other disease prediction model in the M disease prediction models except the mth disease prediction model.

[0226] Based on this, the embodiment of the present application also provides a possible implementation of the above "mth disease prediction model", which can specifically include a text vector extraction network and an mth text classification network, and the input data of the mth text classification network includes the output data of the text vector extraction network.

[0227] The above-mentioned "text vector extraction network" is used for text vector representation processing on data of the text vector extraction network; and the embodiment of the present application does not limit the "text vector extraction network", for example, which can be implemented by using the architecture of 12-layer transformer network (for example, 12-layer Encoder+12-layer Decoder) in Bidirectional Encoder Representation from Transformers (BERT).

[0228] The network structure of the above-mentioned "mth text classification network" is different from the network structure of the text classification network of any one of the other disease prediction models except the mth disease prediction model. For example, when M is 4, the 1st text classification network (that is, the text classification network of the 1st disease prediction model) can be implemented by using Congolutional Neural Networks (CNN), the 2nd text classification network (that is, the text classification network of the 2nd disease prediction model) can be implemented by using Recurrent Neural Network (RNN), the 3rd text classification network (that is, the text classification network of the 3rd disease prediction model) can be implemented by using Deep pyramid Congolutional Neural Networks for Text Categorization (DPCNN), and the 4th text classification network (that is, the text classification network of the 4th disease prediction model) can be implemented by using Rich feature hierarchies for accurate object detection and semantic segmentation (RCNN).

[0229] The embodiment of the present application does not limit the construction process of the above-mentioned "mth disease prediction model", for example, which can specifically include steps 91-step 92:

[0230] Step 91: After obtaining the pre-training model, the mth to-be-trained model is obtained according to the pre-training model and the mth text classification network.

[0231] The above-mentioned "pre-training model" is used for text vector representation processing on input data of the pre-training model; and the embodiment of the present application does not limit the "pre-training model", for example, which can be the BERT pre-training model published by Google.

[0232] It can be seen that after the BERT pre-training model released by Google is obtained, the mth text classification network (for example, CNN, RNN, DPCNN, or RCNN) can be added to the 12-layer transformer architecture of the BERT to obtain the mth to-be-trained model, so that the mth disease prediction model can be obtained by means of the training process (for example, the parameter fine-tuning process) for the mth to-be-trained model.

[0233] Step 92: training the mth to-be-trained model by using the third symptom sample and the actual disease diagnosis result of the third symptom sample to obtain the mth disease prediction model.

[0234] The above-mentioned "third symptom sample" is used to represent the sample data required when training the mth to-be-trained model; and the embodiments of the present application do not limit the "third symptom sample", for example, the "third symptom sample" can be a symptom set (for example, a symptom set similar to {cough, fever, sneezing}).

[0235] The above-mentioned "actual disease diagnosis result of the third symptom sample" is used to represent the disease actually diagnosed by the doctor referring to the symptom information carried by the third symptom sample; and the embodiments of the present application do not limit the acquisition method of the "actual disease diagnosis result of the third symptom sample", for example, it can be extracted from the medical record sample. For another example, it can be manually annotated by the doctor.

[0236] The embodiments of the present application do not limit the implementation of step 92, for example, any existing or future model training method can be used for implementation. For another example, in order to further improve the disease prediction performance of the mth disease prediction model, the embodiments of the present application further provide another possible implementation of step 92, which can specifically include steps 101-106:

[0237] Step 101: inputting the third symptom sample into the mth to-be-trained model to obtain the predicted disease diagnosis result of the third symptom sample output by the mth to-be-trained model.

[0238] The above-mentioned "predicted disease diagnosis result of the third symptom sample" is used to represent the possible disease determined by referring to all the symptoms in the third symptom sample; and the "predicted disease diagnosis result of the third symptom sample" is similar to the "to-be-used disease diagnosis result" mentioned above.

[0239] Step 102: judging whether the gap between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample meets the preset reservation condition, if yes, executing steps 103-104; if not, executing step 104.

[0240] The determination process of the "gap between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample" is not limited by the embodiments of the present application. For example, when the "actual disease diagnosis result of the third symptom sample" records the actual occurrence probability of rhinitis as 1, the actual occurrence probability of dermatitis as 0, the actual occurrence probability of breast cancer as 0, and so on, and the "predicted disease diagnosis result of the third symptom sample" records the actual occurrence probability of rhinitis as 0.8, the actual occurrence probability of dermatitis as 0.1, the actual occurrence probability of breast cancer as 0.05, and so on, the "gap between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample" can be determined according to a preset distance calculation formula (for example, Euclidean distance, cosine distance, etc.).

[0241] The "preset reservation condition" can be preset, and the "preset reservation condition" is not limited by the embodiments of the present application. For example, it is lower than a preset gap threshold.

[0242] Based on the related content of the above step 102, after obtaining the predicted disease diagnosis result of the third symptom sample, the gap between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample can be calculated. Then, it is determined whether the gap meets the preset reservation condition. If yes, it is determined that the disease prediction performance of the mth training model is better, and therefore the following steps 103-104 can be executed. However, if not, it is determined that the disease prediction performance of the mth training model is poorer, and therefore the following step 104 can be directly executed.

[0243] Step 103: determining the model parameters of the mth training model as a set of to-be-used model parameters.

[0244] The "set of to-be-used model parameters" is used to represent the model parameters required for constructing the mth disease prediction model. It can be seen that for each round of training process of the mth training model, as long as it is determined that the gap between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample meets the preset reservation condition, all model parameters of the mth training model in the current round can be directly determined as a set of to-be-used model parameters for saving processing. After completing all training processes for the mth training model, at least one set of saved to-be-used model parameters can be referred to for constructing the mth disease prediction model.

[0245] Step 104: determining whether a third stop condition is reached. If yes, step 106 is executed. If not, step 105 is executed.

[0246] The third stop condition can be preset, for example, any existing or future model stop condition (e.g., the number of model updates reaches a preset threshold, the mth to-be-trained model is in a convergent state, the prediction performance of the mth to-be-trained model reaches a preset performance threshold, etc.) can be used for implementation.

[0247] Step 105: updating the mth to-be-trained model according to the disease diagnosis prediction result of the third symptom sample and the actual disease diagnosis result of the third symptom sample, and returning to step 101.

[0248] In the embodiments of the present application, when it is determined that the third stop condition is not reached, it can be determined that the training process for the mth to-be-trained model is not yet completed, and therefore the difference between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample can be referred to for updating the mth to-be-trained model to obtain an updated mth to-be-trained model, so that subsequent step 101 and subsequent steps can be continued to be performed based on the updated mth to-be-trained model, to implement the next round of training process for the mth to-be-trained model.

[0249] Step 106: determining the mth disease prediction model according to the at least one to-be-used model parameter set.

[0250] The embodiments of the present application do not limit the implementation of step 106, for example, it can specifically include: first performing average processing (as shown in formulas (5)-(6)) on the at least one to-be-used model parameter set to obtain an average model parameter set; and then determining the mth disease prediction model according to the average model parameter set and the model structure of the mth to-be-trained model, so that the model structure of the mth disease prediction model is the same as that of the mth to-be-trained model, and all model parameters of the mth disease prediction model belong to the average model parameter set.

[0251]

[0252]

[0253] In the formula, A parameter represents the average model parameter set; P l represents the lth to-be-used model parameter set; represents the bth model parameter in the lth to-be-used model parameter set, b is a positive integer, b≤B, B is a positive integer, and B represents the number of model parameters in the mth to-be-trained model.

[0254] Based on the related content of the above step 106, when it is determined that the third stop condition is reached, it can be determined that all training processes for the mth to-be-trained model have been completed, and therefore the mth disease prediction model can be determined by referring to the model parameters included in the at least one to-be-used model parameter set and the model structure of the mth to-be-trained model, so that the disease prediction performance of the mth disease prediction model is more stable and accurate.

[0255] Based on the related content of the above steps 91 to 92, in some cases, in order to improve the training efficiency, each disease prediction model can be initialized by adding a text classification network to some pre-trained model (for example, a BERT pre-trained model); and then the multi-round model parameter fine-tuning processing is implemented for these disease prediction models by means of the third symptom sample and the actual disease diagnosis result thereof, so that each disease prediction model finally obtained has better disease prediction performance.

[0256] Based on the related content of the above step 81, when it is determined that the first stop condition is reached, it can be determined that the interrogation process for the to-be-diagnosed patient has ended, and therefore the disease prediction processing can be performed on all symptoms in the to-be-used symptom set by using the M disease prediction models constructed in advance, respectively, to obtain M disease prediction results, so that the interrogation result of the to-be-diagnosed patient can be determined by comprehensively considering the M disease prediction results subsequently.

[0257] Step 82: determining the interrogation result of the to-be-diagnosed patient according to the first disease prediction result to the Mth disease prediction result.

[0258] The embodiments of the present application do not limit the implementation manner of step 82, and in order to facilitate understanding, two examples are described below.

[0259] In example 1, when the mth disease prediction result includes the occurrence probability of at least one candidate disease, step 82 can specifically include: first determining the probability of having at least one candidate disease by using the first disease prediction result to the Mth disease prediction result; and then determining the interrogation result of the to-be-diagnosed patient according to the probability of having at least one candidate disease.

[0260] The probability of having the gth candidate disease is used to represent the possibility that the to-be-diagnosed patient has the gth candidate disease, and the determination process of the probability of having the gth candidate disease is not limited in the embodiments of the present application. For example, the third statistical analysis processing is performed on the occurrence probability of the gth candidate disease in the first disease prediction result, the occurrence probability of the gth candidate disease in the second disease prediction result, and the occurrence probability of the gth candidate disease in the Mth disease prediction result, to obtain the probability of having the gth candidate disease. Wherein, g is a positive integer, g≤G, and G is a positive integer, G representing the number of candidate diseases.

[0261] It should be noted that the above-mentioned "third statistical analysis processing" can be pre-set; and the embodiments of the present application do not limit the "third statistical analysis processing", for example, it can be an average value processing, a maximum value processing, or a minimum value processing, etc.

[0262] The embodiments of the present application do not limit the implementation of the above-mentioned step "determining the interrogation result of the patient to be diagnosed according to the probability of having the at least one candidate disease". For example, it can specifically include: first determining the candidate disease with the highest probability of having (or multiple candidate diseases with higher probability of having) as the disease information to be used; and then determining the interrogation result of the patient to be diagnosed according to the disease information to be used.

[0263] Based on the related content of the above-mentioned example 1, after obtaining the first disease prediction result to the Mth disease prediction result, the disease prediction result can be first statistically analyzed to obtain the likelihood of the patient to be diagnosed having each candidate disease; and then the interrogation result of the patient to be diagnosed is determined by referring to the likelihood of the patient to be diagnosed having each candidate disease, so that the interrogation result can more accurately represent the physical condition of the patient to be diagnosed.

[0264] Example 2, step 82 can specifically include steps 821-823:

[0265] Step 821: determining the mth disease to be used according to the mth disease prediction result. Wherein, m is a positive integer, m≤M.

[0266] The above-mentioned "mth disease to be used" refers to the disease with the maximum occurrence probability in the mth disease prediction result.

[0267] Step 822: performing first statistical analysis processing on the first disease to be used to the Mth disease to be used to obtain a statistical analysis result.

[0268] The above-mentioned "first statistical analysis processing" can be pre-set; and the embodiments of the present application do not limit the "first statistical analysis processing", for example, it can be an occurrence frequency statistical processing.

[0269] The above-mentioned "statistical analysis result" is used to represent the first statistical analysis processing result for the first disease to be used to the Mth disease to be used; and the embodiments of the present application do not limit the "statistical analysis result", for example, it can specifically be: the occurrence frequency of the first disease is w1, the occurrence frequency of the second disease is w2, the occurrence frequency of the third disease is w3, and w1+w2+w3=M. Wherein, w1, w2, w3 are positive integers.

[0270] Step 823: determining the interrogation result of the patient to be diagnosed according to the statistical analysis result.

[0271] As an example, when the statistical analysis result includes the occurrence frequency of at least one disease, and the inquiry result of the patient to be diagnosed carries the disease information, step 823 can specifically be: first, determining the disease with the highest occurrence frequency in the statistical analysis result as the disease information to be used; and then, determining the inquiry result of the patient to be diagnosed according to the disease information to be used.

[0272] Based on the related content of steps 81 to 82, when it is determined that the first stop condition is reached, it can be determined that the inquiry process for the patient to be diagnosed has ended, so the disease prediction model constructed in advance can be used to perform disease prediction processing on all symptoms in the symptom set to be used, to obtain multiple disease prediction results; and then, the inquiry result of the patient to be diagnosed is determined by referring to the disease prediction results, so that the disease information carried by the inquiry result is more stable and accurate, which is conducive to improving the inquiry effect.

[0273] In addition, in order to further improve the inquiry effect, the embodiment of the present application also provides another possible implementation manner of S5, which can specifically include steps 111-115:

[0274] Step 111: determining at least one disease prediction result according to the symptom set to be used and at least one disease prediction model.

[0275] It should be noted that step 111 can be implemented by any implementation manner of step 81 or any implementation manner of S51.

[0276] Step 112: determining whether the at least one disease prediction result meets a preset confidence condition, if yes, executing step 113; if no, executing steps 114-115.

[0277] The "preset confidence condition" can be preset, and the embodiment of the present application does not limit the "preset confidence condition", for example, it can specifically include: there is a disease prediction result meeting the high-quality prediction condition in the at least one disease prediction result. The "high-quality prediction condition" refers to the difference between the highest occurrence probability and the second highest occurrence probability in a disease prediction result is greater than a preset difference threshold (that is, the highest occurrence probability is much greater than the second highest occurrence probability).

[0278] It can be seen that after obtaining the at least one disease prediction result, it can be determined whether the disease prediction result meets the preset confidence condition; if it meets, it can be determined that there is at least one disease prediction result with high confidence in the disease prediction result, so the inquiry result of the patient to be diagnosed can be directly determined according to the disease prediction result; if it does not meet, it can be determined that the confidence of the disease prediction result is relatively low, so in order to improve the inquiry effect, steps 114-115 can be used to determine the inquiry result of the patient to be diagnosed.

[0279] Step 113: determining the inquiry result of the patient to be diagnosed according to the at least one disease prediction result.

[0280] It should be noted that step 113 can be implemented by any embodiment of step 82 or any embodiment of S52.

[0281] Step 114: selecting at least one reference symptom that meets the preset differential condition from at least one symptom corresponding to the at least one candidate disease according to the differential characteristic data of the at least one symptom corresponding to the at least one candidate disease.

[0282] The at least one symptom corresponding to the gth candidate disease belongs to the performance symptom corresponding to the gth candidate disease.

[0283] The differential characteristic data of the vth symptom corresponding to the gth candidate disease is used to represent the diagnosis influence degree of the vth symptom on the gth candidate disease. For example, as shown in Figure 7 “0.4610805162361092” is the differential characteristic data of the itching corresponding to the dermatitis. Wherein, g is a positive integer, g≤G, G is a positive integer, G represents the number of candidate diseases; v is a positive integer, v≤V g , V g is a positive integer, V g represents the number of symptoms in the “at least one symptom corresponding to the gth candidate disease”.

[0284] The embodiment of the present application does not limit the determination process of the above-mentioned “differential characteristic data”, for example, it can be realized by term frequency-inverse document frequency (tf-idf). In order to facilitate understanding, the following will be described in conjunction with examples.

[0285] As an example, when the above-mentioned “at least one symptom corresponding to the gth candidate disease” includes an evaluated symptom, the determination process of the differential characteristic data of the evaluated symptom can specifically include steps 121-124:

[0286] Step 121: determining a reference file corresponding to the gth candidate disease from the corpus.

[0287] As an example, when the corpus includes a large number of electronic medical records, and the electronic medical records include disease information diagnosed by doctors for patients, then step 121 can specifically include steps of: first searching at least one electronic medical record including the gth candidate disease from the corpus to obtain at least one electronic medical record corresponding to the gth candidate disease; and then determining the reference file corresponding to the gth candidate disease according to the at least one electronic medical record corresponding to the gth candidate disease. Wherein, g is a positive integer, g≤G, and G is a positive integer.

[0288] It should be noted that the embodiments of the present application do not limit the implementation of the above-mentioned step "determining the reference file corresponding to the gth candidate disease according to the at least one electronic medical record corresponding to the gth candidate disease", for example, the at least one electronic medical record corresponding to the gth candidate disease can be processed by set to obtain the reference file corresponding to the gth candidate disease. For another example, the at least one electronic medical record corresponding to the gth candidate disease can be processed by splicing to obtain the reference file corresponding to the gth candidate disease.

[0289] Step 122: determining the word frequency of the symptom to be evaluated according to the reference file corresponding to the gth candidate disease.

[0290] The above-mentioned "word frequency of the symptom to be evaluated" is used to represent the frequency of the symptom to be evaluated appearing in the above-mentioned "reference file corresponding to the gth candidate disease".

[0291] In addition, the embodiments of the present application do not limit the implementation of step 122, for example, it can specifically include: determining the ratio between the number of occurrences of the symptom to be evaluated in the "reference file corresponding to the gth candidate disease" and the total number of words in the "reference file corresponding to the gth candidate disease" as the word frequency of the symptom to be evaluated.

[0292] In addition, since the above-mentioned "reference file corresponding to the gth candidate disease" is generated based on a plurality of electronic medical records, in order to improve the accuracy of the discriminative representation data of the symptom to be evaluated, the embodiments of the present application also provide another possible implementation of step 122, in which the above-mentioned "reference file corresponding to the gth candidate disease" includes J g medical record files, and step 122 can specifically include steps 1221-1222:

[0293] Step 1221: determining the jth occurrence frequency of the symptom to be evaluated according to the jth medical record file in the above-mentioned "reference file corresponding to the gth candidate disease". Wherein, j is a positive integer, j≤J g , and J g is a positive integer.

[0294] The "jth occurrence frequency" is used to represent the frequency of the to-be-evaluated symptom in the jth medical record file in the "reference file corresponding to the gth candidate disease".

[0295] In addition, the embodiment of the present application does not limit the determination process of the "jth occurrence frequency", for example, it can specifically include: determining the ratio between the number of occurrences of the to-be-evaluated symptom in the jth medical record file and the total number of words in the jth medical record file as the jth occurrence frequency of the to-be-evaluated symptom.

[0296] Step 1222: performing second statistical analysis processing on the 1st occurrence frequency of the to-be-evaluated symptom to the Jth occurrence frequency of the to-be-evaluated symptom to obtain the word frequency of the to-be-evaluated symptom. g

[0297] The "second statistical analysis processing" can be pre-set, and the embodiment of the present application does not limit the "second statistical analysis processing", for example, it can be specifically taking the average value, taking the maximum value, taking the minimum value, etc.

[0298] Based on the related content of the above step 122, for the to-be-evaluated symptom, after obtaining the reference file corresponding to the gth candidate disease, the word frequency of the to-be-evaluated symptom can be determined by referring to the reference file corresponding to the gth candidate disease, so that the word frequency of the to-be-evaluated symptom can represent the frequency of the to-be-evaluated symptom in the reference file, so that the discriminative representation data of the to-be-evaluated symptom can be determined by referring to the word frequency of the to-be-evaluated symptom subsequently.

[0299] Step 123: determining the inverse document frequency of the to-be-evaluated symptom according to the reference files corresponding to the G candidate diseases.

[0300] The "inverse document frequency of the to-be-evaluated symptom" is used to measure the importance of the to-be-evaluated symptom to the diagnosis process of the gth candidate disease.

[0301] In addition, the embodiment of the present application does not limit the implementation manner of step 123, for example, it can specifically be: determining the ratio between G and the number of reference files including the to-be-evaluated symptom as the inverse document frequency of the to-be-evaluated symptom.

[0302] In addition, since each reference file is generated based on a plurality of electronic medical records, in order to improve the accuracy of the discriminative representation data of the to-be-evaluated symptom, the embodiment of the present application further provides two other possible implementation manners of step 123, which will be introduced respectively.

[0303] In one possible implementation manner, when the "reference file corresponding to the gth candidate disease" includes J g candidate diseases, step 123 can specifically include steps 131-133:​

[0304] Step 131: determining a first number value of the gth candidate disease according to a ratio between the number of the medical record files including the to-be-evaluated symptom and J g g , wherein g is a positive integer, and g≤G.

[0305] In the embodiments of the present application, for the J g medical record files included in the "reference files corresponding to the gth candidate disease", all target medical records including the to-be-evaluated symptom are first searched out from the J g medical record files; and then a ratio between the number of the target medical records and J g is determined as the gth first number value, so that the gth first number value can represent the importance of the to-be-evaluated symptom to the gth candidate disease; and the greater the gth first number value is, the more important the to-be-evaluated symptom is to the gth candidate disease. Wherein g is a positive integer, and g≤G.

[0306] Step 132: summing the first number value to the Gth first number value to obtain a second number value.

[0307] In the embodiments of the present application, for the to-be-evaluated symptom, after the first number value corresponding to the to-be-evaluated symptom to the Gth first number value are obtained, the G first number values can be summed to obtain a second number value corresponding to the to-be-evaluated symptom, so that the second number value can replace the "number of reference files including the to-be-evaluated symptom" to participate in the determination of the "inverse document frequency of the to-be-evaluated symptom".

[0308] Step 133: determining the inverse document frequency of the to-be-evaluated symptom according to a ratio between G and the second number value.

[0309] In the embodiments of the present application, for the to-be-evaluated symptom, after the second number value corresponding to the to-be-evaluated symptom is obtained, a ratio between G and the second number value can be determined as the inverse document frequency of the to-be-evaluated symptom.

[0310] Based on the related content of the above steps 131 to 133, for the to-be-evaluated symptom, the inverse document frequency of the to-be-evaluated symptom can be determined by referring to the occurrence of the to-be-evaluated symptom in the medical record texts corresponding to each disease, so that the inverse document frequency can better represent the importance of the to-be-evaluated symptom to the diagnosis process of the gth candidate disease.

[0311] In another possible implementation, the step 123 can include steps 141-143:

[0312] ​Step 141: Determine each reference file including the symptom to be evaluated in the G candidate disease corresponding reference files as a file to be used.

[0313] The "file to be used" is used to represent the reference file including the symptom to be evaluated; and the embodiment of the present application does not limit the determination process of the "file to be used", for example, it can specifically include: judging whether the symptom to be evaluated appears in the gth candidate disease corresponding reference file, if yes, determining the gth candidate disease corresponding reference file as the file to be used; if not, discarding the gth candidate disease corresponding reference file. Wherein, g is a positive integer, g≤G.

[0314] Step 142: Sum the occurrence probability of the symptom to be evaluated in all files to be used to obtain a third value. Wherein, the file to be used refers to the reference file including the symptom to be evaluated.

[0315] The occurrence probability of the symptom to be evaluated in the dth file to be used refers to the frequency of the symptom to be evaluated appearing in the dth file to be used. Wherein, d is a positive integer, d≤D, D is a positive integer, and D represents the number of files to be used.

[0316] In addition, the embodiment of the present application does not limit the determination process of the "occurrence probability of the symptom to be evaluated in the dth file to be used", for example, any of the above step 122 embodiments can be implemented, only need to replace "the gth candidate disease corresponding reference file" with "the dth file to be used" and "the word frequency of the symptom to be evaluated" with "the occurrence probability of the symptom to be evaluated in the dth file to be used" in any of the above step 122 embodiments.

[0317] Based on the related content of the above step 142, for the symptom to be evaluated, after obtaining D files to be used, the occurrence probability of the symptom to be evaluated in the dth file to be used is determined first; wherein, d is a positive integer, d≤D, D is a positive integer. Then, the occurrence probability of the symptom to be evaluated in the 1st file to be used, the occurrence probability of the symptom to be evaluated in the 2nd file to be used, …, and the occurrence probability of the symptom to be evaluated in the Dth file to be used are summed to obtain a third value corresponding to the symptom to be evaluated, so that the third value can replace the "number of reference files including the symptom to be evaluated" in the subsequent determination process of the "inverse file frequency of the symptom to be evaluated".

[0318] Step 143: Determine the inverse file frequency of the symptom to be evaluated according to the ratio between G and the third value.

[0319] In the embodiments of the present application, for the to-be-evaluated symptom, after the third numerical value corresponding to the to-be-evaluated symptom is obtained, the ratio between G and the third numerical value can be determined as the inverse document frequency of the to-be-evaluated symptom.

[0320] Based on the related content of steps 131 to 133, for the to-be-evaluated symptom, the inverse document frequency of the to-be-evaluated symptom can be determined by referring to the frequency of the to-be-evaluated symptom in the reference text corresponding to each disease, so that the inverse document frequency can better represent the importance of the to-be-evaluated symptom to the diagnosis process of the gth candidate disease.

[0321] Based on the related content of step 123, for the to-be-evaluated symptom, after the reference files corresponding to the G candidate diseases are obtained, the inverse document frequency of the to-be-evaluated symptom can be determined by referring to the G reference files, so that the inverse document frequency can represent the importance of the to-be-evaluated symptom to the diagnosis process of the gth candidate disease, so that the discriminative representation data of the to-be-evaluated symptom can be determined by referring to the inverse document frequency in the subsequent process.

[0322] Step 124: determining the discriminative representation data of the to-be-evaluated symptom according to the product between the word frequency of the to-be-evaluated symptom and the inverse document frequency of the to-be-evaluated symptom.

[0323] In the embodiments of the present application, for the to-be-evaluated symptom, after the word frequency of the to-be-evaluated symptom and the inverse document frequency of the to-be-evaluated symptom are obtained, the product between the word frequency of the to-be-evaluated symptom and the inverse document frequency of the to-be-evaluated symptom can be determined as the discriminative representation data of the to-be-evaluated symptom, so that the discriminative representation data can represent the diagnosis influence degree of the to-be-evaluated symptom on the gth candidate disease.

[0324] Based on the related content of steps 121 to 124, for a symptom corresponding to a disease, the discriminative representation data of the symptom can be determined by means of the calculation idea of tf-idf, so that the discriminative representation data can represent the diagnosis influence degree of the symptom on the disease.

[0325] It should be noted that the above "discriminative representation data of the vth symptom corresponding to the gth candidate disease" can be determined by steps 121 to 124, only by replacing "to-be-evaluated symptom" in steps 121 to 124 with "vth symptom". Wherein, g is a positive integer, g≤G, G is a positive integer; v is a positive integer, v≤V g , V g is a positive integer.

[0326] The preset identification condition can be preset, and the embodiment of the present application does not limit the preset identification condition. For example, it can be greater than a preset representation threshold (for example, 0.2).

[0327] The reference symptom is used to represent a symptom that meets the preset identification condition (that is, a symptom that is strongly representative of a certain disease). For example, as shown in the following table, when the preset identification condition is greater than 0.2, “itching”, “rash”, “breast pain”, and the like can be determined as reference symptoms. Figure 7

[0328] Based on the related content of 114, after it is determined that the at least one disease prediction result does not meet the preset confidence condition, some symptoms that are strongly representative of the candidate diseases can be selected from all symptoms of the candidate diseases as reference symptoms according to the identification representation data of the symptoms corresponding to the candidate diseases, so that the disease that the patient to be diagnosed is likely to have can be quickly predicted based on the reference symptoms subsequently.

[0329] Step 115: determining the interrogation result of the patient to be diagnosed according to the symptom matching result between the at least one reference symptom and the symptom set to be used.

[0330] The symptom matching result is used to represent whether there is a reference symptom that matches at least one symptom in the symptom set to be used successfully in the at least one reference symptom.

[0331] In addition, the embodiment of the present application does not limit the implementation manner of step 115. For example, when the at least one reference symptom includes a symptom to be matched, and the symptom matching result indicates that there is a symptom in the symptom set to be used that matches the symptom to be matched successfully, the candidate disease corresponding to the symptom to be matched can be determined as the disease information to be used, and the interrogation result of the patient to be diagnosed can be determined according to the disease information to be used.

[0332] Based on the related content of steps 111 to 115, when it is determined that the first stop condition is reached, it can be determined that the interrogation process for the patient to be diagnosed has ended, so at least one disease prediction model constructed in advance can be used to perform disease prediction processing on all symptoms in the symptom set to be used respectively to obtain at least one disease prediction result, and then it is determined whether the disease prediction results are credible, so that when the credibility of the disease prediction results is relatively low, the symptom matching result between the symptoms that are strongly representative of a large number of candidate diseases and the symptom set to be used can be used to determine the interrogation result of the patient to be diagnosed, so that the disease information carried by the interrogation result is more accurate and credible, which is beneficial to improve the interrogation effect.

[0333] ​Based on the method provided in the above method embodiment, the present embodiment further provides a diagnosis device, which is explained and described below in combination with the drawings.

[0334] Device embodiment

[0335] The device embodiment introduces a diagnosis device, and the related content can be found in the above method embodiment.

[0336] Referring to Figure 8 The figure is a structural schematic diagram of a diagnosis device provided in the present embodiment.

[0337] The diagnosis device 800 provided in the present embodiment comprises:

[0338] The first determination unit 801 is configured to, after obtaining the illness self-description information of the patient to be diagnosed, determine a symptom set to be used according to the illness self-description information.

[0339] The second determination unit 802 is configured to determine a symptom to be inquired according to the symptom set to be used.

[0340] The third determination unit 803 is configured to, after obtaining the inquiry feedback information of the patient to be diagnosed on the symptom to be inquired, update the symptom set to be used according to the inquiry feedback information, and continue to execute the step of determining the symptom to be inquired according to the symptom set to be used until a first stop condition is reached, and determine the diagnosis result of the patient to be diagnosed according to the symptom set to be used.

[0341] In a possible implementation, the second determination unit 802 comprises:

[0342] The first determination sub-unit is configured to determine a disease diagnosis result to be used according to the symptom set to be used.

[0343] The second determination sub-unit is configured to determine a symptom to be inquired according to the symptom set to be used and the disease diagnosis result to be used.

[0344] The third determination unit 803 is specifically configured to, after obtaining the inquiry feedback information of the patient to be diagnosed on the symptom to be inquired, update the symptom set to be used according to the inquiry feedback information, and continue to execute the step of determining the disease diagnosis result to be used according to the symptom set to be used until a first stop condition is reached, and determine the diagnosis result of the patient to be diagnosed according to the symptom set to be used.

[0345] In a possible implementation, the first determining subunit is specifically configured to: input the to-be-used symptom set into a pre-constructed disease pre-diagnosis model to obtain a first disease pre-diagnosis result output by the disease pre-diagnosis model; and determine the to-be-used disease diagnosis result according to the first disease pre-diagnosis result.

[0346] and / or,

[0347] The second determining subunit is specifically configured to: input the to-be-used symptom set and the to-be-used disease diagnosis result into a pre-constructed symptom prediction model to obtain a first symptom prediction result output by the symptom prediction model; and determine the to-be-asked symptom according to the first symptom prediction result.

[0348] In a possible implementation, the disease pre-diagnosis model comprises a feature vector determining module, N pooling modules, a splicing module, a full connection module, and a decision module; where N is a positive integer.

[0349] The first determining subunit comprises:

[0350] The third determining subunit is configured to: input the to-be-used symptom set into the feature vector determining module to obtain a vector feature result output by the feature vector determining module; input the vector feature result into an nth pooling module to obtain an nth pooling result output by the nth pooling module; where n is a positive integer, and n≤N; input the 1st pooling result to the Nth pooling result into the splicing module to obtain a splicing result output by the splicing module; input the splicing result into the full connection module to obtain a full connection result output by the full connection module; and input the full connection result into the decision module to obtain the first disease pre-diagnosis result output by the decision module.

[0351] In a possible implementation, the inquiry device 800 further comprises:

[0352] The first construction unit is configured to train a to-be-optimized pre-diagnosis model by using a first symptom sample and an actual disease diagnosis result of the first symptom sample, obtain a sample symptom set according to a second symptom sample, input the sample symptom set into the to-be-optimized pre-diagnosis model to obtain a second disease pre-diagnosis result output by the to-be-optimized pre-diagnosis model, obtain a second symptom prediction result according to the sample symptom set, the second disease pre-diagnosis result and a to-be-optimized symptom model, update the to-be-optimized pre-diagnosis model according to the second disease pre-diagnosis result and an actual associated symptom corresponding to the second symptom sample, update the to-be-optimized symptom model according to the second symptom prediction result and the actual associated symptom corresponding to the sample symptom set, and continue to perform the step of inputting the sample symptom set into the to-be-optimized pre-diagnosis model until a second stop condition is reached, and then determine the disease pre-diagnosis model and the symptom prediction model according to the to-be-optimized pre-diagnosis model and the to-be-optimized symptom model.

[0353] In a possible implementation, the interrogation device 800 further includes:

[0354] The symptom set updating sub-unit is configured to determine a to-be-added symptom according to the second symptom prediction result before the step of continuing to perform the inputting of the sample symptom set into the to-be-optimized pre-diagnosis model, and add the to-be-added symptom to the sample symptom set if it is determined that the to-be-added symptom belongs to the actual associated symptom corresponding to the sample symptom set.

[0355] In a possible implementation, the model architecture of the disease pre-diagnosis model is the same as that of the symptom prediction model.

[0356] In a possible implementation, the second determining sub-unit includes:

[0357] The fourth determining sub-unit is configured to obtain at least one candidate symptom according to the first symptom prediction result.

[0358] The fifth determining sub-unit is configured to perform screening processing on the at least one candidate symptom by using a pre-set invalid symptom screening rule to obtain at least one valid symptom.

[0359] The sixth determining sub-unit is configured to determine the to-be-asked symptom from the at least one valid symptom.

[0360] In a possible implementation, the fifth determining subunit is specifically configured to: if it is determined that the at least one candidate symptom includes at least one first symptom, determine each of the first symptoms as an invalid symptom; the first symptom belongs to the set of symptoms to be used; if it is determined that the at least one candidate symptom includes at least one second symptom, determine each of the second symptoms as an invalid symptom; the second symptom belongs to a historical inquiry symptom set; the historical inquiry symptom set is used to record symptoms that have been inquired from the patient to be diagnosed; if it is determined that the at least one candidate symptom includes at least one third symptom, determine each of the third symptoms as an invalid symptom; the degree of association between the third symptom and any symptom in the set of symptoms to be used satisfies a preset irrelevant condition; and after it is determined that there is at least one invalid symptom in the at least one candidate symptom, delete the at least one invalid symptom from the at least one candidate symptom to obtain the at least one valid symptom.

[0361] In a possible implementation, the set of symptoms to be used includes a symptom to be compared, and the inquiry device 800 further includes:

[0362] The degree of association determining subunit is configured to: if the symptom to be compared satisfies a first condition, determine the point mutual information between the third symptom and the symptom to be compared as the degree of association between the third symptom and the symptom to be compared; and if the symptom to be compared satisfies a second condition, determine a preset degree of association value as the degree of association between the third symptom and the symptom to be compared.

[0363] In a possible implementation, the third determining unit 803 includes:

[0364] The seventh determining subunit is configured to input the set of symptoms to be used into a pre-constructed disease prediction model to obtain a disease prediction result output by the disease prediction model.

[0365] The eighth determining subunit is configured to determine an inquiry result of the patient to be diagnosed according to the disease prediction result.

[0366] In a possible implementation, the number of disease prediction models is M; M is a positive integer.

[0367] The seventh determining subunit is specifically configured to: input the set of symptoms to be used into an mth disease prediction model that is pre-constructed to obtain an mth disease prediction result output by the mth disease prediction model; m is a positive integer, and m≤M.

[0368] The eighth determining subunit is specifically configured to: determine the inquiry result of the patient to be diagnosed according to the 1st disease prediction result to the Mth disease prediction result.

[0369] In one possible implementation, the eighth determining subunit is specifically used to: determine the probability of having at least one candidate disease using the first to the Mth disease prediction results; and determine the consultation result of the patient to be diagnosed based on the probability of having the at least one candidate disease.

[0370] In one possible implementation, the eighth determining subunit is specifically used to: determine the m-th disease to be used based on the m-th disease prediction result; where m is a positive integer and m≤M; perform a first statistical analysis on the 1st to the Mth diseases to be used to obtain a statistical analysis result; and determine the consultation result of the patient to be diagnosed based on the statistical analysis result.

[0371] In one possible implementation, the consultation device 800 further includes:

[0372] The second construction unit is used to obtain the m-th model to be trained based on the pre-trained model and the m-th text classification network after obtaining the pre-trained model; and to train the m-th model to be trained using the third symptom sample and the actual disease diagnosis result of the third symptom sample to obtain the m-th disease prediction model.

[0373] In one possible implementation, the network structure of the m-th text classification network is different from the network structure of any other text classification network among the M text classification networks except for the m-th text classification network.

[0374] In one possible implementation, the second building unit includes:

[0375] The model training subunit is used to input the third symptom sample into the m-th model to be trained, and obtain the predicted disease diagnosis result of the third symptom sample output by the m-th model to be trained; if the difference between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample meets a preset retention condition, then the model parameters of the m-th model to be trained are determined as a set of model parameters to be used; based on the disease diagnosis prediction result of the third symptom sample and the actual disease diagnosis result of the third symptom sample, the m-th model to be trained is updated, and the step of inputting the third symptom sample into the m-th model to be trained is continued until a third stopping condition is reached, and the m-th disease prediction model is determined based on at least one set of model parameters to be used.

[0376] In one possible implementation, the model training subunit includes:

[0377] The ninth determining subunit is configured to perform averaging processing on the at least one to-be-used model parameter set to obtain an average model parameter set; and determine the mth disease prediction model according to the average model parameter set and a model structure of the mth to-be-trained model.

[0378] In a possible implementation, the inquiry device 800 further includes:

[0379] The fourth determining unit is configured to, when determining that the disease prediction result does not satisfy the preset confidence condition, select at least one reference symptom that satisfies a preset discrimination condition from at least one symptom corresponding to at least one candidate disease according to the discrimination feature data of the at least one symptom corresponding to the at least one candidate disease.

[0380] The fifth determining unit is configured to determine the inquiry result of the patient to be diagnosed according to a symptom matching result between the at least one reference symptom and the to-be-used symptom set.

[0381] In a possible implementation, the number of candidate diseases is G, and at least one symptom corresponding to a gth candidate disease includes a to-be-evaluated symptom; wherein g is a positive integer, g≤G, and G is a positive integer.

[0382] The inquiry device 800 further includes:

[0383] The sixth determining unit is configured to determine reference files corresponding to G candidate diseases from a corpus; determine a term frequency of the to-be-evaluated symptom according to the reference files corresponding to the gth candidate disease; determine an inverse document frequency of the to-be-evaluated symptom according to the reference files corresponding to the G candidate diseases; and determine discrimination feature data of the to-be-evaluated symptom according to a product between the term frequency of the to-be-evaluated symptom and the inverse document frequency of the to-be-evaluated symptom.

[0384] In a possible implementation, the reference files corresponding to the gth candidate disease include J g medical record files; wherein J g is a positive integer, g is a positive integer, g≤G, and G is a positive integer.

[0385] The sixth determining unit includes:

[0386] The tenth determining subunit is configured to determine a gth first value according to a ratio between a number of medical record files including the to-be-evaluated symptom in the J g medical record files and the J g ; wherein g is a positive integer, g≤G; perform summing processing on a first first value to a Gth first value to obtain a second value; and determine the inverse document frequency of the to-be-evaluated symptom according to a ratio between the G and the second value.

[0387] In a possible implementation, the sixth determining unit comprises:

[0388] The eleventh determining sub-unit is configured to determine each reference file including the to-be-evaluated symptom in the reference files corresponding to the G candidate diseases as a to-be-used file;

[0389] The occurrence probabilities of the to-be-evaluated symptom in all to-be-used files are summed to obtain a third numerical value;

[0390] The inverse document frequency of the to-be-evaluated symptom is determined according to a ratio between the G and the third numerical value.

[0391] In a possible implementation, the reference files corresponding to the gth candidate disease comprise J g medical record files;

[0392] The sixth determining unit comprises:

[0393] The twelfth determining sub-unit is configured to determine a jth occurrence frequency of the to-be-evaluated symptom according to a jth medical record file; where j is a positive integer, j≤J g , J g is a positive integer; and a first occurrence frequency of the to-be-evaluated symptom to a J g th occurrence frequency of the to-be-evaluated symptom are subjected to second statistical analysis processing to obtain a term frequency of the to-be-evaluated symptom.

[0394] In a possible implementation, the eighth determining sub-unit is specifically configured to: determine to-be-used disease information according to the disease prediction result; and perform set processing on the to-be-used disease information and the to-be-used symptom set to obtain an interrogation result of the to-be-diagnosed patient.

[0395] In a possible implementation, the interrogation device 800 further comprises:

[0396] The interrogation interaction unit is configured to: generate symptom interrogation information according to the to-be-interrogated symptom; send the symptom interrogation information to the to-be-diagnosed patient, so that the to-be-diagnosed patient replies to the symptom interrogation information; and after obtaining reply content of the to-be-diagnosed patient to the symptom interrogation information, determine the interrogation feedback information according to the reply content.

[0397] Further, an apparatus is also provided in the embodiments of the present application, which comprises: a processor, a memory, and a system bus;

[0398] The processor and the memory are connected through the system bus;

[0399] The memory is configured to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform any of the above-mentioned methods for inquiring.

[0400] Further, the embodiment of the present application further provides a computer readable storage medium, wherein instructions are stored in the computer readable storage medium, and when the instructions run on a terminal device, the terminal device executes any of the above-mentioned methods for inquiring.

[0401] Further, the embodiment of the present application further provides a computer program product, wherein the computer program product runs on a terminal device, and the terminal device executes any of the above-mentioned methods for inquiring.

[0402] From the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment methods can be implemented by means of software and necessary universal hardware platforms. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) execute the methods described in the various embodiments or some parts of the embodiments of the present application.

[0403] It should be noted that the various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0404] It should also be noted that, in this document, the terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.

[0405] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for taking medical history, characterized in that, Applied to electronic devices, the method includes: After obtaining the patient's self-reported medical condition information, the symptom set to be used is determined based on the self-reported medical condition information; Based on the symptom set to be used, the symptoms to be inquired about are determined. The symptoms to be inquired about are determined based on the output data of the symptom prediction model. The output data of the symptom prediction model is determined based on the symptom set to be used and the output data of the disease pre-diagnosis model. The output data of the disease pre-diagnosis model is determined based on the symptom set to be used. The model architecture of the symptom prediction model is the same as that of the disease pre-diagnosis model. The symptom prediction model and the disease pre-diagnosis model are jointly trained. After obtaining the inquiry feedback information of the patient to be diagnosed regarding the symptoms to be inquired about, the symptom set to be used is updated according to the inquiry feedback information, and the step of determining the symptoms to be inquired about based on the symptom set to be used continues to be executed until the first stopping condition is met, and the consultation result of the patient to be diagnosed is determined according to the symptom set to be used.

2. The method according to claim 1, characterized in that, The step of determining the symptoms to be inquired about based on the symptom set to be used includes: Based on the set of symptoms to be used, determine the diagnostic results of the diseases to be used; Based on the symptom set to be used and the disease diagnosis results to be used, determine the symptoms to be inquired about; The step of continuing to perform the step of determining the symptoms to be questioned based on the symptom set to be used includes: Continue with the step of determining the disease diagnosis result to be used based on the set of symptoms to be used.

3. The method according to claim 2, characterized in that, The step of determining the disease diagnosis result to be used based on the symptom set to be used includes: The symptom set to be used is input into a pre-built disease prediagnosis model to obtain a first disease prediagnosis result output by the disease prediagnosis model; the diagnosis result of the disease to be used is determined based on the first disease prediagnosis result. And / or, The step of determining the symptoms to be inquired about based on the symptom set to be used and the disease diagnosis results to be used includes: The symptom set to be used and the disease diagnosis results to be used are input into a pre-built symptom prediction model to obtain a first symptom prediction result output by the symptom prediction model; based on the first symptom prediction result, the symptom to be inquired about is determined.

4. The method according to claim 3, characterized in that, The disease prediagnosis model includes a representation vector determination module, N pooling modules, a splicing module, a fully connected module, and a decision module; where N is a positive integer. The process of determining the first disease prediagnosis result includes: The symptom set to be used is input into the representation vector determination module to obtain the vector representation result output by the representation vector determination module; The vector representation result is input into the nth pooling module to obtain the nth pooling result output by the nth pooling module; where n is a positive integer and n≤N; The first to Nth pooling results are input into the splicing module to obtain the splicing result output by the splicing module; The splicing result is input into the fully connected module to obtain the fully connected result output by the fully connected module; The fully connected result is input into the decision module to obtain the first disease prediagnosis result output by the decision module.

5. The method according to claim 3, characterized in that, The construction process of the disease prediagnosis model and the symptom prediction model includes: Using the first symptom sample and the actual disease diagnosis results of the first symptom sample, the model to be used is trained to obtain the pre-diagnosis model to be optimized; Based on the second symptom sample, determine the sample symptom set; The sample symptom set is input into the pre-diagnosis model to be optimized, and the second disease pre-diagnosis result output by the pre-diagnosis model to be optimized is obtained. Based on the sample symptom set, the second disease prediagnosis results, and the symptom model to be optimized, the second symptom prediction results are obtained; Based on the second disease prediagnosis result and the actual disease diagnosis result of the second symptom sample, the prediagnosis model to be optimized is updated. Based on the second symptom prediction result and the actual associated symptoms corresponding to the sample symptom set, the symptom model to be optimized is updated. The step of inputting the sample symptom set into the prediagnosis model to be optimized is continued until the second stopping condition is reached. Based on the prediagnosis model to be optimized and the symptom model to be optimized, the disease prediagnosis model and the symptom prediction model are determined.

6. The method according to claim 5, characterized in that, Before proceeding to the step of inputting the sample symptom set into the pre-diagnosis model to be optimized, the method further includes: Based on the second symptom prediction results, determine the symptoms to be added; If it is determined that the symptom to be added belongs to the actual associated symptom corresponding to the sample symptom set, then the symptom to be added is added to the sample symptom set.

7. The method according to claim 3, characterized in that, The model architecture of the disease prediagnosis model is the same as that of the symptom prediction model.

8. The method according to claim 3, characterized in that, The step of determining the symptom to be inquired about based on the first symptom prediction result includes: Based on the first symptom prediction result, at least one candidate symptom is obtained; Using pre-defined invalid symptom screening rules, at least one candidate symptom is screened to obtain at least one valid symptom; The symptom to be questioned is determined from the at least one valid symptom.

9. The method according to claim 8, characterized in that, The process of determining the at least one valid symptom includes: If it is determined that the at least one candidate symptom includes at least one first symptom, then each of the first symptoms is determined to be an invalid symptom; wherein the first symptom belongs to the set of symptoms to be used; If it is determined that the at least one candidate symptom includes at least one second symptom, then each of the second symptoms is determined to be an invalid symptom; wherein the second symptom belongs to the historical inquiry symptom set; the historical inquiry symptom set is used to record the symptoms that have been asked of the patient to be diagnosed; If it is determined that the at least one candidate symptom includes at least one third symptom, then each of the third symptoms is determined to be an invalid symptom; wherein the degree of correlation between the third symptom and any symptom in the set of symptoms to be used satisfies a preset no-correlation condition; After determining that at least one invalid symptom exists among the at least one candidate symptom, the at least one invalid symptom is removed from the at least one candidate symptom to obtain the at least one valid symptom.

10. The method according to claim 9, characterized in that, The symptom set to be used includes symptoms to be compared, and the process of determining the degree of association between the third symptom and the symptoms to be compared includes: If the symptom to be compared meets the first condition, then the point mutual information between the third symptom and the symptom to be compared is determined as the degree of correlation between the third symptom and the symptom to be compared. If the symptom to be compared meets the second condition, then the preset correlation value is determined as the degree of correlation between the third symptom and the symptom to be compared.

11. The method according to claim 1, characterized in that, The step of determining the consultation results of the patient to be diagnosed based on the symptom set to be used includes: The symptom set to be used is input into a pre-built disease prediction model to obtain the disease prediction results output by the disease prediction model. Based on the disease prediction results, the consultation results of the patient to be diagnosed are determined.

12. The method according to claim 11, characterized in that, The number of disease prediction models is M; where M is a positive integer; The step of inputting the symptom set to be used into a pre-built disease prediction model and obtaining the disease prediction result output by the disease prediction model includes: The symptom set to be used is input into the pre-constructed m-th disease prediction model to obtain the m-th disease prediction result output by the m-th disease prediction model; where m is a positive integer, m≤M; The step of determining the consultation results of the patient to be diagnosed based on the disease prediction results includes: Based on the first to the Mth disease prediction results, the consultation results of the patient to be diagnosed are determined.

13. The method according to claim 12, characterized in that, The step of determining the medical history of the patient to be diagnosed based on the first to the Mth disease prediction results includes: Using the first to the Mth disease prediction results, determine the probability of having at least one candidate disease; based on the probability of having at least one candidate disease, determine the consultation results of the patient to be diagnosed. or, The step of determining the medical history of the patient to be diagnosed based on the first to the Mth disease prediction results includes: Based on the prediction result of the m-th disease, the m-th disease to be used is determined; where m is a positive integer, m≤M; the first statistical analysis is performed on the first to the m-th diseases to be used to obtain the statistical analysis results; based on the statistical analysis results, the consultation results of the patients to be diagnosed are determined.

14. The method according to claim 12, characterized in that, The construction process of the m-th disease prediction model includes: After obtaining the pre-trained model, the m-th model to be trained is obtained based on the pre-trained model and the m-th text classification network. Using the third symptom sample and the actual disease diagnosis results of the third symptom sample, the m-th model to be trained is trained to obtain the m-th disease prediction model.

15. The method according to claim 14, characterized in that, The network structure of the m-th text classification network is different from the network structure of any other text classification network among the M text classification networks except for the m-th text classification network.

16. The method according to claim 14, characterized in that, The process of training the m-th model to obtain the m-th disease prediction model using the third symptom sample and the actual disease diagnosis result of the third symptom sample includes: The third symptom sample is input into the m-th training model to obtain the predicted disease diagnosis result of the third symptom sample output by the m-th training model. If the difference between the predicted disease diagnosis result of the third symptom sample and the actual disease diagnosis result of the third symptom sample meets the preset retention condition, then the model parameters of the m-th model to be trained are determined as the set of model parameters to be used. Based on the disease diagnosis prediction result of the third symptom sample and the actual disease diagnosis result of the third symptom sample, the m-th model to be trained is updated, and the step of inputting the third symptom sample into the m-th model to be trained continues until the third stopping condition is met. Based on at least one set of model parameters to be used, the m-th disease prediction model is determined.

17. The method according to claim 16, characterized in that, The step of determining the m-th disease prediction model based on at least one set of model parameters to be used includes: Average the at least one set of model parameters to be used to obtain an average set of model parameters. Based on the average model parameter set and the model structure of the m-th model to be trained, the m-th disease prediction model is determined.

18. The method according to claim 11, characterized in that, The method further includes: When it is determined that the disease prediction result does not meet the preset confidence conditions, at least one reference symptom that meets the preset identification conditions is selected from at least one symptom corresponding to at least one candidate disease, based on the discriminative characterization data of at least one symptom corresponding to at least one candidate disease. The consultation results of the patient to be diagnosed are determined based on the symptom matching results between the at least one reference symptom and the symptom set to be used.

19. The method according to claim 18, characterized in that, The number of candidate diseases is G, and at least one symptom corresponding to the g-th candidate disease includes the symptom to be evaluated; where g is a positive integer, g≤G, and G is a positive integer; The process of determining the differential characterization data of the symptoms to be evaluated includes: Identify reference documents corresponding to G candidate diseases from the corpus; Based on the reference document corresponding to the g-th candidate disease, determine the word frequency of the symptom to be evaluated; Based on the reference files corresponding to the G candidate diseases, determine the reverse file frequency of the symptom to be evaluated; The discriminative characterization data of the symptom to be evaluated are determined by multiplying the word frequency of the symptom to be evaluated with the reverse document frequency of the symptom to be evaluated.

20. The method according to claim 19, characterized in that, The reference document corresponding to the g-th candidate disease includes J. g One medical record file; wherein, the J g Let g be a positive integer, g ≤ G, and G be a positive integer; The process of determining the frequency of the reverse file of the symptom to be evaluated includes: J g The number of medical record files containing the symptoms to be evaluated and the number of J g The ratio between them is determined as the g-th first value; where g is a positive integer, and g≤G; The first to the Gth first values ​​are summed to obtain the second value; The reverse file frequency of the symptom to be evaluated is determined based on the ratio between G and the second value.

21. The method according to claim 19, characterized in that, The process of determining the frequency of reverse files for the symptoms to be evaluated includes: Each reference file containing the symptoms to be evaluated in the reference files corresponding to the G candidate diseases is identified as a file to be used; The probability of occurrence of the symptoms to be evaluated described in all the files to be used is summed to obtain a third value; The reverse file frequency of the symptom to be evaluated is determined based on the ratio between G and the third value.

22. The method according to claim 19, characterized in that, The reference document corresponding to the g-th candidate disease includes J. g The process of determining the word frequency of the symptom to be evaluated includes: [Number of medical records] Based on the j-th medical record file, determine the j-th frequency of occurrence of the symptom to be evaluated; where j is a positive integer, j≤J g J g It is a positive integer; The frequency of the first occurrence of the symptom to be evaluated up to the Jth occurrence of the symptom to be evaluated g The frequency of occurrence is then subjected to a second statistical analysis to obtain the word frequency of the symptom to be evaluated.

23. The method according to claim 11, characterized in that, The step of determining the consultation results of the patient to be diagnosed based on the disease prediction results includes: Based on the disease prediction results, determine the disease information to be used; The disease information to be used and the symptom set to be used are combined and processed to obtain the consultation results of the patient to be diagnosed.

24. The method according to claim 1, characterized in that, The method further includes: Based on the symptoms to be inquired about, generate symptom inquiry information; The symptom inquiry information is sent to the patient to be diagnosed, so that the patient to be diagnosed can respond to the symptom inquiry information; After obtaining the patient's response to the symptom inquiry, the inquiry feedback information is determined based on the response.

25. A medical consultation device, characterized in that, include: The first determining unit is used to determine the symptom set to be used based on the patient's self-reported medical condition information after obtaining the information. The second determining unit is used to determine the symptoms to be inquired about based on the symptom set to be used. The symptoms to be inquired about are determined based on the output data of the symptom prediction model, which is determined based on the output data of the disease pre-diagnosis model, which is determined based on the symptom set to be used. The model architecture of the symptom prediction model is the same as that of the disease pre-diagnosis model, and the symptom prediction model and the disease pre-diagnosis model are jointly trained. The third determining unit is used to update the set of symptoms to be used based on the inquiry feedback information of the patient to be diagnosed regarding the symptoms to be inquired about after obtaining the inquiry feedback information, and continue to execute the step of determining the symptoms to be inquired about based on the set of symptoms to be used, until the first stopping condition is met, and then determine the consultation result of the patient to be diagnosed based on the set of symptoms to be used.

26. A device, characterized in that, The device includes: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 24.

27. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the method described in any one of claims 1 to 24.

28. A computer program product, characterized in that, When the computer program product is run on a terminal device, the terminal device causes the terminal device to perform the method described in any one of claims 1 to 24.

Images