An intelligent prediction method for disease staging, an electronic device, and a storage medium

Abstract

The application discloses an intelligent disease staging prediction method, an electronic device and a storage medium, and belongs to the technical field of intelligent medical treatment. The application collects case data of a certain disease, performs data cleaning, and constructs a medical data set; performs feature value extraction to obtain a feature value set F1; constructs a medical knowledge graph; extracts a keyword set of disease staging for one-hop query in the medical knowledge graph to obtain a subgraph set of the medical knowledge graph, which is input into a GDLN model to obtain a node feature set R1 output by the GDLN model; the F1 is divided into a numerical type feature set F2 and a text type feature set F3, the F2 is normalized, the F3 is shielded, and a fused set R2 is obtained; the R1 and the R2 are fused in a transverse connection mode, and then input into a deep feedforward neural network model to perform prediction on different stages of the disease. The application is used for disease staging prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent medical treatment, and particularly relates to an intelligent disease staging prediction method, an electronic device and a storage medium. BACKGROUND

[0002] The diagnosis and treatment process of a doctor is based on the patient's condition, combined with his own clinical experience accumulated for many years and medical professional knowledge in related fields, and a symptomatic diagnosis and treatment plan is proposed. In this process, the specific details of the patient's condition and professional medical knowledge are very important. The specific details of the patient's condition are obtained from the electronic medical record provided by the hospital, which is the basis for diagnosis. With the development of machine learning and the improvement of electronic medical records, data-driven intelligent medical diagnosis and treatment methods have become mainstream. Intelligent medical treatment is a hot topic in the academic field in recent years, and it is also a popular concern in the field of computer and medical treatment. Therefore, how to help disease staging prediction through intelligent medical treatment is a problem to be solved.

[0003] A knowledge graph is a structured semantic knowledge base used to describe concepts and their relationships in the physical world in symbolic form. In the medical field, a large amount of medical knowledge can be accurately stored, which is the basis for diagnosis. Knowledge graph has a wide range of applications in knowledge-driven systems, and entities and relationships are organized in the form of triples. Compared with traditional knowledge representation forms, knowledge graphs have a wider coverage and more diverse semantic information, and can quickly replicate the domain knowledge and clinical experience of medical experts. The intelligent medical prediction method based on it can provide medical decision support for general practitioners and improve medical quality. SUMMARY

[0004] The problem to be solved by the present application is to help disease staging diagnosis through intelligent medical treatment, and an intelligent disease staging prediction method, an electronic device and a storage medium are proposed.

[0005] To achieve the above-mentioned purpose, the technical scheme is as follows:

[0006] An intelligent disease staging prediction method comprises the following steps:

[0007] S1, collecting case data of a certain disease, performing data cleaning, and constructing a medical data set;

[0008] S2, extracting feature values from the medical data set obtained in step S1 to obtain a feature value set F1;

[0009] S3, constructing a medical knowledge graph of the certain disease in step S1;

[0010] S4. Extract the set of keywords K for disease staging from the feature value set F1 obtained in step S2. Perform a one-hop query on the medical knowledge graph of a certain disease constructed in step S3 using the obtained keyword set K to obtain the subgraph set G of the medical knowledge graph of a certain disease.

[0011] S5. Input the set of subgraphs G of the medical knowledge graph of a certain disease obtained in step S4 into the GDLN model for training, and obtain the set of node features R1 output by the GDLN model.

[0012] S6. Divide the feature set F1 obtained in step S2 into a numerical feature set F2 and a text feature set F3. Normalize the numerical feature set F2 and input the text feature set F3 into the MH-BERT model for masking to obtain the node features output by the MH-BERT model. Then, fuse the processed F2 and F3 in a horizontal connection manner to obtain the fused set R2.

[0013] S7. Fuse R1 and R2 in a horizontal connection manner, and then input them into a deep feedforward neural network model to predict different stages of the disease.

[0014] Furthermore, the characteristic values ​​mentioned in step S2 include the patient's age, gender, body mass index, chief complaint, present medical history, other medical history, and examination results, including blood, urine, and biochemical test results.

[0015] Furthermore, the medical knowledge graph described in step S3 is constructed based on the diagnostic criteria for diseases and authoritative medical books related to diseases, including disease-related graphs, etiology-related graphs, clinical manifestation-related graphs, laboratory test-related graphs, imaging test-related graphs, and auxiliary test-related graphs.

[0016] Furthermore, the method for extracting the keyword set K for disease staging in step S4 includes the following steps:

[0017] S4.1 Collect disease-related medical terms to build a keyword database;

[0018] S4.2 Then, the data in the feature value set F1 is matched with the data in the keyword lexicon to extract the keyword set K in the medical record data, K = {k1,k2,…,kn}, where kn is the keyword in the nth medical record data.

[0019] Furthermore, the specific implementation method of step S5 includes the following steps:

[0020] S5.1 Set the subgraph set G of the medical knowledge graph of a certain disease obtained in step S4 to include n nodes, and the feature of each node is represented as a D-dimensional vector;

[0021] S5.2, input the features of one node in G into the GDLN model, set the adjacency matrix of D as A, A is an n*n matrix, and the expression of the output node features is:

[0022]

[0023] wherein, H (L) is the node features of the Lth layer, sigmoid is the activation function of the GDLN model, W (L-1) is the weight matrix from the (L-1)th layer to the Lth layer, is the degree matrix of H , H (L-1) is the node features of the (L-1)th layer, is the adjacency matrix of the graph;

[0024] S5.3, input all the node features in the subgraph set G of the medical knowledge graph of a certain disease obtained in step S4 into the GDLN model in step S5.2 for training, and the node feature set output by the GDLN model is R1.

[0025] Further, the specific implementation method of step S6 includes the following steps:

[0026] S6.1, normalize the numerical type feature set F2: subtract the actual value of the feature value in the numerical type feature set F2 from the mean value and divide by the standard deviation, and the normalized feature z obtained satisfies the normal distribution with a mean of 0 and a standard deviation of 1, and the expression is:

[0027]

[0028] wherein, x2 is the feature value in the numerical type feature set F2, μ is the mean, and σ is the standard deviation;

[0029] S6.2, input the text type feature set F3 into the MH-BERT model for masking processing:

[0030] S6.2.1, construct the MH-BERT model, and the expression is:

[0031] M1=masking(x3)

[0032] H1=transformer(M1)

[0033] wherein, x3 is the feature value in the text feature set F3, M1 is the input of the first masking layer of the MH-BERT model, and H1 is the output of the first masking layer of the MH-BERT model;

[0034] S6.2.2 The output of each masking layer is passed to the next masking layer for further processing, as expressed in the following expression:

[0035] M L =masking(H L-1 )

[0036] H L =transformer(M L )

[0037] Among them, H L-1 M is the output of the (L-1)th layer. L H is the input of the Lth masking layer. L This is the output of the Lth masking layer;

[0038] The entire sequence is output from the last layer, yielding the node features output by the MH-BERT model.

[0039] S6.3. Merge F2 after processing in step S6.1 and F3 after processing in step S6.2 by horizontal connection to obtain the merged set R2.

[0040] Furthermore, the specific implementation method of step S7 includes the following steps:

[0041] S7.1 Merge R1 obtained in step S5 and R2 obtained in step S6 by horizontal connection to obtain set R;

[0042] S7.2. Input the features in set R obtained in step S7.1 into the deep feedforward neural network model to predict different stages of the disease;

[0043] S7.2.1 Construct a deep feedforward neural network model, the expression of which is:

[0044] h1 = f(W1r + b1)

[0045] Where r is the input data, h1 is the output of the first hidden layer of the feedforward neural network model, W1 is the weight of the first layer of the feedforward neural network model, and b1 is the bias term of the first layer of the feedforward neural network model.

[0046] The ReLU function is constructed as the activation function of the hidden layer. The expression for the ReLU function is:

[0047]

[0048] Where, f(z) i ) represents the result of performing a nonlinear transformation operation on the output of the hidden layer of the feedforward neural network model, where h is the output of the hidden layer of the feedforward neural network model;

[0049] The Softmax function is constructed as the activation function of the output layer to solve the multi-classification problem, and the expression of the Softmax function is as follows:

[0050]

[0051] Wherein, z i is the output value of the i th node, and C is the number of disease stages;

[0052] S7.2.2, the output of the previous hidden layer is used as the input of the subsequent hidden layer for subsequent calculation, and the expression is as follows:

[0053] h L-1 = f(W L-1 h L-2 +b L-1 )

[0054] Wherein, h L-2 is the output of the L-2 hidden layer, h L-1 is the output of the L-1 hidden layer, W L-1 and b L-1 are the weight and bias term of the L-1 layer respectively;

[0055] S7.2.3, the number of neurons of the input layer corresponds to the number of characteristic values in R1 and R2, and the number of neurons of the output layer corresponds to the number of disease stages, so that the label D={d1,d2,…,dn} is obtained, wherein d1 to dn are the diagnosed disease stages, and the classification cross-entropy is used as the loss function to further improve the accuracy of the prediction of different disease stages.

[0056] Electronic equipment, comprising a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of the intelligent disease stage prediction method.

[0057] A computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to realize the intelligent disease stage prediction method.

[0058] The beneficial effects of the present application are:

[0059] The intelligent disease stage prediction method provided by the present application uses a large number of electronic medical records to provide sufficient data sources for intelligent medical treatment, and combines knowledge graph to quickly copy the field knowledge and clinical experience of medical experts, thereby providing reliable basis for disease diagnosis and assisting clinical doctors in disease diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1A flow chart of an intelligent disease staging prediction method according to the present application. DETAILED DESCRIPTION

[0061] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application, that is, the described specific embodiments are only a part of the embodiments of the present application, but not all the specific embodiments. The components of the specific embodiments of the present application generally described and shown in the accompanying drawings can be arranged and designed in various different configurations, and the present application can also have other embodiments.

[0062] Therefore, the detailed description of the specific embodiments of the present application provided below in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected specific embodiments of the present application. All other specific embodiments obtained by those skilled in the art based on the specific embodiments of the present application without making creative efforts are within the scope of protection of the present application.

[0063] In order to further understand the inventive content, characteristics and effects of the present application, the following specific embodiments are exemplified, and the accompanying drawings are combined Figure 1 The detailed description is as follows: Specific embodiment one

[0065] An intelligent disease staging prediction method, comprising the following steps:

[0066] S1, collecting case data of a certain disease, performing data cleaning, and constructing a medical data set;

[0067] Further, it is found through observation of patient medical records that the medical records contain all data of the patient's current visit, part of the data involves patient privacy and does not affect the final diagnosis result, and part of the data cannot be directly used, so after obtaining the patient's medical record, the data needs to be cleaned for subsequent use;

[0068] S2, extracting feature values from the medical data set obtained in step S1 to obtain a feature value set F1;

[0069] Further, the feature values in step S2 include the patient's age, gender, body mass index, chief complaint, present illness history, other illness history, and examination results, and the examination results include blood, urine, and biochemical test results;

[0070] S3, constructing a medical knowledge graph of the certain disease described in step S1;

[0071] Further, the medical knowledge graph in step S3 is constructed based on the diagnostic criteria of diseases and authoritative medical books related to diseases, and includes disease class graphs, etiology class graphs, clinical manifestation class graphs, laboratory examination class graphs, imaging examination class graphs, and auxiliary examination class graphs.

[0072] Further, each class graph includes at least one entity, and the disease class graph has a connection relationship with all other class graphs.

[0073] S4, extract the key word set K of the disease stage in the feature value set F1 obtained in step S2, and perform one-hop query on the medical knowledge graph of a certain disease constructed in step S3 to obtain a subgraph set G of the medical knowledge graph of the certain disease.

[0074] Further, the method for extracting the key word set K of the disease stage in step S4 includes the following steps:

[0075] S4.1, collect disease-related medical professional terms to construct a key word library;

[0076] S4.2, then match the data in the feature value set F1 with the data in the key word library to extract the key word set K in the medical record data, K={k1, k2, …, kn}, kn is the key word in the nth medical record data.

[0077] Further, when writing medical records, doctors usually use some professional terms to accurately describe the patient's condition. These terms are often closely related to the patient's condition. Therefore, these terms are extracted as key words. For example, when describing a painful joint, professional joint names such as left knee joint, right knee joint, left wrist joint, and right wrist joint are used. These joint names are the key words that need to be extracted.

[0078] S5, input the subgraph set G of the medical knowledge graph of a certain disease obtained in step S4 into the GDLN model for training to obtain the node feature set R1 output by the GDLN model (Graph Deep Learning Network, GDLN).

[0079] Further, the specific implementation method of step S5 includes the following steps:

[0080] S5.1, set the subgraph set G of the medical knowledge graph of a certain disease obtained in step S4 to include n nodes, and the feature of each node is represented as a D-dimensional vector.

[0081] S5.2, input the feature of a node in G into the GDLN model, set the adjacency matrix of D as A, A is an n*n matrix, and the expression of the output node feature is:

[0082]

[0083] where H (L) is the node feature of the Lth layer, sigmoid is the activation function of the GDLN model, W (L-1) is the weight matrix from the (L-1)th layer to the Lth layer, is the degree matrix of H , H (L-1) is the node feature of the (L-1)th layer, is the adjacency matrix of the graph;

[0084] S5.3, input all the node features in the subgraph set G of the medical knowledge graph of a certain disease obtained in step S4 into the GDLN model in step S5.2 for training, and the node feature set output by the GDLN model is R1;

[0085] Further, the above formula can be regarded as a weighted sum of the features of the node and its neighbors, where the weight is the corresponding element of the adjacency matrix A. In this way, more feature information can be obtained through the interaction with the neighbor nodes on the basis of preserving the original features.

[0086] The training process of GDLN usually adopts a supervised learning method, which uses known node label information to train the network. Common loss functions include cross-entropy loss function and mean square error loss function, etc.

[0087] GDLN is a neural network model that can learn the interaction relationship between nodes, and can well process data related to nodes, and is suitable for feature extraction and classification tasks of graph structure data.

[0088] S6, divide the feature value set F1 obtained in step S2 into a numerical type feature set F2 and a text type feature set F3, normalize the numerical type feature set F2, input the text type feature set F3 into the MH-BERT model for masking processing, obtain the node features output by the MH-BERT model, and then fuse F2 and F3 processed in a transverse connection manner to obtain a fused set R2;

[0089] Further, the specific implementation method of step S6 includes the following steps:

[0090] S6.1, normalize the numerical type feature set F2: subtract the actual value of the feature value in the numerical type feature set F2 from the mean value and divide by the standard deviation, and the normalized feature z obtained satisfies the normal distribution with a mean of 0 and a standard deviation of 1, and the expression is:

[0091]

[0092] wherein x2 is a feature value in the numerical type feature set F2, μ is the mean, and σ is the standard deviation;

[0093] After standardization, the features between different dimensions can be compared numerically, and the size of the parameter value can reflect the contribution of different features to the sample.

[0094] S6.2, input the text type feature set F3 into the MH-BERT model for masking processing:

[0095] S6.2.1, construct the MH-BERT model, the expression is:

[0096] M1 = masking(x3)

[0097] H1 = transformer(M1)

[0098] wherein x3 is a feature value in the text feature set F3, M1 is the input of the first masking layer of the MH-BERT model, and h1 is the output of the first masking layer of the MH-BERT model;

[0099] S6.2.2, the output of each masking layer is passed to the next masking layer for further processing, the expression is:

[0100] M L =masking(H L-1 )

[0101] H L =transformer(M L )

[0102] wherein H L-1 is the output of the L-1 layer, M L is the input of the L layer masking layer, and H L is the output of the L layer masking layer;

[0103] until the last layer outputs the entire sequence, obtaining the node features output by the MH-BERT model;

[0104] MH-BERT is trained using medical data based on the BERT model. Medical domain knowledge is introduced into the BERT model, which can better handle the long tail distribution and other problems existing in medical vocabulary.

[0105] Different from the standard BERT model, the MH-BERT uses multiple masking layers in the encoding layer. The masking layers include a word group masking layer and a professional medical text masking layer based on the BERT model. In each masking layer, the model divides the input text sequence into multiple blocks according to a fixed rule, and a part of the words in each block are masked, while the remaining words can participate in the training of the model.

[0106] The BERT model with multiple masking layers can enhance the ability of the model to capture local information in the text sequence, because each masking layer can focus on different word blocks and learn specific language features. In addition, since each masking layer can be trained independently, the MH-BERT also has a certain parallel training capability.

[0107] S6.3, fuse F2 processed in step S6.1 and F3 processed in step S6.2 in a transverse connection manner to obtain a fused set R2;

[0108] S7, fuse R1 and R2 in a transverse connection manner, and then input to a deep feedforward neural network model to predict different stages of the disease;

[0109] Further, the specific implementation method of step S7 includes the following steps:

[0110] S7.1, fuse R1 obtained in step S5 and R2 obtained in step S6 in a transverse connection manner to obtain a set R;

[0111] S7.2, input the features in the set R obtained in step S7.1 to a deep feedforward neural network model to predict different stages of the disease;

[0112] S7.2.1, construct a deep feedforward neural network model, and the expression is:

[0113] h1=f(W1r+b1)

[0114] wherein, r is the input data, h1 is the output of the first hidden layer of the feedforward neural network model, W1 is the weight of the first layer of the feedforward neural network model, and b1 is the bias term of the first layer of the feedforward neural network model;

[0115] A Relu function is constructed as the activation function of the hidden layer, and the expression of the Relu function is:

[0116]

[0117] wherein, f(z i) is the result of the nonlinear transformation operation on the output of the hidden layer of the feedforward neural network model, h is the output of the hidden layer of the feedforward neural network model;

[0118] The Softmax function is constructed as an activation function of the output layer to solve the multi-classification problem, and the expression of the Softmax function is:

[0119]

[0120] Wherein, z i is the output value of the i th node, and C is the number of disease stages;

[0121] S7.2.2, the output of the previous hidden layer is used as the input of the subsequent hidden layer for subsequent calculation, and the expression is:

[0122] h L-1 = f(W L-1 h L-2 +b L-1 )

[0123] Wherein, h L-2 is the output of the L-2 hidden layer, h L-1 is the output of the L-1 hidden layer, W L-1 and b L-1 are the weight and bias of the L-1 layer, respectively;

[0124] S7.2.3, the number of neurons of the input layer corresponds to the number of characteristic values in R1 and R2, and the number of neurons of the output layer corresponds to the number of disease stages, so that the label D={d1, d2, …, dn} is obtained, wherein d1 to dn are the diagnosed disease stages, and the classification cross-entropy is used as the loss function to further improve the accuracy of the prediction of different disease stages. Specific implementation two:

[0126] The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of the intelligent disease stage prediction method.

[0127] The computer device of the application can be a device comprising a processor and a memory, such as a single-chip microcomputer comprising a central processing unit. Moreover, the processor is used to execute the computer program stored in the memory to realize the steps of the intelligent disease stage prediction method.

[0128] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0129] The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required by at least one function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices. Specific embodiment three

[0131] A computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the intelligent disease staging prediction method.

[0132] The computer readable storage medium of the present application can be any form of storage medium readable by the processor of the computer device, including but not limited to a non-volatile memory, a volatile memory, a ferroelectric memory, etc., and the computer readable storage medium has a computer program stored thereon. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-mentioned intelligent disease staging prediction method can be implemented.

[0133] The computer readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a USB (Universal Serial Bus) flash disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a ROM (Read-Only Memory), a RAM (Random Access Memory), an electrical carrier signal, a telecommunications signal, and a software distribution medium, and the like. It should be noted that the computer readable medium contains contents that can be appropriately added, deleted or modified according to requirements of legislation and patent practices in a jurisdiction, for example, in some jurisdictions, according to legislation and patent practices, the computer readable medium does not include an electrical carrier signal and a telecommunications signal.

[0134] It should be noted that the relational terms herein, such as first and second, and the like, are used solely to distinguish one from another entity or action, without necessarily requiring or implying any actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0135] Although the present application has been described with reference to specific implementations, it is apparent that various modifications and changes can be made thereto without departing from the scope of the present application. In particular, various features of the specific implementations of the present application can be interchanged and otherwise used in other specific implementations, without departing from the scope of the present application. Such modifications and changes are intended to fall within the scope of the application. Therefore, the present application is not limited to the specific implementations described herein, but only by the scope of the appended claims, and their equivalents.

Claims

1. An intelligent prediction method for disease staging, characterized in that, The method comprises the following steps: S1, collecting case data of a certain disease, performing data cleaning, and constructing a medical data set; S2, extracting feature values from the medical data set obtained in step S1 to obtain a feature value set F1; S3, constructing a medical knowledge graph of the certain disease described in step S1; S4, extracting a keyword set K of disease staging from the feature value set F1 obtained in step S2, and performing one-hop query on the keyword set K in the medical knowledge graph of the certain disease constructed in step S3 to obtain a subgraph set G of the medical knowledge graph of the certain disease; S5, inputting the subgraph set G of the medical knowledge graph of the certain disease obtained in step S4 into the GDLN model for training to obtain a node feature set R1 output by the GDLN model; The specific implementation method of step S5 comprises the following steps: S5.1, setting the subgraph set G of the medical knowledge graph of the certain disease obtained in step S4 to include n nodes, and the feature of each node is represented as a D-dimensional vector; S5.2, inputting the feature of a node in G into the GDLN model, setting the adjacency matrix of D as A, and A is an n*n matrix, then the expression of the output node feature is: ； wherein, is a node feature of the Lth layer, is an activation function of the GDLN model, is a weight matrix from the L-1th layer to the Lth layer, is a degree matrix of is a node feature of the L-1th layer, is an adjacency matrix of the graph;​ S5.3, inputting all node features in the subgraph set G of the medical knowledge graph of the certain disease obtained in step S4 into the GDLN model in step S5.2 for training, and the node feature set R1 output by the GDLN model is obtained; S6, dividing the feature value set F1 obtained in step S2 into a numerical type feature set F2 and a text type feature set F3, normalizing the numerical type feature set F2, inputting the text type feature set F3 into the MH-BERT model for masking processing, obtaining the node feature output by the MH-BERT model, and then fusing the processed F2 and F3 in a horizontal connection manner to obtain a fused set R2; The specific implementation method of step S6 comprises the following steps: S6.1, normalizing the numerical type feature set F2: subtracting the mean value of the actual value of the feature value in the numerical type feature set F2 from the actual value and dividing by the standard deviation to obtain the normalized feature satisfying the normal distribution with the mean value of 0 and the standard deviation of 1, and the expression is: ； wherein is a feature value in the numerical type feature set F2, is the mean, is the standard deviation; S6.2, inputting the text type feature set F3 into the MH-BERT model for masking processing: S6.2.1, constructing the MH-BERT model, and the expression is: ； ； wherein, is a feature value in the text feature set F3, is an input of the first masking layer of the MH-BERT model, is an output of the first masking layer of the MH-BERT model; S6.2.2, the output of each masking layer is passed to the next masking layer for further processing, and the expression is: ； ； wherein, is the output of the L-1 layer, is the input of the L layer of the masking layer, is the output of the L layer of the masking layer; until the last layer outputs the entire sequence, obtaining the node feature output by the MH-BERT model; S6.3, fusing the F2 processed in step S6.1 and the F3 processed in step S6.2 in a horizontal connection manner to obtain a fused set R2; S7, fusing R1 and R2 in a horizontal connection manner, and then inputting into a deep feedforward neural network model to predict different stages of the disease; The specific implementation method of step S7 comprises the following steps: S7.1, fusing R1 obtained in step S5 and R2 obtained in step S6 in a horizontal connection manner to obtain a set R; S7.2, inputting the features in the set R obtained in step S7.1 into a deep feedforward neural network model to predict different stages of the disease; S7.2.1, constructing a deep feedforward neural network model, and the expression is: ； wherein, is an input data, is an output of a first hidden layer of the feedforward neural network model, is a weight of the first layer of the feedforward neural network model, is a bias term of the first layer of the feedforward neural network model; The Relu function is constructed as the activation function of the hidden layer, and the expression of the Relu function is: ； wherein, is a result of a nonlinear transformation operation on an output of a hidden layer of the feedforward neural network model, h is the output of the hidden layer of the feedforward neural network model; The Softmax function is constructed as the activation function of the output layer to solve the multi-classification problem, and the expression of the Softmax function is: ； wherein, Yi is the output value of the i-th node, and C is the number of disease stages. S7.2.

2. The output of the previous hidden layer is used as input for the subsequent computation of the next hidden layer, expressed as: ; wherein, is the output of the L-2 hidden layer, is the output of the L-1 hidden layer, and are the weight and bias terms, respectively, of the L-1 layer. S7.2.3, finally, the number of neurons in the input layer corresponds to the number of characteristic values in R1 and R2, and the number of neurons in the output layer corresponds to the number of disease stages, that is, the label D={d1, d2,…, dn} is obtained, wherein d1 to dn are the diagnosed disease stages, and the classification cross-entropy is used as the loss function to improve the accuracy of the prediction of different disease stages.

2. The intelligent prediction method of disease staging according to claim 1, characterized in that, The characteristic values in step S2 include the age, gender, body mass index, chief complaint, history of present illness, history of other diseases, and examination results of the patient, and the examination results include blood, urine, and biochemical test results.

3. The intelligent prediction method of disease staging according to claim 1 or 2, characterized in that, The medical knowledge graph in step S3 is constructed based on the diagnostic criteria of diseases and authoritative medical books related to diseases, and includes disease class graphs, etiology class graphs, clinical manifestation class graphs, laboratory examination class graphs, imaging examination class graphs, and auxiliary examination class graphs.

4. The intelligent prediction method of disease staging according to claim 3, characterized in that, The method for extracting the keyword set K of the disease stage in step S4 includes the following steps: S4.1, collect medical professional terms related to diseases to construct a keyword library; S4.2, then match the data in the characteristic value set F1 with the data in the keyword library to extract the keyword set K in the medical record data, K={k1, k2, …, kn}, kn is the keyword in the nth medical record data.

5. An electronic device, characterized by The computer program is executed by the processor to realize the steps of the intelligent disease stage prediction method according to any one of claims 1-4.

6. A computer readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to realize the steps of the intelligent disease stage prediction method according to any one of claims 1-4.

Images