A method and system for predicting stroke risk

Through deep learning algorithms, the stroke risk prediction model is constructed, which solves the problems of low efficiency and poor accuracy of stroke risk assessment in the existing technology, and realizes efficient and accurate risk prediction and personalized preventive measures, which improves the stroke risk assessment Data value.

CN117976200BActive Publication Date: 2024-11-05SHUGUANG HOSPITAL AFFILIATED WITH SHANGHAI UNIV OF T C M +1
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202311321246.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-10-12
Publication Date
2024-11-05
Estimated Expiration
2043-10-12

AI Technical Summary

Technical Problem

In the prior art, the low efficiency and poor accuracy of stroke risk assessment and prediction have led to waste of medical resources and low data value, and the inability to effectively discover the deep correlation between clinical data and stroke risk, and cannot provide guiding opinions for prevention strategies.

Method used

A deep learning algorithm is used to construct a stroke risk prediction model, and personalized preventive measures are generated by preprocessing and knowledge graphs of stroke risk assessment knowledge big data and historical clinical big data, combined with real-time clinical data to reduce dimensionality and predict.

Benefits of technology

It improves the efficiency and accuracy of stroke risk prediction, explores the deep relationship between clinical data, physiological indicators and stroke risk, provides personalized preventive measures, and enhances the data value of the predicted results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117976200B_ABST
    Figure CN117976200B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of risk prediction technology, and discloses a method and system for predicting stroke risk. The method comprises the following steps: capturing stroke risk assessment knowledge big data and historical clinical big data; performing data preprocessing; constructing a stroke risk knowledge graph; performing data secondary processing; using a deep learning algorithm to construct a stroke risk prediction model; performing data dimension reduction on real-time clinical data; performing stroke risk prediction; and obtaining stroke prevention measures. The system comprises a big data capture module, a data preprocessing module, a knowledge graph construction module, a data secondary processing module, a stroke risk prediction model construction module, a data dimension reduction module, a stroke risk level prediction module, and a stroke prevention measures acquisition module, which are sequentially connected. The present invention solves the problems of low efficiency, poor accuracy, waste of medical resources, and low data value in the prior art.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of risk prediction, and in particular relates to a method and system for predicting stroke risk. Background Art

[0002] Stroke, also known as cerebrovascular accident, is a local brain dysfunction caused by acute cerebrovascular disease. Its clinical symptoms last for more than 24 hours. It has the characteristics of high morbidity, high mortality, high disability rate, and high recurrence rate. It often occurs in people aged 50 and above. The clinical manifestations are symptoms and signs of transient or permanent brain dysfunction. It is often characterized by sudden weakness, clumsiness, heaviness or numbness of one side of the limbs, numbness of one side of the face or crooked corner of the mouth, slurred speech accompanied by impaired consciousness or convulsions, and ranks first among the causes of death among urban residents.

[0003] With the high-paced work, over-nutritious diet and long-term lack of exercise, the age group of stroke has shown a trend of developing towards middle-aged and young people, becoming one of the major health issues that must be paid attention to in all age groups. People are paying more and more attention to the risk of stroke. The risk of stroke only represents the possibility of stroke, and provides reference opinions for people to assess their own condition or formulate prevention strategies.

[0004] In the existing technology, the assessment and prediction of stroke risk mainly relies on medical staff to conduct preliminary analysis of clinical data based on their professional knowledge. This method is inefficient and inaccurate, and cannot discover the deep correlation between clinical data and stroke risk. It also easily takes up the working time of medical staff, resulting in excessive pressure on medical resources. The stroke risk prediction result data obtained is of low value and is only a simple indicator that cannot provide guidance for prevention strategies. Summary of the invention

[0005] In order to solve the problems of low efficiency, poor accuracy, waste of medical resources and low data value in the prior art, the present invention aims to provide a method and system for predicting stroke risk.

[0006] The technical solution adopted by the present invention is:

[0007] A method for predicting stroke risk comprises the following steps:

[0008] Capture stroke risk assessment knowledge big data and historical clinical big data;

[0009] Preprocessing the stroke risk assessment knowledge big data and the historical clinical big data to obtain preprocessed stroke risk assessment knowledge big data and preprocessed historical clinical big data;

[0010] Construct a stroke risk knowledge graph based on the pre-processed stroke risk assessment knowledge big data;

[0011] Using the stroke risk knowledge graph, the pre-processed historical clinical big data was reprocessed to obtain a training sample set of historical stroke clinical key data;

[0012] Based on the historical stroke clinical key data training sample set, a stroke risk prediction model was constructed using a deep learning algorithm;

[0013] Collect users' real-time clinical data, perform data dimension reduction on the real-time clinical data, and obtain real-time key clinical data of stroke;

[0014] Input the real-time key clinical data of stroke into the stroke risk prediction model to predict the stroke risk and obtain the predicted result of the user's stroke risk level;

[0015] The user's stroke risk level prediction results and corresponding real-time clinical data are input into the stroke risk knowledge graph to obtain corresponding stroke prevention measures.

[0016] Furthermore, the stroke risk assessment knowledge big data includes a number of stroke risk assessment knowledge consisting of stroke professional terminology knowledge, stroke clinical manifestation knowledge, stroke risk judgment knowledge, and stroke prevention measures knowledge;

[0017] Historical clinical big data includes a number of historical clinical data consisting of user basic information, health consultation information, historical clinical indicators and historical clinical parameters.

[0018] Furthermore, the basic information of users includes age, gender, ethnicity, and permanent area of ​​residence;

[0019] Health information included whether the patient smoked, drank alcohol, had diabetes, had a stroke, and had a family history of stroke;

[0020] Historical clinical indicators include brain oxygen data, blood pressure data, pulse wave data, skin electricity data, sleep data, systolic blood pressure data after treatment, systolic blood pressure data without treatment, total cholesterol data, high-density lipoprotein cholesterol data, rectal temperature data, mean arterial pressure data, ventricular rate data, respiratory rate data, arterial oxygen partial pressure data, alveolar arterial oxygen saturation data, arterial pH data, serum sodium data, serum potassium data, blood creatinine data, hematocrit data, white blood cell count data, eye opening reaction data, speech reaction data and movement reaction data.

[0021] Further, data preprocessing is performed on the stroke risk assessment knowledge big data and the historical clinical big data to obtain the preprocessed stroke risk assessment knowledge big data and the preprocessed historical clinical big data, including the following steps:

[0022] The stroke risk assessment knowledge big data is sequentially subjected to error data elimination and duplicate data deletion to obtain the stroke risk assessment knowledge big data after preprocessing;

[0023] The historical clinical big data is processed in sequence by eliminating erroneous data, deleting duplicate data, desensitizing privacy data, and normalizing to obtain standardized historical clinical big data;

[0024] Data clustering and data balancing are performed on the standardized historical clinical big data to obtain the pre-processed historical clinical big data.

[0025] Further, the standardized historical clinical big data includes a number of standardized historical clinical data;

[0026] Performing data clustering and data balancing on the standardized historical clinical big data to obtain pre-processed historical clinical big data includes the following steps:

[0027] The membership of the FCM clustering algorithm is initialized using the Circle chaotic map to obtain the Circle chaotic map membership;

[0028] According to the membership of Circle Chaos Map, several cluster centers and corresponding data categories of FCM clustering algorithm are generated;

[0029] According to the Euclidean distance between the standardized historical clinical big data and the cluster center, the standardized historical clinical data within the preset distance range are divided into corresponding data categories;

[0030] If the merge function value of the cluster center or the change value of the merge function value is less than the preset threshold, the same amount of standardized historical clinical data in each data category is integrated to obtain the preprocessed historical clinical big data;

[0031] Otherwise, the cluster center and the corresponding data category are updated to obtain updated cluster centers and the corresponding updated data categories, and the data clustering process and the data balancing process are repeated according to the updated cluster centers and the corresponding updated data categories.

[0032] Furthermore, based on the pre-processed stroke risk assessment knowledge big data, a stroke risk knowledge graph is constructed, including the following steps:

[0033] Use the BERT-BILSTM-GAT-CRF algorithm to build a stroke named entity extraction model, and use BiGRU-Attention to build a stroke entity relationship extraction model;

[0034] A stroke named entity extraction model is used to extract several stroke named entities from the preprocessed stroke risk assessment knowledge big data, and a stroke entity relationship extraction model is used to extract several stroke entity relationships from the preprocessed stroke risk assessment knowledge big data;

[0035] A stroke risk knowledge graph is constructed based on several stroke named entities and corresponding stroke entity relationships.

[0036] Furthermore, based on the historical stroke clinical key data training sample set, a stroke risk prediction model was constructed using a deep learning algorithm, including the following steps:

[0037] The initial values ​​of the network parameters of the DBN network are used as the optimization target of the IGWO optimization algorithm;

[0038] According to the optimization goal, the IGWO optimization algorithm is used to iteratively optimize and obtain the optimal initial values ​​of network parameters;

[0039] According to the optimal initial values ​​of network parameters, the network structure of DBN network is set to obtain the initial stroke risk prediction model;

[0040] The historical stroke clinical key data training sample set was input into the initial stroke risk prediction model for optimization training to obtain the optimal stroke risk prediction model.

[0041] Furthermore, the chaotic mapping initialization mechanism, dynamic reverse learning mechanism and Gaussian mutation mechanism are introduced to improve the GWO optimization algorithm, and the IGWO optimization algorithm is obtained.

[0042] According to the optimization goal, the IGWO optimization algorithm is used to iteratively optimize and obtain the optimal initial value of the network parameters, including the following steps:

[0043] Set the IGWO population parameters, maximum number of iterations and fitness function of the IGWO optimization algorithm, and take the optimization target as the position of the IGWO individual in the IGWO population;

[0044] According to the IGWO population parameters, the Tent-Logistic-Cosine chaotic mapping sequence is used to initialize the IGWO population to obtain the initialized IGWO population;

[0045] According to the fitness function, calculate the fitness value of the IGWO individuals in the initialized IGWO population;

[0046] Perform encirclement, hunting and attack operations on the initialized IGWO population to obtain an updated IGWO population;

[0047] The IGWO individuals with the best fitness value in the updated IGWO population are subjected to dynamic reverse learning to obtain dynamic reverse IGWO individuals;

[0048] Perform Gaussian mutation on several IGWO individuals with poor fitness values ​​in the updated IGWO population to obtain several Gaussian mutated IGWO individuals;

[0049] If the current number of iterations is less than the maximum number of iterations, the IGWO population is updated based on the IGWO individuals with the best fitness value, the dynamically reversed IGWO individuals, and several Gaussian mutated IGWO individuals;

[0050] Otherwise, the position of the IGWO individual with the best response value is output as the optimal initial value of the network parameter.

[0051] Furthermore, the real-time clinical data of the user is collected, and the real-time clinical data is subjected to data dimension reduction to obtain real-time key clinical data of stroke, including the following steps:

[0052] Collect users' real-time clinical data and convert the real-time clinical data into a real-time clinical data matrix;

[0053] Obtaining a real-time clinical data core matrix corresponding to the real-time clinical data matrix;

[0054] Obtaining real-time clinical data eigenvalues ​​and real-time clinical data eigenvectors corresponding to the real-time clinical data kernel matrix;

[0055] Select the real-time clinical data eigenvalues ​​and real-time clinical data eigenvectors with the top 95% of cumulative variance contribution rates, perform dimensionality reduction processing on the real-time clinical data matrix, and obtain the real-time clinical data dimensionality reduction matrix;

[0056] According to the dimension reduction matrix of real-time clinical data, real-time key clinical data of stroke are obtained.

[0057] A stroke risk prediction system, used to implement a stroke risk prediction method, the system includes a big data capture module, a data preprocessing module, a knowledge graph construction module, a data secondary processing module, a stroke risk prediction model construction module, a data dimension reduction module, a stroke risk level prediction module and a stroke prevention measure acquisition module, which are connected in sequence;

[0058] Big data capture module, used to capture stroke risk assessment knowledge big data and historical clinical big data;

[0059] A data preprocessing module is used to preprocess the stroke risk assessment knowledge big data and the historical clinical big data to obtain the preprocessed stroke risk assessment knowledge big data and the preprocessed historical clinical big data;

[0060] A knowledge graph construction module is used to construct a stroke risk knowledge graph based on the pre-processed stroke risk assessment knowledge big data;

[0061] The data secondary processing module is used to use the stroke risk knowledge graph to perform secondary data processing on the pre-processed historical clinical big data to obtain a historical stroke clinical key data training sample set

[0062] The stroke risk prediction model building module is used to build a stroke risk prediction model based on the historical stroke clinical key data training sample set and using deep learning algorithms;

[0063] The data dimension reduction module is used to collect users' real-time clinical data, perform data dimension reduction on the real-time clinical data, and obtain real-time key clinical data of stroke;

[0064] The stroke risk level prediction module is used to input real-time key clinical data of stroke into the stroke risk prediction model to predict the stroke risk and obtain the user's stroke risk level prediction result;

[0065] The stroke prevention measures acquisition module is used to input the user's stroke risk level prediction results and corresponding real-time clinical data into the stroke risk knowledge graph to obtain corresponding stroke prevention measures.

[0066] The beneficial effects of the present invention are:

[0067] The present invention provides a method and system for predicting stroke risk. By analyzing historical clinical big data and using a deep learning algorithm to construct a stroke risk prediction model, the system performs automated stroke risk prediction, thereby avoiding the occupation of medical resources and improving the efficiency of stroke risk prediction. The system mines the deep relationship between the user's clinical data, physiological indicators, medical history and other information and stroke risk, thereby improving the accuracy of stroke risk prediction. The stroke risk knowledge graph constructed based on stroke risk assessment knowledge big data can formulate personalized stroke prevention measures according to the patient's individual characteristics and physiological conditions, thereby improving the data value of the stroke risk level prediction results.

[0068] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a flowchart of the method for predicting stroke risk in the present invention.

[0070] Figure 2 It is a structural block diagram of the stroke risk prediction system of the present invention. DETAILED DESCRIPTION

[0071] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0072] Embodiment 1:

[0073] like Figure 1 As shown, this embodiment provides a method for predicting stroke risk, comprising the following steps:

[0074] Capture stroke risk assessment knowledge big data and historical clinical big data;

[0075] The stroke risk assessment knowledge big data includes a number of stroke risk assessment knowledge consisting of stroke professional terminology knowledge, stroke clinical manifestation knowledge, stroke risk judgment knowledge, and stroke prevention measures knowledge;

[0076] Historical clinical big data includes several historical clinical data consisting of basic user information, health consultation information, historical clinical indicators and historical clinical parameters;

[0077] Basic user information includes age, gender, ethnicity, and permanent location;

[0078] Stroke is related to the user's age and gender, and there are also certain differences depending on the ethnicity's eating habits and permanent residence area. Therefore, age, gender, ethnicity, and permanent residence area are used as reference features;

[0079] Health information included whether the patient smoked, drank alcohol, had diabetes, had a stroke, and had a family history of stroke;

[0080] According to the research on stroke incidence, stroke is related to bad living habits such as smoking and drinking as well as genetic diseases, so the above information is used as a reference feature;

[0081] Historical clinical indicators include brain oxygen data, blood pressure data, pulse wave data, skin electricity data, sleep data, systolic blood pressure data after treatment, systolic blood pressure data before treatment, total cholesterol data, high-density lipoprotein cholesterol data, rectal temperature data, mean arterial pressure data, ventricular rate data, respiratory rate data, arterial oxygen partial pressure data, alveolar arterial oxygen saturation data, arterial blood pH data, serum sodium data, serum potassium data, blood creatinine data, hematocrit data, white blood cell count data, eye opening reaction data, speech reaction data and movement reaction data;

[0082] Clinical indicators take into account several indicators currently included in clinical examinations, and can dig out the deep relationship between physiological data and stroke risk as much as possible;

[0083] Data preprocessing is performed on the stroke risk assessment knowledge big data and the historical clinical big data to obtain the preprocessed stroke risk assessment knowledge big data and the preprocessed historical clinical big data, including the following steps:

[0084] The stroke risk assessment knowledge big data is sequentially subjected to error data elimination and duplicate data deletion to obtain the stroke risk assessment knowledge big data after preprocessing;

[0085] The historical clinical big data is sequentially processed by eliminating erroneous data, deleting duplicate data, desensitizing privacy data, and normalizing to obtain standardized historical clinical big data; the standardized historical clinical big data includes several standardized historical clinical data;

[0086] Eliminate erroneous data and duplicate data to reduce the volume of big data, improve data processing efficiency, and reduce computer computing and storage pressure. Desensitize privacy data by performing fuzzy desensitization on privacy data such as names and ID numbers contained in clinical data.

[0087] Performing data clustering and data balancing on the standardized historical clinical big data to obtain pre-processed historical clinical big data includes the following steps:

[0088] The membership of the FCM clustering algorithm is initialized using the Circle chaotic map, and the Circle chaotic map membership formula is:

[0089]

[0090] In the formula, u b,j are the Circle chaos mapping membership and initial membership of the b-th standardized historical clinical data to the j-th cluster center; the traditional FCM clustering algorithm randomly generates an initial membership of (0,1), which easily causes abnormal points of the initial membership or clustering of the distribution, resulting in poor clustering effect. The membership generated by the Circle chaos mapping is evenly distributed in the range of (0,1), which improves the clustering effect and accuracy;

[0091] According to the Circle chaos map membership, several cluster centers and corresponding data categories of the FCM clustering algorithm are generated. The formula is:

[0092]

[0093]

[0094] d bj =||x b -z j || 2

[0095] In the formula, z j is the jth cluster center; x b is the bth standardized historical clinical data in the standardized historical clinical big data; b, j and k are all indicators; m is the total number of standardized historical clinical data; c is the total number of cluster centers; d bj d bk is the Euclidean distance from the i-th standardized historical clinical data to the j-th and k-th cluster centers; α is a hyperparameter;

[0096] According to the Euclidean distance between the standardized historical clinical big data and the cluster center, the standardized historical clinical data within the preset distance range are divided into corresponding data categories;

[0097] If the combined function value of the cluster center or the change value of the combined function value is less than the preset threshold, the same amount of standardized historical clinical data in each data category is integrated to obtain the preprocessed historical clinical big data, the formula is:

[0098]

[0099] ΔJ t =J t -J t-1

[0100] In the formula, J t , J t-1 is the combined function value of the Lagrange multiplier method at time t and t-1; ΔJ t is the change value; b is the characteristic parameter of the membership of the bth Circle chaotic map;

[0101] Otherwise, the cluster center and the corresponding data category are updated to obtain updated cluster centers and the corresponding updated data categories, and the data clustering process and the data balancing process are repeated according to the updated cluster centers and the corresponding updated data categories;

[0102] According to the pre-processed stroke risk assessment knowledge big data, a stroke risk knowledge graph is constructed, including the following steps:

[0103] Use the BERT-BILSTM-GAT-CRF algorithm to build a stroke named entity extraction model, and use BiGRU-Attention to build a stroke entity relationship extraction model;

[0104] The bidirectional encoder representation from Transformers (BERT) pre-trained language sub-model performs vector representation on the pre-processed stroke risk assessment knowledge big data, the bidirectional long short-term memory network (BILSTM) network extracts the semantic features of the pre-processed stroke risk assessment knowledge big data represented by the vector, and the graph attention mechanism network (GAT) network extracts the graph features of the pre-processed stroke risk assessment knowledge big data represented by the vector, and the semantic features and graph features are interactively collaborated at a fine-grained level to obtain an interactive collaboration feature sequence, which is then input into the linear chain conditional random field (CRF) module to complete the annotation of stroke named entities and output stroke named entities;

[0105] The gated recurrent unit (GRU) structure converts the preprocessed stroke risk assessment knowledge big data into a vector representation. Then the sentence-level attention mechanism allocates attention to each sentence, reduces the impact of sentences with incorrect relationship annotations, and outputs the stroke entity relationship.

[0106] A stroke named entity extraction model is used to extract several stroke named entities from the preprocessed stroke risk assessment knowledge big data, and a stroke entity relationship extraction model is used to extract several stroke entity relationships from the preprocessed stroke risk assessment knowledge big data;

[0107] Construct a stroke risk knowledge graph based on several stroke named entities and corresponding stroke entity relationships;

[0108] Using the stroke risk knowledge graph, the pre-processed historical clinical big data was reprocessed to obtain a historical stroke clinical key data training sample set with alternative stroke risk level labels;

[0109] The stroke risk knowledge graph can sequentially reduce the data dimension and generate candidate stroke risk level labels for the pre-processed historical clinical big data, which can be used as label data for model training and provide data support for the deep learning algorithm of the supervised learning mechanism.

[0110] The traditional Grey Wolf Optimization (GWO) algorithm is improved by introducing the chaotic mapping initialization mechanism, dynamic reverse learning mechanism and Gaussian mutation mechanism, and the Improved Grey Wolf Optimization (IGWO) algorithm is obtained. The IGWO optimization algorithm is used to optimize the Deep Belief Network (DBN) in the deep learning algorithm, and the IGWO-DBN algorithm is obtained.

[0111] Based on the historical stroke clinical key data training sample set, a stroke risk prediction model was constructed using a deep learning algorithm, including the following steps:

[0112] The initial values ​​of the network parameters of the DBN network are used as the optimization target of the IGWO optimization algorithm;

[0113] According to the optimization goal, the IGWO optimization algorithm is used to iteratively optimize and obtain the optimal initial value of the network parameters, including the following steps:

[0114] Set the IGWO population parameters, maximum number of iterations and fitness function of the IGWO optimization algorithm, and take the optimization target as the position of the IGWO individual in the IGWO population;

[0115] According to the IGWO population parameters, the Tent-Logistic-Cosine chaotic mapping sequence is used to initialize the IGWO population, and the initialized IGWO population is obtained. The formula is:

[0116]

[0117] In the formula, q i+1 The initial position of the IGWO population generated by the Tent-Logistic-Cosine chaotic mapping sequence; q i is the initial position of the randomly generated GWO population; r is the preset parameter; i is the indicator;

[0118] According to the fitness function, the fitness value of the IGWO individual in the initialized IGWO population is calculated. The formula is:

[0119]

[0120] In the formula, f(q i ) is the IGWO individual q i The fitness value; MSE is the prediction mean square error function; Y' i , Y i are the predicted value and the true value; m' is the total number of IGWO individuals;

[0121] Perform encirclement, hunting and attack operations on the initialized IGWO population to obtain an updated IGWO population;

[0122] The formula for encirclement is:

[0123]

[0124] Where q(t+1) and q(t) are the positions of the IGWO individuals in the encirclement activity at time t+1 and t; q p (t) is the prey position; A is the convergence influence coefficient vector; C is the swing influence coefficient vector; D is the distance between the IGWO individual and the prey; a is the convergence factor; r1 and r2 are both random numbers between [0,1];

[0125] The formula for hunting is:

[0126]

[0127] The position update formula is:

[0128]

[0129] Where, q(t+1) is the position of the IGWO individual in the encirclement activity at the tth moment; q1, q2, q3 are the updated positions of the optimal leader wolves α, β and δ in the encirclement activity respectively; q α ,q β ,q δ are the positions of the optimal leader wolves α, β and δ respectively; D α , D β , D δ are the distances between the IGWO individual and the optimal leader wolf α, β and δ respectively;

[0130] Attack activity: After the gray wolf has finished hunting, it needs to attack the food to obtain the optimal solution. The attack of the gray wolf is determined by the convergence influence coefficient vector A, and A∈[-a,a]. When |A|>1, the IGWO group searches for prey globally. When |A|≤1, it conducts local precise attacks.

[0131] The IGWO individuals with the best fitness value in the updated IGWO population are subjected to dynamic reverse learning to obtain the dynamically reversed IGWO individuals. The formula is:

[0132]

[0133] In the formula, is a dynamically reversed IGWO individual; X best is the IGWO individual with the best fitness value; γ is the decreasing inertia coefficient, γ = 0.9-0.5t / T; L max , Lmin are the maximum and minimum values ​​of the vector space respectively;

[0134] Perform Gaussian mutation on several IGWO individuals with poor fitness values ​​in the updated IGWO population to obtain several Gaussian mutated IGWO individuals. The formula is:

[0135]

[0136] S=r4*(X best -X bad )*G(0,1)

[0137] In the formula, is the IGWO individual of Gaussian mutation; X bad is an IGWO individual with poor fitness value; S is the update step size; r4 is a random number between [0,1]; G(0,1) is a Gaussian function with a mean of 0 and a variance of 1; X best IGWO individuals with the best fitness value;

[0138] If the current number of iterations is less than the maximum number of iterations, the IGWO population is updated based on the IGWO individuals with the best fitness value, the dynamically reversed IGWO individuals, and several Gaussian mutated IGWO individuals;

[0139] Otherwise, the position of the IGWO individual with the best response value is output as the optimal initial value of the network parameter;

[0140] According to the optimal initial values ​​of network parameters, the network structure of DBN network is set to obtain the initial stroke risk prediction model;

[0141] Input the historical stroke clinical key data training sample set into the initial stroke risk prediction model for optimization training to obtain the optimal stroke risk prediction model;

[0142] Collect users' real-time clinical data, perform data dimension reduction on the real-time clinical data, and obtain real-time key clinical data of stroke, including the following steps:

[0143] Collect users' real-time clinical data and convert the real-time clinical data into a real-time clinical data matrix;

[0144] Obtaining a real-time clinical data core matrix corresponding to the real-time clinical data matrix;

[0145] Obtaining real-time clinical data eigenvalues ​​and real-time clinical data eigenvectors corresponding to the real-time clinical data kernel matrix;

[0146] Select the real-time clinical data eigenvalues ​​and real-time clinical data eigenvectors with the top 95% of cumulative variance contribution rates, perform dimensionality reduction processing on the real-time clinical data matrix, and obtain the real-time clinical data dimensionality reduction matrix;

[0147] According to the real-time clinical data dimension reduction matrix, real-time key clinical data of stroke are obtained;

[0148] Input the real-time key clinical data of stroke into the stroke risk prediction model to predict the stroke risk and obtain the predicted result of the user's stroke risk level;

[0149] The user's stroke risk level prediction results and corresponding real-time clinical data are input into the stroke risk knowledge graph to obtain corresponding stroke prevention measures.

[0150] Embodiment 2:

[0151] like Figure 2 As shown, this embodiment provides a stroke risk prediction system for implementing a stroke risk prediction method, the system includes a big data capture module, a data preprocessing module, a knowledge graph construction module, a data secondary processing module, a stroke risk prediction model construction module, a data dimension reduction module, a stroke risk level prediction module and a stroke prevention measure acquisition module connected in sequence;

[0152] Big data capture module, used to capture stroke risk assessment knowledge big data and historical clinical big data;

[0153] A data preprocessing module is used to preprocess the stroke risk assessment knowledge big data and the historical clinical big data to obtain the preprocessed stroke risk assessment knowledge big data and the preprocessed historical clinical big data;

[0154] A knowledge graph construction module is used to construct a stroke risk knowledge graph based on the pre-processed stroke risk assessment knowledge big data;

[0155] The data secondary processing module is used to use the stroke risk knowledge graph to perform secondary data processing on the pre-processed historical clinical big data to obtain a historical stroke clinical key data training sample set with alternative stroke risk level labels.

[0156] The stroke risk prediction model building module is used to build a stroke risk prediction model based on the historical stroke clinical key data training sample set and using deep learning algorithms;

[0157] The data dimension reduction module is used to collect users' real-time clinical data, perform data dimension reduction on the real-time clinical data, and obtain real-time key clinical data of stroke;

[0158] The stroke risk level prediction module is used to input real-time key clinical data of stroke into the stroke risk prediction model to predict the stroke risk and obtain the user's stroke risk level prediction result;

[0159] The stroke prevention measures acquisition module is used to input the user's stroke risk level prediction results and corresponding real-time clinical data into the stroke risk knowledge graph to obtain corresponding stroke prevention measures.

[0160] The present invention provides a method and system for predicting stroke risk. By analyzing historical clinical big data and using a deep learning algorithm to construct a stroke risk prediction model, the system performs automated stroke risk prediction, thereby avoiding the occupation of medical resources and improving the efficiency of stroke risk prediction. The system mines the deep relationship between the user's clinical data, physiological indicators, medical history and other information and stroke risk, thereby improving the accuracy of stroke risk prediction. The stroke risk knowledge graph constructed based on stroke risk assessment knowledge big data can formulate personalized stroke prevention measures according to the patient's individual characteristics and physiological conditions, thereby improving the data value of the stroke risk level prediction results.

[0161] The present invention is not limited to the above optional implementations, and anyone can derive other various forms of products under the enlightenment of the present invention. The above specific implementations should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. A method for predicting stroke risk, characterized in that: The steps include: Capture stroke risk assessment knowledge big data and historical clinical big data; The stroke risk assessment knowledge big data includes a number of stroke risk assessment knowledge consisting of stroke professional terminology knowledge, stroke clinical manifestation knowledge, stroke risk judgment knowledge, and stroke prevention measure knowledge; The historical clinical big data includes several historical clinical data consisting of user basic information, health consultation information, historical clinical indicators and historical clinical parameters; Data preprocessing is performed on the stroke risk assessment knowledge big data and the historical clinical big data to obtain the preprocessed stroke risk assessment knowledge big data and the preprocessed historical clinical big data, including the following steps: The stroke risk assessment knowledge big data is sequentially subjected to error data elimination and duplicate data deletion to obtain the stroke risk assessment knowledge big data after preprocessing; The historical clinical big data is processed in sequence by eliminating erroneous data, deleting duplicate data, desensitizing privacy data, and normalizing to obtain standardized historical clinical big data; The standardized historical clinical big data includes several standardized historical clinical data; Performing data clustering and data balancing on the standardized historical clinical big data to obtain pre-processed historical clinical big data includes the following steps: The membership of the FCM clustering algorithm is initialized using the Circle chaotic map to obtain the Circle chaotic map membership; According to the membership of Circle Chaos Map, several cluster centers and corresponding data categories of FCM clustering algorithm are generated; According to the Euclidean distance between the standardized historical clinical big data and the cluster center, the standardized historical clinical data within the preset distance range are divided into corresponding data categories; If the merge function value of the cluster center or the change value of the merge function value is less than the preset threshold, the same amount of standardized historical clinical data in each data category is integrated to obtain the preprocessed historical clinical big data; Otherwise, the cluster center and the corresponding data category are updated to obtain updated cluster centers and the corresponding updated data categories, and the data clustering process and the data balancing process are repeated according to the updated cluster centers and the corresponding updated data categories; According to the pre-processed stroke risk assessment knowledge big data, a stroke risk knowledge graph is constructed, including the following steps: Use the BERT-BILSTM-GAT-CRF algorithm to build a stroke named entity extraction model, and use BiGRU-Attention to build a stroke entity relationship extraction model; A stroke named entity extraction model is used to extract several stroke named entities from the preprocessed stroke risk assessment knowledge big data, and a stroke entity relationship extraction model is used to extract several stroke entity relationships from the preprocessed stroke risk assessment knowledge big data; Construct a stroke risk knowledge graph based on several stroke named entities and corresponding stroke entity relationships; Using the stroke risk knowledge graph, the pre-processed historical clinical big data was reprocessed to obtain a training sample set of historical stroke clinical key data; Based on the historical stroke clinical key data training sample set, a stroke risk prediction model was constructed using a deep learning algorithm, including the following steps: The initial values ​​of the network parameters of the DBN network are used as the optimization target of the IGWO optimization algorithm; According to the optimization goal, the IGWO optimization algorithm is used to iteratively optimize and obtain the optimal initial values ​​of network parameters; According to the optimal initial values ​​of network parameters, the network structure of DBN network is set to obtain the initial stroke risk prediction model; Input the historical stroke clinical key data training sample set into the initial stroke risk prediction model for optimization training to obtain the optimal stroke risk prediction model; Collect users' real-time clinical data, perform data dimension reduction on the real-time clinical data, and obtain real-time key clinical data of stroke; Input the real-time key clinical data of stroke into the stroke risk prediction model to predict the stroke risk and obtain the predicted result of the user's stroke risk level; The user's stroke risk level prediction results and corresponding real-time clinical data are input into the stroke risk knowledge graph to obtain corresponding stroke prevention measures.

2. A method for predicting stroke risk according to claim 1, characterized in that: The basic information of the user includes age, gender, ethnicity and permanent place of residence; The health information includes whether the patient smokes, drinks, has diabetes, has experienced a stroke, and has a family history of stroke; The historical clinical indicators include brain oxygen data, blood pressure data, pulse wave data, skin electricity data, sleep data, systolic blood pressure data after treatment, systolic blood pressure data without treatment, total cholesterol data, high-density lipoprotein cholesterol data, rectal temperature data, mean arterial pressure data, ventricular rate data, respiratory rate data, arterial oxygen partial pressure data, alveolar arterial oxygen saturation data, arterial blood pH data, serum sodium data, serum potassium data, blood creatinine data, hematocrit data, white blood cell count data, eye opening reaction data, speech reaction data and movement reaction data.

3. The method for predicting stroke risk according to claim 1, characterized in that: The chaotic mapping initialization mechanism, dynamic reverse learning mechanism and Gaussian mutation mechanism are introduced to improve the GWO optimization algorithm and obtain the IGWO optimization algorithm. According to the optimization goal, the IGWO optimization algorithm is used to iteratively optimize and obtain the optimal initial value of the network parameters, including the following steps: Set the IGWO population parameters, maximum number of iterations and fitness function of the IGWO optimization algorithm, and take the optimization target as the position of the IGWO individual in the IGWO population; According to the IGWO population parameters, the Tent-Logistic-Cosine chaotic mapping sequence is used to initialize the IGWO population to obtain the initialized IGWO population; According to the fitness function, calculate the fitness value of the IGWO individuals in the initialized IGWO population; Perform encirclement, hunting and attack operations on the initialized IGWO population to obtain an updated IGWO population; The IGWO individuals with the best fitness value in the updated IGWO population are subjected to dynamic reverse learning to obtain dynamic reverse IGWO individuals; Perform Gaussian mutation on several IGWO individuals with poor fitness values ​​in the updated IGWO population to obtain several Gaussian mutated IGWO individuals; If the current number of iterations is less than the maximum number of iterations, the IGWO population is updated based on the IGWO individuals with the best fitness value, the dynamically reversed IGWO individuals, and several Gaussian mutated IGWO individuals; Otherwise, the position of the IGWO individual with the best response value is output as the optimal initial value of the network parameter.

4. The method for predicting stroke risk according to claim 1, characterized in that: Collect users' real-time clinical data, perform data dimension reduction on the real-time clinical data, and obtain real-time key clinical data of stroke, including the following steps: Collect users' real-time clinical data and convert the real-time clinical data into a real-time clinical data matrix; Obtaining a real-time clinical data core matrix corresponding to the real-time clinical data matrix; Obtaining real-time clinical data eigenvalues ​​and real-time clinical data eigenvectors corresponding to the real-time clinical data kernel matrix; Select the real-time clinical data eigenvalues ​​and real-time clinical data eigenvectors with the top 95% of cumulative variance contribution rates, perform dimensionality reduction processing on the real-time clinical data matrix, and obtain the real-time clinical data dimensionality reduction matrix; According to the dimension reduction matrix of real-time clinical data, real-time key clinical data of stroke are obtained.

5. A stroke risk prediction system, used to implement the stroke risk prediction method according to any one of claims 1 to 4, characterized in that: The system comprises a big data capture module, a data preprocessing module, a knowledge graph construction module, a data secondary processing module, a stroke risk prediction model construction module, a data dimension reduction module, a stroke risk level prediction module and a stroke prevention measure acquisition module which are connected in sequence; Big data capture module, used to capture stroke risk assessment knowledge big data and historical clinical big data; A data preprocessing module is used to preprocess the stroke risk assessment knowledge big data and the historical clinical big data to obtain the preprocessed stroke risk assessment knowledge big data and the preprocessed historical clinical big data; A knowledge graph construction module is used to construct a stroke risk knowledge graph based on the pre-processed stroke risk assessment knowledge big data; The data secondary processing module is used to use the stroke risk knowledge graph to perform secondary data processing on the pre-processed historical clinical big data to obtain a historical stroke clinical key data training sample set The stroke risk prediction model building module is used to build a stroke risk prediction model based on the historical stroke clinical key data training sample set and using deep learning algorithms; The data dimension reduction module is used to collect users' real-time clinical data, perform data dimension reduction on the real-time clinical data, and obtain real-time key clinical data of stroke; The stroke risk level prediction module is used to input real-time key clinical data of stroke into the stroke risk prediction model to predict the stroke risk and obtain the user's stroke risk level prediction result; The stroke prevention measures acquisition module is used to input the user's stroke risk level prediction results and corresponding real-time clinical data into the stroke risk knowledge graph to obtain corresponding stroke prevention measures.

Citation Information

Patent Citations

  • Cerebral apoplexy risk prediction method and device based on hybrid deep transfer learning

    CN111968746A

  • Knowledge graph construction method for cerebral apoplexy

    CN112541086A

  • Improved grey wolf optimization DBN-ELM sub-health state recognition method

    CN113190931A

  • Cerebral stroke risk prediction method and system based on transfer learning

    CN114512239A