Cerebral stroke prediction method, system and equipment

Abstract

The invention relates to the technical field of cerebral apoplexy prediction, in particular to a cerebral apoplexy prediction method, a cerebral apoplexy prediction system and cerebral apoplexy prediction equipment. According to the cerebral apoplexy prediction method provided by the invention, the facial information of the patient is identified, the grip strength difference information and the historical medical record information of the patient are also acquired, and the grip strength difference information and the historical medical record information of the patient are interactively processed and then are combined with the facial feature information to serve as the input of the prediction model to predict the cerebral apoplexy, so that the accuracy of the cerebral apoplexy prediction is improved. In this way, the face information and the grip strength difference information of the patient are combined, meanwhile, the historical record data of the user are compared, the risk prediction result can reflect the change of the illness state of the patient, and the cerebral apoplexy prediction success rate is greatly increased.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of stroke prediction, and specifically to a stroke prediction method, system and device. Background Art

[0002] Stroke is a serious neurological disease with high morbidity and disability, which has brought heavy burden to patients and society. In recent years, some progress has been made in the research of stroke prediction models. Existing research mainly focuses on the analysis of biomarkers, clinical characteristics and imaging data, while visual features are relatively less valued.

[0003] Visual features may have important indicative significance in the clinical presentation of stroke patients. Certain visual symptoms may be associated with specific types of stroke, especially in patients with migraine. For example, brief visual aura is considered a characteristic of young and middle-aged stroke patients.

[0004] The inventors of the present application found in their research that the existing system for predicting stroke through visual features relies on single facial recognition data without reference to other features of the human body, which results in poor prediction accuracy of the system. Summary of the invention

[0005] In view of the above problems, the embodiments of the present application provide a stroke prediction method, system, device and computer-readable storage medium to solve the above technical problems existing in the prior art.

[0006] One aspect of the embodiments of the present application provides a method for predicting stroke, comprising: Obtain facial recognition information, grip strength difference information, and historical medical records; Generate facial feature information using a pre-trained visual feature prediction model based on the facial recognition information; Standardizing the facial feature information, grip strength difference information, and historical medical record information; Interactively processing the standardized grip strength difference information and the standardized historical medical record information to generate interactive feature information; Constructing a multivariate regression model based on the facial feature information and the interactive feature information; Estimate the multivariate regression model according to the least squares method to generate estimated model parameters; Adjusting the estimation model parameters according to a preset facial feature weight factor and a preset interactive feature weight factor, wherein the preset facial feature weight factor is greater than the preset interactive feature weight factor; generating a stroke risk assessment value through the multivariate regression model according to the adjusted estimated model parameters; Stroke prediction is performed according to the stroke risk assessment value.

[0007] Preferably, in some embodiments, generating facial feature information according to the facial recognition information through a pre-trained visual feature prediction model includes: Extracting the facial recognition information to obtain left and right eye gaze images, a facial symmetry image, and a smile symmetry image; normalizing the left and right eye gaze images, the facial symmetry image, and the smile symmetry image; Performing convolution processing on the normalized image to generate a feature map, and performing a pooling operation on the feature map; Processing the feature map after convolution and pooling operations through an activation function to generate a probability value corresponding to the feature map; Facial feature information is generated according to the probability value.

[0008] Preferably, in some embodiments, the grip strength difference information includes a grip strength difference change rate.

[0009] Preferably, in some embodiments, the standardized grip strength difference information and the standardized historical medical record information are interactively processed to generate interactive feature information, including: generating a grip force difference matrix according to the standardized grip force difference information; Generate a historical medical record matrix according to the standardized historical medical record information; Multiplying the grip strength difference matrix with the historical medical record matrix to generate interactive feature information; The constructing a multivariate regression model according to the facial feature information and the interactive feature information comprises: The multiple regression model is: ; Among them, Y is the stroke risk assessment value, is the facial feature information, … is the interactive feature information, is the error term, is the intercept term, are the initial model parameters.

[0010] Preferably, in some embodiments, the estimating the multivariate regression model according to the least squares method to generate estimated model parameters includes: The estimated model parameters are determined by minimizing the sum of squared errors using the following formula: ; Among them, m is the number of samples.

[0011] Preferably, in some embodiments, the adjusting the estimation model parameters according to the preset weight factors includes: Constructing a weighted loss function according to the facial feature weight factor and the interactive feature weight factor; Determining a weighted error corresponding to the facial feature information and the interactive feature information according to the weighted loss function; The weighted loss function is optimized by a gradient descent algorithm according to the weighted error, and the estimation model parameters are adjusted.

[0012] Another aspect of the embodiment of the present application further provides a stroke management system, including a facial recognition unit, a grip strength acquisition unit, a questionnaire unit and a data prediction unit; The facial recognition unit is used to obtain facial recognition information and send the facial recognition information to the data prediction unit; The grip force acquisition unit is used to acquire grip force difference information and send the grip force difference information to the data prediction unit; The questionnaire unit is used to obtain historical medical record information and send the historical medical record information to the data prediction unit; The data prediction unit is used to execute the stroke prediction method described in the above embodiment according to the facial recognition information, the grip strength difference information and the historical medical record information.

[0013] Preferably, in some embodiments, an automatic registration unit is also included; The automatic registration unit is used to automatically generate registration information according to the stroke prediction result, wherein the registration information includes the stroke prediction result.

[0014] According to a third aspect of the embodiment of the present application, a stroke prediction device is provided, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least a program, and the program enables the processor to execute the operations of the stroke prediction method described in the above embodiment.

[0015] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, wherein at least one program is stored in the storage medium. When the program is executed on a stroke prediction device, the stroke prediction device performs the operations of the stroke prediction method described in the above embodiments.

[0016] In summary, in addition to identifying the patient's facial information, the embodiments of the present application also obtain the patient's grip strength difference information and historical medical history information, and after interactive processing, the grip strength difference information and the patient's historical medical history information are combined with the facial feature information and used as the input of the prediction model to predict stroke. In this way, the patient's facial information and grip strength difference information are combined, and the user's historical record data is also compared. The predicted risk results can reflect changes in the patient's condition, greatly improving the success rate of stroke prediction.

[0017] The above description is only an overview of the technical solution of the embodiment of the present application. In order to more clearly understand the technical means of the embodiment of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiment of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are only used to illustrate the embodiments and are not to be considered as limiting the present application. In addition, the same reference symbols are used to represent the same components throughout the accompanying drawings. In the accompanying drawings: Figure 1 A schematic diagram of the structure of a stroke management system provided in an embodiment of the present application; Figure 2 A schematic diagram of a stroke prediction method provided in an embodiment of the present application; Figure 3 A schematic diagram of a loss function optimization process provided in an embodiment of the present application; Figure 4 A schematic diagram of the structure of a stroke prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein.

[0020] With the development of deep learning technology, it is possible to use visual data to improve the accuracy of stroke prediction. The application of facial features in stroke prediction is an emerging research field that uses artificial intelligence and machine learning techniques to analyze facial images to identify and predict the risk of stroke in patients.

[0021] Clinically, before a stroke occurs, the body will experience many symptoms, such as numbness in the legs, weakness in the hands and feet, etc. At the same time, the patient's own diseases will also increase the risk of causing a stroke. The inventors of this application found in their research that the current stroke prediction system relies on single facial recognition data and does not refer to other medical history features of the patient, resulting in poor system prediction accuracy. The existing prediction system does not adequately process feature engineering and fails to identify and extract high-quality features that are helpful for prediction, which can cause misdiagnosis and lead to missing the best time to treat a stroke.

[0022] In view of this, the embodiment of the present application proposes a method and system for predicting stroke. Through the method and system for predicting stroke proposed in the embodiment of the present application, in addition to identifying the patient's facial information, the patient's grip difference information and historical medical record information are also obtained, and the grip difference information and the patient's historical medical record information are interactively processed to generate interactive feature information, and the interactive feature information is combined with the facial feature information as the input of the prediction model to predict the stroke. In this way, the patient's facial information and grip difference information are combined, and the user's historical medical record data is also referred to. The predicted risk results obtained in this way can reflect the changes in the patient's condition and greatly improve the success rate of stroke prediction. At the same time, the stroke prediction system provided by the embodiment of the present application is also connected to the hospital's medical treatment process, automatically provides registration services for users according to the prediction results, and includes the stroke prediction situation in the registration information, so that doctors can quickly and accurately diagnose the patient's condition.

[0023] like Figure 1 , which is a schematic diagram of the structure of a stroke prediction system proposed in an embodiment of the present application, and a stroke prediction method proposed in an embodiment of the present application can be applied to the stroke prediction system.

[0024] The stroke prediction system 100 includes a face recognition unit 101 , a grip strength acquisition unit 102 , a questionnaire unit 103 and a data prediction unit 104 .

[0025] The facial recognition unit 101 is usually a camera, which is used to obtain the patient's facial recognition information. When the patient visits the doctor, the stroke prediction system can remind the patient to stand in a specific position and collect information about the patient's face through the camera, including taking static pictures and taking dynamic videos. The facial recognition unit 101 can also identify facial features that need to be acquired, such as eye features, mouth features, and smile features. When the above key information is not acquired, the facial recognition unit 102 can remind the user to make corresponding actions to acquire it again. After the facial recognition unit 102 acquires the above facial recognition information, it sends the facial recognition information to the data prediction unit.

[0026] The grip strength acquisition unit 102 is used to obtain the grip strength difference information of the patient's left and right hands. According to clinical experience, when a patient suffers from a stroke, his left and right hands will have muscle spasms or atrophy to a certain extent, resulting in a large difference in the grip strength of his left and right hands. Obtaining the difference in the grip strength of the patient's left and right hands is of great significance for determining the risk of stroke. In an embodiment of the present application, in order to obtain the grip strength difference of the patient's left and right hands, the grip strength acquisition unit 102 can provide a handgrip dynamometer, which is connected to the data prediction unit 104 through an interface. When the patient clenched the handgrip dynamometer, the data prediction unit 104 obtained the grip strength information of the left and right hands, and took the difference to obtain the difference between the two.

[0027] Furthermore, in order to more accurately obtain the impact of the left and right hand grip strength difference on stroke, multiple left and right hand grip strength difference information can be periodically obtained, and the change rate of the left and right hand grip strength difference can be determined based on the multiple left and right hand grip strength difference information. In practice, some patients do have a large difference in grip strength between their left and right hands, and the grip strength difference alone may not be able to truly reflect the patient's stroke risk. Clinical observations have found that the persistence of hand grip strength in early stroke patients will be affected. Therefore, by obtaining the change rate of the grip strength difference, the risk of stroke can be more accurately reflected.

[0028] The questionnaire unit 103 is used to obtain the patient's historical medical information. The stroke management system can investigate the patient's historical medical information through voice inquiry or interface display. The historical medical information includes whether the patient suffers from hypertension, diabetes, hyperlipidemia, etc., whether there is a family history of cardiovascular and cerebrovascular diseases, living habits, etc.

[0029] The data prediction unit 104 is used to predict the probability of stroke by executing the stroke prediction method proposed in the embodiment of the present application according to the facial recognition information obtained by the facial recognition unit 101, the grip strength difference information obtained by the grip strength acquisition unit 102, and the historical medical record information obtained by the questionnaire unit 103. At the same time, the data prediction unit 104 can also present the stroke prediction result to the patient through the display screen.

[0030] The data prediction unit 104 generates facial feature information through a pre-trained visual feature prediction model based on the facial recognition information; standardizes the facial feature information, grip strength difference information and historical medical record information; interactively processes the standardized grip strength difference information and the standardized historical medical record information to generate interactive feature information; constructs a multivariate regression model based on the facial feature information and the interactive feature information; estimates the multivariate regression model based on the least squares method to generate estimated model parameters; adjusts the estimated model parameters based on preset facial feature weight factors and preset interactive feature weight factors, wherein the preset facial feature weight factor is greater than the preset interactive feature weight factor; generates a stroke risk assessment value through the multivariate regression model based on the adjusted estimated model parameters; and predicts a stroke based on the stroke risk assessment value.

[0031] Furthermore, the stroke management system also includes an automatic registration unit 105, which is used to automatically generate registration information according to the stroke prediction result, wherein the registration information includes the stroke prediction result. The automatic registration unit 105 automatically connects to the hospital treatment platform and can provide patients with a one-stop automatic registration service, so that stroke patients can receive rapid treatment. By including the stroke prediction result in the registration information, the doctor can directly obtain the patient's relevant information after seeing the registration information, which helps the doctor to provide accurate and timely treatment to the patient.

[0032] The embodiment of the present application combines facial recognition information, the difference in grip strength between the left and right hands, and historical medical records to achieve a multi-dimensional assessment of stroke risk. This comprehensive method can more comprehensively reflect the individual's physiological state, thereby improving the accuracy and reliability of prediction. Compared with the traditional single-indicator prediction method, this multi-feature fusion method will provide more accurate decision support for clinicians. At the same time, through the stroke management system, one-click appointment registration is achieved, saving medical and social resource costs while building a solid barrier for possible stroke risks.

[0033] In the above embodiment, the data prediction unit 104 is used to execute the stroke prediction method proposed in the embodiment of the present application, specifically, Figure 2 As shown, the stroke prediction method includes: Step S10: Obtaining facial recognition information, grip strength difference information and historical medical record information; As described in the above-mentioned embodiment of the stroke management system, the facial recognition information can be obtained through the facial recognition unit of the stroke management system. The facial recognition information is obtained by taking photos or videos of the patient with a camera. The facial recognition information includes eye features, mouth features, and smile features. The eye features can be used to determine whether the patient has symptoms of strabismus, the mouth features can be used to determine whether the patient has symptoms of crooked mouth, the smile features can be used to determine whether the patient has symptoms of facial paralysis, etc. The above-mentioned features are typical symptoms of stroke patients. Obtaining facial recognition information by photographing the patient's face helps to determine whether the patient is at risk of stroke.

[0034] The grip strength difference information can be obtained by the grip strength acquisition unit in the above embodiment. In the embodiment of the present application, in addition to obtaining the patient's facial information, the grip strength difference information of the patient's left and right hands is further obtained. Combining the grip strength difference information with the facial recognition information can more accurately determine whether the patient has a stroke risk. In addition to including the difference between the grip strength of the left and right hands, the grip strength difference information can also include the rate of change of the grip strength difference of the left and right hands. The rate of change of the grip strength difference represents the change of the patient's muscles and can more accurately reflect whether the patient has a stroke risk.

[0035] In an embodiment of the present application, in addition to the patient's facial information and grip strength difference information, the patient's historical medical record information is further obtained. The historical medical record information is mainly used to understand the patient's basic information and past medical history, including basic information such as age, gender, height, weight, and whether the patient has hypertension, diabetes, hyperlipidemia, cardiovascular and cerebrovascular family history, etc. By obtaining historical medical record information, a more comprehensive understanding of the patient's condition can be obtained in order to more accurately judge whether the patient is at risk of stroke.

[0036] Step S20: Generate facial feature information through a pre-trained visual feature prediction model according to the facial recognition information; In the stroke management system proposed in the above embodiment, a trained visual feature prediction model is pre-stored. The visual feature prediction model realizes the extraction and recognition of facial recognition information through a deep learning algorithm. For example, the acquired facial picture or video can be processed through the openCV function, including recognizing the left and right eye gaze, facial symmetry, smile symmetry, mouth symmetry and other photo or video formats into picture formats, and then normalizing the left and right eye gaze images, facial symmetry images, smile symmetry images and mouth symmetry images, for example: rescaling the pixel value of the input data from the range of [0, 255] to [0, 1], etc.

[0037] The normalized image is convoluted to generate a feature map, and the feature map is pooled. In an embodiment of the present application, the deep learning algorithm for processing facial recognition information includes a seven-layer CNN network with two convolutional layers (activation function is rule), two pooling layers, one Flatten layer, and two fully connected layers (activation functions are rule and sigmoid, respectively).

[0038] Among them, the convolutional layer is responsible for extracting the features of the input data. By defining a set of convolutional kernels (filters), the input data is convolved to obtain a series of feature maps. These feature maps can capture local features and spatial structures.

[0039] The pooling layer is used for dimensionality reduction and sampling, reducing model parameters and computational complexity. Common pooling operations include maximum pooling and average pooling, which retain important information and reduce the amount of data by taking the maximum value or average value in a specific area.

[0040] The Flatten layer is used to convert a multidimensional array, for example, a two-dimensional image matrix or a three-dimensional convolutional layer output, into a one-dimensional array so that subsequent layers can process it.

[0041] The fully connected layer is used to combine and classify the features extracted by the convolutional layer and the pooling layer. Each neuron in the fully connected layer is connected to all neurons in the previous layer to achieve high-level feature learning and classification of the input data.

[0042] The feature map after convolution and pooling is processed by an activation function to generate a probability value corresponding to the feature map, wherein the activation function is linear in the positive interval and is always zero in the negative interval, which helps to solve the gradient vanishing problem and speed up the training process.

[0043] Through the above process, facial feature information is generated according to the probability value. Assuming that the AUC of the visual feature model is 92.538, the prediction results greater than 0.92 can be converted to 1, and the rest of the results can be converted to 0.

[0044] In this step, the acquired facial recognition information is firstly subjected to deep learning and facial feature information extraction through a pre-trained visual feature prediction model, so as to be combined with the acquired grip strength difference information and historical medical record information to predict the outcome of stroke.

[0045] Step S30: standardizing the facial feature information, grip strength difference information and historical medical record information; Combined with the questionnaire data, the historical medical record information obtained is standardized. For example, if the patient has a family history of hypertension, diabetes, hyperlipidemia, or cardiovascular and cerebrovascular diseases in the questionnaire, it is converted to 1, and if not, it is converted to 0. For the collected grip strength difference information, it can be judged according to the pre-set threshold. For example, if the pre-set threshold is 15%, then if the grip strength difference > 15%, it is converted to 1, and if the grip strength difference is less than or equal to the rest, it is converted to 0.

[0046] Step S40: interactively processing the standardized grip strength difference information and the standardized historical medical record information to generate interactive feature information; Creating interactive features is an effective way to improve the prediction ability of the model, especially in the multivariate linear regression model. Interactive features can help the model capture the complex relationship between different features. In the embodiment of the present application, although the grip strength difference information can reflect the risk of stroke in patients to a certain extent, in real life, there are indeed some people whose grip strength of the left and right hands is quite different. If these people are identified as high-risk stroke patients, it will obviously lead to wrong results. In order to solve this problem, the embodiment of the present application interactively processes the grip strength difference information and the historical medical record information, and interactively processes these two pieces of information so that they can more accurately reflect the relationship between the features.

[0047] A grip force difference matrix G is generated based on the standardized grip force difference information, wherein the grip force difference matrix includes multiple grip force difference information; a historical medical record matrix is ​​generated based on the standardized historical medical record information, wherein the historical medical record matrix may include age (A), gender (S), blood pressure (B), diabetes (D), etc., and an interactive feature matrix is ​​generated by multiplying the grip force difference matrix with the historical medical record matrix. For example, the interactive feature matrices G*A*B, G*A*D, G*S*B, G*S*D, etc. are obtained by multiplying the above multiple features.

[0048] The interactive feature matrix combines the grip strength difference information and historical medical record information, and can capture the complex relationships between different features, such as the relationship between grip strength difference and age, the relationship between age and blood pressure, the relationship between age and diabetes, and so on.

[0049] Step S50: constructing a multivariate regression model according to the facial feature information and the interactive feature information; A multiple regression model is an analytical method in statistics that is used to study the relationship between a dependent variable (often called a response variable or outcome variable) and multiple independent variables (also called explanatory variables or predictor variables). In multiple regression analysis, a linear equation is established that predicts the value of the dependent variable using the values ​​of multiple independent variables.

[0050] In the embodiment of the present application, we use the stroke risk assessment value as the dependent variable, the facial feature information and the interactive feature information as the dependent variable, and the basic form of the multiple regression model is: ; (Formula 1) Among them, Y is the stroke risk assessment value, is the facial feature information, … is the interactive feature information; is the error term, which represents the random variation that cannot be explained by the model; is the intercept term, is the initial model parameter, which represents the expected change in the dependent variable when the independent variable changes by one unit.

[0051] Step S60: Estimate the multivariate regression model according to the least squares method to generate estimated model parameters; In the multivariate regression model, the least squares method is used for parameter estimation in the embodiment of the present application. This method finds the best coefficient estimate by minimizing the residual sum of squares. The residual refers to the difference between the actual observed value and the model predicted value.

[0052] Estimated model parameters are determined by minimizing the sum of squared errors: ; Among them, m is the number of samples. When the sum of squared errors is minimized, the estimated model parameters can be determined. The value of .

[0053] Step S70: adjusting the estimation model parameters according to a preset facial feature weight factor and a preset interaction feature weight factor, wherein the preset facial feature weight factor is greater than the preset interaction feature weight factor; In practical applications, the patient's facial features can usually better reflect the risk of stroke. Interactive feature information such as grip strength difference and historical medical history information can be combined with facial features to improve and supplement them. In clinical significance, facial features are more important than grip strength difference and historical medical history information.

[0054] Therefore, in order to more accurately reflect the relationship between facial feature information and interactive feature information, in the embodiment of the present application, the weights of each model parameter in the least squares method are adjusted to have different weight values. Feature weighting is a technique to increase the influence of specific features in the model. In the context of the least squares method, this can be achieved by modifying the loss function so that facial features are given higher weights during the optimization process. Figure 3 As shown, specifically including: Step S71: constructing a weighted loss function according to the facial feature weight factor and the interactive feature weight factor; When constructing the weighted loss function, we first define different weight factors for facial feature information and interactive feature information. Facial feature information corresponds to facial feature weight factors , the interactive feature information corresponds to the interactive feature weight factor , since the weight of facial feature information is greater than the weight of interactive feature information, .

[0055] In the loss function, different weight factors are applied to the error terms of facial feature information and interactive feature information. The weighted loss function can be expressed as: in, is the facial feature weight factor, is the interaction feature weight factor; is the actual value of the i-th sample;

[0056]

[0057]

[0058]

[0059] Step S72: determining a weighted error corresponding to the facial feature information and the interaction feature according to the weighted loss function; For each sample, the weighted error of facial feature information and interaction feature information is calculated: ; ; in, is the weighted error of facial feature information, is the weighted error of the interactive feature information.

[0060] Multiply each weighted error by the corresponding weight factor and calculate the squared weighted error, which are: and .

[0061] Sum the squared weighted errors of all samples to get the total weighted loss: .

[0062] Step S73: Optimizing the weighted loss function through a gradient descent algorithm according to the weighted error, and adjusting the estimation model parameters.

[0063] The gradient descent algorithm is a commonly used numerical optimization method. In the embodiment of the present application, the weighted loss function is optimized by the gradient descent algorithm to adjust the estimated model parameters.

[0064] In each iteration, the model parameters are updated according to the gradient of the loss function. In each step of gradient descent, the facial features and interaction features are updated while taking into account the weight factors: ; in, is the learning rate, are the estimated model parameters to be updated.

[0065] Through the above steps, the embodiment of the present application ensures that facial feature information has a higher weight, thereby improving the accuracy of the multivariate regression model in predicting stroke risk.

[0066] Step S80: generating a stroke risk assessment value through the multivariate regression model according to the adjusted estimated model parameters; The final stroke risk assessment value was generated by the multivariate regression model of formula 1 using the adjusted estimated model parameters.

[0067] Step S90: performing stroke prediction according to the stroke risk assessment value.

[0068] In summary, the stroke prediction method proposed in the embodiment of the present application, in addition to identifying the patient's facial information, also obtains the patient's grip strength difference information and historical medical history information, and interactively processes the grip strength difference information and the patient's historical medical history information, and combines them with facial feature information as the input of the prediction model to predict the stroke. In this way, the patient's facial information and grip strength difference information are combined, and the user's historical record data is also compared. The predicted risk results can reflect changes in the patient's condition, greatly improving the success rate of stroke prediction.

[0069] According to another aspect of the embodiments of the present application, a stroke prediction device is provided. Figure 4 , Figure 4 A schematic diagram of the structure of a stroke prediction device according to an embodiment of the present application is shown. The specific embodiment of the present application does not limit the specific implementation of the stroke prediction device.

[0070] like Figure 4 As shown, the stroke prediction device may include: a processor 402 , a memory 406 , a communication interface 404 and a communication bus 408 .

[0071] The processor 402, the memory 406 and the communication interface 404 communicate with each other via a communication bus 408. The memory 406 is used to store at least one program 410, and the program 410 enables the processor 402 to execute the relevant steps in the above-mentioned stroke prediction method embodiment.

[0072] Specifically, the program 410 may include program code including computer executable instructions.

[0073] The processor 402 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the stroke prediction device may be processors of the same type, such as one or more CPUs; or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0074] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0075] The program 410 can be specifically called by the processor 402 to enable the stroke prediction device to execute any operation method provided in the above-mentioned stroke prediction method embodiment, and the details are not repeated here.

[0076] An embodiment of the present application further provides a computer-readable storage medium, in which executable instructions are stored. When the executable instructions are executed on a stroke prediction device, the stroke prediction device executes the stroke prediction method provided in any of the above embodiments.

[0077] The embodiment of the present application also provides a stroke prediction program, which is used to execute the stroke prediction method provided in the above embodiment.

[0078] The algorithm or display provided here are not inherently related to any specific computer, virtual system or other equipment. Various general systems can also be used together with the teaching based on this. According to the above description, it is obvious to construct the structure required for this type of system. In addition, the present application embodiment is not directed to any specific programming language yet. It should be understood that various programming languages ​​can be utilized to realize the content of the present application described here, and the above description of specific languages ​​is to disclose the best mode of implementation of the present application.

[0079] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0080] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various aspects of the invention, in the above description of exemplary embodiments of the present application, the various features of the embodiments of the present application are sometimes grouped together into a single embodiment, figure, or description thereof.

[0081] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying abstract and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying abstract and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0082] It should be noted that the above embodiments illustrate the present application rather than limit the present application, and those skilled in the art may design alternative embodiments without departing from the scope. The steps in the above embodiments, unless otherwise specified, should not be understood as limiting the execution order.

Claims

1. A method for predicting stroke, characterized in that: include: Obtain facial recognition information, grip strength difference information, and historical medical records; Generate facial feature information using a pre-trained visual feature prediction model based on the facial recognition information; Standardizing the facial feature information, grip strength difference information, and historical medical record information; Interactively processing the standardized grip strength difference information and the standardized historical medical record information to generate interactive feature information; Constructing a multivariate regression model based on the facial feature information and the interactive feature information; Estimate the multivariate regression model according to the least squares method to generate estimated model parameters; Adjusting the estimation model parameters according to a preset facial feature weight factor and a preset interactive feature weight factor, wherein the preset facial feature weight factor is greater than the preset interactive feature weight factor; generating a stroke risk assessment value through the multivariate regression model according to the adjusted estimated model parameters; Stroke prediction is performed according to the stroke risk assessment value.

2. The stroke prediction method according to claim 1, characterized in that: The generating facial feature information according to the facial recognition information by using a pre-trained visual feature prediction model comprises: Extracting the facial recognition information to obtain left and right eye gaze images, a facial symmetry image, and a smile symmetry image; normalizing the left and right eye gaze images, the facial symmetry image, and the smile symmetry image; Performing convolution processing on the normalized image to generate a feature map, and performing a pooling operation on the feature map; Processing the feature map after convolution and pooling operations through an activation function to generate a probability value corresponding to the feature map; Facial feature information is generated according to the probability value.

3. The stroke prediction method according to claim 1, characterized in that: The grip strength difference information includes a grip strength difference change rate.

4. The method for predicting stroke according to any one of claims 1 to 3, characterized in that: The step of interactively processing the standardized grip strength difference information and the standardized historical medical record information to generate interactive feature information includes: generating a grip force difference matrix according to the standardized grip force difference information; Generate a historical medical record matrix according to the standardized historical medical record information; Multiplying the grip strength difference matrix with the historical medical record matrix to generate interactive feature information; The constructing a multivariate regression model according to the facial feature information and the interactive feature information comprises: The multiple regression model is: ； Among them, Y is the stroke risk assessment value, is the facial feature information, … is the interactive feature information, is the error term, is the intercept term, are the initial model parameters.

5. The stroke prediction method according to claim 4, characterized in that: The step of estimating the multivariate regression model according to the least squares method to generate estimated model parameters includes: The estimated model parameters are determined by minimizing the sum of squared errors using the following formula: ； Among them, m is the number of samples.

6. The method for predicting stroke according to claim 5, characterized in that: The adjusting the estimation model parameters according to the preset weight factors includes: Constructing a weighted loss function according to the facial feature weight factor and the interactive feature weight factor; Determining a weighted error corresponding to the facial feature information and the interactive feature information according to the weighted loss function; The weighted loss function is optimized by a gradient descent algorithm according to the weighted error, and the estimation model parameters are adjusted.

7. A stroke management system, characterized in that: It includes a facial recognition unit, a grip strength acquisition unit, a questionnaire survey unit and a data prediction unit; The facial recognition unit is used to obtain facial recognition information and send the facial recognition information to the data prediction unit; The grip force acquisition unit is used to acquire grip force difference information and send the grip force difference information to the data prediction unit; The questionnaire unit is used to obtain historical medical record information and send the historical medical record information to the data prediction unit; The data prediction unit is used to execute the stroke prediction method as described in any one of claims 1-6 based on the facial recognition information, the grip strength difference information and the historical medical record information.

8. The stroke management system according to claim 7, characterized in that: Also included is an automatic registration unit; The automatic registration unit is used to automatically generate registration information according to the stroke prediction result, wherein the registration information includes the stroke prediction result.

9. A stroke prediction device, characterized in that: include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least a program, and the program enables the processor to perform the operation of the stroke prediction method according to any one of claims 1-6.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one program, and when the program is executed on the stroke prediction device, the stroke prediction device performs the operation of the stroke prediction method according to any one of claims 1 to 6.

Images