Cerebrovascular disease dynamic prediction method and system based on artificial intelligence

Abstract

The invention discloses a cerebrovascular disease dynamic prediction method and system based on artificial intelligence, and belongs to the field of cerebrovascular disease prediction.The method comprises the steps that feature extraction is conducted on historical cerebrovascular disease related data to obtain cerebrovascular disease related features; performing feature selection based on the cerebrovascular disease related features to obtain effective related features; constructing a risk factor model according to a plurality of preset features and the weight value of the cerebrovascular disease, substituting a plurality of effective related features into the risk factor model for fitting, and configuring an AMS Grad optimization algorithm according to a fitting result; training the preset cerebrovascular disease prediction model through an AMS Grad optimization algorithm to obtain a cerebrovascular disease prediction model; and predicting historical cerebrovascular disease related data of the patient through the cerebrovascular disease prediction model, and adjusting the cerebrovascular disease prediction model according to a prediction result. According to the scheme, the cerebrovascular disease prediction accuracy can be improved by constructing the cerebrovascular disease prediction model through feature extraction and feature selection in combination with the optimization algorithm.

Description

Technical Field

[0001] The present invention belongs to the technical field of cerebrovascular disease detection, and in particular relates to a cerebrovascular disease dynamic prediction method and system based on artificial intelligence. Background Art

[0002] With the aging of the population and the change of lifestyle, the incidence of cerebrovascular disease is on the rise worldwide and has become one of the major diseases that seriously endanger human health. Traditional cerebrovascular disease prediction methods often rely on the experience of doctors and static clinical information of patients, such as age, blood pressure, blood lipids, etc. These methods are limited in prediction accuracy and timeliness. In addition, the pathogenesis of cerebrovascular disease is complex and involves the interaction of multiple physiological systems and environmental factors. Therefore, it is particularly important to develop a method that can dynamically monitor and predict the risk of cerebrovascular disease.

[0003] At present, although some AI-based cerebrovascular disease prediction models have been proposed and applied, these models still have certain limitations in performance. On the one hand, due to the complex pathogenesis of cerebrovascular disease, existing prediction models often find it difficult to fully capture all relevant factors and their interactions, resulting in low accuracy of prediction results. On the other hand, traditional model training methods, such as the gradient descent method and its variants, are prone to fall into local optimal solutions during the optimization process, affecting the generalization ability and stability of the model, resulting in relatively low accuracy in cerebrovascular disease prediction. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of low accuracy of cerebrovascular disease prediction methods in the prior art and to provide a cerebrovascular disease dynamic prediction method and system based on artificial intelligence.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: According to a first aspect of the present invention, there is provided a method for dynamic prediction of cerebrovascular disease based on artificial intelligence, comprising the following steps: S1, extracting features of historical cerebrovascular disease related data in a historical medical record database to obtain cerebrovascular disease related features; S2. Based on the cerebrovascular disease related features, feature selection is performed by a feature selection algorithm to obtain effective related features; S3, constructing a risk factor model according to a plurality of preset features of cerebrovascular disease and a weight value of each preset feature, substituting the plurality of effective related features into the risk factor model for fitting, and configuring an AMSGrad optimization algorithm according to the fitting result; S4. Training the preset cerebrovascular disease prediction model by the AMS Grad optimization algorithm to obtain a trained cerebrovascular disease prediction model; S5. Predicting the patient's historical cerebrovascular disease-related data using the cerebrovascular disease prediction model, and adjusting the cerebrovascular disease prediction model according to the prediction result.

[0006] In the above technical solution, through feature extraction and feature selection, key features that are highly correlated with cerebrovascular disease can be identified, and irrelevant or noise data can be removed, thereby building a more accurate and concise prediction model, which significantly improves the prediction accuracy of the model. The AMS Grad optimization algorithm is used to train the model, which not only accelerates the convergence speed of the model, but also enhances the generalization ability of the model on unseen data. The model can not only accurately predict known types of data, but also better adapt to emerging data patterns, improving the practicality and reliability of the model. In the prediction process, the model can be dynamically adjusted according to the prediction results to form a closed-loop feedback mechanism, so that the model can continuously learn and optimize itself to adapt to the real-time changes in the patient's condition and provide more personalized prediction results. Through early prediction and timely intervention, accurate prediction of cerebrovascular disease can help reduce the number and severity of cerebrovascular disease attacks, thereby reducing the physical and psychological burden of patients and improving the quality of life of patients.

[0007] Preferably, S1 comprises the following steps: S11, calling up a historical medical record database and extracting historical cerebrovascular related data from the historical medical record database; S12, preprocessing the historical cerebrovascular-related data, wherein the preprocessing at least includes outlier processing, missing value processing and data standardization; S13, extracting features from the preprocessed historical cerebrovascular-related data to obtain corresponding time series features, text data features, and image features; S14, fusing the time series features, text data features and image features through a feature concatenation method to obtain final cerebrovascular-related features.

[0008] In the above technical solution, by including the preprocessing steps of processing outliers, missing values ​​and data standardization, not only can the prediction error caused by data quality problems be reduced, but also the efficiency and stability of subsequent model training can be improved. Multi-dimensional feature extraction is performed on the preprocessed data, including time series features, text data features and image features, which can more comprehensively capture various information related to cerebrovascular disease. The features of different dimensions are fused by the feature splicing method to obtain the final cerebrovascular-related features. The process of feature fusion helps to explore the potential relationship between different features, further enhance the expression and prediction capabilities of the features, and enable the prediction model to more comprehensively understand and predict the complex changes of cerebrovascular disease. With high-quality data and comprehensive feature extraction methods, the prediction model finally constructed will be able to more accurately capture the incidence patterns and trends of cerebrovascular disease, help identify potential high-risk patients in advance, and provide them with more timely and effective intervention measures. Multi-dimensional feature extraction and feature fusion also help to achieve more personalized predictions. By analyzing the specific characteristics of different patients, the model can give more accurate prediction results for different situations, thereby providing patients with more personalized treatment recommendations and health management plans.

[0009] Preferably, S2 comprises the following steps: S21, calculating the correlation characteristic coefficients between the prediction targets corresponding to the cerebrovascular related characteristics; S22, setting preset coefficients of relevant features by expert experience method, and eliminating cerebrovascular related features that are lower than the preset coefficients of relevant features; S23. Recursively eliminate the cerebrovascular related features in combination with the machine learning model to obtain effective related features.

[0010] In the above technical solution, by calculating the correlation feature coefficient between the cerebrovascular related features and the prediction target, the degree of influence of each feature on the prediction target can be quantified, which helps to identify the features that are highly correlated with the prediction target and provide a strong basis for subsequent feature selection. By setting the preset coefficient of the relevant features through the expert experience method and eliminating the cerebrovascular related features below the coefficient, the features that are not closely related or irrelevant to the prediction target can be effectively removed, the complexity and calculation amount of the model can be reduced, and the training efficiency and prediction performance of the model can be improved. Recursive feature elimination is performed in combination with the machine learning model to further screen out the most effective related features for the prediction target. Recursive feature elimination is an effective feature selection method. It continuously and iteratively builds the model and eliminates the least important features to finally obtain the optimal feature subset. It can make full use of the powerful learning ability of the machine learning model and further improve the accuracy and effectiveness of feature selection.

[0011] Preferably, S3 comprises the following steps: S31. defining a preset feature set, including a plurality of features highly correlated with cerebrovascular disease; S32, setting a weight value for each preset feature according to a statistical analysis method; S33, constructing a risk factor model by weighted summation method according to the preset characteristics and corresponding weight values; S34. Fit the risk factor model using the effective correlation features, and configure the AMS Grad optimization algorithm according to the fitting results.

[0012] In the above technical scheme, by defining a preset feature set, multiple features that are highly correlated with cerebrovascular disease are clarified. The selection of these features is based on previous research and clinical experience, and can accurately reflect the risk of cerebrovascular disease. The weight value of each preset feature is set by statistical analysis, and the importance and contribution of each feature in the prediction are taken into account, so that the model can pay more attention to the influence of key features when predicting. The risk factor model is constructed by the weighted summation method, and the preset features and their weight values ​​are incorporated into the model, which can comprehensively consider the influence of multiple factors and improve the prediction accuracy and scientificity of the model. The risk factor model is fitted using effective related features, and the AMS Grad optimization algorithm is configured according to the fitting results, which can ensure that the algorithm configuration is consistent with the actual data and improve the training efficiency and convergence speed of the model. The AMS Grad optimization algorithm is an adaptive learning rate algorithm suitable for non-convex optimization problems. It can automatically adjust the learning rate during the training process to meet the training needs of different stages. This solution can further enhance the adaptability and stability of the model by configuring the AMS Grad optimization algorithm.

[0013] Preferably, the S34 comprises the following steps: S341, randomly selecting a plurality of the effective correlation features as initial correlation features, inputting the initial correlation features into the risk factor model, performing preliminary fitting on the risk factor model, and obtaining preliminary fitting results; S342, verifying the preliminary fitting result by a cross-validation method, if the verification passes, taking the preliminary fitting result as the final fitting result and executing S343, if the verification fails, executing S341; S343. Adjust the parameters of the AMS Grad optimization algorithm according to the final fitting result, including the learning rate and momentum parameters.

[0014] In the above technical solution, multiple valid related features are randomly selected as initial related features for preliminary fitting. This randomness helps to avoid falling into the local optimal solution and gradually approach the global optimal solution. At the same time, multiple iterative attempts can ensure that the model can obtain good fitting effects under multiple feature combinations. The cross-validation method is used to verify the preliminary fitting results. This is an effective model evaluation method. Through cross-validation, the performance of the model on different data sets can be evaluated to ensure that the model has good generalization ability. According to the final fitting results, the parameters of the AMS Grad optimization algorithm are adjusted, including the learning rate and momentum parameters. This parameter adjustment method based on data feedback can ensure that the optimization algorithm is more adapted to the characteristics of the data during the training process, and improve the training efficiency and convergence speed of the model. By accurately adjusting the optimization algorithm parameters, the training effect of the model can be further improved. The optimized algorithm can better deal with noise and outliers in the data and improve the robustness and stability of the model.

[0015] Preferably, S4 comprises the following steps: S41, constructing a preset model for cerebrovascular disease prediction through a neural network model and initializing it; S42, setting a training data set and a verification data set based on the AMS Grade optimization algorithm; S43, inputting the AMS Grade optimization algorithm into the preset model for cerebrovascular disease prediction to perform model training to obtain an initial model for cerebrovascular disease prediction.

[0016] In the above technical solution, a cerebrovascular disease prediction preset model is constructed through a neural network model, and a large amount of data is used for training, which can capture complex nonlinear relationships and improve the accuracy of prediction. Through the introduction of the AMS Grade optimization algorithm, the training data set and the validation data set can be selected and optimized more effectively to ensure that the model can maintain good generalization ability in a variety of scenarios. The AMS Grade optimization algorithm is not only used to divide the data set, but also may play a role in the data preprocessing stage, such as removing noise data, processing missing values, etc., to ensure the data quality of the input model. The optimized data set is more representative of the actual situation, which helps the model learn more representative features and further improve the prediction performance. The neural network model itself has a strong learning ability. Combined with the AMS Grade optimization algorithm, the model can show stronger adaptability and robustness when facing new data or different scenarios. Even when the data distribution changes or there is noise interference, the model can maintain good prediction performance. The accuracy and robustness of the cerebrovascular disease prediction model are improved, which can more accurately assess the patient's risk status and formulate more reasonable treatment plans and preventive measures.

[0017] Preferably, S5 comprises the following steps: S51. Acquire the patient's real-time cerebrovascular disease-related data; S52, using the real-time cerebrovascular disease related data as input of the cerebrovascular disease prediction model to obtain a prediction result; S53, evaluating the cerebrovascular disease risk level corresponding to the patient according to the prediction result and a preset cerebrovascular disease risk level classification mechanism; S54: adaptively adjusting the model parameters of the cerebrovascular disease prediction model according to the prediction result to obtain an updated cerebrovascular disease prediction model.

[0018] In the above technical solution, by obtaining the patient's real-time cerebrovascular disease-related data, the model can make predictions based on the latest patient conditions, thereby improving the timeliness and accuracy of the prediction results. By inputting real-time data into the prediction model, the prediction results obtained can more accurately reflect the patient's current condition and future risk trends, providing doctors with a reliable basis for decision-making. Through the preset cerebrovascular disease risk level classification mechanism, the prediction results are converted into easy-to-understand and easy-to-operate risk levels, helping doctors to quickly assess the patient's risk status, thereby formulating more scientific and reasonable treatment plans and preventive measures. According to the prediction results, the parameters of the model are adaptively adjusted, so that the model can continuously learn and optimize when facing new data and situations, and improve its prediction ability and robustness under different conditions. The input of real-time data, the prediction of the model, the evaluation of the risk level, and the adaptive adjustment of the model constitute a closed-loop feedback system, which can continuously obtain new information and knowledge from clinical practice and promote the continuous improvement and upgrading of the model. Through continuous learning and adjustment, the prediction accuracy of the model will be further improved, thereby better serving clinical practice and patient health management.

[0019] Preferably, the S53 comprises the following steps: S531. Obtain different pathological types and severity levels based on historical cerebrovascular disease-related data; S532, setting a cerebrovascular disease risk level classification mechanism in combination with the pathological type and the severity; S533. Classify the prediction result by using the cerebrovascular disease risk level classification mechanism to obtain the cerebrovascular disease risk level corresponding to the patient.

[0020] In the above technical solution, emphasis is placed on obtaining different pathological types and severity levels based on historical cerebrovascular disease-related data, ensuring that the basic data of the risk level division mechanism is comprehensive, accurate and representative, thereby improving the scientific nature of the risk level division. The risk level division mechanism is set up in combination with the pathological type and severity, realizing the refined division of risks for different patient groups, which not only takes into account the differences in pathological types, but also takes into account the severity of the disease, making the risk level assessment more accurate and detailed. The prediction results are divided using the established risk level division mechanism, realizing the standardized operation of risk level assessment, which helps to reduce the interference of human factors and improve the consistency and reliability of the assessment results.

[0021] Preferably, the S54 comprises the following steps: S541, calculating the evaluation index of the cerebrovascular disease prediction model according to the prediction result; S542, evaluating the cerebrovascular disease prediction model according to the evaluation index to obtain a prediction deviation; S543, adjusting the model parameters of the cerebrovascular disease prediction model to optimize the prediction deviation; S544: When the prediction deviation is less than a deviation threshold, update the cerebrovascular disease prediction model based on the adjusted model parameters.

[0022] In the above technical solution, by calculating the evaluation indicators of the prediction model, such as accuracy, recall rate, F1 score, etc., the prediction performance of the model can be comprehensively measured, providing a quantitative basis for subsequent model evaluation and optimization. Evaluating the model according to the evaluation indicators and obtaining the prediction deviation can intuitively reflect the difference between the model prediction results and the actual situation, and help identify the deficiencies in the model. By adjusting the model parameters to optimize the prediction deviation, the prediction accuracy and generalization ability of the model can be improved. By continuously evaluating, adjusting and optimizing the model parameters, an iterative optimization mechanism is formed, which can gradually eliminate the unstable factors in the model and improve the stability and reliability of the model. The deviation threshold is set as the judgment criterion for updating the model. Only when the prediction deviation is less than this threshold, the prediction model will be updated based on the adjusted model parameters, ensuring that the updated model has higher prediction accuracy and stability.

[0023] According to a second aspect of the present invention, there is provided a cerebrovascular disease dynamic prediction system based on artificial intelligence, comprising: a feature extraction module for extracting features of historical cerebrovascular disease related data in a historical medical record database to obtain cerebrovascular disease related features; A feature selection module, used to select features based on cerebrovascular disease related features through a feature selection algorithm to obtain effective related features; AMS Grad optimization algorithm configuration module, which builds a risk factor model based on multiple preset features of cerebrovascular disease and the weight value of each preset feature, is used to substitute multiple valid related features into the risk factor model for fitting, and configures the AMS Grad optimization algorithm based on the fitting results; A model training module, used to train the preset cerebrovascular disease prediction model through the AMS Grad optimization algorithm to obtain a trained cerebrovascular disease prediction model; The model prediction module is used to predict the patient's historical cerebrovascular disease related data through the cerebrovascular disease prediction model, and adjust the cerebrovascular disease prediction model according to the prediction results.

[0024] In the above technical solution, the feature extraction module can extract features that are highly relevant to cerebrovascular disease from the historical medical record database. These features are more comprehensive and representative, providing a solid data foundation for subsequent analysis and prediction. The feature selection module uses an algorithm to screen out the effective relevant features that have the greatest impact on the prediction results, reducing the impact of redundant information on model performance and improving the accuracy and efficiency of prediction. The AMS Grad optimization algorithm configuration module constructs a risk factor model based on the specific characteristics and weight values ​​of cerebrovascular disease, and configures an optimization algorithm based on the model, making the model training process more efficient, able to fit data more accurately, and improve prediction accuracy. During the prediction process, the model prediction module not only outputs the prediction results, but also adjusts the cerebrovascular disease prediction model based on the prediction results. This dynamic optimization mechanism enables the model to continuously adapt to new data changes and improve its applicability and accuracy in different scenarios.

[0025] The beneficial effects of the present invention are as follows: through feature extraction and selection, the key features of cerebrovascular disease can be identified, noise data can be removed, a more accurate and concise prediction model can be constructed, and the prediction accuracy can be significantly improved. The AMS Grad optimization algorithm not only accelerates the convergence of the model, but also enhances the generalization ability of the model on unseen data, making the model more practical and reliable. During the prediction process, the model can be dynamically adjusted to form a closed-loop feedback, optimize the prediction results in real time, and provide personalized predictions for patients. In addition, the preprocessing step reduces the errors caused by data quality problems, and multi-dimensional feature extraction and fusion further enhance the feature expression ability and prediction ability of the model. The model finally constructed can accurately capture the pathogenesis of cerebrovascular disease and provide timely intervention for high-risk patients. Feature selection is performed by combining expert experience and machine learning to ensure that the model is simple and efficient. The evaluation and optimization mechanism continuously iterates to improve the stability and reliability of the model, ensure that the prediction deviation is within a controllable range, and provide strong support for clinical decision-making.

[0026] The above invention content is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0028] Figure 1 A flowchart of a method for dynamic prediction of cerebrovascular disease based on artificial intelligence disclosed in an embodiment of the present invention; Figure 2 The present invention is a schematic diagram of the structure of a cerebrovascular disease dynamic prediction system based on artificial intelligence disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the operations (or steps) as sequential processes, many of the operations (or steps) therein can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the operations can be rearranged. A process can be terminated when its operations are completed, but can also have additional steps not included in the accompanying drawings; a process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0031] Example: According to a first aspect of this embodiment, a method for dynamic prediction of cerebrovascular disease based on artificial intelligence is provided, such as Figure 1 As shown, the following steps are included: S1. Extract features of historical cerebrovascular disease related data in the historical medical record database to obtain cerebrovascular disease related features.

[0032] Specifically, S1 includes the following steps: S11, calling up a historical medical record database and extracting historical cerebrovascular related data from the historical medical record database; S12. Preprocessing the historical cerebrovascular related data, the preprocessing at least includes outlier processing, missing value processing and data standardization; S13, extracting features from the preprocessed historical cerebrovascular-related data to obtain corresponding time series features, text data features, and image features; S14. The time series features, text data features and image features are fused by a feature concatenation method to obtain the final cerebrovascular related features.

[0033] In this embodiment, it is assumed that an electronic medical record system (EMR) of a large tertiary hospital is used, which stores patient data for many years. First, a SQL query statement is used to retrieve all patient records diagnosed with "cerebrovascular disease" from the database. The query statement may include fields such as patient ID, name, age, gender, admission date, discharge date, diagnosis code (such as ICD-10 code), examination report ID (MRI, CT, etc.), and doctor's clinical notes. Then, abnormal values ​​are processed. For example, when checking age data, it is found that there is a record showing that the patient is 200 years old. Replace it with the median age in the data set (assuming it is 65 years old) to eliminate its impact on subsequent analysis. A practical example of missing value processing: For blood pressure records of some patients, if missing values ​​are found, since blood pressure is an important indicator for assessing the risk of cerebrovascular disease, the mean blood pressure of the patient at other time points is used to fill these missing values. If the data at other time points are also missing, consider using the mean or median of blood pressure in the entire data set for filling. For the patient's blood pressure data (such as systolic and diastolic blood pressure), standardization is performed. Assuming that the original mean of systolic blood pressure is 140 mmHg and the standard deviation is 20 mmHg, the systolic blood pressure of each patient is subtracted from the mean and divided by the standard deviation to obtain the standardized systolic blood pressure value. Perform time series feature extraction, and extract the mean, standard deviation, maximum value, minimum value, and change trend (such as the rate of blood pressure drop at night) of systolic and diastolic blood pressure from the patient's 24-hour dynamic blood pressure monitoring data. These features reflect the fluctuation of blood pressure and circadian rhythm changes of patients, which are of great significance for assessing the risk of cerebrovascular disease. Use natural language processing technology (such as the BERT model) to convert doctors' clinical notes into vector representations. For example, for a clinical note containing "the patient complained of headache, dizziness, accompanied by nausea and vomiting", the semantic information of these keywords is captured, and a high-dimensional vector is generated to represent the content of the note. For the patient's MRI or CT images, use a deep learning model (such as ResNet) for feature extraction, identify the lesion area in the image, and extract high-level features such as the shape, size, and location of the lesion. In the feature fusion stage, the feature vectors extracted from time series, text, and images are concatenated. Assuming that the length of the time series feature vector is 10, the length of the text feature vector is 300, and the length of the image feature vector is 2048, the final feature vector length is 2358. This long vector contains information on multiple dimensions of the patient, such as physiological indicators, clinical symptoms, and imaging manifestations.

[0034] In the above implementation, by including the preprocessing steps of processing outliers, missing values ​​and data standardization, not only can the prediction error caused by data quality problems be reduced, but also the efficiency and stability of subsequent model training can be improved. Multi-dimensional feature extraction is performed on the preprocessed data, including time series features, text data features and image features, which can more comprehensively capture various information related to cerebrovascular disease. The features of different dimensions are fused by the feature splicing method to obtain the final cerebrovascular-related features. The process of feature fusion helps to explore the potential relationship between different features, further enhance the expression and prediction capabilities of the features, and enable the prediction model to more comprehensively understand and predict the complex changes of cerebrovascular disease. With high-quality data and comprehensive feature extraction methods, the prediction model finally constructed will be able to more accurately capture the incidence patterns and trends of cerebrovascular disease, help identify potential high-risk patients in advance, and provide them with more timely and effective intervention measures. Multi-dimensional feature extraction and feature fusion also help to achieve more personalized predictions. By analyzing the specific characteristics of different patients, the model can give more accurate prediction results for different situations, thereby providing patients with more personalized treatment suggestions and health management plans.

[0035] S2. Based on the cerebrovascular disease related features, feature selection algorithm is used to select effective related features.

[0036] Specifically, S2 includes the following steps: S21, calculating the correlation characteristic coefficients between the prediction targets corresponding to the cerebrovascular related characteristics; S22. Set preset coefficients of relevant features through expert experience method, and eliminate cerebrovascular related features that are lower than the preset coefficients of relevant features; S23. Combine the machine learning model to perform recursive feature elimination on cerebrovascular related features to obtain effective relevant features.

[0037] In this embodiment, the prediction target is defined as "the probability of cerebrovascular events", and then statistical methods (such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) are used to calculate the correlation between each cerebrovascular-related feature and the prediction target. For example, by calculating the correlation coefficient between features such as age, blood pressure, blood sugar level, lesion size and the probability of cerebrovascular events. After obtaining the correlation coefficients of all features, a reasonable preset coefficient threshold is set based on expert experience. The preset coefficient threshold represents the minimum standard for the feature to have sufficient influence on the prediction target. Then eliminate those features whose correlation coefficients are lower than the preset coefficients. After eliminating low-correlation features, use a machine learning model to further perform feature selection.

[0038] In the above implementation, by calculating the correlation feature coefficient between the cerebrovascular related features and the prediction target, the degree of influence of each feature on the prediction target can be quantified, which helps to identify the features that are highly correlated with the prediction target and provide a strong basis for subsequent feature selection. By setting the preset coefficient of the relevant features through the expert experience method and eliminating the cerebrovascular related features below the coefficient, the features that are not closely related to or irrelevant to the prediction target can be effectively removed, the complexity and calculation amount of the model can be reduced, and the training efficiency and prediction performance of the model can be improved. Recursive feature elimination is performed in combination with the machine learning model to further screen out the most effective related features for the prediction target. Recursive feature elimination is an effective feature selection method. It continuously and iteratively builds the model and eliminates the least important features to finally obtain the optimal feature subset. It can make full use of the powerful learning ability of the machine learning model and further improve the accuracy and effectiveness of feature selection.

[0039] S3. Construct a risk factor model based on multiple preset features of cerebrovascular disease and the weight value of each preset feature, substitute multiple valid related features into the risk factor model for fitting, and configure the AMS Grad optimization algorithm according to the fitting results.

[0040] Specifically, S3 includes the following steps: S31. defining a preset feature set, including a plurality of features highly correlated with cerebrovascular disease; S32, setting a weight value for each preset feature according to a statistical analysis method; S33, constructing a risk factor model by weighted summation method according to preset characteristics and corresponding weight values; S34. Fit the risk factor model through effective correlation features and configure the AMS Grad optimization algorithm based on the fitting results.

[0041] In this embodiment, first, a set of preset feature sets highly correlated with cerebrovascular disease is defined based on medical research and clinical data. These features may include but are not limited to age, gender, history of hypertension, history of diabetes, smoking history, blood lipid levels (such as total cholesterol, triglycerides, HDL-C, LDL-C, etc.), body mass index (BMI), family history of cerebrovascular disease, etc. Statistical analysis methods are used to evaluate the degree of influence of each preset feature on the risk of cerebrovascular disease, and the weight value of each feature is set accordingly, and the weight value reflects the importance of the feature in the prediction. Using the weighted summation method, the value of each preset feature is multiplied by its corresponding weight, and then all products are added to obtain a comprehensive score or probability value, which is used to represent the risk level of cerebrovascular disease in an individual. Use the relevant feature set to fit the constructed risk factor model to improve the accuracy and generalization ability of the model. At the same time, in order to optimize the training process of the model, the AMS Grad optimization algorithm is configured to adjust the parameters of the model, including setting hyperparameters such as learning rate and momentum parameter.

[0042] In the above implementation, by defining a preset feature set, multiple features that are highly correlated with cerebrovascular disease are clarified. The selection of these features is based on previous research and clinical experience, and can accurately reflect the risk of cerebrovascular disease. The weight value of each preset feature is set by statistical analysis, and the importance and contribution of each feature in the prediction are taken into account, so that the model can pay more attention to the influence of key features when predicting. The risk factor model is constructed by the weighted summation method, and the preset features and their weight values ​​are incorporated into the model, which can comprehensively consider the influence of multiple factors and improve the prediction accuracy and scientificity of the model. The risk factor model is fitted using effective relevant features, and the AMS Grad optimization algorithm is configured according to the fitting results, which can ensure that the algorithm configuration is consistent with the actual data and improve the training efficiency and convergence speed of the model. The AMS Grad optimization algorithm is an adaptive learning rate algorithm suitable for non-convex optimization problems. It can automatically adjust the learning rate during the training process to meet the training needs of different stages. This solution can further enhance the adaptability and stability of the model by configuring the AMS Grad optimization algorithm.

[0043] Specifically, S34 includes the following steps: S341, randomly selecting a plurality of effective correlation features as initial correlation features, inputting the initial correlation features into the risk factor model, performing preliminary fitting on the risk factor model, and obtaining preliminary fitting results; S342, verifying the preliminary fitting result by cross-validation method, if the verification is passed, taking the preliminary fitting result as the final fitting result and executing S343, if the verification is not passed, executing S341; S343. Adjust the parameters of the AMS Grad optimization algorithm according to the final fitting result, including the learning rate and momentum parameters.

[0044] In this implementation, a random number generator is first used to select initial relevant features, the selected initial relevant features are input into the risk factor model, the model is preliminarily fitted using the training data set, and the preliminary fitting results (such as loss function value, accuracy, etc.) are recorded. The training data set is divided into K subsets. For each subset, the remaining K-1 subsets are used as training sets and the current subset is used as a validation set to train and validate the model. The results of each validation are recorded, and performance indicators such as average accuracy and average loss function value are calculated. Whether the preliminary fitting result passes the validation is determined based on the performance indicators. If the validation is passed, S343 is executed; if not, the initial relevant features are reselected by returning to S341. The performance indicators of the final fitting result are analyzed, and the learning rate and momentum parameters are gradually adjusted based on the analysis results. The learning rate can start from a larger value and then gradually decrease; the momentum parameter can be fine-tuned according to the convergence of the model. In the process of adjusting the parameters, the cross-validation method is used to evaluate the impact of different parameter combinations on the model performance. The parameter combination that optimizes the model performance is selected and saved as the final configuration. Retrain the model using the AMS Grad optimization algorithm with the final configuration to ensure the model achieves optimal performance.

[0045] In the above implementation, a plurality of valid related features are randomly selected as initial related features for preliminary fitting. This randomness helps to avoid falling into a local optimal solution and gradually approach the global optimal solution. At the same time, multiple iterative attempts can ensure that the model can obtain a good fitting effect under multiple feature combinations. The cross-validation method is used to verify the preliminary fitting results. This is an effective model evaluation method. Through cross-validation, the performance of the model on different data sets can be evaluated to ensure that the model has good generalization ability. The parameters of the AMS Grad optimization algorithm are adjusted according to the final fitting results, including the learning rate and momentum parameters. This parameter adjustment method based on data feedback can ensure that the optimization algorithm is more adapted to the characteristics of the data during the training process, and improve the training efficiency and convergence speed of the model. By accurately adjusting the optimization algorithm parameters, the training effect of the model can be further improved. The optimized algorithm can better deal with noise and outliers in the data, and improve the robustness and stability of the model.

[0046] S4. The preset cerebrovascular disease prediction model is trained by the AMS Grad optimization algorithm to obtain a trained cerebrovascular disease prediction model.

[0047] Specifically, S4 includes the following steps: S41, constructing a preset model for cerebrovascular disease prediction through a neural network model and initializing it; S42. Setting training data set and validation data set based on AMS Grade optimization algorithm; S43, inputting the AMS Grade optimization algorithm into the preset model for cerebrovascular disease prediction to perform model training to obtain an initial model for cerebrovascular disease prediction.

[0048] In this embodiment, the structure of the neural network model is first determined, including the configuration of the input layer, hidden layer, and output layer. Then, the model parameters, such as weights and biases, are initialized, usually using random small values ​​or specific initialization methods, such as He initialization, Xavier initialization, etc. The training parameters of the model are configured, such as the learning rate, batch size, number of iterations, etc.

[0049] Select or define appropriate activation functions and loss functions. Randomly divide the original dataset into training dataset and validation dataset in the ratio of 80% training set and 20% validation set. Make sure that the training dataset and validation dataset are as consistent as possible in data distribution so that the model performance can be accurately evaluated. Configure the parameters of the AMS Grad optimization algorithm, such as the initial learning rate, momentum decay coefficient, etc. Prepare data loaders or batch functions to load training data and validation data in batches during training. Initialize the training loop and set the number of iterations or stop conditions. In each iteration, load a batch of data from the training dataset. Perform forward propagation, input the data into the neural network model, and calculate the prediction results. Calculate the loss value between the prediction results and the true label. Perform backpropagation, calculate the gradient based on the loss value, and update the parameters of the model using the AMS Grad optimization algorithm. After each iteration or every few iterations, use the validation dataset to evaluate the performance of the model and record performance indicators (such as accuracy, loss value, etc.). Adjust hyperparameters or perform early stopping based on the verification results. When the stopping condition is met, the stopping condition can be set to reach the maximum number of iterations or the validation loss no longer decreases, ending the training process and obtaining the initial model for cerebrovascular disease prediction.

[0050] In the above implementation, a cerebrovascular disease prediction preset model is constructed through a neural network model, and a large amount of data is used for training, which can capture complex nonlinear relationships, thereby improving the accuracy of the prediction. Through the introduction of the AMS Grade optimization algorithm, the training data set and the validation data set can be selected and optimized more effectively to ensure that the model can maintain good generalization ability in a variety of scenarios. The AMS Grade optimization algorithm is not only used to divide the data set, but also may play a role in the data preprocessing stage, such as removing noise data, processing missing values, etc., so as to ensure the data quality of the input model. The optimized data set is more representative of the actual situation, which helps the model learn more representative features and further improve the prediction performance. The neural network model itself has a strong learning ability. Combined with the AMS Grade optimization algorithm, the model can show stronger adaptability and robustness when facing new data or different scenarios. Even when the data distribution changes or there is noise interference, the model can maintain good prediction performance. The accuracy and robustness of the cerebrovascular disease prediction model are improved, which can more accurately assess the patient's risk status and formulate more reasonable treatment plans and preventive measures.

[0051] S5. Predict the patient's historical cerebrovascular disease related data using a cerebrovascular disease prediction model, and adjust the cerebrovascular disease prediction model according to the prediction results.

[0052] Specifically, S5 includes the following steps: S51. Acquire the patient's real-time cerebrovascular disease-related data; S52, using the real-time cerebrovascular disease related data as input of a cerebrovascular disease prediction model to obtain a prediction result; S53, evaluating the patient's corresponding cerebrovascular disease risk level according to the prediction result and a preset cerebrovascular disease risk level classification mechanism; S54. Adaptively adjust the model parameters of the cerebrovascular disease prediction model according to the prediction results to obtain an updated cerebrovascular disease prediction model.

[0053] In the above implementation, by obtaining the patient's real-time cerebrovascular disease-related data, the model can make predictions based on the latest patient conditions, thereby improving the timeliness and accuracy of the prediction results. By inputting real-time data into the prediction model, the prediction results obtained can more accurately reflect the patient's current condition and future risk trends, providing doctors with a reliable basis for decision-making. Through the preset cerebrovascular disease risk level classification mechanism, the prediction results are converted into risk levels that are easy to understand and operate, helping doctors to quickly assess the patient's risk status, thereby formulating more scientific and reasonable treatment plans and preventive measures. According to the prediction results, the parameters of the model are adaptively adjusted, so that the model can continuously learn and optimize when facing new data and situations, and improve its prediction ability and robustness under different conditions. The input of real-time data, the prediction of the model, the evaluation of the risk level, and the adaptive adjustment of the model constitute a closed-loop feedback system that can continuously obtain new information and knowledge from clinical practice and promote the continuous improvement and upgrading of the model. Through continuous learning and adjustment, the prediction accuracy of the model will be further improved, thereby better serving clinical practice and patient health management.

[0054] Specifically, S53 includes the following steps: S531. Obtain different pathological types and severity levels based on historical cerebrovascular disease-related data; S532. Set up a risk classification mechanism for cerebrovascular disease based on pathological type and severity; S533. Classify the prediction results using the cerebrovascular disease risk level classification mechanism to obtain the cerebrovascular disease risk level corresponding to the patient.

[0055] In this embodiment, historical cerebrovascular disease-related data are first collected, including the patient's age, gender, underlying disease, cerebrovascular disease type, lesion location, lesion size, treatment, prognosis and other data. Then, these data are sorted and analyzed to identify different cerebrovascular disease pathological types. According to the classification criteria, cerebrovascular disease can be divided into ischemic stroke (such as cerebral infarction, transient ischemic attack, etc.), hemorrhagic stroke (such as cerebral hemorrhage, subarachnoid hemorrhage, etc.), intracranial aneurysm, intracranial vascular malformation, cerebral vasculitis and other types. At the same time, according to the patient's clinical manifestations, imaging examination results and prognosis, the severity of each pathological type is evaluated, and the severity is divided into different levels such as mild, moderate and severe. After obtaining information on different pathological types and severity, it is necessary to set up a cerebrovascular disease risk level division mechanism. This mechanism comprehensively considers multiple factors such as pathological type, severity, patient age, underlying disease, etc. to comprehensively assess the patient's cerebrovascular disease risk. The risk level is divided into four different levels: no risk, low risk, medium risk and high risk. The specific classification criteria are formulated based on medical expertise and clinical experience, for example: No risk: The patient has no risk factors related to cerebrovascular disease and no symptoms or signs of cerebrovascular disease. Low risk: The patient has fewer risk factors for cerebrovascular disease, or has symptoms but the condition is mild, and the prognosis is good. Medium risk: The patient has multiple risk factors for cerebrovascular disease, or the condition is severe but not yet life-threatening. High risk: The patient has multiple serious risk factors for cerebrovascular disease, or the condition is critical and may endanger life at any time. After establishing the cerebrovascular disease risk level classification mechanism, the prediction results are divided. For each patient's prediction result, it is evaluated and divided according to the risk level classification mechanism. First, determine the patient's pathological type and severity; then, combine the patient's age, underlying disease and other factors to comprehensively evaluate the patient's cerebrovascular disease risk level. Finally, each patient is divided into the corresponding risk level, and the corresponding prevention and treatment strategies are formulated accordingly. For example, for high-risk patients, monitoring and intervention measures should be strengthened to reduce their risk of cerebrovascular disease; for low-risk patients, the occurrence of cerebrovascular disease can be prevented by improving lifestyles and controlling risk factors.

[0056] In the above implementation, emphasis is placed on obtaining different pathological types and severity levels based on historical cerebrovascular disease-related data, ensuring that the basic data of the risk level division mechanism is comprehensive, accurate and representative, thereby improving the scientific nature of the risk level division. The risk level division mechanism is set up in combination with the pathological type and severity, and a refined division of risks for different patient groups is achieved. Not only the differences in pathological types are considered, but also the severity of the disease is taken into account, making the risk level assessment more accurate and detailed. The prediction results are divided using the established risk level division mechanism, which realizes the standardized operation of risk level assessment, helps reduce the interference of human factors, and improves the consistency and reliability of the assessment results.

[0057] Specifically, S54 includes the following steps: S541. Calculate the evaluation index of the cerebrovascular disease prediction model according to the prediction results; S542, evaluating the cerebrovascular disease prediction model according to the evaluation index to obtain the prediction deviation; S543, adjusting the model parameters of the cerebrovascular disease prediction model to optimize the prediction deviation; S544. When the prediction deviation is less than the deviation threshold, the cerebrovascular disease prediction model is updated based on the adjusted model parameters.

[0058] In this embodiment, the evaluation index of the cerebrovascular disease prediction model is first calculated according to the prediction results, and the calculated evaluation index is compared with a pre-set threshold or industry standard to identify the deficiencies in the model, such as inaccurate prediction of a specific type of case, insensitivity to certain features, etc. The size and distribution of the prediction deviation are quantified by systematic means. The hyperparameters of the model are adjusted by grid search, and the model is retrained using the adjusted parameters or structure. After the model has been adjusted for many times, if the prediction deviation is reduced to an acceptable range (i.e., less than the set deviation threshold), the cerebrovascular disease prediction model is updated using these optimal model parameters.

[0059] In the above implementation, by calculating the evaluation indicators of the prediction model, such as accuracy, recall rate, F1 score, etc., the prediction performance of the model can be comprehensively measured, providing a quantitative basis for subsequent model evaluation and optimization. The model is evaluated according to the evaluation indicators to obtain the prediction deviation, which can intuitively reflect the difference between the model prediction result and the actual situation, and help identify the deficiencies in the model. By adjusting the model parameters to optimize the prediction deviation, the prediction accuracy and generalization ability of the model can be improved. By continuously evaluating, adjusting and optimizing the model parameters, an iterative optimization mechanism is formed, which can gradually eliminate the unstable factors in the model and improve the stability and reliability of the model. By setting the deviation threshold as the judgment criterion for updating the model. Only when the prediction deviation is less than this threshold, the prediction model will be updated based on the adjusted model parameters, ensuring that the updated model has higher prediction accuracy and stability.

[0060] According to a second aspect of this embodiment, a cerebrovascular disease dynamic prediction system based on artificial intelligence is provided. Figure 2 As shown. Includes: Feature extraction module M01, used to extract features from historical cerebrovascular disease related data in the historical medical record database to obtain cerebrovascular disease related features; A feature selection module M02 is used to select features based on cerebrovascular disease related features through a feature selection algorithm to obtain effective related features; AMS Grad optimization algorithm configuration module M03, which constructs a risk factor model based on multiple preset features of cerebrovascular disease and the weight value of each preset feature, is used to substitute multiple valid related features into the risk factor model for fitting, and configures the AMS Grad optimization algorithm according to the fitting results; The model training module M04 is used to train the preset cerebrovascular disease prediction model through the AMS Grad optimization algorithm to obtain a trained cerebrovascular disease prediction model; The model prediction module M05 is used to predict the patient's historical cerebrovascular disease related data through the cerebrovascular disease prediction model, and adjust the cerebrovascular disease prediction model according to the prediction results.

[0061] Beneficial effects of this embodiment: This embodiment fully considers the multi-dimensional characteristics of historical medical record data, effectively extracts and selects features that are highly correlated with cerebrovascular disease, and constructs a scientific and reasonable risk factor model. The AMS Grad optimization algorithm improves the training efficiency and prediction performance of the model. In addition, the model parameters can be dynamically adjusted according to real-time data and prediction results to continuously improve the prediction accuracy and stability. This embodiment can significantly improve the accuracy of cerebrovascular disease prediction, optimize treatment plans, and reduce patient risks. It provides strong support for clinical decision-making and helps to achieve personalized medical care and health management.

[0062] The above specific implementation is a preferred implementation of a method and system for dynamic prediction of cerebrovascular disease based on artificial intelligence of the present invention, and is not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to this specific implementation. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A method for dynamic prediction of cerebrovascular disease based on artificial intelligence, characterized in that: The steps include: S1. Extracting features from historical cerebrovascular disease related data in the historical medical record database to obtain cerebrovascular disease related features; S2. Based on the cerebrovascular disease related features, feature selection is performed by a feature selection algorithm to obtain effective related features; S3. constructing a risk factor model according to a plurality of preset features of cerebrovascular disease and a weight value of each preset feature, substituting the plurality of effective related features into the risk factor model for fitting, and configuring an AMS Grad optimization algorithm according to the fitting result; S4. Training the preset cerebrovascular disease prediction model by the AMS Grad optimization algorithm to obtain a trained cerebrovascular disease prediction model; S5. Predicting the patient's historical cerebrovascular disease-related data using the cerebrovascular disease prediction model, and adjusting the cerebrovascular disease prediction model according to the prediction result.

2. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 1, characterized in that: The S1 comprises the following steps: S11, calling up a historical medical record database and extracting historical cerebrovascular related data from the historical medical record database; S12, preprocessing the historical cerebrovascular-related data, wherein the preprocessing at least includes outlier processing, missing value processing and data standardization; S13, extracting features from the preprocessed historical cerebrovascular-related data to obtain corresponding time series features, text data features, and image features; S14, fusing the time series features, text data features and image features through a feature concatenation method to obtain final cerebrovascular related features.

3. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 1, characterized in that: The S2 comprises the following steps: S21, calculating the correlation characteristic coefficients between the prediction targets corresponding to the cerebrovascular related characteristics; S22, setting preset coefficients of relevant features by expert experience method, and eliminating cerebrovascular related features that are lower than the preset coefficients of relevant features; S23. Recursively eliminate the cerebrovascular related features in combination with the machine learning model to obtain effective related features.

4. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 1, characterized in that: The S3 comprises the following steps: S31. defining a preset feature set, including a plurality of features highly correlated with cerebrovascular disease; S32, setting a weight value for each preset feature according to a statistical analysis method; S33, constructing a risk factor model by weighted summation method according to the preset characteristics and corresponding weight values; S34. Fitting the risk factor model through the effective correlation features, and configuring the AMSGrad optimization algorithm according to the fitting results.

5. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 3, characterized in that: The S34 comprises the following steps: S341, randomly selecting a plurality of the effective correlation features as initial correlation features, inputting the initial correlation features into the risk factor model, performing preliminary fitting on the risk factor model, and obtaining preliminary fitting results; S342, verifying the preliminary fitting result by a cross-validation method, if the verification passes, taking the preliminary fitting result as the final fitting result and executing S343, if the verification fails, executing S341; S343. Adjust the parameters of the AMS Grad optimization algorithm according to the final fitting result, including the learning rate and momentum parameters.

6. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 1, characterized in that: The S4 comprises the following steps: S41, constructing a preset model for cerebrovascular disease prediction through a neural network model and initializing it; S42, setting a training data set and a verification data set based on the AMS Grade optimization algorithm; S43, inputting the AMS Grade optimization algorithm into the preset model for cerebrovascular disease prediction to perform model training to obtain an initial model for cerebrovascular disease prediction.

7. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 1, characterized in that: The S5 comprises the following steps: S51. Acquire the patient's real-time cerebrovascular disease-related data; S52, using the real-time cerebrovascular disease related data as input of the cerebrovascular disease prediction model to obtain a prediction result; S53, evaluating the cerebrovascular disease risk level corresponding to the patient according to the prediction result and a preset cerebrovascular disease risk level classification mechanism; S54: adaptively adjusting the model parameters of the cerebrovascular disease prediction model according to the prediction result to obtain an updated cerebrovascular disease prediction model.

8. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 7, characterized in that: The S53 comprises the following steps: S531. Obtain different pathological types and severity levels based on historical cerebrovascular disease-related data; S532, setting a cerebrovascular disease risk level classification mechanism in combination with the pathological type and the severity; S533. Classify the prediction result by using the cerebrovascular disease risk level classification mechanism to obtain the cerebrovascular disease risk level corresponding to the patient.

9. The method for dynamic prediction of cerebrovascular disease based on artificial intelligence according to claim 7, characterized in that: The S54 comprises the following steps: S541, calculating the evaluation index of the cerebrovascular disease prediction model according to the prediction result; S542, evaluating the cerebrovascular disease prediction model according to the evaluation index to obtain a prediction deviation; S543, adjusting the model parameters of the cerebrovascular disease prediction model to optimize the prediction deviation; S544: When the prediction deviation is less than a deviation threshold, update the cerebrovascular disease prediction model based on the adjusted model parameters.

10. A cerebrovascular disease dynamic prediction system based on artificial intelligence, applicable to a cerebrovascular disease dynamic prediction method based on artificial intelligence as claimed in any one of claims 1 to 9, characterized in that: include: A feature extraction module, used to extract features of historical cerebrovascular disease related data in the historical medical record database to obtain cerebrovascular disease related features; A feature selection module, used to select features based on cerebrovascular disease related features through a feature selection algorithm to obtain effective related features; AMS Grad optimization algorithm configuration module, which builds a risk factor model based on multiple preset features of cerebrovascular disease and the weight value of each preset feature, is used to substitute multiple valid related features into the risk factor model for fitting, and configures the AMS Grad optimization algorithm based on the fitting results; A model training module, used to train the preset cerebrovascular disease prediction model through the AMS Grad optimization algorithm to obtain a trained cerebrovascular disease prediction model; The model prediction module is used to predict the patient's historical cerebrovascular disease related data through the cerebrovascular disease prediction model, and adjust the cerebrovascular disease prediction model according to the prediction results.

Images