Method and system for constructing dynamic prediction model of cerebrovascular disease based on AI

Abstract

The present invention discloses an AI-based cerebrovascular disease dynamic prediction model construction method and system, which relate to the field of disease management technology. The system includes an information aggregation module, a data processing module, a learning prediction module, a risk assessment module and a frequency detection prediction module, and the entire system can monitor the patient's brain abnormalities. The technical points are as follows: effective integration of historical data and image data is achieved, and comprehensive data support is provided for stroke risk prediction. Through data cleaning and standardization, the quality of model input data is ensured, and the prediction accuracy of the model is improved. The patient's condition estimate is compared with a preset evaluation threshold, and a corresponding strategy is executed according to the comparison result. Under the condition that the patient's condition is judged to be serious, a frequency detection adjustment strategy is executed, and the inspection frequency prediction value is calculated according to the patient's condition estimate and the evaluation threshold, so as to increase the inspection density, thereby achieving targeted management of the patient's condition and improving the patient's safety.

Description

Technical Field

[0001] The present invention relates to the field of disease management technology, and specifically to a method and system for constructing a dynamic prediction model for cerebrovascular disease based on AI. Background Art

[0002] Disease management is a comprehensive approach to health care that aims to prevent, treat and control diseases through systematic strategies and measures; it involves a full range of services and management from disease prevention, early screening, diagnosis, treatment, rehabilitation to long-term care. The main goals of disease management include: preventing the occurrence of diseases: promoting people to develop healthy living habits and reduce the risk of disease through education, publicity, lifestyle intervention and other means; early screening and diagnosis: timely identification of potential health problems or disease risks through regular health examinations and screenings for early treatment and intervention; providing effective treatment and management: taking scientific and effective treatment measures for confirmed diseases and conducting comprehensive disease management, including drug treatment, surgical treatment, rehabilitation training, etc.

[0003] The technical solution pointed out in the document with the existing authorization announcement number CN113628754B and the name of the document entitled Method and System for Constructing a Dynamic Prediction Model of Cerebrovascular Disease Based on Artificial Intelligence includes: performing feature extraction on the test result item values ​​of the target patient to obtain feature extraction results, performing feature transformation on multiple first features in the feature extraction results to obtain multiple transformed second features, constructing a risk factor model based on multiple preset features of cerebrovascular disease and the weight value of each preset feature, substituting the multiple second features into the risk factor model for fitting, constructing a preset Adam optimization algorithm based on the fitting results, pre-constructing a cerebrovascular disease prediction model, and training the cerebrovascular disease prediction model using the preset Adam optimization algorithm to obtain a trained cerebrovascular disease prediction model; accurate cerebrovascular disease assessment can be performed for the patient's commonly used examination items, ensuring the accuracy and objectivity of the assessment results; reducing costs and improving user experience, but it cannot provide effective follow-up examination management measures for different patients.

[0004] In combination with the above documents and existing technologies, patients with cerebrovascular disease need to undergo regular examinations after treatment, such as brain CT scans and blood tests. Although the examination is relatively comprehensive, it still requires judgment and analysis by doctors, and it is impossible to get a rough estimate of the condition as soon as possible. If it is determined that the patient's condition is more serious, treatment is required as soon as possible, and the original examination cycle should be shortened. The degree of shortening still needs to be controlled manually, and it is impossible to accurately give the duration of the shortened cycle, which affects the management of the patient's condition. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an AI-based method and system for constructing a dynamic prediction model for cerebrovascular disease, which realizes the effective integration of historical data and image data, provides comprehensive data support for stroke risk prediction, ensures the quality of model input data through data cleaning and standardization, and improves the prediction accuracy of the model; compares the patient's condition estimate with a preset evaluation threshold, executes a corresponding strategy based on the comparison result, executes a frequency adjustment strategy under the condition that the patient's condition is judged to be serious, calculates the inspection frequency prediction value based on the patient's condition estimate and the evaluation threshold, increases the inspection density, thereby realizing targeted management of the patient's condition, improving the patient's safety, and solving the problems raised in the background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] The AI-based cerebrovascular disease dynamic prediction model construction system includes:

[0008] Information aggregation module, which obtains and aggregates the historical data and image data of each patient;

[0009] A data processing module pre-processes the acquired historical data and image data;

[0010] The learning prediction module uses AI technology to predict the risk of stroke in patients and analyze the image data of the patient's brain, outputs the probability value and image analysis results, and the image analysis results include normal and abnormal, and obtains the abnormal area and location when it is determined to be abnormal;

[0011] The risk assessment module calculates the patient's condition estimate based on the dimensionless probability value of the corresponding patient when the image analysis result is normal. When the image analysis result is abnormal, the patient's condition estimate is calculated based on the dimensionless probability value and abnormal area of ​​the corresponding patient. ;

[0012] Frequency detection prediction module, which estimates the patient's condition With the preset evaluation threshold Make a comparison and implement corresponding adjustment strategies based on the comparison results.

[0013] Furthermore, the historical data includes demographic characteristics, medical examination data and medical history data. The demographic characteristics include age, gender and race. The medical examination data include blood pressure, blood sugar, cholesterol and body mass index. The medical history data includes the history of hypertension, diabetes and heart disease, that is, whether there is a corresponding medical history. The image data includes several MRI images of the corresponding patient's brain.

[0014] Furthermore, the process of preprocessing the acquired historical data and image data is: data cleaning and data standardization.

[0015] Furthermore, AI techniques include random forest algorithms and convolutional neural networks.

[0016] Furthermore, the process of predicting stroke risk using the random forest algorithm is as follows:

[0017] Extract medical examination data from the corresponding patient historical data, perform feature selection and label definition before splitting the data, and divide the data into training set and test set. Use Python's scikit-learn library to build a random forest model, use the random forest algorithm to train the model on the training set, adjust the model parameters through cross-validation and grid search, deploy the trained model to the clinical system, and output the classification results, which are displayed in the form of probability values.

[0018] Further, the process of analyzing the patient's brain image data is as follows:

[0019] Use convolutional neural networks to analyze image data, collect and annotate several brain MRI images, divide them into normal and abnormal categories, build a convolutional neural network model, including convolutional layers, pooling layers, and fully connected layers, use annotated data sets to train the model, adjust model parameters to optimize classification performance, and output image analysis results.

[0020] Furthermore, when it is determined to be abnormal, the abnormal area and position are obtained, and the process is as follows:

[0021] Under the condition of determining the abnormality, an image processing algorithm is used to extract lesion features from the MRI image, and the location and area of ​​the abnormality are extracted and obtained. The image processing algorithm includes the use of OpenCV and scikit-image libraries.

[0022] Further, the estimated value of the patient's condition is calculated The formula is as follows:

[0023] ;

[0024] In formula (1), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is normal. is a constant correction factor, and The value range is 1~2. is the probability value of the corresponding patient's image analysis result being normal;

[0025] In formula (2), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is abnormal. is a constant correction factor, and The value range is 1~2. is the probability value of abnormal image analysis result of the corresponding patient, represents the abnormal area, are the probability value and the preset proportional coefficient of the abnormal area when the image analysis result of the corresponding patient is abnormal, respectively, and .

[0026] Furthermore, the process of executing the corresponding adjustment strategy according to the comparison results is as follows:

[0027] Estimated patient condition The assessment threshold is not exceeded Under the condition of the patient's condition, no response action is taken and the established adjustment strategy is implemented; Exceeding the evaluation threshold Under the condition of , a first-level warning signal is issued and the frequency detection adjustment strategy is implemented;

[0028] Among them, the implementation of the established adjustment strategy is:

[0029] The corresponding patients go to the hospital for regular check-ups according to the established check-up frequency T;

[0030] The process of executing the frequency detection adjustment strategy is as follows:

[0031] When a level 1 warning signal is received, the patient's condition is estimated and evaluation threshold , calculate the predicted inspection frequency , and check the frequency prediction value The formula is as follows:

[0032] ;

[0033] In the formula, represents the error correction factor, and The value range is 0~1;

[0034] Corresponding patients according to the predicted value of examination frequency , go to the hospital for regular check-ups.

[0035] The method for constructing a dynamic prediction model for cerebrovascular disease based on AI includes the following steps:

[0036] S1. Obtain and summarize the historical data and image data of each patient;

[0037] The historical data includes demographic characteristics, medical examination data and medical history data. The demographic characteristics include age, gender and race. The medical examination data include blood pressure, blood sugar, cholesterol and body mass index. The medical history data include the history of hypertension, diabetes and heart disease, that is, whether there is a corresponding medical history. The image data includes several MRI images of the brain of the corresponding patient.

[0038] S2, preprocessing the acquired historical data and image data;

[0039] S3. Use AI technology to predict the risk of stroke in patients and analyze the image data of the patient's brain, output probability values ​​and image analysis results, and the image analysis results include normal and abnormal, and obtain the abnormal area and location when it is determined to be abnormal;

[0040] S4. Under the condition that the image analysis result is normal, calculate the estimated value of the patient's condition based on the probability value of the corresponding patient after dimensionless processing When the image analysis result is abnormal, the patient's condition estimate is calculated based on the dimensionless probability value and abnormal area of ​​the corresponding patient. ;

[0041] Among them, calculate the estimated value of the patient's condition The formula is as follows:

[0042] ;

[0043] In formula (1), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is normal. is a constant correction factor, and The value range is 1~2. is the probability value of the corresponding patient's image analysis result being normal;

[0044] In formula (2), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is abnormal. is a constant correction factor, and The value range is 1~2. is the probability value of abnormal image analysis result of the corresponding patient, represents the abnormal area, are the probability value and the preset proportional coefficient of the abnormal area when the image analysis result of the corresponding patient is abnormal, respectively, and .

[0045] S5. Estimate the patient's condition With the preset evaluation threshold Make a comparison;

[0046] Estimated patient condition The assessment threshold is not exceeded Under the condition of the patient's condition, no response action is taken and the established adjustment strategy is implemented; Exceeding the evaluation threshold Under the condition of , a first-level warning signal is issued and the frequency detection adjustment strategy is implemented;

[0047] Among them, the implementation of the established adjustment strategy is:

[0048] The corresponding patients go to the hospital for regular check-ups according to the established check-up frequency T;

[0049] The process of executing the frequency detection adjustment strategy is as follows:

[0050] When a level 1 warning signal is received, the patient's condition is estimated and evaluation threshold , calculate the predicted inspection frequency , and check the frequency prediction value The formula is as follows:

[0051] ;

[0052] In the formula, represents the error correction factor, and The value range is 0~1;

[0053] Corresponding patients according to the predicted value of examination frequency , go to the hospital for regular check-ups.

[0054] The present invention provides a method and system for constructing a dynamic prediction model for cerebrovascular disease based on AI, which has the following beneficial effects:

[0055] 1. It realizes the effective integration of historical data and image data, providing comprehensive data support for stroke risk prediction. Through data cleaning and standardization, it ensures the quality of model input data and improves the prediction accuracy of the model. It uses the random forest algorithm to realize efficient stroke risk prediction, and outputs probability values ​​to help doctors make accurate risk assessments. It realizes efficient analysis of brain MRI images through the CNN model, which can accurately identify brain abnormalities and locate abnormal areas, providing an important reference for clinical diagnosis.

[0056] 2. Calculate the patient's condition estimate based on the image analysis results and the corresponding data parameters. , can realize targeted condition prediction and treatment, further ensure the accuracy of the patient's condition prediction, compare the patient's condition prediction value with the preset evaluation threshold, execute the corresponding strategy according to the comparison result, and execute the frequency adjustment strategy under the condition that the patient's condition is judged to be serious. According to the patient's condition prediction value and the evaluation threshold, the inspection frequency prediction value is calculated to increase the inspection density, thereby realizing targeted management of the patient's condition and improving the patient's safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of the module structure of the AI-based cerebrovascular disease dynamic prediction model construction system in the present invention;

[0058] Figure 2 The figure is a flowchart of the overall steps of the method for constructing a dynamic prediction model for cerebrovascular disease based on AI in the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Example 1: Please refer to Figure 1 , this embodiment provides an AI-based cerebrovascular disease dynamic prediction model construction system, the system includes an information aggregation module, a data processing module, a learning prediction module, a risk assessment module and a frequency detection prediction module, and the entire system can monitor the patient's brain abnormalities;

[0061] Information aggregation module, which obtains and aggregates the historical data and image data of each patient;

[0062] The patient's historical data and image data are all for a single patient, and the historical data include demographic characteristics, medical examination data, and medical history data; the historical data is obtained or extracted from an electronic health record (EHR) or a medical database;

[0063] Demographic characteristics included age, gender, and race;

[0064] Medical examination data included blood pressure, blood sugar, cholesterol, and body mass index (BMI);

[0065] Medical history data included history of hypertension, diabetes, and heart disease, that is, whether there was a corresponding medical history;

[0066] It should be noted that the medical history data refers to past medical history data;

[0067] The image data includes a number of MRI images corresponding to the patient's brain;

[0068] The data processing module preprocesses the acquired historical data and image data, and the preprocessing includes data cleaning and data standardization. Data cleaning is to remove noise and erroneous data and fill in missing values. Data standardization is to statistically calculate data from different sources to the same scale.

[0069] For example: Use Python's Pandas library for data cleaning and standardization; use image processing algorithms to extract lesion features from MRI images; Pandas is a Python data analysis library that is widely used for data cleaning and processing; image processing algorithms: use OpenCV and scikit-image libraries for image preprocessing, segmentation, and feature extraction; in addition, these two parts can be combined to form a complete data processing and analysis process for predicting stroke risk.

[0070] The learning prediction module uses AI technology to predict the risk of stroke in patients and analyze the image data of the patient's brain, outputs the probability value and image analysis results, and the image analysis results include normal and abnormal, and obtains the abnormal area and location when it is determined to be abnormal;

[0071] Among them, AI technologies include random forest algorithms and convolutional neural networks;

[0072] The process of predicting stroke risk using the random forest algorithm is as follows:

[0073] Extract medical examination data from the corresponding patient's historical data, and divide the data into training set and test set. Use Python's scikit-learn library to build a random forest model. Use the random forest algorithm to train the model on the training set, evaluate the model performance on the test set, use indicators such as accuracy, recall, and F1 score, adjust model parameters through cross-validation and grid search to improve performance, deploy the trained model to the clinical system, and use it to predict the stroke risk of new patients in real time. Output the classification results, and display the classification results in the form of probability values;

[0074] Split data: When dividing the data into training set and test set, the ratio is 80% training set and 20% test set;

[0075] Among them, feature selection and label definition are performed before data segmentation: blood pressure, blood sugar, cholesterol, and body mass index are selected as features, and stroke is set as the label with 0 indicating no risk and 1 indicating risk;

[0076] The sample code for building a random forest model using Python's scikit-learn library is as follows:

[0077] import pandas as pd

[0078] from sklearn.model_selection import train_test_split

[0079] from sklearn.ensemble import RandomForestClassifier

[0080] from sklearn.metrics import accuracy_score, recall_score, f1_score

[0081] # Assume the data has been loaded into dataframe df, labelled 'target'

[0082] X = df.drop(columns=['target'])

[0083] y = df['target']

[0084] # Data segmentation

[0085] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

[0086] # Build a random forest model

[0087] model = RandomForestClassifier(n_estimators=100, random_state=42)

[0088] # Train the model on the training set

[0089] model.fit(X_train, y_train);

[0090] Through the above steps, the random forest algorithm can be used to predict the risk of stroke. This process includes data collection and preprocessing, model training, performance evaluation and parameter adjustment, and finally the practical application of the model. In clinical practice, such a model can help doctors identify high-risk patients and take preventive and treatment measures in time.

[0091] Probability value: that is, output a probability value between 0 and 1, indicating the patient's risk of stroke. This probability value can help doctors assess the degree of risk, rather than simply "high" or "low"; values ​​close to 0: indicate a very low risk; values ​​close to 1: indicate a very high risk; intermediate values: indicate a medium risk; probability values ​​can provide a more detailed risk assessment and help doctors make more accurate decisions;

[0092] The process of analyzing the patient's brain image data is as follows:

[0093] Use convolutional neural networks (CNNs) to analyze image data, i.e., brain MRI images of patients: collect and annotate a large number of brain MRI images, dividing them into two categories: normal and abnormal (with lesions); build a convolutional neural network model, including convolutional layers, pooling layers, and fully connected layers; use the annotated dataset, i.e., a large number of annotated brain MRI images, to train the model and adjust the model parameters to optimize the classification performance; evaluate the performance of the model on an independent test set to ensure that it can accurately identify cerebrovascular lesions;

[0094] When an abnormality is determined, the abnormal area and location are obtained. The process is as follows:

[0095] Under the condition of determining the abnormality, the image processing algorithm is used to extract the lesion features from the MRI image, and the position and area of ​​the abnormality are extracted and obtained. Specifically, the lesion features are extracted from the MRI image using OpenCV and scikit-image libraries, and the lesion features include the position and area of ​​the abnormality;

[0096] Image processing example:

[0097] import cv2

[0098] from skimage import io, color, filters

[0099] import numpy as np

[0100] import matplotlib.pyplot as plt

[0101] # Read MRI images

[0102] image = io.imread('mri_image.jpg')

[0103] # Image preprocessing: grayscale

[0104] gray_image = color.rgb2gray(image)

[0105] # Image preprocessing: denoising

[0106] smoothed_image = filters.gaussian(gray_image, sigma=1)

[0107] # Image segmentation: Using Otsu thresholding method

[0108] thresh_val = filters.threshold_otsu(smoothed_image)

[0109] binary_image = smoothed_image>thresh_val

[0110] # Display preprocessing and segmentation results

[0111] plt.figure(figsize=(12, 6))

[0112] plt.subplot(1, 3, 1)

[0113] plt.title('Original Image')

[0114] plt.imshow(image, cmap='gray')

[0115] plt.subplot(1, 3, 2)

[0116] plt.title('Smoothed Image')

[0117] plt.imshow(smoothed_image, cmap='gray')

[0118] plt.subplot(1, 3, 3)

[0119] plt.title('Binary Image')

[0120] plt.imshow(binary_image, cmap='gray')

[0121] plt.show()

[0122] # Feature extraction: Shape feature example (area, perimeter)

[0123] contours,_=cv2.findContours((binary_image*255).astype(np.uint8),cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

[0124] for contour in contours:

[0125] area = cv2.contourArea(contour)

[0126] perimeter = cv2.arcLength(contour, True)

[0127] print(f'Area: {area}, Perimeter: {perimeter}');

[0128] The risk assessment module calculates the patient's condition estimate based on the dimensionless probability value of the corresponding patient when the image analysis result is normal. When the image analysis result is abnormal, the patient's condition estimate is calculated based on the dimensionless probability value and abnormal area of ​​the corresponding patient. , and the patient's condition is estimated The larger it is, the more abnormal the patient's condition is;

[0129] Among them, calculate the estimated value of the patient's condition The formula is as follows:

[0130] ;

[0131] In formula (1), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is normal. is a constant correction coefficient, the specific value of which can be adjusted by the user or generated by analytical function fitting, and The value range is 1~2. is the probability value of the corresponding patient's image analysis result being normal;

[0132] In formula (2), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is abnormal. is a constant correction coefficient, the specific value of which can be adjusted by the user or generated by analytical function fitting, and The value range is 1~2. is the probability value of abnormal image analysis result of the corresponding patient, represents the abnormal area, are the probability value and the preset proportional coefficient of the abnormal area when the image analysis result of the corresponding patient is abnormal, respectively, and .

[0133] Frequency detection prediction module, which estimates the patient's condition With the preset evaluation threshold The comparison results are as follows:

[0134] Estimated patient condition The assessment threshold is not exceeded Under the condition of , it means that the corresponding patient's condition is normal, no response action is taken, and the established adjustment strategy is executed; Exceeding the evaluation threshold Under the condition of , it indicates that the corresponding patient's condition is serious, a first-level warning signal is issued, and the frequency detection adjustment strategy is executed;

[0135] Among them, the implementation of the established adjustment strategy is: the corresponding patient goes to the hospital for regular examination according to the established examination frequency T;

[0136] The process of executing the frequency detection adjustment strategy is as follows:

[0137] When a level 1 warning signal is received, the patient's condition is estimated and evaluation threshold , calculate the predicted inspection frequency , and check the frequency prediction value The formula is as follows:

[0138] ;

[0139] In the formula, represents the error correction factor, and The value range is 0~1;

[0140] Corresponding patients according to the predicted value of examination frequency , go to the hospital for regular check-ups;

[0141] For example, a patient's scheduled examination frequency T is once every 3 days.

[0142] Specifically, according to the different image analysis results, the patient's condition estimate is calculated based on the data parameters under the corresponding conditions. , can realize targeted condition prediction and treatment, further ensure the accuracy of the patient's condition prediction, compare the patient's condition prediction value with the preset evaluation threshold, execute the corresponding strategy according to the comparison result, and execute the frequency adjustment strategy under the condition that the patient's condition is judged to be serious. According to the patient's condition prediction value and the evaluation threshold, the inspection frequency prediction value is calculated to increase the inspection density, thereby realizing targeted management of the patient's condition and improving the patient's safety.

[0143] Example 2: Please refer to Figure 2 Based on Example 1, this embodiment also provides a method for constructing a dynamic prediction model for cerebrovascular disease based on AI, comprising the following specific steps:

[0144] S1. Obtain and summarize the historical data and image data of each patient;

[0145] S2, preprocessing the acquired historical data and image data;

[0146] S3. Use AI technology to predict the risk of stroke in patients and analyze the image data of the patient's brain, output probability values ​​and image analysis results, and the image analysis results include normal and abnormal, and obtain the abnormal area and location when it is determined to be abnormal;

[0147] S4. Under the condition that the image analysis result is normal, calculate the estimated value of the patient's condition based on the probability value of the corresponding patient after dimensionless processing When the image analysis result is abnormal, the patient's condition estimate is calculated based on the dimensionless probability value and abnormal area of ​​the corresponding patient. ;

[0148] Among them, calculate the estimated value of the patient's condition The formula is as follows:

[0149] ;

[0150] In formula (1), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is normal. is a constant correction factor, and The value range is 1~2. is the probability value of the corresponding patient's image analysis result being normal;

[0151] In formula (2), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is abnormal. is a constant correction factor, and The value range is 1~2. is the probability value of abnormal image analysis result of the corresponding patient, represents the abnormal area, are the probability value and the preset proportional coefficient of the abnormal area when the image analysis result of the corresponding patient is abnormal, respectively, and .

[0152] S5. Estimate the patient's condition With the preset evaluation threshold Make a comparison;

[0153] Estimated patient condition The assessment threshold is not exceeded Under the condition of the patient's condition, no response action is taken and the established adjustment strategy is implemented; Exceeding the evaluation threshold Under the condition of , a first-level warning signal is issued and the frequency detection adjustment strategy is implemented;

[0154] Among them, the implementation of the established adjustment strategy is:

[0155] The corresponding patients go to the hospital for regular check-ups according to the established check-up frequency T;

[0156] The process of executing the frequency detection adjustment strategy is as follows:

[0157] When a level 1 warning signal is received, the patient's condition is estimated and evaluation threshold , calculate the predicted inspection frequency , and check the frequency prediction value The formula is as follows:

[0158] ;

[0159] In the formula, represents the error correction factor, and The value range is 0~1;

[0160] Corresponding patients according to the predicted value of examination frequency , go to the hospital for regular check-ups.

[0161] In the application, the several formulas involved are all calculated by removing the dimensions and taking their numerical values. The formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The formula is set by technical personnel in this field according to actual conditions.

[0162] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0163] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. The AI-based cerebrovascular disease dynamic prediction model construction system is characterized by: include: Information aggregation module, which obtains and aggregates the historical data and image data of each patient; A data processing module pre-processes the acquired historical data and image data; The learning prediction module uses AI technology to predict the risk of stroke in patients and analyze the image data of the patient's brain, outputs the probability value and image analysis results, and the image analysis results include normal and abnormal, and obtains the abnormal area and location when it is determined to be abnormal; The risk assessment module calculates the patient's condition estimate based on the dimensionless probability value of the corresponding patient when the image analysis result is normal. , when the image analysis result is abnormal, the patient's condition estimate is calculated based on the dimensionless probability value and abnormal area of ​​the corresponding patient ; Frequency detection prediction module, which estimates the patient's condition With the preset evaluation threshold Make a comparison and implement corresponding adjustment strategies based on the comparison results; The AI ​​technology includes a random forest algorithm and a convolutional neural network; the process of predicting stroke risk using the random forest algorithm is as follows: Extract medical examination data from the corresponding patient's historical data, perform feature selection and label definition before splitting the data, and divide the data into training set and test set. Use Python's scikit-learn library to build a random forest model, use the random forest algorithm to train the model on the training set, adjust the model parameters through cross-validation and grid search, deploy the trained model to the clinical system, and output the classification results, which are displayed in the form of probability values. The process of analyzing the patient's brain image data is as follows: Use convolutional neural networks to analyze image data, collect and annotate several brain MRI images, classify them into normal and abnormal categories, build a convolutional neural network model, including convolutional layers, pooling layers, and fully connected layers, use annotated data sets to train the model, adjust model parameters to optimize classification performance, and output image analysis results; When an abnormality is determined, the abnormal area and location are obtained. The process is as follows: Under the condition of determining the abnormality, using image processing algorithms to extract lesion features from MRI images, extract and obtain the location and area of ​​the abnormality, and the image processing algorithms include using OpenCV and scikit-image libraries; Calculate the estimated patient condition The formula is as follows: ； In formula (1), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is normal, is a constant correction factor, and The value range is 1~2. is the probability value of the corresponding patient's image analysis result being normal; In formula (2), the estimated value of the patient's condition is calculated The prerequisite is that the image analysis result is abnormal. is a constant correction factor, and The value range is 1~2. is the probability value of abnormal image analysis result of the corresponding patient, represents the abnormal area, are the probability value and the preset proportional coefficient of the abnormal area when the image analysis result of the corresponding patient is abnormal, respectively, and .

2. The AI-based cerebrovascular disease dynamic prediction model construction system according to claim 1, characterized in that: The historical data includes demographic characteristics, medical examination data and medical history data. The demographic characteristics include age, gender and race. The medical examination data include blood pressure, blood sugar, cholesterol and body mass index. The medical history data includes the history of hypertension, diabetes and heart disease, that is, whether there is a corresponding medical history. The image data includes several MRI images of the corresponding patient's brain.

3. The AI-based cerebrovascular disease dynamic prediction model construction system according to claim 2, characterized in that: The process of preprocessing the acquired historical data and image data is: data cleaning and data standardization.

4. The AI-based cerebrovascular disease dynamic prediction model construction system according to claim 1, characterized in that: The process of implementing corresponding adjustment strategies based on the comparison results is as follows: Estimated patient condition The assessment threshold is not exceeded Under the condition of the patient's condition, no response action is taken and the established adjustment strategy is implemented; Exceeding the evaluation threshold Under the condition of , a first-level warning signal is issued and the frequency detection adjustment strategy is implemented; Among them, the implementation of the established adjustment strategy is: The corresponding patients go to the hospital for regular check-ups according to the established check-up frequency T; The process of executing the frequency detection adjustment strategy is as follows: When a level 1 warning signal is received, the patient's condition is estimated and evaluation threshold , calculate the predicted inspection frequency , and check the frequency prediction value The formula is as follows: ； In the formula, represents the error correction factor, and The value range is 0~1; Corresponding patients according to the predicted value of examination frequency , go to the hospital for regular check-ups.

5. A method for constructing a dynamic prediction model for cerebrovascular disease based on AI, using any system described in claims 1 to 4, characterized in that: The steps include: S1. Obtain and summarize the historical data and image data of each patient; S2, preprocessing the acquired historical data and image data; S3. Use AI technology to predict the risk of stroke in patients and analyze the image data of the patient's brain, output probability values ​​and image analysis results, and the image analysis results include normal and abnormal, and obtain the abnormal area and location when it is determined to be abnormal; S4. Under the condition that the image analysis result is normal, calculate the estimated value of the patient's condition based on the probability value of the corresponding patient after dimensionless processing , when the image analysis result is abnormal, the patient's condition estimate is calculated based on the dimensionless probability value and abnormal area of ​​the corresponding patient ; S5. Estimate the patient's condition With the preset evaluation threshold Make a comparison and implement corresponding adjustment strategies based on the comparison results.

Images