Method for constructing cerebrovascular and brain function disease prediction model by applying big data and AI technology

Abstract

The invention discloses a method for constructing a cerebrovascular and brain function disease prediction model by using big data and an AI technology, and relates to the field of artificial intelligence enabled medical health, and the method integrates an advanced medical imaging technology, an artificial intelligence driven quantitative analysis method and a large-scale multi-center medical big data resource. The method aims to carry out early accurate screening, risk stratification and individualized intervention on high-risk people suffering from cerebrovascular and brain function diseases. A brain disease risk prediction model based on an error back-propagation neural network is constructed by deeply fusing clinical medical history, lifestyle, blood biochemical indexes, radiomics characteristics of carotid artery and intracranial vascular plaques of a subject and functional connection indexes of brain volume and a brain network. And a technical support is provided for realizing accurate prevention and individualized treatment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence-enabled medical health, and specifically to a method for using big data and AI technology to construct a prediction model for cerebrovascular and brain function diseases. Background Art

[0002] Currently, cerebrovascular and brain functional diseases are among the major health threats posing a serious threat to public health, and the threat they pose is increasing. Cerebrovascular diseases primarily include cerebral infarction, aneurysms, and vascular lesions, and they primarily occur in middle-aged and elderly individuals. Studies have shown that nearly one-third of deaths worldwide are related to cerebrovascular disease, primarily due to stroke caused by ruptured atherosclerotic plaques. Currently, stroke remains highly prevalent in my country, resulting in a significant disease burden. A 2013 special survey on cerebrovascular disease epidemiology in my country revealed that at any given time, there are approximately 11 million stroke survivors nationwide, over 2.4 million first-time stroke patients each year, and over 1.1 million stroke deaths. These data indicate that the prevention and control of cerebrovascular diseases remains critical. Data from recent domestic epidemiological surveys have also shown that there are approximately 270 million patients with hypertension, 110 million with diabetes, and 160 million with dyslipidemia in my country, and the number of these high-risk individuals for cerebrovascular disease continues to rise. The accumulation of these risk factors further increases the risk of cerebrovascular disease. Brain functional diseases mainly include Alzheimer's disease, Parkinson's disease, depression, and cranial nerve diseases, which are characterized by long-term chronic development and irreversible neurological damage. Among them, the progression of many neurological diseases is accompanied by atrophy of different structural areas of the brain. For neurodegenerative diseases, this atrophy is a gradual process. Therefore, it is difficult to achieve early identification and accurate diagnosis of the disease through human visual observation or image estimation and measurement. Currently, the incidence of brain functional diseases is increasing year by year. These data show that the prevention and control of brain functional diseases is a very arduous task and seriously affects the quality of life of the general public.

[0003] Currently, clinical data such as basic physical characteristics, blood biochemical markers, and metabolic levels are generally easy to obtain. However, the collection of quantitative organ imaging indicators is often limited by the current state of medical care and the level of clinical diagnostic technology. Furthermore, clinical practice primarily focuses on one-sided information from a single examination site, rarely extracting functional characteristics of organs or lesions based on imaging. Therefore, cardiovascular examinations, whether ultrasound, CT, or MRI, currently focus primarily on the morphological structure of the lesion itself, particularly its geometry. This may be due to two factors: first, extracting functional features of vascular plaques is time-consuming and complex; second, effective tools for extracting these quantifiable features are lacking. Taking atherosclerotic plaques as an example, atherosclerotic stenosis is a recognized risk factor for ischemic stroke, accounting for 10%-20% of strokes or transient ischemic attacks. Currently, in clinical practice, the degree of luminal stenosis based on angiography (ultrasound Doppler or CT) is the primary basis for assessing lesion risk. However, clinical trials have shown that the majority of clinical events occur in patients with mild to moderate vascular stenosis, and that surgical benefit is low in patients with stenosis >70%. Therefore, solely studying the degree of luminal stenosis in the diagnosis of cerebrovascular disease is clearly insufficient for assessing disease characteristics and preventing stroke. For example, in predicting the risk of brain function disorders, the development and progression of various neurological diseases is accompanied by atrophy of various brain structures, such as hippocampal volume, ventricular volume, lateral ventricular volume, and cerebral lobe volume. Brain volume is a rapid, effective, and noninvasive measurement indicator. Therefore, the brain volume of the lesion area, combined with the subject's basic physical information and blood biochemical parameters, can be used to effectively assess the risk of brain function disorders.

[0004] With the rapid development of medicine and computing, artificial intelligence has begun to influence all areas of the medical industry. "Artificial intelligence + imaging medicine" has achieved image recognition performance that surpasses traditional methods, providing new opportunities for early diagnosis and precise treatment of diseases. Therefore, the use of artificial intelligence and deep learning algorithms to establish a whole-brain quantitative analysis tool for neuroimaging, as well as a high-resolution magnetic resonance imaging vascular plaque quantitative analysis tool, can provide accurate quantitative indicators of organ function. Based on artificial intelligence and medical big data technology, with the quantitative assessment of atherosclerotic plaques, brain region segmentation, and quantitative assessment of brain atrophy as the entry point, combined with basic physical information and blood index information, it can provide a rapid and intelligent method for clinically accurately assessing the risk of cerebrovascular and brain function diseases. Summary of the Invention

[0005] The purpose of this invention is to provide a method for constructing a prediction model for cerebrovascular and brain functional diseases using big data and AI technology. By combining the subject's basic physical information, blood biochemical indicators, carotid artery and intracranial vascular plaques, and functional quantitative indicators of brain volume, a disease risk prediction system is provided based on the BP neural network model to accurately screen, risk stratify, and individualize interventions for people at high risk of cerebrovascular and brain functional diseases.

[0006] The invention objectives of the present invention are mainly achieved through the following technical solutions: a method for constructing a prediction model for cerebrovascular and brain functional diseases using big data and AI technology, including a method for constructing a cerebrovascular disease risk assessment model and a method for constructing a brain functional disease risk assessment model;

[0007] The method for constructing the cerebrovascular disease risk assessment model comprises the following specific steps:

[0008] Step 1: Analyze the contrast MRI imaging data of the patient's vascular plaques to achieve quantitative analysis of vascular morphology and plaque composition, and provide a plaque vulnerability assessment; the plaque vulnerability assessment includes assessment of plaque vulnerability properties, vulnerability types, vulnerability levels, AHA plaque classification, and quantitative data of various components within the plaque; the display results of the plaque vulnerability assessment include detailed reports of typical layers, 3D reconstructions, and fused images;

[0009] Step 2: Extract cerebral vascular analysis parameters based on the image processing technology in step 1;

[0010] Step 3: Based on the ADL scale, simple physical examination, blood biochemical indicators, genetics and other test results, combined with clinical diagnosis, cerebrovascular disease is graded and grouped;

[0011] Step 4: Use the random forest algorithm in machine learning to screen the main causes of disease after hierarchical grouping in Step 3. Rank feature importance using OOB error estimation. Use the Gini impurity reduction quantification to quantify feature contribution, and set a threshold to retain the top 30% of key predictors (such as hippocampal volume and default mode network connectivity). Finally, construct a BP neural network algorithm model for important variables that contribute to cerebrovascular disease risk to predict the individual's probability of cerebrovascular disease.

[0012] The method for constructing the brain function disease risk assessment model comprises the following specific steps:

[0013] Step 1: Using voxel-based morphological measurements, the method mainly includes the following steps: spatial normalization, brain tissue segmentation, image smoothing, statistical analysis, and generation of overlay images. Spatial normalization refers to the need to spatially normalize brain structure images of different subjects due to differences in brain volume between different subjects, thereby reducing the impact of differences in brain volume between individuals on experimental results. Brain tissue segmentation involves applying an artificial intelligence algorithm to segment the standardized image into gray matter, white matter, and cerebrospinal fluid. The purpose of image smoothing is to reduce noise in the segmented image, improve the image signal-to-noise ratio and image quality, and then perform statistical analysis on the smoothed image data. The obtained experimental data is then overlaid on the standardized brain image to obtain an overlay image expressed by statistical stars for accurate assessment of brain atrophy and cognitive impairment.

[0014] Step 2: Based on the image processing technology used in Step 1, brain disease risk assessment is performed. To construct a brain disease risk assessment model, the subjects' Geriatric Depression Scale (GDS), Mini-Mental State Examination (MMSE), and Hasegawa Dementia Rating Scale (HDS) are first collected as part of a simple physical examination, along with blood biochemical parameters and brain structural parameters extracted through visual brain machine learning (VBM). Next, a random forest algorithm is used for feature selection. The random forest algorithm constructs multiple decision trees using bootstrap sampling, and the average information gain of each feature across all decision trees is evaluated to select the most important features. Finally, a BP neural network model is constructed for the selected important variables. The BP neural network model calculates the output value through forward propagation, adjusts the weights and bias through backpropagation of the error, and iterates training until the error reaches a preset threshold, thereby predicting the probability of brain disease in individuals. Random forests, as an ensemble learning method, can effectively reduce the risk of overfitting. The BP neural network, as a multi-layer feedforward network, can approximate complex nonlinear functions, enabling accurate prediction of brain disease.

[0015] Furthermore, the image processing technology in step one is as follows: first, image preprocessing is performed, including adaptive filtering noise reduction, motion artifact correction and N4ITK bias field correction, to improve image quality; then, multi-scale Hessian filtering is applied to enhance the vascular structure, and deep learning segmentation technology (such as V-Net) is used to accurately extract the three-dimensional model of cerebral blood vessels, and perform topological structure correction; for the lumen, wall and plaque components, multi-sequence magnetic resonance image information such as T1WI and T2WI is integrated, and deformation models or deep learning methods are used for accurate segmentation to identify plaque components such as lipid core (high signal T1WI), fibrous cap (low signal T1 / T2WI), and calcification (low signal in each sequence); according to the AHA guidelines, the various components in the plaque are quantitatively analyzed, and vulnerability indicators such as lipid core area and fibrous cap thickness are calculated to evaluate the vulnerability properties of the plaque (such as thin fibrous cap), vulnerability type (such as lipid-rich core) and vulnerability degree (low, medium and high risk); finally, Marching is used The Cubes algorithm performs three-dimensional reconstruction, fusing the three-dimensional models of blood vessels, lumens, walls, and plaque components to generate a complete three-dimensional structure of cerebral blood vessels and present it visually. The results are presented in the form of detailed reports of typical levels, three-dimensional reconstructions, and fused images, providing an intuitive basis for plaque vulnerability assessment.

[0016] Furthermore, the image processing technology in step one also includes medical image denoising, which is a medical image denoising technology based on an improved non-local means (NLM) filtering algorithm, which is used to effectively suppress the noise introduced during the medical image acquisition process. This method innovatively combines an improved Facet operator and a region-adaptive self-similarity calculation strategy. Specifically, the improved Facet operator is first used to accurately extract the edge features of the image, and different self-similarity measurement functions are adaptively selected based on the local statistical characteristics of the pixel distribution within the image; at the same time, a variable-size search window mechanism is adopted to maximize the search range of similar neighborhoods and reduce the interference of noise on the self-similarity evaluation, thereby significantly improving the ability to preserve the edge structure of the image while effectively removing noise.

[0017] The beneficial effects of the present invention are:

[0018] 1. The BP neural network model constructed in this study has good classification / prediction performance and can accurately predict the occurrence of cerebral ischemic symptom outcomes based on 20 characteristic independent variables collected clinically, including gender, age group, smoking, drinking, hypertension, high-density cholesterol, vascular stenosis and plaque length.

[0019] 2. The BP neural network model constructed in this study can predict individuals who will develop cerebral ischemic symptoms and reduce their exposure to risk factors (quitting smoking, drinking, etc.) in a timely manner. At the same time, it can strengthen protective measures and achieve etiology prevention, aiming to prevent the occurrence of cerebral ischemic symptoms and achieve the effect of precise prevention and control.

[0020] 3. The BP neural network model constructed in this study increases the frequency of clinical follow-up monitoring of people at high risk of cerebral ischemic symptoms, and can achieve secondary prevention of early detection, early diagnosis and early treatment.

[0021] 4. The BP neural network model constructed in this study predicts that the frequency of active health monitoring of individuals who will not develop cerebral ischemic symptoms can be appropriately reduced, and limited medical resources can be mainly used for people who are predicted to be at high risk of cerebral ischemic symptoms. Ultimately, the precise allocation of medical resources can be achieved, and precise prevention and control of cerebral ischemic symptoms can be carried out to achieve maximum health economic benefits.

[0022] 5. This system integrates advanced medical imaging technology, AI-driven quantitative analysis methods, and large-scale, multi-center medical big data resources. It aims to provide early, accurate screening, risk stratification, and personalized intervention for individuals at high risk for cerebrovascular and functional brain diseases. By deeply integrating the subject's clinical history, lifestyle, blood biochemical indicators, imaging features of carotid and intracranial vascular plaques, and brain volume and functional connectivity indicators, a brain disease risk prediction model based on a back propagation neural network (BP neural network) is constructed, providing technical support for precise prevention and personalized treatment. DETAILED DESCRIPTION

[0023] The BP neural network is one of the most common neural network models and a commonly used deep learning model. It has the following advantages: 1. The BP neural network model can process a variety of data types, including numerical, categorical, and sequential data. It can handle nonlinear problems because its neuron activation functions can be nonlinear. Mathematical theory proves that a three-layer neural network can approximate any nonlinear continuous function with arbitrary accuracy, making it particularly suitable for solving problems with complex internal mechanisms. 2. The BP neural network model can automatically learn and extract features from input data, eliminating the need for manual feature engineering. It abstracts the input data layer by layer to obtain higher-level feature representations. 3. During training, the BP neural network can automatically extract "reasonable rules" between input and output data through learning and adaptively memorize these learned features in the network weights. This means that BP neural networks have a high degree of self-learning and self-adaptation capabilities. 4. The BP neural network has the ability to apply learning results to new knowledge. That is, after training, the network can accurately classify unseen patterns or patterns contaminated by noise. 5. The BP neural network model can explain the model's decision-making process through weights and bias values, help understand the model's importance to different features, and perform feature selection and ranking.

[0024] Research content:

[0025] 1. Construction of a cerebrovascular disease risk assessment model

[0026] Contrast-contrast MRI data of vascular plaques from patients were analyzed. Image preprocessing, including adaptive filtering for noise reduction, motion artifact correction, and N4ITK bias field correction, was performed to enhance image quality. Subsequently, multiscale Hessian filtering was applied to enhance vascular structure. Deep learning segmentation techniques (such as V-Net) were used to accurately extract 3D models of cerebral vessels and perform topological correction. Deformable models or deep learning methods were used to accurately segment the lumen, wall, and plaque components by integrating information from multiple MRI sequences, including T1WI and T2WI. These techniques identified plaque components such as the lipid core (high signal intensity on T1WI), fibrous cap (low signal intensity on T1 / T2WI), and calcification (low signal intensity on all sequences). Based on the American Heart Association (AHA) guidelines, plaque components were quantified and analyzed, and vulnerability indicators such as lipid core area and fibrous cap thickness were calculated. Plaque vulnerability was assessed for nature (e.g., thin fibrous cap), vulnerability type (e.g., lipid-rich core), and vulnerability severity (low, intermediate, or high risk). Finally, a Marching Cubes algorithm is used for 3D reconstruction, fusing 3D models of vessels, lumen, wall, and plaque components to generate a complete 3D cerebral vascular structure for visualization. Results are presented as detailed reports of representative layers, 3D reconstructions, and fused images, providing an intuitive basis for plaque vulnerability assessment. Using these image processing techniques, cerebrovascular analysis parameters such as vessel diameter, degree of lumen stenosis, plaque volume, and component ratios are extracted. Quantitative data for each plaque component is classified according to the American Heart Association (AHA) guidelines. Results are displayed in detailed reports of representative layers, 3D reconstructions, and fused images. Using these image processing techniques, cerebrovascular analysis parameters are extracted. Cerebrovascular disease risk assessment: Cerebrovascular disease is graded based on the ADL scale, a simple physical examination, blood biochemical markers, genetic testing, and other test results, combined with clinical diagnosis. First, the random forest algorithm in machine learning was used to screen these major causes. Feature importance was ranked using out-of-band error estimation. Feature contribution was quantified using the Gini impurity reduction metric, and a threshold was set to retain the top 30% of key predictors (such as hippocampal volume and default mode network connectivity). Finally, a BP neural network algorithm model was constructed for variables that were important for cerebrovascular disease risk, predicting the probability of cerebrovascular disease in individuals.

[0027] 2. Construction of a risk assessment model for brain function diseases

[0028] Voxel-based morphometric measurement, a volumetric measurement method, primarily involves the following steps: spatial normalization, brain tissue segmentation, image smoothing, statistical analysis, and generation of overlay images. Spatial normalization involves spatially normalizing brain structural images due to differences in brain volume between subjects, thereby minimizing the impact of inter-individual brain volume variations on experimental results. Brain tissue segmentation involves applying artificial intelligence algorithms to segment the normalized images into gray matter, white matter, and cerebrospinal fluid. Image smoothing aims to reduce noise in the segmented images, improving the signal-to-noise ratio and image quality. Statistical analysis is then performed on the smoothed image data, and the resulting experimental data is then overlaid onto the normalized brain images to create an overlay image represented by statistical stars, which is used to accurately assess brain atrophy and cognitive impairment. Based on these image processing techniques, brain disease risk assessment is performed. To construct a brain disease risk assessment model, participants' Geriatric Depression Scale (GDS), Mini-Mental State Examination (MMSE), and Hasegawa Dementia Rating Scale (HDS) were first collected, along with blood biochemical parameters and brain structural parameters extracted through visual brain machine learning (VBM) analysis. Next, a random forest algorithm was used for feature screening. This algorithm constructs multiple decision trees through bootstrap sampling and evaluates the average information gain of each feature across all decision trees, selecting the most important features. Finally, a BP neural network model was constructed for the selected important variables. This model calculates the output value through forward propagation, adjusts the weights and bias through backpropagation of the error, and iterates training until the error reaches a preset threshold, thereby predicting the probability of brain function disorders in individuals. Random forests, as an ensemble learning method, can effectively reduce the risk of overfitting; BP neural networks, as multi-layer feedforward networks, can approximate complex nonlinear functions, enabling accurate prediction of brain function disorders.

[0029] Example 1

[0030] Medical image data acquisition: Based on the center's independent MRI scan sequence parameters, vascular atherosclerotic plaque images and brain volume images are collected.

[0031] Medical image denoising: A medical image denoising technique based on an improved non-local means (NLM) filtering algorithm is used to effectively suppress noise introduced during the medical image acquisition process. This method innovatively combines an improved Facet operator with a region-adaptive self-similarity calculation strategy. Specifically, the improved Facet operator is first used to accurately extract image edge features, and different self-similarity metric functions are adaptively selected based on the local statistical characteristics of the pixel distribution within the image. At the same time, a variable-size search window mechanism is adopted to maximize the search range of similar neighborhoods and reduce the interference of noise on the self-similarity assessment. This effectively removes noise while significantly improving the ability to preserve image edge structure.

[0032] Plaque component identification: Based on machine learning and combining the characteristics of magnetic resonance image histograms and edge gradients, reasonable data augmentation is performed by adding Gaussian noise, random brightness adjustment, random image blurring, and other methods to the magnetic resonance images to achieve data balance and improve the stability of the training model. Utilizing a parallel branch plaque target detection algorithm based on a deep convolutional neural network, feature maps are extracted from the horizontal and vertical sections of the magnetic resonance images using the parallel branches of two pre-trained convolutional neural networks. Classification and positioning are completed based on the plaque area features in the feature maps. During the network training process, batch regularization and residual neural networks are used to improve the network training effect. Finally, the rectangular area of ​​the magnetic resonance image plaque is obtained according to the deep learning target detection algorithm. The magnetic resonance image plaque area is further instantiated and segmented to identify the plaque components.

[0033] Cerebrovascular disease risk assessment: Cerebrovascular disease was stratified based on baseline data (gender, age group, smoking, alcohol consumption, hypertension, high-density lipoprotein cholesterol, vascular stenosis, and plaque morphology), combined with clinical diagnosis. To construct a cerebrovascular disease prediction model based on these grouping results, this study first conducted a univariate analysis to identify candidate variables significantly associated with cerebrovascular disease. To mitigate the impact of multicollinearity on model stability, the variance inflation factor (VIF) was used to diagnose collinearity. Variables with VIF values ​​greater than 4 were removed, and the remaining variables were retained. Subsequently, univariate logistic regression analysis was performed on these variables to identify protective and risk factors for cerebrovascular disease. Based on this data, a BP neural network model was constructed. This model is a multilayer feedforward neural network that uses a backpropagation algorithm to adjust network weights to minimize error. The model was configured with 10 hidden nodes, a maximum number of iterations of 1×10^8, an error threshold of 1×10^4, and a random seed of 123 to ensure reproducibility. Finally, the original data were randomly divided into a training set (70%) and a test set (30%), which were used for the construction and verification of the BP neural network model, respectively, to evaluate the model's ability to predict the outcomes of cerebrovascular diseases.

[0034] Brain region segmentation: Based on the risk prediction model for plaque component identification, brain magnetic resonance images are studied and a brain region segmentation model is trained. Finally, the deep learning target detection algorithm is used to obtain rectangular regions of brain functional areas in the magnetic resonance images. The brain functional areas in the magnetic resonance images are further instantiated and segmented, resulting in 96 finely segmented brain regions.

[0035] Risk Assessment for Brain Function Disorders: To construct a precise risk assessment model for brain function disorders, this study first categorized participants' brain function disorders based on baseline data, including GDS, MMSE, HDS scores, physical examination results, biochemical markers, and clinical diagnoses. Subsequently, univariate analysis was used to preliminarily identify candidate variables significantly associated with brain function disorders. To ensure model stability and avoid multicollinearity, the variance inflation factor (VIF) was used to diagnose collinearity. Variables with VIF values ​​greater than a set threshold (e.g., 4 or 5) were removed, and the remaining variables were retained for subsequent analysis. To further clarify the relationship between each variable and the risk of brain function disorders, univariate logistic regression analysis was performed to identify protective and risk factors, and the corresponding odds ratios (ORs) and 95% confidence intervals were calculated. Based on this, a multilayer perceptron (MLP) neural network model based on the backpropagation (BP) algorithm was constructed. This model has strong nonlinear fitting capabilities and can effectively capture complex data relationships. During model construction, grid search or Bayesian optimization methods were used to optimize model hyperparameters, such as the number of hidden layers, the number of neurons per layer, the learning rate, the activation function (e.g., ReLU, Sigmoid), and the regularization parameter, to improve model generalization. This study used a single hidden layer with 10 neurons, a maximum number of iterations of 1×10^8, an error threshold of 1×10^4, and a random seed of 123 to ensure reproducibility. To evaluate the predictive performance of the model, the original dataset was randomly divided into a training set (70%) and a test set (30%). The training set was used for model parameter learning, and the test set was used to evaluate the model's generalization ability. The model's predictive performance for brain disease risk was comprehensively evaluated using metrics such as the area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, and specificity. K-fold cross-validation and other methods were also used to further verify the model's stability and reliability.

[0036] Establish a prediction mechanism for cerebrovascular diseases and brain functional diseases: By establishing an effective risk prediction model for cerebrovascular diseases and brain functional diseases, in the early stages of the disease, through the best intervention methods and diagnosis and treatment mechanisms, we can achieve efficient and accurate prevention and stratified management of brain functional diseases and cerebrovascular diseases, and effectively reduce the risk of cerebrovascular and brain functional diseases.

[0037] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for constructing a prediction model for cerebrovascular and brain functional diseases using big data and AI technology, characterized by: Including the construction method of cerebrovascular disease risk assessment model and the construction method of brain function disease risk assessment model; The method for constructing the cerebrovascular disease risk assessment model comprises the following specific steps: Step 1: Analyze the contrast MRI imaging data of the patient's vascular plaques to achieve quantitative analysis of vascular morphology and plaque composition, and provide a plaque vulnerability assessment; the plaque vulnerability assessment includes assessment of plaque vulnerability properties, vulnerability types, vulnerability levels, AHA plaque classification, and quantitative data of various components within the plaque; the display results of the plaque vulnerability assessment include detailed reports of typical layers, 3D reconstructions, and fused images; Step 2: Extract cerebral vascular analysis parameters based on the image processing technology in step 1; Step 3: Cerebral vascular disease is graded and grouped based on the ADL scale, simple physical examination, blood biochemical indicators, and genetic test results; Step 4: Use the random forest algorithm in machine learning to screen the main causes of disease after hierarchical grouping in Step 3. Rank feature importance using OOB error estimation. Use the Gini impurity reduction metric to quantify feature contribution, and set a threshold to retain the top 30% of key predictors. Finally, construct a BP neural network algorithm model for important variables that contribute to cerebrovascular disease risk to predict the probability of cerebrovascular disease in individuals. The method for constructing the brain function disease risk assessment model comprises the following specific steps: Step 1: Using voxel-based morphological measurements, the method mainly includes the following steps: spatial normalization, brain tissue segmentation, image smoothing, statistical analysis, and generation of overlay images. Spatial normalization refers to the need to spatially normalize brain structure images of different subjects due to differences in brain volume between different subjects, thereby reducing the impact of differences in brain volume between individuals on experimental results. Brain tissue segmentation involves applying an artificial intelligence algorithm to segment the standardized image into gray matter, white matter, and cerebrospinal fluid. The purpose of image smoothing is to reduce noise in the segmented image, improve the image signal-to-noise ratio and image quality, and then perform statistical analysis on the smoothed image data. The obtained experimental data is then overlaid on the standardized brain image to obtain an overlay image expressed by statistical stars for accurate assessment of brain atrophy and cognitive impairment. Step 2. Based on the image processing technology in step 1, perform risk assessment of brain function diseases: To construct a risk assessment model for brain function diseases, first collect the subjects' Geriatric Depression Scale, Mini-Mental State Examination Scale, Hasegawa Dementia Scale simple physical examination and blood biochemical indicators, as well as brain structure parameters extracted by VBM analysis; then, use the random forest algorithm for feature screening. The random forest algorithm constructs multiple decision trees through Bootstrap sampling, and evaluates the average information gain of each feature in all decision trees, and selects features with higher importance; finally, construct a BP neural network model for the important variables screened out. The BP neural network model calculates the output value through forward propagation, adjusts the weight and bias through backpropagation error, and iterates training until the error reaches the preset threshold, thereby predicting the probability of individual brain function diseases.

2. The method for constructing a prediction model for cerebrovascular and brain functional diseases using big data and AI technology according to claim 1, characterized in that: The image processing techniques in step 1 are as follows: First, image preprocessing is performed, including adaptive filtering noise reduction, motion artifact correction, and N4ITK bias field correction, to improve image quality. Subsequently, multiscale Hessian filtering is applied to enhance vascular structure, and deep learning segmentation technology is used to accurately extract the 3D model of the cerebral vasculature and perform topological structure correction. For the lumen, wall, and plaque components, T1WI and T2WI multi-sequence magnetic resonance image information is fused, and deformation models or deep learning methods are used for precise segmentation to identify the lipid core, fibrous cap, and calcified plaque components. According to the American Heart Association (AHA) guidelines, the various components within the plaque are quantitatively analyzed, and the vulnerability indicators of lipid core area and fibrous cap thickness are calculated to assess the nature, type, and degree of vulnerability of the plaque. Finally, 3D reconstruction is performed using the Marching Cubes algorithm, and the 3D models of the blood vessels, lumen, wall, and plaque components are fused to generate a complete 3D cerebral vascular structure, which is then visualized. The results are presented in the form of detailed reports of typical layers, 3D reconstructions, and fused images, providing an intuitive basis for plaque vulnerability assessment.

3. The method for constructing a prediction model for cerebrovascular and brain functional diseases using big data and AI technology according to claim 2, characterized in that: The image processing technology in step one also includes medical image denoising, which is a medical image denoising technology based on an improved non-local mean filtering algorithm. First, the Facet operator is used to accurately extract the edge features of the image, and different self-similarity measurement functions are adaptively selected according to the local statistical characteristics of the pixel distribution within the image; at the same time, a variable-size search window mechanism is adopted to maximize the search range of similar neighborhoods and reduce the interference of noise on the self-similarity evaluation, thereby effectively removing noise while significantly improving the ability to maintain the edge structure of the image.