A stroke prediction method, device, medium and product based on a cerebrovascular image feature set

Abstract

The application discloses a stroke prediction method and device based on a cerebrovascular image feature group, a medium and a product, relates to the field of image processing, and comprises the following steps: constructing a clinical database; constructing a clinical cause classification prediction model according to multi-modal image feature group data, clinical feature data and patient clinical cause classification; determining cerebrovascular image features by adopting a cerebrovascular tissue segmentation model and voxel-based morphological analysis according to TOF-MRA, and performing positioning diagnosis on a responsible blood vessel area according to the cerebrovascular image features; and constructing a stroke recurrence risk prediction model according to the cerebrovascular image features in the TOF-MRA, SWI, T1WI, T2-FLAIR and corresponding clinical feature data. The application can quickly and accurately realize cause classification of an acute stroke patient in a clinical scene, accurately identify a responsible blood vessel, and realize recurrence risk prediction in an early stage of the disease.

Description

Technical Field

[0001] This application relates to the field of image processing, and in particular to a method, device, medium and product for predicting stroke based on cerebrovascular imaging characterization. Background Technology

[0002] With the increasing aging of the global population, stroke has become one of the leading causes of death and disability worldwide. The etiological classification of acute ischemic stroke (such as large artery atherosclerosis, cardioembolism, and small vessel occlusion) and the location of the responsible vessel are crucial for early diagnosis, treatment selection, and prognostic assessment. However, existing methods for etiological classification and culprit vessel location are mostly based on traditional clinical indicators and single-modality imaging data, such as computed tomography (CT) and conventional magnetic resonance imaging (MRI). While these methods can provide some structural information, they cannot comprehensively capture the aging state of the cerebral blood vessels and subtle lesions. Especially for acute ischemic stroke, accurate early lesion localization is crucial, and single-modality data often falls short of this requirement, leading to insufficient diagnostic accuracy and reliability. Furthermore, the subgroup of patients experiencing recurrence exhibits worse long-term prognosis. Therefore, achieving objective and accurate quantitative analysis of early etiological classification, culprit vessel location, and recurrence risk in acute ischemic stroke has extremely high clinical guiding significance.

[0003] Therefore, developing a method capable of quantitatively assessing cerebrovascular aging and achieving precise classification and localization of the responsible vessel is of significant clinical and social value. Early identification of the responsible vessel in acute ischemic stroke is crucial for the selection of treatment options.

[0004] Existing classification and localization models are often trained on small-scale clinical data, lacking support from large-scale population cohorts. This results in poor generalization ability, making it difficult to adapt to patient populations in different regions and limiting their widespread application in real-world clinical settings. Furthermore, existing methods for locating the responsible vessel typically rely on manual annotation or simple image processing techniques, which are easily influenced by subjective factors and struggle to accurately match the brain region where the stroke occurred with the responsible vessel supplying it.

[0005] Current stroke classification methods are mainly based on traditional clinical indicators (such as age, blood pressure, and blood lipids). While these indicators can reflect some risk factors, they cannot directly assess the degree of aging of cerebrovascular vessels. Cerebrovascular aging is an important precursor to the occurrence and development of stroke. Especially in the early stages, changes in microvessels may be the earliest pathological features. However, existing assessment methods cannot effectively capture these subtle changes, resulting in inaccurate and incomplete classification of stroke and localization of the responsible vessel.

[0006] Early identification of the culprit vessel in acute ischemic stroke is crucial for treatment selection. Existing methods for culprit vessel localization typically rely on manual annotation or simple image processing techniques, which are easily influenced by subjective factors and struggle to accurately match the infarcted brain region with its supplying vessel. Furthermore, traditional feature extraction methods depend on manually designed algorithms, which are susceptible to subjective factors and fail to fully utilize the complex information in multimodal images.

[0007] Most existing stroke recurrence risk prediction models use uniform assessment criteria and fail to adequately consider individual differences. The degree of cerebrovascular aging and recurrence risk may vary significantly from patient to patient, and existing models cannot provide personalized prediction results, affecting clinicians' ability to accurately manage patients.

[0008] How to quickly and accurately classify the causes of acute stroke patients in clinical settings, accurately identify the responsible blood vessel, and predict the risk of recurrence in the early stages of the disease, so as to assist clinicians in achieving individualized and precise diagnosis and treatment, is an urgent problem to be solved. Summary of the Invention

[0009] The purpose of this application is to provide a method, device, medium and product for predicting stroke based on cerebrovascular imaging characterization, which can quickly and accurately classify the etiology of acute stroke patients in clinical scenarios, accurately identify the responsible blood vessel, and predict the risk of recurrence in the early stage of the disease.

[0010] To achieve the above objectives, this application provides the following solution:

[0011] In a first aspect, this application provides a stroke prediction method based on cerebrovascular imaging characterization groups, the stroke prediction method based on cerebrovascular imaging characterization groups comprising:

[0012] A clinical database is constructed, comprising: multimodal imaging feature group data and corresponding clinical feature data for each patient; the multimodal imaging feature group data includes: TOF-MRA, SWI, T1WI and T2-FLAIR; the clinical feature data includes: gender, age, height, weight and disease status;

[0013] Based on the multimodal imaging feature group data, corresponding clinical feature data, and corresponding clinical etiology classification of each patient, a simple fully connected neural network architecture with a single hidden layer is used to obtain a clinical etiology classification prediction model.

[0014] Based on each patient's TOF-MRA, a cerebral vascular tissue segmentation model and voxel-based morphological analysis were used to determine cerebral vascular imaging features, and the responsible vessel region was located and diagnosed based on these features. The cerebral vascular tissue segmentation model was used to obtain segmented cerebral vascular images based on TOF-MRA. The cerebral vascular imaging features included: basic cerebral vascular features, geometric features, and topological features. The basic cerebral vascular features included: the length and volume of individual tubular structures, and the diameter and length of local brain regions. The geometric features included: the radius of curvature and degree of tortuosity of cerebral vessels. The topological features included: cerebral vascular bifurcation points.

[0015] Based on the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical features of each patient in TOF-MRA, a simple fully connected neural network architecture with a single hidden layer is used to obtain a stroke recurrence risk prediction model based on the cerebrovascular aging imaging characterization group; the stroke recurrence risk prediction model based on the cerebrovascular aging imaging characterization group is used to output the recurrence probability value of stroke patients.

[0016] Obtain multimodal image feature group data and corresponding clinical feature data of the subject to be predicted;

[0017] Clinical etiology classification prediction model and stroke recurrence risk prediction model based on cerebrovascular aging imaging characterization group were used to predict clinical etiology classification and stroke recurrence risk; and the localization diagnosis of the responsible vascular area was carried out.

[0018] Optionally, the construction of the clinical database specifically includes:

[0019] The multimodal imaging feature group data and corresponding clinical feature data of each patient are preprocessed; the preprocessing includes: removing outlier data, multiple imputation of missing data, and data cleaning.

[0020] Optionally, both the clinical etiology classification prediction model and the stroke recurrence risk prediction model based on cerebrovascular aging imaging characterization group are trained using the early training cessation method.

[0021] Optionally, the step of determining cerebral vascular imaging features based on each patient's TOF-MRA using a cerebral vascular tissue segmentation model and voxel-based morphological analysis, and then locating and diagnosing the responsible vessel region based on these features, specifically includes:

[0022] The TOF-MRA magnetic resonance sequence image preprocessing includes: brain tissue extraction, magnetic resonance sequence image registration, magnetic resonance sequence image denoising and B0 field correction, contrast-limited adaptive histogram equalization, blood vessel edge signal enhancement, and transform coefficient Gamma correction.

[0023] Using a cerebral vascular tissue segmentation model, cerebral vascular tissue segmentation was performed on the preprocessed magnetic resonance sequence images to obtain cerebral vascular segmentation images.

[0024] Based on the segmented images of cerebral blood vessels, a three-dimensional conditional random field model was used for three-dimensional conditional random field processing.

[0025] The final cerebral blood vessel segmentation image is determined based on the three-dimensional maximum connectivity component in the cerebral blood vessel segmentation image after processing with a three-dimensional conditional random field.

[0026] The final cerebral blood vessel segmentation image is nonlinearly registered with the cerebral blood vessel distribution density map to obtain the transformed Jacobian matrix.

[0027] The transformed Jacobian matrix is ​​used to perform Jacobian matrix normalization on the nonlinearly registered cerebral blood vessel segmentation image to obtain a cerebral blood vessel distribution density image; the cerebral blood vessel distribution density image is used to characterize the distribution density of individual cerebral blood vessels in the standard atlas space.

[0028] Based on the cerebral vascular distribution density image and cerebral vascular distribution density atlas, a voxel-level one-sample t-test was performed, and the Bonferroni method was used for multiple comparison correction to achieve the localization diagnosis of the responsible vascular region; the responsible vascular region is the vascular region with the greatest signal difference in the individual vascular density distribution in the cerebral vascular distribution density image.

[0029] Optionally, the cerebral vascular tissue segmentation model is a deep neural network model that introduces an attention gating module into the skip connections of the 3D-UNet convolutional neural network.

[0030] Optionally, the stroke recurrence risk prediction model based on cerebrovascular aging image characterization group includes: a convolutional neural network encoder, a cerebrovascular image feature fusion module, and a vascular biological age prediction module;

[0031] The convolutional neural network encoder is used to extract convolutional features from SWI, T1WI, and T2-FLAIR using a combination of asymmetric convolution modules and Squeeze-and-Excitation modules.

[0032] The cerebrovascular imaging feature fusion module is used to compress convolutional features using a global pooling compression method, obtaining convolutional feature groups of SWI, T1WI, and T2-FLAIR with each feature dimension of 1. Demographic indicators and cerebrovascular imaging features are then directly concatenated with these convolutional feature groups to generate a comprehensive image feature group. This comprehensive image feature group includes TOF-MRA cerebrovascular imaging features that specifically reflect cerebrovascular characteristics, as well as convolutional feature groups of SWI, T1WI, and T2-FLAIR that reflect brain aging characteristics. The convolutional feature groups include: white matter high signal volume, number of cavities, cortical thickness, gray matter volume, microbleeds, and ventricular volume. The demographic indicators include: age and sex.

[0033] The vascular biological age prediction module is used to comprehensively predict the patient's brain aging age based on the comprehensive imaging feature group, and compare the comprehensive prediction result with the individual's physical age to obtain the comprehensive index of cerebrovascular aging.

[0034] Secondly, this application provides a stroke prediction device based on a cerebrovascular imaging characterization group, the stroke prediction device based on the cerebrovascular imaging characterization group comprising:

[0035] A clinical database construction module is used to construct a clinical database; the clinical database includes: multimodal imaging feature group data and corresponding clinical feature data for each patient; the multimodal imaging feature group data includes: TOF-MRA, SWI, T1WI and T2-FLAIR; the clinical feature data includes: gender, age, height, weight and disease status;

[0036] The clinical etiology classification prediction model determination module is used to obtain the clinical etiology classification prediction model based on the multimodal image feature group data, corresponding clinical feature data and corresponding patient clinical etiology classification of each patient, using a simple fully connected neural network architecture with a single hidden layer.

[0037] The localization diagnosis module is used to determine the cerebral vascular imaging features based on each patient's TOF-MRA using a cerebral vascular tissue segmentation model and voxel-based morphological analysis, and to perform localization diagnosis of the responsible vessel region based on these features. The cerebral vascular tissue segmentation model is used to obtain segmented cerebral vascular images based on TOF-MRA. The cerebral vascular imaging features include: basic cerebral vascular features, cerebral vascular geometric features, and cerebral vascular topological features. The basic cerebral vascular features include: the length and volume of individual tubular structures, and the diameter and length of local brain regions. The cerebral vascular geometric features include: the radius of curvature and degree of tortuosity of cerebral vessels. The cerebral vascular topological features include: cerebral vascular bifurcation points.

[0038] The stroke recurrence risk prediction model determination module is used to determine the stroke recurrence risk prediction model based on the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data of each patient in TOF-MRA, using a simple fully connected neural network architecture with a single hidden layer; the stroke recurrence risk prediction model based on the cerebrovascular aging imaging characterization group is used to output the recurrence probability value of stroke patients.

[0039] The data acquisition module is used to acquire multimodal image feature group data and corresponding clinical feature data of the subject to be predicted;

[0040] The stroke prediction module is used to predict the clinical etiology and stroke recurrence risk using a clinical etiology classification prediction model and a stroke recurrence risk prediction model based on cerebrovascular aging imaging characterization group; and to perform localization diagnosis of the responsible vascular area.

[0041] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the stroke prediction method based on cerebrovascular imaging characterization groups.

[0042] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the stroke prediction method based on cerebrovascular imaging characterization groups.

[0043] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the stroke prediction method based on cerebrovascular imaging characterization groups.

[0044] According to the specific embodiments provided in this application, this application has the following technical effects:

[0045] This application provides a method, device, medium, and product for stroke prediction based on cerebrovascular imaging characterization. It constructs a clinical database including multimodal imaging feature sets and corresponding clinical feature data for each patient. Through time-of-flight magnetic resonance angiography (TOF-MRA) segmentation and automatic extraction of feature data, a comprehensive cerebrovascular imaging characterization set is built to fully capture the aging state and subtle lesions of cerebrovascular systems. Furthermore, it integrates susceptibility-weighted imaging (SWI), T1-weighted imaging (T1WI), and T2 fluid attenuated inversion recovery imaging. This study constructs a stroke recurrence risk prediction model based on cerebral vascular aging imaging characterization, combining T2-FLAIR (Taiwanese Imaging Recovery) and convolutional feature sets reflecting brain aging characteristics with features automatically extracted from cerebral vessels by TOF-MRA. This model comprehensively captures the patient's overall health status, particularly the aging state of cerebral vessels, reflecting important indicators of cerebral vascular pathophysiology and thus improving the accuracy of clinical risk classification prediction. Quantitative assessment of cerebral vascular aging reflects structural changes in large vessels and captures aging characteristics of microvessels, providing a more comprehensive evaluation perspective. By comparing with an individual's physical age, the cerebral vascular aging imaging characterization model accurately estimates the patient's cerebral vascular biological age, revealing the potential risk of vascular aging. The fusion of multimodal data from a clinical database not only improves the model's accuracy but also enhances its robustness and generalization ability. Based on the multimodal imaging feature set data of the subjects to be predicted, voxel-based morphological analysis is used to locate and diagnose the responsible vessel region, accurately matching the brain region where cerebral infarction occurred with its supplying vessel. This application provides personalized stroke classification and culprit vessel localization results, as well as individualized predictions of recurrence risk, based on multimodal imaging feature data and corresponding clinical characteristics of the patient. By integrating information on the degree of cerebrovascular aging in patients, this application performs individual-level etiological classification and culprit vessel localization for stroke patients, and finally uses the combined information on the degree of cerebrovascular aging to predict the risk of recurrence. This helps in developing personalized treatment plans, improving the quality of patient prognosis, and further assisting clinicians in understanding the underlying pathological mechanisms of stroke pathogenesis and recurrence. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of a stroke prediction method based on cerebrovascular imaging characterization group in one embodiment of this application. Detailed Implementation

[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0049] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] In one exemplary embodiment, such as Figure 1 As shown, a stroke prediction method based on cerebrovascular imaging characterization is provided, which includes the following steps S101 to S106. Wherein:

[0051] S101, Construct a clinical database; the clinical database includes: multimodal imaging feature group data and corresponding clinical feature data for each patient; the multimodal imaging feature group data includes: TOF-MRA, SWI, T1WI and T2-FLAIR; the clinical feature data includes: gender, age, height, weight and disease status.

[0052] S101 specifically includes:

[0053] The multimodal imaging feature group data and corresponding clinical feature data of each patient are preprocessed; the preprocessing includes: removing outlier data, multiple imputation of missing data, and data cleaning.

[0054] As a specific implementation, outliers are detected and removed using the Median Absolute Deviation (MAD) method. Simultaneously, multiple imputation is performed on samples with missing data. Finally, a standard data frame structure is constructed to complete the data cleaning and organization.

[0055] S102. Based on the multimodal image feature group data, corresponding clinical feature data, and corresponding clinical etiology classification of each patient, a simple fully connected neural network architecture with a single hidden layer is used to obtain a clinical etiology classification prediction model.

[0056] The loss function CE of the clinical etiology classification prediction model adopts the softmax cross-entropy loss function, and the specific formula is as follows:

[0057] .

[0058] Where, f(s) i t represents the result of the linear combination of variables in the multimodal imaging feature set data of the i-th patient after transformation by the softmax function. i Let N represent the actual etiology classification value of the i-th patient, where N represents the number of samples and C represents the number of etiology classifications.

[0059] The specific form of the Softmax function is: .

[0060] Data from the clinical database was used to construct a dataset, which was then randomly sampled in a 7:3 ratio to form a training set and a test set. The training set was used for 10-fold cross-validation training, and the optimal model was selected as the final output model, i.e., the clinical etiology classification prediction model. Validation was performed on the test set. To prevent overfitting, early stopping was used during training to ensure the model was sufficiently trained, greatly reducing the risk of overfitting. The final output of the clinical etiology classification prediction model is the probability value of each clinical etiology for the patient. The etiology with the highest probability is the final predicted clinical etiology classification value for that patient.

[0061] S103, based on each patient's TOF-MRA, a cerebral vascular tissue segmentation model and voxel-based morphometry (VBM) analysis are used to determine cerebral vascular imaging features, and the responsible vessel region is located and diagnosed based on these features. The cerebral vascular tissue segmentation model is used to obtain segmented cerebral vascular images based on TOF-MRA. The cerebral vascular imaging features include: basic cerebral vascular features, cerebral vascular geometric features, and cerebral vascular topological features. The basic cerebral vascular features include: the length and volume of individual tubular structures, and the diameter and length of local brain regions. The cerebral vascular geometric features include: the radius of curvature and degree of tortuosity of cerebral vessels. The cerebral vascular topological features include: cerebral vascular bifurcation points.

[0062] Before extracting basic cerebral vascular features from TOF-MRA, the process includes: estimating the centerline, normal plane, and performing 3D reconstruction; the specific steps include:

[0063] S11. First, centerline estimation is performed on the three-dimensional tubular tree structure of cerebral blood vessels in the segmented image. Based on the Voronoi Covariance Measure (VCM), the centerline of the segmented cerebral blood vessels is stably extracted, reducing the drift of center point identification and voxel discontinuity at the bifurcation of cerebral blood vessels.

[0064] S12, based on the extraction of the cerebral vascular centerline, the tangent corresponding to each voxel on the centerline is calculated using the λ-maximal segment tangent method (λ-MST), thereby further calculating the normal plane on each voxel of the centerline. A robust estimate of the normal plane of each voxel of the centerline is obtained by weighted averaging of the normal planes of neighboring voxels.

[0065] S13, for each voxel along the centerline, fit the largest circle in its normal plane and calculate the diameter corresponding to that voxel.

[0066] S14, combining centerline and diameter information, achieves 3D reconstruction of the tubular tree structure. This effectively prevents minor local errors introduced by the structure segmentation algorithm from being amplified in the cascade of blood vessel feature estimation, ensuring the robustness of the 3D reconstruction.

[0067] The process of extracting basic features of cerebral blood vessels is as follows:

[0068] S21. Count the number of positively labeled voxels in the 3D image of cerebral blood vessel segmentation to obtain the absolute length of cerebral blood vessels. At the same time, considering that the size of individual brain tissue can significantly affect the length of cerebral blood vessel structure, this algorithm divides the absolute length by the volume of individual brain tissue to obtain the relative length of cerebral blood vessels.

[0069] S22: Count the number of voxels with positive markers in the 3D image of cerebral blood vessel segmentation, and multiply them by the voxel volume to obtain the absolute volume of cerebral blood vessels. At the same time, considering that the size of individual brain tissue can also significantly affect the volume of cerebral blood vessel structure, this algorithm also divides the absolute volume of cerebral blood vessels by the size of individual brain tissue to obtain the relative volume, which reflects the average distribution density of cerebral blood vessel tissue in the overall brain tissue.

[0070] S23 utilizes a two-step registration method to nonlinearly register the MNI standard brain template with cerebral angiography magnetic resonance imaging (MRI) sequences, and transforms the Harvard-Oxford brain region map in the MNI standard brain space to the individual-level MRI angiography image space. Local vessel length and diameter are calculated in each brain region to reflect the heterogeneity of basic vascular features in the spatial distribution of different brain regions.

[0071] The process of extracting geometric features of cerebral blood vessels is as follows:

[0072] In the segmented central line image of cerebral blood vessels, the radius of curvature and degree of tortuosity of each voxel point on the central line in three-dimensional space are calculated based on the tangent vector flip angle and the Frenet frame. Furthermore, the average radius of curvature and degree of tortuosity of blood vessels in each brain region corresponding to the Harvard-Oxford brain region map are calculated to measure the geometric distribution characteristics of cerebral blood vessels in brain regions.

[0073] The process of extracting cerebral vascular topological features is as follows:

[0074] S31. Based on the segmented central line image of cerebral blood vessels, the adjacency relationship of each voxel of the central line is represented by a tree structure, thereby realizing the automatic identification of the bifurcation points in the tree structure of blood vessels, and counting the number of bifurcations and their spatial distribution.

[0075] S32, calculate the angle between the adjacent tangent vectors of the bifurcation point, which is the bifurcation angle of the bifurcation point;

[0076] S33 calculates the average bifurcation angle and bifurcation density in each brain region, reflecting the spatial distribution characteristics of vascular bifurcation statistics in brain regions.

[0077] S103 specifically includes:

[0078] S311, Perform magnetic resonance sequence image preprocessing on multimodal image feature group data; the magnetic resonance sequence image preprocessing includes: brain tissue extraction, magnetic resonance sequence image registration, magnetic resonance sequence image denoising and B0 field correction, contrast-limited adaptive histogram equalization, blood vessel edge signal enhancement, and transformation coefficient Gamma correction.

[0079] Magnetic resonance image preprocessing specifically includes:

[0080] S100, Brain Tissue Extraction: Brain tissue-specific identification and segmentation are performed using TOF-MRA to remove non-brain tissue structures such as the scalp, skull, facial features, and neck. The purpose of this step is to eliminate interference from non-brain tissue image signals in subsequent processing.

[0081] S200, Magnetic Resonance Imaging Sequence Registration: Using a linear rigid body transformation with 6 degrees of freedom and spline interpolation, T1-weighted images and TOF-MRA of the same individual are registered and resampled to the same image space. The purpose of this step is to transform images from different voxel coordinates to the same anatomical coordinates, facilitating the integration of cross-sequence signal information.

[0082] S300, Magnetic Resonance Imaging Sequence Image Denoising and B0 Field Correction: An adaptive nonlocal means algorithm is used to denoise the TOF-MRA image, removing Gaussian noise that may be present in the magnetic resonance imaging sequence. Subsequently, a signal intensity remapping method after Gaussian deconvolution is used to remove artifacts in the magnetic resonance imaging sequence caused by B0 field inhomogeneity. The purpose of this step is to remove common artifact signals in magnetic resonance imaging sequences and improve the image signal-to-noise ratio.

[0083] S400, Contrast-Limited Adaptive Histogram Equalization: Due to the limitations of TOF-MRA and other vascular imaging magnetic resonance sequences, the signal enhancement amplitude of small vessels is significantly lower than that of large vessels, making small vessel identification difficult. A contrast-limited adaptive histogram equalization method is employed to adaptively enhance local cerebral vascular signals. This step allows cerebral vascular signals of different scales to achieve similar enhancement amplitudes.

[0084] S500, Enhancement of Vascular Edge Signal: Due to the partial volume effect, the signal at the vascular edge is significantly lower than that at the vascular center, making it difficult to identify. The Sobel convolution method is employed to identify edge locations where signal intensity abruptly changes in the image, achieving specific enhancement of the vascular edge signal and improving the contrast between the blood vessel and surrounding non-vascular tissue. This step allows cerebral vascular signals at different distances from the centerline to achieve similar enhancement.

[0085] S600, Gamma Correction: Finally, Gamma correction is used. By adjusting the transform coefficient Gamma, the signal of vascular tissue (high signal) in the magnetic resonance image is nonlinearly enhanced while the signal of non-vascular tissue (low signal) is suppressed. The transform coefficient Gamma, as a hyperparameter of the processing, can control the relative amplitude of signal enhancement and suppression according to the nature of the processed image, thereby obtaining better image contrast.

[0086] S332 utilizes a cerebral vascular tissue segmentation model to segment cerebral vascular tissue from multimodal image feature group data after preprocessing of magnetic resonance sequence images, resulting in cerebral vascular segmentation images.

[0087] The training and testing dataset for the cerebral vascular tissue segmentation model was determined as follows: T1WI and TOF-MRA images of 100 healthy individuals were collected using routine clinical scanning sequences from different devices. After preprocessing the image data, two neuroradiologists with extensive experience in image annotation independently annotated the cerebral vascular tissue signals in the MRI images. A third senior neuroradiologist then comprehensively reviewed and evaluated the annotation results, completing the annotation of cerebral vascular tissue signals from multimodal MRI images to ensure consistency and accuracy. This resulted in the construction of a supervised training and testing dataset for the model. The cerebral vascular tissue signals are specific signal features in MRI that can identify and distinguish cerebral vascular tissue. These specific signal features mainly include low signal intensity (flowing blood) in T1WI and high signal intensity (flowing blood) in TOF-MRA. Through the combination of multimodal images and expert annotation, a high-quality dataset for training and testing the cerebral vascular tissue segmentation model was constructed.

[0088] The cerebral vascular tissue segmentation model is a deep neural network model that incorporates an attention gating module into the skip connections of a 3D-UNet convolutional neural network. The overall 3D-UNet convolutional neural network employs an adaptive design, automatically adjusting hyperparameters such as batch size and network depth / width based on the spatial resolution and size of the input image. Simultaneously, this model introduces an attention gating module into the skip connections of the 3D-UNet convolutional neural network, the basic form of which can be expressed as:

[0089] .

[0090] .

[0091] Among them, f i g i σ1 represents the feature tensors on the encoding and decoding modules connected by the jump, σ2 represents the ReLU activation function, σ3 represents the Sigmoid activation function, and φ, w f , w g Represents the linear transformation coefficients, b φ and b g This represents the bias coefficient. Finally, the resampled SAct image is compared with the feature tensor f on the decoding module. i Multiplying them together allows for the fusion of feature information at different scales, greatly reducing background noise.

[0092] Five-fold cross-validation was used to train and test the model. To address the differences in specificity and sensitivity of vessel segmentation results at different scales during the segmentation process, the Tversky function was introduced as the model's loss function. Based on the Tversky function, a weighted Tversky loss was further proposed, which uses the reciprocal of the Euclidean distance from the voxel to the center line as the mismatch loss weight. This strengthens the topological constraints of the segmentation results, resulting in more consistent performance of vessel segmentation results across various scales.

[0093] S333, based on the cerebral blood vessel segmentation image, a three-dimensional conditional random field model is used for three-dimensional conditional random field processing. The three-dimensional conditional random field model considers the spatial continuity of vascular tissue. By establishing potential functions between pairs and at individual points, neighboring points with similar signal intensities on magnetic resonance images with different segmentation labels on the cerebral blood vessel segmentation image are penalized, resulting in a more continuous and reasonable cerebral blood vessel segmentation image in three-dimensional space.

[0094] S334, determine the final cerebral blood vessel segmentation image based on the three-dimensional maximum connectivity component in the cerebral blood vessel segmentation image after processing with a three-dimensional conditional random field; that is, remove small, disconnected signal noise in three-dimensional space.

[0095] S335, the final cerebral blood vessel segmentation image is nonlinearly registered with the cerebral blood vessel distribution density map to obtain the transformed Jacobian matrix.

[0096] S336, the Jacobian matrix of the nonlinearly registered cerebral blood vessel segmentation image is normalized using the transformed Jacobian matrix to obtain a cerebral blood vessel distribution density image; the cerebral blood vessel distribution density image is used to characterize the distribution density of individual cerebral blood vessels in the standard atlas space.

[0097] S337. Based on the cerebral vascular distribution density image and cerebral vascular distribution density atlas, a voxel-level single-sample t-test was performed, and the Bonferroni method was used for multiple comparison correction to achieve the localization diagnosis of the responsible vascular region; the responsible vascular region is the vascular region with the greatest signal difference in the individual vascular density distribution in the cerebral vascular distribution density image.

[0098] S104 uses a simple fully connected neural network architecture with a single hidden layer to obtain a stroke recurrence risk prediction model based on the cerebral vascular aging image characterization group, based on the cerebral vascular aging image characterization group of each patient's TOF-MRA cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data; the stroke recurrence risk prediction model based on the cerebral vascular aging image characterization group is used to output the recurrence probability value of stroke patients.

[0099] Among them, T1WI, T2-FLAIR, and SWI reflect the overall characteristics of various tissue structures such as white matter and gray matter in an individual's brain. Therefore, they can reflect the impact of vascular aging on these brain tissue structures and contain rich information about cerebral vascular aging.

[0100] The stroke recurrence risk prediction model uses the cross-entropy loss function, the specific formula of which is as follows:

[0101] .

[0102] Among them, y （i） The two-component variable x represents the actual relapse status of the i-th patient. （i） Let m represent the variable in the i-th patient's imaging representation group, and h represent the total sample size. θ (x) （i） ) represents the result predicted by the model for the i-th patient.

[0103] The dataset was randomly divided into training and test sets in a 7:3 ratio. Ten-fold cross-validation was performed on the training set to select the optimal model as the stroke recurrence risk prediction model. The model was then validated on the test set. To prevent overfitting, early stopping was used during training to ensure sufficient training of the stroke recurrence risk prediction model, significantly reducing the risk of overfitting. The final output of the stroke recurrence risk prediction model represents the probability of recurrence for the patient. This model can quantitatively and objectively measure the relationship between the degree of cerebrovascular aging on imaging and the risk of stroke recurrence.

[0104] The stroke recurrence risk prediction model based on cerebrovascular aging imaging characterization group includes: a convolutional neural network encoder, a cerebrovascular imaging feature fusion module, and a vascular biological age prediction module.

[0105] The convolutional neural network encoder is used to extract convolutional features from SWI, T1WI, and T2-FLAIR by combining an asymmetric convolution module and a squeeze-and-excitation module. That is, it takes advantage of the neural network's ability to adaptively encode and extract features to automatically and comprehensively extract vascular aging features from SWI, T1WI, and T2-FLAIR.

[0106] The asymmetric convolution module increases the image receptive field of the convolution kernel, and its specific formula is as follows:

[0107] .

[0108] in, f It is the input feature map.g The output feature map is , where Conv1×1×k, conv1×k×1, convk×1×1, and convk×k×k are the convolution kernels of 1×1×k, 1×k×1, k×1×1, and k×k×k, respectively.

[0109] The Squeeze-and-Excitation module can introduce an attention mechanism into neural networks, adaptively adjusting the weight parameters of each feature based on the correlation between different convolutional features. The specific formula is as follows:

[0110] .

[0111] Where SE(F) is the output of the Squeeze-and-Excitation module, F is the input feature map, σ is the sigmoid nonlinear activation function, ReLU is the Rectified LieanerUnit nonlinear activation function, W1 is the linear parameter matrix of the input feature map after global pooling compression, b1 is the bias term of the input feature map after global pooling compression, W2 is the linear parameter matrix after ReLU nonlinear activation, and b2 is the corresponding bias term.

[0112] GlobalAvgPool represents global pooling compression, which averages the width and height of the input feature map. Its specific formula is as follows:

[0113] .

[0114] in, Let be the pooled value of channel i, H be the height of the input feature map, W be the width of the input feature map, j be the number of rows of the feature map in channel i, and k be the number of columns of the feature map in channel k. is the pixel value of the feature map at position (j,k) in channel i.

[0115] By connecting the two asymmetric convolution modules and the Squeeze-and-Excitation module in the convolutional neural network encoder, the convolutional features of SWI, T1WI, and T2-FLAIR are extracted.

[0116] The cerebrovascular imaging feature fusion module is used to compress convolutional features using a global pooling compression method to obtain convolutional feature groups of SWI, T1WI, and T2-FLAIR with each feature dimension of 1. Demographic indicators and cerebrovascular imaging features are then directly concatenated with the convolutional feature groups to generate a comprehensive image feature group. This comprehensive image feature group includes TOF-MRA cerebrovascular imaging features that specifically reflect cerebrovascular characteristics, as well as convolutional feature groups of SWI, T1WI, and T2-FLAIR that reflect brain aging characteristics. The convolutional feature groups include: white matter high signal volume, number of cavities, cortical thickness, gray matter volume, microbleeds, and ventricular volume. The demographic indicators include: age and sex.

[0117] The vascular biological age prediction module is used to comprehensively predict the patient's brain aging age based on the comprehensive imaging feature group, and compare the comprehensive prediction result with the individual's physical age to obtain the comprehensive index of cerebrovascular aging.

[0118] The loss function for the vascular biology age prediction module adopts the MSE loss function. The specific formula is as follows:

[0119] .

[0120] in, Estimation of vascular age, y i N represents the patient's actual physical age, and N represents the total sample size.

[0121] This application proposes a comprehensive cerebrovascular aging index based on a stroke recurrence risk prediction model using cerebrovascular aging imaging representations. This index can quantitatively and objectively measure the degree of brain aging in patients using multimodal magnetic resonance imaging data. Simultaneously, the model obtains the effect coefficient corresponding to each imaging feature. This coefficient can quantitatively assess the contribution of imaging features to the overall cerebrovascular aging imaging representation group, indicating the imaging features with the greatest impact on vascular aging, and has strong clinical indicative significance.

[0122] S105, Obtain the multimodal image feature group data and corresponding clinical feature data of the person to be predicted.

[0123] S106, the vascular biological age prediction module is used to comprehensively predict the patient's brain aging age based on the comprehensive imaging feature group, and compare the comprehensive prediction result with the individual's physical age to obtain the comprehensive index of cerebrovascular aging.

[0124] This application combines multimodal magnetic resonance imaging (MRI) and automatically extracted feature data from cerebral vascular segmentation to construct a comprehensive image characterization set. By fusing brain aging features from SWI, T1WI, and T2-FLAIR, and combining them with cerebral vascular features automatically extracted from TOF-MAR, the model can more comprehensively capture the patient's overall health status, especially the aging state of cerebral blood vessels. This allows for the use of these important indicators reflecting the pathophysiological state of cerebral blood vessels to predict clinical risk classification. This multimodal data fusion not only improves the model's accuracy but also enhances its robustness and generalization ability. The concept of a cerebral vascular aging image characterization set is introduced, using neural networks to automatically extract vascular aging features and utilizing multimodal imaging data to quantitatively assess the degree of cerebral vascular aging. This index not only reflects structural changes in large vessels but also captures the aging features of microvessels, providing a more comprehensive assessment perspective. By comparing with an individual's physical age, the cerebral vascular aging image characterization set can accurately estimate a patient's cerebral vascular biological age, revealing the potential risks of vascular aging. This method employs nonlinear registration to align individual-level cerebral vascular segmentation images with population-level cerebral vascular distribution density maps, and then uses a difference matrix standardized by the Jacobian matrix to locate the responsible vessel. This approach not only accurately locates the brain region experiencing infarction and its supplying vessel but also quantitatively assesses the local or global differences between individual and population-level vascular distribution network density, significantly improving the accuracy and reliability of responsible vessel localization. The model is trained on a large cohort clinical database covering 20,000 stroke patients, ensuring its generalization ability and reliability. Through rigorous outlier detection and data cleaning, the model performs well across different regions and patient groups, demonstrating broad application prospects. To prevent overfitting, ten-fold cross-validation and early stopping strategies are used during training. These methods not only ensure model stability but also significantly reduce the risk of overfitting, improving predictive performance. Personalized stroke recurrence risk scores are provided based on each patient's imaging characteristics and clinical features. This helps clinicians better understand patients' conditions, develop personalized treatment plans, and improve patient prognosis.

[0125] Based on the same inventive concept, this application also provides a stroke prediction device based on cerebrovascular imaging characterization groups for implementing the stroke prediction method based on cerebrovascular imaging characterization groups described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more stroke prediction device embodiments based on cerebrovascular imaging characterization groups provided below can be found in the limitations of the stroke prediction method based on cerebrovascular imaging characterization groups described above, and will not be repeated here.

[0126] In one exemplary embodiment, a stroke prediction device based on cerebrovascular imaging characterization groups is provided, comprising:

[0127] A clinical database construction module is used to construct a clinical database; the clinical database includes: multimodal imaging feature group data and corresponding clinical feature data for each patient; the multimodal imaging feature group data includes: TOF-MRA, SWI, T1WI and T2-FLAIR; the clinical feature data includes: gender, age, height, weight and disease status.

[0128] The clinical etiology classification prediction model determination module is used to obtain a clinical etiology classification prediction model based on the multimodal imaging feature group data, corresponding clinical feature data, and corresponding clinical etiology classification of each patient, using a simple fully connected neural network architecture with a single hidden layer.

[0129] The localization diagnosis module is used to determine the cerebral vascular imaging features based on each patient's TOF-MRA using a cerebral vascular tissue segmentation model and voxel-based morphological analysis, and to perform localization diagnosis of the responsible vessel region based on these features. The cerebral vascular tissue segmentation model is used to obtain segmented cerebral vascular images based on time-of-flight magnetic resonance angiography images. The cerebral vascular imaging features include: basic cerebral vascular features, cerebral vascular geometric features, and cerebral vascular topological features. The basic cerebral vascular features include: the length and volume of individual tubular structures, and the diameter and length of local brain regions. The cerebral vascular geometric features include: the radius of curvature and degree of tortuosity of cerebral vessels. The cerebral vascular topological features include: cerebral vascular bifurcation points.

[0130] The stroke recurrence risk prediction model determination module is used to obtain a stroke recurrence risk prediction model based on the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data of each patient using a simple fully connected neural network architecture with a single hidden layer; the stroke recurrence risk prediction model based on the cerebrovascular aging imaging characterization group is used to output the recurrence probability value of stroke patients.

[0131] The data acquisition module is used to acquire multimodal image feature group data and corresponding clinical feature data of the subject to be predicted.

[0132] The stroke prediction module is used to predict the clinical etiology and stroke recurrence risk using a clinical etiology classification prediction model and a stroke recurrence risk prediction model based on cerebrovascular aging imaging characterization group; and to perform localization diagnosis of the responsible vascular area.

[0133] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a clinical database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a stroke prediction method based on a cerebrovascular imaging characterization group.

[0134] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0135] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0136] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0137] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, clinical databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0138] The clinical databases involved in the embodiments provided in this application may include at least one type of relational clinical database and non-relational clinical database. Non-relational clinical databases may include, but are not limited to, blockchain-based distributed clinical databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0139] In this application, all actions to acquire signals, information, or data are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with the authorization granted by the owner of the relevant device.

[0140] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0141] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A stroke prediction method based on a cerebrovascular image feature set, characterized in that, The stroke prediction method based on the cerebrovascular image feature set comprises the following steps: A clinical database is constructed, and the clinical database comprises multi-modal image feature set data and corresponding clinical feature data of each patient, wherein the multi-modal image feature set data comprises TOF-MRA, SWI, T1WI and T2-FLAIR, and the clinical feature data comprises gender, age, height, weight and disease conditions; According to the multi-modal image feature set data, the corresponding clinical feature data and the corresponding patient clinical etiological typing of each patient, a single-hidden-layer simple full-connection neural network framework is used to obtain a clinical etiological typing prediction model; According to the TOF-MRA of each patient, a cerebrovascular tissue segmentation model and voxel-based morphological analysis are used to determine cerebrovascular image features, and the cerebrovascular image features are used to perform positioning diagnosis on the responsible blood vessel region; the cerebrovascular tissue segmentation model is used to obtain a cerebrovascular segmentation image according to the TOF-MRA; the cerebrovascular image features comprise cerebrovascular basic features, cerebrovascular geometric features and cerebrovascular topological features; the cerebrovascular basic features comprise the length and volume of individual tubular structures, the diameter and length of local brain regions; the cerebrovascular geometric features comprise the radius of curvature and bending degree of the cerebrovascular; the cerebrovascular topological features comprise cerebrovascular bifurcation points; the cerebrovascular tissue segmentation model is a deep neural network model in which an attention gate module is introduced in the skip connection of a 3D-UNet convolutional neural network; According to the cerebrovascular image features in the TOF-MRA of each patient, SWI, T1WI, T2-FLAIR and the corresponding clinical feature data, a single-hidden-layer simple full-connection neural network framework is used to obtain a stroke recurrence risk prediction model based on a cerebrovascular aging image feature set; the stroke recurrence risk prediction model based on the cerebrovascular aging image feature set is used to output a stroke patient recurrence probability value; Multi-modal image feature set data and corresponding clinical feature data of a to-be-predicted person are obtained; The clinical etiological typing prediction model and the stroke recurrence risk prediction model based on the cerebrovascular aging image feature set are used to perform clinical etiological typing prediction and stroke recurrence risk prediction, and positioning diagnosis on the responsible blood vessel region is performed; The cerebrovascular image features are determined according to the TOF-MRA of each patient, the cerebrovascular tissue segmentation model and the voxel-based morphological analysis, and the cerebrovascular image features are used to perform positioning diagnosis on the responsible blood vessel region, and the positioning diagnosis specifically comprises the following steps: Magnetic resonance sequence image preprocessing is performed on the TOF-MRA; the magnetic resonance sequence image preprocessing comprises brain tissue extraction, magnetic resonance sequence image registration, magnetic resonance sequence image noise reduction and B0 field correction, limited contrast adaptive histogram equalization, blood vessel edge signal enhancement and transform coefficient Gamma correction; The cerebrovascular tissue segmentation model is used to perform cerebrovascular tissue segmentation on the image after the magnetic resonance sequence image preprocessing, and a cerebrovascular segmentation image is obtained; A three-dimensional conditional random field model is used to perform three-dimensional conditional random field processing according to the cerebrovascular segmentation image. Determine the final cerebral vessel segmentation image according to the three-dimensional maximum connected component in the cerebral vessel segmentation image processed according to the three-dimensional conditional random field; Nonlinearly register the final cerebral vessel segmentation image with the cerebral vessel distribution density atlas to obtain a transformed Jacobian matrix; Perform Jacobian matrix standardization processing on the cerebral vessel segmentation image after nonlinear registration by using the transformed Jacobian matrix to obtain a cerebral vessel distribution density image; the cerebral vessel distribution density image is used to represent the distribution density of the individual cerebral vessel in the standard atlas space; Perform a single-sample t-test at the voxel level according to the cerebral vessel distribution density image and the cerebral vessel distribution density atlas, and perform multiple comparison correction by using the Bonferroni method to realize positioning diagnosis of the responsible vessel region; the responsible vessel region is a vessel region in which the individual vessel density has the largest signal difference in the cerebral vessel distribution density image. 2.The stroke prediction method based on the cerebrovascular image feature set according to claim 1, characterized in that, The method for constructing the clinical database specifically comprises: Preprocessing the multi-modal image feature group data and the corresponding clinical feature data of each patient; the preprocessing comprises: removing abnormal value data, multiple missing data filling, and data cleaning. 3.The stroke prediction method based on the brain vascular image feature set according to claim 1, characterized in that, The clinical etiological typing prediction model and the stroke recurrence risk prediction model based on the cerebral vessel imaging representation group are trained by using an early stopping method. 4.A cerebral stroke prediction device based on a cerebral vascular image feature set, which implements the cerebral stroke prediction method based on the cerebral vascular image feature set according to any one of claims 1-3, characterized in that, The stroke prediction device based on the cerebral vessel imaging representation group comprises: A clinical database construction module is configured to construct a clinical database; the clinical database comprises: multi-modal image feature group data and corresponding clinical feature data of each patient; the multi-modal image feature group data comprises: TOF-MRA, SWI, T1WI, and T2-FLAIR; and the clinical feature data comprises: gender, age, height, weight, and disease condition. A clinical etiological typing prediction model determination module is configured to obtain a clinical etiological typing prediction model by using a single-hidden-layer simple fully connected neural network framework according to the multi-modal image feature group data, the corresponding clinical feature data, and the corresponding patient clinical etiological typing of each patient. A positioning diagnosis module is configured to determine cerebral vessel imaging features by using a cerebral vessel tissue segmentation model and voxel-based morphological analysis according to TOF-MRA of each patient, and to perform positioning diagnosis of a responsible vessel region according to the cerebral vessel imaging features; the cerebral vessel tissue segmentation model is used to obtain a cerebral vessel segmentation image according to a time-of-flight magnetic resonance angiography image; the cerebral vessel imaging features comprise: cerebral vessel basic features, cerebral vessel geometric features, and cerebral vessel topological features; the cerebral vessel basic features comprise: length and volume of individual tubular structures, local brain region diameter and length; the cerebral vessel geometric features comprise: cerebral vessel curvature radius and bending degree; and the cerebral vessel topological features comprise: cerebral vessel bifurcation points. The stroke recurrence risk prediction model determination module is configured to determine a stroke recurrence risk prediction model based on the cerebral vascular image features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data of each patient in the TOF-MRA, and adopt a single-hidden-layer simple full-connection neural network framework to obtain the stroke recurrence risk prediction model based on the cerebral vascular aging image feature group; the stroke recurrence risk prediction model based on the cerebral vascular aging image feature group is configured to output a stroke patient recurrence probability value; The data acquisition module is configured to acquire the multi-modal image feature group data and corresponding clinical feature data of a to-be-predicted person. The stroke prediction module is configured to perform clinical etiological typing prediction and stroke recurrence risk prediction by adopting the clinical etiological typing prediction model and the stroke recurrence risk prediction model based on the cerebral vascular aging image feature group, and perform positioning diagnosis of the responsible blood vessel region.

5. A computer device comprising: The memory, the processor, and the computer program stored in the memory and executable on the processor are characterized in that the processor executes the computer program to implement the stroke prediction method based on the cerebral vascular image feature group according to any one of claims 1-3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the stroke prediction method based on the cerebral vascular image feature group according to any one of claims 1-3.

7. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the stroke prediction method based on the cerebral vascular image feature group according to any one of claims 1-3.

Images