Cerebral stroke prediction method and device based on cerebrovascular image representation group, medium and product

Abstract

The invention discloses a cerebral apoplexy prediction method and device based on a cerebrovascular image representation group, a medium and a product, and relates to the field of image processing, and the method comprises the steps: constructing a clinical database; according to the multi-modal image feature group data, the clinical feature data and the clinical cause typing of the patient, constructing a clinical cause typing prediction model; according to TOF-MRA, a cerebrovascular tissue segmentation model and voxel-based morphological analysis are adopted to determine cerebrovascular image features, and positioning diagnosis of a responsible blood vessel area is carried out according to the cerebrovascular image features; according to the cerebrovascular image features in the TOF-MRA, the SWI, the T1WI, the T2-FLAIR and the corresponding clinical feature data, a cerebral apoplexy recurrence risk prediction model is constructed; according to the application, the pathogenesis classification of the acute cerebral apoplexy patient can be quickly and accurately realized in a clinical scene, the responsible blood vessel can be accurately identified, and the recurrence risk prediction can be realized in the early stage of the disease.

Description

Technical Field

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

[0002] With the aging of the global population, stroke has become one of the diseases with the highest mortality and disability rates worldwide. The etiology classification (such as large artery atherosclerosis, cardioembolism, and small vessel occlusion) and responsible vessel localization of acute ischemic stroke are crucial for early diagnosis, treatment selection, and prognosis evaluation of patients. However, most of the existing etiology classification and responsible vessel localization methods are based on traditional clinical indicators and single-modality imaging data; such as computed tomography (CT) and conventional magnetic resonance imaging (MRI). Although these methods can provide certain structural information, they cannot fully capture the aging state and microlesions of the patient's cerebrovascular system. Especially for acute ischemic stroke, the precise localization of early lesions is crucial, and single-modality data often cannot meet this demand, resulting in insufficient diagnostic accuracy and reliability; and the subgroup of patients with recurrence shows a worse long-term prognosis. Therefore, it is of great clinical significance to achieve early etiology classification, responsible vessel localization, and objective and accurate quantitative analysis of the risk of recurrence of acute ischemic stroke.

[0003] Therefore, developing a method that can quantitatively evaluate cerebrovascular aging and achieve accurate classification and location of responsible vessels has important clinical significance and social value. Early identification of responsible vessels in acute ischemic stroke is crucial for the selection of treatment options.

[0004] Existing classification and localization models are often trained based on small-scale clinical data and lack the support of large-scale population cohorts. This makes the model's generalization ability poor and difficult to adapt to patient groups in different regions, limiting its promotion in actual clinical applications. In addition, existing responsible vessel localization methods usually rely on manual annotation or simple image processing technology, which is easily affected by subjective factors and difficult to accurately match the brain area where cerebral infarction occurs with the responsible blood vessels that supply it. Existing stroke classification methods are mainly based on traditional clinical indicators (such as age, blood pressure, blood lipids, etc.). Although these indicators can reflect some risk factors, they cannot directly assess the degree of cerebrovascular aging; cerebrovascular aging is an important precursor to the occurrence and development of stroke. Especially in the early stages, microvascular changes may be the earliest pathological features. However, existing evaluation methods cannot effectively capture these subtle changes, resulting in inaccurate and incomplete stroke classification and responsible vessel location. Early identification of the responsible vessel for acute ischemic stroke is crucial for the selection of treatment options. Existing responsible vessel localization methods usually rely on manual annotation or simple image processing technology, which is easily affected by subjective factors and difficult to accurately match the brain area where cerebral infarction occurs with the responsible vessel that supplies it. In addition, traditional feature extraction methods rely on manually designed algorithms, which are easily affected by subjective factors and difficult to fully utilize the complex information in multimodal images. Most existing stroke recurrence risk prediction models use a unified evaluation standard and fail to fully consider individual differences. The degree of cerebrovascular aging and recurrence risk of each patient may vary significantly, and the existing models cannot provide personalized prediction results, which affects the precise management of patients by clinicians.

[0005] How to quickly and accurately classify the causes of acute stroke patients in clinical scenarios, accurately identify the responsible blood vessels, predict the risk of recurrence in the early stages of the disease, and assist clinical workers in achieving individualized and precise diagnosis and treatment is an urgent problem that needs to be solved. Summary of the invention

[0006] The purpose of this application is to provide a stroke prediction method, device, medium and product based on a cerebrovascular image characterization group, which can quickly and accurately classify the causes of acute stroke patients in clinical scenarios, accurately identify responsible blood vessels, and predict recurrence risk in the early stages of the disease.

[0007] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a method for predicting stroke based on a cerebrovascular image characterization group, the method for predicting stroke based on a cerebrovascular image characterization group comprising: Constructing a clinical database; the clinical database includes: multimodal imaging feature group data and corresponding clinical feature data of 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 condition; According to each patient's multimodal imaging feature group data, corresponding clinical feature data and corresponding patient clinical etiology classification, a simple fully connected neural network architecture with a single hidden layer was used to obtain a clinical etiology classification prediction model; According to the TOF-MRA of each patient, the cerebrovascular tissue segmentation model and voxel-based morphological analysis are used to determine the cerebrovascular image characteristics, and the positioning diagnosis of the responsible vascular area is performed according to the cerebrovascular image characteristics; the cerebrovascular tissue segmentation model is used to obtain the cerebrovascular segmentation image according to TOF-MRA; the cerebrovascular image characteristics include: basic cerebrovascular characteristics, cerebrovascular geometric characteristics and cerebrovascular topological characteristics; the basic cerebrovascular characteristics include: the length and volume of individual tubular structures, the diameter and length of local brain areas; the cerebrovascular geometric characteristics include: the cerebrovascular curvature radius and the degree of bending; the cerebrovascular topological characteristics include: cerebrovascular bifurcation points; According to the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data in TOF-MRA of each patient, 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 the stroke patient; Acquire multimodal imaging feature group data and corresponding clinical feature data of the subject to be predicted; 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 localization diagnosis of responsible vascular areas was performed.

[0008] Optionally, the building of a clinical database specifically includes: The multimodal imaging feature group data and the corresponding clinical feature data of each patient are preprocessed; the preprocessing includes: removing outlier data, multiple filling of missing data and data cleaning.

[0009] Optionally, the clinical etiology classification prediction model and the stroke recurrence risk prediction model based on the cerebrovascular aging imaging characterization group are both trained by the early stopping training method.

[0010] Optionally, the method of determining the cerebrovascular imaging features based on the TOF-MRA of each patient by using a cerebrovascular tissue segmentation model and voxel-based morphological analysis, and performing localization diagnosis of the responsible vascular region based on the cerebrovascular imaging features, specifically includes: Performing magnetic resonance sequence image preprocessing on TOF-MRA; the magnetic resonance sequence image preprocessing includes: 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 transformation coefficient Gamma correction; Using the cerebrovascular tissue segmentation model, the cerebrovascular tissue is segmented on the image preprocessed by the magnetic resonance sequence image to obtain the cerebrovascular segmentation image; According to the cerebral vascular segmentation image, a three-dimensional conditional random field model is used to perform three-dimensional conditional random field processing; Determine the final cerebrovascular segmentation image according to the three-dimensional maximum connected component in the cerebrovascular segmentation image after the three-dimensional conditional random field processing; The final cerebrovascular segmentation image is nonlinearly registered with the cerebrovascular distribution density map to obtain the transformed Jacobian matrix; Using the transformed Jacobian matrix to perform Jacobian matrix normalization processing on the cerebrovascular segmentation image after nonlinear registration, to obtain a cerebrovascular distribution density image; the cerebrovascular distribution density image is used to characterize the distribution density of individual cerebrovascular vessels in the standard atlas space; A single-sample t-test at the voxel level was performed based on the cerebral vascular distribution density image and the cerebral vascular distribution density atlas, and the Bonferroni method was used for multiple comparison correction to achieve the localization diagnosis of the responsible vascular area; the responsible vascular area was the vascular area with the largest signal difference in the cerebral vascular distribution density image among the individual vascular density distributions.

[0011] Optionally, the cerebrovascular tissue segmentation model is a deep neural network model that introduces an attention gating module into the jump connection of the 3D-UNet convolutional neural network.

[0012] Optionally, the stroke recurrence risk prediction model based on the cerebrovascular aging image characterization group includes: a convolutional neural network encoder, a cerebrovascular image feature fusion module and a vascular biological age prediction module; The convolutional neural network encoder is used to extract convolutional features of SWI, T1WI and T2-FLAIR by combining asymmetric convolution modules and Squeeze-and-Excitation modules; The cerebrovascular image feature fusion module is used to compress the convolution features by a global pooling compression method to obtain a convolution feature group of SWI, T1WI and T2-FLAIR with each feature dimension of 1, and directly connect the demographic indicators and cerebrovascular image features with the convolution feature group to generate a comprehensive image feature group; the comprehensive image feature group includes the cerebrovascular image features of TOF-MRA that specifically reflect the cerebrovascular characteristics, and the convolution feature groups of SWI, T1WI and T2-FLAIR that reflect the brain aging characteristics; the convolution feature group includes: white matter high signal volume, lacunar number, cortical thickness, gray matter volume, microbleeding and ventricular volume; the demographic indicators include: age and gender; The vascular biological age prediction module is used to comprehensively predict the patient's brain aging age based on the comprehensive image feature group, and compare the comprehensive prediction result with the individual's physical age to obtain a comprehensive cerebrovascular aging index.

[0013] In a second aspect, the present application provides a stroke prediction device based on a cerebrovascular image characterization group, wherein the stroke prediction device based on a cerebrovascular image characterization group comprises: A clinical database construction module is used to construct a clinical database; the clinical database includes: multimodal imaging feature group data of each patient and corresponding clinical feature data; 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 condition; A 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 of each patient, the corresponding clinical feature data and the corresponding patient clinical etiology classification, using a simple fully connected neural network architecture with a single hidden layer; A positioning diagnosis module is used to determine the cerebrovascular image features according to the TOF-MRA of each patient by using a cerebrovascular tissue segmentation model and voxel-based morphological analysis, and to perform positioning diagnosis of the responsible vascular area according to the cerebrovascular image features; the cerebrovascular tissue segmentation model is used to obtain a cerebrovascular segmentation image according to TOF-MRA; the cerebrovascular image features include: basic cerebrovascular features, cerebrovascular geometric features, and cerebrovascular topological features; the basic cerebrovascular features include: the length and volume of individual tubular structures, and the diameter and length of local brain regions; the cerebrovascular geometric features include: the cerebrovascular curvature radius and the degree of bending; the cerebrovascular topological features include: cerebrovascular bifurcation points; A stroke recurrence risk prediction model determination module is used to obtain a stroke recurrence risk prediction model based on a cerebrovascular aging image characterization group according to the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data in TOF-MRA of each patient, using a simple fully connected neural network architecture with a single hidden layer; the stroke recurrence risk prediction model based on a cerebrovascular aging image characterization group is used to output a stroke patient recurrence probability value; A data acquisition module, used to acquire multimodal imaging feature group data and corresponding clinical feature data of the subject to be predicted; The stroke prediction module is used to perform clinical etiology classification prediction and stroke recurrence risk prediction using a clinical etiology classification prediction model and a stroke recurrence risk prediction model based on a cerebrovascular aging imaging characterization group; and to perform localization diagnosis of the responsible vascular area.

[0014] In a third aspect, the present application provides a computer device, comprising: 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 a cerebrovascular image characterization group.

[0015] In a fourth aspect, the present 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 a cerebrovascular image characterization group.

[0016] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the stroke prediction method based on a cerebrovascular image characterization group.

[0017] According to the specific embodiments provided in this application, this application has the following technical effects: The present application provides a method, device, medium and product for predicting stroke based on a cerebrovascular image characterization group. By constructing a clinical database including multimodal image feature group data and corresponding clinical feature data of each patient, a comprehensive cerebrovascular image feature characterization group is constructed by segmenting the feature data automatically extracted by time-of-flight magnetic resonance angiography (TOF-MRA), which comprehensively captures the aging state and microlesions of cerebrovascular vessels; by fusing susceptibility weighted imaging (SWI), T1 weighted imaging (T1WI) and T2 fluid attenuated inversion recovery imaging (T2Fluid Attenuated Inversion Recovery Imaging). Recovery, T2-FLAIR) and convolutional feature groups reflecting brain aging characteristics, combined with the features automatically extracted from TOF-MRA cerebral vessels, to construct a stroke recurrence risk prediction model based on cerebrovascular aging image characterization group. The stroke recurrence risk prediction model based on cerebrovascular aging image characterization group can more comprehensively capture the patient's overall health status, especially the aging state of cerebrovascular vessels, so as to reflect the important indicators of cerebrovascular pathophysiological state, and thus achieve the accuracy of clinical risk classification prediction; by quantitatively evaluating the degree of cerebrovascular aging, reflecting the structural changes of large blood vessels, and capturing the aging characteristics of microvessels, a more comprehensive evaluation perspective is provided. By comparing with the individual's physical age, the cerebrovascular aging image characterization group can accurately estimate the patient's cerebrovascular biological age and reveal the potential risk of vascular aging. The fusion of multimodal data in the clinical database not only improves the accuracy of the model, but also enhances its robustness and generalization ability; according to the multimodal image feature group data of the subject to be predicted, based on voxel-based morphological analysis, the location diagnosis of the responsible vascular area is performed, which can accurately match the brain area with cerebral infarction and the responsible blood vessel supplying it. This application provides personalized stroke classification and responsible vessel location results and individualized prediction of patient recurrence risk based on the multimodal imaging feature group data of the subject to be predicted and the corresponding clinical feature data. Based on the comprehensive information of the patient's cerebrovascular aging degree, this application performs etiology classification and responsible vessel location on stroke patients at the individual level, and finally predicts the patient's recurrence risk by combining the patient's cerebrovascular aging degree, which is helpful to formulate personalized treatment plans, improve the patient's prognosis quality, and help clinical workers understand the deep pathological mechanism of stroke onset and recurrence. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 The present invention is a flowchart of a method for predicting stroke based on a cerebrovascular image characterization group in one embodiment of the present application. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0021] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0022] In an exemplary embodiment, Figure 1 As shown, a method for predicting stroke based on a cerebrovascular image characterization group is provided, and the method includes the following S101 to S106. Among them: S101, constructing a clinical database; the clinical database includes: multimodal imaging feature group data and corresponding clinical feature data of 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 condition.

[0023] S101 specifically includes: The multimodal imaging feature group data and the corresponding clinical feature data of each patient are preprocessed; the preprocessing includes: removing outlier data, multiple filling of missing data and data cleaning.

[0024] As a specific embodiment, outliers are detected according to the Median Absolute Deviation (MAD) method, outlier data are removed, and multiple fillings are performed on samples with missing data; finally, a standard data frame structure data is constructed to complete data cleaning and sorting.

[0025] S102: Based on the multimodal imaging feature group data, the corresponding clinical feature data and the corresponding clinical etiology classification of each patient, a simple fully connected neural network architecture with a single hidden layer (One-Hidden-Layer) is used to obtain a clinical etiology classification prediction model.

[0026] The loss function CE (Loss Function) of the clinical etiology classification prediction model adopts the cross entropy loss function (softmax cross-entropy loss function), and the specific formula is as follows: .

[0027] Among them, f(s) i represents the result of the linear combination of variables in the multimodal imaging feature group data of the i-th patient after being transformed by the softmax function, t i represents the actual etiology classification variable value of the i-th patient, N represents the number of samples, and C represents the number of etiology classifications.

[0028] The specific form of the Softmax function is: .

[0029] The data in the clinical database constitute a data set, and are randomly sampled into a training set and a test set in a ratio of 7:3. The training set is used for ten-fold cross-validation training, and the optimal model is selected as the final output model, that is, the clinical etiology classification prediction model; it is verified in the test set. In order to prevent overfitting, the early stopping method is used in training to fully train the clinical etiology classification prediction model, which greatly reduces the risk of model overfitting. The final output of the clinical etiology classification prediction model is the probability value of each clinical etiology of the patient, among which the etiology corresponding to the maximum probability is the final clinical etiology classification prediction value of the patient.

[0030] S103, according to the TOF-MRA of each patient, the cerebrovascular tissue segmentation model and voxel based morphometry (VBM) analysis are used to determine the cerebrovascular image characteristics, and the positioning diagnosis of the responsible vascular area is performed based on the cerebrovascular image characteristics; the cerebrovascular tissue segmentation model is used to obtain the cerebrovascular segmentation image according to TOF-MRA; the cerebrovascular image characteristics include: basic cerebrovascular characteristics, cerebrovascular geometric characteristics and cerebrovascular topological characteristics; the basic cerebrovascular characteristics include: the length and volume of individual tubular structures, the diameter and length of local brain areas; the cerebrovascular geometric characteristics include: the cerebrovascular curvature radius and the degree of bending; the cerebrovascular topological characteristics include: cerebrovascular bifurcation points.

[0031] Before extracting the basic features of cerebrovascular vessels in TOF-MRA, the following steps are also included: centerline estimation, normal plane estimation and three-dimensional reconstruction. The specific process includes: S11, firstly, the centerline of the cerebrovascular segmentation image of the 3D tubular tree structure of the cerebrovascular is estimated. Based on the Voronoi covariance measure (VCM), the centerline of the cerebrovascular segmentation is stably extracted to reduce the center point identification drift and voxel discontinuity problems at the cerebrovascular bifurcation.

[0032] S12, based on the extraction of cerebral vascular centerline, the λ-maximal segment tangent method (λ-MST) is used to calculate the tangent corresponding to each voxel on the centerline, so as to further calculate the normal plane on each centerline voxel, and the normal plane of each centerline voxel is robustly estimated by weighted averaging of the normal planes of adjacent voxels.

[0033] S13, for each voxel on the center line, fit a maximum circle in its normal plane, and calculate the diameter corresponding to the voxel.

[0034] S14, combining the centerline and diameter information to achieve three-dimensional reconstruction of the tubular tree structure, thereby effectively preventing the local minor errors that may be introduced by the structure segmentation algorithm from being cascaded and amplified in the vascular feature estimation, thereby ensuring the robustness of the three-dimensional reconstruction.

[0035] The process of extracting basic features of cerebral blood vessels is as follows: S21, count the number of voxels with positive labels in the 3D image of cerebrovascular segmentation to obtain the absolute length of cerebrovascular vessels; at the same time, considering that the size of individual brain tissue can significantly affect the length of cerebrovascular structure, this algorithm divides the absolute length by the volume of individual brain tissue to obtain the relative length of cerebrovascular vessels.

[0036] S22, count the number of voxels with positive labels in the 3D image of cerebrovascular segmentation, and multiply it by the voxel volume to obtain the absolute volume of cerebrovascular vessels; at the same time, considering that the size of individual brain tissue can also significantly affect the volume of cerebrovascular structure, this algorithm also divides the absolute volume of cerebrovascular vessels by the size of individual brain tissue to obtain the relative volume, reflecting the average distribution density of cerebrovascular tissue in the whole brain tissue.

[0037] S23, using a two-step registration method to nonlinearly register the MNI standard brain template with the cerebrovascular MRI sequence images, and convert the Harvard-Oxford brain partition map in the MNI standard brain space to the individual level of MRI image space. The local vascular length and diameter were calculated in each brain partition to reflect the heterogeneity of the basic characteristics of blood vessels in the spatial distribution of each brain region.

[0038] The extraction process of cerebral vascular geometric features is as follows: In the cerebrovascular segmentation centerline image, the curvature radius and curvature degree of each voxel point of the centerline in three-dimensional space are calculated according to the tangent vector flip angle and the Frenet frame. The average vascular curvature radius and curvature degree of each brain region corresponding to the Harvard-Oxford brain partition map are calculated to measure the geometric brain distribution characteristics of cerebrovascular.

[0039] The extraction process of cerebrovascular topological features is as follows: S31, based on the cerebral blood vessel segmentation centerline image, the tree structure is used to characterize the adjacency relationship of each centerline voxel, so as to realize automatic identification of bifurcation points in the blood vessel tree structure, and count the number of bifurcations and their spatial distribution.

[0040] S32, calculating the angle between the tangent vectors of the adjacent voxels at the bifurcation point, which is the bifurcation angle of the bifurcation point; S33, calculate the average bifurcation angle and bifurcation density in each brain region to reflect the distribution characteristics of vascular bifurcation statistics in the brain region space.

[0041] S103 specifically includes: S311, performing magnetic resonance sequence image preprocessing on the multimodal image feature group data; the magnetic resonance sequence image preprocessing includes: 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 transformation coefficient Gamma correction.

[0042] Magnetic resonance image preprocessing specifically includes: S100, brain tissue extraction: perform brain tissue-specific identification and segmentation on TOF-MRA, and remove non-brain tissue structures such as scalp, skull, facial features and neck. The purpose of this step is to eliminate the interference of non-brain tissue structure image signals on subsequent processing.

[0043] S200, MRI sequence image registration: The T1-weighted imaging and TOF-MRA of the same individual are registered using a linear rigid body transformation with 6 degrees of freedom and spline interpolation, and resampled to the same image space. The purpose of this step is to transform each sequence image at different voxel coordinates to the same anatomical coordinates to facilitate the integration of cross-sequence signal information.

[0044] S300, Magnetic resonance sequence image denoising and B0 field correction: Use the adaptive non-local mean algorithm to perform image denoising on TOF-MRA to remove Gaussian noise that may exist in the magnetic resonance sequence images. Then, the signal intensity remapping method after Gaussian deconvolution is used to remove the magnetic resonance sequence image artifacts caused by B0 field inhomogeneity. The purpose of this step is to remove common artifact signals in magnetic resonance sequence images and improve the image signal-to-noise ratio.

[0045] S400, limited contrast adaptive histogram equalization: Due to the principle limitation of vascular imaging magnetic resonance sequences such as TOF-MRA, the signal enhancement amplitude of small blood vessels will be significantly lower than that of large blood vessels, making it difficult to identify small blood vessels. The limited contrast adaptive histogram equalization method is used to adaptively enhance local cerebrovascular signals. This step allows cerebrovascular signals of different scales to obtain similar enhancement amplitudes.

[0046] S500, blood vessel edge signal enhancement: Due to the partial volume effect, the blood vessel edge signal is significantly lower than the blood vessel center signal, making it difficult to identify. The Sobel convolution method is used to identify the edge position where the signal intensity mutation occurs in the image, achieve specific enhancement of the blood vessel edge signal, and improve the contrast between the blood vessel and the surrounding non-vascular tissue. This step allows the cerebral vascular signals at different distances from the center line to obtain similar enhancement.

[0047] S600, Gamma correction: Finally, the Gamma correction method is used to nonlinearly increase the vascular tissue signal (high signal) and suppress the non-vascular tissue signal (low signal) of the magnetic resonance sequence image by adjusting the transformation coefficient Gamma. The transformation coefficient Gamma is a processing hyperparameter that can control the relative amplitude of the enhanced and suppressed signals according to the nature of the processed image to obtain better image contrast.

[0048] S332, using the cerebrovascular tissue segmentation model, performing cerebrovascular tissue segmentation on the multimodal image feature group data after preprocessing of the magnetic resonance sequence images to obtain a cerebrovascular segmentation image.

[0049] The training and testing data set for the cerebrovascular tissue segmentation model is determined as follows: T1WI and TOF-MRA of 100 healthy people acquired by different equipment using routine clinical scanning sequences are collected. After preprocessing the image data, two neuroradiologists with rich experience in image annotation independently annotate the cerebrovascular tissue signals in the magnetic resonance images. A third senior neuroradiologist conducts a comprehensive inspection and evaluation of the annotation results to complete the annotation of cerebrovascular tissue signals in multimodal magnetic resonance image data to ensure the consistency and accuracy of data annotation, and finally construct a training and testing data set for a supervised model. Among them, cerebrovascular tissue signals are specific signal features in MRI that can identify and distinguish cerebrovascular tissues. The specific signal features mainly include low signals (flowing blood) in T1WI and high signals (flowing blood) in TOF-MRA. Through the combination of multimodal images and expert annotation, a high-quality data set for training and testing cerebrovascular tissue segmentation models is constructed.

[0050] The cerebrovascular tissue segmentation model is a deep neural network model that introduces an attention gating module into the jump connection of the 3D-UNet convolutional neural network. The overall 3D-UNet convolutional neural network adopts an adaptive design, which automatically adjusts hyperparameters such as batch size and network depth width according to the spatial resolution and size of the input image. At the same time, the model introduces an attention gating module into the jump connection of the 3D-UNet convolutional neural network, and its basic form can be expressed as: .

[0051] .

[0052] Among them, f i , g i They represent the feature tensors of the encoding module and the decoding module connected by the jump, σ1 represents the RELU activation function, σ2 represents the Sigmoid activation function, φ, w f , w g represents the linear transformation coefficient, b φ and b g Represents the bias coefficient. Finally, the resampled SAct image is combined with the feature tensor f on the decoding module. i Multiplying together can fuse feature information at different scales and reduce background noise to a great extent.

[0053] The model was trained and tested using a five-fold cross validation method. In order to address the differences in specificity and sensitivity of vascular segmentation results at different scales during vascular segmentation, the Tversky function was introduced as the loss function of the model. Based on the Tversky function, a weighted Tversky loss was further proposed, which uses the inverse of the Euclidean distance of the voxel from the center line as the mismatch loss weight to strengthen the topological structure restrictions of the segmentation results, so that the vascular segmentation results have a more consistent performance at each scale.

[0054] S333, based on the cerebrovascular segmentation image, a three-dimensional conditional random field model is used to perform three-dimensional conditional random field processing; the three-dimensional conditional random field model takes into account the continuity of vascular tissue in space, and by establishing potential functions between two points and on a single point, adjacent points with similar signal intensities on the magnetic resonance image with different segmentation labels on the cerebrovascular segmentation image are penalized, thereby obtaining a more continuous and reasonable cerebrovascular segmentation image in three-dimensional space.

[0055] S334, determining a final cerebral blood vessel segmentation image according to the three-dimensional maximum connected component in the cerebral blood vessel segmentation image after the three-dimensional conditional random field processing; that is, removing unconnected small blocks of signal noise in the three-dimensional space.

[0056] S335, nonlinearly aligning the final cerebral vascular segmentation image with the cerebral vascular distribution density map to obtain a transformed Jacobian matrix.

[0057] S336, using the transformed Jacobian matrix to perform Jacobian matrix normalization processing on the cerebrovascular segmentation image after nonlinear registration to obtain a cerebrovascular distribution density image; the cerebrovascular distribution density image is used to characterize the distribution density of individual cerebrovascular vessels in the standard atlas space.

[0058] S337, a single-sample t-test at the voxel level was performed based on the cerebral vascular distribution density image and the cerebral vascular distribution density atlas, and the Bonferroni method was used for multiple comparison correction to achieve the localization diagnosis of the responsible vascular area; the responsible vascular area is the vascular area with the largest signal difference in the cerebral vascular distribution density image.

[0059] S104 obtains a stroke recurrence risk prediction model based on the cerebrovascular aging image characterization group according to the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data in TOF-MRA 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 image characterization group is used to output the recurrence probability value of the stroke patient.

[0060] Among them, T1WI, T2-FLAIR and SWI reflect the overall characteristics of various tissue structures such as white matter and gray matter in the individual's brain. Therefore, they can reflect the impact of vascular aging on these brain tissue structures and contain rich information on cerebrovascular aging.

[0061] The stroke recurrence risk prediction model adopts the cross entropy loss function, and the specific formula is as follows: .

[0062] Among them, y （i） The binary variable value representing whether the ith patient actually relapses or not, x （i） represents the variables in the i-th patient image representation group, m represents the total sample size, h θ (x （i） ) represents the result predicted by the model for the i-th patient.

[0063] The data set was randomly sampled into a training set and a test set in a ratio of 7:3; the training set was used for ten-fold cross-validation training, and the optimal model was selected as the stroke recurrence risk prediction model; and it was verified in the test set. In order to prevent overfitting, the early stopping method was used in training to fully train the stroke recurrence risk prediction model, greatly reducing the risk of model overfitting. The final output of the stroke recurrence risk prediction model is the final recurrence probability value of the patient; the stroke recurrence risk prediction model can quantitatively and objectively measure the relationship between the degree of cerebrovascular aging in the patient's imaging and the risk of recurrence of the patient's stroke prognosis.

[0064] The stroke recurrence risk prediction model based on the cerebrovascular aging image characterization group includes: a convolutional neural network encoder, a cerebrovascular image feature fusion module and a vascular biological age prediction module.

[0065] The convolutional neural network encoder is used to extract convolution features of SWI, T1WI and T2-FLAIR by combining an asymmetric convolution module (Asymmetric Convolution) and a squeeze-and-excitation module; that is, it takes advantage of the ability of the neural network to adaptively extract feature encoding to automatically and comprehensively extract the vascular aging features of SWI, T1WI and T2-FLAIR.

[0066] The asymmetric convolution module improves the image receptive field of the convolution kernel. The specific formula is as follows: .

[0067] in, f is the input feature map, gis the output feature map, Conv1×1×k, conv1×k×1 and convk×1×1, convk×k×k are the convolution kernels of 1×1×k, 1×k×1, k×1×1 and k×k×k respectively.

[0068] The Squeeze-and-Excitation module can introduce an attention mechanism into the neural network and adaptively adjust the weight parameters of each feature through the correlation between different convolutional features. The specific formula is as follows: .

[0069] Among them, SE (F) is the output result of the Squeeze-and-Excitation module, F is the input feature map, σ is the sigmoid nonlinear activation function, ReLU is the RectifiedLieanerUnit nonlinear activation function, W1 is the linear parameter matrix of the input feature map after global pooling compression, b1 is the bias term on 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.

[0070] GlobalAvgPool represents global pooling compression, which averages the width and height of the input feature map. The specific formula is: .

[0071] in, is the value after pooling of channel i, H is the height of the input feature map, W is the width of the input feature map, j is the number of rows of the feature map in channel i, k is the number of columns of the feature map in channel k, is the pixel value of the feature map at position (j, k) of channel i.

[0072] By connecting two asymmetric convolution modules with the Squeeze-and-Excitation module in the convolutional neural network encoder, the convolution features of SWI, T1WI and T2-FLAIR are extracted.

[0073] The cerebrovascular image feature fusion module is used to compress the convolution features through a global pooling compression method to obtain a convolution feature group of SWI, T1WI and T2-FLAIR with each feature dimension of 1, and directly connect the demographic indicators and cerebrovascular image features with the convolution feature group to generate a comprehensive image feature group; the comprehensive image feature group includes the cerebrovascular image features of TOF-MRA that specifically reflect cerebrovascular characteristics, and the convolution feature group of SWI, T1WI and T2-FLAIR that reflect brain aging characteristics; the convolution feature group includes: white matter high signal volume, lacunar number, cortical thickness, gray matter volume, microbleeding and ventricular volume; the demographic indicators include: age and gender.

[0074] The vascular biological age prediction module is used to comprehensively predict the patient's brain aging age based on the comprehensive image feature group, and compare the comprehensive prediction result with the individual's physical age to obtain a comprehensive cerebrovascular aging index.

[0075] The loss function of the vascular biological age prediction module adopts the MSE loss function , the specific formula is as follows: .

[0076] in, Estimation of vascular age, y i represents the actual physical age of the patient, and N represents the total sample size.

[0077] The comprehensive cerebrovascular aging index proposed in this application based on the stroke recurrence risk prediction model of the cerebrovascular aging imaging characterization group can use multimodal magnetic resonance imaging data to quantitatively and objectively measure the degree of brain aging in patients. At the same time, the effect coefficient corresponding to each imaging feature is obtained in the model. This coefficient can quantitatively evaluate the contribution of the imaging feature to the overall cerebrovascular aging imaging characterization group, indicating the imaging feature that has the greatest impact on vascular aging, which has a strong clinical indication significance.

[0078] S105, obtaining multimodal image feature group data and corresponding clinical feature data of the subject to be predicted.

[0079] S106, the vascular biological age prediction module is used to comprehensively predict the patient's brain aging age based on the comprehensive image feature group, and compare the comprehensive prediction result with the individual's physical age to obtain a comprehensive cerebrovascular aging index.

[0080] This application combines multimodal magnetic resonance imaging (MRI) and feature data automatically extracted from cerebrovascular segmentation to construct a comprehensive image characterization group. By fusing brain aging features in SWI, T1WI, and T2-FLAIR, combined with cerebrovascular features automatically extracted from TOF-MAR, the model can more comprehensively capture the patient's overall health status, especially the aging state of cerebrovascular vessels, so as to use these important indicators that can reflect the pathophysiological state of cerebrovascular vessels to achieve clinical risk classification prediction. This multimodal data fusion not only improves the accuracy of the model, but also enhances its robustness and generalization ability. The concept of cerebrovascular aging image characterization group is introduced, and the vascular aging features are automatically extracted through neural networks, and the degree of cerebrovascular aging is quantitatively evaluated using multimodal image data. This index not only reflects the structural changes of large blood vessels, but also captures the aging characteristics of microvessels, providing a more comprehensive evaluation perspective. By comparing with the individual's physical age, the cerebrovascular aging image characterization group can accurately estimate the patient's cerebrovascular biological age and reveal the potential risks of vascular aging. The individual-level cerebrovascular segmentation images were registered with the group cerebrovascular distribution density map by nonlinear registration, and the responsible vessels were located using the difference matrix after Jacobian matrix normalization. This method can not only accurately locate the brain area with cerebral infarction and the responsible vessels supplying blood, but also quantitatively evaluate the differences between the individual vascular distribution network density and the local or overall level between the groups, greatly improving the accuracy and reliability of responsible vessel location. The model was trained based on a large population cohort clinical database covering 20,000 stroke patients followed up by the team for a long time, ensuring the generalization ability and reliability of the model. Through strict outlier detection and data cleaning, the model can perform well in different regions and different types of patient groups, and has broad application prospects. In order to prevent the model from overfitting, ten-fold cross validation and early stopping strategies were used for training. These methods not only ensure the stability of the model, but also greatly reduce the risk of overfitting and improve the predictive performance of the model. According to the image representation group and clinical characteristics of each patient, a personalized stroke recurrence risk score is provided. It helps clinicians better understand the patient's condition, formulate personalized treatment plans, and improve the quality of patient prognosis.

[0081] Based on the same inventive concept, the embodiment of the present application also provides a stroke prediction device based on a cerebrovascular image characterization group for implementing the above-mentioned stroke prediction method based on a cerebrovascular image characterization group. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the one or more embodiments of the stroke prediction device based on a cerebrovascular image characterization group provided below can refer to the limitations of the stroke prediction method based on a cerebrovascular image characterization group in the above text, and will not be repeated here.

[0082] In an exemplary embodiment, a stroke prediction device based on a cerebrovascular image characterization group is provided, comprising: 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 of 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 condition.

[0083] The clinical etiology classification prediction model determination module is used to obtain the clinical etiology classification prediction model based on the multimodal imaging feature group data of each patient, the corresponding clinical feature data and the corresponding patient clinical etiology classification, using a simple fully connected neural network architecture with a single hidden layer.

[0084] A positioning diagnosis module is used to determine the cerebrovascular imaging features according to the TOF-MRA of each patient, using a cerebrovascular tissue segmentation model and voxel-based morphological analysis, and to perform positioning diagnosis of the responsible vascular area according to the cerebrovascular imaging features; the cerebrovascular tissue segmentation model is used to obtain a cerebrovascular segmentation image according to the time-of-flight magnetic resonance angiography image; the cerebrovascular imaging features include: basic cerebrovascular features, cerebrovascular geometric features and cerebrovascular topological features; the basic cerebrovascular features include: the length and volume of individual tubular structures, the diameter and length of local brain areas; the cerebrovascular geometric features include: the cerebrovascular curvature radius and the degree of bending; the cerebrovascular topological features include: cerebrovascular bifurcation points.

[0085] The stroke recurrence risk prediction model determination module is used to obtain a stroke recurrence risk prediction model based on the cerebrovascular aging image characterization group according to the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data in the TOF-MRA 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 image characterization group is used to output the recurrence probability value of the stroke patient.

[0086] The data acquisition module is used to obtain the multimodal imaging feature group data and the corresponding clinical feature data of the subject to be predicted.

[0087] The stroke prediction module is used to perform clinical etiology classification prediction and stroke recurrence risk prediction using a clinical etiology classification prediction model and a stroke recurrence risk prediction model based on a cerebrovascular aging imaging characterization group; and to perform localization diagnosis of the responsible vascular area.

[0088] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a clinical database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for predicting stroke based on a cerebrovascular image characterization group is implemented.

[0089] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0090] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0091] 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, stored data, displayed data, 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 relevant data must comply with relevant regulations.

[0092] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, clinical database 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), magnetoresistive 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 may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0093] The clinical database involved in each embodiment provided in this application may include at least one of a relational clinical database and a non-relational clinical database. The non-relational clinical database may include a distributed clinical database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0094] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0095] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, 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.

[0096] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for predicting stroke based on a cerebrovascular imaging characterization group, characterized in that: The stroke prediction method based on the cerebrovascular image characterization group includes: Constructing a clinical database; the clinical database includes: multimodal imaging feature group data and corresponding clinical feature data of 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 condition; According to each patient's multimodal imaging feature group data, corresponding clinical feature data and corresponding patient clinical etiology classification, a simple fully connected neural network architecture with a single hidden layer was used to obtain a clinical etiology classification prediction model; According to the TOF-MRA of each patient, the cerebrovascular tissue segmentation model and voxel-based morphological analysis are used to determine the cerebrovascular image characteristics, and the positioning diagnosis of the responsible vascular area is performed according to the cerebrovascular image characteristics; the cerebrovascular tissue segmentation model is used to obtain the cerebrovascular segmentation image according to TOF-MRA; the cerebrovascular image characteristics include: basic cerebrovascular characteristics, cerebrovascular geometric characteristics and cerebrovascular topological characteristics; the basic cerebrovascular characteristics include: the length and volume of individual tubular structures, the diameter and length of local brain areas; the cerebrovascular geometric characteristics include: the cerebrovascular curvature radius and the degree of bending; the cerebrovascular topological characteristics include: cerebrovascular bifurcation points; According to the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data in TOF-MRA of each patient, 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 the stroke patient; Acquire multimodal imaging feature group data and corresponding clinical feature data of the subject to be predicted; 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 localization diagnosis of responsible vascular areas was performed.

2. The method for predicting stroke based on cerebrovascular image characterization group according to claim 1, characterized in that: The construction of the clinical database specifically includes: The multimodal imaging feature group data and the corresponding clinical feature data of each patient are preprocessed; the preprocessing includes: removing outlier data, multiple filling of missing data and data cleaning.

3. The method for predicting stroke based on cerebrovascular image characterization group according to claim 1, characterized in that: The clinical etiology classification prediction model and the stroke recurrence risk prediction model based on the cerebrovascular aging imaging characterization group are both trained by the early stopping training method.

4. The method for predicting stroke based on cerebrovascular image characterization group according to claim 1, characterized in that: The method uses a cerebrovascular tissue segmentation model and voxel-based morphological analysis based on the TOF-MRA of each patient to determine the cerebrovascular imaging features, and performs localization diagnosis of the responsible vascular area based on the cerebrovascular imaging features, specifically including: Performing magnetic resonance sequence image preprocessing on TOF-MRA; the magnetic resonance sequence image preprocessing includes: 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 transformation coefficient Gamma correction; Using the cerebrovascular tissue segmentation model, the cerebrovascular tissue is segmented on the image preprocessed by the magnetic resonance sequence image to obtain the cerebrovascular segmentation image; According to the cerebral vascular segmentation image, a three-dimensional conditional random field model is used to perform three-dimensional conditional random field processing; Determine the final cerebrovascular segmentation image according to the three-dimensional maximum connected component in the cerebrovascular segmentation image after the three-dimensional conditional random field processing; The final cerebrovascular segmentation image is nonlinearly registered with the cerebrovascular distribution density map to obtain the transformed Jacobian matrix; Using the transformed Jacobian matrix to perform Jacobian matrix normalization processing on the cerebrovascular segmentation image after nonlinear registration, to obtain a cerebrovascular distribution density image; the cerebrovascular distribution density image is used to characterize the distribution density of individual cerebrovascular vessels in the standard atlas space; A single-sample t-test at the voxel level was performed based on the cerebral vascular distribution density image and the cerebral vascular distribution density atlas, and the Bonferroni method was used for multiple comparison correction to achieve the localization diagnosis of the responsible vascular area; the responsible vascular area was the vascular area with the largest signal difference in the cerebral vascular distribution density image among the individual vascular density distributions.

5. The method for predicting stroke based on cerebrovascular image characterization group according to claim 4, characterized in that: The cerebrovascular tissue segmentation model is a deep neural network model that introduces an attention gating module into the jump connection of the 3D-UNet convolutional neural network.

6. The method for predicting stroke based on cerebrovascular image characterization group according to claim 1, characterized in that: The stroke recurrence risk prediction model based on the cerebrovascular aging image characterization group includes: a convolutional neural network encoder, a cerebrovascular image feature fusion module and a vascular biological age prediction module; The convolutional neural network encoder is used to extract convolutional features of SWI, T1WI and T2-FLAIR by combining asymmetric convolution modules and Squeeze-and-Excitation modules; The cerebrovascular image feature fusion module is used to compress the convolution features by a global pooling compression method to obtain a convolution feature group of SWI, T1WI and T2-FLAIR with each feature dimension of 1, and directly connect the demographic indicators and cerebrovascular image features with the convolution feature group to generate a comprehensive image feature group; the comprehensive image feature group includes the cerebrovascular image features of TOF-MRA that specifically reflect the cerebrovascular characteristics, and the convolution feature groups of SWI, T1WI and T2-FLAIR that reflect the brain aging characteristics; the convolution feature group includes: white matter high signal volume, lacunar number, cortical thickness, gray matter volume, microbleeding and ventricular volume; the demographic indicators include: age and gender; The vascular biological age prediction module is used to comprehensively predict the patient's brain aging age based on the comprehensive image feature group, and compare the comprehensive prediction result with the individual's physical age to obtain a comprehensive cerebrovascular aging index.

7. A stroke prediction device based on a cerebrovascular image characterization group, characterized in that: The stroke prediction device based on the cerebrovascular image characterization group includes: A clinical database construction module is used to construct a clinical database; the clinical database includes: multimodal imaging feature group data of each patient and corresponding clinical feature data; 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 condition; A 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 of each patient, the corresponding clinical feature data and the corresponding patient clinical etiology classification, using a simple fully connected neural network architecture with a single hidden layer; A positioning diagnosis module is used to determine the cerebrovascular image features according to the TOF-MRA of each patient by using a cerebrovascular tissue segmentation model and voxel-based morphological analysis, and to perform positioning diagnosis of the responsible vascular area according to the cerebrovascular image features; the cerebrovascular tissue segmentation model is used to obtain a cerebrovascular segmentation image according to the time-of-flight magnetic resonance angiography image; the cerebrovascular image features include: basic cerebrovascular features, cerebrovascular geometric features, and cerebrovascular topological features; the basic cerebrovascular features include: the length and volume of individual tubular structures, and the diameter and length of local brain regions; the cerebrovascular geometric features include: the radius of curvature and the degree of bending of the cerebrovascular; the cerebrovascular topological features include: cerebrovascular bifurcation points; A stroke recurrence risk prediction model determination module is used to obtain a stroke recurrence risk prediction model based on a cerebrovascular aging image characterization group according to the cerebrovascular imaging features, SWI, T1WI, T2-FLAIR and corresponding clinical feature data in TOF-MRA of each patient, using a simple fully connected neural network architecture with a single hidden layer; the stroke recurrence risk prediction model based on a cerebrovascular aging image characterization group is used to output a stroke patient recurrence probability value; A data acquisition module, used to acquire multimodal imaging feature group data and corresponding clinical feature data of the subject to be predicted; The stroke prediction module is used to perform clinical etiology classification prediction and stroke recurrence risk prediction using a clinical etiology classification prediction model and a stroke recurrence risk prediction model based on a cerebrovascular aging imaging characterization group; and to perform localization diagnosis of the responsible vascular area.

8. A computer device comprising: 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 a cerebrovascular image characterization group as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting stroke based on a cerebrovascular image characterization group described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting stroke based on a cerebrovascular image characterization group described in any one of claims 1 to 6 is implemented.

Images