Brain infarction risk assessment method and system based on brain image

Abstract

The application provides a cerebral infarction risk assessment method and system based on brain images. It belongs to the technical field of cross between neural image analysis and cerebral vascular disease prediction. The method comprises the following steps: acquiring multiple modal images of the brain, obtaining multiple modal image data, and generating a multiple modal brain image data set; performing microvessel three-dimensional reconstruction on the multiple modal brain image data set based on generative artificial intelligence technology to generate brain microvessel atlas data; and significantly improving the accuracy of cerebral infarction risk assessment by fusing multiple modal images and reconstructing microvessel atlas, and more carefully capturing subtle brain lesions; the three-element coupling model constructed reduces the one-sidedness of single modal assessment, and makes the assessment result more comprehensive and reliable.

Description

Technical Field

[0001] This invention proposes a method and system for assessing the risk of cerebral infarction based on brain imaging, which belongs to the interdisciplinary field of neuroimaging analysis and cerebrovascular disease prediction. Background Technology

[0002] Cerebral infarction, a cerebrovascular disease that seriously threatens human health, requires accurate early assessment and effective prevention. However, current image-based methods for assessing cerebral infarction risk have many limitations.

[0003] In terms of imaging modal applications, there is a clear disconnect. While CT angiography (CTA), commonly used in clinical practice, can assess large artery stenosis, it cannot reflect cerebral microcirculatory disturbances; arterial spin labeling perfusion imaging (ASL) can measure cerebral blood flow, but its low spatial resolution makes it difficult to accurately locate small ischemic areas; and while diffusion tensor imaging (DTI) can reflect white matter structure, it is not fully integrated with other imaging information.

[0004] The assessment methods are also relatively static. Most artificial intelligence models only provide simple labels such as "high, medium, and low risk," lacking spatial probability heatmaps of cerebral infarction risk and temporal evolution predictions, thus failing to provide comprehensive and dynamic decision-making basis for clinical practice.

[0005] Furthermore, existing methods do not adequately consider white matter integrity, and high white matter signal intensity is often reported separately without establishing an effective causal chain model with inadequate perfusion and vascular stenosis. Moreover, existing patents mostly focus on single tasks, such as using convolutional neural networks to classify carotid artery plaques or using U-Net to segment infarct lesions, failing to deeply integrate multimodal images and making it difficult to achieve accurate, interpretable, and prospective assessment of cerebral infarction risk. Summary of the Invention

[0006] This invention provides a method and system for assessing the risk of cerebral infarction based on brain imaging, in order to solve the problems mentioned in the background section above: The present invention proposes a method for assessing the risk of cerebral infarction based on brain imaging, the method comprising: S1. Perform multimodal image acquisition on the brain to obtain multimodal image data and generate a multimodal brain image dataset; perform three-dimensional reconstruction of microvessels on the multimodal brain image dataset based on generative artificial intelligence technology to generate brain microvessel atlas data; S2. Construct a low-perfusion region identification model based on brain microvascular atlas data, extract abnormal distribution features of cerebral blood flow, and generate low-perfusion spatial heat map data; simultaneously perform white matter high signal segmentation processing on multimodal brain image datasets to generate quantitative assessment data of white matter damage; S3. Establish a grading model for the degree of cerebral artery stenosis using magnetic resonance angiography data to generate spatial localization data of arterial stenosis; perform ternary coupling modeling with hypoperfusion spatial heat map data, white matter injury quantitative assessment data and arterial stenosis spatial localization data to generate a dataset of related features of hypoperfusion, white matter injury and arterial stenosis. S4. Train a spatiotemporal evolution prediction network based on the associated feature dataset, input multimodal brain imaging data at the current time point, output a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months, and generate spatiotemporal risk projection data for cerebral infarction; perform risk level quantification processing on the spatiotemporal risk projection data for cerebral infarction to generate graded early warning threshold data; S5. Generate dynamic risk intervention suggestions based on the graded early warning threshold data, integrate the spatiotemporal risk projection data of cerebral infarction with the dynamic risk intervention suggestions, generate a digital twin assessment report of cerebral infarction risk, and complete the construction of the intelligent assessment engine.

[0007] The present invention proposes a system for implementing the brain imaging-based stroke risk assessment method as described above, the system comprising: Data acquisition module: acquires multimodal images of the brain, obtains multimodal image data, and generates a multimodal brain image dataset; performs three-dimensional reconstruction of microvessels on the multimodal brain image dataset based on generative artificial intelligence technology, and generates brain microvessel atlas data; Feature extraction module: Constructs a low-perfusion region identification model based on brain microvascular atlas data, extracts abnormal distribution features of cerebral blood flow, and generates low-perfusion spatial heat map data; Simultaneously, performs white matter high signal segmentation processing on multimodal brain image datasets to generate quantitative assessment data of white matter damage; Feature association module: Establish a grading model of cerebral artery stenosis degree using magnetic resonance angiography data to generate spatial localization data of arterial stenosis; perform ternary coupling modeling of hypoperfusion spatial heat map data, white matter injury quantitative assessment data and arterial stenosis spatial localization data to generate a dataset of associated features of hypoperfusion, white matter injury and arterial stenosis. The graded early warning module trains a spatiotemporal evolution prediction network based on a related feature dataset, inputs multimodal brain imaging data at the current time point, outputs a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months, and generates spatiotemporal risk projection data for cerebral infarction; it then performs risk level quantification processing on the spatiotemporal risk projection data for cerebral infarction to generate graded early warning threshold data. Report generation module: Generates dynamic risk intervention suggestions based on graded early warning threshold data, integrates spatiotemporal risk projection data of cerebral infarction with dynamic risk intervention suggestions, generates a digital twin assessment report of cerebral infarction risk, and completes the construction of intelligent assessment engine.

[0008] The beneficial effects of this invention are as follows: By fusing multimodal images and reconstructing microvascular atlases, the accuracy of stroke risk assessment is significantly improved, enabling more detailed capture of subtle brain lesions; the constructed ternary coupling model reduces the bias of single-modal assessment, making the assessment results more comprehensive and reliable; it enhances the forward-looking nature of risk prediction, allowing for the extrapolation of risk probability distribution over the next 6-24 months, thus gaining valuable time for prevention; it reduces misdiagnosis and missed diagnosis, avoiding delays in treatment due to inaccurate assessments. It provides spatial probability heatmaps, intuitively displaying risk areas, and offers temporal evolution predictions, giving doctors and patients a clear understanding of disease progression; it also generates dynamic risk intervention suggestions, providing a basis for personalized prevention and treatment plans, effectively preventing the occurrence of serious consequences of stroke and truly protecting patient health. Attached Figure Description

[0009] Figure 1 This is a diagram illustrating the steps of the method described in this invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0010] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0011] One embodiment of the present invention, such as Figure 1 As shown, a method for assessing the risk of cerebral infarction based on brain imaging includes: S1. Perform multimodal image acquisition on the brain to obtain multimodal image data, including T1-weighted imaging, T2-weighted imaging, arterial spin label perfusion imaging, magnetic resonance angiography, and diffusion tensor imaging data, to generate a multimodal brain image dataset; perform three-dimensional reconstruction of microvessels on the multimodal brain image dataset based on generative artificial intelligence technology to generate brain microvessel atlas data; S2. Construct a low-perfusion region identification model based on brain microvascular atlas data, extract abnormal distribution features of cerebral blood flow, and generate low-perfusion spatial heat map data; simultaneously perform white matter high signal segmentation processing on multimodal brain image datasets to generate quantitative assessment data of white matter damage; S3. Establish a grading model for the degree of cerebral artery stenosis using magnetic resonance angiography data to generate spatial localization data of arterial stenosis; perform ternary coupling modeling with hypoperfusion spatial heat map data, white matter injury quantitative assessment data and arterial stenosis spatial localization data to generate a dataset of related features of hypoperfusion, white matter injury and arterial stenosis. S4. Train a spatiotemporal evolution prediction network based on the associated feature dataset, input multimodal brain imaging data at the current time point, output a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months, and generate spatiotemporal risk projection data for cerebral infarction; perform risk level quantification processing on the spatiotemporal risk projection data for cerebral infarction to generate graded early warning threshold data; S5. Generate dynamic risk intervention suggestions based on graded early warning threshold data, including vascular reconstruction priority ranking, microcirculation improvement plan and white matter protection strategy; integrate spatiotemporal risk projection data of cerebral infarction with dynamic risk intervention suggestions to generate a digital twin assessment report of cerebral infarction risk and complete the construction of intelligent assessment engine.

[0012] The working principle and effects of the above technical solution are as follows: Multimodal image acquisition and microvascular 3D reconstruction improve the accuracy of stroke risk assessment and reduce the possibility of missed or misdiagnosis. Ternary coupling modeling enhances the ability to analyze the correlation between different pathological features, reducing the one-sidedness of single-indicator assessments. Spatiotemporal evolution prediction networks predict future risks in advance, avoiding the problem of delayed diagnosis and treatment when stroke occurs, and reducing the probability of poor prognosis. Tiered early warning and dynamic intervention recommendations can both develop targeted treatment plans and optimize the efficiency of medical resource allocation. Digital twin assessment reports make risk information more intuitive, enhance the scientific nature of clinical decision-making, reduce the medical burden caused by blind intervention, and overall improve the foresight and effectiveness of stroke risk prevention and control.

[0013] In one embodiment of the present invention, S1 includes: S11. A multimodal magnetic resonance imaging (MRI) device is used to perform a comprehensive scan of the brain to acquire raw T1-weighted imaging data, raw T2-weighted imaging data, raw arterial spin label perfusion imaging data, raw magnetic resonance angiography data, and raw diffusion tensor imaging data, thereby generating a multimodal brain image raw dataset. S12. Perform grayscale calibration, noise filtering, and image registration on the original multimodal brain image dataset to unify the data resolution and spatial coordinate system, and generate a standardized multimodal brain image dataset. S13. Based on generative artificial intelligence technology, microvascular structural features are extracted from standardized multimodal brain image datasets, and a three-dimensional topological model of microvessels is constructed. S14. Based on the three-dimensional topological model of microvessels, optimize the vascular cavity boundary and complete the branch nodes to generate high-precision brain microvessel atlas data.

[0014] The working principle and effects of the above technical solution are as follows: Multimodal omnidirectional scanning acquisition improves the comprehensiveness of brain imaging data and reduces information loss caused by single imaging methods. Preprocessing operations such as grayscale calibration and noise filtering reduce data interference and improve the consistency and reliability of image data. Generative artificial intelligence technology extracts microvascular structural features, enhancing the ability to identify vascular details and avoiding subjective errors caused by manual extraction. Optimization of vascular cavity boundaries and completion of branch nodes improve the accuracy of brain microvascular atlases, avoiding subsequent analysis biases caused by structural ambiguity or missing structures. The entire process not only lays a solid data foundation for cerebral infarction risk assessment but also reduces the amount of correction work in subsequent processing, improving the efficiency of the overall assessment process.

[0015] In one embodiment of the present invention, S2 includes: S21. Extract vascular hemodynamic characteristic parameters based on high-precision brain microvascular atlas data and construct an intelligent recognition network for low-perfusion areas; S22. Input the standardized multimodal brain image dataset into the intelligent recognition network for low perfusion areas, extract the abnormal distribution features of cerebral blood flow through feature layer overlay analysis, and generate low perfusion spatial heat map data. S23. Perform preliminary gray-white matter boundary segmentation on the standardized multimodal brain image dataset and filter regions of interest in the white matter region; S24. Perform signal intensity quantification analysis and lesion region marking on the selected white matter regions of interest to generate quantitative assessment data of white matter damage.

[0016] The working principle and effects of the above technical solution are as follows: By extracting vascular hemodynamic feature parameters to construct an identification network, the accuracy of identifying low-perfusion areas is improved, and the omission of subtle abnormalities is reduced. Feature layer overlay analysis enhances the identification of abnormal cerebral blood flow distribution, preventing local abnormalities from being masked by the overall image. Gray-white matter boundary segmentation and region of interest screening reduce the interference of irrelevant regions on the analysis, making subsequent processing more targeted. Signal intensity quantification and lesion marking make white matter damage assessment more objective, reducing bias caused by subjective judgment. The entire process can accurately capture the pathological features related to cerebral infarction and lay a solid foundation for subsequent risk extrapolation, improving the efficiency and reliability of the assessment process and avoiding deviations in subsequent assessments due to incomplete feature extraction.

[0017] In one embodiment of the present invention, step S22 includes: Modal feature decomposition was performed on the standardized multimodal brain image dataset, and independent feature layers corresponding to each modality were extracted to generate a set of single-modal independent feature layers; Spatial coordinate alignment and calibration are performed on the set of single-modal independent feature layers to unify the spatial reference benchmark of each layer and generate an aligned multimodal feature layer group. The aligned multimodal feature layer group is input into the intelligent recognition network for low perfusion areas. The network's built-in algorithm is used to enhance the features, strengthen blood flow-related feature signals, and generate an enhanced multimodal feature layer group. The enhanced multimodal feature layers are stacked and fused layer by layer to explore the correlation between different modal features and generate a multimodal fused feature map. By screening abnormal cerebral blood flow associated features from multimodal fusion feature maps, quantifying the spatial distribution range and labeling the intensity, low perfusion spatial heat map data is generated.

[0018] The working principle and effects of the above technical solution are as follows: Modal feature decomposition extracts independent layers, improving the purity of each modal feature and reducing interference between different modal signals. Spatial coordinate alignment and calibration unify the reference benchmark, avoiding feature matching deviations caused by layer misalignment and making subsequent analysis more accurate. Feature enhancement processing strengthens blood flow-related signals, improving the identification of weak abnormal features and reducing the possibility of subtle lesions being masked. Layer-by-layer overlay and fusion uncovers intermodal correlations, making feature information more comprehensive and avoiding the limitations of single-modal analysis. Screening and quantifying abnormal features improves the accuracy of low-perfusion area localization and avoids errors in range and intensity judgment. The entire process accurately captures abnormal cerebral blood flow features and provides reliable support for heatmap generation, making the presentation of low-perfusion areas more intuitive and improving the efficiency and reliability of subsequent risk assessment.

[0019] In one embodiment of the present invention, step S23 includes: Boundary enhancement processing is performed on a standardized multimodal brain image dataset to enhance the signal differences between gray and white matter regions, generating a boundary-enhanced image dataset. Extract gray-level gradient features from boundary-enhanced image datasets to distinguish signal distribution patterns in gray and white matter regions and generate gray-level gradient feature maps. Region thresholding is performed based on gray-level gradient feature map to initially define the gray-white matter boundary range and generate preliminary gray-white matter boundary contour data. The preliminary boundary contour data of gray and white matter is smoothed and optimized to eliminate jagged noise in the contour and generate regularized gray and white matter boundary data. White matter regions are separated based on normalized gray and white matter boundary data, and the core analysis scope is defined in combination with clinical focus to generate regions of interest in white matter regions.

[0020] The working principle and effects of the above technical solution are as follows: Boundary enhancement strengthens the difference between gray and white matter signals, improving the distinguishability of their boundaries and reducing confusion caused by signal overlap. Extracting gray-level gradient features distinguishes distribution patterns, enhancing the rationality of region division and avoiding the problem of blindly defining boundaries. Thresholding initially locks in the boundary range, improving the targeting of subsequent processing and reducing invalid operations. Smoothing optimization eliminates jagged noise, making the boundaries more regular and avoiding misjudgments of regions caused by irregular contours. Separating white matter regions and combining them with clinically relevant delineation not only focuses on the core analysis area but also reduces interference from irrelevant information, improving the efficiency and accuracy of subsequent white matter lesion assessment and avoiding the impact of improper range definition on assessment results.

[0021] In one embodiment of the present invention, S3 includes: S31. Extract magnetic resonance angiography data from a standardized multimodal brain imaging dataset and perform vascular lumen diameter measurement and vascular wall thickness analysis. S32. Based on the measurement and analysis results, construct a grading standard system for the degree of cerebral artery stenosis and establish an intelligent grading model for arterial stenosis; S33. Input the magnetic resonance angiography data into the intelligent grading model for arterial stenosis to generate spatial localization data of arterial stenosis and quantitative results of the degree of stenosis. S34. Perform spatiotemporal alignment processing on low perfusion spatial thermal map data, white matter damage quantitative assessment data and arterial stenosis spatial location data to unify data timestamps and spatial reference benchmarks. S35. A multi-feature fusion algorithm is used to perform correlation analysis on the aligned three types of data, construct a ternary coupling relationship model, and generate a dataset of correlation features of low perfusion, white matter damage and arterial stenosis.

[0022] The working principle and effects of the above technical solution are as follows: By extracting specific data from magnetic resonance angiography and conducting precise measurement and analysis, the accuracy of parameters such as vessel lumen diameter and vessel wall thickness is improved, reducing subjective errors caused by manual measurement. A standardized grading system and an intelligent grading model for arterial stenosis are established, making the judgment of stenosis degree more standardized and avoiding grading bias caused by individual differences in judgment. The spatial positioning data and quantitative results output by the model improve the efficiency of stenosis area identification and reduce the missed detection of minor stenosis. The spatiotemporal dimension alignment processing of the three types of data unifies the temporal and spatial analysis benchmarks, avoiding bias in correlation analysis caused by inconsistent benchmarks. Multi-feature fusion correlation analysis constructs a ternary coupling model, enhancing the correlation mining of different pathological features and reducing the limitations of single feature assessment. The entire process can comprehensively capture the pathological correlations related to cerebral infarction and provide high-quality feature data for subsequent risk prediction, improving the accuracy and reliability of the overall assessment.

[0023] In one embodiment of the present invention, S34 includes: Temporal information was extracted from low perfusion spatial thermogram data, white matter injury quantitative assessment data, and arterial stenosis spatial location data, and independent temporal information lists of the three types of data were generated. Time benchmark calibration is performed on the independent time information lists of the three types of data to unify the time measurement units and starting time nodes, and time alignment benchmark rules are generated. Adjust the time dimension parameters of the three types of data according to the time alignment benchmark rules to eliminate time offset differences and generate a time dimension aligned dataset. Extract spatial coordinate parameters of various data types in the time-dimensional aligned dataset, sort out the differences in spatial reference systems, and generate a list of spatial coordinate features; Based on the list of spatial coordinate features, a unified spatial reference benchmark is established, coordinate projection methods and measurement standards are standardized, and a spatiotemporally aligned dataset is generated. Consistency checks are performed on the spatiotemporally aligned dataset to correct data deviations, ensure the spatiotemporal matching accuracy of the three types of data, and complete the spatiotemporal alignment process.

[0024] The working principle and effects of the above technical solution are as follows: By extracting time stamps from the three types of data to generate independent lists, the clarity of time information organization is improved, and the confusion of different data time records is reduced. The time information lists are calibrated to standardize units of measurement and starting nodes, avoiding alignment deviations caused by inconsistent time bases and making time dimension adjustments more justifiable. Adjusting time parameters according to rules eliminates offsets, enhancing the matching degree of the three types of data in the time dimension and reducing analysis errors caused by time differences. Extracting spatial coordinate parameters and organizing differences to generate a list improves the recognizability of the spatial reference system and avoids spatial base confusion affecting alignment results. Standardizing the projection and measurement of the spatial reference base standard makes spatial dimension alignment more accurate and reduces correlation misalignment caused by inconsistencies in coordinates. Consistency verification corrects deviations, ensuring spatiotemporal matching accuracy and preventing errors in subsequent correlation analysis due to data misalignment. The entire process achieves accurate spatiotemporal alignment of the three types of data and lays a solid foundation for subsequent multi-feature fusion analysis, improving the reliability and efficiency of the overall analysis.

[0025] In one embodiment of the present invention, S35 includes: Feature extraction was performed on the spatiotemporally aligned dataset, and core feature vectors corresponding to low perfusion, white matter injury and arterial stenosis were extracted to generate three types of single feature vector sets. The three types of single feature vector sets are subjected to dimensionality standardization to unify the feature data scale and quantization range, thereby generating a standard feature vector set. A multi-feature fusion algorithm is used to perform cross operations on the standard feature vector set to uncover the intrinsic relationships between different features and generate preliminary fused feature data. Perform quantitative analysis of the correlation strength of the preliminary fused feature data, calculate the correlation coefficient between features, and generate a correlation strength matrix; A ternary coupling relationship model was constructed based on the correlation strength matrix to clarify the interaction pathways of hypoperfusion, white matter injury and arterial stenosis, and to generate a coupling model framework. The preliminary fusion feature data is input into the coupled model framework for screening and optimization, retaining highly correlated and effective features, and generating a dataset of features related to low perfusion, white matter damage, and arterial stenosis.

[0026] The working principle and effects of the above technical solution are as follows: By extracting three types of core feature vectors, the relevance of feature information is improved, and the interference of redundant data on subsequent analysis is reduced. Dimensional standardization unifies the data scale and quantification range, avoiding fusion bias caused by differences in feature scales and enhancing data consistency. Multi-feature fusion and cross-operation deeply explore the inherent correlations, breaking the limitations of single-feature analysis and making the correlation information more comprehensive. Quantitative analysis of correlation strength generates a matrix, making the correlation between features clearer and reducing errors caused by subjective judgment. The ternary coupling model clarifies the interaction path, avoiding confusion in correlation logic and making the analysis more logical. Screening and optimization retain highly correlated and effective features, improving the quality of the dataset and reducing interference from invalid features. The entire process can accurately capture the deep correlations of three types of pathological features and provide high-quality data support for subsequent risk prediction, enhancing the reliability and accuracy of the analysis results and improving the efficiency of the overall assessment process.

[0027] In one embodiment of the present invention, step S4 includes: S41. Perform data augmentation on the associated feature dataset to increase the number of samples and feature diversity, and divide it into training set, validation set and test set; S42. Construct a spatiotemporal evolution prediction network architecture, set the number of network layers, convolutional kernel size and activation function type, and initialize network parameters; S43. Iteratively train the spatiotemporal evolution prediction network using the training set, adjust the network hyperparameters using the validation set, and verify the network prediction accuracy using the test set. S44. Input the standardized multimodal brain image dataset at the current time point, and output a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months through the trained spatiotemporal evolution prediction network to generate spatiotemporal risk projection data for cerebral infarction. S45. Based on the probability distribution range of spatiotemporal risk extrapolation data for cerebral infarction, divide the risk levels into high, medium, and low risk ranges and generate graded early warning threshold data.

[0028] The working principle and effects of the above technical solution are as follows: Data augmentation expands the sample and feature diversity, improving the generalization ability of network training and reducing overfitting caused by insufficient samples. Dividing the training, validation, and test sets makes network training more scientific, avoiding blind parameter adjustments and improving model reliability. Constructing a reasonable spatiotemporal evolution prediction network architecture and initializing parameters enhances the targeting of risk prediction and reduces prediction bias caused by unreasonable network design. Iterative training combined with hyperparameter adjustment and accuracy verification continuously optimizes network performance, improving the accuracy of risk extrapolation and reducing prediction errors. Outputting a risk probability heatmap for the next 6 to 24 months makes the trend of cerebral infarction risk more intuitive, avoiding the passive situation of being unable to predict in advance. Dividing into high, medium, and low risk level intervals makes early warning more hierarchical, allowing for early intervention time for high-risk groups while avoiding over-treatment of low-risk groups, improving the rationality and efficiency of clinical diagnosis and treatment.

[0029] In one embodiment of the present invention, step S5 includes: S51. Combining graded early warning threshold data with clinical diagnosis and treatment guidelines, differentiated intervention directions are formulated for different risk levels, generating priority ranking of vascular reconstruction, microcirculation improvement plans and white matter protection strategies. S52. Visualize the spatiotemporal risk projection data of cerebral infarction, optimize the data display format and information hierarchy, and improve data readability; S53. Integrate the visualized spatiotemporal risk projection data of cerebral infarction with dynamic risk intervention suggestions in a structured manner, and standardize the data fields and reporting framework; S54. Perform logical verification and information completion on the integrated content to generate a digital twin assessment report on the risk of cerebral infarction. S55. Based on feedback from actual clinical applications, iteratively optimize the logic for generating assessment reports and the accuracy of matching intervention recommendations, and complete the construction of an intelligent assessment engine.

[0030] The working principle and effects of the above technical solution are as follows: By combining graded early warning thresholds with clinical treatment guidelines to formulate differentiated intervention directions, the targeted nature of intervention recommendations is improved, avoiding inappropriate interventions caused by a one-size-fits-all approach and reducing the waste of medical resources. Visualizing the spatiotemporal risk projection data of cerebral infarction optimizes the display format and information hierarchy, enhancing data readability and enabling clinicians to quickly grasp core risk information, reducing misunderstandings. Structured integration of visualized data and intervention recommendations standardizes the report framework and data fields, enhancing the logical coherence of the assessment report and avoiding omissions of key content due to information clutter. Logical verification and information completion further improve the accuracy of the report, reducing the impact of incomplete information or logical contradictions on diagnostic judgment. Iterative optimization of assessment logic and intervention matching accuracy based on clinical feedback makes the intelligent assessment engine more aligned with practical application needs, improving its adaptability and reliability. The entire process not only provides clinicians with accurate and easy-to-understand risk assessment reports but also continuously optimizes the assessment system, improving the clinical adaptability and scientific rigor of cerebral infarction risk prevention and control.

[0031] In one embodiment of the present invention, S51 includes: Extract the risk level classification range from the graded early warning threshold data, clarify the quantitative boundaries corresponding to high, medium and low risks, and generate detailed risk level data; We sorted out the core content related to cerebral infarction intervention in clinical practice guidelines, extracted the intervention principles corresponding to different disease severity, and generated a list of hierarchical diagnosis and treatment principles. By matching the detailed risk level data with the list of hierarchical diagnosis and treatment principles, the core intervention direction for each risk level is determined, and a differentiated intervention key catalog is generated. Based on the differentiated intervention key catalog, key influencing factors of vascular reconstruction are screened for high-risk levels, priority ranking criteria are formulated, and vascular reconstruction priority assessment indicators are generated. Based on the intervention needs at the medium-risk level, suitable technical means were selected around the goal of improving microcirculation, the implementation process and parameter requirements were clarified, and a framework for microcirculation improvement was generated. To address the prevention needs of low-risk levels, we integrate clinical measures related to white matter protection, standardize applicable scenarios and implementation standards, and generate detailed white matter protection strategies. The vascular revascularization priority assessment indicators, microcirculation improvement program framework, and white matter protection strategy details are integrated to form a complete differentiated intervention program system, generating a vascular revascularization priority ranking, microcirculation improvement program, and white matter protection strategy.

[0032] The working principle and effects of the above technical solution are as follows: Extracting quantitative boundaries of risk levels to generate detailed data improves the clarity of risk definition and avoids deviations in intervention direction due to ambiguous risk level boundaries. Reviewing clinical practice guidelines to refine intervention principles enhances the standardization of intervention recommendations and reduces blind recommendations divorced from clinical practice. Precisely matching risk levels with treatment principles makes intervention priorities clearer and reduces the probability of intervention going astray. Developing differentiated plans for different risk levels not only accurately addresses the core issues of each risk level, avoiding insufficient intervention for high-risk cases and over-treatment for low-risk cases, but also improves the adaptability of intervention measures. Finally, coordinating and integrating these into a complete system improves the systematic nature of the intervention plan, reduces problems of poor coordination between parts, and enhances the overall operability of intervention recommendations, providing reliable support for precise clinical policy implementation.

[0033] One embodiment of the present invention, such as Figure 2 As shown, a system for implementing the brain imaging-based stroke risk assessment method as described above is characterized in that the system comprises: Data acquisition module: performs multimodal image acquisition of the brain, acquires multimodal image data, including T1-weighted imaging, T2-weighted imaging, arterial spin label perfusion imaging, magnetic resonance angiography, and diffusion tensor imaging data, and generates a multimodal brain image dataset; based on generative artificial intelligence technology, performs three-dimensional reconstruction of microvessels on the multimodal brain image dataset to generate brain microvessel atlas data; Feature extraction module: Constructs a low-perfusion region identification model based on brain microvascular atlas data, extracts abnormal distribution features of cerebral blood flow, and generates low-perfusion spatial heat map data; Simultaneously, performs white matter high signal segmentation processing on multimodal brain image datasets to generate quantitative assessment data of white matter damage; Feature association module: Establish a grading model of cerebral artery stenosis degree using magnetic resonance angiography data to generate spatial localization data of arterial stenosis; perform ternary coupling modeling of hypoperfusion spatial heat map data, white matter injury quantitative assessment data and arterial stenosis spatial localization data to generate a dataset of associated features of hypoperfusion, white matter injury and arterial stenosis. The graded early warning module trains a spatiotemporal evolution prediction network based on a related feature dataset, inputs multimodal brain imaging data at the current time point, outputs a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months, and generates spatiotemporal risk projection data for cerebral infarction; it then performs risk level quantification processing on the spatiotemporal risk projection data for cerebral infarction to generate graded early warning threshold data. Report generation module: Generates dynamic risk intervention recommendations based on graded early warning threshold data, including vascular reconstruction priority ranking, microcirculation improvement programs, and white matter protection strategies; integrates spatiotemporal risk projection data of cerebral infarction with dynamic risk intervention recommendations to generate a digital twin assessment report of cerebral infarction risk, completing the construction of the intelligent assessment engine.

[0034] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for assessing the risk of cerebral infarction based on brain imaging, characterized in that, The method includes: S1. Perform multimodal image acquisition on the brain to obtain multimodal image data and generate a multimodal brain image dataset; perform three-dimensional reconstruction of microvessels on the multimodal brain image dataset based on generative artificial intelligence technology to generate brain microvessel atlas data; S2. Construct a low-perfusion region identification model based on brain microvascular atlas data, extract abnormal distribution features of cerebral blood flow, and generate low-perfusion spatial heat map data; simultaneously perform white matter high signal segmentation processing on multimodal brain image datasets to generate quantitative assessment data of white matter damage. S3. Establish a grading model for the degree of cerebral artery stenosis using magnetic resonance angiography data to generate spatial localization data of arterial stenosis; perform ternary coupling modeling with hypoperfusion spatial heat map data, white matter injury quantitative assessment data and arterial stenosis spatial localization data to generate a dataset of related features of hypoperfusion, white matter injury and arterial stenosis. S4. Train a spatiotemporal evolution prediction network based on the associated feature dataset, input multimodal brain imaging data at the current time point, output a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months, and generate spatiotemporal risk projection data for cerebral infarction; perform risk level quantification processing on the spatiotemporal risk projection data for cerebral infarction to generate graded early warning threshold data; S5. Generate dynamic risk intervention suggestions based on the graded early warning threshold data, integrate the spatiotemporal risk projection data of cerebral infarction with the dynamic risk intervention suggestions, generate a digital twin assessment report of cerebral infarction risk, and complete the construction of the intelligent assessment engine.

2. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 1, characterized in that, S1 includes: S11. A multimodal magnetic resonance imaging (MRI) device is used to perform a comprehensive scan of the brain to acquire raw T1-weighted imaging data, raw T2-weighted imaging data, raw arterial spin label perfusion imaging data, raw magnetic resonance angiography data, and raw diffusion tensor imaging data, thereby generating a multimodal brain image raw dataset. S12. Perform grayscale calibration, noise filtering, and image registration on the original multimodal brain image dataset to unify the data resolution and spatial coordinate system, and generate a standardized multimodal brain image dataset. S13. Based on generative artificial intelligence technology, microvascular structural features are extracted from standardized multimodal brain image datasets, and a three-dimensional topological model of microvessels is constructed. S14. Based on the three-dimensional topological model of microvessels, optimize the vascular cavity boundary and complete the branch nodes to generate high-precision brain microvessel atlas data.

3. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 1, characterized in that, S2 includes: S21. Extract vascular hemodynamic characteristic parameters based on high-precision brain microvascular atlas data and construct an intelligent recognition network for low-perfusion areas; S22. Input the standardized multimodal brain image dataset into the intelligent recognition network for low perfusion areas, extract the abnormal distribution features of cerebral blood flow through feature layer overlay analysis, and generate low perfusion spatial heat map data. S23. Perform preliminary gray-white matter boundary segmentation on the standardized multimodal brain image dataset and filter regions of interest in the white matter region; S24. Perform signal intensity quantification analysis and lesion region marking on the selected white matter regions of interest to generate quantitative assessment data of white matter damage.

4. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 3, characterized in that, S22 includes: Modal feature decomposition was performed on the standardized multimodal brain image dataset, and independent feature layers corresponding to each modality were extracted to generate a set of single-modal independent feature layers; Spatial coordinate alignment and calibration are performed on the set of single-modal independent feature layers to unify the spatial reference benchmark of each layer and generate an aligned multimodal feature layer group. The aligned multimodal feature layer group is input into the intelligent recognition network for low perfusion areas. The network's built-in algorithm is used to enhance the features, strengthen blood flow-related feature signals, and generate an enhanced multimodal feature layer group. The enhanced multimodal feature layers are stacked and fused layer by layer to explore the correlation between different modal features and generate a multimodal fused feature map. By screening abnormal cerebral blood flow associated features from multimodal fusion feature maps, quantifying the spatial distribution range and labeling the intensity, low perfusion spatial heat map data is generated.

5. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 1, characterized in that, The S3 includes: S31. Extract magnetic resonance angiography data from a standardized multimodal brain imaging dataset and perform vascular lumen diameter measurement and vascular wall thickness analysis. S32. Based on the measurement and analysis results, construct a grading standard system for the degree of cerebral artery stenosis and establish an intelligent grading model for arterial stenosis; S33. Input the magnetic resonance angiography data into the intelligent grading model for arterial stenosis to generate spatial localization data of arterial stenosis and quantitative results of the degree of stenosis. S34. Perform spatiotemporal alignment processing on low perfusion spatial thermal map data, white matter damage quantitative assessment data and arterial stenosis spatial location data to unify data timestamps and spatial reference benchmarks. S35. A multi-feature fusion algorithm is used to perform correlation analysis on the aligned three types of data, construct a ternary coupling relationship model, and generate a dataset of correlation features of low perfusion, white matter damage and arterial stenosis.

6. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 5, characterized in that, S34 includes: Temporal information was extracted from low perfusion spatial thermogram data, white matter injury quantitative assessment data, and arterial stenosis spatial location data, and independent temporal information lists of the three types of data were generated. Time benchmark calibration is performed on the independent time information lists of the three types of data to unify the time measurement units and starting time nodes, and time alignment benchmark rules are generated. Adjust the time dimension parameters of the three types of data according to the time alignment benchmark rules to eliminate time offset differences and generate a time dimension aligned dataset. Extract spatial coordinate parameters of various data types in the time-dimensional aligned dataset, sort out the differences in spatial reference systems, and generate a list of spatial coordinate features; Based on the list of spatial coordinate features, a unified spatial reference benchmark is established, coordinate projection methods and measurement standards are standardized, and a spatiotemporally aligned dataset is generated. Perform consistency checks on the spatiotemporally aligned dataset, correct data deviations, and complete the spatiotemporal alignment process.

7. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 1, characterized in that, The S4 includes: S41. Perform data augmentation on the associated feature dataset to increase the number of samples and feature diversity, and divide it into training set, validation set and test set; S42. Construct a spatiotemporal evolution prediction network architecture, set the number of network layers, convolutional kernel size and activation function type, and initialize network parameters; S43. Iteratively train the spatiotemporal evolution prediction network using the training set, adjust the network hyperparameters using the validation set, and verify the network prediction accuracy using the test set. S44. Input the standardized multimodal brain image dataset at the current time point, and output a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months through the trained spatiotemporal evolution prediction network to generate spatiotemporal risk projection data for cerebral infarction. S45. Based on the probability distribution range of spatiotemporal risk extrapolation data for cerebral infarction, divide the risk levels into high, medium, and low risk ranges and generate graded early warning threshold data.

8. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 1, characterized in that, The S5 includes: S51. Combining graded early warning threshold data with clinical diagnosis and treatment guidelines, differentiated intervention directions are formulated for different risk levels, generating priority ranking of vascular reconstruction, microcirculation improvement plans and white matter protection strategies. S52. Visualize the spatiotemporal risk projection data of cerebral infarction, optimize the data display format and information hierarchy, and improve data readability; S53. Integrate the visualized spatiotemporal risk projection data of cerebral infarction with dynamic risk intervention suggestions in a structured manner, and standardize the data fields and reporting framework; S54. Perform logical verification and information completion on the integrated content to generate a digital twin assessment report on the risk of cerebral infarction. S55. Based on feedback from actual clinical applications, iteratively optimize the logic for generating assessment reports and the accuracy of matching intervention recommendations, and complete the construction of an intelligent assessment engine.

9. The method for assessing the risk of cerebral infarction based on brain imaging according to claim 8, characterized in that, S51 includes: Extract the risk level classification range from the graded early warning threshold data, clarify the quantitative boundaries corresponding to high, medium and low risks, and generate detailed risk level data; We sorted out the core content related to cerebral infarction intervention in clinical practice guidelines, extracted the intervention principles corresponding to different disease severity, and generated a list of hierarchical diagnosis and treatment principles. By matching the detailed risk level data with the list of hierarchical diagnosis and treatment principles, the core intervention direction for each risk level is determined, and a differentiated intervention key catalog is generated. Based on the differentiated intervention key catalog, key influencing factors of vascular reconstruction are screened for high-risk levels, priority ranking criteria are formulated, and vascular reconstruction priority assessment indicators are generated. Based on the intervention needs at the medium-risk level, suitable technical means were selected around the goal of improving microcirculation, the implementation process and parameter requirements were clarified, and a framework for microcirculation improvement was generated. To address the prevention needs of low-risk levels, we integrate clinical measures related to white matter protection, standardize applicable scenarios and implementation standards, and generate detailed white matter protection strategies. The vascular revascularization priority assessment indicators, microcirculation improvement program framework, and white matter protection strategy details are integrated to form a complete differentiated intervention program system, generating a vascular revascularization priority ranking, microcirculation improvement program, and white matter protection strategy.

10. A system for implementing the brain imaging-based stroke risk assessment method as described in any one of claims 1-9, characterized in that, The system includes: Data acquisition module: acquires multimodal images of the brain, obtains multimodal image data, and generates a multimodal brain image dataset; performs three-dimensional reconstruction of microvessels on the multimodal brain image dataset based on generative artificial intelligence technology, and generates brain microvessel atlas data; Feature extraction module: Constructs a low-perfusion region identification model based on brain microvascular atlas data, extracts abnormal distribution features of cerebral blood flow, and generates low-perfusion spatial heat map data; Simultaneously, performs white matter high signal segmentation processing on multimodal brain image datasets to generate quantitative assessment data of white matter damage; Feature association module: Establish a grading model of cerebral artery stenosis degree using magnetic resonance angiography data to generate spatial localization data of arterial stenosis; perform ternary coupling modeling of hypoperfusion spatial heat map data, white matter injury quantitative assessment data and arterial stenosis spatial localization data to generate a dataset of associated features of hypoperfusion, white matter injury and arterial stenosis. The graded early warning module trains a spatiotemporal evolution prediction network based on a related feature dataset, inputs multimodal brain imaging data at the current time point, outputs a heat map of the probability distribution of cerebral infarction risk in the next 6 to 24 months, and generates spatiotemporal risk projection data for cerebral infarction; it then performs risk level quantification processing on the spatiotemporal risk projection data for cerebral infarction to generate graded early warning threshold data. Report generation module: Generates dynamic risk intervention suggestions based on graded early warning threshold data, integrates spatiotemporal risk projection data of cerebral infarction with dynamic risk intervention suggestions, generates a digital twin assessment report of cerebral infarction risk, and completes the construction of intelligent assessment engine.

Images