Carotid plaque stability and stroke risk prediction system fusing multiple modes

By integrating multimodal data into a carotid plaque stability and stroke risk prediction system, the problem of large risk assessment bias in single-modal assessment has been solved. This system enables multi-dimensional and accurate plaque risk assessment and individualized intervention, improving the accuracy and timeliness of stroke risk prediction.

CN120932896APending Publication Date: 2025-11-11LINFEN CENT HOSPITAL (THE FOURTH PEOPLES HOSPITAL OF LINFEN)

Patent Information

Application Number
CN202511286064.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-10
Publication Date
2025-11-11

AI Technical Summary

Technical Problem

In existing technologies, single-modal imaging examinations and in vitro biomarker analysis cannot achieve deep fusion and synergistic analysis of multimodal data from plasma proteomics, CT angiography, and magnetic resonance imaging. This results in a lack of judgment on the correlation between the biological activity and morphological characteristics of carotid plaques, large biases in risk assessment, and an inability to guarantee the accuracy of stroke risk prediction.

Method used

The multimodal carotid plaque stability and stroke risk prediction system integrates data acquisition and fusion, feature extraction and association, risk assessment and stratification, and intervention decision-making modules. It combines plasma proteomics, CT angiography, and MRI data, and uses graph neural networks and deep forest algorithms to perform cross-modal feature mapping and risk assessment, generating plaque stability index and stroke risk probability. The system self-optimizes through a closed-loop feedback mechanism.

Benefits of technology

It achieves multi-dimensional, comprehensive and accurate carotid plaque risk assessment, improves the accuracy and individualization of stroke risk prediction, has the ability to continuously learn and adapt to different clinical scenarios, and provides reliable diagnostic basis and individualized intervention strategies.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120932896A_ABST
    Figure CN120932896A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of image recognition, and discloses a carotid plaque stability and stroke risk prediction system fusing multiple modes, and the system comprises a data collection and fusion module, a feature extraction and association module, a risk assessment and layering module, an intervention decision module, and a report generation and feedback module. When carotid plaque stability and cerebral apoplexy risk prediction is carried out, plasma proteomics, radiomics and clinical data are synergistically integrated by establishing a multi-modal data fusion analysis framework, so that the limitation that a traditional method depends on a single data source is overcome; the plaque risk can be evaluated from multiple dimensions of biological activity and morphological features, the comprehensiveness and accuracy of risk prediction are improved, a more reliable diagnosis basis is provided for clinic, interaction and sensitivity influence among different modal features can be adaptively quantified by introducing dynamic feature correlation modeling and a real-time weight calibration mechanism, and the risk prediction accuracy is improved. And objectivity and consistency of risk assessment results are ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to a multimodal carotid plaque stability and stroke risk prediction system. Background Technology

[0002] Stroke is characterized by high incidence, high disability rate, high mortality rate, high recurrence rate, and high economic burden, placing a huge burden on families and society. Carotid artery plaque is one of the main causes of stroke, and early and accurate identification and prediction of carotid artery plaque development are of great significance for timely intervention and reduction of stroke risk. Due to the high disability rate, high recurrence rate, and poor prognosis of stroke, stroke prevention, screening of high-risk groups, and early risk management are crucial. Among these, carotid artery stenosis caused by atherosclerotic plaques is a significant factor in ischemic stroke; therefore, early identification of vulnerable carotid artery plaques through imaging methods is of great significance for stroke prevention and early clinical intervention.

[0003] Currently, clinical assessment of carotid plaque stability and stroke risk mainly relies on single-modality imaging examinations and in vitro biomarker analysis. This fails to achieve deep fusion and synergistic analysis of multimodal data from plasma proteomics, CT angiography, and magnetic resonance imaging. Consequently, the correlation between plaque biological activity and morphological characteristics is lost, leading to significant biases in risk assessment and compromising the accuracy of risk prediction.

[0004] Therefore, we propose a multimodal carotid plaque stability and stroke risk prediction system to address the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multimodal carotid plaque stability and stroke risk prediction system. The carotid plaque stability and stroke risk prediction system provided by this invention solves the problem mentioned in the background technology of large bias in risk assessment and inability to guarantee the accuracy of risk prediction.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multimodal carotid plaque stability and stroke risk prediction system, the system comprising: Data acquisition and fusion module: Collects plasma proteomics, CT angiography and MRI data to construct a multimodal input dataset. When the deviation of key monitoring parameters exceeds the preset warning value, it determines whether the risk assessment activation threshold has been reached. If it has, it enters the feature extraction and association module. Feature extraction and association module: Extracts biomarker concentration features based on plasma proteomics data, quantifies plaque morphology features based on CTA data, analyzes fibrous cap thickness and plaque component distribution features based on MRI data, calls graph neural network to model nonlinear association mapping of cross-modal features, and generates a comprehensive feature vector of plaque stability. Risk assessment and stratification module: The comprehensive feature vector of plaque stability is processed by a dual-channel deep forest algorithm. The first channel uses XGBoost to parse protein data, and the second channel uses a three-dimensional convolutional neural network to process image data. Attention mechanism is applied in the fusion layer to weight and integrate multimodal features, and output the plaque stability index and the probability of stroke risk in 1 year. Intervention decision module: Based on the plaque stability index and stroke risk probability, combined with clinical history and health tolerance, determine whether individualized intervention measures are needed; if so, generate an optimized treatment strategy report. Report generation and feedback module: Integrates risk assessment results and intervention strategies to generate a visual report, and updates the feature association model based on treatment feedback data to achieve system self-optimization.

[0007] Preferably, the process for determining whether the risk assessment activation threshold has been reached is as follows: Set key monitoring parameters, including the plasma matrix metalloproteinase concentration threshold, the plaque volume threshold in CT angiography, and the fibrous cap thickness threshold in magnetic resonance imaging, and calculate the deviation of the current parameter values ​​from the normal range. When the deviation exceeds the preset warning value, it indicates that the plaque is in an abnormal state and it is necessary to assess whether it is close to the activation threshold. Calculate the difference between the current parameter and the activation threshold to obtain the activation proximity value. The specific formula is as follows: ; in Indicates the degree of concentration deviation. This represents the real-time plasma matrix metalloproteinase concentration measurement value. This indicates the preset concentration warning threshold; ; in Indicates activation of the proximity value. Indicates the activation threshold concentration of the digital model; Degree of deviation Time-triggered evaluation, activating proximity value Initiate the risk assessment process in a timely manner. The process is paused at any time, among which The preset activation threshold is approached.

[0008] Preferably, the process of generating the comprehensive feature vector of patch stability is as follows: High-throughput mass spectrometry analysis was performed on plasma proteomics data to extract the concentration values ​​of biomarkers, including matrix metalloproteinases and C-reactive protein, and integrated into a proteome. Three-dimensional reconstruction of CT angiography data was performed to quantify plaque volume, calcification degree and luminal stenosis rate, and integrated into a morphological feature group. Multi-sequence analysis was performed on magnetic resonance imaging data to resolve fibrous cap thickness, lipid core size, and hemorrhage signal intensity, which were then integrated into a component feature set. The protein feature set, morphological feature set, and component feature set are input into a graph neural network to model the mapping relationship between features and output a comprehensive feature vector of plaque stability.

[0009] Preferably, the process of modeling the mapping relationship between features is as follows: Construct a graph structure with biomarkers as nodes and feature associations as edges; Node embeddings are learned through graph neural networks to capture non-linear interactions; The sensitivity of protein concentration changes to image features was calculated, and a dynamic correlation matrix was established.

[0010] The preferred application process of the dual-channel deep forest algorithm is as follows: The first channel receives structured protein data and applies the XGBoost algorithm to sort features by importance and reduce dimensionality. The second channel receives multimodal image data and uses a three-dimensional convolutional neural network to extract spatial-temporal features; An attention mechanism is used in the fusion layer to assign weights to features of different modalities and then weighted and integrated to output risk prediction values.

[0011] Preferably, the process for outputting the plaque stability index and stroke risk probability is as follows: The plaque stability index is set to a range of 0 to 1, where 0 represents high stability and 1 represents high vulnerability. The comprehensive feature vector of plaque stability is input into the risk prediction model to calculate the plaque stability index. Based on the stability index and historical stroke cohort data, regression prediction was used to forecast the probability of stroke risk within 1 year. When the stability index is greater than 0.7, it is judged as a morphological high-risk state. When the stroke risk probability is greater than the preset risk threshold, it is judged as a probabilistic high-risk state. When both of the above states exist at the same time, it is judged as a composite high-risk state.

[0012] Preferably, the process of determining whether individualized intervention measures are needed is as follows: Obtain the patient's clinical history data, including age, history of hypertension, history of diabetes, and current medication regimen; Set health tolerance indicators, including remaining vascular healthy lifespan and risk tolerance thresholds; The formula for calculating the match between plaque risk status and health tolerance is as follows: ; in This represents the Pearson correlation coefficient. This represents the observations in the patch risk status data set. This represents the mean of the data set. This represents the observed values ​​of the health tolerance data set. This represents the mean of the data set; when If the match is insufficient, intervention is required. If the stroke risk probability exceeds the safety threshold, it is considered an independent high-risk condition requiring intervention. If both conditions are met, it is considered a composite high-risk condition requiring intervention. This is the preset tolerance lower limit.

[0013] Preferably, the process for generating an optimized treatment strategy report is as follows: Based on the risk assessment results, an intervention strategy library was matched to screen lipid-lowering therapy, antiplatelet therapy, and surgical intervention options. Optimize strategy parameters by taking into account individual patient factors; Generate a detailed report that includes the timing of the intervention, dosage, and expected results.

[0014] Preferably, the system self-optimization process is as follows: Collect treatment feedback data, including changes in plaque stability and stroke event records after intervention; Update the feature association model parameters; Iterative training improves prediction accuracy.

[0015] Preferably, the system also includes a real-time risk dynamic adjustment mechanism, which is activated after the intervention strategy generation module, and the process is as follows: After the intervention strategy report is generated and implemented, patient data is collected in real time, including changes in plasma proteomics concentrations, updates in imaging features, and feedback on clinical symptoms after the intervention. The data input feature extraction and association module will be updated to regenerate the comprehensive feature vector of patch stability. The risk assessment and stratification module is invoked to recalculate the plaque stability index and stroke risk probability. Compare the reassessed risk probability with the original risk probability before intervention, and calculate the rate of change of risk: ; in Indicates the rate of change of risk. This indicates the reassessed probability of stroke risk. This indicates the probability of stroke risk before intervention; When the absolute value of the risk change rate exceeds the preset adjustment threshold, the intervention strategy generation module is triggered to generate a dynamic adjustment plan based on the updated data, including optimizing drug dosage, adjusting treatment timing, and switching intervention methods. The adjustment plan is integrated into the report generation and feedback module, the visualization report is updated and fed back to the clinical end, and at the same time it is input into the system self-optimization module for model iterative training.

[0016] (III) Beneficial Effects Compared with existing technologies, this invention provides a multimodal carotid plaque stability and stroke risk prediction system, which has the following beneficial effects: 1. In this invention, when predicting carotid plaque stability and stroke risk, a multimodal data fusion analysis framework is established to synergistically integrate plasma proteomics, radiomics, and clinical data. This overcomes the limitations of traditional methods that rely on a single data source and can assess plaque risk from multiple dimensions of biological activity and morphological characteristics, thereby improving the comprehensiveness and accuracy of risk prediction and providing more reliable diagnostic evidence for clinical practice.

[0017] 2. In this invention, by introducing dynamic feature association modeling and real-time weight calibration mechanism, the interaction and sensitivity effects between different modal features can be adaptively quantified, avoiding evaluation errors caused by feature mapping distortion and weight allocation deviation, improving the system's ability to analyze complex medical data relationships and stability, and ensuring the objectivity and consistency of risk assessment results.

[0018] 3. In this invention, by constructing a closed-loop feedback and self-optimization system, clinical intervention results and patient follow-up data can be continuously integrated to achieve dynamic iteration and optimization of the risk prediction model. This enables the system to continuously learn and adapt to different clinical scenarios, enhances the timeliness and individualization of prediction results, and provides strong support for accurate decision-making in stroke prevention and control. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the architecture of the multimodal carotid plaque stability and stroke risk prediction system of this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] For specific implementation examples, please refer to: Figure 1 A multimodal carotid plaque stability and stroke risk prediction system, comprising: Data acquisition and fusion module: Collects plasma proteomics, CT angiography and MRI data to construct a multimodal input dataset. When the deviation of key monitoring parameters exceeds the preset warning value, it determines whether the risk assessment activation threshold has been reached. If it has, it enters the feature extraction and association module. Feature extraction and association module: Extracts biomarker concentration features based on plasma proteomics data, quantifies plaque morphology features based on CTA data, analyzes fibrous cap thickness and plaque component distribution features based on MRI data, calls graph neural network to model nonlinear association mapping of cross-modal features, and generates a comprehensive feature vector of plaque stability. Risk assessment and stratification module: The comprehensive feature vector of plaque stability is processed by a dual-channel deep forest algorithm. The first channel uses XGBoost to parse protein data, and the second channel uses a three-dimensional convolutional neural network to process image data. Attention mechanism is applied in the fusion layer to weight and integrate multimodal features, and output the plaque stability index and the probability of stroke risk in 1 year. Intervention decision module: Based on the plaque stability index and stroke risk probability, combined with clinical history and health tolerance, determine whether individualized intervention measures are needed; if so, generate an optimized treatment strategy report. Report generation and feedback module: Integrates risk assessment results and intervention strategies to generate a visual report, and updates the feature association model based on treatment feedback data to achieve system self-optimization.

[0022] The process for determining whether the risk assessment activation threshold has been reached is as follows: Set key monitoring parameters, including the plasma matrix metalloproteinase concentration threshold, the plaque volume threshold in CT angiography, and the fibrous cap thickness threshold in magnetic resonance imaging, and calculate the deviation of the current parameter values ​​from the normal range. When the deviation exceeds the preset warning value, it indicates that the plaque is in an abnormal state and it is necessary to assess whether it is close to the activation threshold. Calculate the difference between the current parameter and the activation threshold to obtain the activation proximity value. The specific formula is as follows: ; in Indicates the degree of concentration deviation. This represents the real-time plasma matrix metalloproteinase concentration measurement value. This indicates the preset concentration warning threshold; ; in Indicates activation of the proximity value. Indicates the activation threshold concentration of the digital model; Degree of deviation Time-triggered evaluation, activating proximity value Initiate the risk assessment process in a timely manner. The process is paused at any time, among which The preset activation threshold is approached.

[0023] The process of generating the comprehensive feature vector of patch stability is as follows: High-throughput mass spectrometry analysis was performed on plasma proteomics data to extract the concentration values ​​of biomarkers, including matrix metalloproteinases and C-reactive protein, and integrated into a proteome. Three-dimensional reconstruction of CT angiography data was performed to quantify plaque volume, calcification degree and luminal stenosis rate, and integrated into a morphological feature group. Multi-sequence analysis was performed on magnetic resonance imaging data to resolve fibrous cap thickness, lipid core size, and hemorrhage signal intensity, which were then integrated into a component feature set. The protein feature set, morphological feature set, and component feature set are input into a graph neural network to model the mapping relationship between features and output a comprehensive feature vector of plaque stability.

[0024] The process of modeling the mapping relationship between features is as follows: A graph structure was constructed with biomarkers as nodes and feature associations as edges. Node data originated from key biomarkers identified through plasma proteomics analysis and their quantified concentrations. Edge data was weighted based on the statistical correlations and physiological pathway associations among multimodal features. The specific weight calculation process is as follows: Calculate the Pearson correlation coefficient between the concentrations of any two biomarkers in a multimodal dataset. Take its absolute value As a measure of statistical correlation strength; The KEGG database API is used to query whether the two biomarkers exist in the same Kyoto Encyclopedia of Genetics and Genomes KEGG pathway graph. If they exist, a pathway association strength value of 1 is assigned; otherwise, it is 0. The final edge weights are then determined. The result is calculated using the formula: Pathway_Strength; in Here, edge weight is denoted as Pathway_Strength, and pathway_strength is denoted as the path association strength. For statistical correlation weighting coefficients, The weighting coefficients for pathway correlation. , Balancing the contribution of statistical correlation with physiological pathway association; By learning node embeddings through graph neural networks and capturing nonlinear interactions, specifically employing a graph attention network algorithm, node representations are iteratively updated through a multi-head attention mechanism to learn the complex, nonlinear regulatory and co-expression relationships between biomarkers. The specific process of the graph attention network algorithm is as follows: The attention coefficients of the target node i and its first-order neighbor node j are calculated using the following formula: LeakyReLU ; in For a trainable parameter vector, For a trainable weight matrix, and These are the initial feature vectors of nodes i and j, respectively; The attention coefficients are normalized using the softmax function to obtain the normalized attention weights. ; A multi-head attention mechanism is adopted, and the calculation results of K independent attention mechanisms are concatenated and averaged to generate the final output feature vector of the target node i. By iteratively updating the representation of all nodes through multiple graph attention layers, the complex, nonlinear regulatory and co-recognition relationships between biomarkers are learned. The sensitivity of protein concentration changes to image features is calculated, and a dynamic correlation matrix is ​​established. The dynamic correlation matrix is ​​quantified by a sensitivity index, and the calculation formula is as follows: ; in Indicates the sensitivity index. Represents the j-th image feature. Indicates the baseline value of image features. This represents the baseline value of protein concentration. This represents the concentration of the i-th protein.

[0025] The application process of the dual-channel deep forest algorithm is as follows: The first channel receives structured protein data and applies the XGBoost algorithm to sort features by importance and reduce dimensionality. The second channel receives multimodal image data and uses a three-dimensional convolutional neural network to extract spatial-temporal features; An attention mechanism is used in the fusion layer to assign weights to features of different modalities and then weighted and integrated to output risk prediction values.

[0026] The process of outputting the plaque stability index and stroke risk probability is as follows: The plaque stability index is set to a range of 0 to 1, where 0 represents high stability and 1 represents high vulnerability. The comprehensive feature vector of plaque stability is input into the risk prediction model to calculate the plaque stability index. The risk prediction model is a regression model trained based on machine learning algorithms. It is trained using historical plaque data and corresponding clinical outcome labels and is used to map the comprehensive feature vector of plaque stability obtained by multimodal fusion to the stability index range. Based on the stability index and historical stroke cohort data, regression analysis was used to predict the probability of stroke risk within one year. The regression prediction employed a proportional hazards model, incorporating the stability index as a key covariate within the survival analysis framework of the historical stroke cohort. This calculated the individual's probability of stroke occurrence within a specific future time window. The specific steps are as follows: Data preparation and model building: Based on historical stroke cohort data, the data includes patients' plaque stability index, follow-up time, and outcome labels of whether or not a stroke occurred; Using this data, a proportional hazards model was trained, and the plaque stability index was incorporated into the model as a key covariate to construct a survival analysis model that can be used to predict the risk of new stroke. Risk prediction calculation: For a new patient being evaluated, the plaque stability index obtained after multimodal fusion is substituted into the trained proportional hazards model, and the probability value of the patient's stroke risk in a specific time window in the future is obtained through model calculation. Risk status determination: The calculated stroke risk probability value is compared with a preset risk threshold. If the risk probability value is greater than the preset risk threshold, the patient is determined to be in a high-probability risk state. When the stability index is greater than 0.7, it is judged as a morphological high-risk state. When the stroke risk probability is greater than the preset risk threshold, it is judged as a probabilistic high-risk state. When both of the above states exist at the same time, it is judged as a composite high-risk state.

[0027] The process of determining whether individualized interventions are needed is as follows: Obtain the patient's clinical history data, including age, history of hypertension, history of diabetes, and current medication regimen; Set health tolerance indicators, including remaining vascular healthy lifespan and risk tolerance thresholds; The formula for calculating the match between plaque risk status and health tolerance is as follows: ; in This represents the Pearson correlation coefficient. This represents the observations in the patch risk status data set. This represents the mean of the data set. This represents the observed values ​​of the health tolerance data set. This represents the mean of the data set; when If the match is insufficient, intervention is required. If the stroke risk probability exceeds the safety threshold, it is considered an independent high-risk condition requiring intervention. If both conditions are met, it is considered a composite high-risk condition requiring intervention. This is a preset tolerance lower limit; The process of generating an optimized treatment strategy report and implementing the intervention is as follows: Based on the results of the determination of insufficient matching degree, independent high risk and composite high risk, the corresponding clinical intervention plan is matched and recommended from the pre-set intervention strategy library; Clinical intervention plans include recommendations for lifestyle interventions, drug therapy, and surgical interventions. Drug therapy includes lipid-lowering therapy and antiplatelet therapy. Optimize strategy parameters by combining individual factors such as the patient's clinical history, liver and kidney function, and drug tolerance; Generate a personalized intervention report that includes the type of intervention, recommended dose, timing of intervention, and expected goals, and push the report to the clinical decision support terminal; The system self-optimization process is as follows: Collect treatment feedback data, including changes in plaque stability and stroke event records after intervention; Update the feature association model parameters; Iterative training improves prediction accuracy.

[0028] The system also includes a real-time risk dynamic adjustment mechanism, which is activated after the intervention strategy generation module. The process is as follows: After the intervention strategy report is generated and implemented, patient data is collected in real time, including changes in plasma proteomics concentrations, updates in imaging features, and feedback on clinical symptoms after the intervention. The data input feature extraction and association module will be updated to regenerate the comprehensive feature vector of patch stability. The risk assessment and stratification module is invoked to recalculate the plaque stability index and stroke risk probability. Compare the reassessed risk probability with the original risk probability before intervention, and calculate the rate of change of risk: ; in Indicates the rate of change of risk. This indicates the reassessed probability of stroke risk. This indicates the probability of stroke risk before intervention; When the absolute value of the risk change rate exceeds the preset adjustment threshold, the intervention strategy generation module is triggered to generate a dynamic adjustment plan based on the updated data. The specific process is as follows: The system calls upon a pre-defined dynamic adjustment strategy library and matches the optimal adjustment type based on updated data. Adjustment types include optimizing drug dosage, adjusting treatment timing, and switching intervention methods. Based on the matching results and the patient's real-time physiological parameters, a final dynamic adjustment plan is generated; The adjustment plan is integrated into the report generation and feedback module, the visualization report is updated and fed back to the clinical end, and at the same time it is input into the system self-optimization module for model iterative training.

[0029] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multimodal carotid plaque stability and stroke risk prediction system, characterized in that: The system includes: Data acquisition and fusion module: Collects plasma proteomics, CT angiography and MRI data to construct a multimodal input dataset. When the deviation of key monitoring parameters exceeds the preset warning value, it determines whether the risk assessment activation threshold has been reached. If it has, it enters the feature extraction and association module. Feature extraction and association module: Extracts biomarker concentration features based on plasma proteomics data, quantifies plaque morphology features based on CTA data, analyzes fibrous cap thickness and plaque component distribution features based on MRI data, calls graph neural network to model nonlinear association mapping of cross-modal features, and generates a comprehensive feature vector of plaque stability. Risk assessment and stratification module: The comprehensive feature vector of plaque stability is processed by a dual-channel deep forest algorithm. The first channel uses XGBoost to parse protein data, and the second channel uses a three-dimensional convolutional neural network to process image data. Attention mechanism is applied in the fusion layer to weight and integrate multimodal features, and output the plaque stability index and the probability of stroke risk in 1 year. Intervention decision module: Based on the plaque stability index and stroke risk probability, combined with clinical history and health tolerance, determine whether individualized intervention measures are needed; if so, generate an optimized treatment strategy report. Report generation and feedback module: Integrates risk assessment results and intervention strategies to generate a visual report, and updates the feature association model based on treatment feedback data to achieve system self-optimization.

2. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 1, characterized in that: The process for determining whether the risk assessment activation threshold has been reached is as follows: Set key monitoring parameters, including the plasma matrix metalloproteinase concentration threshold, the plaque volume threshold in CT angiography, and the fibrous cap thickness threshold in magnetic resonance imaging, and calculate the deviation of the current parameter values ​​from the normal range. When the deviation exceeds the preset warning value, it indicates that the plaque is in an abnormal state and it is necessary to assess whether it is close to the activation threshold. Calculate the difference between the current parameter and the activation threshold to obtain the activation proximity value. The specific formula is as follows: ; in Indicates the degree of concentration deviation. This represents the real-time plasma matrix metalloproteinase concentration measurement value. This indicates the preset concentration warning threshold; ; in Indicates activation of the proximity value. Indicates the activation threshold concentration of the digital model; Degree of deviation Time-triggered evaluation, activating proximity value Initiate the risk assessment process in a timely manner. The process is paused at any time, among which The preset activation threshold is approached.

3. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 1, characterized in that: The process of generating the comprehensive feature vector of patch stability is as follows: High-throughput mass spectrometry analysis was performed on plasma proteomics data to extract the concentration values ​​of biomarkers, including matrix metalloproteinases and C-reactive protein, and integrated into a proteome. Three-dimensional reconstruction of CT angiography data was performed to quantify plaque volume, calcification degree and luminal stenosis rate, and integrated into a morphological feature group. Multi-sequence analysis was performed on magnetic resonance imaging data to resolve fibrous cap thickness, lipid core size, and hemorrhage signal intensity, which were then integrated into a component feature set. The protein feature set, morphological feature set, and component feature set are input into a graph neural network to model the mapping relationship between features and output a comprehensive feature vector of plaque stability.

4. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 3, characterized in that: The process of modeling the mapping relationship between features is as follows: Construct a graph structure with biomarkers as nodes and feature associations as edges; Node embeddings are learned through graph neural networks to capture non-linear interactions; The sensitivity of protein concentration changes to image features was calculated, and a dynamic correlation matrix was established.

5. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 1, characterized in that: The application process of the dual-channel deep forest algorithm is as follows: The first channel receives structured protein data and applies the XGBoost algorithm to sort features by importance and reduce dimensionality. The second channel receives multimodal image data and uses a three-dimensional convolutional neural network to extract spatial-temporal features; An attention mechanism is used in the fusion layer to assign weights to features of different modalities and then weighted and integrated to output risk prediction values.

6. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 1, characterized in that: The process of outputting the plaque stability index and stroke risk probability is as follows: The plaque stability index is set to a range of 0 to 1, where 0 represents high stability and 1 represents high vulnerability. The comprehensive feature vector of plaque stability is input into the risk prediction model to calculate the plaque stability index. Based on the stability index and historical stroke cohort data, regression prediction was used to forecast the probability of stroke risk within 1 year. When the stability index is greater than 0.7, it is judged as a morphological high-risk state. When the stroke risk probability is greater than the preset risk threshold, it is judged as a probabilistic high-risk state. When both of the above states exist at the same time, it is judged as a composite high-risk state.

7. The multimodal carotid plaque stability and stroke risk prediction system according to claim 1, characterized in that: The process of determining whether individualized intervention is needed is as follows: Obtain the patient's clinical history data, including age, history of hypertension, history of diabetes, and current medication regimen; Set health tolerance indicators, including remaining vascular healthy lifespan and risk tolerance thresholds; The formula for calculating the match between plaque risk status and health tolerance is as follows: ; in This represents the Pearson correlation coefficient. This represents the observations in the patch risk status data set. This represents the mean of the data set. This represents the observed values ​​of the health tolerance data set. This represents the mean of the data set; when If the match is insufficient, intervention is required. If the stroke risk probability exceeds the safety threshold, it is considered an independent high-risk condition requiring intervention. If both conditions are met, it is considered a composite high-risk condition requiring intervention. This is the preset tolerance lower limit.

8. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 1, characterized in that: The process of generating an optimized treatment strategy report is as follows: Based on the risk assessment results, an intervention strategy library was matched to screen lipid-lowering therapy, antiplatelet therapy, and surgical intervention options. Optimize strategy parameters by taking into account individual patient factors; Generate a detailed report that includes the timing of the intervention, dosage, and expected results.

9. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 1, characterized in that: The system self-optimization process is as follows: Collect treatment feedback data, including changes in plaque stability and stroke event records after intervention; Update the feature association model parameters; Iterative training improves prediction accuracy.

10. The carotid plaque stability and stroke risk prediction system integrating multimodal data according to claim 1, characterized in that: The system also includes a real-time risk dynamic adjustment mechanism, which is activated after the intervention strategy generation module. The process is as follows: After the intervention strategy report is generated and implemented, patient data is collected in real time, including changes in plasma proteomics concentrations, updates in imaging features, and feedback on clinical symptoms after the intervention. The data input feature extraction and association module will be updated to regenerate the comprehensive feature vector of patch stability. The risk assessment and stratification module is invoked to recalculate the plaque stability index and stroke risk probability. Compare the reassessed risk probability with the original risk probability before intervention, and calculate the rate of change of risk: ; in Indicates the rate of change of risk. This indicates the reassessed probability of stroke risk. This indicates the probability of stroke risk before intervention; When the absolute value of the risk change rate exceeds the preset adjustment threshold, the intervention strategy generation module is triggered to generate a dynamic adjustment plan based on the updated data, including optimizing drug dosage, adjusting treatment timing, and switching intervention methods. The adjustment plan is integrated into the report generation and feedback module, the visualization report is updated and fed back to the clinical end, and at the same time it is input into the system self-optimization module for model iterative training.

Citation Information

Patent Citations

  • Method and system for predicting recurrence of cerebral apoplexy

    CN118430819A

  • Carotid plaque intelligent identification method and system based on multi-modal image fusion

    CN118799693A

Cited By

  • Method and system for predicting lifetime of non-small cell lung cancer patient after radiotherapy

    CN121148603A

  • Cerebral hemorrhage prognosis analysis system fusing images and clinical data

    CN121617639A

  • A cerebral hemorrhage prognosis analysis system fusing images and clinical data

    CN121617639B

  • Method and system for predicting severe ischemic event of aorta arteritis nervous system based on multi-modal deep learning

    CN122091190A