A method and system for early warning of cerebral infarction risk based on the changing trends of imaging features

Abstract

This invention discloses a data processing method and system based on the changing trends of image features, relating to the field of cerebral infarction early warning technology. The data processing method includes: feature extraction: processing image data and modal data separately using feature analysis methods to obtain image features, image feature change data, modal data features, modal data feature change data, and the regions where the features are located. In this invention, image data provides intuitive information about brain structure and lesions, while modal data supplements key features such as hemodynamics, brain tissue electrical activity, and electrical impedance from different physiological dimensions. TCD data can monitor blood flow velocity and vascular resistance in real time, BI data is sensitive to changes in pathological states such as cerebral edema, and EEG data reflects subtle changes in neuronal electrical activity. Through feature extraction and analysis, the combination of multi-source heterogeneous data achieves the correction of image feature changing trends, thereby ensuring the accuracy of data processing.

Description

Technical Field

[0001] This invention relates to the field of cerebral infarction early warning technology, specifically to a data processing method and system based on the changing trends of image features. Background Technology

[0002] Stroke risk warning technology mainly relies on the analysis of clinical and imaging data.

[0003] The patent publication number "CN119400393A" is titled "Method and System for Predicting Infarction Changes Based on Brain Parenchymal MRI Images." This patent acquires brain parenchymal MRI images and constructs a multimodal image set. It then filters images according to selection criteria, identifies artifacts, calculates quality scores, performs preprocessing to obtain standard image data, extracts features to obtain brain region segmentation masks, extracts multimodal feature data, constructs topological relationships between different brain regions, extracts high-order semantic features to obtain brain region feature data, selects high-quality feature data, constructs a temporal feature matrix, assigns weights, generates residual output, performs prediction through a fully connected neural network, generates infarction change prediction results and prognostic results, and generates a comprehensive prediction report.

[0004] While image feature change trend analysis is based on image boundaries alone, it can only capture the lesion features of cerebral infarction. Simply relying on the change trend features of images can only reflect the dynamic changes of the brain's physiological state in a one-sided manner. Analyzing based on single imaging data lacks correction analysis of multi-source data. Moreover, the manifestations of infarction are not only reflected in images. This makes it impossible for the prediction model to fully understand the complex situation of cerebral infarction, thus limiting the accuracy of prediction. To address this, a data processing method and system based on image feature change trends has been invented. Summary of the Invention

[0005] The purpose of this invention is to provide a data processing method and system based on the changing trends of image features, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data processing method based on the changing trends of image features, the data processing method comprising:

[0007] Feature extraction: Image data and modal data are processed separately using feature analysis methods to obtain image features, image feature variation data, modal data features, modal data feature variation data, and the regions where the features are located. Regional associations between image features and modal data features are established based on the regions where the features are located.

[0008] Feature analysis: The image data is processed using the Dynamic Spatiotemporal Topology Alignment Network (DSTAN) to generate a continuous spatiotemporal image field. The image feature change data and modal data feature change data are mapped onto the spatiotemporal image field. Based on the data, the modal data change trend and the image feature change trend are obtained, and the attribute association between the modal data change trend and the image feature change trend is established.

[0009] Feature correction: Establish a reference library of modal data, and use correction methods to correct the relationship between image features and modal data features, thereby obtaining the trend of changes in corrected image features;

[0010] The correction method includes: obtaining modal data feature change data and image feature change trends in the same region and with attribute relationships based on regional association and attribute relationships; comparing the modal data feature change data with a reference library based on the image feature change trends to obtain a comparison difference; obtaining correction parameters based on the comparison difference; and correcting the image feature change trends with the correction parameters to obtain a corrected image feature change trend.

[0011] Risk assessment: Establish and obtain a personalized threshold set, correct the trend of image feature changes and input it into the assessment model to conduct cerebral infarction risk assessment, obtain an assessment value set, and when the assessment value set reaches or exceeds the set of personalized thresholds, trigger the early warning mechanism and issue a cerebral infarction risk alarm.

[0012] Furthermore, the modal data includes transcranial Doppler ultrasound (TCD) data, bioelectrical impedance analysis (BI) data, and electroencephalography (EEG) data;

[0013] The imaging features include structural features, functional features, and texture features. The trends in changes of imaging features include changes in hemodynamic parameters, lesion volume diffusion rate, and spatial distribution of white matter lesions.

[0014] By processing transcranial Doppler ultrasound (TCD) data through feature extraction and feature analysis, blood flow velocity and vascular resistance index features in the TCD data were obtained. Based on attribute association, a first attribute association was established between the changing trends of blood flow velocity features, the changing trends of vascular resistance index features, and the changes of hemodynamic parameters.

[0015] The feature correction includes a weighted correction for the association of attribute number one, to obtain the feature change trend of the image with attribute number one correction.

[0016] The weighted correction includes comparing the trend of blood flow velocity characteristics with normal blood flow velocity and the trend of vascular resistance index characteristics with normal vascular resistance index to obtain a first comparison difference. Based on the first comparison difference, a first correction parameter is obtained. The first correction parameter corrects the changes in hemodynamic parameters to obtain the corrected changes in hemodynamic parameters.

[0017] Risk assessment is performed by adjusting for changes in hemodynamic parameters.

[0018] Furthermore, by processing EEG data through feature extraction and feature analysis, frequency domain features and time domain features in the EEG data are obtained. The EEG features are then dimensionality reduced and principal component analysis (PCA) is used to extract frequency domain feature change data and time domain feature change data from the EEG data. Based on the data, the frequency domain feature change trend and time domain feature change trend are obtained.

[0019] A second attribute association was established between the frequency domain feature change trend, the time domain feature change trend, and the spatial distribution of white matter lesions based on attribute association.

[0020] The feature correction includes a comparison correction for the association of attribute number two;

[0021] The comparison correction includes: constructing a graph structure with EEG data features and brain regions in the comparison library as nodes; analyzing the first topological relationship between the graph structure and the spatial distribution of white matter lesions using a graph neural network (GNN); obtaining reference EEG data features by combining the EEG data features in the comparison library with the frequency domain feature change trend and the time domain feature change trend; obtaining the second topological relationship through the reference EEG data features; analyzing the difference distribution between the first and second topological relationships; and correcting the spatial distribution of white matter lesions based on the difference distribution to obtain the corrected spatial distribution of white matter lesions.

[0022] Risk assessment was conducted using the modified spatial distribution of white matter lesions.

[0023] Furthermore, by processing bioelectrical impedance analysis (BI) data through feature extraction and feature analysis, the electrical impedance characteristics of BI are obtained, and a third attribute association is established between the trend of BI electrical impedance characteristics and the lesion volume diffusion rate based on attribute association.

[0024] The feature correction includes regression correction for the association of attribute number three:

[0025] Input the electrical impedance characteristic change trend of BI and the lesion volume diffusion rate into the regression model to obtain the corrected lesion volume diffusion rate.

[0026] Risk assessment was performed using the modified lesion volume spread rate.

[0027] Furthermore, feature analysis methods for processing image data include:

[0028] Normal images and continuously monitored images are acquired. Image comparison analysis is used to obtain the difference points and changes in difference points between the normal images and continuously monitored images. The difference points include the difference data between the normal images and continuously monitored images, as well as the difference data between continuously monitored images. Based on the regions where the difference points are located in the image data, the brain regions where the difference points are located are determined, and the location of the image features is obtained.

[0029] Furthermore, the Dynamic Spatiotemporal Topology Alignment Network (DSTAN) processes image data in the following ways:

[0030] Based on the sequential characteristics of image data, image features at the same time but different levels are aligned into a unified spatial coordinate system to obtain a spatial image field.

[0031] A temporal correspondence is established between image data at the same level but at different times, and the temporal correspondence is combined with the spatial image field to obtain a spatiotemporal image field.

[0032] Furthermore, establishing and obtaining a personalized threshold set includes:

[0033] Virtual personal characteristic data of virtual individuals are constructed for comparison. Based on the virtual personal characteristic data and the trend of image characteristic changes, a basic threshold set is established. The basic threshold set includes hemodynamic parameter gradient threshold, lesion spread rate threshold, and white matter lesion distribution entropy threshold.

[0034] The process involves acquiring actual personal characteristic data, comparing it with virtual personal characteristic data to obtain comparative data, and then using this comparative data to correct the basic threshold set to obtain a personalized threshold set.

[0035] A data processing system based on image feature change trends, employing the aforementioned data processing method based on image feature change trends, is characterized in that the data processing system comprises:

[0036] Feature extraction module: Processes image data and modal data separately through feature analysis methods to obtain image features, image feature change data, modal data features, modal data feature change data, and the regions where the features are located. Establishes regional associations between image features and modal data features based on the regions where the features are located.

[0037] Feature Analysis Module: The module uses the Dynamic Spatiotemporal Topology Alignment Network (DSTAN) to process image data, generate a continuous spatiotemporal image field, map image feature change data and modal data feature change data onto the spatiotemporal image field, and calculate the modal data change trend and image feature change trend based on the data, and establish the attribute association between the modal data change trend and the image feature change trend.

[0038] Feature correction module: Establishes a reference library of modal data, and corrects the relationship between image features and modal data features through correction methods to obtain the trend of changes in corrected image features;

[0039] The correction method includes: obtaining modal data feature change data and image feature change trends in the same region and with attribute relationships based on regional association and attribute relationships; comparing the modal data feature change data with a reference library based on the image feature change trends to obtain a comparison difference; obtaining correction parameters based on the comparison difference; and correcting the image feature change trends with the correction parameters to obtain a corrected image feature change trend.

[0040] Risk assessment module: Establishes and obtains a personalized threshold set, corrects the trend of changes in image features and inputs it into the assessment model to conduct stroke risk assessment, obtains an assessment value set, and triggers an early warning mechanism when the assessment value set reaches or exceeds the set of personalized thresholds, and issues a stroke risk alarm.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This data processing method and system based on the changing trends of image features integrates image data with multiple modal data, such as TCD, BI, and EEG, breaking through the limitations of traditional single-data type analysis. Image data provides intuitive information on brain structure and lesions, while modal data supplements key features such as hemodynamics, brain tissue electrical activity, and electrical impedance from different physiological dimensions. TCD data can monitor blood flow velocity and vascular resistance in real time, BI data is sensitive to changes in pathological states such as cerebral edema, and EEG data reflects subtle changes in neuronal electrical activity. Through feature extraction and analysis, the combination of multi-source heterogeneous data achieves the correction of the changing trends of image features, thereby ensuring the accuracy of data processing.

[0043] Meanwhile, by using the changing trends of image features and modal data, compared with the single analysis of static data, it is more sensitive to capture the dynamic evolution of the brain's physiological state. Taking the time gradient of hemodynamic parameters as an example, it can reflect the rate of change of key indicators such as blood flow velocity and blood flow over time. This in-depth mining of changing trends enables the system to issue early warning signals in the early stage of cerebral infarction, when physiological indicators just begin to deviate from the normal range but have not yet reached the degree of obvious lesion.

[0044] By designing a personalized threshold set, the feature data of virtual individuals and real individuals are compared. Taking into account the differences in individual characteristics such as age, vascular elasticity, and collateral circulation capacity among different patients, the basic threshold set is precisely modified. This personalized threshold set avoids the problem of inaccurate warnings that may occur when dealing with different patients using a uniform fixed threshold. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the data processing method of the present invention;

[0046] Figure 2 This is a schematic diagram illustrating the attribute association of the present invention;

[0047] Figure 3 A schematic diagram illustrating the feature modification of this invention;

[0048] Figure 4 This is a schematic diagram of the data processing system of the present invention;

[0049] Figure 5 This is a schematic diagram of the image analysis method of the present invention;

[0050] Figure 6 This is a schematic diagram illustrating the risk assessment of the present invention;

[0051] Figure 7 This is a schematic diagram of the stroke risk assessment report of the present invention;

[0052] Figure 8 This is a trend diagram showing the changes in hemodynamic parameters of the present invention;

[0053] Figure 9 This is a trend graph showing the diffusion of lesion volume in this invention;

[0054] Figure 10 This is a trend diagram showing the distribution of white matter lesions according to the present invention;

[0055] Figure 11 This is a trend chart of the overall risk of this invention. Detailed Implementation

[0056] 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.

[0057] like Figure 1 - Figure 11 As shown, the present invention provides a technical solution: a data processing method based on the changing trends of image features, the data processing method comprising:

[0058] Feature extraction: Image data and modal data are processed separately using feature analysis methods to obtain image features, image feature variation data, modal data features, modal data feature variation data, and the regions where the features are located. Regional associations between image features and modal data features are established based on the regions where the features are located.

[0059] Feature analysis: The image data is processed using the Dynamic Spatiotemporal Topology Alignment Network (DSTAN) to generate a continuous spatiotemporal image field. The image feature change data and modal data feature change data are mapped onto the spatiotemporal image field. Based on the data, the modal data change trend and the image feature change trend are obtained, and the attribute association between the modal data change trend and the image feature change trend is established.

[0060] Feature correction: Establish a reference library of modal data, and use correction methods to correct the relationship between image features and modal data features, thereby obtaining the trend of changes in corrected image features;

[0061] The correction method includes: obtaining modal data feature change data and image feature change trends in the same region with attribute relationships based on regional association and attribute relationships; comparing the modal data feature change data with a reference library based on the image feature change trends to obtain the comparison difference; obtaining correction parameters based on the comparison difference; and correcting the image feature change trends with the correction parameters to obtain the corrected image feature change trends.

[0062] Risk assessment: Establish and obtain a personalized threshold set, correct the trend of image feature changes and input it into the assessment model to conduct cerebral infarction risk assessment, obtain an assessment value set, and when the assessment value set reaches or exceeds the set of personalized thresholds, trigger the early warning mechanism and issue a cerebral infarction risk alarm.

[0063] Modal data include transcranial Doppler ultrasound (TCD) data, bioelectrical impedance analysis (BI) data, and electroencephalography (EEG) data.

[0064] Imaging features include structural features, functional features, and texture features. Trends in imaging features include changes in hemodynamic parameters, lesion volume diffusion rate, and spatial distribution of white matter lesions.

[0065] By processing transcranial Doppler ultrasound (TCD) data through feature extraction and feature analysis, blood flow velocity and vascular resistance index features in the TCD data were obtained. Based on attribute association, a first attribute association was established between the changing trends of blood flow velocity features, the changing trends of vascular resistance index features, and the changes of hemodynamic parameters.

[0066] Feature correction includes weighted correction for association of attribute 1, to obtain the feature change trend of image 1 with correction 1;

[0067] Weighted correction involves comparing the trend of blood flow velocity characteristics with normal blood flow velocity and the trend of vascular resistance index characteristics with normal vascular resistance index to obtain the first comparison difference. The first correction parameter is obtained based on the first comparison difference. The first correction parameter corrects the changes in hemodynamic parameters to obtain the corrected changes in hemodynamic parameters.

[0068] Risk assessment is performed by adjusting for changes in hemodynamic parameters.

[0069] By processing EEG data through feature extraction and feature analysis, frequency domain features and time domain features in EEG data are obtained. Dimensionality reduction of EEG features is performed and principal component analysis (PCA) is used to extract frequency domain feature change data and time domain feature change data in EEG data. Based on the data, the frequency domain feature change trend and time domain feature change trend are obtained.

[0070] A second attribute association was established between the frequency domain feature change trend, the time domain feature change trend, and the spatial distribution of white matter lesions based on attribute association.

[0071] Feature corrections include comparison corrections used for association of attribute number two;

[0072] The comparison correction includes: constructing a graph structure with EEG data features and brain regions in the comparison library as nodes; analyzing the first topological relationship between the graph structure and the spatial distribution of white matter lesions using a graph neural network (GNN); obtaining reference EEG data features by combining the EEG data features in the comparison library with the frequency domain feature change trend and the time domain feature change trend; obtaining the second topological relationship through the reference EEG data features; analyzing the difference distribution between the first and second topological relationships; and correcting the spatial distribution of white matter lesions based on the difference distribution to obtain the corrected spatial distribution of white matter lesions.

[0073] Risk assessment was conducted using the modified spatial distribution of white matter lesions.

[0074] Bioelectrical impedance analysis (BI) data was processed by feature extraction and feature analysis to obtain the electrical impedance characteristics of BI. Based on attribute association, a third attribute association was established between the trend of BI electrical impedance characteristics and the lesion volume diffusion rate.

[0075] Feature corrections include regression corrections for associations of attribute number three:

[0076] Input the electrical impedance characteristic change trend of BI and the lesion volume diffusion rate into the regression model to obtain the corrected lesion volume diffusion rate.

[0077] Risk assessment was performed using the modified lesion volume spread rate.

[0078] Feature analysis methods for processing image data include:

[0079] Normal images and continuously monitored images are acquired. Image comparison analysis is used to obtain the difference points and changes in difference points between the normal images and continuously monitored images. The difference points include the difference data between the normal images and continuously monitored images, as well as the difference data between continuously monitored images. Based on the regions where the difference points are located in the image data, the brain regions where the difference points are located are determined, and the location of the image features is obtained.

[0080] The Dynamic Spatiotemporal Topology Alignment Network (DSTAN) processes image data including:

[0081] Based on the sequential characteristics of image data, image features at the same time but different levels are aligned into a unified spatial coordinate system to obtain a spatial image field.

[0082] A temporal correspondence is established between image data at the same level but at different times, and the temporal correspondence is combined with the spatial image field to obtain a spatiotemporal image field.

[0083] Establishing and obtaining a personalized threshold set includes:

[0084] Virtual personal characteristic data of virtual individuals are constructed for comparison. Based on the virtual personal characteristic data and the trend of image characteristic changes, a basic threshold set is established. The basic threshold set includes hemodynamic parameter gradient threshold, lesion spread rate threshold, and white matter lesion distribution entropy threshold.

[0085] The process involves acquiring actual personal characteristic data, comparing it with virtual personal characteristic data to obtain comparative data, and then using this comparative data to correct the basic threshold set to obtain a personalized threshold set.

[0086] A data processing system based on image feature change trends, employing the aforementioned data processing method based on image feature change trends, is characterized in that the data processing system comprises:

[0087] Feature extraction module: Processes image data and modal data separately through feature analysis methods to obtain image features, image feature change data, modal data features, modal data feature change data, and the regions where the features are located. Establishes regional associations between image features and modal data features based on the regions where the features are located.

[0088] Feature Analysis Module: The module uses the Dynamic Spatiotemporal Topology Alignment Network (DSTAN) to process image data, generate a continuous spatiotemporal image field, map image feature change data and modal data feature change data onto the spatiotemporal image field, and calculate the modal data change trend and image feature change trend based on the data, and establish the attribute association between the modal data change trend and the image feature change trend.

[0089] Feature correction module: Establishes a reference library of modal data, and corrects the relationship between image features and modal data features through correction methods to obtain the trend of changes in corrected image features;

[0090] The correction method includes: obtaining modal data feature change data and image feature change trends in the same region with attribute relationships based on regional association and attribute relationships; comparing the modal data feature change data with a reference library based on the image feature change trends to obtain the comparison difference; obtaining correction parameters based on the comparison difference; and correcting the image feature change trends with the correction parameters to obtain the corrected image feature change trends.

[0091] Risk assessment module: Establishes and obtains a personalized threshold set, corrects the trend of changes in image features and inputs it into the assessment model to conduct stroke risk assessment, obtains an assessment value set, and triggers an early warning mechanism when the assessment value set reaches or exceeds the set of personalized thresholds, and issues a stroke risk alarm.

[0092] The imaging data sources included CT and MRI. The control library consisted of normal modalities without stroke or other abnormalities. EEG data features from the control library were combined with frequency and temporal domain trend changes to obtain reference EEG data features. Since the EEG data features were normal, combining them with frequency and temporal domain trend changes yielded reference EEG data features. A first-order topological relationship based on normal EEG data features and a second-order topological relationship based on reference EEG data features were compared. The first-order topological relationship reflected the initial association pattern between EEG features and imaged brain regions in the control library, and the relationship between this association and the spatial distribution of white matter lesions. The second-order topological relationship reflected the relationship between EEG features combined with frequency and temporal domain trend changes and the imaged brain regions. Like a new association pattern between brain regions, spatial distribution analysis of node features in topological relationships 1 and 2 is performed to calculate the difference in feature values ​​of nodes in each brain region. This can be achieved by subtracting the two topological relationship matrices element by element to obtain a difference matrix. Each element in the difference matrix represents the difference in the association strength of the corresponding node pair in the two topological relationships. Based on the analysis results of the difference distribution, it is determined how to correct the spatial distribution of white matter lesions. For example, if the difference distribution shows that the association between EEG features and the image brain region in some brain regions is significantly enhanced, but the spatial distribution of white matter lesions has not fully reflected this change, the feature values ​​of white matter lesions in these brain regions can be increased accordingly, and vice versa. This is to correct the spatial distribution of white matter lesions. The original spatial distribution data of white matter lesions is fused with the difference distribution to obtain the corrected spatial distribution of white matter lesions.

[0093] like Figure 5 , Figure 6 and Figure 7As shown, a series of images are uploaded and loaded into the image analysis interface. The images are arranged in ascending order of time points. Then, clicking "Analyze Features" allows for risk assessment and trend analysis based on the images. The risk assessment analyzes the trends of various parameters, and the trend analysis displays these trends graphically, connecting actual data into points and forming curves. Since trend prediction exists, the predictive analysis only displays points on the x-axis (3-6). For structural images, baseline reference images and continuous monitoring image sequences are acquired. Image registration technology eliminates displacement deviations caused by changes in patient position. When acquiring continuous monitoring images, the consistency of the scanning protocol is ensured, including fixed imaging planes, field of view, and resolution parameters. The preprocessing stage involves brain tissue extraction and gray / white matter segmentation to provide standardized brain templates for subsequent feature analysis. A difference point detection algorithm identifies dynamically changing regions in the continuous image sequence and positions them based on anatomical brain region atlases.

[0094] Image contrast analysis methods include extracting image features through convolutional neural networks (ICNNs). Gray-level features include gray-level mean and variance; texture features include contrast, correlation, energy, and entropy; and shape features include the area, perimeter, and roundness of the lesion region. Image contrast analysis methods include image comparison, which involves obtaining the difference points between normal images and continuously monitored images. Normal images are used to determine if the influence is on normal brain regions. Continuously monitored images are obtained by taking pictures of the patient at different time points. By continuously taking pictures at multiple time points, changes in the monitored images can be obtained. By comparing normal images and continuously monitored images, difference points are obtained, and the change data of difference points on the continuously monitored images are analyzed to obtain image features and image feature change data. Sequence characteristics include shooting time and shooting level, where shooting level refers to the shooting depth.

[0095] Because of the primary attribute correlation between the trends in blood flow velocity characteristics, vascular resistance index characteristics, and changes in hemodynamic parameters, and because the trends in blood flow velocity characteristics and vascular resistance index characteristics directly affect hemodynamic parameters, a weighted correction is adopted. This is achieved by comparing the trends in blood flow velocity characteristics with normal blood flow velocity, and comparing the trends in vascular resistance index characteristics with normal vascular resistance index, obtaining a primary comparison difference. This primary comparison difference is used to correct the hemodynamic parameters. The change in hemodynamic parameters after correction is calculated as 1 + w2 + w3 × the change in hemodynamic parameters. The weight of the blood flow velocity characteristic is w2, and the weight of the vascular resistance index characteristic is w3. w2 can be obtained by comparing the trends in blood flow velocity characteristics with normal blood flow velocity, and w3 can be obtained by comparing the trends in vascular resistance index characteristics with normal vascular resistance index.

[0096] Dynamic Spatiotemporal Topology Alignment Network (DSTAN) is a deep learning framework specifically designed for processing multi-temporal and multimodal medical imaging data. Its core objective is to achieve accurate modeling and risk prediction of dynamic physiological processes through spatiotemporal registration, topology preservation, and multimodal correction.

[0097] The trends in imaging features include changes in hemodynamic parameters, the rate of lesion volume diffusion, and the spatial distribution of white matter lesions. Changes in hemodynamic parameters: By accurately monitoring the rate of change of parameters such as cerebral blood flow velocity, blood flow rate, and mean transit time over time, dynamic trends such as cerebral vascular stenosis and reduced blood perfusion can be keenly captured, providing core physiological indicators for early warning of cerebral infarction. The calculation of its temporal gradient can be based on continuous multi-temporal imaging data or TCD data, using mathematical methods such as numerical differentiation, which can reflect the evolution speed and direction of hemodynamic state and help to detect potential cerebral infarction risks in advance.

[0098] Lesion volume diffusion rate: For brain tissue regions that have shown mild signs of lesion, the lesion volume diffusion rate can intuitively quantify how fast the lesion region is expanding. By comparing and analyzing the changes in the area and volume of the lesion region at adjacent time points using imaging data, and combining the diffusion rate with geometric models, key information can be provided for judging the progression of cerebral infarction. When the lesion shows an accelerated diffusion trend, even if it has not yet reached the degree of obvious infarction, it can serve as an important basis for triggering early warning and assist doctors in early intervention.

[0099] Spatial distribution of white matter lesions: The spatial distribution pattern of white matter lesions in the brain contains rich pathological information. Spatial distribution can measure the complexity and non-uniformity of its distribution. This index is obtained by using image processing algorithms such as segmentation algorithms to extract white matter lesion areas and spatial statistical methods to calculate the distribution. It can reflect the range and pattern changes of damage to the microstructure of brain tissue. Abnormal changes such as the increase in the spatial distribution of white matter lesions can be used as one of the sensitive indicators for cerebral infarction risk assessment. Therefore, changes in hemodynamic parameters, lesion volume diffusion rate and spatial distribution of white matter lesions are used as references for data processing.

[0100] Brain biometrics (BI) technology acquires data features by applying a weak current to human brain tissue and measuring changes in its electrical impedance. Different brain tissues, such as normal brain tissue, edematous regions, and cerebral infarction lesions, have different electrical impedance characteristics. When pathological changes occur in brain tissue, such as edema and alterations in intracellular and extracellular fluid balance, their electrical impedance values ​​change accordingly. BI technology allows for real-time, non-invasive, and dynamic monitoring of these pathological changes in brain tissue, providing unique supplementary information for early diagnosis and risk warning of cerebral infarction, especially showing significant advantages in early detection of cerebral edema. Based on BI localization technology, BI data features are inverted into image data and spatially matched with brain regions in the images. This combines the electrical impedance changes detected by BI with the brain structure information displayed in the image data, achieving precise spatial correction of multimodal data and providing a more comprehensive assessment of the physiological and pathological state of brain tissue, offering richer feature information for data processing. BI technology is a method of detecting tissue pathophysiological conditions by applying a weak current to human tissue and measuring changes in its electrical impedance. In applying this method to the brain, different brain tissues, such as normal brain tissue, edematous regions, and cerebral infarction lesions, possess different electrical impedance characteristics. When pathological changes occur in brain tissue, such as edema and alterations in intracellular and extracellular fluid balance, their electrical impedance values ​​will change accordingly. Therefore, a linear relationship can be established between the trend of changes in the electrical impedance characteristics of brain tissue and the lesion volume diffusion rate. The lesion volume diffusion rate can be corrected by using the trend of changes in the electrical impedance characteristics of brain tissue. Linear regression is selected in the regression model. The collected data on the trend of changes in the electrical impedance characteristics of brain tissue and the data on the lesion volume diffusion rate are used as input and target variables, respectively. The regression model is trained, and the parameters of the model are adjusted by optimization algorithms such as gradient descent to make the model fit the training data to the greatest extent possible, thereby establishing a mathematical relationship between the two. New data on the trend of changes in the electrical impedance characteristics of brain tissue and the lesion volume diffusion rate are input into the trained regression model. The regression model corrects the lesion volume diffusion rate based on the input data and the previously learned relationship between the two, thus obtaining the correction.

[0101] Image feature change data can directly use the regional association and attribute association of image features. There is a relationship between image features, image feature change data and image feature change trend. Therefore, regional association and attribute association can be used directly among these three without restriction. The same applies to modal data and modal data feature change data.

[0102] In data processing, brain electrical activity (BI) detection can serve as an auxiliary means of imaging features. When mild exudation or early signs of cerebral edema are observed around cerebral blood vessels in images, BI can detect a decrease in the corresponding brain region's electrical impedance value, which shows a progressive change over time. This further confirms the occurrence and development trend of cerebral edema. Combining BI with the trend of changes in imaging features can more accurately assess the pathological state of brain tissue and improve the sensitivity and specificity of data processing. In particular, for the detection of early cerebral edema, it helps to identify potential stroke risks in advance.

[0103] EEG provides real-time monitoring of functional status, and imaging reveals long-term evolution of structure and blood flow. The combination of these two significantly improves sensitivity to early, minor lesions. EEG data analysis is a key technology for understanding brain activity. It records the electrical activity of neurons in the cerebral cortex by placing electrodes on the scalp. EEG signals typically include fluctuations in different frequency ranges, such as delta waves (0.5–4 Hz), theta waves (4–8 Hz), alpha waves (8–13 Hz), beta waves (13–30 Hz), and gamma waves (30–100 Hz). These waveforms can provide information about brain activity. EEG data analysis methods include time-domain analysis, frequency-domain analysis, spatial analysis, and machine learning and decoding. In time-domain analysis, event-related potentials such as P300 and N170 are used for target detection and cognitive task evoked potentials. In frequency-domain analysis, Fourier transform (FFT) is used to calculate the power spectral density (PSD) and analyze the energy of each frequency band. In spatial analysis, topographic mapping (Topoplot) and source localization (SourceLocalization) are used to visualize different brain regions. The distribution of electrical potential and power in brain regions can alter brain electrical activity. Since neuronal electrical activity depends on a sufficient supply of oxygen and energy, insufficient cerebral blood flow affects neuronal activity, manifesting as abnormalities in the electroencephalogram (EEG). In the early stages of cerebral infarction, even if structural imaging such as CT or MRI has not yet shown obvious abnormalities, EEG may already be able to capture subtle changes in electrical activity. For example, neurons in the infarct region may exhibit weakened electrical activity, enhanced slow waves such as delta waves and theta waves, or abnormal local rhythms. By using EEG source localization technology to invert abnormal EEG signals to specific brain regions and spatially match them with high-signal areas of white matter in imaging, a direct functional-structural correlation can be established, such as a second-region correlation. For example, if EEG data shows increased delta wave power in the left parietal lobe and a significant increase in the entropy of white matter lesion distribution in the corresponding region in the imaging, it suggests that neuronal function impairment and structural damage in this region are synergistically worsening. A second-attribute correlation can be established between the frequency domain characteristic change trend, the time domain characteristic change trend, and the spatial distribution of white matter lesions.

[0104] TCD (Transcranial Doppler) is a technique that uses Doppler ultrasound to detect intracranial hemodynamic parameters. It involves placing an ultrasound probe in an extracranial location, such as the temporal or occipital window, to transmit and receive ultrasound signals. This allows for the measurement of parameters such as blood flow velocity, direction, and vascular resistance index in the arteries at the base of the brain. Simultaneously, it acquires the location of the arteries within the brain and maps this location to the imaging data. TCD enables real-time, non-invasive monitoring of dynamic changes in cerebral blood flow, which is valuable for assessing cerebral vascular stenosis and cerebral blood flow reserve. In terms of data processing, TCD can detect early hemodynamic changes at sites of cerebral vascular stenosis, such as increased blood flow velocity, increased vascular resistance, and increased amplitude of blood flow fluctuations. Correcting and analyzing the hemodynamic parameters monitored by TCD against the trends in imaging features can enhance data processing capabilities and visualize the presence of cerebral blood vessels on images. When there is mild stenosis but not yet significant obstruction, if TCD detects a progressive increase in blood flow velocity and a gradually rising resistance index in that vessel segment, it suggests that the stenosis may be worsening and cerebral blood flow perfusion is gradually restricted. Combining this with the changing trends of imaging features can provide early warning of the risk of cerebral infarction. Therefore, it is necessary to establish a primary attribute correlation between the changing trends of blood flow velocity characteristics, the changing trends of vascular resistance index characteristics, and changes in hemodynamic parameters. Imaging data such as CT or MRI can provide morphological and hemodynamic information of blood vessels, allowing observation of vascular morphology and blood flow. When processing imaging data using the Dynamic Spatiotemporal Topology Alignment Network (DSTAN), a continuous spatiotemporal image field can be generated. By analyzing images from different time phases, the changing trends of blood flow velocity characteristics can be calculated, integrating imaging data with multiple modalities such as TCD, BI, and EEG. This approach breaks through the limitations of traditional single-data type analysis. Imaging data provides intuitive information on brain structure and lesions, while modal data supplements key features such as hemodynamics, brain tissue electrical activity, and electrical impedance from different physiological dimensions. TCD data can monitor blood flow velocity and vascular resistance in real time, BI data is sensitive to changes in pathological states such as cerebral edema, and EEG data reflects subtle changes in neuronal electrical activity. Through feature extraction and analysis, the combination of multi-source heterogeneous data enables the correction of trends in imaging feature changes, thereby ensuring the accuracy of data processing.

[0105] By designing personalized threshold sets, a comparison is made between the characteristic data of virtual and real individuals. Considering the differences in individual characteristics such as age, vascular elasticity, and collateral circulation capacity among different patients, the basic threshold set is precisely adjusted. This personalized threshold set avoids the inaccurate warning problems that may occur when using uniform, fixed thresholds for different patients. The model corrects the trend of changes in imaging features for stroke risk assessment. In the assessment model, the corrected trend of changes in imaging features is combined with the latest imaging feature data to predict the imaging feature data. The basic threshold set is used for the predictive imaging feature data. Patient-specific characteristics include age, vascular elasticity, and collateral circulation capacity. The specific threshold set for warnings is adjusted based on these individual characteristics, thus establishing a basic threshold set for judgment, under which comparisons with individual characteristics are made. The system obtains patient information by considering factors such as age, vascular elasticity, and collateral circulation capacity, along with individual patient characteristics and comparisons. This allows for the assessment of specific threshold sets. Since cerebral infarction manifests in multiple ways, such as hemodynamic parameters, lesion volume, and white matter lesion space, a threshold is designed for each aspect. A threshold set is obtained, and an alarm is triggered when any of the imaging features exceeds the threshold. By utilizing the trends in imaging features and modal data, compared to the analysis of static data alone, the system can more sensitively capture the dynamic evolution of the brain's physiological state. For example, the time gradient of hemodynamic parameters reflects the rate of change of key indicators such as blood flow velocity and blood volume over time. This in-depth analysis of trends enables the system to issue timely warning signals in the early stages of cerebral infarction, when physiological indicators begin to deviate from the normal range but have not yet reached a significant lesion level.

[0106] The early warning information includes risk level, warning location, and recommended clinical intervention measures to assist doctors in making diagnostic and treatment decisions. The assessment model captures early signs such as cerebral vascular stenosis and reduced blood flow perfusion by analyzing the changing trends of imaging features and modal data, thus achieving early warning of cerebral infarction risk. The early warning mechanism can monitor changes in the physiological state of the brain in real time to ensure that potential risks are detected in time before cerebral infarction occurs.

[0107] 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 embodiments and their equivalents.

Claims

1. A data processing system based on image feature change trend, characterized in that, The data processing system comprises: a feature extraction module: image data and modal data are respectively processed by a feature analysis method to obtain image features, image feature change data, modal data features, modal data feature change data and feature regions, and a region association between the image features and the modal data features is established according to the feature regions; a feature analysis module: the image data is processed by a dynamic spatio-temporal topology alignment network to generate a continuous spatio-temporal image field, the image feature change data and the modal data feature change data are mapped on the spatio-temporal image field, and a modal data change trend and an image feature change trend are obtained based on data calculation, and an attribute association between the modal data change trend and the image feature change trend is established; a feature correction module: a contrast library of the modal data is established, a correction between the image features and the modal data features is realized by a correction method, and a corrected image feature change trend is obtained; the correction method comprises: obtaining the modal data feature change data and the image feature change trend in the same region and with the attribute association according to the region association and the attribute relationship, comparing the modal data feature change data and the contrast library based on the image feature change trend, obtaining a comparison difference value, and obtaining a correction parameter according to the comparison difference value, the correction parameter correcting the image feature change trend to obtain the corrected image feature change trend; a risk assessment module: a personalized threshold set is established and obtained, the corrected image feature change trend is input into an assessment model for cerebral infarction risk assessment, an assessment value set is obtained, and when the assessment value set reaches or exceeds the set personalized threshold set, a warning mechanism is triggered to issue a cerebral infarction risk warning.

2. The data processing system based on the trend of image feature change according to claim 1, characterized in that: The modal data comprises transcranial Doppler ultrasound TCD data, bioelectrical impedance technology BI data and electroencephalogram EEG data; The image features comprise structural features, functional features and texture features, and the image feature change trend comprises changes in hemodynamic parameters, lesion volume diffusion rate and white matter lesion spatial distribution; The transcranial Doppler ultrasound TCD data is processed by feature extraction and feature analysis to obtain blood flow velocity features and vascular resistance index features in the transcranial Doppler ultrasound TCD data, and a first attribute association between the blood flow velocity feature change trend, the vascular resistance index feature change trend and the changes in hemodynamic parameters is established according to the attribute association; The feature correction comprises a weighted correction for the first attribute association to obtain a first corrected image feature change trend; The weighted correction comprises comparing the blood flow velocity feature change trend with a normal blood flow velocity and comparing the vascular resistance index feature change trend with a normal vascular resistance index to obtain a first comparison difference value, obtaining a first correction parameter according to the first comparison difference value, and correcting the changes in hemodynamic parameters by the first correction parameter to obtain corrected changes in hemodynamic parameters; The risk assessment is performed by correcting the changes in hemodynamic parameters.

3. The data processing system based on the trend of image feature change according to claim 2, characterized in that: The electroencephalogram (EEG) data is processed by feature extraction and feature analysis to obtain frequency domain features and time domain features in the EEG data. The EEG features are processed by dimension reduction, and the principal component analysis (PCA) method is used to extract frequency domain feature change data and time domain feature change data from the EEG data. Based on data calculation, the frequency domain feature change trend and the time domain feature change trend are obtained. A second attribute association is established between the frequency domain feature change trend, the time domain feature change trend, and the white matter lesion spatial distribution according to the attribute association. The feature correction includes a contrast correction for the second attribute association. The contrast correction includes constructing a graph structure with EEG data features in the contrast library and image brain regions as nodes, analyzing the first topological relationship between the graph structure and the white matter lesion spatial distribution through a graph neural network (GNN), combining the EEG data features in the contrast library with the frequency domain feature change trend and the time domain feature change trend to obtain reference EEG data features, obtaining a second topological relationship through the reference EEG data features, analyzing the difference distribution between the first topological relationship and the second topological relationship, and correcting the white matter lesion spatial distribution according to the difference distribution to obtain a corrected white matter lesion spatial distribution. Risk assessment is performed based on the corrected white matter lesion spatial distribution.

4. The data processing system based on image feature change trend according to claim 2, characterized in that: The bioimpedance (BI) data is processed by feature extraction and feature analysis to obtain BI impedance features. A third attribute association is established between the BI impedance feature change trend and the lesion volume diffusion rate according to the attribute association. The feature correction includes a regression correction for the third attribute association. The BI impedance feature change trend and the lesion volume diffusion rate are input into a regression model to obtain a corrected lesion volume diffusion rate. Risk assessment is performed based on the corrected lesion volume diffusion rate.

5. The data processing system based on the trend of image feature change according to claim 1, characterized in that: The feature analysis method processes image data, which includes: Normal images and continuous monitoring images are obtained, and the difference points and difference point change data between the normal images and the continuous monitoring images are obtained by image contrast analysis. The difference points include the difference data between the normal images and the continuous monitoring images, as well as the difference data between the continuous monitoring images. The brain region position of the difference points is determined based on the region where the difference points are located in the image data, and the region position of the image features is obtained.

6. The data processing system based on image feature change trend according to claim 1, characterized in that: The dynamic spatio-temporal topology alignment network (DSTAN) processes image data, which includes: Based on the sequence characteristics of the image data, the image features at the same time but different levels are aligned to a unified spatial coordinate system to obtain a spatial image field. A time correspondence relationship is established between image data at the same level but different times, and a spatio-temporal image field is obtained by combining the time correspondence relationship and the spatial image field.

7. The data processing system based on image feature change trend of claim 1, wherein: The personalized threshold set is established and obtained, which includes: Virtual personal feature data of a virtual contrast individual is constructed, and a basic threshold set is established based on the virtual personal feature data and the image feature change trend. The basic threshold set includes a hemodynamic parameter gradient threshold, a lesion diffusion rate threshold, and a white matter lesion distribution entropy threshold. Actual personal feature data is obtained, and the personal feature data and the virtual personal feature data are compared to obtain comparison data. The comparison data are used to modify the basic threshold set to obtain a personalized threshold set.

8. A data processing method based on image feature change trend, using the data processing system based on image feature change trend according to any one of claims 1-7, characterized in that, The data processing method comprises: Feature extraction: image data and modal data are respectively processed by a feature analysis method to obtain image features, image feature change data, modal data features, modal data feature change data and feature regions, and a region association between the image features and the modal data features is established according to the feature regions; Feature analysis: the image data is processed by a dynamic space-time topology alignment network (DSTAN) to generate a continuous space-time image field, the image feature change data and the modal data feature change data are mapped on the space-time image field, and a modal data change trend and an image feature change trend are obtained based on data calculation, and an attribute association between the modal data change trend and the image feature change trend is established; Feature correction: a contrast database of the modal data is established, a correction between the image features and the modal data features is realized by a correction method, and a corrected image feature change trend is obtained; The correction method comprises: obtaining the modal data feature change data and the image feature change trend in the same region and with the attribute association according to the region association and the attribute relationship, comparing the modal data feature change data and the contrast database based on the image feature change trend, obtaining a comparison difference value, obtaining a correction parameter according to the comparison difference value, correcting the image feature change trend by the correction parameter, and obtaining the corrected image feature change trend; Risk assessment: a personalized threshold set is established and obtained, the corrected image feature change trend is input into an assessment model for cerebral infarction risk assessment, an assessment value set is obtained, and when the assessment value set reaches or exceeds the set personalized threshold set, a warning mechanism is triggered, and a cerebral infarction risk warning is issued.

Images