A cerebral infarction risk prediction method, system, storage medium and program product

Abstract

The application belongs to the technical field of risk prediction, and discloses a cerebral infarction risk prediction method, system, storage medium and program product, which comprises the following steps: acquiring brain magnetic resonance imaging data and clinical characteristic data of a target examinee; performing blood vessel high signal sign interpretation on a liquid attenuated inversion recovery sequence image to generate a blood vessel high signal sign interpretation result; performing cerebral artery ring integrity interpretation on a time fly method sequence image to generate a cerebral artery ring integrity interpretation result; inputting the blood vessel high signal sign interpretation result, the cerebral artery ring integrity interpretation result and the clinical data into a pre-constructed cerebral infarction risk prediction model to calculate a prediction probability of cerebral infarction of the target examinee within a preset time period and generate a risk prediction probability value. TIA The application can identify high-risk groups of cerebral infarction within 30 days in the early stage of the disease, guide individualized hierarchical management of intensified intervention and conventional follow-up, and has good clinical transformation application potential.

Description

Technical Field

[0001] This invention relates to the field of risk prediction technology, and in particular to a method, system, storage medium, and program product for predicting the risk of cerebral infarction. Background Technology

[0002] Transient ischemic attack (TIA) TIA (This is a clinical syndrome characterized by transient, reversible impairment of brain neurological function; symptoms usually resolve completely within 24 hours.) Although... TIA While it does not cause permanent neurological damage, it is a key precursor to ischemic stroke, significantly increasing the risk of cerebral infarction within 30 days of onset, thus placing a heavy burden on patients' quality of life and social medical resources.

[0003] at present, TIA The diagnosis and management of cerebral ischemia mainly rely on clinical manifestations, routine imaging examinations and auxiliary laboratory tests. Although risk assessment tools such as the ABCD2 score commonly used in clinical practice can guide the stratified management of patients to a certain extent, these tools are mostly based on clinical symptoms and some imaging indicators, and do not fully consider the anatomical variations of cerebrovascular system and hemodynamic status. They are difficult to fully reflect the cerebral blood flow reconstruction capacity and local ischemia tolerance, resulting in some high-risk patients not being fully identified and targeted for intervention.

[0004] With the development of magnetic resonance imaging (MRI) technology, the methods for assessing the anatomy and functional status of cerebral blood vessels have become increasingly abundant. Among them, FLAIR vascular hyperintensity (FVH), as an imaging sign of abnormal blood flow in the internal carotid artery or large cerebral arteries, has been proven to be closely related to cerebral hemodynamic disturbances and can indicate potential large vessel obstruction or insufficient bypass circulation. TIA An important predictor of recurrence of ischemic stroke. The Circle of Willis (CoW), an important annular arterial structure at the base of the brain, directly affects the collateral circulation capacity of cerebral blood flow due to its anatomical integrity. An incomplete Circle of Willis structure is considered one of the important risk factors for the occurrence and recurrence of cerebral ischemic events.

[0005] However, in clinical practice, the attention and application of Willis ring variations are still limited, and the combination of FVH sign and Willis ring incompleteness is also problematic. TIA There is a lack of applied research in short-term risk prediction, and its synergistic predictive value and clinical efficacy need to be further elucidated.

[0006] In summary, the existing TIACurrent risk assessment methods suffer from insufficient utilization of cerebrovascular imaging features. There is an urgent need to introduce multi-dimensional imaging indicators that can simultaneously reflect dynamic blood flow status and static vascular structure, and to construct individualized risk prediction models with both high discriminative power and good calibration, in order to achieve [the desired outcome]. TIA Early and accurate identification and intervention of the risk of cerebral infarction in patients.

[0007] Therefore, how to provide a method, system, storage medium, and program product for predicting the risk of cerebral infarction is an urgent problem to be solved. Summary of the Invention

[0008] This invention provides a method, system, storage medium, and program product for predicting the risk of cerebral infarction, in order to solve the problems of the above-mentioned technologies in the prior art.

[0009] According to a first aspect of the present invention, a method for predicting the risk of cerebral infarction is provided.

[0010] In one embodiment, the method for predicting the risk of cerebral infarction includes: Acquire brain magnetic resonance imaging (MRI) data and clinical characteristic data of the target subject; wherein, the brain MRI data includes liquid attenuation inversion recovery sequence images and time-of-flight sequence images; Vascular hyperintensity features were interpreted from liquid attenuation inversion recovery sequence images, and vascular hyperintensity feature interpretation results were generated. The integrity of the circle of arteries in the brain is determined by time-flight sequence images, and the results of the integrity determination of the circle of arteries in the brain are generated. The results of vascular high signal sign interpretation, cerebral arterial circle integrity interpretation, and clinical data are input into a pre-constructed cerebral infarction risk prediction model to calculate the predicted probability of cerebral infarction in the target subject within a preset time period and generate a risk prediction probability value to complete the risk prediction of cerebral infarction.

[0011] According to a second aspect of the present invention, a system for predicting the risk of cerebral infarction is provided.

[0012] In one embodiment, the stroke risk prediction system includes: The data acquisition module is used to acquire brain magnetic resonance imaging data and clinical characteristic data of the target subject; wherein, the brain magnetic resonance imaging data includes liquid attenuation inversion recovery sequence images and time-of-flight sequence images; The vascular high signal sign interpretation module is used to interpret vascular high signal signs in liquid attenuation inversion recovery sequence images and generate vascular high signal sign interpretation results. The cerebral arterial circle integrity interpretation module is used to interpret the cerebral arterial circle integrity in time-flight sequence images and generate cerebral arterial circle integrity interpretation results. The cerebral infarction probability prediction module is used to input the interpretation results of vascular high signal signs, the interpretation results of the integrity of the circle of arteries in the brain, and clinical data into a pre-constructed cerebral infarction risk prediction model, calculate the predicted probability of the target subject having cerebral infarction within a preset time period, and generate a risk prediction probability value to complete the risk prediction of cerebral infarction.

[0013] According to a third aspect of the present invention, the present invention provides a storage medium comprising a stored program, wherein, when the program is executed, the device on which the storage medium is located executes the above-described method for predicting the risk of cerebral infarction.

[0014] According to a fourth aspect of the present invention, the present invention provides a program product that, when executed by a processor, implements the steps of the above-described method for predicting the risk of cerebral infarction.

[0015] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: 1) This invention combines two imaging indicators, FLAIR vascular hyperintensity and Willis circle incompleteness, to reflect the risk of cerebral ischemia from the perspectives of dynamic blood flow status and static vascular structure, respectively. This overcomes the deficiency of insufficient information from a single imaging indicator and significantly improves the comprehensiveness and accuracy of prediction.

[0016] 2) The cerebral infarction risk prediction model constructed based on logistic regression in this invention has an area under the curve of 0.912, and its sensitivity and specificity are superior to those of individual indicators, significantly improving the prediction accuracy. TIA The ability to differentiate the risk of short-term cerebral infarction.

[0017] 3) This invention uses nomograms to visualize multifactor regression models, enabling intuitive and quantitative reading of individualized risk probabilities, and providing a convenient risk quantification tool for clinical practice.

[0018] 4) This invention uses receiver operating characteristic curves, Hosmer-Lemeshow tests, and decision curve analysis to verify the model from multiple dimensions, ensuring its reliability in terms of discrimination ability, calibration, and net clinical benefit.

[0019] 5) This invention can be used in TIA It can identify high-risk individuals for cerebral infarction within 30 days of the onset of the disease, guide individualized stratified management of enhanced intervention and routine follow-up, and has good potential for clinical translational application.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0022] Figure 1 This is a flowchart illustrating a method for predicting the risk of cerebral infarction according to an exemplary embodiment; Figure 2 This is a block diagram illustrating the principle of a stroke risk prediction system according to an exemplary embodiment; Figure 3 This is a prediction illustrated according to an exemplary embodiment. TIA Nodal diagram of patients with concurrent cerebral infarction; Figure 4 This is the ROC curve of a stroke risk prediction model illustrated according to an exemplary embodiment; Figure 5 This is a calibration curve of a stroke risk prediction model illustrated according to an exemplary embodiment; Figure 6 This is a decision curve of a stroke risk prediction model illustrated according to an exemplary embodiment. Detailed Implementation

[0023] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some portions and features of certain embodiments may be included in or replace portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims and all available equivalents thereof. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0024] The modules in the apparatus or system of this application can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0025] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0026] Figure 1 An embodiment of a method for predicting the risk of cerebral infarction according to the present invention is shown.

[0027] In this optional embodiment, the method for predicting the risk of cerebral infarction includes: Step S101: Obtain the target subject ( TIAThe brain magnetic resonance imaging data and clinical features of the patient; wherein the brain magnetic resonance imaging data includes fluid attenuation inversion recovery (FLAIR) sequence images and time-of-flight (3D-TOF-MRA) sequence images; Step S102: Interpret the vascular high signal sign (FVH sign) of the liquid attenuation inversion recovery sequence image and generate the vascular high signal sign interpretation result; Step S103: Perform cerebral arterial circle (Willis circle) integrity assessment on the time-flight sequence images and generate cerebral arterial circle integrity assessment results; Step S104: Input the interpretation results of vascular high signal sign, the interpretation results of cerebral arterial circle integrity and clinical data into the pre-constructed cerebral infarction risk prediction model, calculate the predicted probability of the target subject having cerebral infarction within a preset time period (30 days), and generate a risk prediction probability value to complete the risk prediction of cerebral infarction.

[0028] In this optional embodiment, the step of interpreting vascular high-signal characteristics in the liquid attenuation inversion recovery sequence image and generating vascular high-signal characteristic interpretation results includes: Determine whether localized, punctate, tubular, or serpentine high signals appear on the surface of the gray matter near the liquid attenuation inversion recovery sequence image; Determine whether a high signal appears in two consecutive layers, or in more than two places in the same layer; When a high signal appears in two consecutive layers, or in two or more places in the same layer, the interpretation result of the vascular high signal sign is positive. When a high signal does not appear in two consecutive layers, or appears in fewer than two places in the same layer, the interpretation result of the vascular high signal sign is negative.

[0029] In this optional embodiment, the step of interpreting the integrity of the circle of arteries in the time-lapse sequence images and generating the interpretation result includes: Using the multiplanar reconstruction method, three orthogonal plane reconstruction images—coronal, sagittal, and axial—are generated with the circle of arteries of the brain as the center, and customized oblique reconstruction is performed along the course of each component vessel of the circle of arteries. Using the maximum density projection method, a series of maximum density projection images are generated by rotating around the head and tail axis at preset angle steps to highlight vascular structures and assess the continuity and symmetry of the blood vessels in the cerebral arterial circle. The anterior communicating artery of the circle of arteries, the A1 segment of the bilateral anterior cerebral arteries, the terminal segments of the bilateral internal carotid arteries, the P1 segment of the bilateral posterior cerebral arteries, and the posterior communicating artery were observed on time-lapse sequence images. If any of the following structures is missing or has a diameter smaller than a preset threshold (diameter <1mm): the current communicating artery, the A1 segment of both anterior cerebral arteries, the terminal portion of both internal carotid arteries, the P1 segment of both posterior cerebral arteries, or the posterior communicating artery, a positive result for incomplete cerebral arterial circle is generated. If there are no missing segments or diameters smaller than a preset threshold in the current communicating artery, the A1 segment of both anterior cerebral arteries, the terminal segments of both internal carotid arteries, the P1 segment of both posterior cerebral arteries, and the posterior communicating artery, a negative result indicating an intact cerebral arterial circle is generated.

[0030] In this optional embodiment, the process of constructing the pre-built cerebral infarction risk prediction model includes: Using the occurrence of cerebral infarction within a preset time period as the dependent variable, the results of vascular high signal sign interpretation, cerebral arterial circle integrity interpretation, and clinical data were successively incorporated into a univariate (logistic) regression model for preliminary screening to determine candidate variables; among which, the clinical data included: gender, age, history of diabetes, and transient ischemic attack (TIA). TIA Duration; Multicollinearity is diagnosed for candidate variables, and the variance inflation factor test is used to exclude variables with multicollinearity. The variables diagnosed by collinearity were included in a multivariate (Logistic) regression model. Using the forward and backward stepwise regression method and the minimum Akaike information criterion as the variable selection criterion, a cerebral infarction risk prediction model was constructed, and the risk prediction probability value was output. The discrimination ability, calibration, and net clinical benefit of the cerebral infarction risk prediction model were validated.

[0031] In this optional embodiment, the formula for calculating the linear risk prediction value is: ; In the formula, This represents the probability value for risk prediction. The intercept; These are the regression coefficients; Interpretation results of high signal intensity in blood vessels; The result of interpreting the integrity of the circle of arteries in the brain; For gender; For age; History of diabetes; The duration of a transient ischemic attack (TIA).

[0032] In this optional embodiment, the validation of the infarction risk prediction model's discriminative ability, calibration accuracy, and net clinical benefit includes: The operating characteristic curve (ROC curve) of the target subjects is plotted based on the individualized predictive probability. The individualized predictive probability is divided into a series of cutoff values. The sensitivity and specificity at each cutoff value are calculated, and the optimal cutoff value is determined by using the Youden index method. After sorting the target subjects according to their individualized predicted probabilities, they were grouped, and the average predicted probability and actual observed incidence rate of each group were calculated. Calibration curves were plotted, and the Hosmer-Lemeshow (HL) test was used to evaluate the calibration of the cerebral infarction risk prediction model in order to verify the reliability of the individualized predicted probabilities. Using the optimal cutoff value as the clinical decision threshold, the net benefit under different threshold probabilities was calculated, and the clinical net benefit of the cerebral infarction risk prediction model was evaluated using a pre-constructed decision curve analysis.

[0033] In this optional embodiment, the step of calculating the net benefit at different threshold probabilities using the optimal cutoff value as the clinical decision threshold, and evaluating the clinical net benefit of the cerebral infarction risk prediction model using a pre-constructed decision curve analysis includes: Using the optimal cutoff value as the clinical decision boundary, a range of threshold probabilities is set, the true positive and false positive results are calculated at each threshold probability, and the net benefit is calculated at different threshold probabilities using the net benefit formula, and a decision curve is plotted. The decision curve was compared with reference curves for both full intervention and no intervention strategies to assess the clinical net benefit of the stroke risk prediction model.

[0034] Figure 2 An embodiment of a stroke risk prediction system according to the present invention is shown.

[0035] In this optional embodiment, the cerebral infarction risk prediction system includes: The data acquisition module 201 is used to acquire brain magnetic resonance imaging data and clinical characteristic data of the target subject; wherein, the brain magnetic resonance imaging data includes liquid attenuation inversion recovery sequence images and time-of-flight sequence images; The high signal intensity vascular feature interpretation module 202 is used to interpret the high signal intensity vascular feature in liquid attenuation inversion recovery sequence images and generate the high signal intensity vascular feature interpretation results. The cerebral arterial circle integrity interpretation module 203 is used to interpret the cerebral arterial circle integrity in time-flight sequence images and generate cerebral arterial circle integrity interpretation results. The cerebral infarction probability prediction module 204 is used to input the interpretation results of vascular high signal signs, the interpretation results of the integrity of the circle of arteries in the brain, and clinical data into a pre-constructed cerebral infarction risk prediction model, calculate the predicted probability of the target subject having cerebral infarction within a preset time period, and generate a risk prediction probability value to complete the risk prediction of cerebral infarction.

[0036] To facilitate understanding of the above technical solutions of the present invention, the following further explains the above technical solutions of the present invention from the perspective of architecture and principle, as follows: This invention explores the application of a logistic regression model based on the Flair vascular hyperintensity (FVH) sign and Willis circle incompleteness in transient ischemic attacks (TIAs). TIA The predictive value of cerebral infarction in patients within 30 days.

[0037] A United Imaging UMR588 1.5T superconducting MRI system was selected, and all patients underwent routine cranial MRI within 24 hours of admission or visit. Patients were placed in the supine position, and a 16-channel head coil was used to cover the entire cranial region from the top of the head to the foramen magnum. FSE sequences were used for T1WI and T2WI imaging. T1WI parameters were TR 2085 ms, TE 11.72 ms, slice thickness 5 mm, and slice interval 1.5 mm; T2WI parameters were TR 4100 ms, TE 104 ms, slice thickness 5 mm, and slice interval 1.5 mm. For DWI imaging, EPI sequences were used with parameters of TR 3206 ms, TE 91.1 ms, slice thickness 5 mm, slice interval 1.5 mm, and FOV 230 mm × 230 mm. The FLAIR sequence parameters were: TR 8000ms, TE 109.78ms, TI 2300ms, slice thickness 5mm, interslice spacing 1.5mm, and FOV 230mm×230mm; the MRA sequence parameters were: TR 23.6ms, TE 6.8ms, slice thickness 27.09mm, and FOV 250mm×250mm. Patient comfort and safety were closely monitored during the scan. After imaging, post-processing techniques were used for 3D reconstruction and image analysis to provide clearer views of brain tissue structure and lesion areas.

[0038] The United Imaging UMR588 1.5T superconducting magnetic resonance imaging system used in this invention is a new generation 1.5T magnetic resonance imaging device launched by United Imaging Healthcare, equipped with the Light Shuttle 2.0 (uCS2.0) imaging platform. This device has the following core technical features: Light Shuttle 2.0 imaging technology based on compressed sensing (CS), integrating mainstream acceleration technologies such as parallel acquisition (PI) and partial Fourier transform (PF) to achieve intelligent optimization of data acquisition and image reconstruction; equipped with a 16-channel head-specific phased array coil to provide high signal-to-noise ratio signal reception; incorporating deep learning reconstruction technology, using convolutional neural networks for intelligent noise reduction while maintaining image sharpness; a built-in uReal 3D intelligent rendering engine to support high-resolution rendering and display of intracranial blood vessels; and the accompanying uWS United Imaging Medical Image Processing Software (post-processing workstation), providing functional modules such as vascular analysis, simulated endoscopy, image fusion, and brain perfusion analysis.

[0039] The 3D image reconstruction and post-processing of this invention are performed on the United Imaging uWS medical image processing workstation. This workstation integrates the following functional modules directly related to this invention: a vascular analysis module, providing automatic extraction of vascular centerlines, vascular contour measurement, and multi-angle 3D reconstruction; a simulated endoscope module, supporting automatic or manual extraction of vascular centerlines and simulating vascular lumen views through 3D display from different angles; an image fusion module, employing image registration technology to fuse different types of image sequences (such as FLAIR and MRA), supporting adjustments to the fusion ratio, window width, window level, and pseudo-color scheme; a uReal 3D intelligent rendering engine, supporting high-resolution rendering of intracranial vessels to achieve clear and realistic presentation of aneurysms and vascular spatial structures; and employing ray projection, maximum density projection, and surface reconstruction to achieve visualized rendering of 3D data structures.

[0040] The complete technical workflow for 3D reconstruction and image analysis is as follows: First, 3D-TOF MRA raw data is acquired. Intracranial arterial imaging is performed using the 3D-TOF sequence on the United Imaging UMR 588. This sequence utilizes the inflow enhancement effect, making flowing blood appear as a high signal, while stationary background tissue is saturated with low signal due to multiple radiofrequency pulses, thus obtaining high-contrast cerebral vascular images. Simultaneously, FLAIR sequence images are acquired for FVH sign interpretation and image fusion reference. All raw data are automatically transferred to the UWS post-processing workstation in DICOM 3.0 standard format. The data includes complete DICOM header information, including patient information, scan parameters, spatial coordinates, and other metadata. On the workstation, the raw images are quality-assessed, including checking for motion artifacts to eliminate blurring or ghosting caused by patient head movement; evaluating the signal-to-noise ratio to ensure the contrast between vessels and background meets interpretation requirements; and confirming the scan range to verify complete coverage of the cranial region from the top of the head to the foramen magnum, especially the level of the Circle of Willis.

[0041] The raw images were quality assessed on the workstation, including: (a) checking for motion artifacts—excluding blurring or ghosting caused by patient head movement; (b) assessing the signal-to-noise ratio—ensuring that the contrast between blood vessels and the background met the interpretation requirements; and (c) confirming the scan range—verifying whether the brain region from the top of the head to the foramen magnum was fully covered, especially the level of the Circle of Willis.

[0042] Multiplanar reconstruction is a fundamental step in 3D post-processing. Its core principle is to rearrange the original axial 3D volumetric data into any specified planar direction using mathematical interpolation algorithms, generating coronal, sagittal, and arbitrary oblique 2D reconstructed images without altering the original voxel data. Using the Willis ring as the center, reconstructed images in three orthogonal planes (coronal, sagittal, and axial) are generated. The reconstruction slice thickness is set to 1.0-1.5 mm, and the interlayer spacing is 0.5-0.8 mm. A linear interpolation algorithm is used to reduce staircase artifacts and improve the smoothness and visual quality of the reconstructed images.

[0043] The calculation formula using the linear interpolation algorithm is as follows: ; In the formula, To reconstruct the signal strength; and The signal strength of adjacent known voxels; , , The coordinates are the spatial location coordinates. Compared with Nearest Neighbor Interpolation, this algorithm can significantly reduce staircase artifacts and improve the smoothness and visual quality of the reconstructed image.

[0044] Customized oblique reconstruction was performed based on the course characteristics of the vessels constituting the Circle of Willis: an anterior circulation oblique reconstruction plane was set along the course of the anterior communicating artery-anterior cerebral artery A1 segment; a posterior circulation oblique reconstruction plane was set along the course of the posterior communicating artery-posterior cerebral artery P1 segment; and a terminal internal carotid artery oblique reconstruction plane was set along the supraclinoid segments of the bilateral internal carotid arteries to their bifurcation. By adjusting the rotation and tilt angles of the reconstruction planes, the continuous course of each vessel was traced segment by segment.

[0045] Maximum Intensity Projection (MIP) is the core 3D visualization technique in TOF MRA cerebral angiography. Its basic algorithm involves traversing the 3D volumetric data along a set of parallel rays from the observer's perspective, extracting the voxel value with the highest signal intensity on each ray, and projecting it to form a 2D image. Because blood flow exhibits high signal intensity in TOF MRA, MIP images can highlight vascular structures while suppressing background tissue. MIP reconstruction is performed on the entire 3D-TOF MRA volumetric data, rotating around the head-to-tail axis in 5-10 degree increments to generate a 360-degree series of MIP images. The projection thickness is set to the full volume to ensure complete display of the Circle of Willis and its major branches. Rotating and observing MIP images from various angles allows for a comprehensive assessment of the continuity, symmetry, and presence of developmental defects or significant stenosis in the constituent vessels of the Circle of Willis.

[0046] This solution integrates the following unique processing steps, distinct from existing technologies, on the basis of conventional 3D reconstruction: Optical Shuttle 2.0 reconstruction fidelity based on uCS compressed sensing, employing a pseudo-random undersampling mode during K-space data acquisition, and restoring the complete image through a nonlinear iterative reconstruction algorithm, achieving equivalent resolution improvement without increasing scan time; Intelligent denoising and super-resolution reconstruction based on deep learning, using a convolutional neural network to intelligently denoise the original image, improving the signal-to-noise ratio while maintaining image edge sharpness, particularly beneficial for the clear display of fine blood vessels; FLAIR-MRA multimodal registration and fusion, utilizing the uWS image fusion module to achieve precise spatial registration of FLAIR and MRA sequences, employing a mutual information-based registration algorithm to automatically align the spatial positions of the two sequences, superimposing the FVH high-signal region with the stenosis / occlusion location of the responsible vessel, and dynamically switching the display of the two modalities by alternately adjusting the fusion weights, intuitively confirming the correspondence between the spatial distribution of FVH and the vascular abnormalities of the Circle of Willis, providing a more accurate variable assignment basis for subsequent Logistic regression modeling.

[0047] Two radiologists at the deputy director level or above independently interpret the images. In case of disputes, the two physicians shall reach a consensus through consultation. Interpretation of FVH sign: FVH sign is determined according to the following criteria: (1) localized punctate, tubular or serpentine high signal on the surface of gray matter near the brain on FLAIR sequence images; (2) appearing on two consecutive layers, or more than two on the same layer; if the above two criteria are met, it is considered a positive FVH sign, otherwise it is negative. Criteria for determining the incompleteness of the Circle of Willis: On 3D-TOF-MRA images, observe the anterior communicating artery, the A1 segment of the bilateral anterior cerebral arteries, the terminal end of the bilateral internal carotid arteries, the P1 segment of the bilateral posterior cerebral arteries and the posterior communicating artery. If any structure is missing or significantly narrowed, it is determined to be a positive incompleteness.

[0048] The R4.4.3 software package was used for statistical analysis of the data. The mean ± standard deviation was used to describe the distribution of quantitative data. The independent samples t test was used to compare the differences between two groups, and one-way ANOVA was used to compare the differences among multiple groups.

[0049] One-way ANOVA was used to compare differences in quantitative data among multiple groups. The specific steps are as follows: Step 1: Test the preconditions. (a) Normality test: Use the Shapiro-Wilk test to assess the normality of the data in each group. The Shapiro-Wilk test is a normality test method based on order statistics, suitable for small sample (n<50) and medium sample (n=50-2000) data. Test the null hypothesis H0: The data follows a normal distribution; if P<0.05, reject the null hypothesis, and consider the data to be non-normally distributed. For large sample data (n>2000), the Kolmogorov-Smirnov test or Anderson-Darling test can be used.

[0050] (b) Homogeneity of Variance Test: The Levene test is used to assess whether the variances among multiple groups are homogeneous. The Levene test is not sensitive to the normality assumption and is suitable for non-normally distributed data. The null hypothesis H0 is tested: the variances of each group are equal; if P < 0.05, the null hypothesis is rejected, indicating that the variances are unequal. If the variances are unequal, Welch-corrected ANOVA is used. This method does not require homogeneity of variance and corrects for the effects of unequal variances by adjusting the degrees of freedom.

[0051] Step 2: Calculation The formula for calculating the statistic is: ; In the formula, The mean square between groups (reflecting the variation between different groups) is calculated using the following formula: ; In the formula, The sum of squares between groups; Number of groups; The within-group mean square (reflecting random variation within the same group) is calculated using the following formula: ; In the formula, The sum of squares within the group; This represents the total sample size. The larger the value, the more significant the difference between groups is relative to the random variation within groups.

[0052] Step 3: Determine overall significance. At a significance level of α=0.05, if P<0.05, it indicates that there is a statistically significant difference between at least two groups, triggering subsequent pairwise comparisons (post-hoc analysis).

[0053] Step 4: Post-hoc pairwise comparisons. After the overall ANOVA is significant, the LSD-t test is used to perform targeted comparisons between any two groups to identify which specific pairs of groups have differences. This invention uses the Willis ring integrity status (complete / partially missing / multiple missing subgroups, etc.) as the grouping variable and continuous variables such as age and BMI as the dependent variables for analysis.

[0054] use LSD-t The test involves pairwise comparisons between groups. LSD-t Least Significant Difference t-test The pooled mean squared error test (MSE), proposed by Fisher, is a classic method for post-hoc pairwise comparisons in ANOVA. Its core principle is to use the pooled mean squared error (MSE) to compare pairs of samples after the overall significance of ANOVA. MSE ) build Statistical measure. The specific steps are as follows: Step 1: The formula for calculating the Pooled Standard Error is as follows: ; In the formula, This is the mean squared error within groups in ANOVA (reflecting the joint variation within each group). and The first Group and No. Sample size of the group. Use merging. Instead of using the variances of just two groups, this method can improve estimation accuracy (with greater degrees of freedom) and is consistent with ANOVA's assumption of homogeneity of overall variance.

[0055] Step 2: The formula for calculating the LSD-t statistic is: ; In the formula, and This represents the mean of two groups. The degrees of freedom of the statistic are ( The total sample size is... (Number of groups).

[0056] Step 3: The formula for calculating the LSD critical value is as follows: ; In the formula, For degrees of freedom of The distribution has two-sided critical values.

[0057] Step 4: Significance assessment. If If the difference is not statistically significant (P < α), then the difference between the two groups is statistically significant. The observed differences for each pair of comparisons are then compared with... LSD The critical values ​​are compared, and all significant pairs are marked.

[0058] Step 5: Application in this invention: using TIA Taking the duration grouping (<1 hour, 1-6 hours, >6 hours) as an example, first perform ANOVA (if P<0.05), then perform pairwise analysis on each subgroup. LSD-t The test was conducted to identify which two duration intervals showed a significant difference in patient age or BMI.

[0059] Assuming the ANOVA results show a significant age difference among the three groups (P=0.012), then three pairwise comparisons will be performed: (1) <1 hour group vs. 1-6 hour group; (2) <1-hour group vs.>6-hour group; (3) 1-6 hour group vs. >6 hour group.

[0060] If the P-value for comparison (1) is 0.031, the P-value for comparison (2) is 0.008, and the P-value for comparison (3) is 0.274, it indicates that the age of patients in the <1 hour group is significantly lower than that of the other two groups, while there is no statistically significant age difference between the 1-6 hour group and the >6 hour group.

[0061] The split-card method, after the overall chi-square test shows significance, decomposes the total chi-square value into multiple independent single-degree-of-freedom components according to a pre-designed orthogonal comparison (each component corresponds to a specific pairwise comparison or linear combination comparison), thereby accurately locating the source of differences between groups without increasing the cumulative risk of Type I errors. The specific steps are as follows: Inspection steps: Step 1: Overall Chi-Square Test. First, perform a Pearson chi-square test on the R×C contingency table (R rows × C columns). The calculation formula is: ; In the formula, To observe the frequency; Expected frequency ( = Row total × Column total / Total total). If the overall... The significance at the α=0.05 level (P<0.05) indicates that there are differences in the composition ratios between at least two groups, triggering subsequent segmentation.

[0062] Step Two: Design the Orthogonal Comparison Matrix. Construct a set of mutually orthogonal comparison vectors (the sum of the comparison coefficients is zero, and the dot product of any two comparison vectors is zero), each comparison vector corresponding to a specific inter-group comparison hypothesis. For example, three comparisons (AvsB, AvsC, BvsC) can be designed with the following orthogonal comparisons: (a) Comparison 1: A vs (B+C) — Comparison coefficient: [2 / 3, -1 / 3, -1 / 3]; (b) Comparison 2: B vs C - Comparison coefficient: [0, 1 / 2, -1 / 2].

[0063] Orthogonality ensures the independence of each comparison, and the overall... It can be strictly decomposed into various comparisons The sum is: ; Step 3: Calculate the chi-square value for each segment. For each orthogonal contrast component, construct a 2×2 sub-table (or a 2×k sub-table) and calculate the corresponding... Components. The degrees of freedom of each component are... .

[0064] Step 4: Significance assessment. (Check...) The distribution table determines the p-value for each component. In this invention, taking the comparison of FVH positivity rate between the cerebral infarction group and the non-cerebral infarction group as an example, after confirming the difference in positivity rates between the two groups by the overall chi-square test, further segmentation chi-square tests can be performed by gender stratification (male subgroup, female subgroup) to assess whether the predictive value of FVH is affected by gender.

[0065] Improved Scheme – Weighted Orthogonal Decomposition and Sample Size Adaptive Correction Factor: This scheme introduces the following improvements based on the conventional chi-square split: (a) Weighted orthogonal decomposition: When constructing the orthogonal comparison matrix, the square root of the sample size of each group is used as the weight. This allows the contributions of the large sample group to be appropriately weighted, preventing the random fluctuations of the small sample group from dominating the segmentation results. (b) Sample Size Adaptive Correction Factor (ACF): Definition In the formula, This represents the total sample size. Number of groups; To minimize the sample size of the subgroup. Each The components are adjusted by multiplying by the ACF, when the sample sizes of each group are unbalanced (as in this invention). TIA For groups with a duration of less than 1 hour and a smaller sample size, ACF can automatically amplify the correction strength and control the false positive rate.

[0066] Bonferroni correction is a conservative multiple comparison correction method that controls the overall Type I error rate (FWER) by adjusting the significance level of each comparison. The correction process of this invention is as follows: Step 1: Determine the number of comparisons. Calculate the total number of all pairwise comparisons. In this invention, ; This represents the number of groups. For example, if we group... TIA The duration is divided into 3 groups (<1h, 1-6h, >6h). =3 × 2 / 2 = 3 times. If the four subgroups of FVH positive / negative and Willis circle intact / incomplete are compared simultaneously, then =4×3 / 2=6 times.

[0067] Step 2: The formula for calculating the level after correction is as follows: ; This invention takes the original =0.05, =6, then That is, the p-value for each pairwise comparison must be less than 0.0083 to be considered statistically significant.

[0068] Step 3: Perform the corrected significance test. For each... LSD-t The original p-value obtained from the test (or split chi-square test) is compared with... Comparison: (a) If Praw < 0.0083, the difference is statistically significant; (b) If Praw ≥ 0.0083, the difference is not statistically significant.

[0069] Step 4: Report the corrected results. Use different markers in the results table to distinguish the significance before and after correction: (a) Marked with "*" indicates =0.05 level is significant (before correction); (b) The symbol “†” indicates =0.0083 level is significant (after Bonferroni correction).

[0070] Both-sided P-values ​​report the original value and the significance determination after correction.

[0071] Univariate and multivariate binary logistic regression analyses were used to analyze the association between FVH and Willis loop incompleteness and the occurrence of cerebral infarction. Step 1: Univariate Logistic Regression Analysis.

[0072] Using "whether a cerebral infarction has occurred (0=no, 1=yes)" as the binary dependent variable, each independent variable (FVH sign, Willis loop incompleteness, gender, age, BMI, smoking history, drinking history, history of hypertension, history of diabetes, etc.) is categorized one by one. TIA Duration was included in a univariate logistic regression model for initial screening. Odds ratio (OR), 95% confidence interval (95% CI), and Wald test p-value were calculated for each variable.

[0073] Logistic regression models are suitable for binary outcome variables, mapping linear predictions to a (0,1) probability space through logistic transformation: ; In the formula, This represents the probability value for risk prediction. The intercept; These are the regression coefficients; for Interpretation results of high signal intensity in blood vessels; for Results of interpretation of the integrity of the circle of arteries in the brain; for Duration of transient ischemic attack (TIA). , representing the independent variable For every additional unit (for a binary variable, i.e., from 0 to 1), the risk of stroke becomes OR times the original risk.

[0074] Step 2: Variable selection criteria.

[0075] Using P<0.10 as the cutoff value for univariate analysis (a relatively lenient threshold to avoid omitting potentially important variables), variables with P<0.10 in the univariate analysis were included in the subsequent multivariate analysis. The results of this invention are shown in Table 1.

[0076] Table 1 Multifactor Analysis Table Step 3: Collinearity diagnosis; Two methods are used to assess the degree of collinearity among independent variables: (a) Spearman / Pearson correlation matrix: Calculate the Pearson correlation coefficient (normal distribution) or Spearman rank correlation coefficient (non-normal distribution) between continuous variables to assess the degree of multicollinearity among independent variables. Set |r|>0.7 as the threshold for strong correlation—if the correlation coefficient between two independent variables exceeds 0.7, it indicates the existence of multicollinearity, requiring selection in the multifactor model or the use of methods such as ridge regression.

[0077] (b) Variance Inflation Factor (VIF) Test: The formula for calculating the VIF value of each independent variable after fitting the multifactor model is as follows: ; In the formula, For the first The regression determination coefficients are: one independent variable as the dependent variable and the remaining independent variables as predictors. A VIF > 5 (or a more stringent standard VIF > 10) indicates significant multicollinearity, and related variables should be removed or combined.

[0078] A multivariate logistic regression prediction model for cerebral infarction was constructed using forward-backward stepwise selection and the Akaike Information Criterion (AIC) as the variable selection criterion. The specific steps are as follows: The theoretical basis of the AIC criterion: Proposed by Japanese statistician Hirotugu Akaike in 1973, AIC is a standard for measuring the goodness of fit of a statistical model for predicting the risk of stroke. Based on the Kullback-Leibler divergence in information theory, AIC measures the information loss of the stroke risk prediction model. A smaller AIC value indicates a better model fit. The formula for calculating the penalty that also considers the complexity (number of parameters) of the stroke risk prediction model is as follows: ; In the formula, The maximum likelihood value for the stroke risk prediction model. This represents the number of parameters (including the intercept) in the stroke risk prediction model. The first term... The second term measures the goodness of fit of the stroke risk prediction model; a smaller value indicates a better fit. The complexity of the stroke risk prediction model is penalized, with the penalty increasing as the number of parameters increases. AIC strikes a balance between model goodness of fit and simplicity, avoiding overfitting.

[0079] Stepwise regression steps: Step 1: Define the initial model and candidate variable pool.

[0080] Dependent variable: Occurrence of cerebral infarction (0 / 1); Candidate independent variable pool: FVH sign (yes / no), Willis loop incompleteness (yes / no), gender (male / female), age (continuous), BMI (continuous), smoking history (yes / no), alcohol consumption history (yes / no), history of hypertension (yes / no), history of diabetes (yes / no) TIA Duration (≥1 hour / <1 hour).

[0081] Step 2: Forward Selection.

[0082] Start with an empty model containing only the intercept: (a) Try adding each variable from the candidate variable pool to the model one by one; (b) Calculate the AIC value after adding each univariate; (c) Select the variable that causes the largest decrease in AIC and add it to the model; (d) Repeat the above process until no variable is added that can further reduce AIC.

[0083] Step 3: Backward Elimination. After each variable is added: (a) Try removing one variable from the current model one by one; (b) Calculate the AIC value after removal; (c) If removing a variable causes the AIC to decrease (i.e., the variable has a negative contribution to the model), then remove it; (d) Repeat the test until the AIC cannot be reduced by removing any variable.

[0084] Step 4: Bidirectional Iteration Termination. Alternately perform forward selection and backward elimination until: (a) It is not possible to further reduce AIC by adding new variables or removing existing variables; (b) The model reaches convergence. The infarction risk prediction model obtained at this point is the optimal AIC model.

[0085] like Figure 3 As shown, with the occurrence of cerebral infarction as the dependent variable, the following factors were considered: FVH sign, Willis circle incompleteness, gender, age, BMI, smoking history, alcohol consumption history, history of hypertension, history of diabetes, and... TIA Duration and other variables were identified as potential predictors. Minimum AIC was used as the variable selection criterion. A multivariate stepwise regression model for predicting stroke risk was constructed using a forward-backward approach. A nomogram was used to visualize the stroke risk prediction model.

[0086] The results showed that the optimal model for predicting the risk of cerebral infarction was obtained when AIC=201.6. At this point, the independent variables included FVH sign, Willis loop incompleteness, gender, age, history of diabetes, and TIA Duration.

[0087] As shown in Table 2, the optimal cerebral infarction risk prediction model was obtained when AIC=201.6, which included 6 independent predictors.

[0088] Table 2. Predictor Factors of the Optimal Cerebral Infarction Risk Prediction Model Final prediction probability formula: ; In the formula, This represents the probability value for risk prediction. The intercept; These are the regression coefficients; Interpretation results of high signal intensity in blood vessels; is the interpretation result of cerebral arterial circle integrity; is gender; is age; is history of diabetes mellitus; is duration of transient ischemic attack.

[0089] As shown in Figure 4 , ROC curve analysis shows that the AUC value of the cerebral infarction risk prediction model is 0.913 (0.877-0.948), indicating that the model has excellent distinguishing ability. Receiver Operating Characteristic (ROC) curve is a standard method for evaluating the distinguishing ability of binary cerebral infarction risk prediction models. The specific analysis steps of the present invention are as follows: ROC curve construction: Step 1: Calculate the prediction probability. Substitute the values of 6 predictors of each patient into the final Logistic regression model to calculate the individual prediction probability of cerebral infarction occurring within 30 days: ; wherein, is the linear prediction value of the -th patient.

[0090] Step 2: Generate ROC curve data points: (a) Divide the prediction probability from 0 to 1 into a series of cutoff values, which are usually 101 points from 0.00, 0.01, 0.02,..., 0.99, to 1.00; (b) For each cutoff value c, calculate the four-fold table: true positive (TP) where the prediction probability is ≥ c and the onset actually occurs, false positive (FP) where the prediction probability is ≥ c and no onset occurs, false negative (FN) where the prediction probability is < c and the onset actually occurs, and true negative (TN) where the prediction probability is < c and no onset occurs; (c) The formula for calculating sensitivity is: ; (d) With (1-specificity) as the abscissa and sensitivity as the ordinate, connect the data points under all cutoff values to draw the ROC curve.

[0091] AUC calculation and interpretation: Step 3: Calculate AUC (Area Under the Curve).

[0092] The area under the ROC curve is approximately calculated by the Trapezoidal Rule: ; wherein, is the There are 100 data points. The AUC value ranges from 0.5 to 1.0.

[0093] Step 4: 95% confidence interval for AUC.

[0094] The standard error of AUC is estimated using the DeLong nonparametric method, and a 95% CI is constructed. The DeLong nonparametric method is based on the U statistic theory, considers the correlation of paired data, and does not require assumptions about the distribution of the data.

[0095] Model Comparison and Validation: Step 5: Determine the optimal cutoff value.

[0096] The formula for calculating the Youden Index using the Youden Index method is as follows: ; Choose to The maximum cutoff value is taken as the optimal critical point. The Youden exponent of the model in this invention is 0.713, corresponding to a sensitivity of 0.849 and a specificity of 0.864.

[0097] Step Six: Model Comparison.

[0098] (a) Plot the ROC curves of individual indicators (FVH only, Willis loop incomplete only, FVH + Willis loop combined) and the full model (6 indicators) on the same coordinate system; (b) The DeLong test was used to compare the significance of differences in AUC among the models; (c) The results of this invention show that the AUC of the whole model is 0.912 (0.874-0.949), which is significantly higher than that of FVH alone (AUC=0.762), Willis loop alone (AUC=0.675) and FVH+Willis loop combination (AUC=0.831), indicating that the predictive ability of the model is significantly improved after integrating clinical variables.

[0099] Model calibration and clinical utility assessment: Step 7: Model calibration evaluation.

[0100] like Figure 5 As shown, the calibration curve of the prediction model was plotted. The results show that the predicted probability and the actual observed probability are basically consistent, indicating good model calibration (Hosmer-Lemeshow X). 2 =2.989, P=0.935). Plot the calibration curve: (a) Patients were divided into 10 groups (deciles) after being sorted according to their predicted probabilities. (b) Calculate the average predicted probability and the actual observed incidence rate for each group; (c) Plot a scatter plot with the average predicted probability on the x-axis and the actual incidence rate on the y-axis, and fit a Loess smoothing curve; (d) The ideal calibration curve should coincide with the 45-degree diagonal line—the greater the deviation, the worse the calibration.

[0101] The Hosmer-Lemeshow test for this model showed χ²=2.989, P=0.935 (P>0.05 indicates that the model is well calibrated).

[0102] Step 8: As Figure 6 As shown, the decision curve results indicate that when the threshold probability is within the range of 0-90%, the predictive model consistently achieves a positive net benefit greater than that of the full intervention model and the no-intervention model. The clinical decision curve analysis (DCA) is as follows: (a) Define the threshold probability pt as the clinical decision threshold—that is, when the predicted probability exceeds this value, intervention measures are taken; (b) Calculation in different Net profit below ; In the formula The weight of false positives relative to true positives; (c) Simultaneously plot the net benefit curves of the "full intervention" strategy (all patients are intervened) and the "no intervention" strategy (no one is intervened) as a reference; (d) The net benefit of this model is positive and better than the reference strategy in the threshold probability range of 0-90%, indicating that it has broad clinical applicability.

[0103] The comparison between the two groups showed no statistically significant differences in BMI, smoking history, alcohol consumption history, and history of hypertension. Gender, age, and history of diabetes were also considered. TIA Statistically significant differences were observed in duration, FVH positivity rate, and positivity rate of incomplete Willis rings. This indicates that male sex, older age, and a history of diabetes are more likely to be associated with these factors. TIA Patients with a duration of ≥1 hour, positive FVH, or incomplete Willis circle have a higher risk of developing cerebral infarction within 30 days.

[0104] Using Logistic Regression to Perform Pairing of Various Indicators TIA Correlation analysis of patients developing cerebral infarction within 30 days. Univariate logistic regression analysis showed that gender, age, history of diabetes, TIA Duration, FVH positivity, and incomplete Willis circle are TIA Risk factors for cerebral infarction within 30 days in patients were identified. Multivariate binary logistic regression analysis was used to determine factors such as gender, age, and history of diabetes. TIA Duration (≥1 hour), positive FVH sign, and positive Willis ring incompleteness are TIAIndependent risk factors for cerebral infarction within 30 days were identified. A nomogram model was constructed based on these indicators, and the model showed good calibration (Hosmer-Lemeshow). =2.989, P=0.935).

[0105] The predictive value and significance of each factor for cerebral infarction were compared. The AUC of FVH alone was 0.762, with a sensitivity of 0.698, a specificity of 0.826, and a Youden index of 0.524. The AUC of Willis loop alone was 0.675, with a sensitivity of 0.453, a specificity of 0.897, and a Youden index of 0.350. The AUC of FVH combined with Willis loop was 0.831, with a sensitivity of 0.849, a specificity of 0.745, and a Youden index of 0.593. The combined AUC of the full model (6 factors) was 0.912, with a sensitivity of 0.849, a specificity of 0.864, and a Youden index of 0.713, which was superior to the individual predictions of each factor.

[0106] Using the occurrence of cerebral infarction as the dependent variable, the following factors were considered: FVH sign, incomplete Willis circle, sex, age, BMI, smoking history, alcohol consumption history, history of hypertension, and history of diabetes. TIA Duration and other variables were considered potential predictors. Minimum AIC was used as the variable selection criterion. A multivariate stepwise regression model was constructed using a forward-backward approach to predict the occurrence of cerebral infarction. The predictive model was visualized using a nomogram. Results showed that the optimal model was obtained when AIC = 201.6. At this point, the independent variables included FVH sign, Willis loop incompleteness, gender, age, and history of diabetes were... TIA Duration.

[0107] ROC curve analysis showed that the model's AUC value was 0.913, indicating that the model has good discriminative ability. The calibration curve of the prediction model was plotted, and the results showed that the predicted probabilities were basically consistent with the actual observed probabilities, indicating good model calibration (Hosmer-Lemeshow). =2.989, P=0.935). The decision curve results show that when the threshold probability is in the range of 0-90%, the prediction model can obtain a positive net benefit that is greater than that of the full intervention model and the no-intervention model.

[0108] In summary, the method for predicting the probability of cerebral infarction in this invention retrospectively included 270 cases from a certain hospital during a certain period. TIA Patients were enrolled and followed up for 30 days. Based on whether they developed cerebral infarction within 30 days, they were divided into a non-cerebral infarction group (n=184) and a cerebral infarction group (n=86). The positive rates of Flair vascular hyperintensity sign, Willis circle incompleteness, and clinical indicators were compared between the two groups. A variable selection strategy using logistic regression analysis was employed to construct... TIA A predictive model for the risk of stroke in patients with concurrent cerebral infarction was used to analyze the predictive model's effectiveness by comparing the area under the receiver operating characteristic (AUC) curve. TIA The predictive value of cerebral infarction within 30 days in patients. Results: Difference analysis revealed that gender (79 (42.93) vs 48 (55.81)) =3.902, P=0.048), age (66.43±11.76 vs 71.8±9.16, t=-4.084, P<0.001), history of diabetes (31 (16.85) vs 32 (37.21), =13.583, P<0.001), TIA Duration (6 (3.26) vs 32 (37.21)) =55.853, P<0.001), FVH positivity rate (32 (17.39) VS 60 (69.77), =71.571, P<0.001), Willis loop incomplete positive rate (19(10.33)VS39(45.35), =42.620, P<0.001) indicates a statistically significant difference. Logistic regression analysis determined the differences in gender, age, history of diabetes, and other relevant factors. TIA Duration (≥1 hour), positive FVH sign, and positive Willis ring incompleteness are TIA Independent risk factors for cerebral infarction within 30 days were identified in the patients. A nomogram prediction model was constructed based on the above indicators, and the model showed good calibration (Hosmer-Lemeshow). =2.989, P=0.935). Decision curve analysis showed that the cerebral infarction risk prediction model had a high net clinical benefit. ROC curve analysis showed that the AUC value of the cerebral infarction risk prediction model was 0.912 (0.874-0.949), the sensitivity was 0.849, and the specificity was 0.864, which was better than each indicator alone. Conclusion: Based on gender, age, history of diabetes, TIA A stroke risk prediction model established based on duration (≥1 hour), Flair vascular hyperintensity sign, and Willis circle incompleteness has good predictive value and can be used for... TIA Quantitative tools are provided for the early identification and intervention of cerebral infarction in patients.

[0109] According to a third embodiment of the present invention, a storage medium for a method for predicting the risk of cerebral infarction is provided, the storage medium including a stored program, wherein the program controls the device where the storage medium is located to execute the method for predicting the risk of cerebral infarction described above when the program is running.

[0110] Since the processing and functions implemented by the storage medium in this embodiment are basically the same as the embodiments, principles and examples of the aforementioned methods, any details not covered in the description of this embodiment can be found in the relevant descriptions in the aforementioned embodiments, and will not be elaborated upon here.

[0111] According to a fourth embodiment of the present invention, a program product of a method for predicting the risk of cerebral infarction is provided, comprising a computer program that, when executed by a processor, implements the steps of the method for predicting the risk of cerebral infarction described above.

[0112] Since the processing and functions implemented by the computer program product in this embodiment are basically the same as the embodiments, principles and examples of the aforementioned methods, any details not covered in the description of this embodiment can be found in the relevant descriptions in the aforementioned embodiments, and will not be elaborated upon here.

[0113] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.

Claims

1. A method for predicting the risk of cerebral infarction, characterized in that, include: Acquire brain magnetic resonance imaging (MRI) data and clinical characteristic data of the target subject; wherein, the brain MRI data includes liquid attenuation inversion recovery sequence images and time-of-flight sequence images; Vascular hyperintensity features were interpreted from liquid attenuation inversion recovery sequence images, and vascular hyperintensity feature interpretation results were generated. The integrity of the circle of arteries in the brain is determined by time-flight sequence images, and the results of the integrity determination of the circle of arteries in the brain are generated. The results of vascular high signal sign interpretation, cerebral arterial circle integrity interpretation, and clinical data are input into a pre-constructed cerebral infarction risk prediction model to calculate the predicted probability of cerebral infarction in the target subject within a preset time period and generate a risk prediction probability value to complete the risk prediction of cerebral infarction.

2. The method for predicting the risk of cerebral infarction according to claim 1, characterized in that, The process of interpreting vascular high-signal characteristics in liquid attenuation inversion recovery sequence images and generating vascular high-signal characteristic interpretation results includes: Determine whether localized, punctate, tubular, or serpentine high signals appear on the surface of the gray matter near the liquid attenuation inversion recovery sequence image; Determine whether a high signal appears in two consecutive layers, or in more than two places in the same layer; When a high signal appears in two consecutive layers, or in two or more places in the same layer, the interpretation result of the vascular high signal sign is positive. When a high signal does not appear in two consecutive layers, or appears in fewer than two places in the same layer, the interpretation result of the vascular high signal sign is negative.

3. The method for predicting the risk of cerebral infarction according to claim 1, characterized in that, The process of interpreting the integrity of the circle of arteries in time-flight sequence images and generating the interpretation results includes: Using the multiplanar reconstruction method, three orthogonal plane reconstruction images—coronal, sagittal, and axial—are generated with the circle of arteries of the brain as the center, and customized oblique reconstruction is performed along the course of each component vessel of the circle of arteries. Using the maximum density projection method, a series of maximum density projection images are generated by rotating around the head and tail axis at preset angle steps to highlight vascular structures and assess the continuity and symmetry of the blood vessels in the cerebral arterial circle. The anterior communicating artery of the circle of arteries, the A1 segment of the bilateral anterior cerebral arteries, the terminal segments of the bilateral internal carotid arteries, the P1 segment of the bilateral posterior cerebral arteries, and the posterior communicating artery were observed on time-lapse sequence images. If any of the following structures is missing or has a diameter smaller than a preset threshold: the current communicating artery, the A1 segment of both anterior cerebral arteries, the terminal portion of both internal carotid arteries, the P1 segment of both posterior cerebral arteries, or the posterior communicating artery, a positive result for incomplete cerebral arterial circle is generated. If there are no missing segments or diameters smaller than a preset threshold in the current communicating artery, the A1 segment of both anterior cerebral arteries, the terminal segments of both internal carotid arteries, the P1 segment of both posterior cerebral arteries, and the posterior communicating artery, a negative result indicating an intact cerebral arterial circle is generated.

4. The method for predicting the risk of cerebral infarction according to claim 1, characterized in that, The construction process of the pre-built cerebral infarction risk prediction model includes: Using whether cerebral infarction occurs within a preset time period as the dependent variable, the results of vascular high signal sign interpretation, the results of cerebral arterial circle integrity interpretation, and clinical data are incorporated into a univariate regression model for preliminary screening to determine candidate variables; among which, the clinical data include: gender, age, history of diabetes, and duration of transient ischemic attack. Multicollinearity is diagnosed for candidate variables, and the variance inflation factor test is used to exclude variables with multicollinearity. The variables diagnosed by collinearity were incorporated into a multivariate regression model. Using the forward and backward stepwise regression method, and with the minimum Akaike information criterion as the variable selection criterion, a cerebral infarction risk prediction model was constructed, and the risk prediction probability value was output. The discrimination ability, calibration, and net clinical benefit of the cerebral infarction risk prediction model were validated.

5. The method for predicting the risk of cerebral infarction according to claim 4, characterized in that, The formula for calculating the linear prediction value of the risk is as follows: ； In the formula, This represents the probability value for risk prediction. The intercept; These are the regression coefficients; Interpretation results of high signal intensity in blood vessels; The result of interpreting the integrity of the circle of arteries in the brain; For gender; For age; History of diabetes; The duration of a transient ischemic attack (TIA).

6. The method for predicting the risk of cerebral infarction according to claim 4, characterized in that, The validation of the infarction risk prediction model's discriminative ability, calibration accuracy, and net clinical benefit includes: The operating characteristic curve of the target subjects was plotted based on the individualized predictive probability. The individualized predictive probability was divided into a series of cutoff values. The sensitivity and specificity at each cutoff value were calculated, and the optimal cutoff value was determined using the Youden index method. After sorting the target subjects according to their individualized predicted probabilities, they were grouped, and the average predicted probability of each group was calculated as well as the actual observed incidence rate. A calibration curve was plotted, and the calibration degree of the cerebral infarction risk prediction model was evaluated using a test to verify the reliability of the individualized predicted probabilities. Using the optimal cutoff value as the clinical decision threshold, the net benefit under different threshold probabilities was calculated, and the clinical net benefit of the cerebral infarction risk prediction model was evaluated using a pre-constructed decision curve analysis.

7. The method for predicting the risk of cerebral infarction according to claim 6, characterized in that, The process of using the optimal cutoff value as the clinical decision threshold, calculating the net benefit under different threshold probabilities, and using pre-constructed decision curve analysis to evaluate the clinical net benefit of the stroke risk prediction model includes: Using the optimal cutoff value as the clinical decision boundary, a range of threshold probabilities is set, the true positive and false positive results are calculated at each threshold probability, and the net benefit is calculated at different threshold probabilities using the net benefit formula, and a decision curve is plotted. The decision curve was compared with reference curves for both full intervention and no intervention strategies to assess the clinical net benefit of the stroke risk prediction model.

8. A system for predicting the risk of cerebral infarction, characterized in that, include: The data acquisition module is used to acquire brain magnetic resonance imaging data and clinical characteristic data of the target subject; wherein, the brain magnetic resonance imaging data includes liquid attenuation inversion recovery sequence images and time-of-flight sequence images; The vascular high signal sign interpretation module is used to interpret vascular high signal signs in liquid attenuation inversion recovery sequence images and generate vascular high signal sign interpretation results. The cerebral arterial circle integrity interpretation module is used to interpret the cerebral arterial circle integrity in time-flight sequence images and generate cerebral arterial circle integrity interpretation results. The cerebral infarction probability prediction module is used to input the interpretation results of vascular high signal signs, the interpretation results of the integrity of the circle of arteries in the brain, and clinical data into a pre-constructed cerebral infarction risk prediction model, calculate the predicted probability of the target subject having cerebral infarction within a preset time period, and generate a risk prediction probability value to complete the risk prediction of cerebral infarction.

9. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the storage medium to perform the stroke risk prediction method according to any one of claims 1 to 7.

10. A program product comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the stroke risk prediction method according to any one of claims 1 to 7.

Images