Cerebrovascular disease risk prediction method
By collecting and integrating basic clinical, dynamic physiological, and imaging data, a fusion prediction model is constructed, which solves the problem of the one-sidedness of single data assessment in existing technologies and achieves multi-dimensional coverage and accurate prediction of cerebrovascular disease risk.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-03-27
AI Technical Summary
Existing methods for predicting the risk of cerebrovascular diseases rely on a single type of data, which makes it difficult to comprehensively cover the multidimensional influencing factors of disease risk, resulting in one-sided assessments and insufficient early warning capabilities.
We collect basic clinical data, dynamic physiological parameter data, and imaging feature data. Through feature extraction and fusion, we construct a fusion prediction model and output risk prediction results.
It enables multi-dimensional coverage of cerebrovascular disease risk from static health baseline, dynamic physiological changes and anatomical structure status, improving the accuracy of early risk identification and prediction, and generating differentiated risk reports.
Smart Images
Figure CN121747918A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of disease risk prediction, and more particularly, to a cerebral vascular disease risk prediction method. BACKGROUND
[0002] As a major chronic disease with high mortality and morbidity worldwide, the pathogenesis of cerebral vascular disease is complex and the early symptoms are occult. Once acute onset, it often leads to irreversible neurological damage, causing heavy burden to the patient's family and the social medical system. Therefore, early and accurate prediction of the risk of cerebral vascular disease is the key to reducing the incidence and improving the prognosis, and is also the research focus in the field of disease prevention.
[0003] Existing cerebral vascular disease risk prediction methods mostly rely on a single type of data to construct an evaluation system, such as a scoring scale model using only basic clinical data (such as age, blood pressure, medical history, etc.) or a structural analysis method based only on static image data. This single data-driven model cannot fully cover the multidimensional influencing factors of disease risk. Basic clinical data can only reflect the static health baseline and cannot capture the dynamic fluctuation of physiological indicators such as heart rate and blood pressure, which are often early warning signals of vascular dysfunction. Although image data can present anatomical structure characteristics, it lacks correlation analysis with individual physiological state and medical history information, resulting in one-sidedness of risk assessment and poor prediction of cerebral vascular disease risk.
[0004] To solve the above problems, a technical solution is provided. SUMMARY
[0005] To overcome the above-mentioned defects of the prior art, embodiments of the present application provide a cerebral vascular disease risk prediction method to solve the problems raised in the background art.
[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution:
[0007] A cerebral vascular disease risk prediction method, characterized in that it comprises the following steps:
[0008] Step S1: Collecting basic clinical data, dynamic physiological parameter data and image feature data of a target object, and constructing a data set;
[0009] Step S2: Extracting features from the basic clinical data, dynamic physiological parameter data and image feature data respectively, and fusing the extracted features to obtain a fused feature set;
[0010] Step S3: Constructing a fused prediction model through the fused feature set;
[0011] Step S4, inputting the data set into the fusion prediction model, and outputting a risk prediction result.
[0012] In a preferred embodiment, in step S1, the basic clinical data of the target object includes age, gender, height, weight, blood pressure, blood glucose, blood lipid related indicators, and history of hypertension, diabetes, coronary heart disease and family history of stroke.
[0013] The dynamic physiological parameter data is obtained by a wearable device for 7 consecutive days of heart rate, 24-hour ambulatory blood pressure and night-time blood oxygen saturation.
[0014] The image feature data covers carotid ultrasound image and brain CT image.
[0015] In a preferred embodiment, the three types of data are quantitatively processed, the classification information in the basic clinical data is converted into a calculable numerical form, the dynamic physiological parameter data is arranged in a time sequence into a continuous data set, and the image feature data is labeled with key structure parameters.
[0016] The proportion of abnormal data is calculated, the number of indicators exceeding the normal range of medicine is compared with the total number of indicators of this type of data, a first ratio value is obtained, and the overall abnormality of the data is reflected.
[0017] The fluctuation of the dynamic physiological parameter is analyzed, the ratio of the mean value and the standard deviation is calculated, a second ratio value is obtained, and the relative strength of the data fluctuation is reflected.
[0018] The second ratio values of all data types are sorted by size, the maximum value is determined as a reference benchmark, and the first ratio value, the second ratio value and the maximum reference benchmark are combined to construct the data set.
[0019] In a preferred embodiment, in step S2, when extracting features from the basic clinical data, the abnormal mean value and the abnormal median value of each indicator are calculated, the information of both is fused by mean value calculation to obtain a comprehensive indicator reflecting the trend of the data set, and the core indicators significantly affecting the risk are screened out.
[0020] For the dynamic physiological parameter data, the mean value, standard deviation and fluctuation amplitude features in the time sequence are extracted, for the image feature data, the texture and structure features are extracted, the mean value and dispersion degree of the key structure parameters are calculated, and they are combined with the abnormal fluctuation features to form the image feature set.
[0021] In a preferred embodiment, the features extracted from the three types of data are fused, the features are sorted by importance, the basic clinical core indicators are taken as a benchmark, the dynamic parameter trend features and the image structure features are associated, and the features are integrated into a unified dimension feature vector through comprehensive operation.
[0022] The mean value and the median value of the fused features are calculated, the feature distribution is optimized by mean value processing, and a fused feature set capable of comprehensively reflecting the risk features of the target object is obtained.
[0023] In a preferred embodiment, in step S3, when building the fusion prediction model, the model base parameters are first initialized and configured, and the fusion feature set is divided into several independent training sample groups;
[0024] The first sample group is imported into the neural network structure, and the hidden node and explicit node state parameters of the initial layer are updated synchronously.
[0025] The hidden node state value is calibrated again according to the real-time state data of the explicit node.
[0026] In a preferred embodiment, all sample groups are sorted in chronological order to form a time sequence, and the long-term correlation feature chain between the sample groups is extracted by analyzing the time sequence, which is used to reflect the long-term correlation and inherent attributes of the cumulative input samples.
[0027] Combined with the long-term correlation feature chain and the preset prediction period, the deviation feature information is generated, and each sample group is used for multiple rounds of iteration optimization, and finally the fusion prediction model is built.
[0028] In a preferred embodiment, in step S4, the data set is imported into the built fusion prediction model, and the fusion prediction model outputs the quantitative evaluation values of each risk dimension after operation;
[0029] The relationship between each evaluation value and the preset risk threshold value is compared; if there is an evaluation value exceeding the limit, a special risk warning report is generated for the exceeding dimension.
[0030] If all evaluation values are within the threshold value range, a comprehensive risk evaluation report is generated by integrating all-dimensional information.
[0031] The technical effects and advantages of the cerebral vascular disease risk prediction method of the present application are as follows:
[0032] 1. Breakthrough the limitations of traditional single data prediction, simultaneously collect basic clinical data, dynamic physiological parameter data and image feature data, cover the key influencing factors of cerebral vascular disease risk from three core dimensions of static health baseline, dynamic physiological change and anatomical structure state, make up for the information short board of single data type, convert different types of data into unified computable form through standardized quantitative processing, combine abnormal data proportion analysis and dynamic fluctuation intensity evaluation, effectively screen out core information with risk indication significance, eliminate redundant interference data, provide high-quality data support for subsequent prediction, and significantly improve the identification ability of early potential risk of diseases.
[0033] 2. Through feature importance ranking and multi-dimensional fusion to form a standardized feature vector, optimize feature distribution and strengthen key risk signals; in the model construction process, through sample grouping training, node parameter secondary calibration and long-term correlation feature mining, the problem of overfitting is effectively avoided, the prediction accuracy of the model to individual differences is improved, and the quantized evaluation value and differentiated report of each risk dimension are output, which can accurately locate the high-risk dimension and generate special early warning, comprehensively present the overall risk situation, and significantly improve the effectiveness of cerebrovascular disease prevention and intervention. BRIEF DESCRIPTION OF DRAWINGS
[0034] Figure 1 FIG. 1 is a flowchart of a cerebrovascular disease risk prediction method according to an embodiment of the present application. DETAILED DESCRIPTION
[0035] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0036] Embodiment 1
[0037] Figure 1 The present application provides a cerebrovascular disease risk prediction method, which specifically comprises the following steps:
[0038] Step S1, collecting the basic clinical data, dynamic physiological parameter data and image feature data of the target object, and constructing a data set;
[0039] Step S2, performing feature extraction on the basic clinical data, dynamic physiological parameter data and image feature data respectively, and fusing the extracted features to obtain a fused feature set;
[0040] Step S3, constructing a fusion prediction model through the fused feature set;
[0041] Step S4, inputting the data set into the fusion prediction model, and outputting a risk prediction result.
[0042] In step S1, the basic clinical data of the target object includes age, gender, height, weight, blood pressure, blood glucose, blood lipid related indicators, and history of hypertension, diabetes, coronary heart disease and family history of stroke;
[0043] The dynamic physiological parameter data is obtained by a wearable device for 7 consecutive days of heart rate, 24-hour ambulatory blood pressure and night-time blood oxygen saturation;
[0044] The image feature data covers brain carotid ultrasound image and brain CT image.
[0045] Quantitative processing is performed on various types of data, and the classification information in the basic clinical data is converted into a calculable numerical form. The dynamic physiological parameter data is arranged in a time sequence to form a continuous data set, and the image feature data is labeled with key structure parameters.
[0046] The proportion of abnormal data is calculated, and the number of indicators that exceed the normal range of medicine is compared with the total number of indicators of this type of data to obtain a first ratio value, which reflects the overall abnormality of the data.
[0047] The fluctuation of the dynamic physiological parameters is analyzed, and the ratio of the mean value to the standard deviation is calculated to obtain a second ratio value, which reflects the relative strength of the data fluctuation.
[0048] The second ratio values of all data types are sorted by size, and the maximum value is determined as a reference benchmark. The first ratio value, the second ratio value, and the maximum reference benchmark are combined to construct the data set.
[0049] The collection of basic clinical data, dynamic physiological parameter data, and image feature data achieves comprehensive coverage of the health status of the target object. The basic clinical data, which includes age, gender, basic signs, and medical history, is the traditional and key basis for judging the risk of cerebrovascular disease and can reflect the individual's basic health baseline. The dynamic physiological parameter data, which is collected through continuous 7-day heart rate, 24-hour dynamic blood pressure, and night-time blood oxygen saturation, captures the dynamic change rule of physiological indicators. These dynamic fluctuations often serve as potential signals of early disease risk, making up for the shortcomings of traditional static data. The image feature data, including carotid ultrasound and CT images of the brain, visually presents the vascular status from the anatomical structure level, such as carotid intima thickness and plaque conditions, providing an objective structural basis for risk assessment.
[0050] The quantitative processing and abnormal analysis of various types of data improve the accuracy of risk identification. The numericalization of classification information, the time sequencing of dynamic data, and the structurization of image data facilitate subsequent feature extraction. The first ratio value, which calculates the proportion of abnormal data, can clearly quantify the overall proportion of abnormal indicators in the data and quickly judge the overall risk level of the data. The second ratio value, which calculates the ratio of the mean value to the standard deviation of dynamic physiological parameters, and determines the maximum reference benchmark, combined with the first ratio value, can identify the fluctuation intensity of single data and effectively filter out the real abnormality with risk indicating significance, improving the early identification ability of cerebrovascular disease risk prediction.
[0051] In step S2, when extracting features from the basic clinical data, the abnormal mean and median of each indicator are calculated. The mean value is calculated to fuse the information of the two, obtaining a comprehensive indicator reflecting the trend of the data set, and screening out the core indicators that significantly affect the risk.
[0052] For dynamic physiological parameter data, the mean, standard deviation and fluctuation amplitude features in the time series are extracted, and for image feature data, the texture and structure features are extracted, and the mean and dispersion degree of key structure parameters are calculated, which are combined with abnormal fluctuation features to form an image feature set.
[0053] The features extracted from the three types of data are fused, sorted according to feature importance, and based on the basic clinical core indicators, the dynamic parameter trend features and image structure features are associated, and integrated into a unified dimensional feature vector through comprehensive operation;
[0054] The mean and median of the fused features are calculated to optimize the feature distribution through mean processing, and a fused feature set that can fully reflect the risk characteristics of the target object is obtained.
[0055] Differential precise extraction methods are used for the three types of data to ensure the full mining of risk information in each dimension. For basic clinical data, comprehensive indicators are obtained through the fusion calculation of abnormal mean and abnormal median, which retains the core information of the trend in the data set, and the interference of extreme abnormal values is weakened through mean fusion, and core indicators such as age and blood pressure that have a significant impact on risk are selected. For dynamic physiological parameter data, focus on the mean, standard deviation and fluctuation amplitude of the time series, and capture the long-term change law of indicators such as heart rate and blood pressure. These dynamic trends are important signals of early risk of cerebrovascular disease, making up for the process risk information that static data cannot reflect. For image feature data, texture, structure features and key parameter mean and dispersion degree are extracted simultaneously, and combined with abnormal fluctuation features to form an image feature set, so that anatomical information such as vascular status can directly participate in risk modeling.
[0056] The features are sorted according to importance and associated with clinical core indicators as a benchmark to ensure the dominance of core risk factors in the fusion process, while organically linking dynamic trends and image structure features, and integrating them into a unified dimensional feature vector through comprehensive operation to provide standardized input for subsequent model training; After fusion, the feature distribution is optimized through mean and median calculation to improve the stability of the feature set, and precise selection and integration strengthen key risk signals, improving the model's ability to capture early risk and distinguish different risk levels.
[0057] In step S3, when building the fused prediction model, first initialize the model base parameters, divide the fused feature set into several independent training sample groups;
[0058] Import the first sample group into the neural network structure, and update the hidden node and explicit node state parameters of the initial layer simultaneously;
[0059] According to the real-time state data of the explicit node, the hidden node state value is calibrated again.
[0060] The sample groups are sorted in chronological order to form a time sequence, and the long-term correlation feature chain between the sample groups is extracted by analyzing the time sequence, and the feature chain is used to reflect the long-term correlation and inherent attributes of the cumulative input samples;
[0061] The deviation feature information is generated by combining the long-term correlation feature chain with the preset prediction period, and each sample group is used for multiple rounds of iterative optimization, and finally a fusion prediction model is constructed.
[0062] The model base parameters are initialized and configured, and the fusion feature set is divided into independent training sample groups, which avoids model overfitting caused by too large training data volume; after the first sample group is imported into the neural network, the initial layer hidden and explicit node state parameters are updated synchronously, and then the implicit node value is calibrated again according to the explicit node state, which can quickly optimize the initial layer parameters.
[0063] The sample groups are sorted in chronological order to form a time sequence, and the long-term correlation feature chain is extracted by analyzing the time sequence, which excavates the time dimension correlation between samples, adapts to the feature attributes of dynamic physiological parameter time series data, and enables the model to identify potential risk patterns in long-term changes of indicators such as heart rate and blood pressure; the deviation feature information is generated by combining the long-term correlation feature chain with the preset prediction period, which injects the model with the targeted adaptation ability of the prediction period, so that the model can adjust the feature weight according to different prediction needs, and improve the accuracy of risk prediction in a specific period. Use each sample group for multiple rounds of training, so that the model can continuously optimize parameters under different sample distributions, effectively reduce the prediction deviation caused by individual data differences, and finally construct a fusion prediction model that can capture long-term dynamic correlation and period adaptation needs.
[0064] In step S4, the data set is imported into the constructed fusion prediction model, and the fusion prediction model outputs the quantitative evaluation value of each risk dimension after operation;
[0065] Compare the relationship between each evaluation value and the preset risk threshold value; if there is an evaluation value exceeding the limit, generate a special risk warning report for the exceeding dimension;
[0066] If all evaluation values are within the threshold value range, integrate the full-dimensional information to generate a comprehensive risk assessment report.
[0067] The fusion prediction model respectively outputs quantitative evaluation values for multiple risk dimensions such as basic clinical, dynamic physiological, and image features. Each value corresponds to the risk intensity of a specific dimension, allowing medical personnel to clearly understand the specific risk level of the target object in terms of blood pressure fluctuations, vascular structure, and basic indicators. By comparing the evaluation values with preset risk threshold values, the high-risk dimensions can be quickly located, and special risk warning reports can be generated for the over-limit dimensions, directly focusing on the core risk points. For example, if the dynamic blood pressure fluctuation dimension evaluation value is over-limit, the report can clearly indicate the risk level and possible pathogenesis of this dimension, helping medical personnel to grasp the intervention points in a short time. When all dimensions meet the standards, the comprehensive risk assessment report generated by integrating all dimensions of information can comprehensively present the overall risk status of the target object, including the synergistic relationship and overall risk level of each dimension risk.
[0068] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
[0069] Finally, the above is only a preferred embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims
1. A method for predicting the risk of cerebrovascular diseases, characterized in that, Includes the following steps: Step S1: Collect basic clinical data, dynamic physiological parameter data, and imaging feature data of the target subjects to construct a dataset; Step S2: Extract features from basic clinical data, dynamic physiological parameter data, and imaging feature data respectively, and fuse the extracted features to obtain a fused feature set; Step S3: Construct a fusion prediction model by fusing feature sets; Step S4: Input the dataset into the fusion prediction model and output the risk prediction results.
2. The method for predicting the risk of cerebrovascular diseases according to claim 1, characterized in that: In step S1, the basic clinical data of the target subjects are collected, including age, gender, height, weight, blood pressure, blood sugar, blood lipid-related indicators, as well as history of hypertension, diabetes, coronary heart disease and family history of stroke. Dynamic physiological parameter data were obtained through wearable devices, including heart rate, 24-hour ambulatory blood pressure, and nocturnal blood oxygen saturation for 7 consecutive days. The imaging feature data includes ultrasound images of the carotid arteries in the brain and CT images of the brain.
3. The method for predicting the risk of cerebrovascular diseases according to claim 2, characterized in that: Quantitative processing is performed on various types of data, converting the categorical information in basic clinical data into calculable numerical forms, organizing dynamic physiological parameter data into continuous datasets according to time series, and annotating key structural parameters in image feature data. The proportion of abnormal data is calculated by dividing the number of indicators that exceed the normal medical range by the total number of indicators in this category. This first ratio is used to reflect the overall degree of abnormality in the data. The fluctuations of dynamic physiological parameters are analyzed, and the ratio of their mean to standard deviation is calculated to obtain the second type ratio, which is used to reflect the relative intensity of data fluctuations. Sort the second analogy values of all data types by size, determine the maximum value as the reference benchmark, and construct the dataset by combining the first analogy value, the second analogy value, and the maximum reference benchmark.
4. The method for predicting the risk of cerebrovascular diseases according to claim 3, characterized in that: In step S2, when extracting features from basic clinical data, the abnormal mean and abnormal median of each indicator are first calculated. The information from both is then fused through mean calculation to obtain a comprehensive indicator that reflects the central trend of the data, and core indicators that have a significant impact on risk are selected. For dynamic physiological parameter data, the mean, standard deviation and fluctuation amplitude features are extracted from the time series. For image feature data, texture and structural features are extracted, the mean and dispersion of key structural parameters are calculated, and these are combined with abnormal fluctuation features to form an image feature set.
5. The method for predicting the risk of cerebrovascular diseases according to claim 4, characterized in that: The features extracted from the three types of data are fused together. First, the features are sorted by importance. Based on the basic clinical core indicators, dynamic parameter trend features and imaging structural features are associated and integrated into a feature vector of a unified dimension through comprehensive calculation. The mean and median of the fused features are calculated, and the feature distribution is optimized by mean processing to obtain a fused feature set that can comprehensively reflect the risk characteristics of the target object.
6. The method for predicting the risk of cerebrovascular diseases according to claim 5, characterized in that: In step S3, when constructing the fusion prediction model, the basic parameters of the model are first initialized and configured, and the fusion feature set is divided into several independent training sample groups. The first sample group is imported into the neural network structure, and the state parameters of the hidden and explicit nodes in the initial layer are updated synchronously. Based on the real-time status data of the explicit nodes, the status values of the implicit nodes are calibrated a second time.
7. The method for predicting the risk of cerebrovascular diseases according to claim 6, characterized in that: All sample groups are sorted in chronological order to form a time series. The time series is then parsed to extract the long-term correlation feature chain between sample groups. This feature chain is used to reflect the long-term correlation and intrinsic attributes of the cumulative input samples. By combining long-term correlation feature chains with preset prediction time periods, deviation feature information is generated. Multiple rounds of iterative optimization are then performed using each sample group to finally construct a fusion prediction model.
8. The method for predicting the risk of cerebrovascular diseases according to claim 7, characterized in that: In step S4, the dataset is imported into the constructed fusion prediction model, and the fusion prediction model outputs the quantitative assessment values of each risk dimension after calculation. Compare the relationship between each assessment value and the preset risk threshold; if any assessment value exceeds the limit, generate a special risk warning report for the dimension that exceeds the limit. If all assessment values are within the critical range, then the comprehensive risk assessment report is generated by integrating all dimensions of information.