Intelligent cardiovascular system disease prediction system based on pathophysiology
By performing classification feature extraction and weighting calculation of multiple influencing factors in the cardiovascular disease prediction system, and combining the mean-optimized symmetric point pattern graph convolutional neural network module for feature extraction, the lack of comprehensiveness and comprehensiveness of the prediction of central vascular disease in the existing technology is solved. The problem of low accuracy achieves more efficient and accurate cardiovascular disease prediction.
Patent Information
- Application Number
- CN202411934872.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-02-14
AI Technical Summary
Existing artificial intelligence technologies usually rely on single variables or simple electronic medical record-related characteristics in cardiovascular disease prediction, which lacks comprehensiveness, resulting in low accuracy of prediction results.
Intelligent cardiovascular system disease prediction system based on pathophysiology, by extracting the classification features of multiple influencing factors, weighted calculation and genetic optimization for all influencing factors, combined with the mean optimization of the symmetric point pattern graph convolutional neural network module, Blood pressure timing chart, blood cholesterol timing chart and ECG signal timing chart are time-frequency conversion and feature extraction to generate representative features for prediction.
It improves the accuracy and efficiency of cardiovascular disease prediction, provides more comprehensive and multi-angle time series data characterization, and enhances the reliability of predicted results.
Smart Images
Figure CN119441957A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of disease prediction, and in particular to an intelligent cardiovascular system disease prediction system based on pathophysiology. Background Art
[0002] Cardiovascular disease is one of the most common fatal diseases worldwide. Early detection and intervention of cardiovascular disease risk factors can reduce morbidity and mortality. In recent years, with the development of data science and artificial intelligence, the use of big data and machine learning technologies to predict cardiovascular disease has become more feasible. These technologies can help healthcare institutions more accurately assess patient risks; however, existing artificial intelligence technologies usually predict cardiovascular disease through a single variable or simple electronic medical record-related features. Predicting cardiovascular disease through a simple combination of influencing factors lacks comprehensiveness and has a low accuracy rate. Summary of the invention
[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention creatively proposes an intelligent cardiovascular disease prediction system based on pathophysiology. In view of the lack of comprehensiveness in predicting cardiovascular diseases through a simple combination of influencing factors and the low accuracy of the prediction results, the present invention creatively extracts features of multiple influencing factors of pathophysiology, and performs weighted calculation and genetic optimization for all influencing factors to obtain a discriminative and comprehensive feature extraction scheme, so that the prediction of cardiovascular diseases for patients is more accurate and more efficient; the present invention creatively adopts a mean-optimized symmetrical point pattern graph convolutional neural network module to perform time-frequency conversion on the blood pressure time series graph, blood cholesterol time series graph and electrocardiogram signal time series graph among the disease influencing factors, so as to obtain a more comprehensive and multi-angle time series data representation, and at the same time, the time-frequency graph is matrix-converted, the time-frequency graph is optimized by matrix pixel similarity, and a deep convolutional neural network is used to extract features from the optimized time-frequency graph to obtain representative features, and the representative features are used to participate in the prediction of cardiovascular diseases in patients, thereby further increasing the accuracy of prediction.
[0004] The intelligent cardiovascular disease prediction system based on pathophysiology provided by the present invention includes a pathophysiological data acquisition module, a feature extraction module, an integrated prediction module and a real-time feedback module;
[0005] The pathophysiological data acquisition module collects numerical data, classification data and daily data of the pathophysiology of all patients tested for cardiovascular diseases and sends them to the feature extraction module;
[0006] The numerical data includes age, height, weight, and blood pressure timing diagram, blood cholesterol timing diagram and electrocardiogram timing diagram of the test patient collected at different time periods in group p, and the blood pressure timing diagram, blood cholesterol timing diagram and electrocardiogram timing diagram at each moment are a group;
[0007] The categorized data include gender, medical history and family genetic disease history;
[0008] The daily data includes health status, smoking status, drinking status and diet status;
[0009] The feature extraction module includes a numerical data processing branch, a classification data processing branch and a daily data processing branch;
[0010] The numerical data processing branch includes a one-dimensional convolutional neural network module, an optimized symmetric point pattern graph convolutional neural network module and a feature splicing module;
[0011] The one-dimensional convolutional neural network module uses a one-dimensional convolutional neural network to extract features of age, height, and weight to obtain one-dimensional features;
[0012] The optimized symmetric point pattern graph convolutional neural network module extracts features from the blood pressure timing graph, the blood cholesterol timing graph and the electrocardiogram timing graph to obtain two-dimensional features;
[0013] The feature concatenation module concatenates the one-dimensional feature with the two-dimensional feature to obtain a numerical data feature;
[0014] The classification data processing branch uses three convolutional layers to extract features from the classification data to obtain classification features;
[0015] The three convolution layers each include a convolution kernel, a normalization layer, an activation function layer, and a maximum pooling layer, wherein the three convolution layers include 32, 64, and 128 3×3 convolution kernels, respectively;
[0016] The daily data processing branch includes an embedding layer, a concatenation layer, and a convolution layer;
[0017] The embedding layer performs feature extraction and low-dimensional mapping on all daily data to obtain key features of all daily data;
[0018] The splicing layer splices all key features row by row to obtain comprehensive features;
[0019] The convolution layer uses a one-dimensional convolutional neural network to perform local feature extraction, nonlinear activation, standardization and maximum pooling on the comprehensive features in sequence to obtain daily data features;
[0020] The integrated prediction module performs weighted classification prediction output on numerical data features, classification features and daily data features to obtain a comprehensive prediction result, and optimizes the weights using a genetic algorithm to obtain the optimal classification weights;
[0021] The real-time feedback module collects pathophysiological numerical data, classification data and daily data of the patient to be predicted, and uses the pathophysiological data collection module, feature extraction module, integrated prediction module and optimal classification weights to predict cardiovascular system diseases in the patient to be predicted, and obtains the final prediction result.
[0022] Furthermore, the construction of the mean-optimized symmetric point pattern graph convolutional neural network specifically includes the following steps:
[0023] Step S1: constructing a time-frequency diagram, using a symmetric polar coordinate SDP algorithm to convert the blood pressure time series diagram, blood cholesterol time series diagram and electrocardiogram signal time series diagram into a snowflake-style time-frequency diagram;
[0024] Step S2: Digital matrix construction, constructing a digital matrix based on the time-frequency diagram:
[0025] ;
[0026] In the formula, Represents the coordinates of the pixel points corresponding to the time-frequency graph, The pixel value representing the coordinates of the pixel point. Represents the total number of pixels in the time-frequency graph, Represents a digital matrix;
[0027] Step S3: Matrix similarity calculation, similarity calculation is performed on the time-frequency graphs of p different groups:
[0028] R(N,M)= ∑ m ∑ n ( N mn - N a )( M mn - M a ) [ ∑ m ∑ n ( N mn - N a ) 2 ][ ∑ m ∑ n ( M mn - M a ) 2 ] ;
[0029] In the formula, Representative A digital matrix of time-frequency plots, Representative The average value of the pixels in the time-frequency graph is Representative The horizontal axis in the time-frequency diagram is And the vertical axis is The pixel of the pixel, Representative The time-frequency diagram and The similarity of the time-frequency diagrams;
[0030] Step S4: Average similarity calculation:
[0031] ;
[0032] In the formula, Representative The digital matrix of the time-frequency diagram is The similarity of the digital matrix of the time-frequency diagram, Represents the total number of digital matrices, represents the average similarity;
[0033] Step S5: In-group optimization, set the threshold with the average similarity, remove the time-frequency graphs of different groups whose similarities are outside the threshold, and obtain d groups of optimized time-frequency graphs
[0034] Step S6: feature extraction, using a deep convolutional neural network, extracting features from d different groups of time-frequency graphs to obtain d groups of features, and performing weighted summation on the d groups of features to obtain two-dimensional features.
[0035] The beneficial results achieved by the present invention using the above scheme are as follows:
[0036] (1) The present invention creatively extracts features from multiple influencing factors of pathophysiology by classification, and performs weighted calculation and genetic optimization for all influencing factors to obtain a discriminative and comprehensive feature extraction scheme, thereby making the prediction of cardiovascular disease in patients more accurate and efficient;
[0037] (2) The present invention creatively adopts a mean-optimized symmetrical point pattern graph convolutional neural network module to perform time-frequency conversion on the blood pressure time series graph, blood cholesterol time series graph and electrocardiogram signal time series graph among the disease influencing factors, so as to obtain a more comprehensive and multi-angle time series data representation. At the same time, the time-frequency graph is transformed into a matrix, and the time-frequency graph is optimized by matrix pixel similarity. A deep convolutional neural network is used to extract features from the optimized time-frequency graph to obtain representative features, and the representative features are used to participate in the prediction of cardiovascular diseases in patients, thereby further increasing the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 A module diagram of the intelligent cardiovascular disease prediction system based on pathophysiology provided by the present invention;
[0039] Figure 2 Schematic diagram of the process of a mean-optimized symmetric point pattern graph convolutional neural network.
[0040] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION
[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0042] Example 1, see Figure 1 , the intelligent cardiovascular disease prediction system based on pathophysiology provided by the present invention includes a pathophysiological data acquisition module, a feature extraction module, an integrated prediction module and a real-time feedback module;
[0043] The pathophysiological data acquisition module collects numerical data, classification data and daily data of the pathophysiology of all patients tested for cardiovascular diseases and sends them to the feature extraction module;
[0044] The feature extraction module includes a numerical data processing branch, a classification data processing branch and a daily data processing branch, which respectively extract numerical data features, classification features and daily data features, and send them to the integrated prediction module;
[0045] The integrated prediction module performs weighted classification prediction output on numerical data features, classification features and daily data features to obtain a comprehensive prediction result, and optimizes the weights using a genetic algorithm to obtain the optimal classification weights;
[0046] The real-time feedback module collects pathophysiological numerical data, classification data and daily data of the patient to be predicted, and uses the pathophysiological data collection module, feature extraction module, integrated prediction module and optimal classification weights to predict cardiovascular system diseases in the patient to be predicted, and obtains the final prediction result.
[0047] By executing the above operations, in order to address the problem that the prediction of cardiovascular diseases through a simple combination of influencing factors lacks comprehensiveness and the accuracy of the prediction results is low, the present invention creatively extracts classified features of multiple influencing factors of pathophysiology, and performs weighted calculation and genetic optimization for all influencing factors to obtain a discriminative and comprehensive feature extraction scheme, thereby making the prediction of cardiovascular diseases for patients more accurate and efficient.
[0048] Embodiment 2: This embodiment is based on the above embodiment, in the pathophysiological data acquisition module:
[0049] The numerical data includes age, height, weight, and blood pressure timing diagram, blood cholesterol timing diagram and electrocardiogram timing diagram of the test patient collected at different time periods in group p, and the blood pressure timing diagram, blood cholesterol timing diagram and electrocardiogram timing diagram at each moment are a group;
[0050] The categorized data include gender, medical history and family genetic disease history;
[0051] The daily data includes health status, smoking status, drinking status and diet status.
[0052] Embodiment 3: This embodiment is based on the above embodiment, in the feature extraction module:
[0053] The numerical data processing branch includes a one-dimensional convolutional neural network module, an optimized symmetric point pattern graph convolutional neural network module and a feature splicing module;
[0054] The one-dimensional convolutional neural network module uses a one-dimensional convolutional neural network to extract features of age, height, and weight to obtain one-dimensional features;
[0055] The optimized symmetric point pattern graph convolutional neural network module extracts features from the blood pressure timing graph, the blood cholesterol timing graph and the electrocardiogram timing graph to obtain two-dimensional features;
[0056] The feature concatenation module concatenates the one-dimensional feature with the two-dimensional feature to obtain a numerical data feature.
[0057] Embodiment 4: This embodiment is based on the above embodiment, in the feature extraction module:
[0058] The classification data processing branch uses three convolutional layers to extract features from the classification data to obtain classification features;
[0059] The three convolution layers each include a convolution kernel, a normalization layer, an activation function layer, and a maximum pooling layer, wherein the three convolution layers include 32, 64, and 128 3×3 convolution kernels, respectively.
[0060] Embodiment 5, this embodiment is based on the above embodiment, in the feature extraction module:
[0061] The daily data processing branch includes an embedding layer, a concatenation layer, and a convolution layer;
[0062] The embedding layer performs feature extraction and low-dimensional mapping on all daily data to obtain key features of all daily data;
[0063] The splicing layer splices all key features row by row to obtain comprehensive features;
[0064] The convolution layer uses a one-dimensional convolutional neural network to perform local feature extraction, nonlinear activation, standardization and maximum pooling on the comprehensive features in sequence to obtain daily data features.
[0065] Embodiment 6: This embodiment is based on the above embodiment, in the feature extraction module:
[0066] A mean-optimized symmetric point pattern graph convolutional neural network is constructed, which specifically includes the following steps:
[0067] Step S1: constructing a time-frequency diagram, using a symmetric polar coordinate SDP algorithm to convert the blood pressure time series diagram, blood cholesterol time series diagram and electrocardiogram signal time series diagram into a snowflake-style time-frequency diagram;
[0068] Step S2: Digital matrix construction, constructing a digital matrix based on the time-frequency diagram:
[0069] ;
[0070] In the formula, Represents the coordinates of the pixel points corresponding to the time-frequency graph, The pixel value representing the coordinates of the pixel point. Represents the total number of pixels in the time-frequency graph, Represents a digital matrix;
[0071] Step S3: Matrix similarity calculation, similarity calculation is performed on the time-frequency graphs of p different groups:
[0072] R(N,M)= ∑ m ∑ n ( N mn - N a )( M mn - M a ) [ ∑ m ∑ n ( N mn - N a ) 2 ][ ∑ m ∑ n ( M mn - M a ) 2 ] ;
[0073] In the formula, Representative A digital matrix of time-frequency plots, Representative The average value of the pixels in the time-frequency graph is Representative The horizontal axis in the time-frequency diagram is And the vertical axis is The pixel of the pixel, Representative The time-frequency diagram and The similarity of the time-frequency diagrams;
[0074] Step S4: Average similarity calculation:
[0075] ;
[0076] In the formula, Representative The digital matrix of the time-frequency diagram is The similarity of the digital matrix of the time-frequency diagram, Represents the total number of digital matrices, represents the average similarity;
[0077] Step S5: In-group optimization, set the threshold with the average similarity, remove the time-frequency graphs of different groups whose similarities are outside the threshold, and obtain d groups of optimized time-frequency graphs
[0078] Step S6: feature extraction, using a deep convolutional neural network, extracting features from d different groups of time-frequency graphs to obtain d groups of features, and performing weighted summation on the d groups of features to obtain two-dimensional features.
[0079] By performing the above operations, the present invention creatively adopts a mean-optimized symmetrical point pattern graph convolutional neural network module to perform time-frequency conversion on the blood pressure timing graph, blood cholesterol timing graph and electrocardiogram signal timing graph among the disease influencing factors, so as to obtain a more comprehensive and multi-angle time series data representation. At the same time, the time-frequency graph is matrix-converted, the time-frequency graph is optimized by matrix pixel similarity, and a deep convolutional neural network is used to extract features from the optimized time-frequency graph to obtain representative features, and the representative features are used to participate in the prediction of cardiovascular diseases in patients, thereby further increasing the prediction accuracy.
[0080] Embodiment 7: This embodiment is based on the above embodiment. In step S1, the symmetric polar coordinate SDP algorithm specifically includes the following steps:
[0081] Step S11: Input data preparation:
[0082] Given one-dimensional time series data, namely, blood pressure time series, blood cholesterol time series and ECG signal time series, the time series are denoised and normalized:
[0083] Remove the global trend (such as mean or linear trend) from the time series.
[0084] Normalization, mapping the time series to the interval [0, 1];
[0085] Step S12: Calculate polar coordinate parameters:
[0086] Map each time series point to a point on the polar coordinate plane. The two parameters of the polar coordinates are:
[0087] Angle: determines the direction of a point in the polar coordinate plane;
[0088] Radius: determines the distance from the point to the origin;
[0089] The specific calculation method is as follows:
[0090] Calculate the angle: map the index of the time series to an angle and evenly distribute the time series in the polar coordinate plane;
[0091] Calculate the radius: Map the amplitude in the time series value to an angle
[0092] Step S13: Mapping to polar coordinate plane:
[0093] Convert the time series points to Cartesian coordinates using the polar coordinate formula;
[0094] Step S14: Add mirror symmetry points:
[0095] Add a point symmetrical about the origin to each point to generate an image with mirror symmetry (such as a snowflake pattern). Therefore, each time series point is mapped to a positive point and a symmetrical point.
[0096] Step S15: Generate a two-dimensional image:
[0097] Plot all points in the form of a scatter plot on a polar coordinate plane;
[0098] By performing the above operations, the generated 2D mirror-symmetric images have unique geometric patterns, and their patterns are closely related to the characteristics of the time series, such as:
[0099] Periodic time series produce regular, symmetrical patterns;
[0100] A time series with higher randomness will produce more dispersed patterns;
[0101] A time series with a certain frequency may produce a snowflake pattern with a certain shape.
[0102] Embodiment 8, this embodiment is based on the above embodiment, in step S6, the deep convolutional neural network specifically includes the following processing steps:
[0103] Input image: Input time-frequency graph;
[0104] Convolutional Layer:
[0105] Extract image features through convolution operations. Each convolution layer contains multiple filters, each of which slides on the input image and performs convolution operations to generate feature maps;
[0106] Activation function:
[0107] After the convolutional layer, an activation function is applied to introduce nonlinearity and increase the expressive power of the network;
[0108] Pooling layer:
[0109] Downsample the feature map to reduce its size while retaining the main features;
[0110] Repeated convolution and pooling:
[0111] Repeatedly stack multiple convolutional layers and pooling layers to obtain higher-level features;
[0112] Fully connected layer:
[0113] After passing through multiple convolutional and pooling layers, the extracted features are flattened and mapped to output categories through a fully connected layer;
[0114] Output layer:
[0115] Use the softmax function to convert the output of the network into a probability distribution corresponding to each category;
[0116] Training and Optimization:
[0117] Training is performed through genetic algorithms to optimize network parameters to minimize the loss function;
[0118] Feature extraction:
[0119] After training is completed, the output of the middle layer of the network is used as a feature extractor.
[0120] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0121] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
[0122] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.
Claims
1. An intelligent cardiovascular disease prediction system based on pathophysiology, characterized by: It includes pathophysiological data acquisition module, feature extraction module, integrated prediction module and real-time feedback module; The pathophysiological data acquisition module collects numerical data, classification data and daily data of the pathophysiology of all patients tested for cardiovascular diseases and sends them to the feature extraction module; The feature extraction module includes a numerical data processing branch, a classification data processing branch and a daily data processing branch, which respectively extract numerical data features, classification features and daily data features, and send them to the integrated prediction module; The integrated prediction module performs weighted classification prediction output on numerical data features, classification features and daily data features to obtain a comprehensive prediction result, and optimizes the weights using a genetic algorithm to obtain the optimal classification weights; The real-time feedback module collects pathophysiological numerical data, classification data and daily data of the patient to be predicted, and uses the pathophysiological data collection module, feature extraction module, integrated prediction module and optimal classification weights to predict cardiovascular system diseases in the patient to be predicted, and obtains the final prediction result.
2. The intelligent cardiovascular disease prediction system based on pathophysiology according to claim 1, characterized in that: In the pathophysiological data acquisition module: The numerical data include age, height, weight, and blood pressure timing diagram, blood cholesterol timing diagram and electrocardiogram timing diagram of the test patients collected in different time periods of group p. The blood pressure timing diagram, blood cholesterol timing diagram and electrocardiogram timing diagram at each moment are a group.
3. The intelligent cardiovascular disease prediction system based on pathophysiology according to claim 1, characterized in that: In the feature extraction module: The numerical data processing branch includes a one-dimensional convolutional neural network module, an optimized symmetric point pattern graph convolutional neural network module and a feature splicing module; The one-dimensional convolutional neural network module uses a one-dimensional convolutional neural network to extract features of age, height, and weight to obtain one-dimensional features; The optimized symmetric point pattern graph convolutional neural network module extracts features from the blood pressure timing graph, the blood cholesterol timing graph and the electrocardiogram timing graph to obtain two-dimensional features; The feature concatenation module concatenates the one-dimensional feature with the two-dimensional feature to obtain a numerical data feature.
4. The intelligent cardiovascular disease prediction system based on pathophysiology according to claim 1, characterized in that: In the feature extraction module: The classification data processing branch uses three convolutional layers to extract features from the classification data to obtain classification features; The three convolution layers each include a convolution kernel, a normalization layer, an activation function layer, and a maximum pooling layer, wherein the three convolution layers include 32, 64, and 128 3×3 convolution kernels, respectively.
5. The intelligent cardiovascular disease prediction system based on pathophysiology according to claim 1, characterized in that: In the feature extraction module: The daily data processing branch includes an embedding layer, a concatenation layer, and a convolution layer; The embedding layer performs feature extraction and low-dimensional mapping on all daily data to obtain key features of all daily data; The splicing layer splices all key features row by row to obtain comprehensive features; The convolution layer uses a one-dimensional convolutional neural network to perform local feature extraction, nonlinear activation, standardization and maximum pooling on the comprehensive features in sequence to obtain daily data features.
6. The intelligent cardiovascular disease prediction system based on pathophysiology according to claim 3 is characterized by: In the feature extraction module: The construction of a mean-optimized symmetric point pattern graph convolutional neural network module specifically includes the following steps: Step S1: constructing a time-frequency diagram, using a symmetric polar coordinate SDP algorithm to convert the blood pressure time series diagram, blood cholesterol time series diagram and electrocardiogram signal time series diagram into a snowflake-style time-frequency diagram; Step S2: Digital matrix construction, constructing a digital matrix based on the time-frequency diagram: ； In the formula, Represents the coordinates of the pixel points corresponding to the time-frequency graph, The pixel value representing the coordinates of the pixel point. Represents the total number of pixels in the time-frequency graph, Represents a digital matrix; Step S3: Matrix similarity calculation, similarity calculation is performed on the time-frequency graphs of p different groups: ； In the formula, Representative A digital matrix of time-frequency plots, Representative The average value of the pixels in the time-frequency graph is Representative The horizontal axis in the time-frequency diagram is And the vertical coordinate is The pixel of the pixel, Representative The time-frequency diagram and The similarity of the time-frequency diagrams; Step S4: Average similarity calculation: ； In the formula, Representative The digital matrix of the time-frequency diagram is The similarity of the digital matrix of the time-frequency diagram, Represents the total number of digital matrices, represents the average similarity; Step S5: In-group optimization, set the threshold with the average similarity, remove the time-frequency graphs of different groups whose similarity is outside the threshold, and obtain d groups of optimized time-frequency graphs Step S6: feature extraction, using a deep convolutional neural network, extracting features from d different groups of time-frequency graphs to obtain d groups of features, performing weighted summation on the d groups of features to obtain two-dimensional features.