An intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events
By building a feature library and using the YOLOv4 neural network model for identification and analysis, a personalized health management solution is provided, which solves the problem that the existing technology cannot effectively manage people with acute cardiovascular and cerebrovascular events, and improves users' confidence and initiative in health management.
Patent Information
- Application Number
- CN202210249953.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-14
- Publication Date
- 2025-05-13
- Estimated Expiration
- 2042-03-14
AI Technical Summary
The existing health management methods cannot provide personalized and effective health management solutions for people at risk of acute cardiovascular and cerebrovascular events, resulting in insufficient confidence and initiative in users' self-health management, and reduce the quality effect on acute cardiovascular events.
By constructing a feature library, including a demographic feature library, a key health monitoring feature library and a dynamic behavior feature library, the Pareto law is used to filter key features, form a feature matrix and a matrix library, convert it into grayscale images, and use the YOLOv4 neural network model for identification and analysis, and output the recommended health management solution.
It realizes intelligent health management based on user personalized characteristics, improves users' self-confidence and initiative in self-health management, and enhances the prevention and management effect of acute cardiovascular and cerebrovascular events.
Smart Images
Figure CN114678125B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of acute cardiovascular health management, and in particular to an intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events. Background Art
[0002] Acute cardiovascular and cerebrovascular diseases refer to a group of diseases with sudden onset of cerebral vascular circulation disorders. They can be sudden thrombosis of cerebral blood vessels, cerebral embolism leading to ischemic cerebral infarction, or rupture of cerebral blood vessels leading to cerebral hemorrhage. They are often accompanied by neurological symptoms, hemiplegia of limbs, aphasia, mental symptoms, dizziness, ataxia, choking and coughing. In severe cases, they can lead to coma and death. Clinically, they are also called cerebrovascular accidents, strokes or infarctions.
[0003] Epidemiological surveys have shown that acute cerebrovascular disease, heart disease, and tumors are the three leading causes of death in humans, with extremely high case fatality rates and lethality rates. In many areas of my country, the prevalence of acute and vascular diseases ranks first.
[0004] Since acute cardiovascular and cerebrovascular diseases occur suddenly, develop quickly, and cause great harm to the human body, the best health management methods need to be provided to people at risk of acute cardiovascular events so that they can achieve the best treatment effects. However, current health management methods are too complex and cannot provide users with better health management methods, which reduces the confidence and initiative of users in self-health management, thereby reducing the quality effect of acute cardiovascular events. Summary of the Invention
[0005] The purpose of the present invention is to provide an intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events, so as to solve the problems raised in the above background technology.
[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events, comprising the following steps:
[0007] Step S1, constructing a feature library;
[0008] Step S2, constructing the feature library in step S1 into a feature matrix;
[0009] Step S3, merging the feature matrices in step S2 into a new matrix library;
[0010] Step S4, constructing the matrix library in step S3 into a grayscale image;
[0011] Step S5, constructing the grayscale image in step S4 into a neural network model;
[0012] Step S6, using the neural network model in step S5 to determine the user's current status and output a recommended health management plan.
[0013] Preferably, in step S1, the feature library includes a demographic feature library, a key health monitoring feature library and a dynamic behavior feature library, and the collection and generation of the demographic feature library, the key health monitoring feature library and the dynamic behavior feature library all adopt the Pareto principle (80 / 20 principle).
[0014] Preferably, in step S2, the feature matrix includes a demographic feature matrix, a key health monitoring feature matrix and a dynamic behavior feature matrix.
[0015] Preferably, the demographic feature matrix is constructed by converting the demographic feature library into a decimal number in the interval [0, 255] and finally constructing it into an a*n matrix; the key health monitoring feature matrix is constructed by converting the key health monitoring feature library into a decimal number in the interval [0, 255] and finally constructing it into a b*n matrix; the dynamic behavior feature matrix is constructed by converting the dynamic behavior feature library into a decimal number in the interval [0, 255] and finally constructing it into a c*n matrix.
[0016] Preferably, in step S3, the matrix library is an n*n matrix fused by a demographic feature matrix, a key health monitoring feature matrix and a dynamic behavior feature matrix.
[0017] Preferably, in step S4, the grayscale image is constructed by converting the numbers in the matrix library into decimal grayscale values.
[0018] Preferably, in step S5, the neural network model is obtained by performing recognition analysis on the grayscale image using a neural network algorithm, and the neural network algorithm adopts YOLOv4.
[0019] Preferably, the YOLOv4 analysis process includes the following steps:
[0020] Input: The grayscale image is used as the input analysis picture;
[0021] Backbone: CSPDarkNet53 is used to extract the preliminary features of the grayscale image and build its backbone structure (extracting the grayscale feature matrix);
[0022] Neck: Processes and enhances the grayscale image features extracted by the Backbone layer, so that the features learned by the model are the desired features;
[0023] Head: Based on the enhanced grayscale feature matrix, a deep learning algorithm is used to determine the user's current status and output recommendation results, providing the most urgently needed scenario-based guidance methods for people at risk of acute cardiovascular and cerebrovascular events.
[0024] Preferably, the method for processing and enhancing the grayscale image features extracted by the Backbone layer comprises the following steps:
[0025] Geometric enhancement: Methods of changing the image geometry such as translation, rotation, and shearing can enhance the generalization ability of the model.
[0026] Color enhancement: mainly brightness transformation, such as using HSV (HueSaturationValue) enhancement.
[0027] Blurring: Enhance the model's generalization ability for blurred images through methods such as Gaussian filtering, box filtering, and median filtering.
[0028] Mixup: An algorithm for mixed-class enhancement of images in computer vision. It can mix images of different classes to expand the training data set.
[0029] Technical effects and advantages of the present invention:
[0030] The present invention can create user portraits based on three major categories of features: demographic characteristics, health monitoring characteristics, and dynamic behavioral characteristics, generate user pictures, and then help users recommend the most urgently needed methods through picture recognition, guide users to establish the most urgent lifestyle, make breakthroughs one by one based on priority, and improve users' confidence and initiative in self-health management. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 Schematic diagram of the steps of the method of the present invention
[0032] Figure 2 Schematic diagram of feature matrix fusion of the present invention.
[0033] Figure 3 Schematic diagram of constructing grayscale images for the matrix library of the present invention.
[0034] Figure 4 This is a schematic diagram of the CSPDarkNet53 network structure of the present invention.
[0035] Figure 5 Schematic diagram of the analysis process of the neural network model based on YOLOv4 in the present invention. DETAILED DESCRIPTION
[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0037] The present invention provides Figure 1-5 The intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events shown includes the following steps:
[0038] Step S1, constructing a feature library;
[0039] Step S2, constructing the feature library in step S1 into a feature matrix;
[0040] Step S3, merging the feature matrices in step S2 into a new matrix library;
[0041] Step S4, constructing the matrix library in step S3 into a grayscale image;
[0042] Step S5, constructing the grayscale image in step S4 into a neural network model;
[0043] Step S6, using the neural network model in step S5 to determine the user's current status and output a recommended health management plan.
[0044] In step S1, the feature library includes a demographic feature library, a key health monitoring feature library, and a dynamic behavior feature library. The collection and generation of the demographic feature library, the key health monitoring feature library, and the dynamic behavior feature library all adopt the Pareto principle (the 80 / 20 principle). The Pareto principle is often called the 80 / 20 principle, which means that 80% of the problems are caused by 20% of the reasons. Therefore, it is not necessary to analyze all health monitoring features. On the one hand, too many health monitoring features will evolve into interference factors, thereby affecting the recognition accuracy. On the other hand, the analysis of all health monitoring features will also cause the expansion of the feature matrix, thereby slowing down the recognition speed of acute cardiovascular and cerebrovascular events.
[0045] In the demographic feature database, based on the 80 / 20 principle, 20% of demographic features that are more helpful in identifying acute cardiovascular and cerebrovascular events were screened from the entire population, including but not limited to: gender, age, region of residence, waist circumference, whether taking antihypertensive medication, history of diabetes, smoking, family history of cardiovascular and cerebrovascular diseases, etc.; in the key health monitoring feature database, based on the 80 / 20 principle, 20% of demographic features that are more helpful in identifying acute cardiovascular and cerebrovascular events were screened from the entire population, including but not limited to: total cholesterol, high-density lipoprotein, low-density lipoprotein, fasting blood glucose, glycated hemoglobin, systolic blood pressure, diastolic blood pressure, etc.;
[0046] In the dynamic behavioral feature library, based on the 80 / 20 principle, 20% of demographic characteristics that are more helpful in identifying people with acute cardiovascular and cerebrovascular events are screened out from the full amount of dynamic behaviors, including but not limited to: exercise duration, number of exercises, protein intake, fat intake, sugar intake, salt intake, vegetable intake, water intake, smoking amount, alcohol intake, etc.
[0047] In step S2, the feature matrix includes a demographic feature matrix, a key health monitoring feature matrix, and a dynamic behavior feature matrix. The demographic feature matrix is constructed by converting the demographic feature library into a decimal number in the interval [0, 255] and finally constructing it into an a*n matrix. The key health monitoring feature matrix is constructed by converting the key health monitoring feature library into a decimal number in the interval [0, 255] and finally constructing it into a b*n matrix. The dynamic behavior feature matrix is constructed by converting the dynamic behavior feature library into a decimal number in the interval [0, 255] and finally constructing it into a c*n matrix.
[0048] In step S3, the matrix library is an n*n matrix fused by the demographic feature matrix, the key health monitoring feature matrix and the dynamic behavior feature matrix;
[0049] In step S4, the grayscale image is constructed by converting the numbers in the matrix library into decimal grayscale values. This transforms the dynamic behavior into an image analysis problem for the first time, reducing noise in the analysis process, shrinking the image size, and improving recognition efficiency.
[0050] In step S5, the neural network model is generated by analyzing the grayscale image using a neural network algorithm. YOLOv4, the currently best-performing object image detector, was used. Testing has shown that YOLOv4 achieves comparable performance with EfficientDet while being twice as fast. Furthermore, compared to YOLOv3, the new version improves AP and FPS by 10% and 12%, respectively. This algorithm enables the construction of an efficient and powerful model for identifying individuals experiencing acute cardiovascular and cerebrovascular events.
[0051] The YOLOv4 analysis process includes the following steps:
[0052] Input: The grayscale image is used as the input analysis picture;
[0053] Backbone: CSPDarkNet53 is used to extract the preliminary features of the grayscale image and build its backbone structure (extracting the grayscale feature matrix);
[0054] Neck: Processes and enhances the grayscale image features extracted by the Backbone layer, so that the features learned by the model are the desired features;
[0055] Head: Based on the enhanced grayscale feature matrix, a deep learning algorithm is used to determine the user's current status and output recommendation results, providing the most urgently needed scenario-based guidance methods for people at risk of acute cardiovascular and cerebrovascular events.
[0056] The method for processing and enhancing grayscale image features extracted by the Backbone layer includes the following steps:
[0057] Geometric enhancement: Methods of changing the image geometry such as translation, rotation, and shearing can enhance the generalization ability of the model.
[0058] Color enhancement: mainly brightness transformation, such as using HSV (HueSaturationValue) enhancement.
[0059] Blurring: Enhance the model's generalization ability for blurred images through methods such as Gaussian filtering, box filtering, and median filtering.
[0060] Mixup: An algorithm for mixed-class enhancement of images in computer vision. It can mix images of different classes to expand the training data set.
[0061] After 150 rounds of iterative training, the model was verified on a test set consisting of 10,000 samples. The final model had an accuracy of 96% and a recall rate of 85%. The recognition speed of the matrix features increased by 100% after Pareto analysis compared with before analysis.
[0062] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. An intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events, characterized in that: The following steps are involved: Step S1, constructing a feature library; the feature library includes a demographic feature library, a key health monitoring feature library and a dynamic behavior feature library, and the collection and generation of the demographic feature library, the key health monitoring feature library and the dynamic behavior feature library all adopt the Pareto principle, and each includes the following 20% features: 20% demographic characteristics, including but not limited to: gender, age, place of residence, waist circumference, whether taking antihypertensive drugs, history of diabetes, smoking, and family history of cardiovascular and cerebrovascular diseases; 20% key health monitoring features, including but not limited to: total cholesterol, HDL, LDL, fasting blood glucose, A1C, systolic blood pressure, diastolic blood pressure; 20% dynamic behavior characteristics, including but not limited to: exercise duration, exercise frequency, protein intake, fat intake, sugar intake, salt intake, vegetable intake, water intake, smoking amount, and alcohol intake; The demographic feature matrix is constructed by converting the demographic feature library into a decimal number in the interval of [0, 255], and finally constructing an a*n matrix; the key health monitoring feature matrix is constructed by converting the key health monitoring feature library into a decimal number in the interval of [0, 255], and finally constructing a b*n matrix; the dynamic behavior feature matrix is constructed by converting the dynamic behavior feature library into a decimal number in the interval of [0, 255], and finally constructing a c*n matrix; Step S2, constructing the feature library in step S1 into a feature matrix; Step S3, merging the feature matrices in step S2 into a new matrix library; Step S4, constructing the matrix library in step S3 into a grayscale image; Step S5, constructing the grayscale image in step S4 into a neural network model; the neural network model is obtained by using a neural network algorithm to identify and analyze the grayscale image, and the neural network algorithm adopts YOLOv4; Step S6, using the neural network model in step S5 to obtain the user's current status and output a recommended health management plan.
2. The intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events according to claim 1, characterized in that: In step S2, the feature matrix includes a demographic feature matrix, a key health monitoring feature matrix and a dynamic behavior feature matrix.
3. The intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events according to claim 1, characterized in that: In step S3, the matrix library is an n*n matrix fused by a demographic feature matrix, a key health monitoring feature matrix and a dynamic behavior feature matrix.
4. The intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events according to claim 1, characterized in that: In step S4, the grayscale image is constructed by converting the numbers in the matrix library into decimal grayscale values.
5. The intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events according to claim 1, characterized in that: The YOLOv4 analysis process includes the following steps: Input: the grayscale image is used as the input analysis picture; Backbone: CSPDarkNet53 is used to extract the preliminary features of the grayscale image, build its backbone structure, and extract the grayscale feature matrix; Neck: Processes and enhances the grayscale image features extracted by the Backbone layer, so that the features learned by the model are the desired features; Head: Based on the enhanced grayscale feature matrix, the deep learning algorithm is used to give the user's current status and output the recommendation results, outputting the most urgently needed scenario-based guidance methods for people at risk of acute cardiovascular and cerebrovascular events.
6. The intelligent health management method for people at risk of acute cardiovascular and cerebrovascular events according to claim 5, characterized in that: The method for processing and enhancing the grayscale image features extracted by the Backbone layer comprises the following steps: Geometric enhancement: Enhance the generalization ability of the model by changing the image geometry through translation, rotation or shearing; Color enhancement: mainly brightness transformation, using HSV enhancement; Blurring: Enhance the model's generalization ability for blurred images through Gaussian filtering, box filtering or median filtering; Mixup: An algorithm for mixed-class enhancement of images in computer vision. It can mix images of different classes to expand the training data set.