A health data grouping processing method, system and terminal for multiple scenarios
Through the combination of self-organized mapping and weighted K-mean model, combined with ftSOM algorithm and visualization method, the problems of low training efficiency and lack of visual results in large-scale health data processing are solved, and rapid grouping and visual analysis are realized, suitable for medical and public Health field.
Patent Information
- Application Number
- CN202410910538.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-09
AI Technical Summary
When processing large-scale health data, the existing technology has low training efficiency, consumes a large amount of computing resources, relies on a large amount of labeled data, and lacks visual results, which affects data analysis and public health research in the medical field.
The self-organized mapping model and weighted K-mean model are used for rapid training and clustering, and the training process is accelerated through the ftSOM algorithm to achieve unsupervised learning, and the results of healthy data clustering are displayed using visual methods.
It realizes rapid clustering of large-scale health data, reduces computing resource consumption, does not rely on labeled data, and provides visual results to facilitate quick understanding and analysis of data patterns.
Smart Images

Figure CN118471536B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of artificial intelligence technology and health big data technology, and in particular to a health data grouping processing method, system, terminal and computer-readable storage medium for multiple scenarios. Background Art
[0002] Explore the diet, exercise and other lifestyle habits of the regional population to assess the health of the population's lifestyle. Lifestyle factors such as diet and exercise. By studying these factors, we can better understand how certain lifestyle habits are associated with specific health problems (such as cardiovascular disease, diabetes, obesity, etc.). This helps to reveal the potential causes of diseases and prevention strategies.
[0003] The above scenarios often require a large amount of health data for analysis, including disease conditions, clinical symptoms, treatment methods, daily behaviors, social information and other information. In previous medical analysis, it is usually necessary to extract this information into features, group it through classical statistics, data analysis methods, machine learning methods, etc., and explore the patterns and subgroups. Statistically, the features are directly described, the incidence and coexistence of the features are analyzed; variance analysis is used to determine whether there are significant differences in certain quantitative indicators between different disease subgroups. Some common data analysis methods include: using methods such as Pearson or Spearman correlation coefficients to analyze the correlation between different features; using logistic regression or multivariate regression models to explore the dependencies and influencing factors between different features; using methods such as Cox proportional-hazards model (Cox, coxproportional-hazards model, a semi-parametric regression model) to analyze the impact of comorbidity on patient prognosis; using network analysis to construct disease networks, analyze interactions and clusters between diseases, and reveal complex comorbidity patterns and potential common biological pathways. Machine learning uses methods such as cluster analysis, principal component analysis, factor analysis, etc. to further explore patterns and structures in the data, or uses advanced techniques such as random forests or artificial neural networks to identify complex comorbidity patterns and predict risks.
[0004] The disadvantages of existing technical methods are: low training efficiency, difficulty in completing training when the amount of data is large, or excessive resource consumption. When the amount of data is large, the use of classical data analysis, machine learning, and deep learning methods requires more computing resources, and requires a lot of time and computing resources to complete model construction. In addition, in the case of too many features, the model fitting effect may decline, affecting the clustering effect.
[0005] Reliance on large amounts of labeled data. Supervised learning methods rely on large amounts of labeled data (i.e., each input sample has a corresponding output label) to train the model. In the healthcare field, obtaining this labeled data is usually time-consuming and costly.
[0006] Lack of visualization results. In actual medical application scenarios, many analysis methods lack visualization results, which is not conducive to public health researchers, public policy makers, and medical practitioners to quickly understand and analyze data conditions and patterns in the data.
[0007] Therefore, the prior art still needs to be improved and developed. Summary of the invention
[0008] The main purpose of the present invention is to provide a method, system, terminal and computer-readable storage medium for health data clustering processing in multiple scenarios, aiming to solve the problems in the prior art of clustering large-scale population health data, which requires a lot of time and computing resources to build models, relies on a large amount of labeled data, and lacks visualization results.
[0009] To achieve the above object, the present invention provides a health data grouping processing method for multiple scenarios, and the health data grouping processing method for multiple scenarios comprises the following steps:
[0010] Obtaining a variety of health information data of multiple sample individuals;
[0011] Inputting a variety of health information data of a plurality of sample individuals into the self-organizing map model to be trained, and accelerating the training of the self-organizing map model to be trained by a fast training self-organizing map algorithm to obtain a trained self-organizing map model;
[0012] Inputting a variety of health information data of multiple sample individuals into the trained self-organizing map model to obtain a two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and inputting all sample numbers into a weighted K-means model to obtain a trained weighted K-means model, and outputting a variety of health information data of multiple sample individuals with group labels;
[0013] The distribution of multiple health information data of multiple sample individuals with group labels and multiple health information data of multiple sample individuals with group labels on a two-dimensional self-organizing mapping network is statistically analyzed to obtain statistical results of health data grouping processing and visualize them.
[0014] In addition, to achieve the above-mentioned purpose, the present invention also provides a health data grouping processing system for multiple scenarios, wherein the health data grouping processing system for multiple scenarios includes:
[0015] A data acquisition module, used to acquire a variety of health information data of multiple sample individuals;
[0016] The first model processing module is used to input a plurality of health information data of a plurality of sample individuals into the self-organizing map model to be trained, and to accelerate the training of the self-organizing map model to be trained by a fast training self-organizing map algorithm to obtain a trained self-organizing map model;
[0017] The second model processing module is used to input various health information data of multiple sample individuals into the trained self-organizing map model to obtain the two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and input all sample numbers into the weighted K-means model to obtain the trained weighted K-means model, and output various health information data of multiple sample individuals with group labels;
[0018] The statistics and visualization module is used to statistically analyze the distribution of multiple health information data of multiple sample individuals with group labels and multiple health information data of multiple sample individuals with group labels on a two-dimensional self-organizing mapping network, obtain the statistical results of health data grouping processing and perform visual display.
[0019] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a health data grouping processing program for multiple scenarios stored on the memory and runnable on the processor, and when the health data grouping processing program for multiple scenarios is executed by the processor, the steps of the health data grouping processing method for multiple scenarios as described above are implemented.
[0020] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a health data grouping processing program for multiple scenarios, and when the health data grouping processing program for multiple scenarios is executed by a processor, the steps of the health data grouping processing method for multiple scenarios as described above are implemented.
[0021] In the present invention, a plurality of health information data of a plurality of sample individuals are obtained; the plurality of health information data of a plurality of sample individuals are input into a self-organizing map model to be trained, and the self-organizing map model to be trained is accelerated through a fast training self-organizing map algorithm to obtain a trained self-organizing map model; the plurality of health information data of a plurality of sample individuals are input into the trained self-organizing map model, and a two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node are obtained, and all sample numbers are input into a weighted K-means model to obtain a trained weighted K-means model, and a plurality of health information data of a plurality of sample individuals with group labels are output; the distribution of the plurality of health information data of a plurality of sample individuals with group labels and the plurality of health information data of a plurality of sample individuals with group labels on a two-dimensional self-organizing map network is statistically analyzed, and statistical results of health data grouping processing are obtained and visualized. The model training speed of the present invention is fast, does not rely on labeled data, has visualization results, realizes rapid grouping of large-scale population health data and provides visualization grouping results, so that relevant personnel can quickly understand and analyze data conditions and patterns in the data. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a flow chart of a preferred embodiment of the method for grouping and processing health data in multiple scenarios of the present invention;
[0023] Figure 2 This is an example diagram of the ftSOM algorithm in a preferred embodiment of the method for health data clustering processing in multiple scenarios of the present invention;
[0024] Figure 3 It is a SOM (self-organizing map) network neuron weight map in a preferred embodiment of the health data grouping processing method for multiple scenarios of the present invention;
[0025] Figure 4 It is a heat map of the clustering results on the SOM network in a preferred embodiment of the health data clustering processing method for multiple scenarios of the present invention;
[0026] Figure 5 It is a cluster group radar chart in a preferred embodiment of the health data grouping processing method for multiple scenarios of the present invention;
[0027] Figure 6 It is a structural diagram of a preferred embodiment of the health data grouping processing system for multiple scenarios of the present invention;
[0028] Figure 7 It is a structural diagram of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION
[0029] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0030] The health data grouping processing method for multiple scenarios described in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the health data grouping processing method for multiple scenarios includes the following steps:
[0031] Step S10: Acquire a variety of health information data of a plurality of sample individuals.
[0032] Specifically, multiple sample individuals from multiple groups are determined, and multiple health information data of the multiple sample individuals are obtained; the multiple health information data of the multiple sample individuals are de-identified to obtain multiple de-identified health information data, and the de-identification is used to process privacy information in personal information so that the personal identity in the information cannot be identified or restored; the multiple de-identified health information data are pre-processed through feature engineering to obtain training data, and the pre-processing includes feature extraction, feature construction, data normalization and standardization, missing value processing, category data processing, feature dimensionality reduction and scale-invariant feature processing.
[0033] Among them, the sample refers to the individual in the population to be grouped, and each sample individual may be a patient or a subject. Sample individuals can come from a variety of groups, including: a group of patients with specific diseases in a hospital, a group of physical examinees in a health center, and a group of subjects selected in a regional or community health survey. The types of various health data will also vary with different groups, and may include: basic social information (gender, age, education level, economic status, etc.), living habits and ability information (types and frequencies of diet and exercise, bad habits, sleep conditions, self-care ability, etc.), scale information (information or scores obtained from commonly used scales and scoring tables), mental health status (whether anxious, depressed, etc.), medical information (disease history, disease type, symptoms, examination and imaging data, diagnostic information, medication information, follow-up information, etc.), health equipment collection information (such as portable blood pressure, blood sugar, blood oxygen monitoring equipment collected information) and other information obtained from multiple channels. It should be explained that when using this model for data processing, sample data from multiple groups can be used at one time, but the types and quantities of health information contained in the sample data need to be similar or the same.
[0034] The multiple sample individuals can be multiple elderly people from the same region who have received health surveys. The multiple health data include basic social information, living habits and ability information, mental health information, scale information, and medical information obtained from the elderly in the past at medical institutions.
[0035] It should be noted that health data contains a large amount of personal health information, and the use of this information involves privacy and confidentiality issues. All data acquisition and processing are carried out in compliance with requirements and are kept confidential.
[0036] Therefore, a variety of health data must be de-identified. De-identification refers to the process of processing personal information through technical means so that the personal identity in the information cannot be identified or restored. It deletes or modifies all personal identification information (such as name, address, telephone number, etc.) in the relevant data so that the personal information cannot be restored. After the data is anonymized, it is no longer possible to associate the data with a specific individual.
[0037] In order to improve the training efficiency of the model, the present invention also pre-processes multiple sample individual data through feature engineering (feature engineering refers to the process of creating useful features from raw data for use by machine learning algorithms) to obtain training data. The process is described in detail as follows: through feature extraction, new features are extracted from the raw data, including extracting keywords from text data and extracting information such as day, month, and year from date data. Through feature construction, new features are constructed based on existing data, such as calculating body mass index (BMI) from height and weight data. Normalize and standardize the data to scale the features to a uniform range or distribution. Process missing values, directly delete rows or columns containing missing values, or fill missing values with statistical values (mean, median, mode, etc.) or use model predictions. Process categorical data, label encode the data, unique hot encode the data, etc. According to the type of data obtained and the usage scenario, the data can also be processed by feature dimensionality reduction, scale-invariant feature processing, etc.
[0038] By preprocessing the various health information data of multiple elderly people, training data is obtained. In this process, the complex and irregular information is converted into a variety of regular and easy-to-process features. Since a sample data corresponds to multiple features, each of which can be regarded as a dimension of the data space, such data is called high-dimensional data. The high-dimensional data mentioned below refers to the preprocessed training data. In order to facilitate the demonstration of the results, in this embodiment, the data features are divided into several categories: indicators, diseases, symptoms, and treatment methods, and several subcategories are determined in these categories, such as: disease 1, disease 2, and disease 3.
[0039] Step S20: input multiple health information data of multiple sample individuals into the self-organizing map model to be trained, and accelerate the training of the self-organizing map model to be trained by the fast training self-organizing map algorithm to obtain a trained self-organizing map model.
[0040] Specifically, taking the SOM model to be trained (SOM model, Self-organizing Maps) as an example, it is necessary to select a grid structure and initialize weights; select the network structure of the self-organizing map model to be trained, and determine the shape, size and topology type of the network; common network shapes are rectangular and hexagonal. In this embodiment, a hexagonal network is selected, and each node is directly neighbored by six other nodes. The topological method selects the most common planar topology, and all neurons are arranged on a two-dimensional plane. Among them, neurons are the basic units of the SOM network. Each neuron is associated with a weight vector with the same dimension as the input data. Neurons are also referred to as nodes or units below. The initialization weight determines the starting state of the network training. In order to speed up the convergence of the training, principal component analysis (PCA) is used to find the first two principal components of the input data, and these two principal components are used as the basis for initializing the weights.
[0041] The size of the network is determined by the number of horizontal neurons and vertical neurons in the network. If the number of horizontal neurons is , the number of longitudinal neurons is The total number of samples is , adjust the value according to the degree of fit of the network to the data:
[0042] ; (1)
[0043] Furthermore, the training iteration of SOM needs to determine the weight update rule, and determine the weight update rule for the training iteration of the self-organizing map model to be trained, and the weight update rule includes the neighborhood function, the learning rate and the neighborhood radius:
[0044] ; (2)
[0045] in, It is The change in the neuron weight vector, It is The current weight vector of the neuron, is the learning rate, a positive constant that controls how much the weights change at each step, is the current input vector, is the set of all neurons, is the neighborhood function, defining the Neurons and winner neurons The topological relationship between is the neighborhood radius of the neighborhood function, which defines the size of the neighborhood. Indicates time, During training, usually over time Reduce.
[0046] Neighborhood functions for:
[0047] ; (3)
[0048] in, and They are neurons and winner neurons The position coordinates in the self-organizing map network lattice, is the Euclidean distance, which measures and The physical distance in the neuron grid, the larger the distance, the smaller the value of the neighborhood function, which means that the The winner neuron The smaller the impact, is the standard deviation of the Gaussian function, which controls the width of the neighborhood function, that is, the size of the impact range. is the exponent of the natural exponential function; in the learning process of SOM, It usually decreases over time, so that in the early stages, the network can make global adjustments to adapt to the overall structure of the data; in the later stages of training, the network pays more attention to local fine-tuning.
[0049] Neighborhood radius for:
[0050] ; (4)
[0051] in, is the initial neighborhood radius, is the current time step, is the maximum number of time steps during training; Formula (4) shows that the neighborhood radius decays exponentially with time. When , the neighborhood radius is , as time goes by As increases, the neighborhood radius gradually decreases. The rate of exponential decay is given by and Formula (4) ensures that at the beginning of training, a larger neighborhood radius allows the network to capture more global data structures, while as training progresses, a smaller neighborhood radius allows the network to refine local features.
[0052] In order to speed up the training, this model uses the ftSOM (fast-training SOM) algorithm to speed up the training. This method gradually "unfreezes" and trains units in batches, and only trains a part of the network in each batch, reducing the computational cost of training while still being able to effectively train the SOM structure, as detailed below:
[0053] The training process of SOM is to find the winner neuron (Best Matching Unit, the best matching unit, is conceptually equivalent to the winner neuron, which is the previous ) is completed. In the general SOM point-by-point iteration process, the neuron with the smallest distance or maximum similarity to the input data set is selected as the winner neuron (Here it means finding the winner neuron in each batch after batching):
[0054] ; (5)
[0055] in, is a function that represents the value of the independent variable that makes the function reach the minimum value. Represents the set of all neurons.
[0056] The present invention adopts the ftSOM algorithm, which is a batch training process: on a square SOM network, each unit is divided into The whole training process divides SOM into batches, each batch trains a different set of neural units. In each batch, a portion of the units on the SOM grid will be trained, and the rest of the units are considered "frozen", that is, they do not participate in the training.
[0057] When the self-organizing map model to be trained is accelerated by using the fast training self-organizing map algorithm, batch training is adopted to divide each unit of the self-organizing map model to be trained into Training is done in batches, with each batch training a different set of neural units, and finding the winner neuron in each batch The process is:
[0058] ; (6)
[0059] in, is the index of the batch, Representative batch The collection of neurons in For batch of neurons, It is neuron weight vector, is the number of modes in the dataset, n For the n Input modes, For Mode n The input vector in .
[0060] like Figure 2 As shown, Figure 2 Here is an example of the algorithm process. In this example, N=3 and the radius of the search range d=2. For each batch, only units (including Represents the batch index, i.e. batches). This means that in the first batch, 3x3 units will be trained, gradually training the entire network. The first batch consists of Figure 2 The first batch consists of black cells in the network and is trained as a regular SOM in a 3x3 geometry. Each input pattern is assigned to the nearest black cell. The second batch includes both gray and black cells, while the white cells are still "frozen". The weights of the gray cells are initialized using the average of the surrounding cells. In the third batch, the white cells are "unfrozen", added to the network and reassigned input patterns before training.
[0061] Step S30: Input the various health information data of multiple sample individuals into the trained self-organizing map model to obtain the two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and input all sample numbers into the weighted K-means model to obtain the trained weighted K-means model, and output the various health information data of multiple sample individuals with group labels.
[0062] Specifically, all sample data are input into the trained two-dimensional SOM model, which is equivalent to converting high-dimensional data into two-dimensional data. After that, the number of samples on each SOM network node is counted as the weight value. Next, the weights are normalized so that the sum of all weights is 1, and the two-dimensional data with weight values are input into the weighted K-Means model (weighted K-means model).
[0063] The first step is to initialize and select the initial cluster center , Represents the cluster center, which contains the coordinates of the center point of each cluster. The superscript 0 represents the initial value of the cluster center, which is randomly selected. Initial cluster centers represent the initial cluster locations.
[0064] Next, generate an initial weight vector ;
[0065] in, is the weight vector, the superscript 0 indicates the initial value of the weight vector, is the weight The total number of The superscript 0 in the Initialized value, yes The index of weights that determine the relative importance of each variable in calculating the distance.
[0066] Furthermore, the objective function to be minimized is determined :
[0067] ; (7)
[0068] in, represents the objective function, that is, the function to be minimized, Represents the membership matrix of the data points, which is used to represent the probability that each data point belongs to each cluster center, where each membership is recorded as , is the index of the data in the dataset, is the index of the variable at the data point, is the total number of data points, is the index of the cluster center, is the total number of initial cluster centers, is a variable at the data point, It is Of the data points The value of a variable, Refers to a variable on the cluster center (a data point is selected as the cluster center, so the cluster center is also a data point, and both have a length of ), It is The cluster center is in The value of a variable, It is Data points to The membership degree of the cluster centers is It is Data points In the The value of the variable and the The cluster center is in The square of the difference between the values on the variables (i.e., one component of the Euclidean distance), It is Weight, superscript It is a hyperparameter that adjusts the influence of variable weights and needs to be selected according to the data situation. The present invention hopes that variables with small differences within the cluster (i.e., variables with high stability) will obtain higher weights. Since each variable should have a corresponding weight to indicate its relative importance in calculating the distance between data points, it can be understood that the weights and variables are one-to-one corresponding. Therefore, the dimension of the weight vector, that is, the number of weights, is equal to the number of variables, so Also the total number of variables.
[0069] Further, update the membership, membership Represents data points For cluster centers In each iteration of the algorithm, the membership is updated according to the latest cluster center and variable weight. The process of updating the membership follows the following formula:
[0070] ; (8)
[0071] in, The expression that reaches the minimum value The value of are the indices of cluster centers traversed during minimization, It is The cluster center is in The value of a variable, It is Data points In the The value of the variable and the The cluster center is in The square of the difference between the values of the variables.
[0072] The update process is:
[0073] (1) For each data point , for all cluster centers Calculate the weighted distance, that is, for each variable Distance Multiply by the corresponding weight of Raise to the power and sum.
[0074] (2) For each data point , find the cluster center index with the smallest weighted distance .
[0075] (3) For each data point , the membership matrix The corresponding elements in Set to 1, indicating data points Belong to the cluster center For all other The cluster center of Set to 0.
[0076] Further, the weights are updated to complete the training. The total intra-cluster distance of each variable is calculated : For each variable , calculate the sum of squares of the differences between all data points and their respective cluster centers on the corresponding variables, and update the weight of each variable :
[0077] ; (9)
[0078] ; (10)
[0079] in, is the iteration index, is the total number of iterations, For the The distance from the data point to the cluster center in iterations.
[0080] By repeatedly updating the membership and weights until the model converges, the input data is clustered into multiple groups, and each sample individual is labeled with the group to which it belongs to distinguish which group it belongs to, and multiple health information data of multiple sample individuals with group labels are output.
[0081] Step S40, statistically analyze the distribution of multiple health information data of multiple sample individuals with group labels and multiple health information data of multiple sample individuals with group labels on the two-dimensional self-organizing map network, obtain the statistical results of health data grouping processing and perform visual display.
[0082] Specifically, two-dimensional SOM node data with clustering results are obtained through weighted KMeans model clustering. At the same time, high-dimensional sample individual data with clustering results (i.e., group labels) can be obtained according to the high-dimensional training data at the corresponding nodes of SOM.
[0083] The statistical methods used include direct counting, prevalence calculation and characteristic density calculation. Direct counting refers to counting the total number of a specific event (such as disease occurrence, death cases, etc.) in the research subjects. Prevalence refers to the presence of a certain symptom or disease in a specific population at a specific time point or time period. It is usually expressed as the ratio of the number of patients to the total population. The formula for prevalence is: The feature density is usually used to describe the average distribution of a certain feature (such as obesity, drinking, sleep index, etc.) in a specific area. In this embodiment, the formula for the feature density is: , use this formula to calculate the distribution density of a certain feature in the population.
[0084] In this embodiment, the SOM network used in the visualization legend has 10 horizontal neurons, 10 vertical neurons, and a hexagonal topology. The features are respectively indicator 1, indicator 2, indicator 3, disease 1, disease 2, disease 3, symptom 1, symptom 2, symptom 3, treatment 1, treatment 2, and treatment 3.
[0085] like Figure 3 As shown, high-dimensional sample individual data with group labels are mapped to a two-dimensional SOM network, and a neuron weight map on the SOM network is drawn to observe the feature distribution on each neuron. Each circle represents a neuron, and different colored sectors in the neuron represent the difference in weight types. In this embodiment, indicator 1, disease 1, symptom 1, and treatment method 1 are selected. In actual applications, the type and number of features represented by the weight can be selected by yourself. The size of a certain color sector in the neuron represents the size of the weight.
[0086] The specific statistical and visualization process is as follows: Data selection includes high-dimensional sample individual data with clustering results (i.e., group labels) and their corresponding SOM model data. First, draw a SOM network diagram, which consists of 100 neurons, 10 horizontal neurons and 10 vertical neurons. Draw a neuron into a circle, and neurons in different rows are staggered to ensure that there are six neurons around a neuron to achieve a hexagonal topological structure. For each neuron in the SOM network, extract its weight vector. Represent the features in each weight vector (such as indicator 1, disease 1, symptom 1, treatment 1) with sectors of different grayscale or color. The size of the sector reflects the relative size of the feature weight, and the angle of the sector is the same, and the sum should form a complete circle.
[0087] like Figure 4 As shown in the figure, high-dimensional sample individual data with group labels are mapped to a two-dimensional SOM network, and a certain indicator in the health data is used as a feature. A heat map of the clustering results on the SOM network is drawn to observe the distribution of the feature in each group. Each circle represents a neuron, and the color depth of the neuron represents the average density of the feature value mapped on the neuron, such as population density, morbidity, etc. The bold black line represents the cluster boundary, and you can observe which group the feature is distributed in.
[0088] The specific statistical and visualization process is as follows: Select a certain health indicator as a feature from the high-dimensional data. Draw a SOM network diagram, which consists of 100 neurons, 10 horizontal neurons and 10 vertical neurons. Draw a neuron as a circle, and neurons in different rows are staggered to ensure that there are six neurons around a neuron to achieve a hexagonal topology. Map each sample to its best matching neuron. Next, draw a heat map. For each neuron in the grid, calculate the feature density of all samples mapped to the neuron, and select the color depth according to the feature density of the neuron. For example, a color gradient from light (low value) to dark (high value) can be used. Each neuron is represented by a circle and drawn at the corresponding position of the SOM grid. The color of the circle reflects the average value of the feature on the neuron. The boundaries of different clusters are represented by bold black lines on the heat map, which helps to visually distinguish different groups.
[0089] like Figure 5 As shown in the figure, the high-dimensional sample individual data will be grouped according to the clustering results, and a radar chart will be drawn for each group to compare the relative relationship between the features. Figure 5 In the figure, a radar chart is drawn for a group of elderly people. The coordinate axes are the characteristics of the selected data: indicator 1, indicator 2, indicator 3, disease 1, disease 2, disease 3, symptom 1, symptom 2, symptom 3, treatment 1, treatment 2, treatment 3. The coordinate axis is the normalized proportion of a certain indicator, such as the prevalence rate, the proportion of treatment measures taken, etc. Figure 5 The health status in the group can be analyzed more intuitively.
[0090] The specific statistical and visualization process is as follows: Group the original data according to the group labels obtained from the clustering results, and draw a radar chart for each group of data. First, select the features to be represented by the radar chart: indicator 1, indicator 2, indicator 3, disease 1, disease 2, disease 3, symptom 1, symptom 2, symptom 3, treatment 1, treatment 2, treatment 3. Create coordinate axes based on the number of features. Each axis in the radar chart represents a feature, and the number of coordinate axes is equal to the number of features. Calculate the feature density of each feature in the group as the data point on each coordinate axis. Connect the points on each coordinate axis to form a closed polygon. Different colors or line styles can be used for different groups to facilitate comparison and distinction.
[0091] The present invention utilizes population health data to construct a SOM model, uses the ftSOM algorithm to accelerate training to adapt to large-scale data, and innovatively uses the nodes of the SOM corresponding to the population health data and the number of samples on the nodes as coordinates and weights to construct a weighted KMeans model to better reflect the differences among people in the actual population and achieve better clustering effects.
[0092] The embodiments of the present invention have at least the following beneficial effects:
[0093] (1) Use population health data to build a SOM model, use the ftSOM algorithm to accelerate training, to adapt to large-scale population health data, the training speed is faster, use the SOM model to reduce the data dimension, and then perform Kmeans clustering to speed up the clustering speed. Use the ftSOM algorithm to accelerate training and optimize the training speed.
[0094] (2) Not relying on labeled data, the present invention adopts unsupervised learning to group data points into clusters with high internal similarity and low inter-group similarity by calculating the similarity or distance between data points, without the need for labeled data.
[0095] (3) The weighted Kmeans algorithm is used to adjust the clustering effect by taking into account the density of population distribution on the network, so as to better reflect the differences among the actual population.
[0096] (4) Use the nodes of SOM corresponding to the population health data and the number of samples on the nodes as coordinates and weights to construct a weighted KMeans model to achieve better clustering results.
[0097] (5) Using a visualization method, the present invention visualizes a variety of health information data of multiple sample individuals with group labels and their distribution on the SOM network, and visualizes the health status of the group to facilitate analysis by relevant personnel.
[0098] Furthermore, if Figure 6 As shown, based on the above-mentioned health data grouping processing method for multiple scenarios, the present invention also provides a health data grouping processing system for multiple scenarios, wherein the health data grouping processing system for multiple scenarios includes:
[0099] A data acquisition module 51 is used to acquire a variety of health information data of a plurality of sample individuals;
[0100] The first model processing module 52 is used to input a plurality of health information data of a plurality of sample individuals into the self-organizing map model to be trained, and to accelerate the training of the self-organizing map model to be trained by a fast training self-organizing map algorithm to obtain a trained self-organizing map model;
[0101] The second model processing module 53 is used to input the various health information data of multiple sample individuals into the trained self-organizing map model to obtain the two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and input all sample numbers into the weighted K-means model to obtain the trained weighted K-means model, and output the various health information data of multiple sample individuals with group labels;
[0102] The statistics and visualization module 54 is used to statistically analyze the distribution of multiple health information data of multiple sample individuals with group labels and multiple health information data of multiple sample individuals with group labels on a two-dimensional self-organizing mapping network, obtain statistical results of health data grouping processing and perform visual display.
[0103] Furthermore, if Figure 7 As shown, based on the above-mentioned health data grouping processing method and system for multiple scenarios, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Figure 7 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.
[0104] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Furthermore, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a health data grouping processing program 40 for multiple scenarios is stored on the memory 20, and the health data grouping processing program 40 for multiple scenarios can be executed by the processor 10, thereby realizing the health data grouping processing method for multiple scenarios in the present application.
[0105] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing the health data grouping processing method for multiple scenarios.
[0106] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.
[0107] In one embodiment, when the processor 10 executes the health data grouping processing program 40 for multiple scenarios in the memory 20, the steps of health data grouping processing for multiple scenarios as described above are implemented.
[0108] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a health data grouping processing program for multiple scenarios, and when the health data grouping processing program for multiple scenarios is executed by a processor, the steps of the health data grouping processing method for multiple scenarios as described above are implemented.
[0109] In summary, the present invention provides a health data clustering processing method, system, terminal and storage medium for multiple scenarios, the method comprising: obtaining multiple health information data of multiple sample individuals; inputting the multiple health information data of multiple sample individuals into a self-organizing map model to be trained, and accelerating the training of the self-organizing map model to be trained through a fast training self-organizing map algorithm to obtain a trained self-organizing map model; inputting the multiple health information data of multiple sample individuals into the trained self-organizing map model to obtain a two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and inputting all sample numbers into a weighted K-means model to obtain a trained weighted K-means model, and outputting multiple health information data of multiple sample individuals with group labels; statistically analyzing the distribution of multiple health information data of multiple sample individuals with group labels and multiple health information data of multiple sample individuals with group labels on the two-dimensional self-organizing map network to obtain statistical results of health data clustering processing and visualize them. The model of the present invention has a fast training speed, does not rely on labeled data, has visual results, realizes the rapid clustering of large-scale population health data and provides visual clustering results, so that relevant personnel can quickly understand and analyze the data situation and patterns in the data.
[0110] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal 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 terminal. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or terminal including the element.
[0111] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium that can be read by a computer, and the program can include the processes of the above-mentioned method embodiments when executed. The computer-readable storage medium can be a memory, a disk, an optical disk, etc.
[0112] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.
Claims
1. A health data grouping processing method for multiple scenarios, characterized in that: The health data grouping processing method for multiple scenarios includes: Obtaining a variety of health information data of multiple sample individuals; Inputting a variety of health information data of a plurality of sample individuals into the self-organizing map model to be trained, and accelerating the training of the self-organizing map model to be trained by a fast training self-organizing map algorithm to obtain a trained self-organizing map model; Inputting a variety of health information data of multiple sample individuals into the trained self-organizing map model to obtain a two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and inputting all sample numbers into a weighted K-means model to obtain a trained weighted K-means model, and outputting a variety of health information data of multiple sample individuals with group labels; The distribution of multiple health information data of multiple sample individuals with group labels and multiple health information data of multiple sample individuals with group labels on a two-dimensional self-organizing map network is statistically analyzed to obtain statistical results of health data grouping processing and visualize them; The method of inputting a plurality of health information data of a plurality of sample individuals into a self-organizing map model to be trained, and accelerating the training of the self-organizing map model to be trained by a fast training self-organizing map algorithm to obtain a trained self-organizing map model specifically includes: Select the network structure of the self-organizing map model to be trained, and determine the shape, size and topology type of the network; Use principal component analysis to obtain the first two principal components of the input training data and use the first two principal components as the basis for initializing weights; The size of the network is determined by the number of horizontal neurons and vertical neurons in the network. If the number of horizontal neurons is , the number of longitudinal neurons is The total number of samples is , adjust the value according to the degree of fit of the network to the data: ;(1) Determine a weight update rule for training iterations of the self-organizing map model to be trained, wherein the weight update rule includes a neighborhood function, a learning rate, and a neighborhood radius: ;(2) in, It is The change in the neuron weight vector, It is The current weight vector of the neuron, is the learning rate, which controls the magnitude of the weight change in each step, is the current input vector, is the set of all neurons, is the neighborhood function, defining the Neurons and winner neurons The topological relationship between is the neighborhood radius of the neighborhood function, which defines the size of the neighborhood. Indicates time; Neighborhood functions for: ;(3) in, and They are neurons and winner neurons The position coordinates in the self-organizing map network lattice, is the Euclidean distance, which measures and The physical distance in the neuronal grid, is the standard deviation of the Gaussian function, is the exponent of the natural exponential function; Neighborhood radius for: ;(4) in, is the initial neighborhood radius, is the current time step, is the maximum number of time steps during training; The neuron with the smallest distance or largest similarity to the input data set is selected as the winner neuron : ;(5) in, is a function, representing the value of the independent variable that makes the function reach the minimum value, represents the set of all neurons; When the fast training self-organizing map algorithm is used to accelerate the training of the self-organizing map model to be trained, batch training is adopted to divide each unit of the self-organizing map model to be trained into Training is done in batches, with each batch training a different set of neural units, and finding the winner neuron in each batch The process is: ;(6) in, is the index of the batch, Representative batch The collection of neurons in For batch of neurons, It is neuron weight vector, is the number of modes in the dataset, n For the n Input modes, For Mode n The input vector in ; Statistics and visualization include: Select a health indicator as a feature from the high-dimensional data, draw a SOM network diagram, draw a neuron into a circle, and stagger neurons in different rows to ensure that there are six neurons around a neuron to achieve a hexagonal topology structure; Map each sample to the most matching neuron and draw a heat map. For each neuron in the grid, calculate the feature density of all samples mapped to the corresponding neuron and choose the color depth according to the feature density of the neuron.
2. The method for multi-scenario health data grouping processing according to claim 1, characterized in that: The health information data includes: basic social information of sample individuals, living habits and ability information, scale information, mental health status, medical information and health equipment collection information.
3. The method for multi-scenario health data grouping processing according to claim 1 or 2, characterized in that: The obtaining of a plurality of health information data of a plurality of sample individuals specifically includes: Determine a plurality of sample individuals from a plurality of groups, and obtain a plurality of health information data of the plurality of sample individuals; De-identifying a plurality of health information data of a plurality of sample individuals to obtain a plurality of de-identified health information data, wherein the de-identification processing is used to process the privacy information in the personal information so that the personal identity in the information cannot be identified or restored; A variety of de-identified health information data are preprocessed through feature engineering to obtain training data, wherein the preprocessing includes feature extraction, feature construction, data normalization and standardization, missing value processing, category data processing, feature dimensionality reduction and scale-invariant feature processing.
4. The method for multi-scenario health data grouping processing according to claim 1, characterized in that:The method of inputting the various health information data of multiple sample individuals into the trained self-organizing map model to obtain the two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and inputting all sample numbers into the weighted K-means model to obtain the trained weighted K-means model, and outputting the various health information data of multiple sample individuals with group labels, specifically includes: Select initial cluster centers , Represents the cluster center, which contains the coordinates of the center point of each cluster. The superscript 0 represents the initial value of the cluster center, which is randomly selected. Initial cluster centers, representing the initial cluster positions; Generate an initial weight vector ; in, is the weight vector, the superscript 0 indicates the initial value of the weight vector, is the weight The total number of The superscript 0 in the Initialized value, yes The index of Determine the objective function to minimize : ;(7) in, represents the objective function, Represents the membership matrix of the data points, which is used to represent the probability that each data point belongs to each cluster center, where each membership is recorded as , is the index of the data in the dataset, is the index of the variable at the data point, is the total number of data points, is the index of the cluster center, is the total number of initial cluster centers, It is Of the data points The value of a variable, It is The cluster center is in The value of a variable, It is Data points to The membership degree of the cluster centers is It is Data points In the The value of the variable and the The cluster center is in The square of the difference between the values of the variables, It is Weight, superscript is a hyperparameter that adjusts the influence of variable weight; Update membership, membership Represents data points For cluster centers In each iteration of the algorithm, the degree of membership is updated according to the latest cluster center and variable weight: ;(8) in, The expression that reaches the minimum value The value of are the indices of cluster centers traversed during minimization, It is The cluster center is in The value of a variable, It is Data points In the The value of the variable and the The cluster center is in The square of the difference between the values on the variables; Calculate the total within-cluster distance for each variable : For each variable , calculate the sum of squares of the differences between all data points and their respective cluster centers on the corresponding variables, and update the weight of each variable : ;(9) ;(10) in, is the iteration index, is the total number of iterations, For the The distance from the data point to the cluster center in the iteration; By repeatedly updating the membership and weights until the model converges, the input data is clustered into multiple groups, each sample individual is labeled with the group to which it belongs, and a variety of health information data of multiple sample individuals with group labels are output.
5. The method for multi-scenario health data grouping processing according to claim 1, characterized in that: The statistics include direct counting, prevalence calculation and characteristic density calculation; The direct counting refers to counting the total number of a certain event in the research object; The prevalence calculation refers to the presence of a condition or disease in a specific population at a specific point in time or over a period of time; The feature density calculation is used to describe the average distribution of a certain feature in a specific area.
6. The method for multi-scenario health data grouping processing according to claim 5, characterized in that: The prevalence is calculated as: ; The characteristic density is calculated as: 。 7. A health data grouping processing system for multiple scenarios, characterized in that: The health data grouping processing system for multiple scenarios is applied to the health data grouping processing method for multiple scenarios according to any one of claims 1 to 6, and the health data grouping processing system for multiple scenarios includes: A data acquisition module is used to acquire a variety of health information data of multiple sample individuals; The first model processing module is used to input a plurality of health information data of a plurality of sample individuals into the self-organizing map model to be trained, and to accelerate the training of the self-organizing map model to be trained by a fast training self-organizing map algorithm to obtain a trained self-organizing map model; The second model processing module is used to input various health information data of multiple sample individuals into the trained self-organizing map model to obtain the two-dimensional self-organizing map network corresponding to all sample data sets and the number of samples corresponding to each network node, and input all sample numbers into the weighted K-means model to obtain the trained weighted K-means model, and output various health information data of multiple sample individuals with group labels; The statistics and visualization module is used to statistically analyze the distribution of multiple health information data of multiple sample individuals with group labels and multiple health information data of multiple sample individuals with group labels on a two-dimensional self-organizing mapping network, obtain the statistical results of health data grouping processing and perform visual display.
8. A terminal, characterized in that: The terminal includes: a memory, a processor, and a health data grouping processing program for multiple scenarios stored in the memory and executable on the processor. When the health data grouping processing program for multiple scenarios is executed by the processor, the steps of the health data grouping processing method for multiple scenarios as described in any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a health data grouping processing program for multiple scenarios, and when the health data grouping processing program for multiple scenarios is executed by a processor, the steps of the health data grouping processing method for multiple scenarios as described in any one of claims 1-6 are implemented.
