Diabetic complication prediction method based on k-means clustering analysis

Through the feature selection and threshold calculation module combined with the convolution prediction module, using k-means clustering analysis and RNN algorithm, the problem of single early warning grading effect of diabetes prediction methods in the existing technology is solved, and more refined prediction of diabetes complications is achieved and diagnosis.

CN118942703AInactive Publication Date: 2024-11-12BEIJING HOSPITAL

Patent Information

Application Number
CN202411073211.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing diabetes prediction method based on k-means clustering analysis has a single effect in early warning grading, and cannot fully utilize the data set, making it difficult to meet the prediction needs of diabetes application fields.

Method used

The feature selection module is used to collect and filter data, combine the threshold calculation module and the convolution prediction module, and generate diabetes characteristic data through the k-means clustering analysis algorithm and the RNN convolution algorithm, and use Otsu threshold weighted calculation and neural network algorithm to perform diabetes. Prediction of subcomplications.

Benefits of technology

A more refined prediction of diabetes complications is achieved, more convincing prediction results are provided, and the accuracy of diabetes diagnosis and prevention effect are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118942703A_ABST
    Figure CN118942703A_ABST
Patent Text Reader

Abstract

The invention discloses a diabetic complication prediction method based on k-means clustering analysis, which comprises a feature selection module, the feature selection module comprises a collection unit and a cleaning and filtering unit, and the feature selection module collects patient data through the collection unit and filters the patient data through the cleaning and filtering unit. The feature selection module is used for cleaning and filtering data conforming to k-means clustering analysis through a cleaning and filtering unit; the invention discloses a threshold calculation module and relates to the technical field of diabetes prediction. According to the invention, a K-means model is refined and improved, and data conforming to k-means clustering analysis is collected and filtered out specifically through the feature selection module in combination with a current medical data refining operation process. The method comprises the steps that firstly, k-means clustering analysis is carried out on a diabetes mellitus prediction model, then the threshold value calculation module gathers feature data of k-means clustering analysis and an RNN convolution algorithm, a cohesion result of k-means clustering analysis is calculated in a weighted mode, finally, the RNN convolution algorithm intervenes in and calculates the cohesion result at night, the diabetes mellitus prediction model for doctor diagnosis is achieved and sketched, and the prediction method is advanced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of diabetes prediction, and in particular to a method for predicting diabetes complications based on k-means cluster analysis. Background Art

[0002] K-means cluster analysis is a widely used unsupervised learning algorithm. K-means cluster analysis is used to divide data points into K different clusters or groups during the calculation process. The data points are assigned to the nearest cluster center by iteratively optimizing the objective function of the sum of squared errors, and the location of the cluster center is recalculated until the cluster center no longer changes or changes very little, which facilitates further analysis and understanding of the structure and pattern of the data set. The advantages of k-means cluster analysis include simplicity and ease of implementation, fast processing speed, and suitability for processing large-scale data sets.

[0003] For example, the patent with publication number CN107403072A discloses a prediction and early warning method for type 2 diabetes based on machine learning. The patent specifically discloses "a prediction and early warning method for type 2 diabetes based on machine learning, which establishes a two-layer prediction and analysis model for diabetes by clustering first and then classifying through the k-means algorithm and the Logistic Regression algorithm, and performs early warning classification on the classification results of non-diabetes through the rules analyzed by the C4.5 algorithm and the k-means algorithm, and collects and manages the diabetes data set through daily data and updates and maintains it to optimize the original two-layer prediction and analysis model for diabetes. This method includes five modules: data collection module, data preprocessing module, prediction and analysis module, early warning classification module and daily management module." The technical solution realizes the technical effect of "this method includes five modules: data collection module, data preprocessing module, prediction and analysis module, early warning classification module and daily management module. It is mainly used in three aspects: early prediction and analysis of diabetes, early warning classification of high-risk groups and daily health management of diabetes. This method has higher prediction accuracy in practical applications, focuses more on the health parameters of each instance, and performs early warning classification for non-diabetes instances, which can play a better preventive role and achieve early prevention of diseases."

[0004] However, the method has the problem of single warning classification effect during use, and cannot use sufficient data sets to achieve more convincing technical purposes.

[0005] The patent with publication number CN103729798B discloses a coal mine safety evaluation system based on improved k-means clustering. The patent specifically discloses "an improved k-means clustering coal mine safety evaluation system, including a mine safety data input module, a data analysis module based on improved k-means clustering, and a result output module; the mine safety data input module receives mine safety status data submitted by users, and the data analysis module based on improved k-means clustering uses an improved k-means clustering method to perform data analysis based on the mine safety status data, and the result output module displays the results of the data analysis to provide an auxiliary basis for the safety management decision-making of coal mine enterprises." The technical solution realizes that "after adopting the coal mine safety evaluation system of the present invention, the scoring results of the safety evaluation of the mines belonging to the coal enterprise can be analyzed to achieve comprehensive management of the coal mine safety evaluation, and the evaluation results can provide important technical basis for the management decision-making of the coal mine enterprise."

[0006] Similarly, although the k-means model has been refined and improved in this method, it still cannot meet the purpose of diabetes prediction in the diabetes application field, and there is room for improvement. Summary of the invention

[0007] The purpose of the present invention is to provide a method for predicting diabetes complications based on k-means cluster analysis, which analyzes the patient's body data (such as blood sugar, uric acid, TH, etc.) using the k-means cluster analysis algorithm, obtains a convergence value and predicts the specific prediction results of each corresponding diabetes sub-complication after the data changes using a graph, and solves the problems raised in the above background technology.

[0008] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a diabetes complication prediction system based on k-means cluster analysis, comprising a feature selection module, the feature selection module comprising a collection unit and a filtering unit, the feature selection module collects patient data through the collection unit, and the feature selection module cleans and filters out data that conforms to the k-means cluster analysis through the filtering unit; a threshold calculation module, the threshold calculation module calculates the clustering result based on the feature data of the k-means cluster analysis and integrates the Otsu threshold weighting; and a convolution prediction module, the convolution prediction module learns the clustering results and feature data based on a neural network algorithm, and automatically identifies and outlines the diabetes feature data.

[0009] A method for predicting diabetes complications based on k-means cluster analysis, including

[0010] Step S1: The feature selection module collects and filters data that conforms to k-means clustering analysis;

[0011] Step S2: the threshold calculation module combines the characteristic data of the k-means cluster analysis with the RNN convolution algorithm and performs weighted calculation to obtain the inner clustering result of the k-means cluster analysis;

[0012] Step S3: The RNN convolution algorithm is used again to calculate the clustering results at night, and a diabetes prediction model is drawn up for doctors to diagnose.

[0013] As a further solution of the present invention: the collection unit includes a blockchain and a cloud platform, the blockchain stores a unique hash value of the patient's physiological indicators, and the cloud platform is used for blockchain reading, writing, modification and storage.

[0014] As a further solution of the present invention: the cleaning unit includes Smartbi and ETL tools, and the Smartbi and ETL tools work together to generate a visual process configuration that supports k-means clustering analysis data. The visual process configuration filters the k-means clustering analysis data through built-in data preprocessing functions such as sorting, deduplication, mapping, row and column merging, row and column conversion aggregation, and removal of null values.

[0015] As a further solution of the present invention: the k-means clustering analysis is provided with a distinction unit for patient cluster classification and a clustering unit for data analysis, the Otsu threshold includes a digitized adaptive threshold selection unit, and the clustering unit and the adaptive threshold selection unit are weightedly fused to produce a clustering result.

[0016] As a further solution of the present invention: a plurality of cluster centers are provided in the clustering unit, a plurality of two-dimensional weighted points are provided in the adaptive threshold value selection unit, the two-dimensional weighted points are calculated correspondingly with the cluster centers and updated iteratively to obtain inner clustering units, and the inner clustering units are allocated for calculation by the convolution prediction module.

[0017] As a further solution of the present invention: the convolution prediction module includes a composite RNN algorithm unit, the composite RNN algorithm unit includes an RNN algorithm and an inner clustering unit, and the composite RNN algorithm unit learns and predicts diabetes characteristic data through the relationship and feature distribution between data points within the inner clustering unit.

[0018] As a further solution of the present invention: the composite RNN algorithm unit includes a convolutional pooling layer and a fusion layer. The composite RNN algorithm unit obtains the extracted features of the cohesive class units through the convolutional pooling layer. The composite RNN algorithm unit learns the multi-layer abstract representation of the cohesive class units through the fusion layer and captures the patterns and feature data of the cohesive class units.

[0019] As a further solution of the present invention: the composite RNN algorithm unit also includes a prediction layer, a training layer and an evaluation layer. The composite RNN algorithm unit provides classification, regression and spatial mapping of the clustered units through the prediction layer, the composite RNN algorithm unit provides model generation, model training and model optimization of the clustered units through the training layer, and the composite RNN algorithm unit provides model evaluation of the clustered units through the evaluation layer.

[0020] As a further solution of the present invention: the model evaluation includes a verification unit, and the model evaluation performs cyclic simulation and verification on the composite RNN algorithm unit through the verification unit.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. This application refines and improves the K-means model, combines the current medical data refinement operation process and specifically collects and filters data that meets the k-means clustering analysis through the feature selection module, and uses the threshold calculation module to combine the feature data of the k-means clustering analysis and the RNN convolution algorithm and weighted calculation to obtain the inner clustering results of the k-means clustering analysis. Finally, the RNN convolution algorithm intervenes and completes the calculation of the inner clustering results, thereby realizing and outlining a diabetes prediction model for doctors to diagnose, and the prediction method is advanced.

[0023] 2. This application uses Smartbi and ETL tools to screen and classify values, and then uses the Otsu threshold to filter out values, thereby analyzing and outlining the output clustering result parameters and further classifying the clustering results. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0026] In the description of the present invention, it should be noted that, for those skilled in the art, the specific meanings of the above terms in the present invention can be understood in specific situations. The following describes an embodiment of the present invention based on its overall structure.

[0027] In an embodiment of the present invention, a diabetes complication prediction system based on k-means cluster analysis includes a feature selection module, wherein the feature selection module includes a collection unit and a filtering unit, wherein the feature selection module collects patient data through the collection unit, and the feature selection module cleans and filters out data that conforms to the k-means cluster analysis through the filtering unit;

[0028] Other embodiments of the present invention: please refer to Figure 1 ,The feature selection module first collects the data set, which includes various physiological indicators associated with diabetes such as blood glucose level, blood pressure, BMI index and age. Then after the data is collected, missing values, outliers and duplicate values ​​are processed to ensure that the data format is consistent and can be used for cluster analysis. Finally, based on the expertise in the field of diabetes research, features related to diabetes are selected for analysis.

[0029] This application is based on the task background of diabetes classification that "the current common diabetes classification method divides diabetes into type 1 diabetes and type 2 diabetes, of which 90% belong to type 2 diabetes. Type 2 diabetes may present differently in different individuals in terms of etiology, clinical manifestations, prognosis, etc., with high heterogeneity and different clinical outcomes", combined with the background of elderly diabetes classification that "my country has entered an aging society, and the elderly diabetic population has many complications and a high mortality rate, but there is a lack of refined management. A more refined classification method is urgently needed, and the complication risks of various diabetes subgroups must be clarified for early intervention". Based on the patients' various clinical data, cluster analysis and classification are performed on elderly diabetic patients (Age>=65) and the stability is verified. The progression differences of complications of different types of diabetes are analyzed (that is, the distribution of complications between the categories obtained by clustering must be different), to assist in the realization of precision medicine for newly diagnosed diabetic patients and improve clinical outcomes.

[0030] Experimental data

[0031] Data sources—clinical data of Kunshan diabetes cohort patients, validation data—China-U.S. type 2 RCT and HR-SP2086-304RCT, and clustering indicators—BMI, fasting blood glucose, uric acid / creatinine ratio (ND), and TG / HDL (TH);

[0032] Data training

[0033] Training cohort (φ=5472), outlier removal, standardization (recording mean and variance), validation cohort - validation using standardized data from the Kunshan cohort, cyclical outlier removal;

[0034] Algorithm Process

[0035] Based on the idea of ​​division in k-means cluster analysis, the φ training queues are divided into n clusters, and the distance between each training queue and the cluster center point is made as small as possible. Specifically,

[0036] Step A: Select the n points with the longest distance in the sample as the initial cluster centers;

[0037] Step B: Calculate the distance between each sample and the n centers, and classify it into the cluster with the nearest center.

[0038] Step C, calculate the new center point of each cluster / the average point of the samples within the cluster;

[0039] Step D: Repeat steps B and C until the center point no longer changes.

[0040] According to the above method, we can find that the number of clusters (N-cluster) is 3, which is the best.

[0041] When cluster is 1, BMI is maximum and in the MOD stage;

[0042] When cluster is 2, blood glucose is the highest, uric acid / creatinine ratio (ND) is the lowest, and homa-β is the lowest, which is the SIDD stage;

[0043] When cluster is 3, BMI is the second largest, TG / HDL (TH) is the highest, homa-IR is the highest, and it is in the SIRD stage.

[0044] When all clusters (N-cluster) are obtained, the indicator distribution map is formed through the Kruskal-Wallis test method, and the risk simulation view of stroke, coronary heart disease, and fatty amine is obtained to help diagnose and prevent diabetes.

[0045] A threshold calculation module, wherein the threshold calculation module calculates the clustering result based on the characteristic data of the k-means cluster analysis and integrates the Otsu threshold weighted calculation, wherein the k-means cluster analysis is provided with a distinction unit for patient cluster classification and a clustering unit for data analysis, and the Otsu threshold includes a digitized adaptive threshold value selection unit, and the clustering unit and the adaptive threshold value selection unit are weightedly integrated to obtain the clustering result;

[0046] Other embodiments of the present invention: This application performs verification calculations based on the diabetes data set collected by the Kunshan research team, which includes features such as blood sugar levels, blood pressure and BMI index. The patients are divided into two clusters using K-means clustering analysis, and the Otsu threshold method is used to calculate the weight of each cluster, and finally the clustering results are obtained.

[0047] patient Blood sugar levels blood pressure BMI index 1 120 70 25 2 140 90 30 3 130 80 27 4 160 85 35 5 110 75 20 6 150 95 32

[0048] Table 1. Diabetes patient dataset

[0049] K-means cluster analysis

[0050] First, the K-means algorithm is used to divide the patients into two clusters. In this embodiment, the number of clusters selected is 2 and the cluster centers are calculated.

[0051] Cluster 1’s cluster center: (130, 80, 27)

[0052] Cluster 2’s cluster center: (150, 90, 30)

[0053] The distance from each data point to the two cluster centers is then calculated and assigned to the nearest cluster center.

[0054] After this we get two new clusters:

[0055] Cluster 1: Patients 1, 3, 5

[0056] Cluster 2: Patients 2, 4, 6

[0057] Therefore, the weight of each cluster is calculated using the Otsu threshold method. (The Otsu threshold method is an adaptive threshold selection technique that can be used for image segmentation and can also be used for data processing.) Here it is applied to the cluster analysis results to calculate the weight of each cluster.

[0058] Output clustering results

[0059] Cluster 1: Patients 1, 3, 5, weight 0.6

[0060] Cluster 2: Patients 2, 4, 6, weight 0.4

[0061] Therefore, this embodiment completes the feature data based on K-means clustering analysis, and combines the clustering results calculated by the Otsu threshold. Specifically, the Otsu threshold is achieved by outlining the image to distinguish between black and white values ​​in the calculation. In the embodiment of the present application, the data of the output clustering results is transcoded into binary parameters used for image analysis, and then the binary parameters are analyzed and outlined through the Otsu threshold to achieve classification of the clustering results.

[0062] A convolution prediction module, which learns clustering results and feature data based on a neural network algorithm, automatically identifies and outlines diabetes feature data, wherein a plurality of cluster centers are provided in the cluster unit, and a plurality of two-dimensional weighted points are provided in the adaptive threshold value selection unit. The two-dimensional weighted points are calculated correspondingly with the cluster centers and iteratively updated to obtain inner cluster units, which are allocated and calculated by the convolution prediction module. In fact, the convolution prediction module performs iterative calculations on a plurality of two-dimensional weighted points in the iterative adaptive value selection unit, and replaces and updates the inner cluster units after performing several rounds of iterative calculations on the two-dimensional weighted points, thereby achieving the purpose of preliminary screening of clustering results and feature data;

[0063] Other embodiments of the present invention:

[0064] As a further solution of the present invention: the collection unit includes a blockchain and a cloud platform, the blockchain stores a unique hash value (composed of an unalterable combination of numbers and letters) of the patient's physiological indicators, and the cloud platform is used for blockchain reading, writing, modification and storage. In this application, the hash value is used to encode the blockchain, so that the blockchain stores unique and unalterable diabetes data, so that even during the transfer process, garbled characters or errors will not occur, ensuring the healthy use of data.

[0065] As a further solution of the present invention: the cleaning unit includes Smartbi (SmartBI is a business intelligence tool for data analysis and reporting. It can help users extract data from multiple data sources and display data analysis results through visualization and reporting. SmartBI usually includes data dashboards, report design, data mining, predictive analysis and other functions, so that users can better understand the trends and patterns behind the data and support corporate decision-making and strategy formulation) and ETL (ETL tools refer to Extract, Transform, Load tools, which are software used to extract data from multiple different data sources, perform data transformation and load data into a target data warehouse or database) tools. The Smartbi and ETL tools are linked to generate a visualization process configuration that supports k-means clustering analysis data. The visualization process configuration filters the k-means clustering analysis data through built-in sorting, deduplication, mapping, row and column merging, row and column transformation aggregation, and null value removal data preprocessing functions.

[0066] Other embodiments of the present invention: In order to realize the k-means clustering analysis data cleaning and visualization process configuration based on Smartbi and ETL tools, a solution including data extraction, cleaning, conversion, loading and final k-means clustering analysis process is proposed.

[0067] 1. Data Extraction

[0068] Use ETL tools to extract data from various data sources, which can include relational databases, NoSQL databases, file systems (such as CSV or Excel files), etc.

[0069] 2. Data Cleaning (Clean)

[0070] Use ETL tools and Smartbi's built-in functions to clean the data and eliminate the miscellaneous and interfering items in the data;

[0071] Sorting: Sort the data according to the specified fields for subsequent processing. For example, if there are single items such as A and B in the k-means cluster analysis in this application, the internal constraints are used to perform ABA or ABABAB sorting;

[0072] Deduplication: Delete duplicate records to ensure the uniqueness of data, which actually means deleting the order of fields of the same type and nature;

[0073] Mapping: Map the fields in the original data to standardized field names;

[0074] Merge rows and columns: merge data from multiple tables or files into a unified data set;

[0075] Row-column conversion aggregation: perform row-column or column-row conversion operations as needed and aggregate data, such as calculating the sum, average, etc.

[0076] Include Null Value Removal: Find and handle null values, with the option to delete records containing null values ​​or fill null values ​​with default values.

[0077] 3. Data Transformation

[0078] The cleaned data is converted into a data format suitable for k-means clustering analysis. Then the data is transformed through standardization / normalization (standardizing or normalizing numerical data to reduce the impact of different dimensions) and feature selection (selecting meaningful feature fields for clustering analysis).

[0079] In addition, if the data dimension is high, PCA dimensionality reduction can be used (PCA is a commonly used data dimensionality reduction technique used to convert high-dimensional data sets into low-dimensional data sets while retaining the key features of the data sets. The basic idea is to map the original data to a new coordinate system through linear transformation. Each dimension in the new coordinate system is called a principal component. The principal components are arranged in order of decreasing variance. The principal components with larger variance are retained, while those with smaller variance are ignored, thereby achieving data dimensionality reduction), retaining the main data information and improving the clustering effect. Principal component analysis (PCA) can also be used to enhance the effect analysis.

[0080] 4. Data loading

[0081] Load the cleaned and transformed data into a data warehouse or database for analysis and visualization. This step ensures that the data is ready for k-means clustering analysis.

[0082] 5. k-means cluster analysis

[0083] Configure the k-means clustering analysis process in Smartbi:

[0084] Configure the parameters of the k-means algorithm, such as the number of clusters (k value), maximum number of iterations, etc.

[0085] Run k-means cluster analysis to generate clustering results.

[0086] 6. Visual process configuration

[0087] Use Smartbi to create a visualization process to display the results of k-means clustering analysis. The visualization configuration steps include:

[0088] Create a dashboard: Create a new dashboard in Smartbi to display the clustering results.

[0089] Add charts: Add various chart types (such as scatter plots, bar charts, pie charts, etc.) to show the distribution and characteristics of different clusters.

[0090] Configure interactions: Set up interactions between charts, allowing users to view detailed information by clicking or selecting specific clusters.

[0091] Filtering and Sorting: Configure data filters and sorting options so that users can dynamically adjust the scope and order of data displayed.

[0092] 7. Verification and optimization

[0093] Verify the entire process to ensure that the data is accurate and the visualization effect meets the requirements. If any problems are found, return to the corresponding steps for adjustment and optimization.

[0094] 8. Deployment and Monitoring

[0095] Deploy the configured processes to the production environment and set up monitoring and alarm mechanisms to ensure the stable operation of the data processing and analysis processes.

[0096] As a further solution of the present invention: the convolution prediction module includes a composite RNN algorithm unit, the composite RNN algorithm unit includes an RNN algorithm and an inner cluster unit, the composite RNN algorithm unit learns and predicts diabetes feature data through the relationship between data points and feature distribution within the inner cluster unit, the composite RNN algorithm unit includes a convolution pooling layer and a fusion layer, the composite RNN algorithm unit obtains the extracted features of the inner cluster unit through the convolution pooling layer, the composite RNN algorithm unit learns the multi-layer abstract representation of the inner cluster unit through the fusion layer, and captures the pattern and feature data of the inner cluster unit, the composite RNN algorithm unit also includes a prediction layer, a training layer and an evaluation layer, the composite RNN algorithm unit provides classification, regression and spatial mapping of the inner cluster unit through the prediction layer, the composite RNN algorithm unit provides model generation, model training and model optimization of the inner cluster unit through the training layer, and the composite RNN algorithm unit provides model evaluation of the inner cluster unit through the evaluation layer.

[0097] Other embodiments of the present invention: This application constructs a deep learning model for processing data with spatial correlation by combining an inner clustering unit with a convolution prediction module;

[0098] Cohesive unit

[0099] Data analysis and clustering: This application analyzes and preprocesses the input data and uses K-means clustering or other clustering algorithms to divide the data into different clusters (please refer to the scheme proposed in the above embodiment, which is not disclosed here to avoid repetition), and then uses the RNN algorithm to calculate and analyze the clustered data to obtain the internal cluster units.

[0100] Feature extraction: The cluster is represented by extracting statistical features, calculating cluster centers, etc. within each cluster unit, thereby converting the original data into a more representative and high-dimensional feature vector (the feature vector is the data selected after removing outliers and standardization as proposed in the above embodiment).

[0101] Internal relationship analysis: Analyze the relationship and feature distribution between data points within each cluster unit to better understand the structure and characteristics of the data, such as

[0102] Based on the idea of ​​division in k-means cluster analysis, the φ training queues are divided into n clusters, and the distance between each training queue and the cluster center point is made as small as possible. Specifically,

[0103] Step A: Select the n points with the longest distance in the sample as the initial cluster centers;

[0104] Step B: Calculate the distance between each sample and the n centers, and classify it into the cluster with the nearest center.

[0105] Step C, calculate the new center point of each cluster / the average point of the samples within the cluster;

[0106] Step D: Repeat steps B and C until the center point no longer changes.

[0107] According to the above method, we can find that the number of clusters (N-cluster) is 3, which is the best.

[0108] When cluster is 1, BMI is maximum and in the MOD stage;

[0109] When cluster is 2, blood glucose is the highest, uric acid / creatinine ratio (ND) is the lowest, and homa-β is the lowest, which is the SIDD stage;

[0110] When cluster is 3, BMI is the second largest, TG / HDL (TH) is the highest, homa-IR is the highest, and it is in the SIRD stage.

[0111] When all clusters (N-cluster) are obtained, the indicator distribution map is formed through the Kruskal-Wallis test method, and the risk simulation view of stroke, coronary heart disease, and fatty amine is obtained to help diagnose and prevent diabetes.

[0112] Feature representation: After obtaining the feature representation of each clustering unit, it can be used as input for the subsequent convolutional prediction module.

[0113] Convolutional prediction module

[0114] Feature learning: By performing convolution and pooling operations on the features extracted by clustered units, the RNN convolutional network can learn multi-layer abstract representations of the data, thereby better capturing the characteristics and patterns of the data.

[0115] Prediction module: Add a fully connected layer or other prediction layer on top of the RNN convolutional network to map the learned features to the output space and perform the final prediction task (including classification, regression, etc.).

[0116] Model training: The entire model is trained using labeled data, and the model parameters are optimized by minimizing the loss function so that it can better fit the training data and improve the prediction performance.

[0117] Model evaluation: Use validation sets or cross-validation techniques to evaluate the trained model and its generalization ability and performance on unseen data.

[0118] Overall process

[0119] Data input: Input the original data into the cohesive unit, perform data analysis, clustering and feature extraction, and obtain the feature representation of each cohesive unit.

[0120] Feature transfer: The feature representation extracted by the clustering unit is input into the convolutional prediction module, passing the spatial features and patterns of the RNN convolutional network data.

[0121] Prediction output: Add a prediction module on top of the RNN convolutional network to perform the final prediction task and output the prediction results of the model.

[0122] By combining the inner clustering unit with the convolutional prediction module, the features extracted by clustering analysis and the spatial features learned by the RNN convolutional network can be utilized to more effectively process data with spatial correlation and improve the performance and generalization ability of the model.

[0123] The working principle of the present invention is: the present application first collects and filters the data that conforms to the k-means cluster analysis through the feature selection module, and then the threshold calculation module combines the feature data of the k-means cluster analysis and the RNN convolution algorithm, and weightedly calculates the inner clustering results of the k-means cluster analysis. Finally, the RNN convolution algorithm intervenes and calculates the inner clustering results at night to realize and outline a diabetes prediction model for doctors to diagnose.

[0124] What is described above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A diabetes complication prediction system based on k-means cluster analysis, characterized in that: It includes a feature selection module, which includes a collection unit and a filtering unit. The feature selection module collects patient data through the collection unit, and the feature selection module cleans and filters out data that meets the k-means clustering analysis through the filtering unit; A threshold calculation module, wherein the threshold calculation module calculates the clustering result based on the feature data of k-means cluster analysis and integrates Otsu threshold weighting; and A convolution prediction module learns clustering results and feature data based on a neural network algorithm to automatically identify and outline diabetes feature data.

2. A diabetes complication prediction system based on k-means cluster analysis according to claim 1, characterized in that: The collection unit includes a blockchain and a cloud platform. The blockchain stores unique hash values ​​of the patient's physiological indicators, and the cloud platform is used for blockchain reading, writing, modification and storage.

3. A diabetes complication prediction system based on k-means cluster analysis according to claim 2, characterized in that: The cleaning unit includes Smartbi and ETL tools. The Smartbi and ETL tools work together to generate a visual process configuration that supports k-means clustering analysis data. The visual process configuration filters the k-means clustering analysis data through built-in data preprocessing functions such as sorting, deduplication, mapping, row and column merging, row and column conversion aggregation, and removal of empty values.

4. A diabetes complication prediction system based on k-means cluster analysis according to claim 3, characterized in that: The k-means clustering analysis is provided with a distinguishing unit for patient cluster classification and a clustering unit for data analysis. The Otsu threshold includes a digitized adaptive threshold value selection unit. The clustering unit and the adaptive threshold value selection unit are weighted and fused to obtain a clustering result.

5. A diabetes complication prediction system based on k-means cluster analysis according to claim 4, characterized in that: The clustering unit is provided with a plurality of clustering centers, the adaptive threshold value selection unit is provided with a plurality of two-dimensional weighted points, the two-dimensional weighted points are calculated correspondingly with the clustering centers and iteratively updated to obtain inner clustering units, and the inner clustering units are allocated for calculation by the convolution prediction module.

6. A diabetes complication prediction system based on k-means cluster analysis according to claim 5, characterized in that: The convolution prediction module includes a composite RNN algorithm unit, which includes an RNN algorithm and an inner clustering unit. The composite RNN algorithm unit learns and predicts diabetes feature data through the relationship and feature distribution between data points within the inner clustering unit.

7. A diabetes complication prediction system based on k-means cluster analysis according to claim 6, characterized in that: The composite RNN algorithm unit includes a convolutional pooling layer and a fusion layer. The composite RNN algorithm unit obtains the extracted features of the cohesive class unit through the convolutional pooling layer. The composite RNN algorithm unit learns the multi-layer abstract representation of the cohesive class unit through the fusion layer and captures the pattern and feature data of the cohesive class unit.

8. A diabetes complication prediction system based on k-means cluster analysis according to claim 7, characterized in that: The composite RNN algorithm unit also includes a prediction layer, a training layer and an evaluation layer. The composite RNN algorithm unit provides classification, regression and spatial mapping of the clustering unit through the prediction layer, provides model generation, model training and model optimization of the clustering unit through the training layer, and provides model evaluation of the clustering unit through the evaluation layer.

9. A diabetes complication prediction system based on k-means cluster analysis according to claim 8, characterized in that: The model evaluation includes a verification unit, and the model evaluation performs cyclic simulation and verification on the composite RNN algorithm unit through the verification unit.

10. A method for predicting diabetes complications based on k-means cluster analysis according to claim 9, characterized in that: include Step S1: The feature selection module collects and filters data that conforms to k-means clustering analysis; Step S2: the threshold calculation module combines the characteristic data of the k-means cluster analysis with the RNN convolution algorithm and performs weighted calculation to obtain the inner clustering result of the k-means cluster analysis; Step S3: The RNN convolution algorithm is used again to calculate the clustering results at night, and a diabetes prediction model is drawn up for doctors to diagnose.

Citation Information

Patent Citations

  • Clinical data mining analysis and aided decision-making method based on Internet integrated medical platform

    CN109830303A

  • Diabetes early warning model based on improved k-means clustering

    CN111223568A

  • High-risk diabetes complication early-warning system based on deep learning and ensemble learning

    CN111968741A

  • Image binarization method based on joint optimization of Otsu method and K-means clustering algorithm

    CN112037233A

  • Data processing device and system for clustering and typing diabetes

    CN115881307A

Cited By

  • New typing system for metabolic dysfunction-related fatty liver disease based on clustering analysis

    CN119296809A

  • Coronary heart disease risk prediction method based on sleep apnea syndrome

    CN119541869A

  • Risk assessment method and system for type 2 diabetes complications and storage medium

    CN120183705A

  • Non-invasive diabetic nephropathy grading and predicting method supporting convenient screening

    CN120452803A

  • Diabetes clinical test data intelligent clustering analysis system and method based on federal learning

    CN120656747A