Diabetes cardiovascular disease risk assessment feature acquisition method based on multi-modal data fusion

Abstract

The invention provides a diabetic cardiovascular disease risk assessment feature acquisition method based on multi-modal data fusion, and belongs to cardiovascular disease risk assessment, and the method comprises the steps: obtaining data in real time, obtaining a plurality of extreme value trend variations, and fitting and constructing a plurality of trend variation curves of physiological indexes and functional indexes; predicting a trend variation curve of the biochemical indexes through a biochemical index trend variation prediction model according to the plurality of trend variation curves of the physiological indexes and the functional indexes; and extracting trend characteristics of the trend variation curves of the physiological indexes, the functional indexes and the biochemical indexes, and performing weighted fusion to obtain fusion characteristics. According to the method, the change trend of biochemical indexes is predicted through historical detection data in combination with physiological indexes and functional indexes obtained in real time, and fusion features for risk assessment of diabetic cardiovascular diseases are obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to cardiovascular disease risk assessment, in particular to a multi-modal data fusion diabetes cardiovascular disease risk assessment feature acquisition method, system and medium. BACKGROUND

[0002] Diabetes complicated by cardiovascular disease is one of the most serious chronic complications of diabetes, mainly manifested as coronary atherosclerotic heart disease, cardiomyopathy, arrhythmia, heart failure, and stroke (ischemic and hemorrhagic), etc., and its pathological mechanism is closely related to vascular endothelial damage caused by long-term hyperglycemia, metabolic disorders caused by insulin resistance, and chronic inflammatory response. Due to factors such as elevated glycated hemoglobin, dyslipidemia (such as low-density lipoprotein oxidation), and hypercoagulable state, the incidence of cardiovascular disease in diabetic patients is 2-4 times that of non-diabetic patients, and the age of onset is earlier and the condition is more dangerous. Risk assessment has a core value in the prevention and control of diabetes cardiovascular disease. First, by quantitatively assessing key indicators such as blood glucose fluctuation, blood pressure variability, dyslipidemia, and renal function damage, high-risk groups can be identified early, and the transition from "passive treatment" to "active prevention" can be achieved. For example, dynamic blood glucose monitoring combined with carotid ultrasound examination can timely detect asymptomatic myocardial ischemia and vascular plaque formation. Second, risk assessment provides a basis for individualized treatment, such as precise selection of hypoglycemic drugs (SGLT-2 inhibitors are preferred for patients with heart failure) and adjustment of statin doses based on the patient's cardiovascular risk stratification (low risk, medium risk, high risk). In addition, regular assessment helps monitor the effectiveness of interventions, and by comparing indicators such as vascular elasticity index and inflammatory factor levels before and after treatment, treatment plans can be dynamically optimized. From a public health perspective, establishing a comprehensive risk assessment system can significantly reduce medical costs. Therefore, incorporating risk assessment into routine diabetes management is not only a key support for clinical decision-making, but also a strategic initiative to improve patient quality of life and reduce the social burden of disease.

[0003] In the prior art, there are many reference physiological data and biochemical and functional indicators for characterizing diabetes cardiovascular disease. Physiological data and functional indicators can be easily obtained through wearable devices or simple detection devices, but biochemical indicators can only be obtained through regular laboratory analysis and detection, and it is not possible to dynamically analyze the trend of changes in biochemical indicators in high-risk patients in real time. SUMMARY

[0004] To solve the above problems, the present application provides a multi-modal data fusion diabetes cardiovascular disease risk assessment feature acquisition method, which predicts biochemical indicators by combining historical detection data with real-time physiological indicators and functional indicators, thereby achieving real-time dynamic analysis of the trend of changes in biochemical indicators in high-risk patients.

[0005] To achieve the above purpose, the present application provides the following technical solutions.

[0006] The method for acquiring the risk assessment features of diabetic cardiovascular disease through multi-modal data fusion comprises the following steps: Obtain multi-class physiological indicators, functional indicators and biochemical indicators related to diabetic cardiovascular disease in a fixed time period; traverse the multi-class physiological indicators, functional indicators and biochemical indicators to determine a plurality of dynamic change feature segments with extreme value change rates exceeding a preset threshold. Determine the overlap degree of the time sequence segments corresponding to the plurality of dynamic change feature segments of the physiological indicator and functional indicator data and the biochemical indicator data, filter out physiological indicators and functional indicators of types below a preset threshold, and determine associated physiological indicator data and functional indicator data related to the biochemical indicators. Extract the extreme value trend change amount of the multi-class associated physiological indicators and functional indicators and the biochemical indicators, and respectively fit and construct a plurality of trend change amount curves to obtain a biochemical indicator trend change amount curve prediction model through training of an LSTM model. Real-time acquisition of physiological indicator and functional indicator data, extraction of a plurality of extreme value trend change amounts of the associated physiological indicator data and functional indicator data, fitting and construction of a plurality of trend change amount curves, and prediction of the trend change amount curve of the biochemical indicators through the biochemical indicator trend change amount curve prediction model. Extract the trend features of the trend change amount curves of the physiological indicators, functional indicators and biochemical indicators, and obtain a fusion feature for diabetic cardiovascular disease risk assessment through weighted fusion.

[0007] Preferably, when the multi-class physiological indicators, functional indicators and biochemical indicators are a plurality of discrete data points, a polynomial fitting method is used to convert the discrete data points into smooth time sequence curves.

[0008] Preferably, the step of traversing the multi-class physiological indicators, functional indicators and biochemical indicators through a fixed window to determine a plurality of dynamic change feature segments with extreme value change rates exceeding a preset threshold comprises the following steps: Standardize the time sequence data of the physiological indicators, functional indicators and biochemical indicators; Set the size of the sliding window, and calculate the extreme value change rate in each window: In the formula, and are the local maximum and minimum values of the data in each window; and are the extreme value change rates; When or , the window is marked as a dynamic change feature segment. The threshold for the rate of change of extreme values; Based on the window start time that satisfies the extreme value change rate condition and end time Forming a time series set T : .

[0009] Preferably, the step of determining the overlap between time segments corresponding to multiple dynamic change characteristic segments of physiological and functional indicator data and biochemical indicator data, and filtering out physiological and functional indicator types below a preset threshold, includes the following steps: Determine the time series set of each physiological or functional indicator A and biochemical indicator B. and Time sequence overlap: like , If the threshold for overlap of time segments is used, then physiological or functional indicators A will be screened out.

[0010] Preferably, the construction of the prediction model for the trend change curve of the biochemical indicators includes the following steps: Normalized and biochemical correlation data of historical physiological and functional indicators, as well as historical biochemical indicators; Based on normalized historical physiological, functional, and biochemical data, multiple extreme value trend changes in the time series are obtained, and multiple trend change curves are constructed by polynomial linear fitting. Construct an LSTM model; By taking the trend change curves of multiple physiological and functional indicators associated with related biochemical indicators as input and the trend change curves of biochemical indicators as output, a predictive model for the trend change curves of biochemical indicators is trained.

[0011] Preferably, the step of obtaining multiple extreme value trend changes in the time series and constructing multiple trend change curves using polynomial linear fitting includes the following steps: For the time series data of each historical physiological, functional and biochemical indicator after normalization, a fixed sliding window is used to detect local maximum and extreme points, and the timestamp and corresponding value of the extreme points are determined. For adjacent extreme points and Calculate the change: Generate extreme value trend change series ; Based on the extreme value trend change sequence, a polynomial fitting method is used, and the coefficients are solved by the least squares method to obtain the trend change curve.

[0012] Preferably, the step of extracting the trend characteristics of the trend change curves of physiological indicators, functional indicators, and biochemical indicators, and performing weighted fusion to obtain fusion features for cardiovascular disease risk assessment in diabetes, includes the following steps: The trend slope, mean stability, and fluctuation amplitude of the trend change curves of each physiological, functional, and biochemical indicator are extracted and transformed into three-dimensional feature vectors to construct multiple feature matrices. Multiple parallel attention heads are set up, each head independently calculates the feature weights of the feature matrix of each indicator, and then they are fused to obtain the fused features.

[0013] This invention also provides a system for acquiring features for assessing cardiovascular disease risk in diabetes through multimodal data fusion, the system comprising: processor; A memory on which computer programs that can run on the processor are stored; The computer program, when executed by a processor, implements the steps of the method for obtaining features for assessing the cardiovascular risk of diabetes through multimodal data fusion.

[0014] The present invention also provides a computer-readable storage medium storing a data processing program, wherein when the data processing program is executed by a processor, it implements the steps of the method for obtaining features for multimodal data fusion-based assessment of cardiovascular disease risk in diabetes.

[0015] The beneficial effects of this invention are: This invention proposes a method for acquiring features for assessing cardiovascular disease risk in diabetes through multimodal data fusion. This method predicts the changing trends of biochemical indicators by combining historical test data with real-time acquired physiological and functional indicators, accurately obtaining the intrinsic correlation between physiological and functional indicators and biochemical indicators, and solving the problem of the difficulty in obtaining biochemical indicators in real time. This invention integrates the changing trend characteristics of multimodal indicator data, further combining subtle changes in multiple indicators, using the changing trends as the basis for analysis. This accurately captures the physiological changes of the assessment subject, rather than just information on the surface of the data, and can serve as a basis for assessing the dynamic risk of cardiovascular disease in diabetes. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the screening process for physiological and functional indicator data according to an embodiment of the present invention. Figure 3This is a flowchart illustrating the construction process of the biochemical indicator trend change prediction model according to an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Example 1 This embodiment proposes a method for obtaining features for diabetic cardiovascular disease risk assessment through multimodal data fusion. The specific steps are as follows: Figure 1 As shown, it includes: S1: Obtain historical data of multiple physiological, functional, and biochemical indicators related to diabetic cardiovascular disease within the same fixed time period; where the historical data of multiple physiological, functional, and biochemical indicators are multiple discrete data points, a polynomial fitting method is used to transform the discrete data points into smooth time-series curves.

[0019] S2: By traversing historical data of various physiological, functional, and biochemical indicators through a fixed window, identify multiple dynamic change characteristic segments where the rate of extreme value change exceeds a preset threshold.

[0020] S3: Determine the overlap of time segments corresponding to multiple dynamic change characteristic segments of physiological and functional indicator data and biochemical indicator data, filter out physiological and functional indicator types below the preset threshold, and determine the associated physiological and functional indicator data related to biochemical indicators.

[0021] S4: Extract historical data of various related physiological and functional indicators, as well as the extreme value trend changes of historical biochemical indicators, and fit and construct multiple trend change curves respectively. Then, obtain a prediction model for the trend change curves of biochemical indicators by training an LSTM model.

[0022] S5: Real-time acquisition of physiological and functional indicator data, extraction of multiple extreme value trend changes of related physiological and functional indicator data, fitting and constructing multiple trend change curves, and prediction of biochemical indicator trend change curves through a biochemical indicator trend change curve prediction model.

[0023] S6: Extract the trend characteristics of the trend change curves of physiological indicators, functional indicators and biochemical indicators, and perform weighted fusion to obtain fusion characteristics for cardiovascular risk assessment of diabetes.

[0024] The specific physiological, functional, and biochemical indicator data for which this invention is applied are as follows: (1) Physiological indicators: Blood pressure: Monitor systolic and diastolic blood pressure, with a target of <130 / 80 mmHg.

[0025] Heart rate and blood oxygen saturation: Continuous monitoring via wearable devices to detect arrhythmias or hypoxemia in a timely manner.

[0026] Height: Measure height regularly and calculate BMI based on weight to assess growth and development or obesity risk.

[0027] Weight: Weigh yourself regularly to monitor weight changes.

[0028] Waist circumference: Measured horizontally around the navel using a soft measuring tape. The target for men is <90cm and for women is <85cm (Chinese population standard).

[0029] BMI: Calculated by dividing weight (kg) by height (m) squared. The target is usually BMI 18.5-24.9 kg / m² (reference range for Asian populations). Overweight or obesity requires a comprehensive assessment of metabolic risk in conjunction with waist circumference.

[0030] (2) Cardiovascular function indicators Electrocardiogram (ECG): Real-time detection of arrhythmias, myocardial ischemia and other abnormalities, which can be achieved through a portable ECG monitor or smartwatch.

[0031] (3) Biochemical indicators Blood glucose: including fasting blood glucose, postprandial blood glucose, and glycated hemoglobin (HbA1c), reflects short-term and long-term blood glucose control.

[0032] Blood lipids: including total cholesterol, low-density lipoprotein (LDL-C), high-density lipoprotein (HDL-C), and triglycerides, should be tested in the laboratory regularly (every 3-6 months).

[0033] Kidney function indicators, such as urinary microalbumin and serum creatinine, are used to assess the cardiovascular risk of diabetes.

[0034] Physiological and functional indicators can reflect the changing trends of biochemical data to a certain extent, especially in long-term health monitoring and disease prediction, where the correlation between the two is crucial. Physiological data (such as weight, heart rate, and blood pressure) and cardiovascular function indicators (such as electrocardiograms) are closely related to biochemical data (such as blood glucose, blood lipids, and liver and kidney function indicators). Analyzing these relationships can help identify potential trends in biochemical data. Therefore, this invention provides a method for trend analysis of biochemical data using physiological and functional indicators.

[0035] Specifically, the first step is to perform correlation analysis between the data, and then screen the physiological and functional indicator data that are related to the biochemical indicators, as in steps S2-S3. Figure 2As shown, specifically: S2.1: Obtain time-series data of various historical physiological indicators, functional indicators, and biochemical indicators related to diabetes and cardiovascular disease within the same fixed time period.

[0036] S2.2: By traversing physiological time-series data, functional indicator time-series data, and biochemical indicator time-series data through a fixed window, determine multiple dynamic change characteristic segments and their corresponding time segments where the extreme value change rate exceeds a preset threshold.

[0037] S2.3: In the time series data of physiological indicators and functional indicators, the physiological and functional indicators whose time series data have multiple dynamic change feature segments and the time series data of biochemical indicators have an overlap of less than a preset threshold are screened out, and the physiological and functional indicators associated with biochemical indicators are determined.

[0038] Among them, the dynamic change characteristic segment is determined by the extreme value change rate: Set the sliding window size and calculate the rate of change of extreme values ​​within each window: In the formula, and For the data within each window Local maximum and minimum values; and The rate of change of the extreme value; when ,or Then mark the window as a dynamically changing feature segment; The threshold for the rate of change of extreme values; Based on the window start time that satisfies the extreme value change rate condition and end time Forming a time series set T : .

[0039] Filtering based on time sequence overlap: Determine the time series set of each physiological or functional indicator A and biochemical indicator B. and Time sequence overlap: like , If the threshold for overlap of time segments is used, then physiological or functional indicators A will be screened out.

[0040] Specifically, the construction of a predictive model for the trend changes of biochemical indicators from S4 to S5, such as... Figure 3As shown, it includes the following steps: S4.1: Normalized and correlated historical physiological and functional data with biochemical indicators, as well as historical biochemical data.

[0041] S4.2: Based on the normalized historical physiological, functional, and biochemical data, obtain multiple extreme value trend changes in the time series, and construct multiple trend change curves using polynomial linear fitting.

[0042] S4.3: Construct an LSTM model; take the trend change curves of multiple physiological and functional indicators associated with the relevant biochemical indicators as input, and the trend change curves of the biochemical indicators as output, and train to obtain a prediction model for the trend change curves of biochemical indicators.

[0043] The process of obtaining multiple extreme value trend changes in a time series and constructing multiple trend change curves using polynomial linear fitting includes the following steps: For the time series data of each historical physiological, functional and biochemical indicator after normalization, a fixed sliding window is used to detect local maximum and extreme points, and the timestamp and corresponding value of the extreme points are determined. For adjacent extreme points and Calculate the change: Generate extreme value trend change series ; Based on the extreme value trend change sequence, a polynomial fitting method is used, and the coefficients are solved by the least squares method to obtain the trend change curve.

[0044] This invention employs a neural network model for risk assessment, serving as a reference for healthcare professionals. A training set is constructed using historical test and laboratory data, and the risk level corresponding to each indicator in the training set is confirmed through expert risk assessment. Notably, the biochemical indicator trend prediction model and the diabetes cardiovascular disease risk assessment model for each assessment subject are unique, not general prediction models, and require prior training or fitting before use.

[0045] During dynamic assessment, physiological and functional indicators associated with biochemical indicators are acquired in real time for the assessed object. For discrete data points, linear fitting is required. Trend curves of physiological and functional indicators are extracted from the data to predict the trend changes of biochemical indicators. The trend characteristics of the curves representing the changes in physiological, functional, and biochemical indicators are then fused using an adaptive weighted fusion method based on a multi-head attention mechanism to obtain fusion features. The risk assessment result is then obtained through a pre-trained diabetes cardiovascular disease risk assessment model.

[0046] The basic model architecture in this embodiment is as follows: Multiple convolutional layers are used to extract the trend slope, mean stability, and fluctuation amplitude of the trend change curves for each physiological, functional, and biochemical indicator, and these are converted into three-dimensional feature vectors to construct multiple feature matrices. Multiple parallel attention heads are set up, each independently calculating the feature weights of the feature matrix for each indicator, and these are then fused to obtain the fused features. A fully connected layer maps the fused features to the dynamic risk probability output of diabetic cardiovascular disease. Similarly, other neural network models can be used, which will not be elaborated upon here.

[0047] The above is one embodiment of the method for obtaining features for diabetic cardiovascular disease risk assessment based on multimodal data fusion. Based on the same idea, this embodiment also provides a corresponding dynamic risk assessment system for diabetic cardiovascular disease based on data fusion. Each module in the above-mentioned dynamic risk assessment system for diabetic cardiovascular disease based on multimodal data fusion can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0048] This embodiment also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The provided method is a feature acquisition method for diabetic cardiovascular disease risk assessment based on multimodal data fusion.

[0049] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0050] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for obtaining features for assessing cardiovascular disease risk in diabetes through multimodal data fusion, characterized in that, Includes the following steps: Acquire historical data of various physiological, functional, and biochemical indicators related to diabetic cardiovascular disease within the same fixed time period; traverse the historical data of various physiological, functional, and biochemical indicators through a fixed window to identify multiple dynamic change characteristic segments where the rate of extreme value change exceeds a preset threshold. Determine the overlap of time segments corresponding to multiple dynamic change characteristic segments of physiological and functional indicators and biochemical indicators, screen out physiological and functional indicators below a preset threshold, and identify associated physiological and functional indicators related to biochemical indicators. Historical data of various related physiological and functional indicators, as well as historical data of biochemical indicators, were extracted. The extreme value trend changes were then fitted and constructed into multiple trend change curves. An LSTM model was trained to obtain a prediction model for the trend change curves of biochemical indicators. Real-time acquisition of physiological and functional indicator data, extraction of multiple extreme value trend changes of related physiological and functional indicator data, fitting and constructing multiple trend change curves, and prediction of biochemical indicator trend change curves through a biochemical indicator trend change curve prediction model. The trend characteristics of the trend change curves of physiological, functional and biochemical indicators are extracted and weighted to obtain fusion characteristics for cardiovascular risk assessment of diabetes.

2. The method for obtaining features for diabetic cardiovascular disease risk assessment based on multimodal data fusion according to claim 1, characterized in that, When the historical data of the various physiological, functional, and biochemical indicators are multiple discrete data points, a polynomial fitting method is used to transform the discrete data points into smooth time-series curves.

3. The method for obtaining features for diabetic cardiovascular disease risk assessment based on multimodal data fusion according to claim 1, characterized in that, The step of identifying multiple dynamic change characteristic segments where the rate of extreme value change exceeds a preset threshold by traversing historical data of various physiological, functional, and biochemical indicators through a fixed window includes the following steps: Standardize time-series data of physiological, functional, and biochemical indicators; Set the sliding window size and calculate the rate of change of extreme values ​​within each window: In the formula, and For the data within each window Local maximum and minimum values; and The rate of change of extreme values; when ,or Then mark the window as a dynamically changing feature segment; The threshold for the rate of change of extreme values; Based on the window start time that satisfies the extreme value change rate condition and end time Forming a time series set T : 。 4. The method for obtaining features for diabetic cardiovascular disease risk assessment based on multimodal data fusion according to claim 1, characterized in that, The process of determining the overlap between time segments corresponding to multiple dynamic change characteristic segments of physiological and functional indicator data and biochemical indicator data, and filtering out physiological and functional indicator types below a preset threshold, includes the following steps: Determine the time series set of each physiological or functional indicator A and biochemical indicator B. and Time sequence overlap: like , If the threshold for overlap of time segments is used, then physiological or functional indicators A will be screened out.

5. The method for obtaining features for diabetic cardiovascular disease risk assessment based on multimodal data fusion according to claim 1, characterized in that, The construction of the prediction model for the trend change curve of the biochemical indicators includes the following steps: Normalized and biochemical correlation data of historical physiological and functional indicators, as well as historical biochemical indicators; Based on normalized historical physiological, functional, and biochemical data, multiple extreme value trend changes in the time series are obtained, and multiple trend change curves are constructed by polynomial linear fitting. Construct an LSTM model; By taking the trend change curves of multiple physiological and functional indicators associated with related biochemical indicators as input and the trend change curves of biochemical indicators as output, a predictive model for the trend change curves of biochemical indicators is trained.

6. The method for obtaining features for diabetic cardiovascular disease risk assessment based on multimodal data fusion according to claim 5, characterized in that, The process of obtaining multiple extreme value trend changes in a time series and constructing multiple trend change curves using polynomial linear fitting includes the following steps: For the time series data of each historical physiological, functional and biochemical indicator after normalization, a fixed sliding window is used to detect local maximum and extreme points, and the timestamp and corresponding value of the extreme points are determined. For adjacent extreme points and Calculate the change: Generate extreme value trend change series ; Based on the extreme value trend change sequence, a polynomial fitting method is used, and the coefficients are solved by the least squares method to obtain the trend change curve.

7. The method for obtaining features for diabetic cardiovascular disease risk assessment based on multimodal data fusion according to claim 1, characterized in that, The process of extracting trend characteristics from the trend change curves of physiological, functional, and biochemical indicators, and then performing weighted fusion to obtain fusion features for assessing cardiovascular risk in diabetes, includes the following steps: The trend slope, mean stability, and fluctuation amplitude of the trend change curves of each physiological, functional, and biochemical indicator are extracted and transformed into three-dimensional feature vectors to construct multiple feature matrices. Multiple parallel attention heads are set up, each head independently calculates the feature weights of the feature matrix of each indicator, and then they are fused to obtain the fused features.

8. A system for acquiring features for assessing cardiovascular disease risk in diabetes through multimodal data fusion, characterized in that, The system includes: processor; A memory on which computer programs that can run on the processor are stored; When the computer program is executed by the processor, it implements the steps of the method for obtaining features for multimodal data fusion in diabetic cardiovascular disease risk assessment as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a data processing program, which, when executed by a processor, implements the steps of the method for obtaining features for multimodal data fusion-based cardiovascular risk assessment of diabetes as described in any one of claims 1 to 7.

Images