Cardiovascular disease prediction method and system based on heart rate variability

By performing small segment division and gradient fusion on HRV data, the characteristics that reflect changes in different time periods are extracted, which solves the problem of low accuracy caused by ignoring time period differences in the prior art, and improves the accuracy of cardiovascular disease prediction.

CN119028577APending Publication Date: 2024-11-26SHENZHEN QIMAI TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310585957.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-23
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Existing HRV-based cardiovascular disease prediction methods ignore differences in heart rate changes in different time periods, resulting in lower prediction accuracy.

Method used

By dividing the N-N interval data within one day into small segments, calculating the equivalent value of each small segment, establishing a matrix S_all, and calculating the gradient in the X and Y directions, fusing the second-order statistical information, and extracting features to reflect the changes in HRV , establish a correlation with cardiovascular disease.

Benefits of technology

The accuracy of cardiovascular disease prediction based on HRV is improved, and the extracted characteristics are more correlated with disease occurrence.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028577A_ABST
    Figure CN119028577A_ABST
Patent Text Reader

Abstract

The invention provides a cardiovascular disease prediction method and system based on heart rate variability, and the method comprises the following steps: 1, dividing N-N interval data in a day into a plurality of small segments, and calculating the equivalent value of the N-N interval in each small segment; 2, representing the equivalent value of each small section of N-N interval through a matrix Sall, and setting the number of rows and the number of columns of the matrix; and step 3, gradients of the established matrix Sall in the X direction and the Y direction are calculated, and second-order statistical information is adopted to realize fusion of the gradient in the X direction and the gradient in the Y direction so as to reflect change conditions of the HRV in different time periods. The method has the beneficial effects that the extracted features have stronger correlation with the occurrence of the cardiovascular diseases, and the prediction accuracy of the cardiovascular diseases based on the HRV is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of health and medical technology, and in particular to a method and system for predicting cardiovascular diseases based on heart rate variability. Background Art

[0003] Heart rate variability (HRV) is one of the important indicators that reflects the tension of the sympathetic and parasympathetic nerves and the balance between them. It is a commonly used method to evaluate the activity of the cardiac autonomic nerves and an independent risk factor for predicting cardiovascular disease mortality. At present, the HRV-based methods mainly predict cardiovascular disease by analyzing the time domain and frequency domain characteristics of HRV. However, the features extracted based on HRV usually only consider the changes in heart rate at a specific time and a specific duration, ignoring the differences in heart rate changes in different time periods, resulting in low accuracy in cardiovascular disease prediction based on HRV. Summary of the invention

[0004] In view of the defects of the current cardiovascular disease prediction based on HRV, the present invention proposes a feature extraction method that can simultaneously reflect the changes of HRV within a specific duration and in different time periods. By utilizing these features, the accuracy of cardiovascular disease prediction based on HRV is improved.

[0005] A method for predicting cardiovascular disease based on heart rate variability comprises the following steps:

[0006] Step 1: Divide the NN interval data within a day into several small segments, and calculate the equivalent value of the NN interval in each small segment;

[0007] Step 2, the equivalent value of each small segment NN interval is represented by the matrix S_all, and the number of rows and columns of the matrix is ​​set;

[0008] Step 3: Calculate the gradient of the established matrix S_all in the X and Y directions, and fuse the gradients in the two directions. Use second-order statistical information to include the HRV changes in adjacent segments, reflecting the changes in HRV in different time periods.

[0009] Step 4: Create feature F feature Correlation with cardiovascular disease and prediction of cardiovascular disease.

[0010] In an embodiment of the present invention, a cardiovascular disease prediction system based on heart rate variability includes:

[0011] NN interval data acquisition module: obtains the user's NN interval data through wearable devices;

[0012] NN interval abnormal data processing module: pre-processes the acquired NN interval data, including the exclusion of abnormal data and automatic completion of missing data;

[0013] NN interval data matrix module: divide the NN interval data within a day into small segments of 1 minute each, and calculate the equivalent value of the NN interval in each small segment;

[0014] NN interval data gradient calculation module: calculate the gradient of the established matrix S_all in the X and Y directions, and fuse the gradients in the two directions to obtain the matrix G, which contains the HRV changes of each adjacent minute, and can also reflect the changes of HRV in different time periods.

[0015] NN interval data feature extraction module: extracts features from the matrix G so that each element in the matrix G has a different weight;

[0016] Cardiovascular disease prediction module: Establishing feature F feature Correlation with cardiovascular disease, thereby achieving the prediction of cardiovascular disease.

[0017] The beneficial effect of the present invention is that the extracted features have a stronger correlation with the occurrence of cardiovascular diseases, which is beneficial to improving the accuracy of cardiovascular disease prediction based on HRV. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a cardiovascular disease prediction system based on HRV and a flow chart in an embodiment of the present invention.

[0019] Figure 2 4 is a flow chart of a method for predicting cardiovascular disease in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0021] The present invention proposes a cardiovascular disease prediction system based on heart rate variability, such as Figure 1 As shown, it mainly includes six modules, namely, NN interval data acquisition module, NN interval abnormal data processing module, NN interval data matrix module, NN interval data gradient calculation module, NN interval data feature extraction module, and cardiovascular disease prediction module.

[0022] NN interval data acquisition module: obtains the user's NN interval data through wearable devices;

[0023] NN interval abnormal data processing module: pre-processes the acquired NN interval data, including the exclusion of abnormal data and automatic completion of missing data;

[0024] NN interval data matrix module: divide the NN interval data within a day into small segments of 1 minute each, and calculate the equivalent value of the NN interval in each small segment;

[0025] NN interval data gradient calculation module: calculate the gradient of the established matrix S_all in the X and Y directions, and fuse the gradients in the two directions to obtain the matrix G, which contains the HRV changes of each adjacent minute, and can also reflect the changes of HRV in different time periods.

[0026] NN interval data feature extraction module: extracts features from the matrix G so that each element in the matrix G has a different weight;

[0027] Cardiovascular disease prediction module: Establishing feature F feature Correlation with cardiovascular disease, thereby achieving the prediction of cardiovascular disease.

[0028] Based on the above system, a cardiovascular disease prediction method based on heart rate variability is also provided, comprising the following steps:

[0029] S1, first divide the NN interval data of a period of time into several small segments, and calculate the equivalent value of the NN interval in each small segment;

[0030] S2, the equivalent value of each small segment-N interval is represented by a matrix S_all, and the number of rows and columns of the matrix is ​​set;

[0031] S3, calculate the gradient of the established matrix S_all in the X direction and the Y direction, and fuse the gradients in the two directions, using second-order statistical information, including the HRV changes in adjacent segments, reflecting the changes in HRV in different time periods;

[0032] S4. Establish feature F feature Correlation with cardiovascular disease and prediction of cardiovascular disease.

[0033] like Figure 1 As shown, the main functions and implementation schemes of each module are as follows:

[0034] The NN interval data acquisition module obtains the user's NN interval data through the wearable device. The acquisition method is as follows:

[0035] First, the user wears a wearable device such as a smart watch or bracelet with ECG measurement function on the arm / wrist, or obtains ECG signals through an ECG patch on the chest.

[0036] After acquiring the ECG signal, the peak value of the ECG signal in each heartbeat cycle is located, that is, the second-order derivative of the ECG signal is derived, and then the derived signal is squared, and the position of the maximum value of the square of the second-order derivative is the peak position.

[0037] By subtracting the positions of two adjacent peaks, the NN interval data can be obtained. If the peak position of the ECG cannot be detected within 1 minute, a reminder message will be automatically sent to the user to determine whether the ECG sensor has fallen off or is not worn properly.

[0038] The NN interval abnormal data processing module pre-processes the acquired NN interval data, including the exclusion of abnormal data and the automatic completion of missing data.

[0039] The elimination of abnormal data is mainly based on the fusion method of Transformer and Gaussian mixture model.

[0040] Firstly, multiple basis detectors with different parameters based on the Gaussian mixture model principle are randomly generated by changing the connection structure of Transformer to obtain outliers of NN interval data and label outliers (i.e., manually calibrated outliers).

[0041] Then, the Manhattan distance and Euclidean distance between NN interval data are calculated by the nearest neighbor algorithm, and a local region is constructed around each interval data.

[0042] Finally, according to the similarity between the outliers determined by the algorithm and the label outliers, the determination results of the top three base detectors with the strongest detection ability in the area are selected for fusion, and the fused results are used as the final outliers. When a heart rate value is determined to be an outlier based on the autoencoder and Gaussian mixture model, the outlier is corrected by replacing it with the average of its two adjacent non-abnormal heart rates.

[0043] The automatic completion of missing data is mainly based on the iterative filling model of the Bayesian network. First, a Bayesian network is constructed to determine the topological relationship between random variables in the network. Second, a data set without missing data is established for training the Bayesian network.

[0044] After training, the model and parameter library will store the structural network describing the dependency relationship between variables and the conditional probability distribution parameter value of each variable. Again, based on the idea of ​​increasing the likelihood function to find the optimal parameters, the Bayesian network iteration is used to update the values ​​of the missing variables in turn to achieve automatic completion of missing data.

[0045] The NN interval data matrixization module divides the NN interval data within a day into small segments of 1 minute each, and calculates the equivalent value of the NN interval in each small segment.

[0046] Generally speaking, if people's heartbeat cycle is 0.7s, about 123428 NN interval data can be collected in one day (24 hours). Considering that the amount of NN interval data is too large, it may cause problems such as complex calculation and high cost. Before matrix processing the NN interval data, the NN interval data within a day is first divided into small segments of 1 minute, and the equivalent value of the NN interval in each small segment is calculated. The calculation formula of the equivalent value of the NN interval is shown as follows:

[0047] S=∑ω p ·a p

[0048] where a p is a NN interval data value within 1 minute, ω p is the weight ratio of the NN interval data value. The weight ratio is related to the frequency of occurrence of the NN interval data value. The higher the frequency, the greater the weight.

[0049] After calculating the equivalent value of the NN interval in each small segment (1 minute in length), a total of 60×24=1440 equivalent values ​​of NN intervals can be obtained in one day (24 hours). The equivalent values ​​of 1440 NN intervals are expressed as the following matrix:

[0050]

[0051] Among them (S 00,00 ,S 00,01 ...S 00,59 ) represents the NN interval value from 00:00 to 00:59 on the same day, with a total of 60 values, namely S 00,00 is the NN interval value at 00:00, S 00,01 is the NN interval value of 00:01, S 00,59 The NN interval value is 00:59; (S 01,00 ,S 01,01 ...S 01,59 ) represents the NN interval value from 01:00 to 01:59 on the same day, with a total of 60 values, and so on. Therefore, the default number of rows M of the matrix S_all is 24, and the default number of columns N is 60. The number of rows M and the number of columns N of the matrix can be set according to actual needs.

[0052] The NN interval data gradient calculation module calculates the gradients of the established matrix S_all in the X and Y directions and fuses the gradients in the two directions.

[0053] The calculation process of the second-order statistics is as follows:

[0054] The gradient G of the matrix S_all in the X direction x The calculation method is as follows: G x For a 24×60 matrix:

[0055]

[0056] Gradient calculation method G of matrix S_all in the Y direction y As shown below, G y For an m×n matrix:

[0057]

[0058] In the process of calculating the gradient of the matrix S_all in the X direction and the Y direction, only the first-order statistical information is used. In order to further obtain features, the present invention proposes to use the second-order statistical information, which is intended to use the difference of the second-order information to achieve feature differentiation when the first-order information is the same. The calculation process is as follows:

[0059] but:

[0060]

[0061] in is the matrix G y The transposed matrix of Represents the matrix G x and matrix Perform a cross multiplication operation. G is the matrix after the cross multiplication operation, which is a 24×24 matrix. 00,00 Represents the (00,00)th element of the matrix G, whose size is:

[0062] G 00,00 =(S 00,01 -S 00,00 )×(S 01,00 -S 00,00 )+(S 00,02 -S 00,00 ) / 2×(S 01,01 -S 00,01 )+(S 00,03 -S 00,01 ) / 2×(S 01,02 -S 00,02 )+…+(S 00,59 -S 00,58 )×(S 01,59 -S00,59 ).

[0063] The calculation method of other elements of matrix G is similar. From the expression of matrix G, it can be seen that it contains the HRV changes of each adjacent 1 minute, and can also reflect the changes of HRV in different time periods.

[0064] The NN interval data feature extraction module extracts features from the matrix G. The feature extraction method is shown in the following formula:

[0065]

[0066] where α 1 is the element G in the matrix 00,00 The weight factor, α 2 is the element G in the matrix 00,01 The weight factor of α makes each element in the matrix G have a different weight, and so on. 1 , α 2 The initial value of the equal weight factor can be set to 1, which can be optimized by particle swarm algorithm, fish swarm algorithm, etc., so that each element in the matrix G has a different weight.

[0067] Cardiovascular disease prediction module, establish feature F feature The correlation with cardiovascular disease can achieve the prediction of cardiovascular disease. The specific calculation method is as follows: by using wearable devices such as smart watches, bracelets, and ECG stickers to obtain the NN interval data of multiple patients with cardiovascular disease and healthy people in one day, the feature F is extracted based on these data. feature , analysis feature F feature Correlation with cardiovascular disease, features that clearly diagnose cardiovascular disease F feature On this basis, we can obtain relevant information such as the user's age, fasting blood sugar, blood pressure, glycosylated hemoglobin, daily exercise, daily sleep quality, etc., and combine the extracted features F feature , a random forest-based algorithm was used for fusion analysis to further improve the sensitivity and specificity of cardiovascular disease prediction.

[0068] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cardiovascular disease prediction method based on heart rate variability, It is characterized in that The steps include: Step 1, divide the NN interval data within a period of time into several small segments, and calculate the equivalent value of the NN interval in each small segment; Step 2, the equivalent value of each small segment NN interval is represented by the matrix S_all, and the number of rows and columns of the matrix is ​​set; Step 3: Calculate the gradients of the established matrix S_all in the X direction and the Y direction, and use the second-order statistical information to achieve the fusion of the gradients in the X direction and the Y direction in the two directions to reflect the changes in HRV in different time periods; Step 4: Create feature F feature Correlation with cardiovascular disease and prediction of cardiovascular disease.

2. The cardiovascular disease prediction method according to claim 1, It is characterized in that The equivalent value calculation formula of the NN interval is as follows: S=∑ω p ·a p where a p is a NN interval data value within 1 minute, ω p is the weight ratio of the NN interval data value. The weight ratio is related to the frequency of occurrence of the NN interval data value. The higher the frequency, the greater the weight.

3. The cardiovascular disease prediction method according to claim 1, It is characterized in that The matrix S_all is expressed as follows: Among them (S 00,00 S 00,01 S 00,02 ……S 00,57 S 00,58 S 00,59 ) represents the NN interval value from 00:00 to 00:59 on the same day, with a total of 60 values, namely S 00,00 is the NN interval value at 00:00, S 00,01 is the NN interval value of 00:01, S 00,59 The NN interval value is 00:

59.

4. The cardiovascular disease prediction method according to claim 3, It is characterized in that The calculation process of the second-order statistics is as follows: The gradient G of the matrix S_all in the X direction x The calculation method is as follows: G x A 24×60 matrix Gradient calculation method G of matrix S_all in the Y direction y As shown below, G y For a 24×60 matrix: but: in is the matrix G y The transposed matrix of Represents the matrix G x and matrix Perform a cross multiplication operation. G is the matrix after the cross multiplication operation, which is a 24×24 matrix. 00,00 Represents the (00,00)th element of the matrix G, whose size is: G 00,00 =(S 00,01 -S 00,00 )×(S 01,00 -S 00,00 )+(S 00,02 -S 00,00 ) / 2×(S 01,01 -S 00,01 )+(S 00,03 -S 00,01 ) / 2×(S 01,02 -S 00,02 )+…+(S 00,59 -S 00,58 )×(S 01,59 -S 00,59 )。 5. The method for predicting cardiovascular disease according to claim 4, It is characterized in that Extract features from the matrix G. The feature extraction method is shown in the following formula: where α 1 is the element G in the matrix 00,00 The weight factor, α 2 is the element G in the matrix 00,01 The weight factor makes each element in the matrix G have a different weight.

6. The method for predicting cardiovascular disease according to claim 5, It is characterized in that Establish feature F feature Correlation with cardiovascular disease, so as to predict cardiovascular disease. The specific calculation method is as follows: Step 1: Obtain NN interval data of multiple patients with cardiovascular diseases and healthy people in one day, and extract feature F based on these data feature , analysis feature F feature Correlation with cardiovascular disease, features that clearly diagnose cardiovascular disease F feature Best entry point.

7. The method for predicting cardiovascular disease according to claim 6, It is characterized in that By obtaining the user's age, fasting blood sugar, blood pressure, glycosylated hemoglobin, daily exercise, and daily sleep quality, combined with the extracted feature F feature , a random forest-based algorithm was used for fusion analysis to improve the sensitivity and specificity of cardiovascular disease prediction.

8. A cardiovascular disease prediction system based on heart rate variability, It is characterized in that include: NN interval data acquisition module: obtains the user's NN interval data through wearable devices; NN interval abnormal data processing module: pre-processes the acquired NN interval data, including the exclusion of abnormal data and automatic completion of missing data; NN interval data matrix module: divide the NN interval data within a period of time into several small segments, and calculate the equivalent value of the NN interval in each small segment; NN interval data gradient calculation module: calculate the gradient of the established matrix S_all in the X and Y directions, and fuse the gradients in the two directions to obtain the matrix G, which contains the HRV changes of each adjacent minute, and can also reflect the changes of HRV in different time periods. NN interval data feature extraction module: extracts features from the matrix G so that each element in the matrix G has a different weight; Cardiovascular disease prediction module: Establishing feature F feature Correlation with cardiovascular disease, thereby achieving the prediction of cardiovascular disease.