HRV Feature Selection Based On Multi-time Scale Analysis And Application In Heart Failure Diagnosis

The number of people suffering from cardiovascular diseases has increased sharply,and the number of people suffering from cardiovascular diseases in China is about 330 million.Heart failure is a kind of cardiovascular disease with high mortality,difficult to cure and expensive treatment cost,which is widely concerned by clinicians and related experts.In this paper,digital signal processing,HRV analysis,multi-time scale analysis,machine learning algorithm and other technologies are used to carry out correlation analysis on physiological signals with the help of computers,and to build a heart failure diagnosis model based on long-range ECG signals,hoping to provide help for the early diagnosis of heart failure with ECG wearers.In this paper,the long-term RR interval time series in MIT/BIH database is used as the data source.A new feature selection scheme is proposed by using multi-time scale analysis and HRV analysis method,and multi-time scale feature matrix is constructed by using it.And then build the support vector machine algorithm of heart failure diagnosis model.The main contents of this paper are as follows:1)Study on multi-time scale HRV feature selection scheme.HRV features of patients with heart failure and healthy subjects were calculated at different time scales.This paper respectively from the selection of time domain and frequency domain,and nonlinear domain the three commonly used features,a total of nine features of HRV scheme as optional features.Significant differences between candidate HRV characteristics in heart failure samples and normal samples were calculated.Choose features that make a significant difference.Then,the multi-time scale analysis method was used to select the HRV features with no trend change on different time scales.As the applicable HRV features of the feature selection scheme in this paper.Finally,a multi-time scale feature matrix is constructed using the applicable HRV features.2)Study on diagnosis model of heart failure based on multi-time scale feature matrix.Firstly,the diagnostic models of heart failure were constructed by using the single-time scale HRV feature matrix and the multi-time scale HRV feature matrix respectively for comparative analysis.The classification effects of the models constructed by different single-time scale feature matrices and multi-time scale feature matrices were compared.It is found that the self-diagnosis model of heart failure constructed by multi-time scale feature matrix is better than single time scale in performance evaluation.3)In-depth analysis of multi-time scale diagnostic models for heart failure.Firstly,the time scale in the multi-time scale analysis method is analyzed in detail.Determine the optimal multi-time scale selection scheme.Then,the HRV features selected in the feature selection scheme are further analyzed for rationality.The stability and difference were analyzed respectively.Finally,the multi-time scale features were used to construct the heart failure diagnosis model and two kinds of heart failure diagnosis models,namely the index stack model and SMOTE model.The effectiveness of the proposed method is verified by performance comparison.Through in-depth analysis and verification,the new feature selection scheme proposed in this paper and its application in the field of physiological signal processing provide reference significance.