Multiple Time Scales Of Heart Rate Variability And Application In Diagnosis Of Heart Failure

With the development of social economy,the incidence of cardiovascular diseases has increased year by year and the population with the diseases has become younger.At present,the number of patients with cardiovascular diseases in China has increased dramatically and reached 290 million.Cardiovascular diseases have become the number-one killer of human health and life.Heart failure(HF),as a serious and terminal stage of various cardiovascular diseases,is characterized by high prevalence,high mortality and high medical expenditure.It is one of the most important cardiovascular diseases.This study aims to establish an early diagnosis model of heart failure with digital signal processing technology,computer technology and machine learning algorithms based on physiological signals.The early diagnosis model is hopeful to provide a basis for early health screening of heart failure disease and switch the traditional treatment model to prevention model.It is efficient to the early prevention of deterioration of patients with heart failure and greatly reduction of burden on families and medical institutions.In this study,the heartbeat interval time series of long-term ECG signals in the PhysioNet database were analyzed with heart rate variability(HRV)analysis and multiple time scales analysis.Nine HRV indices and seven types of time scale were selected in this study.Nine indices include three time-domain indices:MEAN,SDNN,RMSSD,three frequency-domain indices:LFn,HFn,Ratio-LH,and three nonlinear-domain indices:VAI,VLI,SampEn.The seven types of time scale include 5 minutes,10 minutes,30 minutes,1 hour,2 hours,5 hours and 10 hours.The main research contents and results are as follows:1)The study on differences in HRV indices.We compared the difference of the same HRV index in the heart failure sample and the normal sample at the same time scale,compared the difference of the same HRV index in the same sample at different time scales,and compared the difference of the same HRV index between the two groups of samples.The student's t-test was used to analyze whether there was a statistically significant difference in the same HRV index at the same time scale.The results showed that:MEAN,RMSSD,LFn,HFn and VAI are not sensitive to time scales,while SDNN,Ratio-LH,VLI and SampEn exhibit different sensitivity to time scales.Five HRV indices that are insensitive to time scales are also insensitive to the time scale between the two sets of samples.The time scale has an amplification effect on the difference between the two sets of samples for both SDNN and VLI.It showed a scaling effect on short-term time scales and an amplification effect on long-term time scales for SampEn.2)Research on heart failure diagnosis model of HRV indices at single time scale.The nine HRV indices at the same time scale were used as feature vectors,and the grid search algorithm used to determine the optimal hyper-parameters.The support vector machine learning algorithm was used to establish the heart failure diagnosis model.The ten-fold cross-validation method was used to evaluate the classification performance of the models.The results showed that:The heart failure diagnosis model established at the 2-hour time scale has the best performance with sensitivity of 86.67%,specificity of 98.33%,and accuracy of 94.44%.The three evaluation measurements are better than those of other time scales.It suggests that the 2-hour time scale may be more suitable for the diagnostic analysis of patients with heart failure.3)Research on heart failure diagnosis model of HRV indices at multiple time scales.According to the trends of HRV indices in the same group of samples,three fitting functions including linear function,exponential function and logarithmic function used to fit the trends.The coefficient of the fitting function used as new non-standard HRV indices,and the support vector machine learning algorithm also used to establish the heart failure diagnosis model.The results showed that:The performances of the heart failure diagnosis models based on the non-standard HRV indices are all better than the best performance on the single time scale.By introducing only three non-standard HRV indices,it achieved a best performance of a sensitivity of 93.33%,specificity of 98.33%and an accuracy of 96.67%.Multiple time scale analysis can provide more useful information for heart failure diagnosis,and provide a useful reference to other physiological signal analysis.