Applications Of HRV Analysis On Heart Failure Diagnosis And Pain Detection For Newborns

Abstract:There are fluctuations between consecutive sinus heartbeats, and this phenomenon is known as heart rate variability (HRV). HRV in which a large amount of physiological and pathological information is contained is an important mark for the evaluation of autonomic nervous system. HRV has applications in a broad field including disease diagnosis, emotion identification, mental workload assessment and et al. This study presents researches on problems related to the acquirement, analysis and applications of HRV signals, and the main research contents are as follows:1) An inspection policy for R-waves detection algorithm which is based on the jump of modulus maxima sequence in wavelet coefficient is proposed, and detection of R-waves using continuous wavelet transform is realized. The complex Morlet wavelet and Mexican-hat wavelet are employed to transform the ECG signal, and according to the trait that the modulus maxima of the two wavelet coefficients above correspond with R peaks of ECG signal, a detection sensitivity of99.37%and positive prediction of99.35%are achieved by using the inspection policy of R-waves detection algorithm based on the jump of modulus maxima sequence in the linearly combination of the two wavelet coefficients.2) The method based on CEEMD for detrending the RR interval (RRI) series is proposed. RRI series derived from ECG are the source information to HRV analysis, and they are unevenly sampled. The slow trends in the RRI series should be removed in the preprocessing step to get a reliable result of HRV analysis. Re-sampling is required to convert the unevenly sampled RRI series into evenly sampled time series when using the widely accepted smoothness prior approach (SPA). Noise is introduced in this process and the information quality is thus compromised. To solve this problem, empirical mode decomposition (EMD) and its variants were introduced to directly process the unevenly sampled RRI series, and trends are removed through partial reconstruction. Besides, a RR interval model is also proposed to fascinate the introduction of standard metrics for the evaluation of the detrending performance. Based on standard metrics including signal-to-noise-ratio in dB (ISNR), mean square error (EMS), and percent root square difference (DPRS), the effectiveness of detrending methods in RR interval analysis were determined. The results demonstrated that complementary ensemble EMD (CEEMD, a variant of EMD) based method had a higher ISNR, a lower EMS, and a lower DPRS as well as a better RRI series detrending performance compared with the SPA method, which would in turn lead to a more accurate HRV analysis.3) The comparison of HRV indexes between heart failure patients and the healthy is made, and diagnosis models for heart failure are established based on these indexes. Time domain, frequency domain and non-linear methods are employed for short duration HRV analysis exerts on ECG data belonging to40heart failure patients and40healthy persons, and thus diagnosis models are established based on various combinations of different HRV indexes, and linear discriminant analysis (LDA) and/or support vector machine (SVM). The results demonstrated a diagnostic accuracy of92.5%could be achieved by the model based on indexes including the mean of RR intervals RR, the standard deviation of successive RR interval differences SDNN, the short-term slope of detrended fluctuation analysis (DFA) α1, the long-term slop of DFA a2and approximate entropy ApEn, and LDA. And a diagnostic accuracy of95%could be achieved by the model based on indexes including RR, SDNN, the root mean square of successive differences RMSSD, The short-term parameter of Poincare SD1and ApEn, and SVM. The results suggest that HRV indexes can extract information underlying the cardiac dynamic systems, and thus could be employed for the diagnosis of heart failure.4) The effects on autonomous nervous system caused by pain exposure of newborns due to heel lancing are investigated, and pain detection models for newborns based on the combination of HRV indexes are established. Time domain, frequency domain and non-linear methods are employed for short duration HRV analysis exerts on ECG data belonging to40newborns before and during pain processing, and thus detection models are established based on various combinations of different HRV indexes, and LDA and/or SVM. The results demonstrated a detection accuracy of78.75%could be achieved by the model based on the combination indexes including ApEn, the maximum line length in recurrence plot Lmax, and determinism DET, and LDA. And a detection accuracy of83.75%could be achieved by the model based on the combination of5indexes including RR, relative amount of successive intervals differing more than50ms pNN50, ApEn, correlation dimension D2and recurrence rate REC, and SVM. The results suggest that HRV indexes can reveal the response of autonomous nervous system to pain exposure of newborns, and thus could be employed for the detection of pain for newborns.