Transcript Heart Rate Variability: Measures and Models

Heart Rate Variability:
Measures and Models
指導教授：鄭仁亮
學生：曹雅婷
Outline
 Introduction
 Methods
 Conventional Point Process
 Fractal Point Process
 Measure
 Standard Measures
 Novel Measures
Introduction
 ECG
a recording of the cardiac-induced skin
potentials at the body’s surface
 HRV
called heart rate variability, the
variability of the RR-interval sequence
Methods
 The heartbeat sequence as a point process.
 The sequence of heartbeats can be studied
by replacing the complex waveform of an
individual heartbeat recorded in the ECG.
 The sequence of heartbeats is represented
by
h(t )    (t  ti )
i
ECG Analysis
Conventional Point Process
 Simplest
 homogeneous Poisson point process
 Related point process
 nonparalyzable fixed-dead-time modified
Poisson point process
 gamma-γ renewal process
Homogeneous Poisson point process
 The interevent-interval probability
density function
p ( )   exp(  )
where λ is the mean number of events
per unit time.
 interevent-interval mean=1/λ
 interevent-interval variance=1/λ2
Dead-time modified Poisson point process
 The interevent-interval probability
density function
p ( ) 
0
 exp[  (   d )]
 d
 d
Here τd is the dead time and λ is the
rate of the process before dead time is
imposed.
Fractal Point Process
 Fractal stochastic processes exhibit
scaling in their statistics.
 Suppose changing the scale by any
factor a effectively scales the statistic by
some other factor g(a), related to the
factor but independent of the original
scale:
w(ax) = g(a)w(x).
Fractal Point Process
 The only nontrivial solution of this
scaling equation, for real functions and
arguments, that is independent of a and
x is
w(x) = bg(x) with g(x) = xc
 The particular case of fixed a admits a
more general solution
g(x; a) = xc cos[2πln(x)/ ln(a)]
Standard Frequency-Domain Measures
 A rate-based power spectral density
 Units of sec-1
 An interval-based power spectral
density
 Units of cycles/interval
 To convert the interval-based frequency
to the time-based frequency using
f time  f int / E[ ]
Estimate the spectral density
1. Divided data into K non-overlapping
blocks of L samples
2. Hanning window
3. Discrete Fourier transform of each
block
1
ˆ
S ( f ) 
K
K
2
~
 k ( f )
k 1
Measures in HRV
 VLF. The power in the very-low-frequency
range: 0.003–0.04 cycles/interval.
 LF. The power in the low-frequency range:
0.04–0.15 cycles/interval.
 HF. The power in the high-frequency range:
0.15–0.4 cycles/interval.
 LF/HF. The ratio of the low-frequency-
range power to that in the high-frequency
range.
Standard Time-Domain Measures
 pNN50. proportion of successive NN
intervals
 SDANN. Standard Deviation of the
Average NN interval
 SDNN. Standard Deviation of the NN
interval
Other Standard Measures
 The event-number histogram
 The Fano factor
Novel Scale-Dependent Measures
 Allen Factor [A(T)]
 The Allan factor is the ratio of the eventnumber Allan variance to twice the mean:
2
E{[N i 1 (T) - N i (T)] }
A(T) 
2E{N i 1 (T)}
Wavelet transform using Haar wavelet