Transcript Arrhytmia analysis - Laboratory of Computer and

Arrhythmia analysis
(heart rate variability)
Johanna Uusvuori
25.11.2004
Contents
1. Introduction: one slide of autonomic nervous system
2. Why does heart rate vary?
3. Analysis methods
a) Time domain measures
b) Model of the heart rate
c) Representations of heart rate
d) Spectral methods (introduction)
4. Summary
Human nervous system
Somatic
Autonomic
Somatic nervous
system: controls
Autonomic nervous
system:
organs under voluntary regulates individual organ
function and homeostasis,
control (mainly
and for the most part is not
muscles)
subject to voluntary
control
Parasympathetic:
rest
Sympathetic:
Fight, fright,
flight
Why does heart rate vary?
Why is the variation interesting?
Heart rhythm is due to the
pacemaker cells in the
sinus node
Autonomic nervous system
regulates the sinus node
Analysis of the sinus rhythm
provides information about the state
of the autonomic nervous system
Starting point of the analysis of the heart rate variability
 sinus node → P-wave (hard to detect)
 analysis methods are based on measuring RR-intervals
(RR-interval can be used instead of PP-interval, since PRinterval ~ constant )
 NN-intervals = RR-intervals but non-normal intervals
excluded
RR-interval
Problems in the analysis
- In laboratory analysis is easy.
- 24 h measurement (Holter)
problems: wrong corrects,
undetected beats,
100 000 RR-intervals
- Analysis methods are sensitive to errors
(time domain methods less sensitive,
spectral most sensitive)
- →
Time domain measures of HR
 Long term variations in heart rate
(due to parasympathetic activity)
are described by:
- SDNN = standard deviation of NN-intervals (1 value/ 24 h)
- SDANN = standard deviation of NN-intervals in 5-minute segments
(288 values / 24 h)
 Short term variations in heart rate
(due to sympathetic activity)
- rMSSD = standard deviation of
successive interval differences
- pNN50 = the proportion of intervals
differing more than 50% from the previous
interval (used clinically)
mean int.diff.
Intervals:
d IT (k )  tk  tk 1
Successive interval differences:
Time domain measures of HR…
Histogram approach:
– has been used to study arrhyhtmias (in addition
to spontane variations in HR)
– possible to remove artefacts and ectopic beats
– only for 24 h measurement
– width of the peak determines the variation in
the heart rate
Peak of short intervals due to falsely
detected T-waves
Model of the heart rate
Integral pulse frequency modulation (IPFM) model:
INTEGRATOR
 Main idea:
THRESHOLD
– We have the output: event series d Eu (t )
– We search for input m(t) that modulates the HR
(=autonomic nervous system)
– m0 is the mean heart rate
IPFM-model…
 Bridge to physiology: pacemaker cells collect the
charge until threshold. Then action potential if
fired.
 When this equation is valid, produce a peak to the
event series:
tk
(
m

m
(

))
d


R
0

t k 1
m0 mean heart rate
tk time of QRS-complex
m(t) modulation of heart rate
R threshold
Representations of the heart rate
Quantities to describe the heart rate:
 Lengths of the RR-intervals
 Occurence times of the QRS-complexes
 Deviations of the QRS-complex times
from the times predicted by a model
With IPFM-model we can test which method
is best in finding the modulation m(t).
Representations of the HR…
1. RR-interval series
* Interval tachogram & inverse
d IT (k )  tk  tk 1
d IIT (k ) 
1
tk  tk 1
These are functions of k (# of heart beats). If they
can be changed to functions of time, several
methods from other fields can be used in the
analysis.
* Interval function & inverse (u=unevenly sampled)
d (t )  k 1 (t k t k 1) (t  t k )
u
IT
K
* Interpolated interval fuction & inverse
(evenly sampled, function of t)
- sample and hold – interpolation (and better
methods)
- sample & hold produces high frequency noise
low pass filter → before resampling
Representations of the HR…
2. Event series
K
d E (t )    (t  tk )
 Event series = QRS occurence times:
k 0
 In low frequencies info of HR, in high frequencies
noise → new representation: low-pass filter h
K
d LE (t )   h(t   )d E ( )d   h(t  tk )
k 0
 h =sin(2piFct)/t for example. After some limit the
terms in the sum are allmost zero.
 If in the IPFM-model m(t)=sin(F1t), a proper lowpass filter removes other stuff
except the m(t)
→ estimate for m(t)=dLE(t)
Representations of the HR…
3. Heart timing
- Unlike previous representations, this is based on the IPFM-
model.
- The aim is to find modulation m(t).
K
- Heart timing representation:
u
d HT (t )   (kT0  tk ) (t  tk )
k 0
k = # of heart beat
T0 = average RR-interval length
- dHT is the deviation of the event time tk from the expected time
of occurence. The expected time of occurence is kT0.
- By calculating Fourier transform of the dHT and m(t), one can
see that the spectrum of dHT and m(t) are related, and spectrum
of m(t) can be calculated from the spectrum of dHT.
Representation of the HR…
Performance of the representations
 Best method to predict
m(t) of IPFM-model is
to use heart timing
representation (which is
based on this model…)
 However: heart timing
representation does not
fully explain the heart
rate variability of
humans
→ the IPFM-model
might not be accurate
The End of the representation-part
New topic: what kind of modulating
signals do we have?
Spectral methods
Which kind of information is gained?
Oscillation in heart rate is related to for example:
- body temperature changes 0.05 Hz (once in 20
seconds)
- blood pressure changes 0.1 Hz
- respiration 0.2-0.4 Hz
Power of spectral peaks → information
about pathologies in different
autonomic funtions
Power spectrum of a heart rate signal during rest
Spectral methods…
Which kind of information is gained?
 Peaks of thermal and blood pressure regulation
sometimes hard to detect →
frequency ranges used: 0.04-0.15 Hz and 0.15-0.40 Hz
 Sympathicus increase, low-frequency power increase
 Parasympathicus increase, high-frequency power
increase
 Ratio between two spectral power describes autonomic
balance
Spectral methods…
Problems of spectral analysis
 Stationarity important
 Extrabeats violate the stationarity, but they
can be removed in the analysis
 Undetected beats are a bigger problem
→ spectral analysis can not be conducted, if
they are present
 HR determines the highest frequency that
can be analyzed: 0.5*mean hr
Summary
 Autonomic nervous system → heart rate varies
 Measurment of HR → info about autonomic system
 Analysis methods of HR:
– Time domain methods  standard deviations
– Representations of the heart rate
(intervals, times, heart timing=model based)
– Model that can predict heart rate: IPFM-model
– Spectral analysis (to be continued in the next talk)