The words you are searching are inside this book. To get more targeted content, please make full-text search by clicking here.
Discover the best professional documents and content resources in AnyFlip Document Base.
Search
Published by cloud.project8, 2018-01-23 04:28:41

HRV data mining-ktsangv1

HRV data mining-ktsangv1

Data-mining in (HRV)
Heart Rate Variability
心率变异性 的数据挖掘

20 Jan 2018

Abstract

• Heart rate variability (HRV) is the physiological phenomenon of
variation in the time intervals between heartbeats, which are
referred to as RR intervals and are measured over a period of
anywhere from 10min to 24 hours to form the most commonly
studied HRV time series. Study of HRV as a method of quantifying
cardiac autonomic function began in the 1960s. The importance of
HRV was established in the 1980s when cardiac diseases such as
myocardial infarction and sudden cardiac death were found to be
related to reduced level of HRV. Later, it was recognized that the
complexity in HRV could be used as a tool to diagnose the health of
the patient under study.

• In this talk, we will introduce traditional time-domain and frequency-
domain techniques to characterize the HRV signals, and investigate
how data-mining and unsupervised learning can be used to classify
HRV associated with different cardiac conditions. Application of non-
linear dynamical methodologies to HRV will also be explored.

Human Heart

Controlled by Autonomous Nerve
System (ANS)

The human Heart:
an electric motor

1 SinoAtrial node 窦房结
P-wave
2 AtrioVentricular Node 房室结
 PR interval

SinoAtrial node 窦房结

AtrioVentricular
Node 房室结

Electrical conduction system
of the heart

How the Heart works

Willem Einthoven (21 May 1860 – 29 Sep 1927), a
Dutch doctor and physiologist, invented the first
practical electrocardiogram (ECG or EKG) in 1903
and received the Nobel Prize in Medicine in 1924
for it. His device was much more sensitive than
earlier ones.

He was also the first one to use the letters P, Q, R,
S, and T to denote the different parts of the wave
observed in ECG.

Electrodes in 12-lead ECG

A portable wireless ECG patch, which sticks
on one’s chest, made it possible to measure
the heart rate conveniently, while we
perform daily routines.

Electrocardiography (ECG or EKG)
心电图

In a conventional 12-lead
ECG, 10 electrodes are
placed on the patient's
limbs and on the surface
of the chest.

What do the ECG wave components
represent?

P wave = atrial depolarisation 心房去极化

This is the start of the cycle, setting the pace of the heart beat.

QRS = ventricular depolarisation

T = repolarisation of the ventricles

Heart rate depends on the rate at which the
sinoatrial node [窦房结]produces action
potentials.

At rest, heart rate is between 60 and 100 beats

per minute.

This is a result of the activity of two sets of nerves,

one acting to slow down action potential
production (parasympathetic nerves副交感神经)
and the other acting to speed up action potential
production (sympathetic nerves交感神经)

Heart rate variability (HRV)

is the physiological phenomenon of variation
in the time interval between heartbeats. It is
measured by the variation in the beat-to-beat
interval.

“RR” is the interval between successive Rs.
"NN" is used sometimes in place of “RR” to
emphasize the fact that the processed beats are
"normal" beats.

Premature ventricular contraction

(PVC)

A PVC may be perceived as a "skipped beat“, in which the
ventricles contract first before the atria have optimally
filled the ventricles with blood. It means that circulation is
inefficient.

ECG (5 min) before Qi-gong. PVCs (Premature Ventricular
Contraction) are apparent in the top panel. The lower
panel is the blow-up view of signals from the top

ECG (5 min) after Qi-gong. No PVC is seen in the top panel.
The lower panel is the blow-up view of signals from the top

Physiological Origin of HRV

• Day-night periodicity

related to circadian changes in autonomic nerve
activity

• Respiratory sinus arrhythmia
• 10s rhythm and slower fluctuations

& others …



Respiratory Sinus Arrhythmia
呼吸窦性心律失常



Early works

Is the Normal Heartbeat Chaotic or Homeostatic?
Ary L. Goldberger NIPS Volume 6/April 1991

https://www.physionet.org/tutorials/ndc/
NONLINEAR DYNAMICS, FRACTALS, AND CHAOS THEORY FOR
CLINICIANS - Ary L. Goldberger - 2000 (June 13)

Homeostasis [體內平衡] is the property of a system within
the body of an organism in which a variable, such as the
concentration of a substance in solution, is actively regulated
to remain very nearly constant. The concept was described
by French physiologist Claude Bernard in 1865 and the word
was coined by Walter B. Cannon in 1929.



Heart rate dynamics: Normal sinus rhythm in healthy
subjects (left) shows complex variability with a broad
spectrum and a phase space plot consistent with a strange
(chaotic) attractor. Loss of complex variability is sign of
heart problem (middle & right)

Heart rate time series, from a healthy subject (top) and a
patient with severe congestive heart failure (CHF充血性心力
衰竭) (bottom) have nearly identical means and variances,

yet very different dynamics.

Two review papers

[R1] Heart rate variability: Standards of measurement,
physiological interpretation, and clinical use
Task Force of The European Society of Cardiology and
The North American Society of Pacing and
Electrophysiology, European Heart Journal (1996) 17,
354–381

[R2] Advances in heart rate variability signal analysis:
joint position statement by the e-Cardiology ESC
Working Group and the European Heart Rhythm
Association co-endorsed by the Asia Pacific Heart
Rhythm Society
European Society of Cardiology 2015

“Heart Rate Variability” [coined in R1] has become
the conventionally accepted term to describe
variations of both instantaneous heart rate and RR
intervals.

HRV contains a lot of information about the heart,
but not everything, e.g. it does not tell if you have
Wolff–Parkinson–White syndrome.

Measurement of HRV

1. Time domain methods Statistical measures

Variables Unit Description
SDNN
SDANN ms Standard deviation of all NN intervals.

RMSSD ms Standard deviation of the averages of NN intervals in all 5 min
segments of the entire recording.
SDNN index
ms Square root of the mean of the sum of the squares of differences
SDSD between adjacent NN intervals.
NN50 count
ms Mean of the standard deviations of all NN intervals for all 5 min
pNN50 segments of the entire recording.
HRV
triangular ms Standard deviation of differences between adjacent NN intervals.
index
Number of pairs of adjacent NN intervals differing by more than 50
ms in the entire recording.

% NN50 count divided by the total number of all NN intervals.

Total number of all NN intervals divided by the height of the
histogram of all NN intervals measured on a discrete scale with bins of
7.8125 ms (1/128 s).

Measurement of HRV

2. Frequency domain methods (DFT,
PSD) short-term recordings (5 min)

Variable Unit Description Freq. Range
VLF ms^2
LF ms^2 Power in very low frequency range Below 0·04 Hz
HF ms^2
LF/HF Power in low frequency range 0·04–0·15 Hz
LF norm ms^2
Power in high frequency range 0·15–0·4 Hz
HF norm
Ratio LF [ms^2]/HF [ms^2]
5 min total
power LF power in normalised units
LF/(Total Power–VLF)x100

HF power in normalised units
HF/(Total Power–VLF)x100

The variance of NN intervals over the temporal approximately <0·4 Hz
segment

Physionet

• Physionet(https://www.physionet.org/) is a biomedical
research resources website in the USA.

• It contains EKG database of healthy people, arrhythmia, heart
failure sleep apnea syndrome and many cardiovascular diseases.

• In this project, the data is from MIT-BIH Normal Sinus Rhythm
Database(nsrdb)
(https://www.physionet.org/physiobank/database/nsrdb/),
Congestive Heart Failure RR Interval Database(chf2db)
(https://www.physionet.org/physiobank/database/chf2db/ )
and Long-term AF database(ltafdb)
(https://physionet.nlm.nih.gov/pn3/ltafdb/) in Physionet.

Data Description

• The data used in this project are downloaded from
Physionet.
• There are 76 people, each contains a continuing long-
term 24 hours around electrocardiogram signal.
• Three groups: 16 healthy people, 30 congestive heart
failure patients and 30 atrial fibrillation patients.
• Only healthy people No.1-5, congestive heart failure
patients No.1-5 and atrial fibrillation patients No.1-5 have
beat-to-beat interval data in Physionet.

Long-term RR interval plot





Other Statistical measures

Standard deviation of NN intervals in 5-minute
segment plot

Result of Time domain methods
statistical measures

Principal component analysis

PC1 PC2 PCP2C3 97.912%

Principal component analysis

Principal component analysis
3D plot

Principal component analysis

PC1

PC2

89.53%

Principal component
analysis 2D plot

2D K-means
Clustering Plot

Accuracy of classification

• About 6-7 cases in a total of 76 patients are
classified incorrectly.

• Our classification scheme is very crude.
• Not all time-domain statistics are used.
• No frequency-domain statistics yet.
• No nonlinear analysis yet.

Histograms of HR (healthy & CHF patients)

Non-Linear Analysis:

Poincaré Plot from an RR interval time series

Heart Rate Variability (HRV) Signal Analysis (p.35)







More Non-Linear Analysis

Approximate Entropy and Sample Entropy
Detrended Fluctuation Analysis
Spectrum Power Law Exponent
Fractal Dimensions
Multifractal Analysis


BOOKS on HRV

Heart Rate Variability (HRV) Signal Analysis: Clinical
Applications by Markad V. Kamath, Mari Watanabe,
Adrian Upton (Editors); (CRC Press; Oct 17, 2012)
Poincaré Plot Methods for Heart Rate Variability
Analysis by Ahsan Habib Khandoker, Chandan Karmakar,
Michael Brennan, Marimuthu Palaniswami, Andreas Voss;
(Springer; Aug 15, 2013)

Heart Rate Variability by Gernot Ernst (Springer; Nov 7,
2013)
Heart Rate Variability by Marek Malik, A. John Camm
(Editors), (Wiley-Blackwell; 1 edition November 6, 1995)

Summaries

• Time & frequency domain statistics can be
used to characterize HRV associated with
different cardiac conditions.

• PCA reduces 7 time-domain statistics to 3
components which contains 98% of variance,
or 2 components which contains 90% of
variance, using HRV data from Physionet.

• 2D (PC1 & PC2) KNN classifies the 3 types of
cardiac conditions in our data up to 90%
accuracy.

Future action items

• Frequency-domain analysis
• Non-linear dynamical time series analysis
• Poincare plots
• Supervised machine learning
• More clinical data needed to draw firm

conclusions


Click to View FlipBook Version