Final Signal 3

PROF.DR. Mohamed Aouf

1
[1] Introduction

•(1-A) Motivation
1

• (1-B) Problem
2 Definition

•(1-C) Objectives
3
2
 (1-A) Motivation

(1-A-1) Patient :
• Anterior surface heart ischemia is one of the most
prominent heart diseases.
• Patients with Coronary Artery diseases CAD may
present objective evidence of myocardial ischemia .
• Myocardial ischemia may begin to occur when the
luminal narrowing of the coronary artery exceeds
70%.
• Body surface potential mapping BSPM helps in early
detection of CAD.
3
(1-A-2) Physician:
Facilitate the way of diagnosing by using clean and
efficient maps instead of traditional electrocardiograph
ECG waveforms.

4
 (1-B)Problem Definition

How to Use 3D Wavelet Transform for

Cardiac Mapping Re-construction and De-
noising Of Heart Ischemia specially
anterior surface to facilitate early
diagnostic .

5
 (1-C)Objectives
• (1) ECG peak detection using filter banks

• (2) Using 3D wavelet transform for cardiac maps re-

construction.

• (3) Using 3D wavelet transform for cardiac maps de-

noising.

• (4) Investigate new visualization tool instead of

traditional ECG waveforms to facilitate diagnosis.
6
 (2)
BACKGROUND

Body Surface
Peak Wavelet
Potential
Detection Transform
Mapping
1985-2014 1993-2014 1984-2015

Classification Classification Classification

Theory Theory Theory

Results Results Results

Conclusion Conclusion Conclusion

7
(3)-METHODOLOGY
•(3-1) Filter
banks and Peak
1 detection

• (3-2)-Wavelet
transform
2
8
(3-1) Filter banks and Peak detection:

• A filter bank has two parts: an analysis bank and a

synthesis bank.
• Each filter bank is grouping of band pass filters.
• The analysis filters are used to decompose the
bandwidth of the input signal into sub-band signals
with uniform frequency bands, down-sampling the
input signal as a multi rate digital signal.
• These sub-bands can be then reconstructed using
the synthesis filters. Before introducing the synthesis
filters we use up-sampling.
9
Filter banks
design and
theory

10
Distortions in filter banks :

Magnitude
Aliasing and Phase
distortion

If a filter bank is free from these distortions, the filter

bank is a perfect reconstruction (PR) filter bank. If
these distortions are small, the filter bank is a near-
perfect reconstruction (NPR) filter bank.

11
 Frequency selectivity

Indicates the ability of

a filter bank to
separate frequency sub
bands.

Can evaluate the frequency selectivity by

examining the stop band attenuation and
the transition bands.

12
 (3-2)-Wavelet transform

Wavelet transform important For non-invasive

medical measurements specially the ECG :

A tool for the analysis of transient, non-

stationary, and time varying phenomena

Plays a great role in describing ECG

waves in both the frequency and the
time domain.

13
 Wavelets are

Varying frequency

Short time localized wave

Average value zero

Finite energy
14
The Continuous wavelet transform
CWT

translation parameter, scale parameter normalization

measure of time (measure of frequency) constant

t
t

Mother wavelet
Wavelet Coefficient of (window)
signal f(t) at position b and
scale a
15
 Scaling In Wavelet transform
◼ Scaling means stretching or compressing of the
signal. scale factor (a) for wavelets:

f (t ) = (t) ; a = 1
f (t ) = (2t) ; a = 12
f (t ) = (4t) ; a = 14

16
 Shifting In Wavelet transform

❑A wavelet means delaying its onset. Mathematically,

delaying a function f(t) by k is represented by f(t-k).

17
 Discrete wavelet transform
• The approximation or the representation of The
function f e.g ECG Signal
• cm is the coefficient
• φm is the wavelet function

18
Multidimensional Wavelet Transform

j=0 The original signal f(t) = Ac0

j=0 2D Data

j=1 App. Ac1 Details Dc1

Wavelet Analysis
AJ
j=2 Ac2 Dc2
j=J
H2
Horizontal
Details (H1)
j=2 V2 D2

j=J Vertical Diagonal

j=1 Details (V1) Details (D1)

19
THE APPROXIMATIONS & THE DETAILS

APPROXIMATIONS
Is generated using an additional
function called the scaling
function (t).

It shows information about slow

changes in the signal(LOW
FREQUENCIES )

Are generated by the wavelet function (t).

It shows the ( HIGH FREQUENCIES )

carried by the signal and hence the rapid
changes in the signal.
DETAILS
20
3D Wavelet transform

• The 3D wavelet transform is

extended from the 1D wavelet
1 transform

• It is the product of three one-

dimensional functions. The analysis is
carried out along the horizontal (x), the
2 vertical (y), and the depth (z)
dimensions of the volumetric data.

Eight coefficients result from the one •

3 level analysis.

21
De-noising using the wavelet transform

Noise in the Details of the signal

A better approach is to threshold the

Details then reconstruct the signal

There are two types of threshold,

hard and soft thresholds

22
(4) Proposed
System
Multi electrode system for chest
leads of ECG (30 Channels)

Preprocessing using FB for Peak

Detection
(QRS Detection)

3D Wavelet Transform

Classification
Denoising using 3D-db N
(Normal/ Abnormal)

Reconstruction of Cardiac
Mapping
23
 System
Identification

PART (1) : Pseudo Body Surface Potential Mapping on Heart Using 3D

Wavelet Transform

PART(2) : Using Wavelet Transform For Cardiac Mapping De-noising Of

Heart Anterior Surface

24
PART (1) : Pseudo Body Surface Potential Mapping on Heart
Using 3D Wavelet Transform

First Publication

25
DATA CASES EQUIPMENT

10 normal and 10 abnormal adult ECG, model Heart Mirror IKO

male cases

Single channel with 12 lead selectors

interfaced with a computer
Aged from 28 to 48

Single channel with 12 lead selectors

interfaced with a computer

weight range from 60 to 90 kg

There is also a 50-Hz line noise
rejection filter and a 35-Hz muscle
noise rejection filter in the circuit.

The sampling frequency is 200 Hz

height range from 1.2 to 1.7 meters.
(assuming the maximum frequency
of the signal fmax = 100 Hz))
26
 Preprocessing filter
banks

Filter banks (FB) are used with finite impulse response filters (FIR)
without any aliasing error and without any magnitude or phase
distortion and Of nonlinear processing and moving window integration.

Optimal thresholding , we can differentiate between QRS complexes

and noise peaks

In some cases the ECG signal may pass without QRS detection; thus
a "search-back" strategy technique is used to check the QRS again

The use filter bandwidth is about 5.6 Hz

27
Initial preparation for mapping

ECG R location
QRS detection

28
 Mapping with New Chest
Positions using 3D WAVELET
TRANSFORM :

The present research introduces a practical

solution for recording body surface potential
maps using the standard 12-lead ECG with
different positions.

The ordinary position (P0), the upper positions

(P1) and (P2), and the lower positions (P3) and
(P4)

The analysis is carried out along the

horizontal (x) chest leads, the vertical (y)
positions, and the depth (z) time.
29
 WHY 3D WAVELET
TRANSFORM

Wavelets provide a tool for time and frequency localization.

If one analyzes a function f(t) by means of wavelets then there will definitely be some
kind of localization.
Transient features (short-time details) of f(t), like jump discontinuities or peaks can
easily be localized in the wavelet coefficients of small scales.
The power of the wavelet technique here is detecting signal discontinuities
Long time trends of f(t) are stored in deeper layers of the coefficient hierarchy and are
automatically represented in larger scales..

30
PART(2) Using Wavelet Transform For Cardiac Mapping De-
noising Of Heart Anterior Surface Ischemia

M ul t i electrode s ys t e m
for chest leads of E C G
(30 channels)

NOISE REDUCTION USING 3D

WAVELET TRANSFORM
(Daubechies (db) )
P r e - p r o c e s s i n g filter
b a n k s (peak detection)

C h 1 C h 2 C h 3 0

Converting signals into cubic

Second f o r m

Publication De-noising using d b n

M a p p i n g
31
NOISE
• The Equation introduces received signal was the
sum for both physical signal S(t) ,the corruption
1 noisy N(t).

• R(t) =S(t)+N(t)
2

• The de-noising process proposed in this

research depends upon the difference
3 between the characteristics of the noise and
signal.

• HARD THRESHOLDING CALCULATION ,we

calculated (SNR) to assess the quality of the
4 reconstructed signal .
32
➢ HARD TRESHOLDING
FORMULA

A universal thresholding rule

that provides an easy, fast
Calculations
33
 •As the morphological shape of the signals before and
after de-noising is different, testing the fidelity and
performance of the de-noising technique is very
important.
• To perform this, we calculated the signal to noise ratio
(SNR) as in equation

•Where VR(n) is the reconstructed ECG signal, and SR(n)

is deformation in reconstructed ECG signal

34
 Filter Banks Introduction

What we have in mind is this… :

H1(z) subband processing

IN H2(z) subband processing OUT

subband processing
+
H3(z)
H4(z) subband processing H1 H2 H3 H4

2
- Signals split into frequency channels/subbands
- Per-channel/subband processing
- Reconstruction : synthesis of processed signal
- Applications : see below (audio coding etc.)
- In practice, this is implemented as a multi-rate structure for higher
efficiency (see next slides)

DSP-II p. 3
 Filter Banks Introduction
Step-1: Analysis filter bank
- collection of M filters (`analysis filters’, `decimation filters’) with a
common input signal
- ideal (but non-practical) frequency responses = ideal bandpass filters

2
- typical frequency responses (overlapping, non-overlapping,…)
H1 H2 H3 H4
M=4
H1(z) 2
H1 H2 H3 H4
IN H2(z)
H3(z) 2
H1 H2 H3 H4
H4(z)

DSP-II Version 2008-2009 Lecture-5 Filter Banks - Preliminaries p. 36

 Filter Banks Introduction

Step-2: Decimators (downsamplers)

- To increase efficiency, subband sampling rate is reduced by factor N
(= Nyquist (bandpass) sampling theorem, see Lecture-6)
- Maximally decimated filter banks (=critically downsampled): N=M
# subband samples= # fullband samples
this sounds like maximum efficiency, but aliasing (see below)!
- Oversampled filter banks (=non-critically downsampled): N<M
# subband samples> # fullband samples
M=4 N=3
H1(z) 3
IN H2(z) 3
H3(z) 3
H4(z) 3
DSP-II Version 2008-2009 Lecture-5 Filter Banks - Preliminaries p. 37
 Filter Banks Introduction

Step-3: Subband processing

- Example :
coding (=compression) + (transmission or storage) + decoding
- Filter bank design mostly assumes subband processing has `unit
transfer function’ (output signals=input signals), i.e. mostly ignores
presence of subband processing

M=4 N=3
H1(z) 3 subband processing

IN H2(z) 3 subband processing

H3(z) 3 subband processing

H4(z) 3 subband processing

DSP-II Version 2008-2009 Lecture-5 Filter Banks - Preliminaries p. 38

 Filter Banks Introduction

Step-4: Expanders (upsamplers)

- restore original fullband sampling rate by N-fold upsampling
(=insert N-1 zeros in between every two samples)

M=4 N=3 N=3

H1(z) 3 subband processing 3
IN H2(z) 3 subband processing 3
H3(z) 3 subband processing 3
H4(z) 3 subband processing 3
DSP-II Version 2008-2009 Lecture-5 Filter Banks - Preliminaries p. 39
 Filter Banks Introduction

Step-5: Synthesis filter bank

- upsampling has to be followed by (interpolation) filtering (see below)
- collection of M `synthesis’ (`interpolation’) filters, with a
`common’ (summed) output signal
- frequency responses : preferably `matched’ to frequency responses of
the analysis filters, e.g., to provide perfect reconstruction (see below)
G1 G2 G3 G4

M=4 N=3 N=3 2

H1(z) 3 subband processing 3 G1(z)

IN H2(z) 3 subband processing 3 G2(z) OUT

subband processing G3(z)
+
H3(z) 3 3
H4(z) 3 subband processing 3 G4(z)

DSP-II Version 2008-2009 Lecture-5 Filter Banks - Preliminaries p. 40

 Aliasing versus Perfect Reconstruction

- Assume subband processing does not modify subband signals

(e.g. lossless coding/decoding)
-The overall aim would be to have y[k]=u[k-d], i.e. that the output signal
is equal to the input signal up to a certain delay
-But: downsampling introduces ALIASING, especially so in maximally
decimated (but also in non-maximally decimated) filter banks
(see also Lecture-6)
M=4 N=3 N=3
y[k]=u[k-d]?
H1(z) 3 output = input 3 G1(z)
u[k]
H2(z) 3 output = input 3 G2(z)
+
H3(z) 3 output = input 3 G3(z)

H4(z) 3 output = input 3 G4(z)

DSP-II Version 2008-2009 Lecture-5 Filter Banks - Preliminaries p. 41

 Aliasing versus Perfect Reconstruction

Question :
Can y[k]=u[k-d] be achieved in the presence of aliasing ?
Answer :
YES !! PERFECT RECONSTRUCTION banks with
synthesis bank designed to remove aliasing effects !

M=4 N=3 N=3

y[k]=u[k-d]?
H1(z) 3 output = input 3 G1(z)
u[k]
H2(z) 3 output = input 3 G2(z)
+
H3(z) 3 output = input 3 G3(z)

H4(z) 3 output = input 3 G4(z)

DSP-II Version 2008-2009 Lecture-5 Filter Banks - Preliminaries p. 42

1.1 RANDOM SIGNALS AND NOISE :
- In (electrical) engineering one often encounters
signals that do not have a precise
mathematical description, since they develop as
random functions of time.
- Sometimes this random development is caused
by a single random variable , but often it is a
consequence of many random variables.
- In other cases the causes of randomness are not
clear and a description is not possible, but the
signal is characterized by means of measurements
only.
-A random time function may be a desired signal,
such as an audio or video signal, or it may be an
unwanted signal that is unintentionally added to a
desired (information) signal and disturbs the desired
signal.

-We call the desired signal a random signal and the

unwanted signal noise.

- However, the latter often does not behave like noise

in the classical sense, but it is more like interference.
Then it is an information bearing signal as well, but
undesired.
-A desired signal and noise (or interference) can, in general, not
be distinguished completely;

-By means of well-defined signal processing in a receiver, the -

desired signal may be favoured in a maximal way where as the
disturbance is suppressed as much as possible .

-In all cases a description of the signals is required in order to be

able to analyse its impact on the performance of the system
under consideration .

- Especially in communication theory this situation often occurs .

-The random character as a function of time makes the signals
difficult to describe and the same holds for signal processing or
filtering.

- Nevertheless, there is a need to characterize these signals by a

few deterministic parameters that enable the system user to
assess the performance of the system.

-The tool to deal with both random signals and noise is the
concept of the stochastic process.

- This COURSE gives an elementary introduction to the methods

used to describe random signals and noise. For that purpose use
is made of the laws of probability, which are extensively
described.