69
Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals By Dr. S. Poornachandra Dean IQAC SNS College of Engineering

Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

  • Upload
    luther

  • View
    38

  • Download
    0

Embed Size (px)

DESCRIPTION

Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals. By Dr. S. Poornachandra Dean IQAC SNS College of Engineering. Objective. Denoising of biomedical signals with better performance. Types of noises. The muscle artifacts Respirator muscles - PowerPoint PPT Presentation

Citation preview

Page 1: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

Wavelet Based Subband Shrinkage Models and their Applications in Denoising of

Biomedical Signals

ByDr. S. Poornachandra

Dean IQACSNS College of Engineering

Page 2: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 2

Objective Denoising of biomedical signals with

better performance

Page 3: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 3

Types of noises The muscle artifacts

Respirator muscles Cardiac muscle Moving artifacts

Electro-magnetic radiations Power line frequency noise Instrument noise Interference of other physiological signals

Page 4: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 4

Statistical Estimations

Mean Variance Risk

M

1k

2

kk θθER ˆ

Page 5: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 5

ECG Specification

The practical ECG was downloaded from the PhysioBank

Sampling rate is 360Hz Resolution is 11 Bits/Samples Bit rate is 3960 bps Length of ECG data: 650000 Length of ECG data considered: 5000

Page 6: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 6

Other biosignals used. . . .

EEG PCG Pulse Waveform

Page 7: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 7

Parameters for Analysis

Signal to Noise Ratio (SNR) =

Percentage Root Mean-Squared Difference(PRD)

= SNR Improvements

= Input SNR – Output SNR RMS Error

= RMS (Recovered Signal – Original Signal) PSNR

=

VarianceNoise VarianceSignal

100% SignalOrigianl

SignalRecovered SignalOriginal

SignalRecovered - SignalOriginal SignalOriginal

Page 8: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 8

Time-domain

Advantages Simple Easy to implement Lower computational complexity

Disadvantages Slow convergence when the input is highly

colored

Page 9: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 9

Need for Transform-domain Advantages

Better convergence Parallism

Disadvantages Complexity increases as order of the filter increases

Exhibit slow convergence High minimum mean square error

Remedy Subbanding – reduced coefficients at each subband

Page 10: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 10

Advantages of Wavelet

Works on non-stationary data Time-frequency aspect gives information

about frequency composition of a signal at a particular time

Short signal pieces also have significance

Page 11: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 11

Wavelets Defined . . . . .

“The wavelet transform is a tool that cuts up data, functions or operators into different frequency components and then studies each component with a resolution matched to its scales”

Dr. Ingrid Daubechies, Lucent, Princeton U

Page 12: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 12

DWT – Demystified

d3: Level 3 DWTCoeff.

Length: 512B: 0 ~

g[n] h[n]

g[n] h[n]

g[n] h[n]

2

d1: Level 1 DWTCoeff.

Length: 256B: 0 ~ /2 Hz

Length: 256B: /2 ~ Hz

Length: 128B: 0 ~ /4 HzLength: 128

B: /4 ~ /2 Hz

d2: Level 2 DWTCoeff.

…a3….

Length: 64B: 0 ~ /8 HzLength: 64

B: /8 ~ /4 Hz

2

2 2

22

|H(j)|

/2-/2

|G(j)|

- /2-/2

a2

a1

Level 3 approximation

Coefficients

Page 13: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 13

Shrinkage ?

A shrinkage method compares empirical wavelet coefficient with a threshold and is set to zero if its magnitude is less than the threshold value.

Page 14: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 14

Condition & Characteristics of Shrinkage

The magnitude of signal component must be larger than existing noise component

It does not introduce artifacts The wavelet transform localizes the most

important spatial and frequential features of a regular signal in a limited number of wavelet coefficients.

Observations suggest that small coefficients should be replace by zero, because they are dominated by noise and carry only a small amount of information.

Page 15: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 15

Pioneers …

Donoho and Johnstone (1994) – Soft Shrinkage Coifman and Donoho (1995) – Cycle Spinning Nason (1996) – Cross Validation Shrinkage Bruce and Gao (1997) –Garrote Shrinkage

Page 16: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 16

Shrinkage functions

xx

xx

x0,

xS

,

,

||

)(

|| ,

|| )(

xx

x0,xH

λ |x| /x)x-(λ

λ|x| 0,xG

,

)(

2

Page 17: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 17

Shrinkage Algorithm

Apply DWT to the vector y and obtain the empirical wavelet coefficients cj,k at scale j, where j = 1, 2, . . , J.

Estimated coefficients are obtained based on the threshold

= [1, 2, . . . . j]T.

Apply shrinkage to the empirical wavelet coefficients at each scale j.

The estimate of the function can be obtained by taking inverse DWT.

Page 18: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 18

Threshold methods

The rigrsure uses for the soft shrinkage estimator, which is a shrinkage solution rule based on Stein’s Unbiased Risk Estimate (SURE).

The sqtwolog threshold uses a fixed form threshold yielding minimax performance multiplied by a small factor proportional to log(length(x)).

The heursure threshold is the hybridization of both rigrsure and sqtwolog threshold.

The minimax threshold uses a fixed threshold chosen to yield minimax performance for MSE against an ideal procedure.

Page 19: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 19

Median Absolute Deviation

Prof. Donoho proposed

Where, is the estimate of noise variance

Median Absolute Deviation MAD(v)=[|v1-vmed|, … , |v1-vmed|]med

)ln(2ˆ N

6745.0

ˆ , jicmedian

Page 20: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 20

Alpha-trim Filter

The alpha-trim filter is a special type of L-filter,

A particular choice of aj coefficient yields a alpha-trim filter

where T is the largest integer which is less than or equal to αM, 0 ≤ α ≤ 0.5. When α = zero, the α-trim filter becomes the running mean filter; When α = 0.5, the α-trim filter becomes the median filter.

jxja ii

M

j1

jxii

TM

Tj TN

1 )(2

1

Page 21: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 21

Threshold at each subband

The Threshold values at each subband for 20% noise level is given in tables

SNR (dB) PRD (%) ECG Signal

Page 22: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 22

Wavelet level analysis

Page 23: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 23

Wavelet level analysis

Page 24: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 24

Hybrid Model ….

Analysis Filter

a3

d2

d3

d1

Hard shrinkage

Xshrinkage

Xshrinkage

Xshrinkage

Page 25: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 25

Basic shrinkage (ECG)

Page 26: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 26

Basic shrinkage (PCG)

Page 27: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 27

BSWTAF-I (Scale-Domain Analysis)

ATI Model

Analysis Filter

Adaptive Filter

Shrinkage Function

Synthesis Filter

Page 28: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 28

BSWTAF-II (Scale-Domain Analysis)

TAI Model

Analysis Filter

Shrinkage Function

Adaptive Filter

Synthesis Filter

Page 29: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 29

ASWTAF Model (Time-Domain Analysis)

TIA Model

Analysis Filter

Shrinkage Function

Adaptive Filter

Synthesis Filter

Page 30: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 30

ECG Simulation….

Page 31: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 31

ECG Simulation….

Page 32: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 32

EEG Simulation….

Page 33: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 33

EEG Simulation….

Page 34: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 34

PCG Simulation….

Page 35: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 35

PCG Simulation….

Page 36: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 36

Shrinkage Distribution

Page 37: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 37

Hyper Shrinkage Function

Where

λ |x| x) tanh(ρ

λ |x| 0txxxhyp ) tanh()(

xmaxΔρ

Page 38: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 38

Modified-hyper shrinkage function

k is the scaling function

x

xxkxXmh

,0

,6

1)(

2

Page 39: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 39

Subband Adaptive shrinkage function

j

j

j

SA

λ |x| 0

λ |x| -

1

--1

x jx2

j

x2

j

)(

Page 40: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 40

ECG Denoising - Noise level is 20%

Page 41: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 41

ECG Denoising - Noise level is 20%

Page 42: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 42

EEG Denoising - Noise level is 20%

Page 43: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 43

EEG Denoising - Noise level is 20%

Page 44: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 44

PCG Denoising - Noise level is 20%

Page 45: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 45

PCG Denoising - Noise level is 20%

Page 46: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 46

Objective…

Reduce the minimum mean square error (MMSE)

between original ECG f and denoised ECG . y = [y1, y2, . . . , yN] N

Thenyi = f(xi) + ni, i = 1, 2, . . ,N

The risk function,

2ˆ1),ˆ( ff

NffR

Page 47: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 47

Estimation of Mean, Variance and Risk

)()( XEM

)()( XVarV

2)()(

XER2

)()(

MV

Page 48: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 48

Mean estimation for Hyper Shrinkage

Let X ~ N(θ,1), and be the probability distribution and the density function for standard Gaussian random variable respectively, then the Mean estimation is given by

kM hyp 13)(

k13

k11 2 k11 2

Page 49: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 49

Variance estimation for Hyper Shrinkage

The Variance estimation is given by

1)(hypV

2)(mhM

Page 50: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 50

Risk estimation for Hyper Shrinkage

The Risk estimation is given by

1)(hypR

Page 51: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 51

Mean estimation for Subband Adaptive Shrinkage

)(SAM 21 32)]()(2[ 2

)]()([2 2

)Φ( )Φ(- 22

The Mean estimation is given by

Page 52: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 52

Variance estimation for Subband Adaptive Shrinkage

The Variance estimation is given by

)(SAV )Φ(- )Φ(

)]()([31 2

)]1)(1([2 22 )Φ( )Φ(

)Φ(- )Φ(- )])[8 3 Φ(-Φ(

)Φ( )Φ(- 24 2)(SAM

Page 53: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 53

Powerline Frequency Interference Cancellation

Page 54: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 54

50 Hz Noise Cancellation in ECG

Page 55: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 55

50 Hz Noise Cancellation ECG

Page 56: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 56

50 Hz Noise Cancellation ECG

Page 57: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 57

50 Hz Noise Cancellation in EEG

Page 58: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 58

50 Hz Noise Cancellation in EEG

Page 59: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 59

50 Hz Noise Cancellation in EEG

Page 60: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 60

50 Hz Noise Cancellation in PCG

Page 61: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 61

50 Hz Noise Cancellation in PCG

Page 62: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 62

50 Hz Noise Cancellation in PCG

Page 63: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 63

Journal publication arise from this work

Poornachandra S. and N. Kumaravel, “A Wavelet coefficient smoothened RLS-Adaptive denoising model for ECG”, Journal of Biomedical Sciences Instrumentation, vol. 39, ISA vol. 437, pp. 154-157, April 2003.

Poornachandra S. and N. Kumaravel, “Hyper-trim shrinkage for denoising of ECG signal”, ELSEVIER Journal of Digital Signal Processing, Vol. 15, Issue-3, pp. 315-327, May 2005.

Poornachandra S. and N. Kumaravel, “Wavelet based Adaptive Denoising Models for Biological Signals”, Journal of Institute of Engineers, Vol. 86, Nov. 2005.

Poornachandra S. and N. Kumaravel, “Subband-Adaptive Shrinkage for Denoising of ECG Signals”, EURASIP Journal on Applied Signal Processing (Available online).

Page 64: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 64

Journal publication arise from this work

Poornachandra S. and N. Kumaravel, “Wavelet Thresholding by -Trim Mean Filter”, Journal of Institute of Engineers, Vol. 87, January 2007

Poornachandra S. and N. Kumaravel, “Statistical Estimation for Hyper Shrinkage”, ELSEVIER Journal of Digital Signal Processing, (Available online).

Poornachandra S. and N. Kumaravel, “A Novel method for the Elimination of Power Line Frequency in ECG Signal using Hyper Shrinkage Function”, ELSEVIER Journal of Digital Signal Processing, (In Press).

Poornachandra S. and N. Kumaravel, “Hyper Shrinkage for Denoising of ECG Signal with Adaptive Filter”, ELSEVIER Journal of Digital Signal Processing, (Under Review).

Page 65: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 65

Conference publication

Poornachandra S. and Dr N. Kumaravel, “A new wavelet co-efficient smoothened Adaptive filtering for bio-signal”, Proc. of ICBME, 4th –7th December 2002, Singapore.

Poornachandra S. and Dr N. Kumaravel “Wavelet coefficient smoothened LMS-adaptive denoising model for electro-cardio graph”, Proc. of NCC, 31st-2nd Jan-Feb 2003, Chennai, India.

Poornachandra S., Dr N. Kumaravel et al., “A modified Wavelet model with RLS-adaptive for denoising Gaussian Noise from ECG signal”, Proc. of NSIP, 8th –11th June 2003, Grudo-Triesta, Italy.

Page 66: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 66

Conference publication

Poornachandra S., Dr N. Kumaravel et al., “Denoising of ECG by -Trim thresholding of Wavelet coefficients”, Proc. of NCC, 31st-2nd Jan-Feb 2004, Bangalore, India.

Poornachandra S., Dr. N. Kumaravel et al, “WaveShrink using Modified Hyper-Shrinkage Function” Proc. of IEEE-EMBC, 1st – 4th September 2005, Shanghai, China.(best session paper award)

Poornachandra S., Dr. N. Kumaravel et al, “A Novel method for the elimination of powerline frequency in ECG signals using modified-hyper shrinkage”, Proc. of IFMBE, ICBME2005, 7th – 10th December 2005, Singapore.

Page 67: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 67

Conclusion…..

In real time bio-signal acquisition, the noise level is always less than the signal level. Hence shrinkage is the better choice for denoising

This thesis suggested that shrinkage can also be used to eliminate powerline frequency from bio-signals.

This thesis proposed following shrinkage function that are better than the existing ones and its mathematical models

Hyper shrinkage Modified shrinkage Subband-Adaptive shrinkage

Page 68: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 68

Suggested for future work

1. Image processing applications2. VLSI implementation3. Better shrinkage models can be proposed4. Communication signal processing5. Iterative shrinkage6. Fussy shrinkage

Page 69: Wavelet Based Subband Shrinkage Models and their Applications in Denoising of Biomedical Signals

04/21/23S. Poornachandra

(2001399722) 69

Bias and Variance