On The Denoising Of Nuclear Medicine Chest Region Images

Faculty of Technical Sciences Bitola, Macedonia

Sozopol 2004

Cvetko D. Mitrovski, Mitko B. Kostov

Sozopol 2004p. 2

Structure

Aim / Problem formulation NM images creation process Wavelet shrinkage The filtration of images Experimental results Conclusion

Sozopol 2004p. 3

AIM: To develop methods for analyzing of anatomical data and ROIs on a basis of a raw NM image (set of raw NM images).

PROBLEM: To find a suitable method for automatic preprocessing of the chest region NM images & extraction of the anatomical data.

Aim of the Work & Problem Formulation

Sozopol 2004p. 4

The raw NM images are based directly on the total counts

a low signal-to-noise ratio (SNR) noisy due to low count levels, scatter,

attenuation, and electronic noises in the detector/camera

One of the major sources of error is Poisson noise due to the quantum nature of the photon detection process

NM Images Creation Process

Sozopol 2004p. 5

DWT (produces two groups of coefficients with low and high SNR)

= wi hi

hihard =

hisoft =

Inverse wavelet transformation

Wavelet Shrinkage Program

iw

i

i

w

w

if,0

if,1

i

ii

i

w

ww

w

if,0

if,sgn

1

Sozopol 2004p. 6

Filtration of Chest Region Images Wavelet shrinkage (threshold for Poisson model?)

the Anscombe variance-stabilizing transformation:

the Donoho’s level dependent threshold:

give up the perfect reconstruction (QMF bank – near PR)

Poisson Gaussian noise model

2,12,1 iiii Ny

JjNJj

j ,...,0,2log22/

Sozopol 2004p. 7

The Algorithm transformation of the image calculation of Donoho’s threshold

( = MAD/0.6745)

MAD is the median of the magnitudes of all the coefficients at the finest decomposition scale

wavelet soft-thresholding inverse wavelet transform square the result removing shadow in the obtained image

Sozopol 2004p. 8

Experimental Results

Sozopol 2004p. 9

The QMF Bank

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.2

0.4

0.6

0.8

1

Normalized frequency

Mag

nitu

de r

espo

nse

QMF bank has overall reconstruction error minimized in the minimax sense; the corresponding QMF filters have least-squares stopband error

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.9985

0.999

0.9995

1

1.0005

1.001

1.0015

Normalized frequency

Mag

nitu

de r

espo

nse

Sozopol 2004p. 10

Comparison

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.2

0.4

0.6

0.8

1

Normalized frequency

Mag

nitu

de r

espo

nse

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.2

0.4

0.6

0.8

1

Normalized frequency

Mag

nitu

de r

espo

nse

with biorthogonal wavelets

Sozopol 2004p. 11

Comparison

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.2

0.4

0.6

0.8

1

Normalized frequency

Mag

nitu

de r

espo

nse

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.2

0.4

0.6

0.8

1

with Daubechies with Symlets

Sozopol 2004p. 12

The QMF Bank

5 10 15 20 25 30-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

10 20 30 40 50 60-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

with meyer

Sozopol 2004p. 13

Conclusions

The presented method offers automatic extracting of the anatomic data from the chest region NM images

The method involves: DWT shrinkage program, variance-stabilizing transformation, QMF filters

Further analyzing of processed data (possible inequality between left and right side)

Questions and discussion

Thank you for your attention