Bonfring International Journal of Advances in Image Processing, Vol. 5, No.

1, February 2015

De-blurring Cardiac SPECT Images by Maximum

Likelihood Approach
Neethu M. Sasi and Jayasree V.K.
Abstract--- This paper presents a blind de-convolution
algorithm for enhancing cardiac SPECT images by reducing
the blur present in the image. The method is based on
maximum likelihood estimate and in particular, the processing
is done in a suitable color space. An iterative algorithm,
without any prior information, is used to estimate the original
image and the point spread function. Blur metric and peak
signal to noise ratio are considered for performance
evaluation of the algorithm. The effect of number of iterations
on the quality of de-blurred image is also studied. Real
medical images are used for appraising the algorithm.
Keywords--- Blind De-convolution, Maximum Likelihood
Estimate, Nuclear Images, Single Photon Emission Computed
Tomography

I.

INTRODUCTION

EARTdiseases contribute to major health problems in the

developed world [1]. More people die annually from
heart diseases than any other cause. About 17.3 million people
died from heart diseases in 2008, representing 30% of all
global deaths [2]. Still, mortality rate could be decreased
considerably by early diagnosis of the disease. The use of noninvasive imaging techniques, like nuclear medicine imaging,
are becoming common nowadays in the diagnosis of heart
diseases [3].
The distribution of blood flow to the heart muscle can be
visually analyzed from a cardiac nuclear image. The
accumulation of plaques creates stenosis in the coronary
arteries which impairs blood flow to the myocardium.
Myocardial perfusion imaging is an accurate test available for
assessing the impact of these plaques on the blood flow to the
heart [4]. One popular technique of nuclear imaging is Single
Photon Emission Computed Tomography (SPECT). In this
technique a radioactive tracer is injected into the human body
and it is taken up to the heart through blood flow. Then the
uptake of the tracer by the heart is being imaged by using a
camera which is rotated around the patient to get multiple
images from multiple directions. Short axis, horizontal long
axis and vertical long axis slices are reconstructed from these
multiple images. Various degrading factors such as patient
motion and photon scatter contribute to the blurring of cardiac
SPECT images which reduces its diagnostic accuracy. In this
Neethu M. Sasi, Research Scholar, Department of Electronics, Model
Engineering College, Kochi, India. E-mail:neethumsasi@gmail.com
Jayasree V.K., Associate Professor, Department of Electronics, Model
Engineering College, Kochi, India.
DOI: 10.9756/BIJAIP.10372

paper we propose a method of restoring the original SPECT

image from its blurred version without knowing the actual
image and the degradation factor.
II.

LITERATURE REVIEW

Image de-blurring is an important issue to be addressed in

the area of image enhancement. Different de-blurring
techniques are available in literature for improving the quality
of medical images. Here the challenging fact is that a specific
algorithm which gives better de-blurring for a particular type
of application may fail in giving good results for another type
of application.
A blurred SPECT image can be viewed as the convolution
between the un-blurred image and the degradation operator.
The degradation operator can be approximated as the point
spread function (PSF) of the system. In the case of SPECT
imaging both the PSF and the actual un-blurred image are not
known. Image de-blurring algorithms are defined by inverse
filtering problems which uses an estimate of the original
image. But blind image de-convolution aims at the
reconstruction of original image without using any prior
information [5], [6]. In the case of SPECT images where little
prior information is available, it can turn out to be more
relevant to apply blind de-convolution algorithms for restoring
the original image from the blurred image.
Many authors have proposed methods to improve the
spatial resolution quality of SPECT images. These methods
can be categorized into two: methods using restoration
algorithms during the reconstruction process, and methods
where the algorithm is performed on the reconstructed images.
In [7] the performance of four widely used filters is compared
for myocardial 99mTc-sestamibi SPECT perfusion studies.
Kundur and Hatzinakos [8] proposed a non-negativity support
constraints recursive inverse filtering (NAS-RIF) algorithm
and it was extended to the 3D SPECT imaging restoration
context in [9]. In [10], [11] the authors proposed Bayesian
tomographic reconstruction methods using structural
information. The SPECT image de-blurring problem was
considered by approximating the blur transfer function with a
two-dimensional symmetric Gaussian function [12] or by
considering a known PSF [13]. In [14] the algorithm imposes
certain spatial-frequency domain constraints on the
reconstruction volume and the PSF.
In this paper we propose a blind de-convolution algorithm
inspired by the idea of maximum likelihood estimation
proposed by [15] and [16]. The algorithm is tested on cardiac
SPECT images, both in gray scale and a suitable color space,
and evaluation is done on the basis of PSNR and blur
reduction.

ISSN: 2277-503X| 2015 Bonfring

Bonfring International Journal of Advances in Image Processing, Vol. 5, No. 1, February 2015

III.

METHODOLOGY

A. Blind De-convolution
A linear space invariant image degradation model is
shown in figure 1. The observed image g(x,y) can be written
as given in (1). The original un-blurred image is represented
by f(x,y), the system transfer function or the PSF is
represented by h(x,y) and n(x,y) represents the noise.

g ( x, y) f ( x, y) h( x, y) n( x, y)

(1)
The original image f(x,y) can be found out by the method
of de-convolution. Since the PSF is unknown the problem
becomes a blind de-convolution problem.

D. Performance Measures Used

Quantitative measures such as peak signal to noise ratio
(PSNR) and blur metric were considered to evaluate the
performance of the proposed method. PSNR is defined as
given in (2).

MaxI2
PSNR 10*log

MSE

MaxIdenotes the maximum intensity in the image and

MSE denotes the mean squared error.
In order to evaluate the amount of reduction in blur using
the proposed method, we consider the blur metric which is
based on the discrimination between different levels of blur
perceptible on the same image [17].
IV.

Figure 1: Image Degradation Model

B. Proposed Method
This paper proposes a method for de-blurring cardiac
SPECT images using maximum likelihood estimate based
blind de-convolution. Both the de-blurred image and the PSF
are estimated using an iterative algorithm without using any
prior information. The maximum likelihood estimate
maximizes the likelihood function p(g/f).

(2)

RESULTS

The application of the proposed method to cardiac nuclear

images obtained from a two collimator single photon emission
computed tomography device was evaluated qualitatively.
Figure 2 shows the blurred image and figure 3 shows the
image de-blurred using the proposed method in YCbCr colour
space. Figure 4 shows the blurred image and figure 5 shows
the de-blurred image using the proposed method in gray scale.

Usually SPECT scans produce both gray scale and colour

images. In colour images the different colours represent
different amount of blood flow. A suitable colour space is
chosen by comparing the performance of different edge
detectors on the image in different colour spaces. Then the
chrominance and luminance channels are separated out. Prefiltering of the channels is done to remove high frequency
drops. The iterative algorithm starts on each of the pre-filtered
channels separately with an initial guess on the actual image
and the PSF, assuming a non-negativity constraint on the
original image.
C. Algorithm
1. Step 1. Identifying a suitable colour space.
A suitable color space is chosen by comparing the
performance of edge detectors in different colour
spaces.
2. Step 2. Converting image into the chosencolour
space.
3. Step 3.Separating luminance and chrominance
channels.
4. Step 4. Making an initial guess on the original image.
5. Step 5. Making an initial guess on the point spread
function.
6. Step 6. Pre-filtering of each channel.
7. Step7. Maximizing the likelihood that the resulting
image, when convolved with the resulting PSF, gives
the blurred image, assuming Poisson noise statistics.

ISSN: 2277-503X| 2015 Bonfring

Figure 2: Blurred Colour Image

Bonfring International Journal of Advances in Image Processing, Vol. 5, No. 1, February 2015

Figure 6: Comparing Blur in the Blurred and De-blurred

Images
Figure 7 shows a plot of PSNR for the blurred and deblurred images. It is found that there is an increase in the
PSNR value for the de-blurred image.

Figure 3: De-blurred Colour Image

Figure 4: Blurred Gray Scale Image

Figure 5: De-blurred Gray Scale Image

In order to evaluate the proposed method quantitatively,
the blur metric [17] and peak signal to noise ratio (PSNR)
were employed. The plots given in figure 6 substantiate
reduction in blur for the proposed method.

Figure 7: Comparing PSNR of the Blurred and De-blurred

Images
The performance of the algorithm is also evaluated for
different number of iterations. Figure 8 gives a plot of
reconstructed PSF for different number of iterations. It is clear
that as the number of iterations increases the reconstructed
PSF becomes closer to the original PSF. But as the number of
iterations increases the PSNR value also increases first but
after a certain number of iterations the PSNR value decreases
as shown in figure 9. As the number of iterations increases
noise gets added to the image. So the number of iterations
needs to be fixed taking into consideration the allowable noise
in the image. The performance of the proposed algorithm in
terms of PSNR is compared with other image de-blurring
techniques as given in table 1. The results show that the
proposed algorithm de-blurs the blurred image with minimum
noise.

ISSN: 2277-503X| 2015 Bonfring

Bonfring International Journal of Advances in Image Processing, Vol. 5, No. 1, February 2015

Figure 8: Reconstructed PSF

Figure 9: PSNR for Different Number of Iterations

Table 1: Comparing the Proposed Method with Other De-blurring Methods
Wiener filter algorithm

Least square error algorithm

Proposed Method

5.56

12.098

29.81

PSNR

V.

CONCLUSION

In this paper a blind de-convolution algorithm suitable for

de-blurring colour as well as gray scale cardiac SPECT images
is proposed. The algorithm makes use of maximum likelihood
estimate approach. As an initial step a suitable colour space is
chosen and then processing is done in that colour space on
luminance and chrominance channels separately. The
performance of the method is evaluated both qualitatively and

quantitatively. Criteria like blur metric and PSNR are used as

quantitative measures. The experimental results show that the
proposed method gives good de-blurring results. The effect of
number of iterations on the image quality is studied. The
effectiveness of the proposed method in de-blurring cardiac
SPECT images is also compared with other de-blurring
methods.

ISSN: 2277-503X| 2015 Bonfring

Bonfring International Journal of Advances in Image Processing, Vol. 5, No. 1, February 2015

REFERENCES
[1]

[2]
[3]

[4]

[5]
[6]
[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

WHO, Cardiovascular diseases (CVDs), Fact sheet number 317 Updated

March 2013 http://www.who.int/ mediacentre/factsheets /fs317/en
/index.html
Global status report on non-communicable diseases 2010.mGeneva,
World Health Organization, 2011.
ArmanRahmim and HabibZaidi, PET versus SPECT: strengths,
limitations and challenges, Nuclear Medicine Communications 2008,
Vol. 29, pp.193207
Paolo G. Camici and Ornella E. Rimoldi,The Clinical Value of
Myocardial Blood Flow Measurement, The Journal of Nuclear
Medicine, Vol. 50, No. 7, July 2009, pp.1076-1087.
G. R. Ayers and J. C. Dainty, Iterative blind deconvolution method and
its applications, Optics Letters, vol. 13, no. 7, pp. 547549, 1988.
R. G. Lane, Blind deconvolution of speckle images, Journal of the
Optical Society of America A, vol. 9, no. 9, pp. 15081514, 1992.
H. Rajabi, A. BitarafanRajabi, N. Yaghoobi, H. Firouzabad, and F.
Rustgou, Determination of the optimum filter function for Tc99msastamibi myocardial perfusion SPECT imaging, Indian Journal of
Nuclear Medicine, vol. 20, no. 3, pp. 7782, 2005.
D. Kundur and D. Hatzinakos, Blind image restoration via recursive
filtering using deterministic constraints, in Proc. Int. Conf. Acoustics
Speech, and Signal Processing, vol. 4, 1996, pp. 547549.
M. Mignotte and J. Meunier, Three-dimensional blind deconvolution of
SPECT images, IEEE Transactions on Biomedical Engineering, vol. 4,
no. 2, pp. 274281, 2000.
G. Gindi, M. Lee, A. Rangarajan, and I. G. Zubal, Bayesian
reconstruction of functional images using anatomical information as
priors, IEEE Transactions on Medical Imaging, vol. 12, no. 4, pp. 670
680, 1993.
X. Ouyang,W. H.Wong, V. E. Johnson, X. Hu, and C. T. Chen,
Incorporation of correlated structural images in PET image
reconstruction, IEEE Transactions on Medical Imaging, vol. 13, no. 4,
pp. 627640, 1994.
M. T. Madsen and C. H. Park, Enhancement of SPECT images by
Fourier filtering the projection set, J. Nucl. Sci., vol. 26, pp. 2687
2690, 1979.
S. Webb, A. P. Long, R. J. Ott, M. O. Leach, and M. A. Flower,
Constrained deconvolution of SPECT liver tomograms by direct digital
image restoration, Med. Phys., vol. 12, no. 1, pp. 5358, 1985
Heng Li and YibinZheng, Application of Blind Deblurring
Reconstruction Technique to SPECT Imaging, Hindawi Publishing
Corporation, International Journal of Biomedical Imaging, Volume
2007, Article ID 63750, pp.1-9
L.A.Shepp, Y.Vardi; Maximum Likelihood Reconstruction for
Emission Tomography, IEEE Transactions on Medical Imaging, Vol.
MI.1, No.2, October 1982, pp.113-122
T. J. Holmes, Blind deconvolution of quantum-limited incoherent
imagery: maximum-likelihood approach, Journal of the Optical Society
of America A, vol. 9, no. 7, pp. 10521061, 1992.
Frederique Crete, Thierry Dolmiere, Patricia Ladret, Marina Nicolas,
The Blur Effect: Perception and Estimation with a New No-Reference
Perceptual Blur Metric, SPIE Electronic Imaging Symposium Conf
Human Vision and Electronic Imaging, San Jose : tats-Unisd'Amrique
(2007)
Neethu M. Sasi is Research Scholar at Model
Engineering College, Kochi, Kerala. She received her
B.Tech Degree in Electronics & Communication
Engineering from Mahatma Gandhi University in 2001
and M.Tech in Signal Processing from National Institute
of Technology in 2010. Her areas of interest include
signal processing, image processing, adaptive signal
processing. (E-mail:neethumsasi@gmail.com)
Jayasree V.K. received M.Tech degree in Biomedical
Engineering from Indian Institute Technology, Chennai,
India in 1998 and PhD degree from Cochin University of
Science and Technology, India in 2009. Currently she is
Head of the Department of Electronics, Govt. Model
Engineering College, Cochin University of Science and
Technology, India. Her research interests include
biomedical image processing and optical communication
techniques
ISSN: 2277-503X| 2015 Bonfring