Volume 2 Special Issue ISSN 2079-8407

Journal of Emerging Trends in Computing and Information Sciences

©2010-11 CIS Journal. All rights reserved.

 
http://www.cisjournal.org 

Performance Evaluation of Rician Noise Reduction Algorithm in

Magnetic Resonance Images
Milindkumar V. Sarode#1, Dr.Prashant R. Deshmukh*2
#
Department of Computer Science and Engineering
Jawaharlal Darda Institute of Engineering& Technology, Yavatmal (M.S.) India
*
Department of Computer Science & Engineering and Information Technology
SIPNA’s College of Engineering Amravati (M.S.) India
1
milind09111970@gmail.com , 2prdeshmukh@ieee.org

ABSTRACT
The objective of this paper is to introduce an estimation of the noise in the Magnetic Resonance images. We propose a
method for Rician noise reduction. This method shows an optimal estimation result that is more accurate in recovering the
true signal from Rician noise. The method proposed specifically for Rician noise reduction, but because Rician noise can
be approximated to Gaussian when SNR is high, therefore, we expect the proposed algorithm also has advantage in
denoising of complex MR images.

Keywords— Recian noise, Speckle noise, Denoising. Signal-to noise ratio, MRI (magnetic Resonance Images).
I. INTRODUCTION
II. DENOISING
Denoising of magnetic resonance (MR) images is
a critical issue, signal-to-noise ratio (SNR), and
acquisition speed, which results in images that still
demonstrate significant noise levels. Sources of MR noise
include thermal noise , sample resolution, etc. To
understand the spatial distribution of noise in an MR
image is critical to any attempt to estimate the true signal.
The investigation of how noise is distributed in MR
images is historical. Pure noise in MR magnitude images
can be modelled as a Rayleigh distribution. To overcome
the problem of smoothing, spatial information can be
incorporated so that spatial regularization of the noise
estimation is performed rather than the simple smoothing
of both signals and noise. In the signal processing
literature, many of the popular denoising algorithms
suggested are based on wavelet thresholding. These
approaches attempt to separate significant features/signals
from noise in the frequency domain and simultaneously
preserve them while removing noise. If the wavelet
transform is applied on MR magnitude data directly, both
the wavelet and the scaling coefficients of a noisy MRI
image are biased estimates of their noise-free counterparts.
The difficulty with wavelet or anisotropic diffusion
algorithms is again the risk of over- smoothing fine
details, particularly in low SNR images . From a statistical
point of view, maximum likelihood estimation (MLE) is
known to yield optimal results . MLE can be used to
accommodate various noise hypotheses. In this paper, we
use intensities of pixels located in the non local
neighbourhoods of a certain pixel to estimate its true noise
free signal. In this method, noise is modeled as Rician and
MLE is deployed within a non local neighbourhood to
predict noise free signals.
Denoising is the process of removing noise from
a signal. Noise reduction techniques are conceptually very
similar regardless of the signal being processed, however a
prior knowledge of the characteristics of an expected
signal can mean the implementations of these techniques
vary greatly depending on the type of signal. Image
denoising is often used in the field of photography or
publishing where an image was somehow degraded but
needs to be improved before it can be printed. For this
type of application we need to know something about the
degradation process in order to develop a model for it.
When we have a model for the degradation
process, the inverse process can be applied to the image to
restore it back to the original form. This type of image
restoration is often used in space exploration to help
eliminate artifacts generated by mechanical jitter in a
spacecraft or to compensate for distortion in the optical
system of a telescope. Image denoising finds applications
in fields such as astronomy where the resolution
limitations are severe, in medical imaging where the
physical requirements for 2 high quality imaging are
needed for analyzing images of unique events, and in
forensic science where potentially useful photographic
evidence is sometimes of extremely bad quality .Noise can
be random or white noise with no coherence or coherent
noise introduced by the devices mechanism or processing
algorithm In the case of photographic film and magnetic
tape noise (both visible and audible) is introduced due to
the grain structure of the medium. In photographic film,
the size of the grains in the film determines the film's
sensitivity, more sensitive film having larger sized grains.
In magnetic tape, the larger the grains of the magnetic
particles, the more prone the medium is to noise.

  39
 Volume 2 Special Issue
Journal of Emerging Trends in Computing and Information Sciences
A. Rician Distribution
The image intensity in magnetic resonance
magnitude images in the presence of noise is to be
governed by a Rician distribution. Low signal intensities
(SNR < 2) are therefore biased due to the noise. The
noise can be estimated from the images and a simple
correction scheme is provided to reduce the bias. The
noise characteristics in phase images are also studied and
shown to be very different from those of the magnitude
images. Common to both, however, is that the noise
distributions are nearly Gaussian for SNR larger than two.
"Rician noise" depends on the data itself, it is not additive,
so to "add" Rician noise to data, what we really mean is
make the data Rician distributed.
B. Non Local Means Algorithm
Non local method does the review of the salient
features of the approach here. We define a lattice L of
points in space. Then a noisy image I is a map of signals
plus noise given by
I   im iL
where im is the noisy image value. A neighbourhood ηi of
pixel is defined as a set of pixels such that for all i, j L
i i (i)
j   i  i  j (ii)
For example, a 5 5 window can be considered as
a neighbourhood for pixel . Correspondingly, {im i
ηs} is denoted by a vector of intensities V(ηs). A Non
Local neighbourhood of pixel i is defined as any set of
pixels j in an image such that ηj is similar to
similarity between two neighbourhoods ηj and is to be
measured
by the quadratic Euclidean distance as given in equation
below. The Non Local method estimates a noise free
image
^ ^
I F  {a i i  I  by a i   j  I Wij mj
where the weights { Wij } satisfy 0 Wij
Wij L, NF is a normalizing factor given by
(
 v ( i ) v ( j )
)2
 je e
h
N F 
The parameter h controls the decay of the
exponential function. This shows that the estimated value
of âi is computed as a weighted average of all the pixels in
the image, and pixels with similar gray level
neighbourhoods to υ( ) will have larger weights in the
average Non Local means will produce an optimal
denoising result if the noise can be modelled as Gaussian.
ISSN 2079-8407
©2010-11 CIS Journal. All rights reserved.
http://www.cisjournal.org 
III. VERITIES OF NOISE
A. Gaussian Noise
Gaussian noise is statistical noise that has a
probability density function (abbreviated pdf) of the
normal distribution (also known as Gaussian distribution).
In other words, the values that the noise can take on are
Gaussian-distributed. It is most commonly used as
additive white noise to yield additive white Gaussian noise
(AWGN) Gaussian noise is properly defined as the noise
with a Gaussian amplitude distribution. This says nothing
of the correlation of the noise in time or of the spectral
density of the noise. Labeling Gaussian noise as 'white'
describes the correlation of the noise. It is necessary to use
the term "white Gaussian noise" to be correct.
B. Salt and Pepper Noise
Salt and pepper noise is an impulse type of noise,
which is also referred to as intensity spikes. This is caused
generally due to errors in data transmission. It has only
 two possible values, a and b. The probability of each is
typically less than 0.1. The corrupted pixels are set
alternatively to the minimum or to the maximum value,
giving the image a “salt and pepper” like appearance.
Unaffected pixels remain unchanged. For an 8-bit image,
the typical value for pepper noise is 0 and for salt noise
255. The salt and pepper noise is generally caused by
malfunctioning of pixel elements in the camera sensors,
faulty memory locations, or timing errors in the
digitization process.
C. Speckle Noise
. The
Speckle noise is a granular noise that inherently
exists in and degrades the quality of the active radar and
synthetic aperture radar (SAR) images. Speckle noise in
conventional radar results from random fluctuations in the
return signal from an object that is no bigger than a single
image-processing element. It increases the mean grey level

of a local area. Speckle noise in SAR is generally more
serious, causing difficulties for image interpretation. It is
1, Σ j caused by coherent processing of backscattered signals
from multiple distributed targets. In SAR oceanography,
for example, speckle noise is caused by signals from
elementary scatterers, the gravity-capillary ripples, and
manifests as a pedestal image, beneath the image of the
sea waves.
IV. MAGNETIC RESONANCE IMAGING
Magnetic resonance imaging (MRI), or nuclear
magnetic resonance imaging (NMRI), is primarily a
medical imaging technique most commonly used in
radiology to visualize detailed internal structure and
limited function of the body. MRI provides much greater
contrast between the different soft tissues of the body than
  40
 Volume 2 Special Issue
Journal of Emerging Trends in Computing and Information Sciences
©2010-11 CIS Journal. All rights reserved.
http://www.cisjournal.org
computed tomography (CT) does, making it especially
useful in neurological (brain), musculoskeletal,
cardiovascular, and ontological (cancer) imaging. Unlike
CT, it uses no ionizing radiation, but uses a powerful
magnetic field to align the nuclear magnetization of
(usually) hydrogen atoms in water in the body. Radio
frequency (RF) fields are used to systematically alter the
alignment of this magnetization, causing the hydrogen
nuclei to produce a rotating magnetic field detectable by
the scanner. This signal can be manipulated by additional
magnetic fields to build up enough information to
construct an image of the body. A radio frequency
transmitter is briefly turned on, producing an
electromagnetic field. In simple terms, the photons of this
field have just the right energy, known as the resonance
frequency, to flip the spin of the aligned protons. As the
intensity and duration of the field increases, more aligned
spins are affected. After the field is turned off, the protons
decay to the original spin-down state and the difference in
energy between the two states is released as a photon. It is
these photons that produce the signal which can be
detected by the scanner. The frequency at which the
protons resonate depends on the strength of the magnetic
field. As a result of conservation of energy, this also
dictates the frequency of the released photons. An image
can be constructed because the protons in different tissues
return to their equilibrium state at different rates. By
changing the parameters on the scanner this effect is used
to create contrast between different types of body tissue or
between other properties, as in fMRI and diffusion MRI.
Contrast agents may be injected intravenously to enhance
the appearance of blood vessels, tumours or inflammation.
Contrast agents may also be directly injected into a joint in
the case of arthrograms, MRI images of joints. Unlike CT,
MRI uses no ionizing radiation and is generally a very safe
procedure. Nonetheless the strong magnetic fields and
radio pulses can affect metal implants, including cochlear
implants and cardiac pacemakers. In the case of cardiac
pacemakers, the results can sometimes be lethal, so
patients with such implants are generally not eligible for
MRI. MRI is used to image every part of the body, and is
particularly useful for tissues with many hydrogen nuclei
and little density contrast, such as the brain, muscle,
connective tissue and most tumours. In clinical practice,
MRI is used to distinguish pathologic tissue (such as a
brain tumour) from normal tissue. One advantage of an
MRI scan is that it is believed to be harmless to the
patient. It uses strong magnetic fields and non-ionizing
radiation in the radio frequency range. Compare this to CT
scans and traditional X-rays which involve doses of
ionizing radiation and may increase the risk of
malignancy, especially in a fetus.While CT provides good
spatial resolution (the ability to distinguish two structures
an arbitrarily small distance from each other as separate),
MRI provides comparable resolution with far better
contrast resolution (the ability to distinguish the
differences between two arbitrarily similar but not
identical tissues). The basis of this ability is the complex
library of pulse sequences that the modern medical MRI
ISSN 2079-8407
 
scanner includes, each of which is optimized to provide
image contrast based on the chemical sensitivity of MRI.
V. NOISE IN MAGNETIC RESONANCE IMAGES
In Magnetic Resonance Images, raw data is
intrinsically complex valued and corrupted with zero mean
Gaussian distributed noise with equal variance. After
inverse Fourier transformation, the real and imaginary
images are still Gaussian distributed given the
orthogonality and linearity of the Fourier transform. MR
magnitude images are formed by simply taking the square-
root of the sum of the square of the two independent
Gaussian random variables (real and imaginary images)
pixel by pixel. After this nonlinear transformation, MR
magnitude data can be shown to be Rician distributed. For
an MR magnitude image defined on a discrete grid L,
I  {mi i  I } , the probability distribution function
(PDF) of I is
mi (mi2  A2 ) / 2 2 A.mi
P (mi A)  e IO( )
 2  2
where I0 is the zeroth-order modified Bessel function of
the first kind, A is the noise-free signal amplitude, and is
the standard deviation of the Gaussian noise in the real and
imaginary images. SNR is defined here as A /  . When
SNR is high, the Rician distribution approaches a
Gaussian; when SNR approaches 0, the Rician distribution
becomes Rayleigh distributed and the PDF becomes
 mi2
mi 2
P (mi )  e
2
VI. EVALUATION AND IMPLEMENTATION
To evaluate and compare the technique with
conventional and Non Local -means algorithms, simulated
MR magnitude data and MR data. Generated MR
magnitude data by adding Rician noise in noise-free
images. Four criteria (display, SNR, method error, and
sharpness measurement) were taken into account in the
method comparison. Each criterion measures a specific
means of the denoising method. Display shows how well
the denoised image can be visually interpreted. The SNR
aims at exploring how much the noise has been reduced.
The method error and sharpness metrics are appropriate
for measuring the sharpness of an image. The perception
of sharpness is related to the clarity of detail and edge
definition of a denoised image.
  41
 Volume 2 Special Issue ISSN 2079-8407
Journal of Emerging Trends in Computing and Information Sciences

©2010-11 CIS Journal. All rights reserved.

 
http://www.cisjournal.org 

  Figure 2. Shows Adding variance for rician noise

Flow Diagram
A. Generation of Rician Noise

Rician noise is not additive, but is instead data-

dependent Consider a set of random numbers, which we
take to be the intensity values of a noise-free MR image A
defined
on a discrete grid M so that A  {ai i  M } . Let σ be
the standard deviation of Gaussian noise. There are two
sets of Gaussian distributed random numbers
X  {xi i  M } and Y  { yi i  M } with zero
mean and identical standard deviations .Then the
following I  {mi i  M } are Rician distributed.

mi  (ai  xi) 2  y
Figure 3. Shows Noised Image
VII. TESTING AND EVALUATION

On execution of our program we get the visual

results as below. The snapshots are given as follows:

Figure 1: Shows the original image of the brain 

Figure 1. Shows Original Image Figure 4. Showing denoised image with NL-Denoising
  42
 Volume 2 Special Issue ISSN 2079-8407
Journal of Emerging Trends in Computing and Information Sciences

©2010-11 CIS Journal. All rights reserved.

 
http://www.cisjournal.org 

VIII. APPLICATIONS

Following are the applications for Denoising of

Magnetic Resonance Images:
1. Denoising of MRI is of importance for clinical
diagnosis such as tissue classification.
2. Denoising of MRI is of importance for computerized
analysis such as Segmentation.
3. NL-Means has many successful application such as in
restoring textures
4. NL-Means is used in denoising image sequences such as
video and movies.
5. Rician noise actually beneficial in Dithering.

IX. CONCLUSION
In this experimental study, we carried out the
brief review of the characteristics of noise in MR images.
The procedure is used for reducing Rician noise in MR
Figure 5. Showing denoised image with LML Denoising
images, and shows how the method visually compared to
the NL means algorithm and also using the term of signal
to noise ratio, sharpness. We focussed on low SNR
situations and demonstrated the outperformed NL means
under the visual, SNR and method error metrics.
Additionally, algorithm outperformed by better preserving
tissue boundaries in the visual comparisons, and also in
the sharpness. This method is used here specifically for
Rician noise reduction, but because Rician noise can be
approximated to Gaussian when SNR is high, therefore,
we expect this algorithm also has advantage in denoising
raw complex MR images. Testing and evaluation section
shows actual implementation results.

REFERENCES
[1] Lili He* and Ian R. Greenshields, “A Nonlocal
Maximum Likelihood Estimation Method for
Figure 6. Showing denoised image with NLML Denoising Rician Noise Reduction in MR Images”, IEEE
Transaction on Mrdical Imaging, Vol. 28, No.
2, February 2009. PP. 165-172.

[2] W. A. Edelstein, P. A. Bottomley, and L. M. Pfeifer,

“A signal-to-noise calibration procedure for nmr
imaging systems,” Med. Phys., vol. 11,no. 2, pp.
180–185, 1984.

[3] E. R. McVeigh, R. M. Henkelman, and M. J.

Bronskill, “Noise and filtration in magnetic
resonance imaging,” Med. Phys., vol. 12, no. 5,pp.
586–591, 1985.

[4] R. M. Henkelman, “Measurement of signal

intensities in the presence of noise in mr
images,” Med. Phys., vol. 12, no. 2, pp. 232–
233, 1985.

[5] M. A. Bernstein, D. M. Thomasson, and W. H.

Figure 7. Showing Performance parameter Perman, “Improved detectability in low signal-
  43
 Volume 2 Special Issue ISSN 2079-8407
Journal of Emerging Trends in Computing and Information Sciences

©2010-11 CIS Journal. All rights reserved.

 
http://www.cisjournal.org 

to-noise ratio magnetic resonance images by
means of phase-corrected real construction,”
Med. Phys., vol. 16,no. 5, pp. 813–817, 1989
[6] A. Achim, E.E. Kuruoglu and J. Zerubia, “ SAR
image filtering based on the heavy – tailed
Rayleigh model”, IEEE Trans. Image Process.,
Vol. 1, No. 9, pp. 2686-2693, Sept. 2006.
[11] P. Perona and J. Malik, “Scale-space and edge
detection using anisotropic diffusion,” IEEE
Trans. Pattern Anal.Mach. Intell., vol. 12, no. 5,
pp. 629–639, May 1990.
[12] C. Tomasi and R. Manduchi, “Bilateral filtering
for gray and color images,” in Proc.IEEE Int.
Conf. Computer Vision, 1998, pp. 839–846.

[7] R. F. Wagner, M. F. Insana, “ Analysis of Ultrasound [13] K. J. Overton and T. E. Weymouth,“A noise
Image Texture via Generalised Rician Statistic”, reducing preprocessing algorithm,” in Proc.
Proc. SPIE 556, pp. 153-159, 1985. IEEE Computer Science Conf. Pattern
Recognition and Image Processing, Chicago,
[8] M. J. Black, G. Sapiro, D.H. Marimont and D. IL, 1979, pp.498–507.
Heeger, “ Robust anisotropic diffusion”, IEEE
Trans. Imag. Proc., vol. 7, No. 3, pp. 412-432, March
[14] H. Lin and A. N.Willson Jr., “Median filters
1998.
with adaptive length,” IEEE Trans. Circuits
Syst., vol. 35, no. 6, pp.675–690, Jun. 1988.
[9] S. G. Chang, Y. Bin, and M. Vetterli, “
Adaptive wavelet thresholding for image
[15] T. Sun and Y. Neuvo, “Detail-preserving
denoising and compression”, IEEE Trans. On
median based filters in image processing,”
Image Processing, vol. 9, No. 9. pp. 1532-1546,
Pattern Recognit. Lett., vol. 15, pp. 341–347,
Sept. 2006.
Apr. 1994.
[10] Michael S. Moore, Moncef Gabbouj, Sanjit K.
Mitra, “Vector SD-ROM Filter For Removal Of
Impulse Noise From Color Images”,
ECMCS’99 EURASIP Conference, DSP for
Multimedia Communications and Services,
Krakow, 24 - 26 June.

  44