Journal of Signal and Information Processing, 2011, 2, 190-195
doi:10.4236/jsip.2011.23026 Published Online August 2011 (http://www.SciRP.org/journal/jsip)
The Role of Combined OSR and SDF Method for
Pre-Processing of Microarray Data That Accounts
for Effective Denoising and Quantification
Jayakishan Meher1, Mukesh Kumar Raval2, Pramod Kumar Meher3, Gananath Dash4
1Department of Computer Science and Engineering, Vikash College of Engineering for Women, Bargarh, Odisha, India; 2Department
of Chemistry, G. M. College, Sambalpur, Odisha, India; 3Department of Embedded Systems, Institute for Infocomm Research, Sin-
gapore; 4Department of Physics, Sambalpur University , Odisha, India.
Received May 8th, 2011; revised June 3rd, 2011; accepted June 11th, 2011.
ABSTRACT
Microarray data is inherently noisy due to the noise contaminated from various sources during the preparation of mi-
croarray slide and thus it greatly affects the accuracy of the gene expression. How to eliminate the effect of the noise
constitutes a challenging problem in microarray analysis. Efficient denoising is often a necessary and the first step to
be taken before the imag e data is analyzed to comp en sa te for data corruption and fo r effective u tiliza tion fo r th ese data .
Hence preprocessing of microarray image is an essential to eliminate the background noise in order to enhance the
image quality and effective quan tification. Existing denoising techniques ba sed on transformed domain have been util-
ized for microarray noise reduction with their own limitations. The objective of this paper is to introduce novel pre-
processing techniques such as optimized spatial resolution (OSR) and spatial domain filtering (SDF) for reduction of
noise from microarray data and reduction of error during quantification process for estimating the microarray spots
accurately to determine expression level of genes. Besides combined optimized spatial resolution and spatial filtering is
proposed and found improved denoising of microarra y data with effective quantifica tion of spots. The propo sed method
has been validated in microarray images of gene expression profiles of Myeloid Leukemia using Stanford Microarray
Database with various quality measures such as signal to noise ratio, peak signal to noise ratio, image fidelity, struc-
tural content, absolute average difference and correlation quality. It was observed by quantitative analysis that the
proposed techniqu e is more efficient for denoising the microarray image which enables to make it suitable for effective
quantification.
Keywords: Denoising, Microarray, Pre-processing, Quantification, Spatial Domain Filtering, Optimized Spatial
Resolution, Quality Measures
1. Introduction
It is well known that microarray technology can monitor
thousand of DNA sequences in a high density array on a
glass [1]. Each microscopic spot represents a single gene.
This technology enables to measure the level of activity
of thousands of genes simultaneously and thus monitor
the whole genome on a single chip so that researchers
can have a big picture of the interactions among those
genes simultaneously. Hence it eliminates “One gene in
one experiment” as in case of wet lab. For a microarray
project, the image quantification makes the transition in
the work flow from wet lab procedures to computational
ones [2,3]. Different technologies are in use to design
microarrays, like microarrays of cDNA, oligonucleotides,
Ink-jet/Bubble jet [4]. In the basic procedure for a mi-
croarray experiment two mRNA samples are reverse-
transcribed into cDNA, labeled using different fluores-
cent dyes (e.g., the red fluorescent dye Cy5 and the green
fluorescent dye Cy3), then mixed and hybridized with the
arrayed DNA sequences. After this competitive hybridi-
zation, the slides are imaged using a scanner which
makes fluorescence measurement for each dye. From the
differential hybridization of the two samples, the relative
abundance of the spotted DNA sequences can be as-
sessed. The results of the microarray experiment are two
16-bit tagged image files, one for each fluorescent dye.
Figure 1 shows the sample subarray of the microarray

The Role of Combined OSR and SDF Method for Pre-Processing of Microarray Data That Accounts for 191
Effective Denoising and Quantification
Figure 1. Microarrary image subarray of myeloid leukemia.
images of Myeloid Leukemia. This technology is widely
used by biologists in the analysis of disease classification,
genome expression in cancer, other genetic diseases, vi-
ral immunology and so on.
The microarray quantification process deals with ex-
traction of spot intensity for each and every probe by
estimating the intensity of foreground and background
values to quantify the expression level of a particular
gene. The images are usually supplied in pairs where one
image shows the intensity of green dye and the other
shows the intensity of a red dye. One of the couple of
most commonly used quantification methods is histo-
gram method, and the other based on pixel intensity [5-7].
Spot detection and quantification based on histogram
method is comprised of three major steps: gridding, seg-
mentation and intensity extraction. One of the major
problems encountered in microarray image processing is
due to the non-uniform spacing between the spots. Find-
ing optimal grid is a time consuming process which re-
quires human intervention. For automatic analysis of
cDNA microarrays, a method using genetic algorithm
approach was implemented to determine optimal grid-
ding and spot segmentation [8]. Dapple has suggested an
intelligent technique to determine the location of spots
using morphological information which is robust to both
variation and artifacts [9]. The authors in [10] present a
fully automatic system for microarray image quantifica-
tion, which locates both subarray grids and individual
spots, without user identification of any image coordi-
nates.
Microarray data is contaminated by noise during the
preparation of slide. The analysis of the scanned images
is not straightforward process since the quality of mi-
croarray images suffer due to noise, artifacts and uneven
background. Without the utilization of a filter, subse-
quent tasks such as spot identification and gene expres-
sion determination cannot be completed. Discrete wave-
let transform (DWT) has been used in image denoising.
The denoising capabilities of DWT is demonstrated in
[11] for removal of noise that is introduced during the
preparation of microarray data. The stationary wavelet
transform (SWT) was introduced in 1996 to make the
wavelet decomposition time invariant [12]. A two-stage
approach for noise removal that processes the additive
and the multiplicative noise component is presented in
[13]. The denoising capabilities of decimated and un-
decimated multiwavelet transforms, DMWT and UMWT
respectively are demonstrated in [14]. The image de-
noising, with spatial filtering techniques as well as hard
and soft thresholding of wavelet coefficients have been
tested in microarray images in [15].
Microarray images consist mostly of low-intensity
features that are not well distinguishable from the back-
ground. These problems lead to errors that propagate to
all the stages of statistical analysis. How to eliminate the
effect of the noise constitutes a challenging problem in
microarray analysis. Denoising the data is required for
effective utilization for these data. Thus there is a need of
developing new methods of denoising for upgradation of
image quality. In this paper we present two novel pre-
processing techniques, namely optimized spatial resolu-
tion and spatial domain filtering to eliminate th e noise in
microarray that helps in more accurate estimation of the
intensity of spots to determine the expression level. Spa-
tial filtering is used for denoising of microarray image
and spatial resolu tion optimizatio n is used to enh ance the
image for accurate quantification of the spots. In order to
improve the quantification result an integrated spatial
domain filtering and optimized spatial resolution has
been used.
The remainder of this paper is organized as follows.
Section 2 presents the propo sed preprocessing techniques
of microarray image. This section presents the principles
of optimized spatial resolution and spatial filtering and
the combined approach. In Section 3 we have discussed
the simulation result and performance analysis. Conclu-
sion is drawn in Section 4.
2. Proposed Pre-processing Techniques of
Microarray Image
The objective of pre-processing a microarray is to elimi-
nate the noise in microarray image that helps in reducing
the error during quantification process for appropriate
interpretation of microarray data. We discuss here the

The Role of Combined OSR and SDF Method for Pre-Processing of Microarray Data That Accounts for 192 Effective Denoising and Quantification
proposed methods for spatial domain filtering, spatial
resolution optimization technique and an integrated ap-
proach of the former two for preprocessing the microar-
ray image for denoising and image enhancement.
2.1. Optimized Spatial Resolution
Spatial resolution is the dens ity of pixels over the image.
The greater the spatial resolution, the more pixels are
used to display the image. The microarray image consists
of large number of pixels. For evaluation of the gene
expression data extensive computation is required be-
cause of the fact that the number of spots is very large
and the data within each spot is also substantial. It is
found that pixel intensities of the microarray are ap-
peared in a particular order in alternate rows. The spatial
resolution can be optimized fo r the entire slide by select-
ing optimal intensity of two consecutive rows without
altering the size of the image in both red plane and green
plane. In this process the red plane and green plane are
separated from the microarray. Each plane is optimized
by selecting the higher values of consecutive rows in red
plane and smaller values of two consecutive rows of
green plane. This property has been selected from the
fact that the signal in tensity is the ratio of red foreground
intensity to th e green foregrou nd inten sity as discu ssed in
Section 3. Blue plane has no role on the signal estimation.
Hence it is left untouched. These modified planes are
restored in a new image resulting in optimized spatial
resolution. This process results in enhanced image qual-
ity and signal intensity.
The histogram plot (Figure 2) shows pixel intensity
having higher magnitude of the effective modified reso
lution as compared to that of original image. When the
optimized image is postprocessed to quantify the spot
intensity, it results better estimation of spot as compared
to processing original raw image.
2.2. Spatial Domain Filtering
In spatial filtering approach, we move a rectangular mask
of the order m × n over the given microarray image. This
process creates a new image with gray values calculated
by the Equation (1 ).
 




12 12
12 12
,,
mn
sm tn
wstpi sj t

 

 (1)
The mask of the order m × n represented as a matrix is
called a filter. A linear filter can be implemented by mul-
tiplying all the elements in the mask by corresponding
elements in the area spanned by the filter mask and add-
ing together all these products [16]. This process is re-
peated for every pixel in the microarray image. For a 3 ×
3 mask with mask values w(i,j) and that corresponding
00.2 0.4 0.6 0.8
1
0
200
400
600
800
1
000
(a)
00.2 0.4 0.6 0.8
1
0
200
400
600
800
1
000
1
200
(b)
Figure 2. Histogram plot of (a) original image and (b) opti-
mized image.
pixel values p(i,j) as shown in Figure 3, the new pixel
value is computed by the Equation (2).
 
11
11 ,,
st
wstpi sj t
 

 (2)
The microarray image obtained by spatial filterin g has
higher intensity of signals and thus background noise is
suppressed thereby improving the signal intensity. This
results higher spot inten sity in the quantification process.
The intensity of the spot depends on the design of the
mask. The mask representing lowpass Spatial filters
based on neighbourhood pixels is denoted as averaging
filter. A mask consisting neighborhood pixe ls of size 3 ×
3 acts as linear filter. The average of all the nine values
within the mask becomes the gray value of the corre-
sponding pixel in the new microarray image. Each pixel
is processed by its neighborhood pixels of mask 3 × 3
and the new pixel is obtai ned by

1
9
Pabcdefghi

   (3)
m(–1,–1) m(–1,0) m(–1,1) p(i–1,j–1) p(i–1,j) p(i–1,j+1)
m(0,–1)m(0,0) m(0,1) p(i,j–1) p(i,j) p(i,j+1)
m(1,–1)m(1,0) m(1,1) p(i+1,j–1) p(i+1,j) p(i+1,j+1)
(a) (b)
Figure 3. (a) mask values, (b) pixel values.

The Role of Combined OSR and SDF Method for Pre-Processing of Microarray Data That Accounts for 193
Effective Denoising and Quantification

The process is continued for the whole image. The
new image obtained is the denoised image. This is called
lowpass spatial filter. Spot intensity is quantified ef-
fecttively when this image is processed. 3 × 3 mask
shows better result in compared to 5 × 5 mask and 3 × 5
mask. Allied to spatial filtering is spatial convolution.
The method for performing a convolution is the same as
that for filtering, except that the filter is rotated by 180˚
before multiplying and adding. Using the w(i,j) and p(i,j)
notation, the output of a convolution with a 3 × 3 mask
for a single pixel is given by Equation (4) or Equation
(5).

11
11
,,
st
wstpisjt
 
 
 (4)
 
11
11 ,, st
wstpi sj t
 

 (5)
Here we have rotated the image pixels by 180˚; this
does not affect the result. In practice, most filter masks
are rotationally symmetric, so that spatial filtering and
spatial convolution will produce the same output. Simi-
larly a 3 × 3 Gaussian filter can effectively eliminate
background noise and image blurring. Gaussian filter of
mask 3 × 3 shows better result than 5 × 5 mask and 3 × 5
mask.
2.3. Combined OSR and SDF Approach
In this paper we have presented two new preprocessing
methods such as the spatial domain filtering and the op-
timized spatial resolution to deal with noise reduction in
microarray image. Both the methods have advantages of
denoising and image enhancement respectively. Spatial
domain method is performed directly on the source im-
ages. Optimization is the simplest spatial domain method,
which needn’t any transformation or decomposition on
the original images. The merit of this method is simple
and fit for real-time processing, Now an integrated spa-
tial domain filtering and optimized spatial resolution ap-
proach is proposed on microarray images to combine the
advantages of both that enables effective quantification.
The flow chart for integrated approach is shown in
Figure 4 that consists of cascaded OSR and SDF stages.
The preprocessing stage encompasses procedures for
optimizing the spatial resolution followed by spatial
lowpass filtering. The combined process not only helps
in denoising but also performs image enhancement re-
sulting in high signal to noise ratio. The resultant image
is now suitable for post processing. The post-processing
entails computing the resulting image for estimation of
spot intensity which follows three steps such as griding,
segmentation and information extraction. It is found that
the integrated approach shows much higher spot intensity
Microarray Image
OSR
SDF
Quantification
(
Post-
p
rocessin
g
)
Figure 4. Flow chart for combined approach.
compared to traditional methods as shown in Table 2.
3. Simulation Result and Performance
Analysis
An important pre-processing step in microarray image is
to eliminate the noise. Extensive study has been made in
eliminating noise and ensures better gene expression by
estimating the spot intensity with the help of novel pro-
posed pre-processing techniques on microarray image.
The denoising methods have been evaluated in 34 mi-
croarray images of gene expression profiles of Myeloid
Leukemia using Stanford Microarray Database [17,18],
each one containing 48 subarrays. The 1632 subarrays
are selected for analysis. In most microarray images ex-
tremely noisy background is met. In a good number of
cases all the proposed methods performed well by elimi-
nating the background noise. The denoising methods
were implemented in Matlab computing environment.
The estimation of microarray spot is quantitatively
analysed using quantification measures known as spot
intensity (SI) or spot quality [5] which is given by
f
g
I
II I
bg
I
fg bg
RR
F
SFB
BG
 
G
(6)
where SI is Spot Intensity, FI is the foreground intensity
and BI is the background intensity. Information is ex-
tracted for each spot on the array in terms of signal in-
tensity which is the mean or median of foreground pixel
intensities within a spot. Spot foreground intensities are
red foreground Rfg and green foreground Gfg. Similarly
spot background intensities are red background Rbg and
green background Gbg intensities. Background informa-
tion is extracted by the existing morphological opening
which is a non-linear filter that generates an image of the

The Role of Combined OSR and SDF Method for Pre-Processing of Microarray Data That Accounts for 194 Effective Denoising and Quantification
estimated background intensity for the entire slide. Spot
quality is the ratio of foreground to the backgro und ratio.
The Spot intensity is computed by several preprocessing
techniques such as spatial domain filtering based on
neighborhood mask processing, Gaussian mask, median
filter, ordered filter, Wiener filter, frequency domain
filter and wavelet transform filter on different microarray
slides and found that the proposed integrated approach
produces high spot intensity that results better gene ex-
pression.
On the other hand, the quality assessment parameters
that ar e used to ev aluate the performance of noise reduc-
tion by different methods employed in this paper are
signal to noise ratio (SNR), peak signal to noise ratio
(PSNR), image fidelity (IF), structural content (SC), ab-
solute average difference (AAD) and Correlatio n Quality
(CQ) [19] which are given by

 
2
,
2
,
,
,,
xy
d
xy
Ixy
SNR
I
xyI xy


 (7)

 

2
,
2
,
max, )
,,
xy
d
xy
M
NIxy
PSNR IxyIxy



 (8)
1
1IF SNR  (9)


2
,
2
,
,
,
xy
d
xy
I
xy
SC
I
xy


 (10)
 
,,,
*
d
xy
I
xyI xy
AAD
M
N


 (11)



,
,
,* ,
,
d
xy
xy
I
xyI xy
CQ Ixy 

 (12)
In general, higher SNR and PSNR are better; the signal
is cleaner and indicates a smaller difference between the
original and denoised image. If IF nears 1, we will obtain
an image of better quality. Similarly if SC spread at 1, we
will obtain image of better quality. A lower AAD gives a
cleaner image as more noise is reduced. A larger value of
CQ usually corresponds to a better quantitative perform-
ance. In this work, we have simulated all the proposed
methods in Matlab 7.0 for preprocessing and postproc-
essing operations.
Table 1 summarizes the quality assessment parameters
for denoising methods on the microarray image. The de-
noising methods have been tested in microarray images
of gene expression profiles of Myeloid Leukemia using
Stanford Microarray Database. In Table 1 the mean val-
ue of the parameters of the each method are shown esti-
mated from the evaluation set of 408 subarrays. Table 2
summarizes spot intensity of one spot in sample mi-
croarray subarray. It is found that the proposed spatial
domain filtering based on neighborhood mask filtering
and Gaussian mask filtering play vital role in reducing
noise in microarray image and thus helps in quantifica-
tion effectively. Further optimized spatial resolution en-
hances the image quality and thus improves the signal
intensity. Hence the optimized image produces improv ed
estimation of spot intensity when post-processed. Simu-
lation result shows that the integrated optimized spatial
resolution and spatial lowpass filtering produces much im-
proved result than that has been done by raw image and
other preprocessing methods. The quantitative results of
Tables 1 and 2 show the optimum assessment parameters
Table 1. Simulation results on assessment parameters for
denoising methods.
Denoising
Methods SNRPSNRIF AAD SC CQ
Gaussian
filter 62.1234.870.98 0.02 1.0277.65
Wiener filter59.3431.260.98 0.06 1.0472.24
Median filter36.2325.730.95 1.004 1.0172.16
Ordered
filter 3.78 15.240.75 15.52 0.56578.64
Frequency
domain 6.25 17.220.79 3.19 0.9953.25
OSR 68.0838.540.98 0.02 1.0779.38
Spatial LPF69.3739.530.99 0.007 1.0980.01
Table 2. Estimation of spot intensity by different methods.
Preprocessing Methods Spot Intensity
Processing raw image 1.23
Frequency domain LPF 1.50
Gaussian mask 1.56
Median filter 1.54
Ordered filter 1.54
Wiener filter 1.55
Optimized spatial resolution (OSR) 1.58
Spatial LPF (neighbourhood mask) 1.59
Integrated approach (OSR+SDF) 2.15

The Role of Combined OSR and SDF Method for Pre-Processing of Microarray Data That Accounts for
Effective Denoising and Quantification
195
for the proposed method. Thus the enhanced image qual-
ity and improved deno ising capability are observed esp e-
cially in integrated approach and hence it helps in quan ti-
fication effectively.
4. Conclusions
In this paper a new approach using integrated optimized
spatial resolution and spatial domain filtering is pre-
sented to deal with enhancement of image quality and re-
duction of noise for effective quantification of microarray
image. Results computed using proposed optimized spa-
tial resolution and spatial lowpass filtering show im-
proved interpretation in contrast to one derived from the
raw data analysis and other preprocessing methods. Var-
ious microarray images from Stanford Microarray Data-
base were examined to validate the performance of our
methods. Experimental results show that the proposed
algorithm provides better performance than the other
methods from quantitative analysis. The information ex-
tracted on down and up regulation of genes under dis-
eased condition can be further analyzed in pharmacoge-
nomics for drug targeting and drug development for the
disease.
REFERENCES
[1] T. K. Attwood and D. J. Parry-Smith, “Introduction to
Bioinformatics,” Addison Wesley Longman Limited, Har-
low, 1999.
[2] R. C. Y. Cheung and C. J. S. Desilva, “Analysis of Gene
Microarray Image,” Neural Information Processing, Pro-
ceedings of ICONIP’99, 6th International Conference,
Perth, Vol. 2, November 1999, pp. 627-632.
[3] R. S. H. Istepanian, “Microarray Image Processing: Cur-
rent Status & Future Directions,” IEEE Transactions on
NanoBioScience, Vol. 2, No. 4, December 2003, pp. 173-
175.
[4] J. Dopazo, “Microarray Data Processing and Analysis,”
Kluwer Academic Publisher, Boston, 2002, pp. 43-63.
[5] D. Vijayan and A. S. Nair, “Microarray Image Processing:
Spot Detection, Quantization and Clustering,” Center for
Bioinformatics, KU, March 2006.
[6] B. Alhadidi, H. N. Fakhouri and O. S. AlMousa, “cDNA
Microarray Genome Image Processing Using Fixed Spot
Position,” American Journal of Applied Sciences, Vol. 3,
No. 2, 2006, pp. 1730-1734.
[7] S. H. Ni, P. Wang, et al., “Spotted cDNA Microarray Im-
age Segmentation Using ACWE,” Romanian Journal of
Information Science a nd Tec hnology , Vol. 12, N o. 2, 2009,
p. 249.
[8] A. Sreedevi and D. S. angamashetti, “Automatically Lo-
cating Spots in DNA Microarray Image Using Genetic
Algorithm without Gridding,” IEEE IACSIT Spring Con-
ference, Singapore, 17-20 April 2009, pp. 178-181.
[9] J. Buhler, T. Ideker and D. Haynor, “Dapple: Improved
Techniques for Finding Sports on DNA Microarrays,”
Technical Reports UWTR 2000-08-05, University of Wa-
shington, 2000.
[10] A. N. Jain, T. A. Tokuyasu, A. M. Snijders, R. Seg raves, D.
G. Albertson and D. Pinkel1, “Fully Automatic Quantifi-
cation of Microarra y Image Data, ” Genome Resear ch, Vol.
12, No. 2, February 2002, pp. 325-332.
doi:10.1101/gr.210902
[11] T. Tokuyasu, D. Albertson, D. Pinkel and A. Jain, “Wa-
velet Transforms for the Analysis of Microarray Experi-
ments,” Proceedings of IEEE Computer Society Bioin-
formatics Conference, 11-14 August 2003, pp. 429-430.
[12] X. H. Wang, R. S. H. Istepanian and Y. H. Song, “Mi-
croarray Image Enhancement by Denoising Using Sta-
tionary Wavelet Transform,” Transactions on Nanobio-
science, Vol. 2, No. 4, December 2003, pp. 184-189.
[13] H. Stefanou1, T. Margaritis, D. Kafetzopoulos, K. Marias
and P. Tsakalides, “Microarray Image Denoising Using a
Two-Stage Multiresolution Technique,” IEEE Interna-
tional Conference on Bioinformatics and Biomedicine,
2007, pp. 383-389.
[14] A. Zifan, M. H. Moradi and S. Gharibzadeh, “Microarray
Image Enhancement by Denoising Using Decimated and
Undecimated Multiwavelet Transforms,” Signal, Image
and Video processing, Vol. 4, No. 2, 2009, pp. 177-185.
[15] A. Mastrogianni, E. Dermatas and A. Bezerianos, “Mi-
croarray Image Denoising Using Spatial Filtering and
Wavelet Transformation,” IFMBE Proceedings, Vol. 23,
No. 1, 2009, pp. 594-597.
[16] R. C. Gonzalez, R. E. Wood and S. L. Eddins, “Digital
Image Processing Using Matlab,” Pearson Education, Inc.,
Berkeley, 2004.
[17] J. Gollub, C. A. Ball, G. Binkley, et al., “The Stanford
Microarray Database: Data Access and Quality Assess-
ment Tools,” Nu cleic Acids Res earch, Vol. 31, No . 1, 2003,
pp. 94-96.
[18] C. A. Ball, I. A. B. Awad, J. Demeter, et al., “The Stanford
Microarray Database Accommodates Additional Mi-
croarray Platforms and Data Formats,” Nucleic Acids Re-
search, Vol. 33, Database Issue, 2005, pp. D580-D582.
[19] M. Mastriani and A. E. Giraldez, “Microa rrays Denoising
via Smoothing of Coefficients in Wavelet Domain,” In-
ternational Journal of Biomedical Sciences, Vol. 1, No. 1,
2006, pp. 7-14.