A new method of gridding for spot detection in microarray images

Computer Engineering and Intelligent Systems www.iiste.org
ISSN 2222-1719 (Paper) ISSN 2222-2863 (Online)
Vol.5, No.3, 2014
25
A New Method of Gridding for Spot Detection in Microarray
Images
J.Harikiran1
, D.RamaKrishna2
, B.Avinash3
, Dr.P.V.Lakshmi4
, Dr.R.KiranKumar5
1. Member IEEE and Assistant professor, Department of IT, GITAM University, Visakhapatnam
2. Assistant professor, Department of CSE, GIT, GITAM University
3. Project Fellow, UGC Major Research Project, GIT, GITAM University, Visakhapatnam
4. Professor and Head, Department of IT, GITAM University, Visakhapatnam.
5. Assistant professor, Department of Computer science, Krishna University, Machilipatnam.
Abstract
A Deoxyribonucleic Acid (DNA) microarray is a collection of microscopic DNA spots attached to a solid
surface, such as glass, plastic or silicon chip forming an array. The analysis of DNA microarray images allows
the identification of gene expressions to draw biological conclusions for applications ranging from genetic
profiling to diagnosis of cancer. The DNA microarray image analysis includes three tasks: gridding,
segmentation and intensity extraction. The gridding process is usually divided into two main steps: sub-gridding
and spot detection. In this paper, a fully automatic approach to detect the location of spots is proposed. Each spot
is associated with a gene and contains the pixels that indicate the level of expression of that particular gene.
After gridding, the image is segmented using fuzzy c-means clustering algorithm for separation of spots from the
background pixels. The result of the experiment shows that the method presented in this paper is accurate and
automatic without human intervention and parameter presetting.
Keywords: Microarray Image, Mathematical Morphology, Image Processing
1. Introduction
Microarrays, widely recognized as the next revolution in molecular biology, enable scientists to analyze genes,
proteins and other biological molecules on a genomic scale [1]. A microarray is a collection of spots containing
DNA deposited on the solid surface of glass slide. Each of the spot contains multiple copies of single DNA
sequence [2]. Microarray expression technology helps in the monitoring of gene expression for tens and
thousands of genes in parallel [3]. The processing of the microarray images [5] usually consists of the following
three steps: (i) gridding, which is the process of segmenting the microarray image into compartments, each
compartment having only one spot and background (ii) Segmentation, which is the process of segmenting each
compartment into one spot and its background area (iii) Intensity extraction, which calculates red and green
foreground intensity pairs and background intensities.
Many approaches have been proposed for spot detection in microarray images. Hirata [6] presented an automatic
sub-array and spot gridding method using the horizontal and vertical profile signal of the image. User assistance
was required in this method to fix image rotation and check if the segmentation is correct. This method is valid
only if the sub-array sizes are equal. Jain [7] proposed a gridding algorithm based on axis projection of image
intensity along the rows and columns of the microarray image. The algorithm requires large number of spots and
is not robust to misalignment of different grids. Y.Wang [8] demonstrated a fully automatic gridding
methodology using intensity projection profile of microarray image. The method is sensitive to contaminations
and large number of missing spots. Shuqing Zhao [13] proposed microarray image processing using
mathematical morphology. An improved gridding method based on mathematical morphology is proposed,
which is characterized by filtering out the block noise and filtering projection plots. Several parameters about the
sub-array and spots are required during the gridding and spotting procedure which can be preset in advance or
acquired from database. Deepa.J [14] proposed automatic gridding of DNA microarray images using optimum
subimage. The approach is based on the selection of optimum subimage and the parameters for gridding are
calculated using the intensity projection profile of the sub-image.
In this paper, a fully automatic gridding algorithm for spot detection is presented. After gridding, fuzzy C-means
clustering algorithm is used for segmentation of microarray image into spots and image background. The
algorithm is automatic and accurate for misalignment of spots in microarray image. Furthermore, when we apply
this algorithm on different microarray images, human intervention and parameter presetting is unnecessary. The
paper is organized as follows: section II presents edge detection using adaptive multi-structure morphological
algorithm, Section III presents Morphological filling, Section IV presents gridding algorithm, Section V presents
the Fuzzy c-means clustering algorithm, Section VI presents the qualitative and quantitative results, and finally
Section VI repots conclusions.

Vol.5, No.3, 2014
26
2. Edge Detection Using Adaptive Multistructure Morphological Algorithm
Because of the unicity and fixity of structural element (SE) in traditional Edge detection using mathematical
morphology, there are two main deficiencies: on the one hand, a single SE can only detect the edge of the same
direction with the SE, but is not sensitive to different directions; on the other hand large-scale SE has strong
ability to restrain noise, but the detected edge image is rough; small-scale SE is good at checking the details of
the edge, but weak at noise suppression. In order to effectively restrain noise and preserve image edge
information, we use adaptive multi-structure morphological algorithm to get the edge images [12]. We calculate
the gray scale distance of original image to adaptively define the weights of SEs. The eight structuring elements
of different directions with the size of 5X5 are shown in figure 1.
0 0 0 0 0
0 0 0 0 0
1 1 1 1 1
0 0 0 0 0
0 0 0 0 0
SE with 00
0 0 0 0 0
0 0 0 0 1
0 0 1 0 0
1 0 0 0 0
0 0 0 0 0
SE with 22.50
0 0 0 0 1
0 0 0 1 0
0 0 1 0 0
0 1 0 0 0
1 0 0 0 0
SE with 450
0 0 0 1 0
0 0 0 0 0
0 0 1 0 0
0 0 0 0 0
0 1 0 0 0
SE with 67.50
0 0 1 0 0
0 0 1 0 0
0 0 1 0 0
0 0 1 0 0
0 0 1 0 0
SE with 900
0 1 0 0 0
0 0 0 0 0
0 0 1 0 0
0 0 0 0 0
0 0 0 1 0
SE with 112.50
1 0 0 0 0
0 1 0 0 0
0 0 1 0 0
0 0 0 1 0
0 0 0 0 1
SE with 1350
0 0 0 0 0
1 0 0 0 0
0 0 1 0 0
0 0 0 0 1
0 0 0 0 0
SE with 157.50
Figure 1. SEs with different directions.
a9 a8 a7 a6 a5
a10 a23 a24 a25 a4
a11 a22 a1 a2 a3
a12 a21 a20 a19 a18
a13 a14 a15 a16 a17
Figure 2. Image Sub-block

Vol.5, No.3, 2014
27
Figure 2 shows the sub-block with the size 5X5, in which a1 is the gray-scale value of the center pixel, and
a2,a3,a4…..,a25 stand for its neighborhood gray scale value, then the gray-scale distance of a1 and its
neighborhood can be performed as:
dk= |a1-ak|, k= 2,3,4,….,25. (1)
The larger the gray-scale distance, the higher extent of salutation, and the bigger possibility that the pixel is an
edge point in the image. The Edge Gray-Scale distances of a1 can be defined as follows.
G1(x,y) = d4 + d5 + d6 + d7 + d8 + d9 + d10 + d12 +d13 + d14 + d15 + d16 + d17 + d18 + d19 + d20 + d21 + d23+ d24 +
d25 ; --for direction 00
G2(x,y) = d4 + d5 + d6 + d7 + d8 + d9 + d10 + d12 +d13 + d14 + d15 + d16 + d17 + d18 + d19 + d20 + d21 + d22 + d23+
d24 + d25; --for direction 22.50
G3(x,y) = d2 + d3+ d4 + d6 + d7 + d8 + d9 + d10 + d11 + d12 + d14 + d15 + d16 + d17 + d18 + d19 + d20 + d22 + d23+ d24
; ----for direction 450
G4(x,y) = d2 + d3 +d4 + d5 + d7 + d8 + d9 + d10 + d11 +d12 + d13 + d14 +d15 + d16 + d17 + d18 + d19 + d21 + d22 +d23
+d24+ d25 ;
---for direction 67.50
G5(x,y) = d2 + d3 +d4 + d5 + d6 + d8 + d9 + d10 + d11+d12 +d13 + d14 + d16 + d17 + d18 + d19 + d20 + d21 + d23+ d25
G6(x,y) = d2 + d3 +d4 + d5 + d6 + d7 + d9 + d10 + d11 +d12 + d13 + d14 + d15 + d17 + d18 + d19 + d20+d21 + d22 +d23
+d24+ d25 ;
G7(x,y) = d2 + d3 +d4 + d5 + d6 + d8 + d9 + d10 + d11+d12 +d13 + d14 + d16 + d17 + d18 + d19 + d20 + d21 + d23+ d25
G8(x,y) = d2 + d3 +d4 + d5 + d6 + d7 + d8 + d10 + d11 +d12 + d13 + d14 + d15 + d16 +d18 + d19 + d20+d21 +d24+ d25
;
As for the whole image, the gray-scale distances of each edge and adaptive weights of SEs can be calculated as
below:
EDk = ∑∑
−
=
−
=
1
2
1
2
N
y
M
x
Gk (x,y), k=1,2,….,8 (1)
wk = EDk / ( ∑=
8
1k
EDk ) , k=1,2,….,8 (2)
The edge E extracted adaptively by multi-structure morphology is given by
E= ∑=
8
1k
wk [(I о bk ) bk -( I • bk )Ɵ bk) ] (3)
3. Morphological Filling
A hole may be defined as a background region surrounded by a connected border of foreground pixels.
The filling holes in an image are based on set dilation, complementation and intersection [11]. The following
procedure fill all the holes with 1’s until Xk=Xk-1.
Xk= (Xk-1 B) ∩ Ac
k=1, 2, 3…. (4)

Vol.5, No.3, 2014
28
Where B is a symmetric structuring element and A is the binary image (output of Edge Detection). X0 is an array
containing 0’s with the same size of A except at the locations corresponding to the point in each hole is 1. The
union of Xk with the image A gives the filled region.
4. Automatic Spot Detection Algorithm
A microarray image contains a number of sub-grids and each sub-grid representing a two dimensional array of
spots. Gridding refers to accurately locating each spot within a microarray image namely sub-gridding and spot
detection. The sub-grid detection is done by the method in [10]. After sub-grid detection, the aim is to separate
the sub-grid into spot regions by means of a grid. The output of gridding is to obtain a 2D matrix G of same size
of sub-grid. Initially the values of G[i,j]= 0, for i=1,….,M and j=1,….,N.
The steps of the automatic spot detection algorithm are as follows:
Step 1: Convert the RGB Microarray image into grayscale image
Step 2: Perform Edge detection using method in section 1 on the grayscale microarray image.
Step 3: Perform morphological filling on the edge image obtained from step 2.
Step 4: Calculation of Horizontal and Vertical Intensity profiles
Horizontal and vertical intensity projection profiles of binary image (Morphological Filled Image) are the sum of
pixel intensities along each row and column respectively. Let Mb indicates the filled image of size MxN . Then
the intensity projection profile along ith
row and jth
column are computed using (3) and (4).
Si=
1
( , )
N
pi b
j
M M i j
=
= ∑ i=1…..M (5)
Sj=
1
( , )
M
pj b
i
M M i j
=
= ∑ j=1……..N (6)
Step 5: Calculation of row width (RW) and column width (CW)
The values in Si are used for identification of row width and values in Sj are used for identification of column
width. The values in Si and Sj looks like zeros followed by nonzero again zeros and so on.
The procedure for row width calculation is as follows:
i. For i = 1 to M,
Count the number of zeroes in Si (between the elements in Si having nonzero) = Wyp, Where p = 1,
2, 3 …, k.
Count the number of non-zeroes in Si (between the elements in Si having zero) = Wxp, Where p =
1, 2, 3 …, k.
ii. For p=1,2,…,k Wzp= Wxp + (Wyp + Wy(p+1))/2,
iii. Row width (RW) = median (Wz).
iv. Using the value RW draw horizontal grid lines at positions of i, where i=1, 2 ….. M with step
increment RW.
for i=1to M step RW
for j=1 to N
G[i,j] =1;
The procedure for column width calculation is as follows:
i. For j = 1 to N,
Count the number of zeroes in Sj (between the elements in Si having nonzero) = Wyp, Where p = 1,
2, 3 …, k.
Count the number of non-zeroes in Sj (between the elements in Si having zero) = Wxp, Where p =
1, 2, 3 …, k.
ii. For p=1,2,…,k Wzp= Wxp + (Wyp + Wy(p+1))/2,
iii. Column width (CW) = median (Wz).
iv. Using the value CW draw horizontal grid lines at positions of j, where j=1, 2 ….. N with step
increment CW.
for j=1to N step RW
for i=1 to M
G[i,j] =1;
Step 6: Map these grid matrix G onto the grayscale image.

Vol.5, No.3, 2014
29
Step 7: Compute all the connected components in the gridded optimum image using 8-adjacency. The number of
connected components is equal to the number of spots. Assign a label for pixels in each connected component
generating a label matrix.
Step 8: The pixels with one particular label represents a region of the single spot. Using region properties,
compute the centroid and area for each region (spot). From the centroids of the spots, we can find the distance
between the spots.
4. Segmentation Using Fuzzy C-Means Clustering Algorithm
The Fuzzy C-means [4][9] is an unsupervised clustering algorithm. The main idea of introducing fuzzy concept
in the Fuzzy C-means algorithm is that an object can belong simultaneously to more than one class and does so
by varying degrees called memberships. It distributes the membership values in a normalized fashion. It does not
require prior knowledge about the data to be segmented. It can be used with any number of features and number
of classes. The fuzzy K-means is an iterative method which tries to separate the set of data into a number of
compact clusters. The segmented microarray image using fuzzy c-means is shown in figure 7.
The Fuzzy K-means algorithm is summarized as follows:
Algorithm Fuzzy K-Means(x,n,c,m)
Input:
N=number of pixels to be clustered; x = {x1, x2 ,..., xN}: pixels of microarray image;
c=2: foreground and background clusters; m=2: the fuzziness parameter;
Output: u: membership values of pixels and segmented Image
Begin
Step_1: Initialize the membership matrix uij is a value in (0,1) and the fuzziness parameter m (m=2). The sum of
all membership values of a pixel belonging to clusters should satisfy the constraint expressed in the following.
∑=
c
j 1
uij =1 (7)
for all i= 1,2,…….N, where c (=2) is the number of clusters and N is the number of pixels in microarray image.
Step_2: Compute the centroid values for each cluster cj. Each pixel should have a degree of membership to those
designated clusters. So the goal is to find the membership values of pixels belonging to each cluster. The
algorithm is an iterative optimization that minimizes the cost function defined as follows:
F= ∑∑ ==
c
i
N
j 11
uij
m
|| xj-ci||2
(8)
where uij represents the membership of pixel xj in the ith cluster and m is the fuzziness parameter.
Step_3: Compute the updated membership values uij belonging to clusters for each pixel and cluster centroids
according to the given formula.
(9)
Step_4: Repeat steps 2-3 until the cost function is minimized.
End.
5. Qualitative and Quantitative Results
The proposed spot detection algorithm is performed on a two different microarray slides drawn from the
Stanford microarray Database corresponds to breast category aCGH tumor tissue. The first sub-grid slide is a
261*289 pixel image (Figure 3) that consists of a total of 75429 pixels. The second sub-grid slide is a 559*489
pixel image (Figure 4) that consists of total 273351 pixels. The output of the proposed automatic spot detection
algorithm on two microarray sub-grids is shown in figure3 and figure 4.

Vol.5, No.3, 2014
30
(a) (b) (c)
(d) (e) (f)
(g) (h)
Figure 3: a) Grey scale Image, b) Edge Detection, c) Morphological Filling d) Plot of piM (for all rows) (Si)
e) Plot of pjM (for all columns)( Sj) f) gridded image g) centroids h) segmented image using fuzzy c-means

Vol.5, No.3, 2014
31
(a) (b) (c)
(d) (e) (f)
(g) (h)
Figure 4:, a) Grey scale Image, b) Edge Detection, c) Morphological Filling d) Plot of piM (for all rows) (Si)
e) Plot of pjM (for all columns)( Sj) f) gridded image g) centroids h) segmented image using fuzzy c-means
The accuracy of the gridding algorithm was calculated as
Percentage accuracy = X100 (10)

Vol.5, No.3, 2014
32
The performance of the proposed spot gridding algorithm was evaluated by comparing the results with the
methods in [6], [7], [8] , [13] ,[14] are shown in table 1 .
Table 1: Percentage accuracy of gridding
Method Percentage Accuracy
figure 1
Percentage Accuracy
figure 2
Hirata [6] 87 79
Jain [7] 89 81
Wang [8] 91 82
Shuqing Zhao[13] 90 84
Deepa .J [14] 92 88
Proposed 96 91
After gridding, the segmentation of spots from the background pixels is done by using fuzzy c-means clustering
algorithm. The method is implemented in such a way that the intensity value of each pixel and the pixels of the
image has been grouped in two clusters. The number of pixels clustered as spot and background for two different
microarray images has been presented in Table 2.
Table 2: The number of pixels clustered as spots and background
Method Total
Number of
Pixels
Spots Background
Image 1 75429 40535 34714
Image 2 273351 111669 161682
6. Conclusion
In this paper, a fully automatic gridding method for separating spot centers in microarray sub-grids has been
proposed. The proposed automatically locates the individual spots without any input parameters and human
intervention. It can be proved that percentage accuracy of gridding is high with the methods that use projection
profile of the entire image. After gridding, the image is segmented using fuzzy c-means clustering algorithm.
The proposed method is accurate and automatic, which takes a microarray sub-grid as input image and makes no
assumptions about the size of the spots, rows and columns in the grid.
References
[1] M.Schena, D.Shalon, Ronald W.davis and Patrick O.Brown, “Quantitative Monitoring of gene expression
patterns with a complementary DNA microarray”, Science, 270,199,pp:467-470.
[2] Wei-Bang Chen, Chengcui Zhang and Wen-Lin Liu, “An Automated Gridding and Segmentation method for
cDNA Microarray Image Analysis”, 19th IEEE Symposium on Computer-Based Medical Systems.
[3] Tsung-Han Tsai Chein-Po Yang, Wei-ChiTsai, Pin-Hua Chen, “Error Reduction on Automatic Segmentation
in Microarray Image”, IEEE 2007.
[4] Volkan Uslan, Ihsan Omur Bucak, “Clustering based Spot Segmentation of cDNA Microarray Images “,
32nd
Annual International Conference of IEEE EMBS , IEEE 2010.
[5] J.Harikiran, Dr.P.V.Lakshmi, Dr.R.Kiran Kumar,” Edge detection using mathematical morphology for
gridding of Microarray image”, International Journal of Advanced Research in Computer Science, vol 3, No 2,
pp- 172-175. March 2012.

Vol.5, No.3, 2014
33
[6] Roberto Hirata Jr, Junior Barrera, R.F. Hashimoto, O.D.Daniel,” Microarray Gridding by mathematical
Morphology”, 2001 IEEE, PP. 112-119.
[7] A.N.Jain, Tokuyasu, Snijders,” Fully Automatic Quantification of microarray image data”, Genome research,
12(2), 32-332.
[8] Y.Wang, F.Y.Shih, M.Ma, “Precise Gridding of microarray images by detecting and correcting rotations in
subarrays”, in proceedings of sixth International conference on computer vision, Patternrecognition and Image
Processing, 2010.
[9] J.Harikiran,” Fuzzy C-means with Bi-dimensional Empirical Mode decomposition for segmentation of
Microarray image”, IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 3, September
2012.
[10] L.Reuda, “Sub-grid Detection in DNA Microarray Images”, Proceedings of the IEEE Pacific-RIM
Symposium on image and video technology, pp 248-259, 2007.
[11] R.Gongalez, R.E.Woods,” Digital Image Processing”, 4th
Edition, Printice Hall India, 2010.
[12] J.Harikiran, Dr.P.V.Lakshmi, Dr.R.Kiran Kumar,” Edge detection using mathematical morphology for
gridding of Microarray image”, International Journal of Advanced Research in Computer Science, Volume 3, No
2, pp- 172-175. March 2012.
[13] Shuqing Zhao, Qiaoning Yang, “Microarray Image Processing Based on Mathematical Morphology”,
Proceedings of ICSP 2006, IEEE.
[14] Deepa J, Tessamma Thomas,“ A New Gridding Technique for High Density Microarray Images Using
Intensity Projection Profile of Best Sub Image”, Computer Engineering and Intelligent systems, Vol 4 No 1 : pp
7-18.

A new method of gridding for spot detection in microarray images

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to A new method of gridding for spot detection in microarray images (20)

More from Alexander Decker (20)

A new method of gridding for spot detection in microarray images