See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/283817592

Leukemia Detection using Digital Image Processing Techniques

Article · November 2015

DOI: 10.5120/ijais2015451461

CITATIONS READS

16 854

4 authors, including:

Hardik Modi Madhukar B. Potdar

Charotar University of Science and Technology 64 PUBLICATIONS   359 CITATIONS   
39 PUBLICATIONS   188 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

emergency Vehicle Rescuing View project

Brain Tumor Classification from MRI Images using Image Processing View project

All content following this page was uploaded by Hardik Modi on 27 February 2016.

The user has requested enhancement of the downloaded file.

 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

Leukemia Detection using Digital Image Processing

Techniques
Himali P. Vaghela Hardik Modi Manoj Pandya M.B. Potdar
M.Tech Student (Member IEEE) PhD PhD
Dept. of Electronics and Faculty of Electronics Project Manager Project Director BISAG
Communication and Communication BISAG Gandhinagar Gandhinagar
Charusat University, Charusat University,
Changa Changa

ABSTRACT
This paper discusses about methods for detection of leukemia.
Various image processing techniques are used for
identification of red blood cell and immature white cells.
Different disease like anemia, leukemia, malaria, deficiency
of vitamin B12, etc. can be diagnosed accordingly. Objective
is to detect the leukemia affected cells and count it. According
to detection of immature blast cells, leukemia can be
identified and also define that either it is chronic or acute. To
detect immature cells, number of methods are used like
histogram equalization, linear contrast stretching, some
morphological techniques like area opening, area closing,
erosion, dilation. Watershed transform, K means, histogram
equalization & linear contrast stretching, and shape based
features are accurate 72.2%, 72%, 73.7 % and 97.8%
respectively.

Keywords
Blood disease detection, leukemia detection, k means
clustering, watershed transform, histogram equalizing, and
shape based features, count number of red and white cells

1. INTRODUCTION
There are different types of white cells in our body. Leukemia Figure 1. The Formation of Myeloid and Lymphoid
is nothing but cancer of blood cells in which number of white Series of Cell
cells is increasing and those are immature cells that destroy
other cells. Today laboratory test takes longer interval of time 2. RATIONALE
to diagnose the disease of leukemia and it is also time Various methods have been applied to automate the task to
consuming, prone to human error and also tedious. The ratio find out leukemia cells and count it.
of white blood cell in our body is 1000:1. It means that 1
white blood cell is present between 1000 red cell. So if 2.1 Watershed transform
number of white blood cells increase remarkably in large Lim Huey Nee et al. proposed methods for segmentation of
number then the person is succumbed to suffer from the white cells based on morphological operation, gradient
leukemia. It further falls into two type: acute and chronic. If magnitude and watershed transform. First image acquisition
number of white cell is increasing in our body then this techniques is used then segmentation is done to separate the
immature cells start destroying another cells of our body. blast cell and back ground. For this method, first the RGB
Here task is to detect immature cell using different image image is converted into HSV color model and saturated
processing techniques and count total number of cells. So we component is extracted for further processing and then find
the gradient magnitude for the saturation component. This is
need to use the technology that identifies different types of
used for edge detection. Moreover, sobel, canny, prewitt
blood cells within short duration of time in emergency.
Furthermore it is vital to study in detail how to differentiate operators are used for the edge detection. After extracting the
different cell and recognize it as immature cell and according white cells from the image and elimination of the background
and red cells, dilation or erosion process is carried out. Then
to it, detect the leukemia. Acute and chronic also have two
watershed transform is carried out to separate the connected
types. 1) Lymphocytic and 2) myeloblastic that both are due
to immature blast of lymphoid and myeloid cell respectively. cell. Thus, leukemic cell can be identified and this method
gives very accurate result [2]. But the exact separation of
The task can be improved and performed in near-real time cells cannot be done using this method. The process flow is
environment using biomedical image processing. Here shown in figure 2. Figure 2(a) has been taken from web
formation of lymphoid and myeloid cells of series is shown in sources [23].
figure 1. Figure 1 has been taken from web sources [22].

43
 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

2.2 K Means Clustering Technique image. After that RGB image is converted to HSI color model
To identify the abnormalities in blood cells or to identify the and then K-means clustering technique is applied for
lymphoblast, Mashiat Fatma and Jaya Sharma proposed the segmentation. Median filter is used to remove the noise from
method of clustering techniques. After pre-processing of image. After feature extraction, image is classified by
image feature extraction is done that gives useful information clustering techniques. Two types of methods can also be
about image. The pre-processing techniques are only used for used.1) supervised learning and 2) unsupervised learning.
image enhancement. It does not give any necessary Feature extraction is used to identify the white cells or the
information of image. So initially acquisition process is done lymphoblast from image. So some features are used [3].
and then contrast enhancement is necessary to see clear

Figure 2(a) original image Figure 2(b) rgb2 hsv color model Figure 2(c) saturation
component extracted

Figure 2(d) derivation of gradient Figure 2(e) area opening: Figure 2(f) opening-closing
magnitude morphological operation :morphological operation

Figure 2(g) regional maxima Figure 2(h) thresholded opening Figure 2(i) watershed ridge
superimposed on original image closing by reconstruction lines

Figure 2(j) marker and object Figure 2(k) detect white cell using Figure 2(l) detection of white
boundaries superimposed on original colored watershed label matrix cell superimposed transparency
image on original image
Figure 2. Process Flow using Watershed Transform

44
 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

Figure 3(a) original image Figure 3(b) image labeled by cluster index

Figure 3 (c) cluster of red cells Figure 3 (d) cluster of white cells
Figure 3. Process Flow for K means Clustering
Process flow is shown in figure 3 that is used to separate white from RGB to gray level and for contrast enhancement,
cell and red cell using K-means clustering method. Figure 3(a) histogram equalization process is used. Then statistical
has been taken from web sources [24]. parameter like mean and standard deviation is calculated and
erosion or dilation technique is used for morphological
2.3 Histogram Equalization and Linear operation. Here sobel operator is used to detect the edge of
Contrast Stretching cells. After that some shape based features can also be
Raje, Chaitali, and Jyoti Rangole suggested method to detect implemented if needed [6]. Process flow s is given in figure 4.
white cells. To perform this operation, image is first converted Figure 4(a) has been taken from web resources [25].

Figure 4(a) original Figure 4(b) rgb2 gray Figure 4(c) linear Figure 4(d) histogram Figure 4(e) addition of
image contrast equalization 4 (c) and 4(d)

Figure4 (f) Figure4(g) Addition of Figure 4(h) Figure 4(i) Removal of Figure 4(j) Edge
Subtraction of 4(c) fig, 4(e) and 4(f) Thresholding method small particles detection using sobel
and 4(d) operator
Figure 4. Process flow for histogram equalization and linear contrast stretching.

45
 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

2.4 HSI Color Model based Segmentation metrical shape of cells, this shape based features are very
For segmentation of WBC and nucleus, linear contrast useful method to detect different type of cells and their
stretching techniques is used and for color based segmentation shapes.
HSI color model is used. For further segmentation, K means In section 3.1 algorithm is given for performing different
clustering techniques is used. For fully segmented WBC, image processing operation on leukemia detected images.
extract the H component and to segment nucleus S component This algorithm is useful to detect number of overlapping and
is necessary. For further segmentation, k means clustering is non-overlapping cells or to count red and white cells. First
used. Then unwanted object or noise from image are removed RGB image is converted into gray level to reduced dimension
using median filter. Thus cytoplasm and nucleus can be of image. After that thresholdig method is used to convert the
extracted from image [15]. image into binary form for more accurate analysis Otsu’s
Process flow is shown in figure 5. Figure 5(a) has been taken method is best for thresholding, After that some
from web sources [23] morphological operation like area opening, closing, erosion,
dilation. Here, area opening is performed to remove connected
3. METHODOLOGY component Dilation is techniques to add pixels to boundary of
After getting the knowledge about all these techniques, it is objects and erosion is used to remove the pixel on object
concluded that shape based features are used for better result boundaries. After detecting the boundary of object, hole
and accuracy. Shape based features is used to detect different filling operation is performed that is used to detect perfect
shapes like circle, rectangle, ellipse, squares etc. Our blood cell.
cells also have different size and shapes. So to detect geo

Figure 5(a) original image Figure 5(b) contrast enhance by histogram equalization

Figure 5 (e) extracted H component Figure 5(d) detection of white cell using K means
Figure 5. Process Flow of HSI Color Model based Segmentation

46
 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

3.1 Algorithm to Count the Cells Roundness =

Standard deviation=
Read rgb image

Where X= (Mean)
Rgb2gray
According to value of major axis and minor axis, number of
conversion over lapping cells and non-overlapping cells can be detected.
According to detection of cells, number of cell can be
counted.
convert into binary image Figure 6(a) shows original image, figure 6(b) is gray image of
original image. Figure. 6(c) is binary image of original image
and after that area opening is performed in figure 6(d).
Hole filling operation is defined in figure 6(e). After that
boundary is defined for each object that shown in figure 6(f).
Area opening Figure 6(g) is defined overlapped cell with the value of major
axis. Figure 6 (h) and figure 6(i) shows radius and roundness
of cell. Figure 6 (j), figure 6(k) and figure 6(l) represent
features like centroid, standard deviation and bounding box of
cells respectively. This algorithm is performed on MATLAB
Hole filling platform. In MATLAB, ’region props’ properties are used to
find area, roundness, centroid, major axis, minor axis,
standard deviation, etc.

Define object Here the number of actual objects or cells (red cells+ white
cells) are 92. Here, number of non over lapping and
boundary overlapping cells are detected is 68 and 11 respectively. So
the cells which are overlapped on each other are calculated as
(X*2). Where X is number of overlapped cells. So
(11*2=22). Thus, according to algorithm total number of cells
are 90. This result is very accurate and nearly equal to actual
Find Area, perimeter,convex hull, result. Because actual count is 92. So shape based features are
roundness, major axis, minor axis, more accurate and gives 97.8 % accuracy. Hence proved.
standard deviation etc
3.2 Experimental Results

Detect overlapping and non

overlapping cell

Count overlapped and non

overlapped cell

After that hole filling operation, boundary is detected around

the cells. To detect object boundaries many operators are used
like sobel, prewitt, canny, etc. after define the boundary of
object, many shape based features, like major axis, minor Figure 6 (a) original image
axis, area, perimeter, standard deviation, radius, roundness
can be found. Radius, roundness, standard deviation can be
found by given formula:

Radius =

47
International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

Figure 6(b) rgb2gray Figure 6(c) thresholding

Figure 6(d) area opening Figure 6 (e) hole filling

Figure 6 (f) detected boundary Figure 6(g) Detected overlapping cell

48
 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

Figure 6 (h) radius of cell

Figure 6(i) roundness of cell Figure 6(j) centroid of cell

Figure 6(k) standard deviation Figure 6(l) bounding box

Figure 6 Process Flow of Shape based Features

4. CONCLUSION accuracy 97.8 % as shown in table below. To detect different

Various methods have been discussed and evaluated. Merits types of geometrical shape of cells like basophils, eosinophil,
and demerits of different methods are derived from lymphocytes, monocytes etc. shape based features are used
experimental results. It is concluded that proposed method: and according to count of immature cells, disease can be
shape based features finding is more accurate than other diagnosed. These are the limitations of watershed transform,
methods for counting leukemic cells and it also gives highest K means clustering, and histogram equalizing.

49
 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

Method Merit/demerits Accuracy

Watershed transform Merit: Easy method for detection of white cell
Demerits: It cannot give accurate result and cannot 72.2 %
implement on each and every image
K means clustering Merits: It is used for clustering and separate the data based
on value of K.
72 %
Demerits: It does not give classification with labeled data
and also not applicable on incremental data.
Edge detection using histogram Merit: This is very useful method to detect white cell and
equalizing method and linear contrast for contrast enhancement. 73.7 %
stretching
Demerit: It is hard to define boundary of overlapping cell.
Shape based features Merit: Very easy for the detection of white and overlapping
cell and shape of cell
97.8 %
Demerit: This is based on statistics so can get approximate
result.

5. REFERENCES IEEE, 2011.

[1] Mohapatra, Subrajeet, Sushanta Shekhar Samanta, Dipti [9] Mazalan, Siti Madihah, Nurul H. Mahmood, and Mohd
Patra, and Sanghamitra Satpathi. "Fuzzy based blood Azhar Abdul Razak. "Automated Red Blood Cells
image segmentation for automated leukemia detection." Counting in Peripheral Blood Smear Image Using
In Devices and Communications (ICDeCom), 2011 Circular Hough Transform." In Artificial Intelligence,
International Conference on, pp. 1-5. IEEE, 2011. Modelling and Simulation (AIMS), 2013 1st International
[2] Lim, Huey Nee, Mohd Yusoff Mashor, and Rosline Conference on, pp. 320-324. IEEE, 2013.
Hassan. "White blood cell segmentation for acute [10] Akrimi, Jameela Ali, Azizah Suliman, Loay E. George,
leukemia bone marrow images." In Biomedical and Abdul Rahim Ahmad. "Classification red blood cells
Engineering (ICoBE), 2012 International Conference on, using support vector machine." In Information
pp. 357-361. IEEE, 2012. Technology and Multimedia (ICIMU), 2014
[3] Fatma, Mashiat, and Jaibir Sharma. "Identification and International Conference on, pp. 265-269. IEEE, 2014.
classification of acute leukemia using neural network." In [11] Mohapatra, Saurav, Dipti Patra, Sudhakar Kumar, and
Medical Imaging, m-Health and Emerging Siddhartha Satpathi. "Kernel induced rough c-means
Communication Systems (MedCom), 2014 International clustering for lymphocyte image segmentation." In
Conference on, pp. 142-145. IEEE, 2014. Intelligent Human Computer Interaction (IHCI), 2012
[4] Madhloom, H. T., S. A. Kareem, H. Ariffin, A. A. 4th International Conference on, pp. 1-6. IEEE, 2012.
Zaidan, H. O. Alanazi, and B. B. Zaidan. "An automated [12] Ge, Jia, Z. Gong, Jiann-Jong Chen, Jiangchuan Liu, John
white blood cell nucleus localization and segmentation Nguyen, Z. Y. Yang, Chingyue Wang, and Yue Sun. "A
using image arithmetic and automatic threshold." (2010). system for automated counting of fetal and maternal red
[5] Halim, NH Abd, M. Y. Mashor, A. S. Abdul Nasir, N. R. blood cells in clinical KB test." In Robotics and
Mokhtar, and H. Rosline. "Nucleus segmentation Automation (ICRA), 2014 IEEE International
technique for acute leukemia." In Signal Processing and Conference on, pp. 1706-1711. IEEE, 2014.
its Applications (CSPA), 2011 IEEE 7th International [13] Supardi, N. Z., M. Y. Mashor, N. H. Harun, F. A. Bakri,
Colloquium on, pp. 192-197. IEEE, 2011. and R. Hassan. "Classification of blasts in acute leukemia
[6] Raje, Chaitali, and Jyoti Rangole. "Detection of blood samples using k-nearest neighbour." In Signal
Leukemia in microscopic images using image Processing and its Applications (CSPA), 2012 IEEE 8th
processing." In Communications and Signal Processing International Colloquium on, pp. 461-465. IEEE, 2012.
(ICCSP), 2014 International Conference on, pp. 255- [14] Mohammed, Emad, Mostaja MA Mohamed, Christopher
259. IEEE, 2014. Naugler, and Behrouz H. Far. "Chronic lymphocytic
[7] Mohapatra, Subrajeet, and Dipti Patra."Automated leukemia cell segmentation from microscopic blood
leukemia detection using hausdorff dimension in blood images using watershed algorithm and optimal
microscopic images." In Emerging Trends in Robotics thresholding." In Electrical and Computer Engineering
and Communication Technologies (INTERACT), 2010 (CCECE), 2013 26th Annual IEEE Canadian Conference
International Conference on, pp. 64-68. IEEE, 2010. on, pp. 1-5. IEEE, 2013.

[8] Berge, Heidi, Dale Taylor, Sriram Krishnan, and Tania S. [15] Abdul Nasir, A. S., M. Y. Mashor, and H. Rosline.
Douglas. "Improved red blood cell counting in thin blood "Unsupervised colour segmentation of white blood cell
smears." In Biomedical Imaging: From Nano to Macro, for acute leukaemia images." In Imaging Systems and
2011 IEEE International Symposium on, pp. 204-207. Techniques (IST), 2011 IEEE International Conference
on, pp. 142-145. IEEE, 2011.

50
 International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868
Foundation of Computer Science FCS, New York, USA
Volume 10 – No.1, November 2015 – www.ijais.org

[16] Rawat, Jyoti, A. Singh, H. S. Bhadauria, and I. Kumar. Method for the Recognition of Lymphocytes versus
"Comparative analysis of segmentation algorithms for Acute Lymphoblastic Leukemia." In Advanced
leukocyte extraction in the acute Lymphoblastic Computer Science Applications and Technologies
Leukemia images." In Parallel, Distributed and Grid (ACSAT), 2012 International Conference on, pp. 330-
Computing (PDGC), 2014 International Conference on, 335. IEEE, 2012.
pp. 245-250. IEEE, 2014.
[21] Putzu, Lorenzo, and Cecilia Di Ruberto. "White blood
[17] Aimi Salihah, A. N., Mohd Yusoff Mashor, Nor Hazlyna cells identification and counting from microscopic blood
Harun, and H. Rosline. "Colour image enhancement images." World Academy of Science, Engineering and
techniques for acute leukaemia blood cell morphological Technology 7, no. 1 (2013): 363-370.
features." In Systems Man and Cybernetics (SMC), 2010
IEEE International Conference on, pp. 3677-3682. IEEE, [22] (2015) Figure 1 website. [Online] Available at:
2010 www.ufrgs.br

[18] Das, Biplab Kanti, Krishna Kumar Jha, and Himadri [23] (2015) Figure 2(a) website. [Online] Available at
Sekhar Dutta. "A New Approach for Segmentation and www.doctortipster.com
Identification of Disease Affected Blood Cells." In [24] (2015) Figure 3(a) website. [Online] Available at:
Intelligent Computing Applications (ICICA), 2014 www.pathologystudent.com
International Conference on, pp. 208-212. IEEE, 2014..
[25] (2015) Figure 4(a) website.[Online] Available at:
[19] Mohapatra, Subrajeet, and Dipti Patra. "Automated cell www.medicalxpress.com
nucleus segmentation and acute leukemia detection in
blood microscopic images." In Systems in Medicine and [26] (2015) Figure 5(a) website.[Online] Available at:
Biology (ICSMB), 2010 International Conference on, pp. www.prezi.com
49-54. IEEE, 2010. [27] (2015) Figure 6(a) website.[Online] Available at:
[20] Madhloom, Hayan T., Sameem Abdul Kareem, and Hany www.commons.wikimedia.org
Ariffin. "A Robust Feature Extraction and Selection

51

View publication stats