Preliminary process in blast cell morphology identification based on image segmentation methods

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 10, No. 6, December 2020, pp. 5714~5725
ISSN: 2088-8708, DOI: 10.11591/ijece.v10i6.pp5714-5725  5714
Journal homepage: http://ijece.iaescore.com/index.php/IJECE
Preliminary process in blast cell morphology identification
based on image segmentation methods
Retno Supriyanti1
, Pangestu F. Wibowo2
, Fibra R. Firmanda3
, Yogi Ramadhani4
, Wahyu Siswandari5
1,2,3,4
Department of Electrical Engineering, Jenderal Soedirman University, Indonesia
5
Department of Medical, Jenderal Soedirman University, Indonesia
Article Info ABSTRACT
Article history:
Received Jan 3, 2019
Revised May 4, 2020
Accepted May 12, 2020
The diagnosis of blood disorders in developing countries usually uses
the diagnostic procedure complete blood count (CBC). This is due to
the limitations of existing health facilities so that examinations use standard
microscopes as required in CBC examinations. However, the CBC process
still poses a problem, namely that the procedure for manually counting blood
cells with a microscope requires a lot of energy and time, and is expensive.
This paper will discuss alternative uses of image processing technology in
blast cell identification by using microscope images. In this paper, we will
discuss in detail the morphological measurements which include the diameter,
circumference and area of blast cell cells based on watershed segmentation
methods and active contour. As a basis for further development, we compare
the performance between the uses of both methods. The results show that
the active contour method has an error percentage 5.15% while the watershed
method has an error percentage 8.25%.
Keywords:
Active contour
Blast cell
Developing countries
Image processing
Watershed
Copyright © 2020 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Retno Supriyanti,
Department of Electrical Engineering,
Jenderal Soedirman University,
Kampus Blater, Jl. Mayjend Sungkono KM 5, Blater, Purbalingga, Central Java.
Email: retno_supriyanti@unsoed.ac.id
1. INTRODUCTION
Identification of blood cells is one of the diagnostic procedures used to identify various diseases.
This diagnostic procedure is commonly called the CBC. CBC is a blood test that provides information to
doctors about five main parts of blood. The five parts are three types of sell (red blood cells, white blood
cells, and platelets) and two types of values (hemoglobin value and hematocrit value) [1]. This CBC
procedure is widely applied in developing countries including Indonesia because of limited facilities and
resources in the medical field. However, the CBC process still poses a problem, namely that the procedure
for manually counting blood cells with a microscope requires a lot of energy and time, and requires a high
cost. Blood cells are classified into several types, namely red blood cells, white blood cells often called as
leukocytes, and blood platelets. The blood component has specific roles and functions in the circulation
process throughout the body. Blood can experience abnormalities that occur mainly in its constituent
structures. This disorder results in a disorder or disease in the body. One blood disorder that can occur is
acute lymphoblastic leukemia (ALL).
Cancer death rates including leukemia are higher in developing countries compared to developed
countries. This difference reflects differences in risk factors and the success of handling detection, as well as
the availability of treatment [2]. According to the latest World Health Organization (WHO) data published in
2017 leukemia deaths in Indonesia reached 9,179 or 0.55% of total deaths. The age adjusted death rate
is 4.20 per 100,000 of population [3]. According to the fact, it is seen that the presence of blood cell
abnormalities such as leukemia is a case that needs serious attention.

Int J Elec & Comp Eng ISSN: 2088-8708 
Preliminary process in blast cell morphology identification … (Retno Supriyanti)
5715
As explained in the paragraph above, one of the causes of the high mortality rate of leukemia in
developing countries including Indonesia is the limited human resources and health facilities. One alternative
to overcome this problem is the implementation of technology that is cheap and easy to use, but also
accurate, and one of them is digital image processing. Research on leukemia that involves the use of digital
image processing itself is quite a lot. Zhang [3] researched leukocyte detection using digital image
processing, specifically a combination of image segmentation and pattern recognition. In the segmentation of
the image, he divided the image of leukocytes into several parts. While for the classification he uses support
vector machine (SVM). Mohapatra [4] introduced the clustering method using shadowed c-means (SCM) in
the segmentation of blood microscopic images. SCM method is used to classify each pixel in 4 clusters.
The algorithm is used to separate the nucleus and cytoplasm in each sub-image. Negm [5] in his research,
he used a decision support system that included panel selection and segmentation using K means clustering
on some datasets on leukimia identification. Ali [6] proposed an algorithm to isolate and count lymphocytes
in the image of white blood cells. The process made includes cell segmentation, scanning algorithms, feature
extraction, and lymphocyte cell recognition. Rawat [7], In his research, he proposed a new method for
distinguishing acute lymphoblastic leukemia from normal lymphocytes. This method separates leukocytes
from other blood cells and then extracts all the information inside. Separation is based on Gray level
co-occurrence matrices (GLCM) and form-based features. Prinyakupt [8] he proposed a system for
segmenting white blood cells into the area of the nucleus and cytoplasm, extracting appropriate features and
classifying them into five types namely basophil, eosinophil, neutrophil, lymphocyte, and monocyte. Liu [9]
he proposed a method of calculating red blood cells in full automatic based on hyperspectral microscopic
images and combining spatial and spectral information to obtain maximum precision. Lin [10] he proposed
a method for classifying five types of leukocytes using a method based on multi-scale regional growth and
grouping mean values. The way to do this is to extract the leukocyte texture feature that is visually visible.
For the classification, he uses the support vector machine (SVM) method. Li [11] he proposed a method for
recognizing leukocytes for human blood smears based on island clustering texture (ICT). The way to do this
is to analyse the features of a typical class of leukocytes to form an ICT model. Sarrafzadeh [12] in his
research, he used texture features in recognizing leukocytes. He focused on seven categories of texture
features in order to get the best category in the classification of leukocytes. Porcu [13] developed
a semi-automatic method of extracting all the information on red blood cells and calculating the amount in
detail. Umamaheswari [14] proposed an algorithm for segmenting nuclei in white blood cells in leukemia
identification. The algorithm developed is based on Otsu's thresholding. Gupta [15] in his research,
he presented the PCSeg Tool for segmenting blood plasma cell images. The algorithm that he uses is
pre-processing the existing input images by removing all information that is not needed so that all that
remains is the region that does provide information about the blood plasma cells only.
Our research also aims to develop a simple system that is efficient and effective in the process of
identifying leukocyte cells, according to the limited facilities and infrastructure in several rural areas in
Indonesia. Our main contribution is the optimization of the use of simple methods but can provide accurate
results in leukocyte identification based on the image of white blood cell smears photographed without
illumination settings. We have done some preliminary research [16-21]. Salmam [22] researched facial
recognition into a form of emotion using artificial neural networks. Carpio et al., [23] they researched an
automatic analysis of bacilli numbers. Also, about calculating the concentration level therein on the sputum
sample image of patients with tuberculosis. However, the results obtained are not optimal, because
the conditions of the input image are acquired without using the light illumination settings. Therefore it is
difficult to apply to input images with various illumination conditions. This paper will discuss the comparison of
the use of watershed and active contour methods in obtaining leukocyte segmentation with input images that
have random light illumination.
2. RESEARCH METHOD
2.1. Data acquisition
All the data used in this research came from the Pathology laboratory, Hospital "Dr. Margono
Soekardjo," Banyumas, Central Java, Indonesia. Figure 1 shows the stages of the data collection for this
research. According to Figure 1, from the pathology laboratory, the medical team examined blood cell smears
using a standard microscope. The primary purpose of using this conventional microscope is to create a pilot
project so that this can be applied in some rural areas in Indonesia where the district does not yet have
a digital microscope facility. Currently, it has only digital microscope facilities, just in large hospitals.
In Figure 1, an example of the input image that will be used in this research is also shown, which shows that
the image input used has a different illumination level.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 10, No. 6, December 2020 : 5714 - 5725
5716
Figure 1. Data acquisition process
2.2. Pre-processing image
2.2.1. Normalization image
The input images used in this research comes from several types of cameras, so the resolution of
each input image is different. According to this case, we set the size of the existing input image so that
the calculation results are according to parameters and do not depend on size, but also to speed up data
processing time. In this research, the image is resized with a calculated scale using the formula in (1).
𝑅 =
400
𝑆
(1)
In (1), R states the resize scale, S is the smallest value of the dimensions of the length and width of
the image. The value of 400 is used as the basis of the calculation to determine the scale where the smallest x
or y dimension value is selected as a comparison with the value 400, and the result is used as a scale for
the remaining values of x or y. The above calculation is used so that can adjust the characteristics of images
that have different lengths and widths. Moreover, to make it easier to recalculate the original pixel size.
Table 1 is a comparison table of original dimensions with resizing dimensions.
Table 1. Comparison of original dimensions and resize dimensions
No Original Dimension Resize Dimension
1 640 x 480 534 x 400
2 960 x 720 534 x 400
3 4000 x 3000 534 x 400
4 4128 x 2322 712 x 401
5 581 x 1032 400 x 711
2.2.2. Cropping
At this stage, the images obtained are cut using the cropping function in order to obtain the Region
of interest (ROI). Cropping is done automatically adjusting the ROI of the image. The process is done by
changing the image to binary with a scale of 0.7 then noise cleaning is done by removing objects that have
a pixel count of fewer than 4000 pixels. This is done to get objects that only contain ROI. Then scan
the object to find the maximum and minimum coordinate values of X and Y. The results are used to calculate
the length and width of the object by subtracting the maximum value of the x and y coordinates with
the minimum value than using the minimum value of the x and y coordinates as the initial coordinate of
cropping. Figure 2 shows the image before and after cropping.
2.2.3. Thresholding
Thresholding in this system aims to remove other objects other than objects of white blood cells in
order to improve accuracy at the stage of segmentation that will be carried out. This step is done by utilizing
RGB value differences in white blood cell objects and other objects. Sampling is carried out on five pieces of
images that represent the condition of each input image. RGB values are taken from 9 pixels of each image
consisting of 3 pixels of white blood cells, 3 pixels of red blood cells, three pieces of background pixels-
Table 2 shown this process.

5717
Figure 2. (a) Before cropping, (b) after cropping
Table 2. RGB value sampling results
No Object Red Green Blue
Img1 White Blood Cell 170 55 86
146 40 72
170 93 97
Red Blood Cell 150 104 88
142 101 95
127 97 88
Background 218 180 109
215 182 122
11 11 11
91 37 133
131 94 144
152 129 157
153 129 161
213 196 206
13 13 13
134 67 134
144 79 145
174 157 275
179 160 180
172 158 175
12 12 12
175 48 125
174 59 126
178 131 139
181 136 141
193 163 159
0 0 0
156 40 103
138 41 110
146 113 130
164 116 128
187 152 150
1 1 1
2.3. Active contour segmentation
Active contour is a segmentation method using a closed curve model that can move wide or narrow
to adjust the boundary of a segmented object [24]. In this experiment, Chan-Vese model contour was used
where this model is a region-based active contour. In this model statistical information is used inside or
outside the curve to determine the direction of evolution and the amount of energy used as described in (2).

 ISSN: 2088-8708
5718
(𝑐1, 𝑐2, 𝐶) = 𝜇. 𝑙𝑒𝑛𝑔𝑡ℎ (𝐶) + 𝑣. 𝑎𝑟𝑒𝑎(𝐶)
+ ∫ |𝜇0. (𝑥, 𝑦) − 𝑐1|2
𝑑𝑥𝑑𝑦 + ∫ |𝜇0(𝑥, 𝑦) − 𝑐2|2
𝑑𝑥𝑑𝑦
𝑜𝑢𝑡𝑠𝑖𝑑𝑒(𝑐)𝑖𝑛𝑠𝑖𝑑𝑒(𝑐)
(2)
According to (2), the first part defines the length of the curve, the second part of the area in
the curve, the third and fourth parts define the difference in intensity of the input image and the average
intensity inside and outside the curve. The value of C is the initial masking curve; the values of c1 and c2 are
the intensities inside and outside the curve; the value of µ0 (x, y) is the input image. The more curves close to
the desired object boundary, the smaller the value F. Figure 3 shows an example of using the active contour
method in segmenting white blood cell images.
Figure 3. Chan-vese model active contour method
2.4. Watershed segmentation
Watershed transformation is a transformation that brings the perspective of two-dimensional
imagery in a three-dimensional viewpoint. Watershed transformation works on grayscale images by using
watershed transformation, the image as if it has a three-dimensional form, x, y, and z. The x and y fields are
the original image itself, while the z plane is the level of the ash scale in the form of valleys. The thicker
the ash level, the deeper the valley [25]. This watershed segmentation method has the advantage of being
able to separate objects which are coiled so that objects can be separated as shown in Figure 4. Referring to
Figure 4, the object which had previously coincided can be separated by the watershed segmentation method.
This is very useful when calculating objects because multicellular cells that are coiled will count as single
cells so that the calculation process becomes wrong.
Figure 4. Separation of Objects Coinciding with Watershed Segmentation
2.5. Post processing
The post-processing stage is a stage that aims to improve the quality of the second stage image after
the segmentation process is carried out. This stage is done because the results of segmentation still allow
producing small objects or noise due to the lack of or excessive process of segmentation. Image improvement
must be made with good and clean image quality. The operations performed at the post-processing stage are
morphological operations, objection elimination on the border edge, elimination of unusual objects, and
object names.

5719
3. RESULTS AND DISCUSSION
In this experiment, the variables to be measured in segmentation with the two methods are
the detection of the number of leukocytes in each image, area, maximum diameter, minimum diameter and
circumference of leukocyte cells. The first experiment was carried out by exploring information using
the results of segmentation of the active contour method. Calculation of cell numbers is done by separating
cell objects. The process of separating objects is done by calculating the distance transformation on the object
to determine the intersection of objects that are overlapped. Distance values are calculated for each nonzero
pixel. The model used in this distance transformation is a city block model. The city block model measures
distance based on four neighborhoods. Where if the pixel is in area 4, the distance is calculated by one but if
it is not then counted two. In order for getting the best result, the distance in the distance transformation
needs to be adjusted by removing the small minima locale on the object. Figure 5 shows the results of
the separation of objects in this method.
The next process is noise elimination, which functions to eliminate other objects other than
the leukocyte cell object. In this process, three parameters are used for three different functions adjusting to
the size of the object of the white blood cell. The parameters used are 10000, 3000, and 300 pixels.
Then remove the object that is tangent to the edge of the image. Finally, which image is selected is an
unusual object and which is not to be removed. The trick is to calculate the degree of roundness of each
object with (3).
𝑅 =
4𝜋𝐴
𝑃2 (3)
Referring to (3), then R states the Degree of Roundness, A states Area, and P states Circumference.
The roundness used in this experiment is 0.78 so if there is an object with a degree of roundness less than
that it will be eliminated. Figure 6 shows some examples of the image results of segmentation by active
contour method.
Figure 5. An example of overlapped object
separation results
Figure 6. Examples of active contour
segmentation result
Referring to Figure 6, the number of leukocyte cells can be segmented using the active contour
method. So that based on the segmentation display, the number of leukocyte cells can be calculated
automatically. For measurements of diameter, circumference, and area, we refer to (2) above. In (2) above,
C1 and C2 are two constants which are the average intensity inside and outside the contour, respectively.
Moreover, µ0 is the input image. On active contour without the edge, it will minimize term fitting and add
some term regulations, such as the length of the C curve and the area of the region in C. From the equation,
the length of the curve can be measured by (4). While the area is formulated in (5).
∫ √1 + (𝑦′)2 𝑑𝑥𝐶
(4)
𝐿 =
1
2
∫ (𝑥𝑑𝑦 − 𝑦𝑑𝑥) =
1
2
∫ (𝑥𝑦′
− 𝑦)𝑑𝑥𝐶𝐶
(5)

 ISSN: 2088-8708
5720
In other hand, the watershed segmentation method is a series of methods consisting of watershed
transformation, distance transformation, morphological operations, and other related operations. The process
of identifying white blood cells in this study is divided into several stages, namely: color threshold,
pre-processing, segmentation, post-processing, calculating object parameters. After going through
pre-processing as described above, noise is eliminated. Noise removal aims to clean binary images from
small objects that are not perfect color thresholds. Also, repairs to the edges of the object are made to be
neater and smoother. This operation can be done using image morphology operations. As explained above,
watershed segmentation also includes distance transformation. This distance transformation function will
produce a set of matrices that contain the transformation value of the distance of each pixel. Distance values
are calculated for each nonzero pixel. The chosen model is very decisive towards the results of segmentation
because it can allow over segmentation when using the incorrect distance transformation method. According
to our experiments using Euclidian, City block, Chess board, Quasy-Euclidian models, and most objects in
the form of round objects are not perfect. Therefore, we need a model that has similar shapes or characteristics
with the aim that the distance transformation is smooth and later does not cause over-segmentation. With that
consideration, it can be seen that the city block method matches the characteristics of the object to be
transformed, besides that this model can produce the right segmentation output, while other methods produce
over segmentation output.
Referring to (3), properties that can be measured using this function include area, diameter,
a circumference of the object. This operation of the elimination of abnormal objects can be done by utilizing
the degree of roundness of the object. In order to determine the roundness value of the object, an analysis of
the value of the roundness of the object is carried out. Figure 7 shows an example of the property of
the degree of roundness of an object. Based on Figure 7, the value of the degree of roundness of strange
objects is lower than the standard object. The highest value of the degree of a strange object in the data is 0.53.
Therefore, the degree of roundness is taken at a value of 0.6. Figure 8 is an example of an image that has
experienced abnormal object elimination.
Figure 7. An example of property of roundness object
Figure 8. Elimination of abnormal objects

5721
The way to calculate the circumference of this function is to use eight neighbors; each pixel that
does not have neighbors will be counted as a circumference of the object. Calculation of diameter is done by
calculating the longest distance and the shortest distance from the edge of the object. The results of diameter
properties in this study will get two values for the diameter, namely the maximum diameter and minimum
diameter. The extensive calculation is done by counting all pixels on the object. Table 3 shows the results of
the calculation of the diameter, circumference and area of leukocytes using the active contour and watershed
methods. Referring to Table 3, we make a graph of analysis that shows the relationship between the results
obtained in each method as shown in Figure 9.
Referring to Table 3 and Figure 9, in general, the results of the measurement of the two methods
give almost the same results for each measurement variable. However, for morphological calculations using
watershed segmentation, calculation of properties which include Amount, Diameter, Area, and circumference
of leukocyte images. The results of the calculation are in the form of a matrix n × 1, where n is the number of
objects detected. The way to calculate the circumference of this function is to use eight neighbors; each pixel
that does not have neighbors will be counted as a circumference of the object. Calculation of diameter is done
by calculating the longest distance and the shortest distance from the edge of the object. The results of
diameter properties in this study will get two values for the diameter, namely the maximum diameter and
minimum diameter. The extensive calculation is done by counting all pixels on the object.
Table 3. The results of the calculation of the number, diameter, circumference,
and area of leukocytes based on the active contour method
Img Active Contour Watershed
Min
D
Max
D
P A Min
D
Max
D
P A
C1 147 178 582 20320 146 179 564 20279
C2 155 186 189 22245 157 191 569 23058
C3 167 184 654 23815 - - - -
C4 184 200 682 28875 189 201 672 29715
C5 143 174 556 19454 143 173 546 19350
C6 139 165 518 17736 139 164 494 17548
C7 128 148 478 14568 127 147 453 14373
134 161 496 16822 133 161 475 16635
C8 125 158 480 15404 124 158 455 15167
138 164 515 17221 137 164 495 17003
C9 143 157 516 17393 142 156 487 17219
C10 112 122 396 10551 112 121 377 10483
M1 26 30 90 596 26 29 85 583
24 27 83 510 24 27 78 505
26 34 98 686 26 34 92 672
23 32 89 569 23 31 85 567
28 32 100 706 28 32 93 694
23 27 81 475 22 27 76 464
27 29 90 617 27 29 86 611
28 41 117 865 28 41 112 856
27 37 105 762 27 37 99 750
32 25 91 620 32 25 87 618
26 31 93 620 26 31 90 621
M2 31 35 105 835 31 35 100 834
32 40 119 980 32 39 114 966
23 30 89 500 28 33 93 706
34 43 126 1143 34 43 120 1133
31 41 117 985 31 43 118 1034
21 32 90 510 - - - -
28 33 98 709 31 41 112 981
M3 23 24 74 425 23 24 71 427
29 36 111 777 29 37 106 779
26 34 99 691 26 34 93 685
37 42 134 1131 36 42 129 1132
M4 26 31 92 614 26 31 87 609
29 31 98 687 28 31 91 675
33 37 118 942 33 38 112 941
P1 95 103 329 7636 95 103 315 7601
P2 84 109 323 7111 83 108 308 6929
80 84 280 5238 78 83 267 5051
P3 83 105 320 6775 82 104 305 6684
P4 82 102 306 6472 82 101 295 6433
P5 87 105 322 7128 87 105 307 7099
P6 89 96 305 6713 89 96 292 6678
P7 83 96 305 6198 82 95 293 6093
56 70 221 3063 56 69 215 2985
P8 84 97 307 6384 84 97 295 6392
56 70 221 3053 57 70 220 3049
Img Active Contour Watershed
Min
D
Max
D
P A Min
D
Max
D
P A
P9 91 105 323 7417 90 104 312 7374
P10 92 104 325 7533 92 104 311 7497
P11 92 101 320 7257 92 101 304 7220
P12 96 99 323 7494 93 96 302 7018
70 93 274 5041 67 91 257 4720
A1 60 89 258 4029 22 25 72 432
69 109 366 5324 70 85 249 4634
A2 73 90 272 5150 75 91 263 5317
70 84 253 4583 71 84 245 4698
- - - - 22 25 73 436
A3 79 89 279 4645 71 84 242 4610
A4 84 216 677 12338 54 80 220 3360
- - - - 65 81 240 4098
- - - - 64 79 239 3924
A5 78 88 283 4057 66 80 232 4088
A6 30 33 102 763 30 34 99 768
26 34 98 703 27 34 96 706
20 27 75 418 20 28 74 422
31 35 110 842 31 36 106 845
28 38 111 834 29 38 106 845
28 34 100 749 28 35 97 754
34 39 117 1032 33 40 114 1026
A7 22 31 86 510 22 31 83 515
30 31 100 728 30 31 95 727
26 32 93 649 26 32 90 652
30 32 102 757 30 32 98 753
22 33 89 545 22 33 85 546
A8 18 29 76 400 18 29 72 400
29 36 106 815 29 36 100 801
32 36 110 912 32 36 106 904
30 30 95 708 30 30 92 700
A9 23 31 90 536 23 31 88 538
22 29 82 493 22 29 80 498
22 27 80 473 22 28 78 477
26 33 94 658 26 33 92 664
24 32 92 597 24 32 89 598
26 31 96 629 26 31 91 627
24 28 84 542 25 29 82 550
25 29 86 581 19 29 75 437
21 31 83 511 26 29 84 587
- - - - 21 31 81 512
A10 27 34 99 714 27 35 96 724
27 32 94 664 18 23 61 318
25 30 92 577 27 32 91 671
23 31 89 553 26 30 90 584
30 32 101 767 23 31 85 551
- - - - 30 33 98 771

 ISSN: 2088-8708
5722
(a)
(b)
(c)
(d)
Figure 9. Graphs of comparison of measurement results based on active contour and watershed methods,
(a) minimum diameter, (b) maximal diameter, (c) perimeter, (d) area

5723
Based on the table above, it can be seen that actual cell count: 97 and total cell deviation calculation: 8.
In order to calculate the percentage of errors can use the following formula as described in (6).
𝐸𝑟𝑟𝑜𝑟 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 = ⌈
𝑇𝑜𝑡𝑎𝑙 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛
𝐴𝑐𝑡𝑢𝑎𝑙 𝐶𝑒𝑙𝑙 𝑁𝑢𝑚𝑏𝑒𝑟
⌉ 𝑥 100 % (6)
Using (6) above, the following results are obtained
𝐸𝑟𝑟𝑜𝑟 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 = |
8
97
| 𝑥 100% = 8.25%
Based on the results of watershed segmentation and cell count calculations, it can be concluded that
some of the causes of this experiment failure are: (a) Overlapping objects have an irregular shape so that
the system does not detect leukocyte cells, so they are not segmented. (b) Other objects besides leukocytes
can be detected as leukocytes. This is because the object has characteristics that are similar in color and shape.
In the morphological calculation using the active contour segmentation method, the way to calculate
the circumference of this function is to use eight neighbours, each pixel that does not have neighbours will be
counted as a circumference of the object. Calculation of diameter is done by calculating the longest distance
and the shortest distance from the edge of the object, so in this experiment, there will be two values for
the diameter. Extensive calculations are carried out by counting all pixels on the object. In general,
the calculation method is the same as the calculation on the watershed segmentation method. We did
the morphological calculation using (6) the error percentage can be calculated from morphological
calculations based on the segmentation method of active contour.
𝐸𝑟𝑟𝑜𝑟 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 = |
5
97
| 𝑥 100% = 5.15%
Based on the results of segmentation using the active contour method and cell count calculations,
it can be concluded that some causes of failure in the experiment are: (a) Overlapped objects have irregular
shapes so that over-segmentation occurs or the object is not even segmented. (b) Cells that are truncated or not
intact remain detected because they have the size of an object that is almost the same as an intact cell object.
4. CONCLUSION
In this experiment, it was found that both for the use of the watershed segmentation method and
the active contour segmentation method, for each image that has different characteristics it will require
different treatments. Objects other than leukocyte cells can be detected as white blood cells because they
have the same intensity of color and shape. Image segmentation using the Watershed Segmentation method
has the advantage of being able to separate the blood cells that are huddled together. While for
the segmentation of active contour, overlapped object separation can use the distance transformation method.
In order to achieve a high percentage of segmentation success, an image that has uniform characteristics and
variable control are needed. In this experiment, the calculation based on the active contour method has
a lower error percentage than using the watershed segmentation method.
ACKNOWLEDGEMENTS
We would like to thank the Pathology laboratory, “Prof. Dr. Margono Soekardjo” Hospital for
the data that is permitted to be used in this research. The research was funded by the Directorate of Research
and Community Service, the Ministry of Research, Technology and Higher Education, Republic of Indonesia
through the "Penelitian Terapan" (applied research) scheme.
REFERENCES
[1] R. Harris, K. Simonsen, and J. Mackay, “Best Tests,” BPAC Organization, New Zealand. [Online]. Available:
www.bpac.org.nz , 2013.
[2] D. M. Parkin, F. Bray, J. Ferlay, and P. Pisani, “Global Cancer Statistics, 2002,” CA. Cancer J. Clin., vol. 55, no. 2,
pp. 74-108, 2009.
[3] W. H. Organization, “World Health Statistics 2017,” 2017.
[4] S. Mohapatra, D. Patra, and K. Kumar, “Fast leukocyte image segmentation using shadowed sets,” Int. J. Comput.
Biol. Drug Des., vol. 5, no. 1, pp. 49-65, 2012.
[5] A. S. Negm, O. A. Hassan, and A. H. Kandil, “A decision support system for Acute Leukaemia classification based
on digital microscopic images,” Alexandria Eng. J., vol. 57, no. 4, pp. 2319-2332, 2018.

 ISSN: 2088-8708
5724
[6] A. B. Ali and M. Z. Othman, “Segmentation and Feature Extraction of Lymphocytes WBC using Microscopic
Images,” Int. J. Eng. Res. Technol., vol. 3, no. 12, pp. 696-701, 2014.
[7] J. Rawat, A. Singh, H. S. Bhadauria, and J. Virmani, “Computer Aided Diagnostic System for Detection of
Leukemia Using Microscopic Images,” Procedia Comput. Sci., vol. 70, pp. 748-756, 2015.
[8] J. Prinyakupt and C. Pluempitiwiriyawej, “Segmentation of white blood cells and comparison of cell morphology
by linear and naïve Bayes classifiers,” Biomed. Eng. Online, vol. 14, no. 1, pp. 1-19, 2015.
[9] X. Liu, M. Zhou, S. Qiu, L. Sun, H. Liu, Q. Li, and Y. Wang, “Adaptive and automatic red blood cell counting
method based on microscopic hyperspectral imaging technology,” Journal of Optics, vol. 19, no. 12, 2017.
doi: 10.1088/2040-8986/aa95d7
[10] L. Lin and W. Wang, “A robust leukocyte recognition method based on multi-scale regional growth and mean-shift
clustering,” J. Algorithms Comput. Technol., vol. 12, no. 3, pp. 208-216, 2018.
[11] X. Li and Y. Cao, “A robust automatic leukocyte recognition method based on island-clustering texture,” J. Innov.
Opt. Health Sci., vol. 09, no. 01, pp. 1650009-1–1650009-13, 2015.
[12] O. Sarrafzadeh, A. M. Dehnavi, Y. Hossein, A. Talebi, and A. Gharibi, “The Best Texture Features for Leukocytes
Recognition,” J. Med. Signals Sens., vol. 7, no. 4, pp. 220-227 2017.
[13] S. Porcu, A. Loddo, L. Putzu, and C. Di Ruberto, “White Blood Cells Counting Via Vector Field Convolution
Nuclei Segmentation,” Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and
Computer Graphics Theory and Applications, vol. 4, pp. 227-234, 2018.
[14] D. Umamaheswari and S. Geetha, “A Framework for Efficient Recognition and Classification of Acute
Lymphoblastic Leukemia with a Novel Customized-KNN Classifier,” J. Comput. Inf. Technol., vol. 26, no. 2,
pp. 131-140, 2018.
[15] A. Gupta, P. Mallick, O. Sharma, R. Gupta, and R. Duggal, “PCSEG: Color model driven probabilistic multiphase
level set based tool for plasma cell segmentation in multiple myeloma,” PLoS One, vol. 13, no. 12, pp. 1-22, 2018.
[16] R. Supriyanti, M. M. Afif, I. T. Hasan, Y. Ramadhani, and W. Siswandari, “A Simple Tool for Identifying Outer
Shape of White Blood Cell Based on Image Processing Techniques in Order To Develop Health Facilities in
Developing Countries,” PONTE Int. Sci. Res. J., vol. 73, no. 12, pp. 314-325, 2017.
[17] R. Supriyanti, A. Chrisanty, Y. Ramadhani, and W. Siswandari, “Computer Aided Diagnosis for Screening the
Shape and Size of Leukocyte Cell Nucleus based on Morphological Image,” International Journal of Electrical and
Computer Engineering (IJECE), vol. 8, no. 1, pp. 150-158, 2018.
[18] R. Supriyanti, B. L. Nababan, Y. Ramadhani, and W. Siswandari, “A Simple and Easy-to-Use Tool for Detecting
Outer Contour of Leukocytes Based on Image Processing Techniques,” 19th International Conference on
Biomedical Applications and Bioinformatics, 2017.
[19] R. Supriyanti, G. Satrio, Y. Ramadhani, and W. Siswandari, “Contour Detection of Leukocyte Cell Nucleus Using
Morphological Image,” J. Phys. Conf. Ser., vol. 824, no. 1, pp. 1-8, 2017.
[20] R. Supriyanti, Rifai. A.rifai, Y. Ramadhani and W. Siswandari, "Influence of camera types in histogram
distribution on morphological identification of myeloblast cell based image segmentation," J. Phys. Conf. Ser.,
vol. 1321, no. 3, pp. 1-6, 2019.
[21] R. Supriyanti, "Technology supporting health services for rural areas based on image processing," J. Phys. Conf.
Ser., vol. 1367, pp. 1-12, 2019.
[22] F. Z. Salmam, A. Madani, and M. Kissi, “Emotion recognition from facial expression based on fiducial points
detection and using neural network,” International Journal of Electrical and Computer Engineering (IJECE),
vol. 8, no. 1, pp. 52-59, 2018.
[23] C. del Carpio et al., “An algorithm for detection of Tuberculosis bacilli in Ziehl-Neelsen sputum smear images,”
International Journal of Electrical and Computer Engineering (IJECE), vol. 9, no. 4, pp. 2968-2981, 2019.
[24] V. Casselles, R. Kimmel, and G. Sapiro, “Geodesic Active Contours,” Int. J. Comput. Vis., vol. 22, pp. 61-79, 1997.
[25] R. C. Gonzales and R. E. Woods, "Digital Image Processing," 3rd editio. New Jersey: Prentice Hall, 2008.
BIOGRAPHIES OF AUTHORS
Retno Supriyanti is a Professor at Electrical Engineering Department, Jenderal Soedirman
University, Indonesia. She received her PhD in March 2010 from Nara Institute of Science and
Technology Japan. Also, she received her M.S degree and Bachelor degree in 2001 and 1998,
respectively, from Electrical Engineering Department, Gadjah Mada University Indonesia.
Her research interests include image processing, computer vision, pattern recognition, biomedical
application, e-health, tele-health and telemedicine.
Pangestu Fajar Wibowo received his Bachelor degree from Electrical Engineering Depratment,
Jenderal Soedirman University Indonesia. His research interest Image Processing field.

5725
Fibra Rhoma Firmanda received his Bachelor degree from Electrical Engineering Depratment,
Jenderal Soedirman University Indonesia. His research interest Image Processing field.
Yogi Ramadhani is an academic staff at Electrical Engineering Department, Jenderal Soedirman
University, Indonesia. He received his MS Gadjah Mada Universirt Indonesia, and his Bachelor
degree from Jenderal Soedirman University Indonesia. His research interest including Computer
Network, Decision Support Syetem, Telemedicine and Medical imaging.
Wahyu Siswandari is an academic staff at Medical Department, Jenderal Soedirman University,
Indonesia. She received her Ph.D from Gadjah Mada University. Also, she received his M.S
degree and bachelor degree from Diponegoro Indonesia. Her research interest including
Pathology, e-health and telemedicine.

Preliminary process in blast cell morphology identification based on image segmentation methods

More Related Content

What's hot (18)

Similar to Preliminary process in blast cell morphology identification based on image segmentation methods (20)

More from IJECEIAES (20)

Recently uploaded (20)

Preliminary process in blast cell morphology identification based on image segmentation methods