JRSSEM 2022, Vol. 01, No. 8, 1188 – 1194
E-ISSN: 2807 - 6311, P-ISSN: 2807 - 6494
DOI : 10.36418/jrssem.v1i8.136 https://jrssem.publikasiindonesia.id/index.php/jrssem/index
MODIFICATION OF IDTCS METHOD FOR TOUCHING
LEUKEMIA CELL GROUPING
Nenden Siti Fatonah
1*
Chastine Fatichah
2
Handayani Tjandrasa
3
1
Esa Unggul University, Jakarta,
2,3
Institut Teknologi Sepuluh Nopember
e-mail: nenden.siti@esaunggul.ac.id
1
, chastine@if.its.ac.id
2
, [email protected]c.id
3
*Correspondence: : nenden.siti@esaunggul.ac.id
Submitted: 25 February 2022, Revised: 05 March 2022, Accepted: 15 March 2022
Abstract. Morphological analysis and calculation of the number of white blood cells on
microscopic images are stages in diagnosing leukemia. Constraints in developing a system for
diagnosing leukemia are white blood cell segmentation and counting of the number single cells in
touching cell. We propose to modify the Iterative Distance Transform For Convex Sets (IDTCS)
method to separate the touching leukemia cells. The IDTCS method is used to determine markers
for each cell in touching cells. The marker results from the IDTCS method are used as cell centroids
and the next process is pixels clustering based on the nearest cell centroid using the euclidean
distance function. The data used are microscopic images of Acute Lymphoblastic Leukemia (ALL).
The experimental results show that using modified IDTCS method for clustering produces better
accuracy compared to the K-Means clustering and Watershed methods.
Keywords: acute leukemia image; touching cell separation; iterative distance transform for
convex sets (IDTCS); clustering; euclidean distance.
Nenden Siti Fatonah, Chastine Fatichah, Handayani Tjandrasa | 1189
DOI : 10.36418/jrssem.v1i8.136 https://jrssem.publikasiindonesia.id/index.php/jrssem/index
INTRODUCTION
Leukemia occurs because the
production of white blood cells out of the
spinal cord is immature and spread
throughout the body. So it is very
dangerous because it disturbs the stability
of blood circulation. The importance of
computerized based research on white
blood cells will greatly help the medical
field in diagnosing leukemia.
One important process in computerized
systems in diagnosing leukemia in
microscopic images is the process of
segmenting white blood cells. A good
segmentation process will produce good
features to produce accurate diagnoses.
Leukemia image segmentation research
has been done in literatures (Labati, Piuri, &
Scotti, 2011); (Zheng, Wang, Wang, & Liu,
2018); (Joshi, Karode, & Suralkar, 2013). The
study used a peripheral blood smear
dataset from ALL patients and not ALL
patients collected at the Tettamaati
Research Center, Monza, Italy research
center (Labati et al., 2011). In literature
(Putzu, Caocci, & Di Ruberto, 2014), the
segmentation process uses Zack Algorithm
for thresholding and uses the arithmetic
process. In literature (Rezatofighi &
Soltanian-Zadeh, 2011) used the Otsu
thresholding, Gram-Schmidt
orthogonalization, and the snake algorithm
for white blood cell segmentation.
The obstacle in the process of leukemia
white blood cells segmentation is the
presence of touching cells. Because the
next step in diagnosing leukemia from
microscopic images is morphological
analysis and calculation of white blood cell
counts. Literature (Fatichah, Purwitasari,
Hariadi, & Effendy, 2014) present white
blood cells counting using features, K-
Means clustering for separation the
overlapping cells.
The segmentation used is the
morphological iteration process of
automatic erosion for tangent objects [5].
Research (Singhal & Singh, 2014)
segmented morphology and thresholding
methods. Research (Tian, Li, Zeng, Evans, &
Zhang, 2019) in segmenting using K-Means
clustering. The research (Tan, He, & Sun,
2010) conducted segmentation using fuzzy
K-means clustering, thresholding and Zack
algorithms. Research (Mohapatra, Patra, &
Satpathy, 2014) in segmenting using the
Shadowed C-means (SCM) and K-Means
Clustering methods. The approach of the
parametric model of the Gaussian mixture
(GM) is used in cell segmentation (Cuevas
& Sossa, 2013). In this study the concave
method was used to determine the
convexity and concomitance of
overlapping cells (Zheng, Wang, Wang, &
Chen, 2014). In research (Mandyartha &
Fatichah, 2016) the watershed method was
used in segmenting leukocyte cells.
Previous studies on the segmentation of
touching leukemia cells still occur in the
presence of oversegmentation and
undersegmentation. It is necessary to
develop a method of segmenting white
blood cells to obtain an accurate area and
number of single cells in the image of
touching cells. The use of clustering
methods such as K-Means for
segmentation of touching leukemia cell
have obstacles in determining the number
of K and the results of clustering are not
optimal if the centroid initialization is not
correct. While the separation of touching
1190 | Modification of IDTCS Method for Touching Leukemia Cell Grouping
cells using the Watershed method still has
oversegment. We propose to modify the
Iterative Distance Transform For Convex
Sets (IDTCS) method to separate touching
leukemic cells. The IDTCS method is used to
determine markers for each cell in the
touching cell. The marker results from the
IDTCS method are used as cell centroids
and then grouping pixels based on the
nearest cell centroid using the euclidean
distance function. The data used for the
experiment are microscopic images of
Acute Lymphoblastic Leukemia (ALL) from
M. Tettamanti Research Center for
childhood leukemias and hematological
diseases, Monza, Italy. The experimental
results of the modified IDTCS for clustering
were compared with the K-Means
clustering and Watershed methods.
METHODS
A. White Blood Cell Segmentation
This stage is to get the area of white
blood cells in leukemia microscopic
images. A flowchart of touching
leukemia cells separation using
Modified IDTCS is shown in Fig. 1.
Before the white blood cell (WBC)
segmentation process is carried out,
the preprocessing is done first. The
stages of preprocessing include
changing RGB to HSV images,
thresholding using the Otsu method to
get the object area of the blood cell to
determine the value based on the
histogram to select candidate cell
objects is WBC.
B. Determining Cell Marker and
Touching Cells Separation using
Modified IDTCS
In detecting touching cell markers,
the Distance Transform based method
is used, namely the Iterative Distance
Transform For Convex Sets (IDTCS)
method (Fatonah, Tjandrasa, &
Fatichah, 2018). Where the IDTCS
method produces good accuracy to
mark cells using the concavity concept.
Concavity formula measure is used as
research (Rosenfeld, 1985); in equation
1.
(1)
Where
and l(L)
are lengths of the line segment L.
Nenden Siti Fatonah, Chastine Fatichah, Handayani Tjandrasa | 1191
DOI : 10.36418/jrssem.v1i8.136 https://jrssem.publikasiindonesia.id/index.php/jrssem/index
Figure 1. Flowchart of Touching Leukemia Cells Separation using Modified IDTCS Method
After getting the marker every
single cell in the overlapping cell then
the cell separation process or clustering
is continued where the marker location
is the cluster centroid.The process of
separating the contacted cells is done
by grouping pixels to the nearest
centroid using the euclidean distance
function as in equation 2.
(2)
Figure 2. Psedocode of IDTCS method (Fatonah et al., 2018)
The IDTCS method process begins with
the input of binary images with the output
of the marker object, with the threshold
parameter
and the concavity threshold
parameter
. Then do smart filling to
remove holes in the cell but do not remove
holes between cells. Then smoothing,
initialization, distance transform and
normalization are done. Do the
threshold in the new binary image and do
Iterative Distance Transform For Convex Set (IDTCS) Algorithm
Input: Binary silhouette image
Output: Object marker (M)
Parameter: distance transform threshold
and concavity
threshold
.
1. Perform smart filling and smoothing object algorithm on
2. Initialize
3. Compute distance transform of
and normalize to [0,1]
4. Create a new binary image by threshold the image using
5. Compute the concavity of all objects.
6. Mark the object if size of the concavity less than
7. Repeat step 3 to 6 until
1192 | Modification of IDTCS Method for Touching Leukemia Cell Grouping
the concavity process on all objects. Make
a mark if the size of the object in concavity
is less than the threshold concavity. Repeat
the process from the distance transform to
the threshold concavity
to get all the
desired objects.
Figure 3. Examples of separation for touching leukemia cells (a1-d1) original Image, (a2-d2)
Modified IDTCS, (a3-d3) Modified K-Means, (a4-d4) Modified Watershed
RESULTS AND DISCUSSION
The table of calculation of the number of
cells in touching leukemia cells with the K-
Means, Watershed, and IDTCS methods, is
23 images from ALL IDB data [16].
Table 1. Comparison of the results of the calculation of the number of cells in touching
leukemia cells in ALL images using K-Means, Watershed, and IDTCS.
From Table 1, it can be seen that
modified IDTCS results have higher
accuracy compared to using the K-Means
[4] and Watershed [13] methods. The
results obtained by modified IDTCS were
73.3%, Watershed 63.3%, and K-Means
30%. Likewise with the occurrence of
undersegment, using IDTCS is very small
for undersegmentation compared to K-
Means and Watershed method. Where
modified IDTCS has 1 undersegmentation,
while Watershed 4 undersegmentation and
Methods
K-Means [4]
Watershed [13]
Modified IDTCS
IDTCS_Watershed
Correct cells
10
15
17
19
Oversegment
8
1
4
3
Under
segment
5
7
0
1
Accuracy (%)
44
65 %
74%
83%
1193 | Modification of IDTCS Method f or Touching Leukemia Cell Grouping
K-Means 13 undersegmentation. For
oversegmentation, IDTCS and Watershed
tend to occur but K-Means are more likely
to occur oversegmentation. The modified
IDTCS, Watershed, have 7, 7, and 8
oversegmentation respectively.
CONCLUSIONS
In this study we propose the modified
IDTCS method for counting and splitting
touching cells in microscopic leukemia
images. Where centroid cells have been
found by the IDTCS method and then
clustering the pixels with the closest
centroid cell using the Euclidean distance
to get the single cells. And by using the
modified IDTCS method, the results are
better than the watershed and K-Means
methods. The comparison results of the
three methods,are the modified IDTCS of
73.3%, Watershed of 63.3%, and K-Means
of 30%. From the results obtained for cell
counting and cell splitting, it is hoped that
it will support accuracy in determining the
overlapping cell counts so that it can
support the pathologist in diagnosing
acute leukemia. In the future research will
be developed towards the cell area and the
more accurate form of White Blood Cell
(WBC) so that it supports the process of
classification of acute leukemia types.
REFERENCES
Cuevas, Erik, & Sossa, Humberto. (2013). A
comparison of nature inspired
algorithms for multi-threshold image
segmentation. Expert Systems with
Applications, 4(4), 1213–1219.
https://doi.org/10.1016/j.eswa.2012.08.
017
Fatichah, Chastine, Purwitasari, Diana,
Hariadi, Victor, & Effendy, Faried.
(2014). Overlapping White Blood Cell
Segmentation And Counting On
Microscopic Blood Cell Images.
International Journal on Smart Sensing
& Intelligent Systems, 7(3).
Fatonah, Nenden Siti, Tjandrasa,
Handayani, & Fatichah, Chastine.
(2018). Automatic Leukemia Cell
Counting using Iterative Distance
Transform for Convex Sets.
International Journal of Electrical and
Computer Engineering (IJECE), 8(3),
1731.
https://doi.org/10.11591/ijece.v8i3.pp1
731-1740
Joshi, Minal D., Karode, Atul H., & Suralkar,
S. R. (2013). White blood cells
segmentation and classification to
detect acute leukemia. International
Journal of Emerging Trends &
Technology in Computer Science
(IJETTCS), 2(3), 147–151.
Labati, Ruggero Donida, Piuri, Vincenzo, &
Scotti, Fabio. (2011). All-IDB: The acute
lymphoblastic leukemia image
database for image processing. 2011
18th IEEE International Conference on
Image Processing, 2045–2048. IEEE.
Mandyartha, Eka Prakarsa, & Fatichah,
Chastine. (2016). Three-level Local
Thresholding Berbasis Metode Otsu
untuk Segmentasi Leukosit pada Citra
Leukemia Limfoblastik Akut.
https://doi.org/10.24002/jbi.v7i1.483
Mohapatra, Subrajeet, Patra, Dipti, &
Satpathy, Sanghamitra. (2014). An
ensemble classifier system for early
diagnosis of acute lymphoblastic
leukemia in blood microscopic images.
1194 | Modification of IDTCS Method for Touching Leukemia Cell Grouping
Neural Computing and Applications,
4(7), 1887–1904.
Putzu, Lorenzo, Caocci, Giovanni, & Di
Ruberto, Cecilia. (2014). Leucocyte
classification for leukaemia detection
using image processing techniques.
Artificial Intelligence in Medicine, 6(3),
179–191.
https://doi.org/10.1016/j.artmed.2014.
09.002
Rezatofighi, Seyed Hamid, & Soltanian-
Zadeh, Hamid. (2011). Automatic
recognition of five types of white blood
cells in peripheral blood. Computerized
Medical Imaging and Graphics, 3(4),
333–343.
https://doi.org/10.1016/j.compmedima
g.2011.01.003
Rosenfeld, Azriel. (1985). Measuring the
sizes of concavities. 3(January), 71–75.
https://doi.org/10.1016/0167-
8655(85)90045-5
Singhal, Vanika, & Singh, Preety. (2014).
Local binary pattern for automatic
detection of acute lymphoblastic
leukemia. 2014 Twentieth National
Conference on Communications (NCC),
1–5. IEEE. 10.1109/NCC.2014.6811261
Tan, Tieniu, He, Zhaofeng, & Sun, Zhenan.
(2010). Efficient and robust
segmentation of noisy iris images for
non-cooperative iris recognition. Image
and Vision Computing, 8(2), 223–230.
https://doi.org/10.1016/j.imavis.2009.0
5.008
Tian, Kai, Li, Jiuhao, Zeng, Jiefeng, Evans,
Asenso, & Zhang, Lina. (2019).
Segmentation of tomato leaf images
based on adaptive clustering number of
K-means algorithm. Computers and
Electronics in Agriculture, 6(5), 1045–
1160.
Zheng, Xin, Wang, Yong, Wang, Guoyou, &
Chen, Zhong. (2014). A novel algorithm
based on visual saliency attention for
localization and segmentation in
rapidly-stained leukocyte images.
Micron, 5(6), 17–28.
https://doi.org/10.1016/j.micron.2013.
09.006
Zheng, Xin, Wang, Yong, Wang, Guoyou, &
Liu, Jianguo. (2018). Fast and robust
segmentation of white blood cell
images by self-supervised learning.
Micron, 10(7), 55–71.
https://doi.org/10.1016/j.micron.2018.
01.010
© 2022 by the authors. Submitted
for possible open access publication
under the terms and conditions of the Creative
Commons Attribution (CC BY SA) license
(https://creativecommons.org/licenses/by-sa/4.0/).
