2015 International Seminar on Intelligent Technology and Its Applications

Automatic Segmentation of Malaria Parasites on

Thick Blood Film using Blob Analysis
Dwi Harini Sulistyawati a, Farah Zakiyah Rahmanti b, I Ketut Eddy Purnama a, Mauridhi Hery Purnomo a
a
Department of Electrical Engineering, Faculty of Industrial Technology ITS Surabaya Indonesia 60111
b
Dian Nuswantoro University, Imam Bonjol 207, Semarang

discovered 2.5 million cases of malaria and more than 2,500

Abstract— Malaria remains a public health problem in people die from it. In 2011 also found 1.5 million cases of
Indonesia. There are still many deaths caused by malaria, malaria in Indonesia, and 400 residents reportedly died of
particularly in eastern Indonesia. There are two types of malaria [1].
blood perform in malaria, thick blood film and thin blood Conventional methods are generally carried out by
film. In Indonesia, thin blood film is used more frequently paramedics is thoroughly examine blood performed using a
than thick blood film. Malaria parasites can be found in microscope (microscopic observation). The diagnosis of
thick blood film rapidly due to the higher volume of the malaria can be done if found malaria parasites in the blood of
blood used and sweeping process is not as much on thin patients. The results obtained in each view overpass later in
blood, still a lot of leukocytes or white blood cells and the analysis, so that would be obtained information on whether
platelets in the thick blood film, making it more difficult to or not there is a malaria parasite in the blood of patients.
identify the malaria parasite . Therefore we need a method However, conventional methods can make a difference if the
can identify malaria parasites in thick blood film with a diagnosis is made by different experts. The detection of this
high percentage of accuracy. This study aims to build a disease is time consuming and subjective factors are very high
segmentation system more objective and reduce subjective [2].
factors of medical personnel in the diagnosis of malaria Diagnose malaria in general by making observations
parasites. This study has two main stages, preprocessing microscopic (observation under a microscope) in blood
and segmentation. We use the HSV color space in the samples of patients suspected of malaria and observed in the
preprocessing and morphological operations and blob laboratory by experts or laboratory personnel. There are two
analysis on the segmentation stage. From the results can be types of human blood performed is performed thin and thick
known that the blob analysis was able to identify malaria blood. In performed thin blood of malaria parasites can
parasites automatically. quickly be identified, it is because less noise and there is little
or almost no blood clots. While in the thick blood film the
Keywords— Thick Blood Film, Malaria Parasite, Preprocessing, identification more difficult because the Giemsa stain
Segmentation, Blob Analysis. highlights not only the malaria parasites but also the white
blood cells, platelets, and artefacts [6][7]. Preprocessing is to
process the original data before entering into the next stage.
Preprocessing goal is to eliminate noise and clarify the
I. INTRODUCTION
features of the data [2].
Malaria, although it has been much reduced each year but it Segmentation process is done by using blob analysis. Blob
still has high levels of mortality. The mortality rate is still high analysis is a method that aims to detect regions of an image. A
to make the handling of this disease remains to be done. blob is defined as the area of pixels connected.
Malaria in Indonesia is more spread out in areas outside Java
and Bali. The largest concentration in Papua and Maluku
Islands followed Borneo, Sumatra, and Sulawesi. This is due II. METHODS
to regions outside Java, there are many expanse of marsh
In performed thick blood of malaria parasites is not as clear
where Anopheles stay. High population in Java and Bali for
in thin blood performed so need preprocessing to get a clearer
decades led to the loss of wetlands converted into paddy
picture. This section will explain the process of image
fields, settlements and so on. Meanwhile, the uneven
acquisition, preprocessing and segmentation.
distribution of population in Papua, Maluku and Kalimantan
make this area still full of a lot of swamp which is the source A. Image acquisition
of origin of the Anopheles mosquito. In addition, treatment of Input image used in this research is the image of thick blood
malaria in areas outside Java government is not maximized. preparations malaria parasite infected. We captured the thick
One reference states that in the last five years has been blood film images using DSLR camera which is connected

978-1-4799-7711-6/15/$31.00 © 2015 IEEE 135

 directly to the microscope, with 1000X magnification. Those in the image. Although greyscale shows a lack of information
images resized into 1800x900 pixels. on the color image, the most important information, related to
the retained features such as edges, regions, blobs and so on
[4]. To convert a greyscale image into a binary image, it has to
be done by applying a known value as a threshold value. This
value is used to determine the intensity will be converted to 0
or 1. Feature detection and processing algorithms then
typically operate on the converted greyscale version of the
image. An RGB colour image, Icolor, is converted to grey
(a)
scale, Igrey-scale, using the following transformation:

Igrey-scale(n,m) = αIcolor (n,m,r) + βIcolor (n,m,g) (1)

+ γIcolor (n,m,b)

Where (n,m) indexes an individual pixel within the grey-scale

image and (n,m,c) the individual channel at pixel location
(b) (n,m) in the colour image for channel c in the red r, blue b and
green g image channels.
Fig. 1. (a) Thick blood image from microscope;
(b) parasites target

B. Segmentation Procedures
Stages of the segmentation can be seen below, there are
4 stages in segmentation procedures, convertion to HSV,
enhancement, morphologi and blob analysis.
(a)

Conversion to HSV
(b)

Fig. 3. Extraction image in RGB to grey: (a) original image;

Enhancement (b) grey image

B.1. Conversion to HSV

Morphologi
HSV is an example of a color space that represents the color
as seen by the human eye. H is derived from the word "hue", S
is derived from the "saturation" or saturation and V is derived
from the "value" or value. H (hue) is the dominant wavelength
Blob Analysis of color, such as red, blue, green. S (saturation) is the 'purity
'colors (in the sense of the amount of white light mixed). V
Fig. 2. Block diagram of segmentation procedures
(value) is the brightness of the color (also known as
luminance) [4]. By examining the individual color channels of
In preprocessing color space selected in accordance with the
images in the HSV space, we can see that image objects are
drawings performed thick blood, which in the color space is
more consistently contained in the resulting hue field than in
most clearly visible parasites then go to the next process. In
the channels of the RGB representation.
this study using RGB color space approach by taking each
component of R component, G component and B component. ( )
Then use the HSV color space, which consists of hue, = tan ( ) ( )
(2)
saturation and values. Converting RGB to greyscale can be
done using a simple transformation. Greyscale conversion is
the first step in several image analysis algorithms, because
basically aims to simplify (reduce) the amount of information

136
 ( , , ) and opening. On morphological operations, the first step we
=1 (3)
eliminate noise with opening operation. This operation is used
to cut the part of the object that is only connected to one or
two dots, or eliminate small objects and generally smoothing
= (4) the limit of large objects without changing the object area
significantly. The opening of set A by structuring element S
denote A o S, structure elements is a small sets or sub-images
However, this way of making hue undefined if S is zero. used to probe an image under study for properties of interest.
The second way by using the following formula: The opening operation is define as:

=( , =( , =( (5) (10)
) ) ) A D S = (A ⊗ S) ⊕ S

= max ( , , ) The next step to filling holes in ROI (Region Of Interest) wich
(6) has been obtained. We use this algorithm to do the region
filling.

0, V=0 Xk = ( Xk−1 ⊕ B ) ∩ Ac ,k=1,2,3,… (11)

= ( , , ) (7)
1 , 0 We start from forming an array X0 of zeros (the same size as
the array containing A), except at the locations in X0
corresponding to the given point in each hole, which is set to
0, =0 one, where B is the symmetric structuring element. The
( )
, = algorithm terminates at the iteration step k if Xk = Xk-1. The set
Xk then contains all the filled holes; the union of Xk and A
= (8)
60 2 , = contains all the filled holes and their boundaries.
60 4 , = After the filling of holes the next step is erosion. Erosion (E) is
the process of elimination of points object (1) to be part of the
background (0), based structuring element S is used. For each
=H 360 H 0 (9) point in A, place the pivot point S at the point A. If there is a
part of S that are outside A, then pivot point removed / used as
background.
B.2. Morphological Operations
Image segmentation is the process to get the objects contained E ( A, S ) = A ⊗ S (12)
in the image or split the image into several regions with each
object or region has the same attributes [5]. Segmentation The next step is dilation. Dilation (D) is the process of
process has a very important role. In this segmentation, merging points background (0) become part of the object (1),
morphological operations is used to reach the maximum based on structuring element S is used. How to dilation is, for
results. The term morphology usually found in the biological each point in A, place the pivot point S at the point A give the
sciences to study the form and structure of animals and plants. number 1 for all points (x, y) which hit / struck by the
The same term is also used in image processing. In the context structure S at position mentioned.
of morphology, mathematics as a tool for extracting image
components that are useful in areas such representation and D( A, S ) = A ⊕ S (13)
description forms the boundary, and the convex hull
framework [5]. Morphological operations are image The results of morphological processes are usually not perfect,
processing techniques that are based on the shape of a segment in sense that still leaves a number of holes in the objects we
or region in the image. This operation is usually applied to the get. so we need a region filling process to fill the hole so we
binary image as focused on the shape of the object. Usually get a good form object in perfect shape of the object. The
segments are based on the object or target destination. algorithm used for filling the operation region binary image
Segmentation is done by distinguishing between the object based on the amount of dilation, complementation and
and the background, such as by utilizing a floating operation intersection. Binary images usually appear in the image
that changes the color image and gray scale into a binary processing application as a result of thresholding or some
image. Binary value represents the image of the object and not other segmentation procedures on the input image gray scale
objects (background). Although more widely used in the or color image. This procedure is rarely perfect (often due to
binary image, morphological operations are often used in uncontrolled variations in lighting intensity) and can leave the
image gray scale and color. The results of morphological hole, the background pixels in a binary object emerging from
operations can be used for decision-making by further the segmentation. Filling holes in these objects are often
analysis. These operations include: dilation, erosion, closing desirable in order to further morphological process can be

137
 effectively implemented. Starting from a point p on the foreground pixel color values (white) the same or adjacent to
boundary, the goal is to fill the entire region with 1. If the the previous pixel will be given the label, as well as so on.
convention that all the non-boundary points are labeled 0, then
we give the value 1 at p to start. The following procedure will
fill the region with a value of 1: III. EXPERIMENTS AND RESULT
The convertion to HSV colorspace can we see the result as
Xk = ( Xk-1 ⊕ B ) Ac , where k = 1; 2; 3; : : : (14) follow, It works by changing the original image to HSV color
space, in which the luminance component (V) is separated
Based on (13) where Xo = p and B is the element structure B. from the chromatic components (H and S). Thus, in the HSV
The algorithm stops at iteration step to K if Xk = (Xk-1). color space we get maximum results, in which the malaria
Association union of X and A fills stuffing and edges. Dilation parasite is more clearly visible. so it is very useful for further
process will fill all areas if not checked. Intersection at every processing.
step with Ac limit makes the results that are always in the
specified region.

B.3. Blob Analysis

Analysis BLOB (Binary Large Object) is a set of areas or
objects we want to observe and have the same pixel value.
BLOB stands for Binary Large Object and refers to a group of (a)
connected pixels in a binary image. The term ‘‘Large’’
indicates that only objects of a certain size are of interest and
that ‘‘small’’ binary objects are usually noise [2]. The purpose
of the blob analysis is to separate the blob objects in binary
form, o and 1. A blob consists of a group of connected pixels.
There are two types of connectivity blob analysis, namely 4 (b) (c)
and 8 connectivity. This research used 8-connectivity for
maximum results. The 8 connectivity is more accurate than 4-
connectivity, but 4-connectivity is faster computation than 8-
connectivity [3].

(d) (e)
T2 T4 T3 T2
Fig. 5. Extraction image in HSV color space; (a) original
T3 P T1 T5 P T1 image; (b) H band; (c) S band; (d) V band; (e) image in HSV

In HSV colorspace we use the Hue (H) and Saturation (S) with
T4 T6 T7 T8 a certain threshold value to obtain a ROI (Region Of Interest)
in a binary image. After it conducted morphological
operations on the binary image. The first morphology
(a) (b) operation is opening, aiming to eliminate noise and smooth the
edges. The form of the parasite that can still leaves a hole, so it
needs to be taken to close the hole by filling holes process.
Fig 4. Blob Analysis; (a) 4 connectivity; (b) 8 connectivity
The next step is the erosion process, the goal to eliminate the
noise or small objects that are not part of the parasite. Then
In 4 connectivity T1, T2, T3, T4 is the nearest neighbor of
performed the dilation process, to maximize the form of the
pixel P. In 8 connectivity T1, T2, T3, T4, T5, T6, T7, T8 is the
parasite after erosion process earlier, so that the shape in
neirest neighbor of pixel P [4].
accordance with the actual parasite.
We use the Grass-fire algorithm at this stage. The algorithm
starts in the upper-left corner of the binary image. Then scans
the entire image from left to right and from top to bottom.
Then scan the whole picture from left to right and from top to
bottom. First, find the dot coordinates of the object. Using 8-
connectivity means having 8 directions to search for
neighboring pixels. The next process is the object labeling.
Labeling the object can be determined by knowing the
relationship of the pixel with another pixel that is determined
by connectivity. Blob calculation is done by searching (a) (b)
(scanning) and labels each pixel of the binary image. If found

138
 Based on malaria parasites images above, the image can be
quickly segmented using blob analysis. To calculate the
degree of accuracy by comparing parasite automatic
segmentation results with the parasite through the manual
retrieval process. Retrieval parasite manually done in
coordination with experts parasitology. From the difference in
the Euclidean distance obtained percentage accuracy. The
(b) (d)
smaller the Euclidean distance is better because the accuracy
will be higher.
Fig. 6. Morphological Operation; (a) Opening; (b) Filling
Holes; (c) Erosion; (d) Dilation Table 1. Accuracy of malaria parasites

After the morphological operation we get the malaria No Image Euclidean Distance Accuracy
parasite, so we can enter the segmentation stage, malaria
parasites can be segmented automatically using blob analysis. 1 Parasite 1 0.0189 98.11%
By using this system, color differences in thick blood films 2 Parasite 2 0.0047 99.53%
can be handled properly. Some of the results of the
3 Parasite 3 0.0024 99.76%
segmentation on thick blood films is as follows.
4 Parasite 4 0.0068 99.32%
5 Parasite 5 0.0048 99.52%
6 Parasite 6 0.0090 99.1%
7 Parasite 7 0.0047 99.52%
8 Parasite 8 0.0027 99.73%
9 Parasite 9 0.0015 99.85%
(a)
10 Parasite 10 0.0052 99.48%

Of the research results can be seen that the high level of

similarity between the parasite segmentation results with the
original parasite with ranges between 0.0015 – 0.0189 dpi (dot
per inch) and degree of accuracy 98.11% - 99.85%.

IV. CONCLUSION
(b)
In this research blob analysis method can be applied in
segmentation of malaria parasites in thick blood films. High
degree of accuracy proved that this method is able to show the
maximum results in the malaria parasite to detect objects with
degree of accuracy 98.11% - 99.85%, an average of accuracy
is 99.392%. Different colors on thick blood films can also be
resolved properly with this system. With results expected to be
useful for the detection of malaria parasites in thick blood
film, so as to reduce the subjective factors of medical
personnel in diagnosing the malaria parasite.
(c)

ACKNOWLEDGEMENT
The author would like to thank Mrs. Suhintam Pusarawati
(Airlangga University) for discussion of parasite and the thick
blood films. Thank to Dinas Kesehatan Provinsi Jawa Timur
(East Java Health Department) for their own thick blood films.

REFERENCES
(d) [1] World Health Organization, “Basic malaria microscopy part 1 learner’s
guide”, WHO library cataloguing in publication data, Switzerland,
Fig. 7. (a), (b), (c), (d) Segmentation results of malaria second edition, 2010.
parasites on thick blood film

139
 [2] I Ketut E.P, Farah Zakiyyah R, Mauridhi Herry P, “Malaria Parasite
Identification on Thick Blood Film Using Genetic Programming’’,
International Conference on Instrumentation, Communications,
Information Technology, and Biomedical Engineering (ICICI-BME),
2013.

[3] Moeslund, T. B, “Introduction to Video Image Processing”, Springer.

2012.

[4] Abdul Kadir, Adi Susanto, “Teori dan Aplikasi Pengolahan

Citra”, 2013.

[5] Solomon, Chris, Toby, B, “Fundamental of Digital Image Processing”,

Wiley-Blackwell. 2011.

[6] F. Borray, Andrew G, Izzet Kale, “Parasite detection and identification

for automated thin blood film malaria diagnosis”, Computer Vision and
Image Understanding 114 (2010) 21-32.

[7] S. S. Savkare, S. P. Narote, “Automatic System for Classification of

Erythrocytes Infected with Malaria and Identification of Parasite’s Life
Stage”, 2nd International Conference on Communication, Computing &
Security [ICCCS-2012].

140