VFAST Transactions on Software Engineering

http://vfast.org/index.php/VTSE@ 2013 ISSN: 2309-0022

Volume 5, Number 1, September-October 2014 pp-10-17

OCCLUDED RED BLOOD CELLS SPLITTING VIA

BOUNDARIES ANALYSIS AND LINES DRAWING
IN MICROSCOPIC THIN BLOOD SMEAR DIGITAL
IMAGES
NAVEED ABBAS1, DZULKIFLI MOHAMAD2
1
Department of Graphics and Multimedia, ViCube Redsearch Lab, Faculty of
Computing, University Technology Malaysia
[email protected]
2
Department of Graphics and Multimedia, ViCube Redsearch Lab, Faculty of
Computing, University Technology Malaysia
[email protected]

Received June 2013

ABSTRACT. Occluded Red Blood Cells are observed frequently in the thin blood
smear digital images. Separating the occluded Red Blood Cells from the single Red
Blood Cells and further the splitting of occluded Red Blood Cells into single Red
Blood Cells is a challenging job in the computer assisted diagnosis of blood for any
disorder in many diseases like Complete Blood Count Test, Anemia, Leukemia and
Malaria etc. The mentioned problem is also highly laborious in manual microscopy
for the hematologist. Many techniques currently existed for the solution but suffered
from both under and over splitting problems when highly complex occlusions of Red
Blood Cells occurred. Also the existed techniques are not computationally efficient.
In this, paper we address the mentioned problems in realistic, efficient and
automated way by considering the boundaries of the occluded Red Blood Cells
through convex hulls to divide the boundaries on the basis of distance calculation in
a very simple and efficient way split the occlusions according to the number of Red
Blood Cells. Further, we draw lines using Digital differential analyzer graphics line
drawing algorithm at the two respective end points to give cuts to split the occluded
Red Blood Cell. The test results of the proposed technique using standard online
data set of thin blood smear digital images (provided by Centers of diseases Control
and prevention, USA) are presented in this paper by qualitatively analysis through
ground truth with visual inspection.
Keywords: Occluded RBCs, Thin Blood Smears, Anemia, Concavity Regions,
Automatic diagnosing, convex hulls, Concavity Points.

1. Introduction. In the last decade, automated microscopy has got much attention of the researchers for the
purposes that the manual microscopy is still considered as a Gold Standard in the diagnosing of various
disorders in blood. Due to ease of availability and low cost still preserve the right of Gold standard with manual
microscopy but it must also be not ignorable that the process requires high expertise and is too laborious. In
automated microscopy of blood the main and most challenging job is that of occluded Red Blood Cells, White
Blood Cells etc. The occlusions of Red Blood Cells are further divided by this study into two types i.e. Clumps
of Red Blood Cells and Overlaps of Red Blood Cells. The word clump means glue and is used for the situation
in which the Red Blood Cells glued each other and formed long chains. The formation of clumps of Red Blood

10
Cells occurs due to iron deficiency in blood and is the frequent observed in diseases like Anaemia, Leukaemia,
Malaria and many other diseases. The degree of severance of these diseases highly dependant on the number of
Red Blood Cells e.g. in Malaria the Parasitemia is the ratio of infected Red Blood Cells to all Red Blood Cells
observed on the slide. In automated diagnosing, the accuracy of diseases in which counting of Red Blood Cells
is involved are highly affected by clumps of Red Blood Cells due to the consideration of a clump as one single
object while a single clump may encompasses more than six Red Blood Cells. Also important information is
hidden in these clumps. On the other hand overlapped Red Blood Cells are few in numbers not more then four
Red Blood Cells combination and formation is just due to improper slide preparation. Both of these problems
affect the counting accuracy in manual as well as automated microscopy while their cleavage in a proper, easy
and computationally less expensive way is the need of the day.
2. Literature Review. Recently, too many efforts have been made by researchers to develop algorithms for
splitting the clumped and overlapped Red Blood Cells i.e. the clustered Red Blood Cells and show a high degree
of success but still there are gaps which are not addressed or addressed in computationally expensive and some
cases impossible ways. The approaches adopted by previous studies to combat the problems are divided into the
following categories i.e. Morphological operation based includes erosion, dilation or opening closing to split
the clusters of Red Blood Cells [1],[2],[3]. However, the main problem in morphological based approach is that
it works well in overlap of Red Blood Cells not more then two cells but in reality we have some clumps which
are very long chains. Concavity based approaches deal the problems in the way to find out the concavity
regions and some cases the concavity points and split the clustered Red Blood Cells through lines cuts or circles
drawing or ellipses drawing as stated in the studies of [4], [5], [6], [7], [8], [9], [10], [11] and [12]. The
concavity based approaches gives good results but in some cases they are computationally very expensive.
Watershed based techniques includes all form of watershed algorithm based etc as presented by the studies of
[13], [14], [15], [14], [16], [17] and [18]. Watershed based approach have certain degree of success but in dense
clumps it results in over segmentation while in some cases also suffered from the problem of under
segmentation. Edges or contour based techniques can gives solution in the form of analysing split edges and
linkages of contours etc as mentioned in the works of [19], [20] and [21]. This approach working well but
required model based on some templates and burden some both in execution as well as in implementation.
Model based approach gives various models in the form of circles through various theories like Gestalt,
geometrical theories etc as presented in the work of [22], [23] and [24]. The problem in this approach seems to
be unrealistic as due to its highly complex nature and implementation. Also it is computationally too much
expensive. While some studies do not consider the clumps and overlaps of Red Blood Cells for splitting but they
relay on guessing Area based estimation approaches as mentioned in the work of [25] and [26]. The problem
in this approach is that in some cases we want to note the disorder as well in the Red Blood Cell in such case this
approach will fails while also the areas of Red Blood Cells by most of the studies considered as circular, which
is not true as because morphology of the Red Blood Cells highly changes due to any disorder. Circular Hough
Transform based approaches as mentioned by [27], [28], [29] and [30] mainly considered the Red Blood Cells
as circles which is not true as mentioned above.

3. Proposed Methodology. In this study we first separate the single Red Blood Cells from occluded Red
Blood Cells for the purpose of efficiency. Separation is considered by this study on the basis of double check
due to the varaying sizes of Red Blood Cells. The whole idea of this study is depicted in the flow diagram
Figure 1while the simulated image of the whole process is presented as Figure 2. The original input image is
converted to binary by considering histogram and on the basis of the histogram an appropriate value for
thresholding is calculated, the values less then the threshold value is replaced as 0s while the greater values
are replaced with 1s. After this small areas are removed as noise while holes in the RBCs are filled. The
formations of holes in the RBCs are due to the similarity of the center of the RBCs with the background and
because of hemoglobin as RBCs have no nucleus. Next, we find out the convex hulls of all the Red Blood
Cells through Equation mentioned as 1, for the purposes to increase the accuracy. The separation of the single
and occluded Red Blood Cells has been made using two methods to increase the accuracy and enhance
consistency. The first method is the comparison of areas (Number of pixels defining the object) through a
median central tendency measure, all the areas of the RBCs are divided by the median value and the results
are combined in array as values near to or one are considered as single RBCs while values results greater are
considered as occluded RBCs, next these masks are passed to the pixels IDX list of the original image for the
purpose to obtain two images one as single having non-occluded RBCs image while the other as occluded

11
Red Blood Cells having only occluded Red Blood Cells. In the same way, for the second check we consider
Elongation instead of Area through Equation presented as 2, while the rest of the process is the same. We
performed the double check because of variation in the sizes of Red Blood Cells the area is not enough to take
decision.

Figure 1Diagramatic Representation of the overall Research Methodology

Figure 2 The simulated diagrammatic reprsentation of the whole process

12
 |X | |X |

 x
i 1
i i | ( i :  i  0) &   i  1
i 1
(1)

where , |X| = finite set of points, is point |X| while is weight assigned to , the sum of
the weights must be equal to 1 mean normalized.
Length
Elongation  (2)
Breadth
where, Length=Major Axis while Breadth= Minor Axis
After separation of single and occluded RBCs, the next process is that of splitting the occlusions (clumped and overlapped) of
Red Blood Cells. In the splitting process we first trace the boundaries of all occluded Red Blood Cells, first we divide the
boundary into halves using equation 3, then taking the first point of the boundary as P1(x1,y1) while, P2(x2,y2) is the last
point of the first half of the boundary. After, finding the points P1 and P2 we calculate the distance between P1 and P2 using
equation 4, once find out the distance the next process is to divide the boundary according to the number of Red Blood Cells in
each occlusion and take these division points and the distances between the consecutive points using equation 5. The same
points are marked on the other half of boundary and then using Digital Differential Analyzer graphics algorithm to draw lines
in between respective end points and this way after applying a slight erosion( Morphological operator) the occlusions are
cleaved into single Red Blood Cells. The idea is simulated in the diagram depicted as Figure 3.

Figure 3 The simulated diagram of the concept of splitting

Length(Boundary )
Index  (3)
2
where, boundary is the boundary of clumped or overlapped RBCs and index is the
index of boundary containing its points.
( x 2  x1 ) 2  ( y 2  y1 ) 2
D (4)
2

13
 D
No.of .Parts  (5)
No.ofRBCs
where, Number of RBCs we can found while dividing the convex hull area by the
median area of single RBC.

4. Results. In this sect we are presenting the results obtained from the
implementation of the above concepts through visual inspection with ground
truth marked by experts. The experimentation has been carried out on 20
images dataset obtained from DPDx [31].

First Input Image

Figure 4 Matlab Results a) presents original RGB input image b) presents the single
RBCs resulted as separation from occluded RBCs c) Occluded RBCs d) Cleaved RBCs
after giving cuts through lines by considering boundary points

14
 Second Input Image

Figure 5 Matlab Results a) presents original RGB input image b) presents the single
RBCs resulted as separation from occluded RBCs c) Occluded RBCs d) Cleaved RBCs
after giving cuts through lines by considering boundary points

Conclusion. The occlusions splitting is an intermediate process of many computers diagnosing

studies of blood and needs to be solved in an efficient and careful way because human health is
involved. In this paper we produce a concept which is simple and efficient. As previously,
researchers approached to the problem of occlusion of RBC successfully but computationally
non-efficient ways. This study considers the mentioned point and solves the problem in a robust and
real way. The proposed method is tested on a 20 images dataset obtained from DPDx [31] and we
found mostly the results are good while in some high clumps i.e. more than 10 RBCs occlusion we
noted a little deviation but this due to the non-smoothen boundaries. In the future goals of this study,
we suggest that to smooth the boundaries through any mean as will further improve the accuracy
because in high occlusion the proposed method is sensitive to noise.

15
 REFERENCES

[1] Buggenthin, F., Marr, C., Schwarzfischer, M., Hoppe, P. S., Hilsenbeck, O., Schroeder, T. &Theis, F.
J.,(2013).An automatic method for robust and fast cell detection in bright field images from
high-throughput microscopy.BMC bioinformatics(vol. 14. pp. 297.).
[2] Prasad, K., Winter, J., Bhat, U. M., Acharya, R. V. &Prabhu, G. K.,(2012).Image analysis approach for
development of a decision support system for detection of malaria parasites in thin blood smear
images.Journal of digital imaging(vol. 25. pp. 542-549.).
[3] Amit Kumar, P. a. C., Prof. P. U. Tembhare, Prof. C. R. Pote,(2012).Enhanced Identification of Malarial
Infected Objects using Otsu Algorithm from Thin Smear Digital Images.International Journal of Latest
Research in Science and Technology(vol. Vol.1. pp. Page No159-163 ).
[4] Latorre, A., Alonso-Nanclares, L., Muelas, S., Peña, J. &Defelipe, J.,(2013).Segmentation of neuronal
nuclei based on clump splitting and a two-step binarization of images.Expert Systems with
Applications(vol. 40. pp. 6521-6530.).
[5] Tafavogh, S., Navarro, K. F., Catchpoole, D. R. &Kennedy, P. J.,(2013).Segmenting Neuroblastoma
Tumor Images and Splitting Overlapping Cells Using Shortest Paths between Cell Contour Convex
Regions.Artificial Intelligence in Medicine (vol. pp. 171-175.).
[6] Zhang, C., Sun, C., Su, R. &Pham, T. D.,(2012).Segmentation of clustered nuclei based on curvature
weighting.Proceedings of the 27th Conference on Image and Vision Computing (vol. pp. 49-54.).
[7] Wang, H., Zhang, H. &Ray, N.,(2011).Clump splitting via bottleneck detection.Image Processing (ICIP),
18th IEEE International Conference on (vol. pp. 61-64.).
[8] Kumarasamy, S. K., Ong, S. &Tan, K. S.,(2011).Robust contour reconstruction of red blood cells and
parasites in the automated identification of the stages of malarial infection.Machine Vision and
Applications(vol. 22. pp. 461-469.).
[9] Wen, Q., Chang, H. &Parvin, B.,(2009).A Delaunay triangulation approach for segmenting clumps of
nuclei.Biomedical Imaging: From Nano to Macro ISBI'09. IEEE International Symposium on (vol. pp.
9-12.).
[10] Makkapati, V. V. &Naik, S. K.,(2009).Clump splitting based on detection of dominant points from
contours.Automation Science and Engineering, 2009. CASE 2009. IEEE International Conference on (vol.
pp. 197-201.).
[11] Gurcan, M. N., Boucheron, L. E., Can, A., Madabhushi, A., Rajpoot, N. M. &Yener,
B.,(2009).Histopathological image analysis: A review.Biomedical Engineering, IEEE Reviews in(vol. 2.
pp. 147-171.).
[12] Cloppet, F. &Boucher, A.,(2008).Segmentation of overlapping/aggregating nuclei cells in biological
images.Pattern Recognition, 2008. ICPR 2008. 19th International Conference on (vol. pp. 1-4.).
[13] Tulsani, H.,(2013).Segmentation using morphological watershed transformation for counting blood
cells.IJCAIT(vol. 2. pp. 28-36.).
[14] Ferro, L., Leal, P., Marques, M., Maciel, J., Oliveira, M. I., Barbosa, M. A. &Quelhas,
P.,(2013).Multinuclear Cell Analysis Using Laplacian of Gaussian and Delaunay Graphs.Pattern
Recognition and Image Analysis (vol. pp. 441-449.).
[15] Hodneland, E., Kögel, T., Frei, D. M., Gerdes, H.-H. &Lundervold, A.,(2013).CellSegm-a MATLAB
toolbox for high-throughput 3D cell segmentation.Source code for biology and medicine(vol. 8. pp.
1-24.).
[16] Schmitt, O. &Reetz, S.,(2009).On the decomposition of cell clusters.Journal of Mathematical Imaging
and Vision(vol. 33. pp. 85-103.).
[17] Schmitt, O. &Hasse, M.,(2009).Morphological multiscale decomposition of connected regions with
emphasis on cell clusters.Computer Vision and Image Understanding(vol. 113. pp. 188-201.).
[18] Špringl, V.,(2009).Automatic Malaria Diagnosis through Microscopy Imaging.FACULTY OF
ELECTRICAL ENGINEERING (vol. Higher Diploma. pp. 128.).
[19] Gonçalves, W. N. &Bruno, O. M.,(2012).Automatic system for counting cells with elliptical shape.arXiv
preprint arXiv:1201.3109(vol. pp.
[20] Kong, H., Gurcan, M. &Belkacem-Boussaid, K.,(2011).Splitting touching-cell clusters on
histopathological images.Biomedical Imaging: From Nano to Macro, 2011 IEEE International
Symposium on (vol. pp. 208-211.).

16
[21] Kong, H., Gurcan, M. &Belkacem-Boussaid, K.,(2011).Partitioning histopathological images: an
integrated framework for supervised color-texture segmentation and cell splitting.Medical Imaging, IEEE
Transactions on(vol. 30. pp. 1661-1677.).
[22] Köppen, M., Yoshida, K. &Valle, P. A.,(2007).Gestalt theory in image processing: A discussion
paper.Proceedings (vol. pp. 1-3.).
[23] Jiang, H., Ngo, C.-W. &Tan, H.-K.,(2006).Gestalt-based feature similarity measure in trademark
database.Pattern recognition(vol. 39. pp. 988-1001.).
[24] Airsang, U., Ghorpade, V. &Rajamani, S. T.,(2013).Contour feature-point tagging as a mechanism for
automated splitting of highly-occluded and dissimilar-sized cells in blood smear images.Image
Information Processing (ICIIP), 2013 IEEE Second International Conference on (vol. pp. 470-475.).
[25] Owais Shaikh, M. G., Neharika Bhat,Roshan Shetty (2013-2014).AUTOMATED RED BLOOD CELLS
COUNT.Department of Computer Engineering, Rizvi College of Engineering (vol. B.E., pp. 29.).
[26] Nguyen, N.-T., Duong, A.-D. &Vu, H.-Q.,(2011).Cell splitting with high degree of overlapping in
peripheral blood smear.Int J Comp Theory Eng(vol. 3. pp.
[27] Mahmood, N. H., Lim, P. C., Mazalan, S. M. &Razak, M. a. A.,(2013).BLOOD CELLS EXTRACTION
USING COLOR BASED SEGMENTATION TECHNIQUE.pp.
[28] Grietinfo.In, grietinfo.in/projects/MAIN/BME2013/cd-8-project%20report_1_.pdf, 20-03-2013.
[29] Mahmood, N. H. &Mansor, M. A.,(2012).Red blood cells estimation using Hough transform
technique.Signal & Image Processing: An International Journal (SIPIJ)(vol. 3. pp. 53-64.).
[30] Ramesh, N., Salama, M. E. &Tasdizen, T.,(2012).Segmentation of haematopoeitic cells in bone marrow
using circle detection and splitting techniques.Biomedical Imaging (ISBI), 2012 9th IEEE International
Symposium on (vol. pp. 206-209.).
[31] Dpdx,(2002).DPDx, Laboratory Identification of Parasites,Centers of diseases Control and Prevention.

17