Analyzing techniques for Detection of Blood Count Related Diseases from Blood Cell Images

Abstract

Medical Image anyalysis plays an important role in quality healthcare. The traditional method of manual count under the microscope yields inaccurate results and put an intolerable amount of stress on the Medical Laboratory Technicians. Although there is hardware solutions such as the Automated Hematology Counter, developing countries like India are not capable of deploying such prohibitively expensive machines in every hospital laboratory in the country. As a solution to this problem, this research project aims to provide a software-based cost effective and an efficient alternative in recognizing and analyzing blood cells.

Full text

International Conference On “Computational Intelligence Applications” -2010, 03-05 March 2010
Analyzing techniques for Detection of Blood Count Related Diseases from
Blood Cell Images
Dr. JayatiGhoshal Dr. R. V. Dharaskar Dr. V.M. Thakare Mr. LekhrajVilhekar
Prof. Dept. of CSE Prof, Head Dept of CSE Prof & Head, PGDCS, III SEM M.Tech
GHRCE, Nagpur GHRCE, Nagpur Amravati University, (C.S.E.) GHRCE,
Amravati Nagpur
lekh.vilhekar@gmail.com
Abstract
Medical Image anyalysis plays an important
role in quality healthcare. The traditional method of
manual count under the microscope yields inaccurate
results and put an intolerable amount of stress on the
Medical Laboratory Technicians. Although there is
hardware solutions such as the Automated
Hematology Counter, developing countries like India
are not capable of deploying such prohibitively
expensive machines in every hospital laboratory in
the country. As a solution to this problem, this
research project aims to provide a software-based
cost effective and an efficient alternative in
recognizing and analyzing blood cells.
Keywords: Medical Image Analysis, Standard
BloodCell Count, Differential White Blood Cell
Count, ImageSegmentation,
1 Introduction
1.1 Introduction to blood cell recognition
& counting
In almost all of the medical laboratories, blood
cell countreports are taken on physicians’
recommendation, inorder to assist the diagnosis of
the particular ailments ofthe patients. In fact this test
is one of the most frequenttests carried out in a
medical laboratory.
Producing blood cell reports can be
broadlycategorized into 2 areas:
1. Standard Count for Red Blood Cells (RBC), White
Blood Cells (WBC) and Platelets (PLT)
2. Differential Count for WBC
1.2 Standard Count for all cell types
For this type of a count, the MLT is produced
with a Hemacytometer shown in the figure, which is
also knownas the Counting Chamber (CC). The count
is obtained forall the different blood cell types (i.e.
RBC, WBC andPLT).Since blood cells are counted
per unit volume (perliter), it is vital that the volume
of blood, in which the cellsare counted, corresponds
to a known quantity. Usually, theCC has a special
objective slide containing counting gridsof the size
3mm x 3mm. In the figure below, there are 2such
counting grids in the middle.
Figure 1-1 Counting Chamber
The schematic representation of the counting grid is
represented in the figure below. The counting grid is
composed of 9 big squares, measuring 1 x 1 mm.
Fromthese squares, the central square contains 25
medium sizedsquares each measuring 0.2 x 0.2 mm.
These are furtherdivided into 16 small squares each
measuring 0.05 x 0.05mm. The large central square is
also called the“erythrocyte” grid. The squares
highlighted in redcorrespond to 80 small squares, and
are used to establishthe RBC and PLT counts. The
large squares marked inblue are used to establish the
leukocyte or the WBC count.
239

International Conference On “Computational Intelligence Applications” -2010, 03-05 March 2010
Figure 1-2 Schematic representation of the
counting grid
In the RBC and Platelet count, the erythrocyte grid in
the middle is examined much closely.
1.3 Differential Count for WBC
White blood cells in human blood can be put in
to fivemain sub categories as follows.
1. Neutrophils
2. Eosinophils
3. Basophils
4. Monocytes
5. Lymphocytes
In differential counting, the white blood
cells arecounted and classified in to the above
categories andoutput the number of each of the above
types as apercentage.
The above five types are broadly divided in
to two categories; the “phils” category and the
“cytes” category.
This division is done based on how the
names of the abovetypes end. Therefore, Neutrophils,
Eosinophils and Basophils fall in to the “phils”
category while Monocytes and Lymphocytes fall in
to the “cytes” category.The nucleus of each of the
above types has a uniqueshape, and this is the most
important feature used in cellclassification. In
addition to the shape of the nucleus, the“phils”
category has granules with in the blood cell whereas
“cytes” category does not have granules.
1.3.1 Types of White blood cells
Neutrophils
Neutrophils are identified based on the blobbed
nucleussmall granules. The nucleus has 3 to 5 blobs.
Figure 1.3-3 A Neutrophil
Eosinophils
The Eosinophil is distinguished by its red granules
andblobbed nucleus. The granules are larger than that
of aNeutrophil.
Figure 1.3-4 An Eosinophil
Basophils
A Basophil is characterized by a lobed nucleus and it
isfilled by large blue-black granules that sometimes
coverthe nucleus.
Figure 1.3-5 A Basophil
Lymphocytes
These have round shaped nucleus. They have large
nucleus to cytoplasm ratio. No granules are present.
Figure 1-6 A Lymphocyte
240

International Conference On “Computational Intelligence Applications” -2010, 03-05 March 2010
Monocytes
These have horse-shoe shaped nucleus. No granules.
Thenucleus to cytoplasm ratio is high but less than
that of aLymphocyte.
Figure 1-7 A Monocyte
2 Current Methods for blood cell
counting
2.1 Manual Method
In general, the MLT would prepare the slide
andexamine it under the microscope. As explained in
HemoSurf [2] and from the requirements of the
LTs,different parameters are taken for each of the cell
typecounts. These inputs are used in different
equations andthe count is obtained.
2.1.1 Manual Red Blood Cell (RBC) Counting
In a manual RBC count, 10 μl of blood is diluted
in1990 μl of dilution solution. This results in a
dilution of1:200. This suspension is usually well-
mixed and beimmediately placed into the counting
chamber. Afterapproximately 3 minutes, the RBCs
will have settled, andthe MLT begins counting the
RBCs in 80 small squares.The calculation is achieved
by following the formulabelow using these factors:
i. The number of RBCs counted in the small
squares
ii. The dilution of the cell solution
iii. The number of counted small squares
iv. The volume above one small square
RBCs/μl=(number of counted
RBCs(i)*dilution(ii))/(number of counted
squares(iii)*volume above one small
square(iv))
Equation 2-1 Calculation of RBC per micro-liter
In 80 small squares, around 400 RBCs are countedfor
normal values. This yields a coefficient of
variation(variability) of ±5%. This constitutes the
highestacceptable amount of random error
(accuracy).
2.1.2 Manual White Blood Cell (WBC)
Counting
In the manual WBC count, 50 μl of blood is mixed
together with 950 μl dilution solution. This
constitutes adilution of 1:20. The RBCs will be lysed
(i.e. cells aredestroyed by bursting), and the WBC
nucleus is stained.The counting chamber is
immediately filled after mixing.After 2 minutes, the
MLT begins counting the WBCs inthe 4 large
squares.
Calculation of the WBC count is achieved
byfollowing the formula below using these factors:
i. The number of WBCs counted in
the big squares
ii. The dilution of the cell solution
iii. The number of counted big squares
iv. The volume above a big square
WBCs/μl=(number of counted
WBCs(i)*dilution(ii))/(number of counted
squares(iii)*volume above one small
square(iv))
Equation 2-2 Calculation of WBC per micro-liter
Since the WBC count shows greater
physiologicvariations, a coefficient of variation of
±10% is accepted.In addition since normoblasts (i.e.
nucleatedprecursors of RBCS which would get
erroneously countedas WBCs) are not recognized in
this procedure, a large number of normoblasts can
invalidate the result. This isalso true for mechanical
WBC counts. The number of normoblasts can only be
accurately identified in a bloodfilm and is expressed
as the number of normoblasts per100 WBCs. If there
are more than 5 normoblasts per 100WBCs, a
correction must be used .The normoblast correction is
achieved by using theformula below:
Corrected WBC count
*100
Equation 2-3 Normoblasts correction for WBC
count
2.1.3 Manual Platelet (PLT) Counting
241

International Conference On “Computational Intelligence Applications” -2010, 03-05 March 2010
The PLT count is very similar to the RBC
count.However unlike in the RBC count, the solution
has adilution of 1:20 and the RBCs are completely
lysed beforeanalysis. The suspension is then mixed
and put into thecounting chamber. The chamber is
left in a moistenvironment for 20-30 minutes so the
platelets can settlewithout the chamber drying. Like
in the RBC count, 80small squares are
counted.Calculation of the PLT count is achieved by
usingthe formula below using these factors:
i. The number of PLTs counted in thesmall squares
ii. The dilution of the cell solution
iii. The number of counted squares
iv. The volume above a square
Equation 2-4 Calculation of PLT per micro-liter
A coefficient of variation of ±10% is acceptable for
PLTs and at least 100 platelets must be counted.
2.1.4 Drawbacks of the manual method
• Visual inspection of microscopic images is
timeconsuming and exhaustive. If the counting
processis interrupted, the MLT has to start over again
fromthe scratch.
• Cell analysis is realized by an experienced MLT
bycomparing what she sees with images of cell
typesshe is familiar with. An amateur MLT would
haveto check with medical literature to confirm on
thecell types to determine the count of a given
sample.Thus these manual methods are susceptible
tohuman fatigue that can easily result in errors.
• After the blood cell slides have been analysed,
theyare kept away. There is no quick and easy way
ofretrieving analyzing lot of images for
futurereference as with a computerized system.
2.2 CellaVision
This is a commercial product developed
byCellaVisionAB. The CellaVision DM Analyzer
(3)consists of a fully-automated system of counting
bloodcells. To analyze a sample of blood the
following steps areundertaken:
1. The vials containing the blood are bar-coded
foridentification.
2. All the vials are fed in to the
AutomatedHematology Counter for the Standard
Count.
3. Samples are taken for morphological review,where
a piece of hardware known as the “slidemaker” or
“stainer” is used to get blood on to athin film on a
slide.
4. These slides are kept in a special container
andplaced in a hardware which is used to
automatethe manual differential count.
5. The slides are then moved under a
microscopeusing a robot arm and images of the
WBCs aretaken.
6. The images are analysed and classifiedaccordingly.
2.2.1 Drawbacks of CellaVision
• This product is not widely available.
• The cost of a CellaVision unit is unbearable for
medical laboratories in developing countries.
3 Aim of the Work
This software can be used for
recognizingand analyzing blood cells and produce
blood countreports. This is capable of performing
standard countswhich comprises of RBC counts,
WBC counts, PLTcounts and differential counts.The
operation is purely based on imageprocessing and
computer vision technologies. The input isan image
of the already prepared slide containing a film
ofblood, taken from a special camera attached to an
ordinarymicroscope. The software would not
consider about thepreparation of the slides to be
viewed through amicroscope. It assumes that the
preparation of the slides isdone by trained
professionals and the image taken througha
microscope is identical to what the MLT sees through
microscope. Therefore, unlike in CellaVision [3],
This workdoes not replace the MLTs. They are
required to preparethe slides as they normally do. The
main objective ofthis work is to provide a software
solution which is costeffective as well as efficient for
under developed countries tobe widely utilized in the
healthcare industry.The software is designed to be
extensible, tiered and ahighly efficient with great
consideration on ease of use forthe end user.
4 Work Functionality
The methodology followed with the
overallprocessing of a microscopic blood smear
image is showing the following figure.
242

International Conference On “Computational Intelligence Applications” -2010, 03-05 March 2010
Figure 4-1 Overall flow of processing
Based on the type, an image going through this flow
ofprocessing would end up in the ‘Standard Count’ or
in the‘Differential Count’. Furthermore, standard
count can beperformed in the automatic mode, where
the applicationwould count the cells in the entire
image or in theinteractive mode, where it would
require the MLT toselect a region of interest and the
locations of few bloodcells in the image.
In the differential count, WBCs are
differentiatedbased on the shape of their nucleus. The
percentage areaof the nucleus and the color of the
granules are used toidentify between the different
types of WBCs.
4.1 Functionality of the System
Basic functionality consists of following:
• Standard count
Standard count is performed for RBC, WBC
andplatelets. Given the image, the user will have
theoption of either going in to the automatic mode
orthe interactive mode. In the automatic mode,
theapplication will determine the standard squares
forthe count operation, whereas in the interactive
modethe user will have to select the regions in the
image(i.e. squares preferred for counting) and
perform thecount.
• Differential count
Differential count is performed for WBC only.
Itprovides a breakdown of the individual counts
forlymphocytes, monocytes, neutrophils,
eosonophilsand basophils along with the percentage
accuracy.The user has the option of selecting the
image ofhis/her preference from the thumbnail view
andloading in to the main image container for
analysis.
• Report generation
After getting the blood count, user can request
toprepare blood count reports. These reports can
beeither standard reports which consist of RBC
andPlatelet count or differential count reports. All
thesereport formats will conform to HL7 or
HIPAA,standards for health care specific data
exchangebetween computer applications.
5 Extensible architecture
The system is designed on the conventional three
tieredarchitecture composed of the Presentation
Logic tierwhich contains all the logic related to the
user interfacesand the display options, the
Application Logic tier whichcontains all the logic
related to blood cell counting,recognition, reporting,
login and logging, and the DataAccess tier which
handles all the interactions theapplication has with
the database.The logic related to counting and
recognition of bloodcells dominates the system. The
presentation logic playsthe next important role and
the system has a very thin dataaccess layer.
Therefore, there was no need to go for morethan
three tiers since that would create a number of
verythin layers and it would be difficult to manage
the system. The datalayer is implemented separately.
Figure 5-1 Abstract System View
243
Acquire
image
from
microsco
pe
Preproc
ess
Obtain
image
data
Std. Count
Differentia
l Count
D
a
t
a
b
a
s
e
Report

International Conference On “Computational Intelligence Applications” -2010, 03-05 March 2010
6 Results
The various image operations can be
perfomed over the images like edge detection, image
enhancement and image segmentation. In image
enhancement , the image is colored with hsv color
space or gray image. In edge detection the edges are
found for the cells in the images, such that they can
be easily extracted. Once the cells are extracted they
needs to be segmented.
7 Conclusion
Here the work aims at developing the
system which can easily detect the disease with less
time complexity with suitable results.
8 References
1. http://www.wadsworth.org/chemheme/heme/
microscope/
2. Gonzalez, Digital Image Prcoessing,
Addison-Wesley, 1993.
3. K.S. Kim, P.K. Kim, J.J.Song, and Y.C. Park,
( 2002), “Analyzing blood cell image to
distinguish its abnormalities”, in Proc., ACM
international conference on Multimedia.
4. BAO, H.F. Den, H.H. Gelsema, Smeulders,
“Automated while blood cell classification
revisited “. Medical Informatics 12, 1(1997),
23-31 .
5. LiboZeng, Hong Zeng, Ningningguo, “
Precise segmentation of white blood cells by
using multispectral imaging analysis
techniques “., at first international conference
on Intelligent networks and intelligent
systems, 491-494.
6. Fabio Scotti , “Robust Segmentation and
Measurements Techniques of White cells in
blood microscope images.” 24-27 April 2006
at IMTC2006 Instrumentation and
Measurement Technology Conference,
Sorrento, Italy, 43-48.
7. Q. Liao and Y. Deng, “ An accurate
segmentation method for white blood cell
images”, in IEEE International symposium on
Biomedical Imaging., 2002.
8. Dempster, Di Ruberto, “Morphological
Processing of Malarial Slide Images”, Matlab
DSP Conference 1999, Nov, 16, Espoo,
Finland.
9. P. Bamford and B. Lovell, “Method for
accurate unsupervised cell nuclueus
segmentation”, in Proc., of the Engineering in
Medicine and Biology society conference,
2001.
10. NeelamSinha, A.G. Ramkrishnan, “Blood
cell segmentation using EM algorithm”.
11. Carboni, DanielaTagliasacchi and Giorgio.
BlOOD CELLS [online] april
1999.http://www.funsci.com/fun3_en/blood/b
lood.htm.Accessed: Sep 2006
12. Leukocyte at eMedicine Dictionary.
eMedicine dictionary.
[Online]http://www.emedicine.com/asp/dictio
nary.asp?keyword=leukocyte. Accessed: Sep
2006OutputInputInput cell identified (as a
percentage)NeutrophilsEosino –
philsBasophilsLympho –
cytesMonocytesNeutrophils85% 14% 0% 0%
1%Eosinophils13% 80% 0% 0%
7%Basophils0% 0% 12% 61%
27%Lymphocytes0% 0% 3% 82%
15%Monocytes0% 2% 3% 18% 77%
13.HemoSurf - An Interactive Hematology
Atlas.[Online] Division of Instructional
244

International Conference On “Computational Intelligence Applications” -2010, 03-05 March 2010
Media, Institute forMedical Education,
School of Medicine, University ofBern.
http://www.aum.iawf.unibe.ch/HemoSurf/D
emo_E/Lab/count_manual.htm. Accessed:
Nov 2006
245