IRJET- Detection of White Blood Sample Cells using CNN

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 03 | Mar 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 3839
DETECTION OF WHITE BLOOD SAMPLE CELLS USING CNN
K.SONA1, C.SRIRAGAVI2, A.VIJAYA3 B.V.VARSHINI 4
1K.SONA Student
2C.SRIRAGAVI Student
3 A.VIJAYA Student
4 B.V.VARSHINI Assistant proffesor Dept. of computer science and Engineering, RMDEngineering college,
Tamil Nadu, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This paper presents a new classification system
for white blood cells to recognize 4 types of white blood cells.
For the segmentation of white blood cellsfromimages, Wecan
segment from an image a white blood cell. Convolutionneural
network has already demonstrated power in many fields of
application and is accepted as a better approach by more and
more people as a better approach than traditional models of
machine learning. The implementation of Convolution Neural
Networks (CNN), in particular, bringsenormousbenefitstothe
medical field, where the processing and analysis of a huge
number of images is required. This paper implements a
Convolution Neural Network for the classification of the four
blood subtypes A CNN-based framework for the automatic
classification of blood cells. Experiments are carried out on a
dataset of 15k images of blood cells with their subtypes, and
the proposed CNN approach generated improved results and
reduced the rate of error compared to other models. A CNN
model based on Deep Learning, where deeplearningenhances
the extraction capability and smooth scaling of features in
case of increased parameters and 81 percent accuracy was
achieved in the classification of WBCs.
Key Words: : White Blood Cells ,Deep Learning,
Convolutional Neural Network.
1.INTRODUCTION
The microscopic inspection of blood provides
diagnostic information concerning patients’ health
status. The differential blood count inspection results
reveal a wide range of significant hematic pathologies.
For example, the presence of infections, leukemia and
certain specific types of cancers can be diagnosed
based on the classification results and the white blood
cell count. Experienced operators perform the
traditional method for differential blood count.They
use a microscope and count the percentage of each
type of cell that is counted within a area of interest.
This manual process of counting is obviously very
tedious and slow. Furthermore, the classification and
accuracy of the cell may depend on the operators '
capabilities and experiences. Consequently, the need
for an automated system of differential counting
becomes inevitable. Recently, a number of different
approaches have been proposed to implement a white
blood cell recognition system based on image
processing. White blood cell classification usually
involves the following three stages: a white blood cell
segmentation from an image,theextractionofeffective
features, and a classifier design.to some extent, the
performance of an automatic white blood cell
classification system depends on a good segmentation
algorithm to segment white blood cells from their
background. We extract three types of characteristics
from the segmented cell region below. These
characteristics are fed into three different neural
networks for the classification of five white blood cell
types. We extract three types of characteristics from
the segmented cell region below.
These characteristics are fed into three different
neural networks for the
classification of five white blood cell types.
Fig.1 Leukemia blood

1.DATASET
Classify it as either polynuclear or mononuclear given s
tained image of a white blood cell. Note that while
lymphocytes and monocytes are
mononuclear, eosinophils and neutrophils are pol
polynuclear.Leukocyte (white
blood cells) evaluation is
the primary step in diagnosing many diseases related t
o the blood. The evaluation of the five
major leukocyte subtypes -Neutrophils, Lymphocytes,
Eosinophils, Monocytes, and Basophils can help to
identify various diseases.Manual counting involves whi
te blood cell counting (WBC) done primarily by medical
operators, whose accuracy is highly dependent on the
skills of the operatorWhile the impedance based hemat
ology analyzer has its advantages, it may mistakenly cla
ssify cell types as white blood cells as their primary cla
ssification parameters are limited to size and particle n
umber.Therefore, precise, time saving diagnostic syste
ms need to be introduced in order to accurately classify
the count of WBC to determine different diseases.
Fig.2 Dataset
2. RELATED WORK
A review based on segmentation techniques (Adollah
et al., 2008) argues that conventional color - based
methods and thresholding methods are simple to
sacrifice accuracy, whereas methods such as region -
growing can offerhighaccuracywithhighcomputation
costs.Some methods work directly on the RGB color
space, while others workdirectlyonHSIor CMYKcolor
space, referring to the color - based segmentation
methods. In general, methods based on the S -
component outperform those based on the RGB.
By leveraging the CMYK color models, Putzu et al.
(2013) attempts to build the feature vector. They find
out that all the other components except white blood
cells have some yellow color in them, while leukocytes
show a good contrast in the CMYK color model's Y
component.Young adopted four characteristics and a
minimum distance classifier to classify 5 cell types[4 ].
Wavelet transform coefficients and artificial neural
networks used by Sheik et al. to recognize white blood
cells, red blood cells, and platelets[8 ].Bikhet et al.
selected 10 features and adopted a minimum distance
classifier to build an automatic classification system
that achieved a 91 percent correct classification rate
for a 71 white blood cell database[6 ].Piuri and Scotti
proposed an automatic classification and detection
system based on 23 morphological characteristics and
a neural classification system[7 ].A classification
system was proposed in [ 5 ] based on own - cell and
parametric characteristics.Asystemthatachieveda77
percent classificationratefortheclassificationofwhite
blood cells in the bone marrow was reported in [ 9 ].
Nilufar et al. proposed a system of classification based
on joint histogram - based characteristics and a vector
support machine[10 ]. Osowski et al. presented a
genetic algorithm and a vector supporting machinefor
the recognition of blood cells in the bone marrow[11 ].
Rezatofighietal.adoptedmorphologicalcharacteristics
and textural characteristicsextractedfromlocalbinary
pattern (LBP) and then trained two types of neural
networks forclassification[1].Tabrizietal.adoptedthe
main component analysis for selection of features and
used a neural network of learning vector quantization
to classify 5 types of white blood cells[2]. Ghosh et al.
provided Naïve Bayes classifier with four statistically
significant features to classify five types of whiteblood
cells with an overall accuracy of 83.2 percent[3 ]. Each
approach has its own considerations for adopting
features and classifiers of what kinds.
Model Representation
CNN should be useful in classifying images and
recognizing objects. It takes the rawpixelsasinputand
produces an outcome indicating the probabilities that

the input belongs to different classes. Instead of
implementing the fully connected structure in each
layer, CNN imposes two additional layers, convolution
and pooling, which can significantly reduce the
parameter magnitude.The convolutionary operation
entitles the Convolution layer to extract the features
from the input images. Kernel modification generates
the features that have the effects of the variant, such as
object identification,edgedetection,imagesharpening,
etc.The pooling procedure is also referred to as sub -
sampling or down - sampling, which is intended to
reduce the convolved characteristics produced by the
convolution operator with the incentive to remain the
significant information.There are various methods of
pooling, such as maximum, average, summation,etc.In
our method we use the maximize pooling, an example
is shown in Fig. 2 A single node is connected to all
nodes in the previous layer in a fully connected layer.
Moreover, more than one hidden layer may apply, and
the differentclassificationoperatormaybeusedbythe
output layer.
i).Convolution
Model Working ConvolutionaryNeuralNetwork
is extensively used to classify images as it uses
neighboring pixel informationtoeffectivelysamplethe
image and then perform predictions resulting in high
accuracy.They also use neural networks that can be
scaled to large datasets. It includes a complex neural
feed forward network that includes convolutions,
pooling, andclassification.Thetermconvolutionrefers
to calculating similarities betweentwofunctionswhen
one function passes (or convolutes) over another
function.
ii).Max pooling
When the image is too large, pooling is used to
reduce the number of parameters, followed by
training-based classification. A computer-basedimage
is perceived as a collection of three-dimensional
numbers or pixels. Width, depth and height. Thus,
CNN's core operations are matrix multiplications.As a
feature extraction and classification, the functioningof
CNN can be divided into two parts. Convolution is
primarily responsible for extraction of features. By
sliding a filter (feature vector) over input data, it
creates a feature map. This is achieved by multiplying
the matrix at each location to extract various parts of
the image and summarize the result to a feature
map.This operation is performed multiple times to
obtain multiple feature maps using different filter
values. This is the convolution layer's output. The
output is non - negative and non - linear in the real
world, so an activation function is applied to it.In this
paper we add a layer of pooling between the layers of
convolution to reduce the number of computations in
the network by reducing the dimensionality.Various
types of pooling such as, max pooling (taking the
maximumofadjacentpixelsafterconvoluting),average
pooling (taking the average of adjacent pixels after
convoluting) and sum pooling (taking into account all
adjacent pixel values) can be used.
Fig 3. Convolutional Neural Network
Various types of pooling like, max pooling (taking the
maximum of neighboring pixels after convoluting),
average pooling (taking the average of neighboring
pixels after convoluting) and sumpooling(considering
all neighboring pixel values) can be used.
iii). Fully Connected Layers
One node is connected to all nodes in the
previous layer in a fully connected layer. Moreover,
more than one hidden layer may apply, and the
different classification operator may be used by the
output layer.

IV. Experimental results:
Setups
For this analysis , these setups have been
utilized to achieve the result and can be seen below.
Processor :GPU
Language used to implement the project: Python
Tools Used: list of Python libraries are Pandas ,
Matplotlib, Scikit Learn , Tensor flow etc.
Development Environment Used :Google Collab.
As shown in Table 1, for precision, a value of 83% is obt
ained, which means that 83% of the times we have the
correct (expected) result. Recall is a measure of a predi
ction model's ability to select instances from a data set
of a certain class.A 78 percent recall average shows tha
t 78 percent of the time the model has been able to corr
ectly classify a specific class. F1 score transmits the bal
ance
between accuracy and memory. We get a 78 percent F
1 score.
Table -1
METHOD CNN
Class Precision
Recall Fscore
NEUTROPHIL 0.60 0.86 0.70
EOSINOPHIL
0.82 0.65 0.71
MONOCYTE
0.84
0.72
0.73
LYMPHOCYTE
1
1
1
AVERAGE
0.81
0.80
0.78
Table 1. Model results Cell Subtype preciseness Recall
F1-scorewhitecorpusclezero.570.880.69whiteblood
cell zero.96 0.53 0.68 white blood corpuscle zero.84
0.81 0.83 white corpuscle zero.97 0.92 0.94 Average /
Total zero.83 0.78 0.78 As shown in table one, a worth
of eighty three is obtained for preciseness, which
suggests that eighty three of the days we have a
tendency to get the proper (expected) result. Recall
may be a live of the flexibility of a prediction model to
pick out instances of a definite category from a
knowledge set. A recall averageofseventy-eightshows
that seventy-eight of the days the model was properly
able to categorify a specific class. F1 score conveys the
balance between preciseness and recall. we tend to
acquire Associate inNursingF1-scoreofseventy-eight.
3. CONCLUSIONS
In this paper, a classification model based on deep lear
ning was implemented using Convolutional Neural Net
work to classify the image dataset into four WBCs —
neutrophils, lymphocytes, eosinophils, and monocytes.
The model achieved 81 percent accuracy on the datase
t.
REFERENCES
[1] Rezatofighi SH, Khaksari K, Soltanian-Zadeh H.
Automatic recognition of five types of white
blood cells in peripheral blood. Proceedings of
the International Conference of Image Analysis
and Recognition; 2010; pp. 161–172.
[2] Tabrizi PR, Rezatofighi SH, Yazdanpanah MJ.
Using PCA and LVQ neural network for
automatic recognition of five types of white
blood cells. Proceedings of the 32nd Annual
International Conference of the IEEE
Engineering in Medicine and Biology Society
(EMBC '10); September 2010; pp. 5593–5596.
[3] Ghosh M, Das D, Mandal S, et al. Statistical
pattern analysis of white blood cell nuclei
morphometry.ProceedingsoftheIEEEStudents’
Technology Symposium (TechSym '10); April
2010; pp. 59–66.
[4] Young IT. The classification of white blood
cells. IEEE Transactions on Biomedical
Engineering. 1972;19(4):291–298.
[5] Yampri P, PintaviroojC,DaochaiS,Teartulakarn
S. White blood cell classification based on the
combination ofeigencellandparametricfeature
detection. Proceedings of the 1st IEEE
Conference on Industrial Electronics and
Applications (ICIEA '06); May 2006; pp. 1–4.

[6] Bikhet SF, Darwish AM, Tolba HA, Shaheen SI.
Segmentation and classification of white blood
cells. Proceedings of the IEEE Interntional
Conference on Acoustics, Speech, and Signal
Processing; June 2000; pp. 2259–2261.
[7] Piuri V, Scotti F. Morphological classification of
blood leucocytes by microscope images.
Proceedings of the IEEE International
Conference on Computational Intelligence for
Measurement Systems and Applications(CIMSA
'04); July 2004; pp. 103–108.
[8] Sheikh H, Zhu B, Tzanakou EM. Blood cell
identification using neural networks.
Proceedings of the IEEE 22nd Annual Northeast
Bioengineering Conference; March 1996; pp.
119–120.
[9] Umpon NT, Dhompongsa S. Morphological
granulometric features of nucleus in automatic
bone marrow white blood cell
classification. IEEE Transactions on Information
Technology in Biomedicine. 2007;11(3):353–
359.
[10] Nilufar S, Ray N, Zhang H. Automatic blood cell
classification based on joint histogram based
feature and BhattacharyaKernel.Proceedingsof
the 42nd Asilomar Conference on Signals,
Systems and Computers (ASILOMAR '08);
October 2008; pp. 1915–1918.
[11] Osowski S, Siroic R, Markiewicz T, Siwek K.
Application of support vector machine and
genetic algorithm for improved blood cell
recognition. IEEE Transactions on
Instrumentation and
Measurement. 2009;58(7):2159–2168.

IRJET- Detection of White Blood Sample Cells using CNN

More Related Content

What's hot (11)

Similar to IRJET- Detection of White Blood Sample Cells using CNN (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Detection of White Blood Sample Cells using CNN