I. Introduction

advertisement
Recognition of White Blood Cells by using
Feature Points Matching
May Thida Aung, Nang Aye Aye Htwe

Abstract—A simple classification scheme using color
information and morphology will be used. As the first step of
WBC segmentation process, the features are extracted from the
blood smear. The nucleus shape is one of the key factors in
deciding how to classify WBCs. The second step (WBC
classification) is to define a set of features using the information
from the cytoplasm and nuclear regions to classify WBCs using
is done by using feature points matching. In feature extraction,
morphological features extract 6 features points. Moreover, to
get more precise recognition result, Principal Component
Analysis (PCA) method extract 30 features points. In
classification, targets classes are five types of WBCs such as
monocytes, lymphocyte, eosinophils, neutrophils and basophils.
The proposed system has been tested successfully to a dataset of
five images for each type and achieves good classification
accuracy for WBC images. The system is implemented with
MATLAB programming language.
Index Terms — Color Conversion, Feature points matching,
Morphological operation, Principle Component Analysis,
Resizing
I. INTRODUCTION
Over the last 15 years, several research groups have
focused on the development of computerized systems that can
analyze different types of medical images and extract useful
information for the medical professional [1]. Most of the
proposed methods use images acquired during a diagnostic
procedure. The major objectives of image analysis in
biomedical instrumentation engineering are to gather the
information, screening or investigating, to diagnose, therapy
and control, monitoring and evaluation. It is important always
to bear in mind that the main purpose of biomedical imaging
and image analysis is to provide a certain benefit to the
subject or patient. Cell classification has widespread
interested especially for clinics and laboratories. For
example, patient’s blood cells counting is used to extract
information about other cells that are not normally presented
in peripheral blood but may be released in certain disease
processes by the hematologist [2]. One of the great challenges
May Thida Aung , Department of Information Technology, Mandalay
Technological University (e-mail: maythidaaung@gamil.com). Mandalay,
Republic of Myanmar, +95-092210794.
Nang Aye Aye Htwe, Department of Information Technology, Mandalay
Technological University (e-mail: htwe.aye@gamil.com). Mandalay, Republic of
Myanmar,+95-095661208.
to engineer especially biomedical engineer is to transform this
human practical task into computer based which the system is
comparable to human performance or better. Thus, the system
must be stable and able to handle uncertain. Up to now,
automatic cells classification systems cannot meet the
complexity of real clinical demands [3].
This paper is organized as follows: related works of the
system are described in section two. In section three,
background theory is explained. In section four, system
design and in section five, implementation results are
presented. Finally, in section six, the paper has been
concluded.
II. RELATED WORKS
V. A. Kovalev, A. Y. Grigoriev, H. Ahn, Robust
proposed recognition of White Blood Cell Images, 1996 [4].
H. T. Madhloom, S. A. Kareem , H. Ariffin, A. A. Zaidan, H.
O.Alanazi, B. B. Zaidan proposed An Automated White
Blood Cell Nucleus Localization and Segmentation using
Image Arithmetic and automated Threshold, 2010[5].
Whole blood is comprised of plasma (the liquid part) and
the formed elements (red blood cells, white blood cells, and
platelets). The process by which all formed elements of the
blood are produced (hematopoiesis) occurs mostly in the bone
marrow, where cells mature from a primitive stem cell.
Billions of red, white blood cells and platelets are produced
per kilogram of body weight daily. Factors important in
regulating blood cell production include the environment of
the bone marrow, interactions among cells, and secreted
chemicals called growth factors. Patients with Waldenstrom’s
Macroglobulinemia experience a reduced capacity to produce
several types of whole blood cells in the bone marrow
(myelosuppression) because the overproduction of immature
WM cells suppresses production of the other blood cell types.
Chemotherapeutic agents which destroy fast growing cells of
the body also contribute to lowered blood cell production. A
Complete Blood Count (CBC) is a measurement of the blood
cells in a specific volume of blood.
III. BACKGROUND THEORY
In this system, there are four main parts: image acquisition,
image preprocessing, feature extraction and recognition.
A. Image Acquisition
This system will be tested using images obtained from the
ARUP Laboratory, University of Utah. The input images are
256 × 256 resolution and types are jpg format.
1
B. Image Preprocessing
Preprocessing step includes resizing, changing color,
opening and hole filling. Preprocessing step is satisfied for
features extraction.
1) Resizing: All of images are resized into 200×150 pixels.
All images are resized to reduce time consuming and money.
2) Changing Color: This function converts the true color
image RGB to the grayscale intensity image. This function is
used to discriminate between nuclear and non-nuclear pixels.
This function converts RGB values to grayscale values by
forming a weighted sum of the R, G, and B components:
 Image Acquisition
 Preprocessing
(i)RGB to Gray
(ii)Gray to Binary
(iii)Opening
(iv)Hole Filling
 Features Extraction
(i)PCA
(ii)Morphological operation
Start
L= 0.2989 * R + 0.5870 * G + 0.1140 * B
The color is changed into grayscale 0 (black) to 255 (white)
gray level transitions for each pixel. Then, the grayscale
image converts into binary image. The output image replaces
all pixels in the input image with luminance greater than level
with the value 1 (white) and replaces all other pixels with the
value 0 (black). Level can be specified in the range [0, 1].
Nuclear pixels leave as black pixels on a white background.
Image Acquisition
Resizing
3) Opening: This process removes from a binary image all
connected components (objects) that have fewer than
specified number of pixels, producing another binary image.
Color
Conversion,Opening,Filling
4) Hole Filling: This process fills holes in the binary image.A
hole may be defined as a background region surrounded by a
connected border of foreground pixels. Dark spots could be
results of reflections. The objective is to eliminate reflections
by hole filling.
Features Extraction(PCA,
Morphological operation)
Combine Features
Features DB
C. Features Extraction
In the features extraction step, PCA and morphological
features are used for 36 features. PCA is used for extracting
30 features and dimensional reduction. Morphological feature
extracted 6 features. Totally, 36 features will be extracted
from two features extraction methods.
1) Principle Component Analysis (PCA)
PCA method is one of the most successful techniques in
image compression. The purpose of PCA is to reduce the
large dimensionality of the data space to the smaller intrinsic
dimensionality of feature space which is needed to describe
the data economically. PCA computes the basis of a space
which is represented by its training vectors [6].
D. Recognition using Feature Points Matching
Feature points matching are the final stage of WBC
recognition process. In this system, 36 features points include
for each training and testing WBC image. In matching step,
the system is done by matching the feature point values of
trained images and tested images. The WBC feature values
from the input test image is same with the feature values from
the feature database, the test image is recognized. If not so, the
test image is not recognized.
IV. SYSTEM DESIGN OF PROPOSED SYSTEM
The entire system can be divided into three parts:
Training
End
Figure 1. System Design of Training Process
The input images are WBC images and types are jpg
format. Input images change to gray scale and binary image.
Features extraction used PCA and Morphological
Operation for 36 features extracted. This system used 25
images in training database.
The entire system can be divided into four parts:
 Image Acquisition
 Preprocessing
(i)RGB to Gray
(ii)Gray to Binary
(iii)Opening
(iv)Hole Filling
 Features Extraction
(i)PCA
(ii)Morphological operation
 Recognition
2
Start
Image Acquisition
Resizing
Color Conversion,Opening,Filling
Figure 4. Main Page of Proposed System
The user must choose “Load Image’ button to load the
input image. And then, user chooses the input WBC image in
Figure 5.
Features Extraction(PCA,
Morphological operation)
Combine Features
Recognized or Not Recognized(feature
points matching)
Features DB
End
Figure 2. System Design of Recognition Process
This system is the same working process of Figure 1. In
recognition, test image must be matched trained image and
then the system produce type of WBC images.
V. IMPLEMENTATION OF PROPOSED SYSTEM
The home page is shown in Figure 3. User can chose
‘MAIN MENU’ button to start the process. Then it is needed
to choose the “Main menu” button. And then Load image
button is chosen as shown in Figure 4.
Figure 3. Home page of the proposed system
Figure 5. Input Image of WBC
After loading the image, the user must choose the
“preprocessing” button. The image is resized into 200×150
pixels. This system includes RGB to gray and gray to binary
conversion.
Figure 6. Pre-processing of Proposed System
3
After pre-processing, the user selects ‘Features
Extraction button. Figure 7 shows Extracted Feature Points.
Figure 10. Recognition Result of Unknown Image
Figure 7. Extracted Feature Points of Proposed System
After features extraction, the user choose the type of WBC
image radio button and then choose “Train’’ button for
successfully trained as shown in Figure 8.
VI. CONCLUSIONS AND FURTHER EXTENSION
By using PCA method, the system can reduce the dimension
size of an image. It can be such that the proposed WBC
recognition system is reliable and provides good accuracy. In
this paper, the recognition accuracy of the system is 100% for
known images. The proposed WBC recognition system is
reliable and provides good accuracy. This system is always
true for trained images. As further study, the hematologists
can define kind of diseases by counting numbers of each type
of WBC.
ACKNOWLEDGMENT
The author would like to give her deepest thanks to her
parents for their unconditional love, never-ceasing support,
encouragement and their artful ways of breeding her
independence and free thinking. The author also thanks to all
her colleagues and friends for their motivation and
encouragement throughout her thesis. The author wishes to
express her gratitude to all persons who helped directly or
indirectly towards the successful completion of this thesis.
REFERENCES
Figure 8.Train WBC Image
[1] Comaniciu, D., Meer, P., Foran, D.: Image Guided Decision Support
System for Pathology, Machine Vision and Applications, Vol. 11,
No.4 (2000) 213-224.
[2] Huang, L.K, Wang, m. J.: Image thresholding by Minimizing the
Measures of Fuzziness, Pattern Recognition, Vol.28 (1995) 1:41-45.
[3] Korea Medical Publisher: Illustrated hematology Book.
KoreaPublishing (1995).
[4] V. A. Kovalev, A. Y. Grigoriev, H. Ahn, Robust Recognition of White
Blood Cell Images, 1996.
[5] H. T. Madhloom, S. A. Kareem , H. Ariffin, A. A. Zaidan, H. O.
Alanazi, B. B. Zaidan An Automated White Blood Cell Nucleus
Localization and Segmentation using Image Arithmetic and
Automated Threshold, 2010.
[6] M.A Turk and A.P. Pentland, “Face Recognition using ., Eigenfaces”,
Principle Component Analysis, IEEE conf. on Computer vision and
Train image and test image are need to features extraction.
Each WBC image of 36 features is matched by using feature
points matching. After feature extraction for the test image,
the user must choose “Recognize’ button to decide that the
system recognize the test image or not. Figure 9 shows
recognition result of known image. Figure 10 shows
recognition result of unknown image.
Figure 9. Recognition Result of Known Image
4
Download