You are on page 1of 4

2018 IEEE Conference on Systems, Process and Control (ICSPC 2018), 14–15 December 2018, Melaka, Malaysia

Automated Detection of Human RBC in


Diagnosing Sickle Cell Anemia with Laplacian of
Gaussian Filter
Ahmad Sabry Mohamad 1, Nur Syahirah Abdul Halim 1, Muhammad Noor Nordin 1, Roszymah Hamzah 2
Jameela Sathar 2
1
Universiti Kuala Lumpur British Malaysian Institute, Bt. 8 Jln Sg Pusu, 53100 Gombak, Selangor
2
Ampang Hospital, Pandan Mewah, 68000 Ampang, Selangor.

sabry@unikl.edu.my

Abstract— Image processing technique has been used to microscopic view. The percentage of sickle cell can be
produce automated detection of Sickle Cell Anemia. A obtained through haemoglobin analysis of electrophoresis
Laplacian of Gaussian (LoG) edge detection algorithm method. However, microscopic view depending on
computed to detect sickle cells diseases at the early stage in competency of laboratory practitioner. The human factor of
diagnosing patient. A MATLAB software able to demonstrate
experience level provides the LoG algorithms to pre-
the abnormalities of the human Red Blood Cell (RBC) in the
single shapes and quantities of sickle cells present in each detection of diagnosing the anemia. Thus, a detection system
dataset. A data samples of sickle cells from government Ampang using image processing technique has been introduced in this
Hospital has contributed this study to validate the results. research work as one of the computation tools that can help
the haematopathologist in reporting of sickle cell anemia.
Index Terms—Sickle Cell Anemia; Laplacian of Gaussian;
Human Red Blood Cell; MATLAB II. MATERIAL AND METHODS
I. INTRODUCTION Laplacian of Gaussian presents the methods consist on the
broad peak at an edge location. An edge was defined as the
Sickle cell anemia is an inherited blood disorder from the boundary that connect two different regions. The edge of an
abnormalities of human red blood cells (RBC). The effect of image filters out unnecessary details and data retained the
this disease causing pain and shortness of breath to the sickle important marks on the shape of an object. LoG operator was
cell traits. Early diagnosis of sickle cell anemia generally quite similar to the first process of filtration for biological
increases the chances for longer survival by focusing on vision system. Gaussian filter as second derivative operator
detecting symptomatic patients as early as possible. responds on the zero-crossing at an edge location [5]. Since
Examination in a public health program are invariably costly the information extracted by incorporation of Gaussian
and are ordinarily not chosen by the patients. Edge detection filtering with Laplacian to link to the edge point into
plays a crucial part in digital signal processing by applying segments. The acts as second order to derivative operator
streaming data and static data. A Laplacian of Gaussian which is a fusion of low and high pass filter with identical
(LoG) or known as Marr-Hildreth operator had been features in edge detection. Laplacian is applied after an image
proposed by Marr and Hildreth in 1980 [1]. The operator has has been smooth using Gaussian. Laplacian emphasize on the
operating at subsequent levels and depends thoroughly at gray level discontinuities in an image and focused in the
smooth low-level feature extraction. An algorithm using the region with slowly varying gray level. The Laplacian were
edge detections of Laplacian of Gaussian method has segregated into two mask form which are positive Laplacian
demonstrated to recognize a segment of abnormalities in and negative Laplacian. The Laplacian L(x,y) of an image as
human red blood cells (RBC) [2,3]. Digital image processing a second derivative operator is defined by [6]:
algorithms was performed by MATLAB programming
software to run the automatic detection[cite paper lama].
𝜕2𝐼 𝜕2 𝐼
Moreover, a normal human red blood cell has two 𝐿(𝑥, 𝑦) = + (1)
haemoglobin alpha and two haemoglobin beta chains. An 𝜕𝑥 2 𝜕𝑦 2
abnormal human RBC resulted in mutation of the human gene
Below are the 2-dimensional (2D) LoG function with
which were diagnosed in Malaysian government hospital
such as Haemoglobin SS, Haemoglobin SC and Haemoglobin Gaussian standard deviation σ:
G-Makassar [4]. In this work, MATLAB algorithm define as
𝑥2 +𝑦2
the edge position which located to zero crossings for second 1 𝑥 2 +𝑦 2 −
derivative of the image to detect the edges [5]. Hence, the
𝐿𝑜𝐺 (𝑥, 𝑦) = − (1 − )𝑒 2𝜎2 (2)
𝜋𝜎4 2𝜎2
conventional approach is used to detect sickle cells by
haemoglobin electrophoresis laboratory test. Clinical LoG method can be done in two way:
Haematology Referral Laboratory Ampang Hospital operates 1. Convolution of the image with Gaussian filter then
blood count by Haematology Analyser Sysmex XE5000 approximate the 2nd derivative with Laplacian
Europe GmbH device. The procedures are followed by operator.
978-1-5386-6327-1/18/$31.00 ©2018 IEEE 214
2018 IEEE Conference on Systems, Process and Control (ICSPC 2018), 14–15 December 2018, Melaka, Malaysia

2. Convolution of the image with a kernel calculated


know as LoG function. Start

Read the input


image
Gaussian filter is the most widely used smoothing filters in
image processing and has an important role for edge detection
and act as a low pass filter. Gaussian filter which also known Smooth the image
as Gaussian smoothing is used to blur an image and reduce with Gaussian filter

noise. In image processing, 2 dimensional (2D) Gaussian


function was used whereby one for each direction (x and y
axis). Gaussian filter takes the Gaussian function as the Examine 2nd
derivative
weighting function.
The Gaussian filter works by convolving the 2D Gaussian ∇2 𝐺𝜎 (𝑥, 𝑦)
distribution with the image, by removing high frequency
noise and smoothing the image as [7];
NO
𝐺𝜎 (𝑥, 𝑦) ∗ 𝐼 (𝑥, 𝑦) (3) Examine 2nd derivative If
2
∇ 𝐺𝜎 (𝑥, 𝑦) for the next 2
∇ 𝐺𝜎 (𝑥, 𝑦) = 0
Where: * represent convolution of image I(x, y) and 2D neighbor
Gaussian filter, G(x, y):
YES

𝑥2 +𝑦2 Mark pixel as edge


1 −
𝐺𝜎 (𝑥, 𝑦) = 2𝑒
2𝜎2 (4) pixel
2𝜋𝜎

End
The Laplacian operator is the applied to the convolution.
Since,
Figure 1: Flowchart for general algorithm of Laplacian operator
∇2 [𝐺𝜎 (𝑥, 𝑦) ∗ 𝐼 (𝑥, 𝑦)] = ∇2 𝐺𝜎 (𝑥, 𝑦) ∗ 𝐼(𝑥, 𝑦) (5)

Both operations can be combined into one known as LoG Sickle cells image dataset were obtained from Ampang
operator: Hospital. All the sickle cells images converted into RGB
format feature in true color of an image.
𝑥2 +𝑦2
1 𝑥 2 +𝑦 2 −
∇2 𝐺𝜎 (𝑥, 𝑦) = (1 − )𝑒 2𝜎2 (6)
𝜋𝜎 4 2𝜎 2

III. RESULT AND ANALYSIS


Figure 1 shows the flowchart for general algorithm
methodology for Laplacian operator. LoG operation represent The second derivative edges operators determined the zero
a second derivative of an image whereby LoG response to the crossed through convolution output. The images convolved
zero when the images made of one region of constant image which the focus measure needs to be computed from
intensity. The Log become positive on darker region and the algorithms. The result that has been obtained by using
negative on the lighter region. After applying Laplacian, the Laplacian of Gaussian (LoG) method in detecting and
gradient image will be obtained. The gradient will be calculating sickle cell from the dataset. There are total of 20
compared with threshold throughout the edges. When the images sample that has been used throughout this research
edges become concentrated, it will portray the second work. All the results from MATLAB has been tabulated in
derivative and find the zero crossings [8]. this paper.
Since Laplacian is a derivative operator, their sensitivity to The methodology of MATLAB coding describes the
noise make the operator produces as ‘double edges’. This computing of the algorithms. The computation coding by
reason create the Laplacian operator relies on the gaussian reading the filename image1.jpg to begin the image analysis
filter. as shown in Figure 2. All the data images were provided by
medical officer from Ampang Hospital. The conversion of
color image to grayscale means to convert RGB values
(24bit) to grayscale values (8bit). All the images prior to
convert into grayscale for further analysis in the image
processing technique with code: grayimage = rgb2gray(x).
[9].
Figure 3 shows the Gaussian filter at the low pass filter that
eliminates high frequency noise of the image. “Imgaussfilt”
is the combination of “fspecial and imfilter” in MATLAB.
Code: Blur=imgaussfilt(grayimage,4);

215
2018 IEEE Conference on Systems, Process and Control (ICSPC 2018), 14–15 December 2018, Melaka, Malaysia

For the image sharpening, Laplacian of Gaussian operator


has computed as shown in Figure 4. LoG is a second
derivative of an image whereby background was preserved,
and edges were enhanced. Code: y=edge(grayimage,'log');
Converting the data images into binary to validate a region of
interest (ROI) which were the human RBC cells in order by
processing in detecting sickle cells. Code: level= graythresh
(grayimage); image= imbinarize (grayimage,level); It fills all
the cells as white and background as black. Code: image
=imfill(~image,'holes');
The cells that connected to the border been removed to
avoid wrong analysis. Objects that is small that the limit been
removed as well. Code: image=imclearborder(image); Code:
image=bwareaopen(image,20);
All the object been labelled with number to make it easier
to detect and counting the human RBC sickle cells. Code:
[L,num] = bwlabel(image);
In order to accomplish automated detection, form factor has
been used with code: %Form Factor Formula: formfactor=
Figure 3: Peripheral blood film and applied Gaussian filter algorithm
Major/(Major-Minor). Counting of sickle cells with code:
formatSpec =There are %d rectangles\n; fprintf
(formatSpec,shpNum);
Figure 5 shows the final MATLAB computing by
counting the target sickle cells using code: formatSpec =
'There are %d rectangles \n';fprintf (formatSpec,shpNum);
disp(['Sickle = ' num2str(shpNum)]) ;h = msgbox(cat(1,
{'Cells counting succeeded.'}, {'Total no. of cells :'},
num2str (length(Boundary)), {'No. of Sickle Cell :'},
num2str(shpNum)), 'Success').

Figure 4: Laplacian of Gaussian filter map on the human RBC cell


morphology

Figure 2: Human RBC cell morphology in the peripheral blood film

(a) (b)

Figure 5: (a) MATLAB detection and (b) sickle cell detected with counted
number of cells

216
2018 IEEE Conference on Systems, Process and Control (ICSPC 2018), 14–15 December 2018, Melaka, Malaysia

Therefore, LoG method using image enhancement provide


a clear and fine edges of the targeted objects. MATLAB can
offer immediate analysis for shape detection condition. The
advantages of using LoG operators allow fewer arithmetic
operation and pre-computed in advance makes one
convolution will perform at run-time of the image. Overall,
the algorithm successfully detects sickle cell anemia in all the
image samples. The form factor is an automatic detection of
sickle cells embedded in kernel. It is a very efficient method
that can be applied for early detection of sickle cell anemia
[11].

IV. CONCLUSION

This paper demonstrated application of the edge detection


Figure 6: The waterfall graph shows the tabulation of human RBC cells technique in signal processing. Heamatopathologist pre-
diagnosis MATLAB computation with LoG provide
Figure 6 shows a waterfall graph to describe the incremental precaution and advance blood film report. Biomedical
changes between sickle cells of the row presented. The data engineering technology is contributing society with robust
also describes the human RBC accumulated by linear and faster diagnostic medical treatment.
increase from effect of column matrices. The Z-axis formed
the colour of the graph extracted from the workspace [10]. ACKNOWLEDGMENT
Table 1 shows all the form factor were less than 1.78 has
been set in the MATLAB coding. The value of major and The authors would like to thank Malaysian government
minor axis length has been obtained using regionprop syntax. Hospital of Clinical Haematology Referral Laboratory and
To evaluate the number of automatic detections, there were UNIKL BMI STRG2018 grant.
15 number of sickle cells detected at image1.jpg from 478
normal human RBCs. REFERENCES
Table 1
Form Factor of sickle cells in data set [1] Marr, David, and Ellen Hildreth. "Theory of edge detection." Proc. R.
Soc. Lond. B 207, no. 1167 (1980): 187-217.
Major Form Factor [2] Pan, X., Yang, D., Li, L., Liu, Z., Yang, H., Cao, Z., ... & Chen, Y.
Sickle
Major Axis Length Major (2018). Cell detection in pathology and microscopy images with multi-
Cells Minor Axis
Length - /(Major-Minor) scale fully convolutional neural networks. World Wide Web, 1-23
Label Length
Minor [3] Mohamad, A. S., Hamzah, R., Mokhtar, A. S., & Sathar, J. (2017,
No
Length September). Sickle cell disease verification via sobel edge algorithms
100 26.9016 10.1050 16.7966 1.60 for image processing. In Engineering Technology and
Technopreneurship (ICE2T), 2017 International Conference on (pp. 1-
129 19.8828 8.6199 11.2629 1.77 4). IEEE.
[4] Mohamad, A. S., Hamzah, R., Selvaratnam, V., Yegapan, S., & Sathar,
130 22.0705 9.3896 12.6809 1.74 J. (2018). Human hemoglobin G-Makassar variant masquerading as
sickle cell anemia. Hematology Reports, 10(3).
[5] Ding, K., Xiao, L., & Weng, G. (2017). Active contours driven by
140 14.9641 4.8334 10.1307 1.48
region-scalable fitting and optimized Laplacian of Gaussian energy for
image segmentation. Signal Processing, 134, 224-233.
149 13.0263 5.4768 7.5495 1.73 [6] Singh, A. K., & Dwivedi, Y. (2015). Application of MATLAB for Edge
Detection in Image Processing. Advanced Science Letters, 21(8), 2621-
220 12.9519 5.6092 7.3427 1.76 2625
[7] Sotak Jr, G. E., & Boyer, K. L. (1989). The Laplacian-of-Gaussian
309 21.2612 9.2012 12.0600 1.76 kernel: a formal analysis and design procedure for fast, accurate
convolution and full-frame output. Computer Vision, Graphics, and
Image Processing, 48(2), 147-189.
334 21.7146 8.0987 13.6159 1.59
[8] Atiquzzaman, M. (1992). Multiresolution Hough transform-an efficient
method of detecting patterns in images. IEEE Transactions on Pattern
340 12.8049 4.7600 8.0449 1.59 Analysis & Machine Intelligence, (11), 1090-1095.
[9] Hiatt, J., Mukwada, G., Barnes, M., Riis, H. L., Huynh, D., &
406 16.2003 6.3903 9.8100 1.65 Rowshanfarzad, P. (2018). MLC positioning verification for small
fields: a new investigation into automatic EPID-based verification
409 24.8937 10.1658 14.7279 1.69 methods. Australasian physical & engineering sciences in medicine, 1-
11
413 15.2259 5.0504 10.1755 1.50 [10] Corke, P. (2017). Images and Image Processing. In Robotics, Vision
and Control (pp. 359-411). Springer, Cham.
[11] Lee, L. K., Liew, S. C., & Thong, W. J. (2015). A review of image
429 25.0345 9.5873 15.4472 1.62 segmentation methodologies in medical image. In Advanced computer
and communication engineering technology (pp. 1069-1080). Springer,
435 16.0350 5.3021 10.7329 1.49 Cham.

470 19.9768 8.2612 11.7156 1.71

217

You might also like