International Journal of Pure and Applied Mathematics
Volume 118 No. 18 2018, 3483-3497
ISSN: 1311-8080 (printed version); ISSN: 1314-3395 (on-line version)
url: http://www.ijpam.eu
Special Issue
ijpam.eu
Automatic Brain Tumor Segmentation Using FPGA
Platform
1
H.M. William Thomas, 2S.C. Prasanna Kumar and 3D. Jayadevappa
1
Department of ECE,
BITM,
Ballari, India.
williamthomas.hm@gmail.com
2
Department of E&IE,
RVCE,
Bengaluru, India.
prasannakumar@rvce.edu.in
3
Department of E&IE,
JSSATE,
Bengaluru, India.
devappa22@gmail.com
Abstract
This paper is an attempt to develop a brain tumor segmentation using
FPGA. The Xilinx platform studio based EDK code is developed on the
FPGA Spartan 3E and the edge detection techniques are used to find the
brain tumor on the MRI images. Matlab based program is used to convert
the image to the bit stream array which would be used as the header file on
the Xilinx platform studio. This bit stream is taken and the calculation for
the tumour segmentation using the level set technique is developed and the
output is sent back to matlab. The experimental results proved that, the
time taken by the FPGA to segment the brain tumor is less as compared to
MATLAB or C++ environment.
Index Terms:Segmentation, MRI, FPGA, fuzzy logic and brain tumor.
3483
International Journal of Pure and Applied Mathematics Special Issue
1. Introduction
Medical image segmentation is always the challenging task of any problem in
computer guided medical procedures in hospitals. Medical image segmentation
is the process of partitioning an object of intersest by labeling each pixel from
the medical image database to identify an anatomical structure. The results
obtained from this process process leads to a wide range of applications in
medical field as well as in visualization of the anatomical structre in 2-D or 3-D
forms. Partioining an anatomical structure from medical datasets gives better
clarity for the identification of a specific boundary or region in a medical image
corresponding to the desired structure. To understand the medical image
segmentation process, an enormous amount of information available in the
literature concentrating on various issues of medical image segmentation
techniques [1], [2], [3] and medical image segmentation has received tremendous
amount of response from the hospitals and the research due to its various
practical applications of segmentation results.
Algorithms for used for medical image segmentation can applied identify
various pathological changes in brain and especially to recognize tumors and
lesions in the brain area. This can be performed by first separating the
recognizable neuro-anatomical structures. Further, these algorithms can also be
used to determine specific disease of human brain. In order to perform this,
identifying the genesis of the disease is quite important to decide the cause and
to workout the options for treatment depending on the affected anatomical
structre of the brain where the pathologis lies. Applications of medical image
segmentation depends on the specific disease, imaging techniques and the other
factors. As we aware that, there is single technique of medical image
segmentation that can cater the medical community who can accept the end
results for every medical image. At present scenario, there is a need for the
solution where the richer information of anatomical structre can be obtained as
compared to the exising systems using medical images automatically with lesser
manual intervention. Medical segmentation is also used for constructing an
anatomical atlases, determining shapes of various tissues structures and tracking
pathological changes in the anatomical structures over the period of time. The
main objective of medical image segmentation is to provide the segmented
image that allows the doctor or clinician for better visualization qualitatively for
shapes and relative positions of complex anatomical structre of the human brain
internally and also to measure accurately their volumes quantitatively.
FPGAs are capably used as a touch of cutting Edge imaging applications picture
sifting pleasing imaging picture weight remote correspondence downside the
vast majority of frameworks is that they utilize a bizarre state for coding.Target
incite to the utilization of Xilinx System Generator (XSG), contraption with a
sporadic state graphical interface under the Matlab. Simulink based squares
which makes it simple to manage concerning other programming for rigging
depiction The unmistakable applications where picture sifting operations related
3484
International Journal of Pure and Applied Mathematics Special Issue
are unsettling influence launch, upgrading edges and structures, blurring etc.This
paper presents framework of sifting pictures for edge territory utilizing System
Generator, which is an improvement of Simulink and includes models called
"XILINX BLOCKS" which are mapped into structures, parts, signs and ports.
2. Background
Most of the image processing applications such as medical image segmentation
algorithms [4], [5] uses FPGAs to segment the desired object with a faster rate.
These systems are are typically programmed with Hardware Description
Languages (HDL) and microprocessor-based DSP design methodologies [6].
The hardware configuaration is designed in such a way that, all the required
functions of image processing applications can be transferred to FPGA that
allows faster processing of the task. It is not necessary to split the individual
instructions for fetching, decoding which is generally used in typical processing
unit of a computer system [7]. However, in most of the low level image
processing applications, the parallelism inherent was exploited to its full extent
for partitioning into subsystems and all of which can run concurrently with each
other. Hence, the objective of this paper will be to propose the medical image
segmentation algorithm and implementing it for segmenting the brain tumor
accurately. Then, for faster segmentation, the program is transferred to hardware
so that the processing speed of the components of the system is bounded within a
specified timing constraint considered typical of such systems. The outcome of
this proposed work is that, implementing image segmentation algorithms on
FPGA hardware reduces the segmkentation time drastically and complex
algorithms can be implemented without debugging and much varification.
Hence, FPGAs based medical image segmentation techniques are the preferred
choice for the implementation of such algorithms [8].
The technique of edge based segmentation by applying wavelet transform in
Synthetic Aperture Radar (SAR) images was introduced by Marivi Tello Alonso
et al. [9]. This technique adapts a novel technique for the detection of edges
automatically. Mohammad Saleh Miri et al. [10] and S.Allin Christe et al. [11]
were proposed the techniques based on curvelet transform and then transformed
the algorithm for the efficient FPGA implementation of segmentation algorithms
for brain MR images for tumour characterization using Xilinx System Generator.
This technique will be able to reduce the amount of time required for the
segmentation process considerable. Alba M. Sanchez et al. [12] came out with an
architecture for segmenting filtered images using Xilinx system generator to
improve the segmentation accuracy as compared to the Aba m et al.. Another
technique based on reconfigurable architecture using C-based hardware
descriptive languages was proposed by Daggu Venkateshwar Rao et al.[13]
which was applied for implementation and evaluation of image segmentation
aigorithms. The improved version of the FPGA used for the MRI brain
segmentation was designed and devleped by Mohd Fauzi Bin Othman et al.[14]
which proved the better specificity as compared to the Daggu Venkateshvar Rao
3485
International Journal of Pure and Applied Mathematics Special Issue
method. Similarly, the FPGA implementation of image realization of beamlet
transform edge detection algorithm using FPGA approach proposed by Selvathi
and Dharani [15].
The above background of current methods, a new automatic brain tumor
segmentation based on FPGA flatform is proposed. The proposed method uses
medical image processing technique to segment the tumor from MR images to
recalculate pixel values.
3. Generic Structre of FPGA
FPGA is an electronic technology implemented with VLSI digital circuits by
means of the software based reconfiguration of a large integrated array,
consisting of similar configurable digital blocks may have variety of elementary
logic gates, look-up tables and flip-flops. The routing system of FPGA [16]
states each application using wide network of horizontal and vertical channels
that may be interconnected in any possible way by transistorized interconnecting
matrices. FPGA is a field reprogrammable leading hardware and software
technologies. The generalized structre of FPGA is shown in figure 1. The whole
execution of corner identification design utilizing Simulink and Xilinx squares
experiences 3 stages, image pre-processing blocks, XSG used for edge detection,
image post-processing blocks. Image pre-processing block sets and the circuit
based layout used for picture pre-get ready squares utilized here are looked into
underneath. Input pictures which could shade or grayscale are given as
obligation to the archive piece. A shading space change square changes over
RGB to grayscalepicture and this data in 2-D changed over to 1-D to get ready of
the required part.
Design change piece sets yield flag to chart based information and
accommodated un bolster square which change over this bundling to scalar
illustrations yield at a higher testing rate [17]. The Xilinx structure generator
instrument is another application in picture planning, and offers a model based
arrangement for taking care of.
Pieces orchestrate the channels and commonly sponsorships with Matlab codes
through client adaptable squares. It also offers straight forwardness of
masterminding with GUI condition.
This instrument bolster programming increase, however particularly it makes
enter documents for use in all Xilinx FPGAs, with the parallelism, vivacious,
fast and modified area diminishing. These parts are stray pieces constantly
picture dealing with.The structure building utilized as a bit of this venture can be
utilized for all Xilinx FPGA pack with legitimate client layout in system
generator square and could be reached out to tireless picture arranging. In
addition to driving any type of high resolution display devices, FPGAs are
capable for image enhancement, segmentation, pattern recognition and image
compression etc.
3486
International Journal of Pure and Applied Mathematics Special Issue
Figure 1: Generalized Structure of FPGA
4. Enhancement and Tumor Boundary
Detection
MR images of human brain contains noise which is very common during the
scanning and trnasmiision of image data through various communication
channels. Generally, preprocessing techniques are used for the removal of such
image noises effectively. For instant, in wavelet transform, wavelet coefficients
are altered or varied in order to remove the noisy points and then an inverse
wavelet transform is applied to reconstruct the image. The smooth and linear
variations Daubechies-4 can be applied for as preprocessing which can give
better resolution for slow changing properties of images [18]. This type of
wavelet are belongs to the orthogonal family wavelets resulting to discrete
wavelet transform domain. These are characterized by a maximal number of
vanishing moments for some given support. The practical applications of
wavelets can be used in fast wavelet ransform domain, therefore the
compational complexity can be reduced.
Figure 2: Two Level Decomposition of the MR Image using Wavelet
Transform
3487
International Journal of Pure and Applied Mathematics Special Issue
In this paper, the preprocessing is performed using DAUB-4 wavelet
approximation coefficients for better resolution of the MR brain images. These
coefficients were used as feature vector for classification. The reason behind
using wavelet transform for image preprocessing is that, MR images of barin are
variable in nature and the tissue structure of the brain is overlap in nature. Such
images required efficient transformation technique.
Figure 3: Result of 2-Level Daubechies 4 on Brain MR Image
The demonatration DAUB-4 wavelet transform which was implemented using
the wavelet toolbox for barin MR image shown in figure 3. The MR
decomposition is illustrated by considering an input MR T1-weighted image.
This image is decomposed into four levels by the DAUB4 wavelet
approximation. Top most band shown in figure 3 is a low pass image. For LP
directions, n is used. At each successive level the number of directional sub-
bands is 4, 8 and 16 respectively. This figure also illustred the process of an
image being decomposed into detailed components having different
decomposition of wavelet transform.
An efficient segmentation of brain tumor boundary is one of the most
challenging task in medical image analysis and to achive this, many algorithms
have been developed and implemented which is established in the available
literature. A boundary is the connected points of sharp change in a region of an
image where pixel locations have abrupt intensity variations i.e. a discontinuity
in gray level values. In other words, the boundary consists of connected edges
between an object and the background.
The proposed work brain tumor segmentation method uses level set [19] for the
detection of the tumor boundary from MR image. The level set technique is
based on the active contour models which offers energy minimization of the
contour. This is a long-range attraction generated by the object boundary and
acting on the evolving contour for solving the segmentation problem for MR
images. This method is very common technique which can be used the image
3488
International Journal of Pure and Applied Mathematics Special Issue
which is blur and has weak edges with intensity in-homogeneity. As we know
that, the Level set is an important group of deformable models.
Level set which is used to segment the brain tumor from MR image is
formulated using active contours was first introduced by Osher and Sethian [20].
The segmentation process using level set uses energy minimization of active
contour [21], [22] surfaces in one higher dimension. In level set, the evolving a
surface is φ instead of contour or curve C. In level set method the contour C ( s )
is represented by φ ( x, y, t ) as an an implicit function and the front is then defined
implicitly as the zero level set φ = 0.
Given an initial φ at t = 0, it would be possible to know φ at any time t with the
∂φ
motion equation .
∂t
Figure 4: Representation of Level Set [17]
From the chain rule,
∂φ ( x ( t ) , t )
=0 (1)
∂t
∂φ ∂x ( t ) ∂φ t
+ =
0 (2)
∂x ( t ) ∂t t t
∂φ
xt + φt =
0 (3)
∂x ( t )
∂φ
Here, = ∇φ also, the speed xt is given by a force F normal to the surface, then
∂x
∇φ
xt = F ( x ( t ) ) n , where n = , now equation (3) can be rewritten as,
∇φ
φt + ∇φ xt = 0 (4)
φt + ∇φ Fn = 0 (5)
3489
International Journal of Pure and Applied Mathematics Special Issue
∇φ
φt + F ∇φ =0 (6)
∇φ
φt + F ∇φ =
0 (7)
Equation (7) is used for the contour evolution of the level set function φ which
is known as level set equation. The parameter F is called the speed function,
the function F depends on the image data and the level set function φ for
effective segmentation.
5. Implementation for the Proposed
Methodology
Multiresolution image analysis has attracted a considerable amount of attention
in medical image processing. Due to the efficiency of multiresolution data
representation, wavelets are considered for medical image segmentation in
conjunction with the level sets. In this method a novel frame work of level set
technique is applied to detect the tumour boundary accurately. In wavelet
transform, the directional information is preserved in each sub-band and is
captured by computing its energy. This energy is capable of enhancing weak and
complex boundaries in details. Further, due its directional image expansion
property, smoothness along the contours can be easily achieved. Therefore,
wavelet transforms are well suited for the further enhancement of the object
boundary.
In the proposed algorithm level set is used in conjunction with the wavelet
transform for the segmentation of tumor boundary. The reason using level set in
this method is that, Snake models cannot handle topological changes and it has a
tendency to produce degenerate contours or self-intersections [24], [25]. The
treatment of the proposed algorithm is divided into two stages. The first stage is
used to compute the energy of the wavelet transform decomposed MR image.
Figure 5: Implementaion of the Propoed Methodology with FPGA
3490
International Journal of Pure and Applied Mathematics Special Issue
The second stage corresponds to integration of this energy using level sets [23].
FPGA execution incorporates building an inserted structure required for the
particular applications here we build up an implanted framework on little scale
affect processor. In this paper an Altera Cyclone chip is incorporated as the
FPGA accelerator and is implemented as a System On a Programmable Chip
(SOPC). However, some of the hardware components are difficult to be re-
designed and transferred on a FPGA board from scratch, when they are already a
functional part of a computer-based system. The following block diagram shown
in figure 5 illustrates the proposed methodology.
The FPGA with Virtex series is desirable one for the proposed methodto
implement the medical image segmentation algorithm. The advantage of
incorporating hardware implementation is that, The segmentation process time
can be saved because of the high parallelism in the algorithm. This transfer will
enable to use the medical image segmentation algorithms for the real time
applications. The following table shows the specifications of the VERTEX 5
FPGA kit used to implement the proposed methodology. A bit stream is loaded
into the device through special configuration pins.
Table I: Vertex 5 FPGA Specifications
Descriptions
Device used XC4VSX25-10FF668
Logic cells 23,040
No of slices 10,240
Extreme DSP slices 128
Speed of RAM 3.9 GHZ
Block RAM 0.99GB
Dedicated multiplier 104
DCMs 4
Max select I/O 320
To implement the proposed segmentation algorithm on FPGA, Xilinx blockset
architecture is derived and subsequently implemented on it. This architecture can
be applied with the suitable image size and for different image sizes, the
parameters has to be modified in the architecture for the desired results. In figure
6, The design of the component’s architecture is illustrated. In the Xilinx M-
Code, logic for the component is encoded, which act as a container that can be
used for Simulink blockset supplied with Mat Lab functions with Simulink [26].
Therefore, this block excutes the MAT Lab functions for the calculations of the
output during the simulation process.
In this paper, the hardware used for embedding the components belongs to the
Xilinx Virtex-5 family with a PCI connector for reliable interface to a PC. This
board is an ideal one for implementing the medical image segmentation
algorithms because of its support in high-speed I/O and also designed for
filtering and other image processing applications.
3491
International Journal of Pure and Applied Mathematics Special Issue
Figure 6: Design of the Component’s Architecture
6. Resuts and Disussion
The proposed methodology was implemented using FGGA platform. In the first
stage, the given input is preproceed by applying wavelet transform. In the second
step, tumor segmentation is carriedout using level sets. Then the segmented
tumor boundary is converted into samples using MAT lab codes and
implemented using FPGA kit. The proposed methodology uses two brain MR
images for testing and evaluation process. System model of the component has
been compiled successfully in the Simulink environment. The hardware co-
simulation block was generated without any errors and the processing speed was
obtained using the synthesis and ISE implementation tool. To fulfill the objective
of our research, the processing speed of the component on the FPGA is
compared with its corresp onding speed in MATLAB and C++. This will enable
us to ensure that the timing constraint imposed on segmentation for the proposed
system is met.
3492
International Journal of Pure and Applied Mathematics Special Issue
(a) (b) (c)
Figure 7: Segmentation Results for the MR Image 1.
(a) Original Image (b) Enhanced by Wavelet Transform (C) Extracted
Boundaries of the Enhanced Image
(a) (b) (c)
Figure 8: Segmentation Results for the MR Image 2. (a) Original Image (b)
Enhanced by Wavelet Transform (C) Extracted Boundaries of the Enhanced
Image
Two images of different sizes (large, medium and small) were selected and run
using the same component on three different platforms namely MATLAB, C++
and FPGA kit.
7. Conclusion
The Xilinx platform studio based EDK code is developed on the FPGA Vertex 5
series and the tumor boundary detection techniques were used to find the brain
tumor on the MRI images. Matlab based program is used to convert the image to
the bit stream array which would be used as the header file on the Xilinx
platform studio. This bit stream is taken and the calculation for the tumour
detection using the level set based technique is developed and the output is sent
back to matlab. The tumour detected image is given to the matlab for displaying
the final results. The results of the proposed technique proves that, there is a
signican improvements in the segmentation accuracy and reduction of
computational time as compared the existing medical image segmentation
algorithms.
References
[1] Clarke L., Velthuizen R., Camacho M., Heine J., Vaydianathan
M., Hall L., Thatcher R., Silbiger M., MRI Segmentation: Methods
3493
International Journal of Pure and Applied Mathematics Special Issue
and Applications, Magnetic Resonance Imaging 13(3) (1995),
343–368.
[2] Pham D.L., Xu C., Prince J.L., Current methods in Medical Image
Segmentation, Annual Review of Biomedical Engineering 2
(2000), 315–337.
[3] Daniel J., Withey Z.J.K., A Review of Medical Image
Segmentation: Methods and Available Software, International
Journal of Bioelectromagnetism 10(3) (2008), 125-148.
[4] Leeser M., Coric S., Miller E., Yu H., Trepanier M., Parallel-Beam
backprojection: An FPGA implementation optimized for medical
imaging, Journal of VLSI Signal Process. Syst. Signal, Image,
Video Technol (2005), 295-311.
[5] Maslennikow O., Sergiyenko A., Mapping DSP Algorithms into
FPGA, Proceedings of the International Symposium on Parallel
Computing in Electrical Engineering, IEEE Xplore Press,
Bialystok (2006), 208-213.
[6] Chang C., Design and Applications of a Reconfigurable
Computing System for High Performance Digital Signal
Processing, Ph.D. Thesis, University of California, Berkeley,
(2005).
[7] Rao D.V., Patil S., Babu N.A., Muthukumar V., Implementation
and Evaluation of Image Processing Algorithms on
Reconfigurable Architecture using C-based Hardware
Description Languages, International Journal of Theoretical and
Applied Computer Sciences 1(1) (2006), 9-34.
[8] Yahia Said, Taoufik Saidani, Fethi Smach, Mohamed Atri and
Hichem Snoussi, Embedded Real-Time Video Processing
System on FPGA, ICISP LNCS, Springer-Verlag Berlin
Heidelberg (2012), 85–92.
[9] Marivi Tello Alonso, Carlos L6pez-Martinez, Jordi Mallorqui,
Philippe Salembier, Edge Enhancement Aigorithm Based on the
Wavelet Transform for Automatie Edge Deteetion in SAR
Images, IEEE Transactions on Geoseienee and Remote Sensing
49 (I) (2011).
[10] Mohammad Saleh Miri, Ali Mahloojifar, Retinal Image Analysis
Using Curvelet Transform and Multistrueture Elements
Morphology by Reeonstruetion, IEEE Transactions on
Biomedieal Engineering 58 (5) (2011).
[11] Allin Christe S., Vignesh M., Kandaswamy A., AnEffieient FPGA
Implementation of MRI Image Filtering and Tumour
3494
International Journal of Pure and Applied Mathematics Special Issue
Charaeterization using Xilinx System Generator, International
Journal of VLSI design & Communieation Systems 2(4) (2011).
[12] Alba M., Sanehez G., Rieardo Alvarez G., Sully Sanehez G,
Arehiteeture for filtering Ximages using Xilinx System Generator,
International Journal of Mathematieal Models and Methods in
Applied Sciences 1(5) (2007).
[13] Daggu Venkateshwar Rao, Shruti Patil, Naveen Anne Babu, V
Muthukumar, Implementation and Evaluation of Image
Processing Aigorithms on Reeonfigurable Arehiteeture using C-
based Hardware Deseriptive Languages, International Journal of
Theoretieal and Applied Computer Seiences 1(1) (2006), 9-34.
[14] Mohd Fauzi Bin Othman, Norarmalina Abdullah, Nur Aizudin Bin
Ahmad Rusli, An Overview of MRI Brain Classification using
FPGA Implementation, IEEE Symposium on Industrial
Eleetronies & Applications (2010).
[15] Selvathi D., Dharani, Realization of Beamlet Transform Edge
Detection Algorithm using FPGA, International Conference on
Signal Processing, Image Processing and Pattern Recognition
(2013).
[16] Sami Hasan, Said Boussakta, Alex Yakovlev, FPGA-Based
Architecture for a Generalized Parallel 2-D MRI Filtering
Algorithm, American J. of Engineering and Applied Sciences 4
(4) (2011), 566-575.
[17] Hemalatha, Santhiyakumari, Suresh S., Implementation of
Medical Image Segmentation using Virtex FPGA kit, SPACES
(2015), 358-362.
[18] Shan Z., Aviyente S., Image denoising based on the wavelet
co-occurrence matrix, Proc. IEEE ICASSP, Philadelphia, USA
(2005), 645–648.
[19] Jayadevappa D., Srinivas Kumar S., Murty D.S., Medical Image
Segmentation Algorithms using Deformable Models: a Review,
IETE Technical Review 28(3) (2011).
[20] Osher S., Sethian J.A., Fronts Propagating with Curvature
Dependent Speed:Algorithms based on Hamiton–Jacobi
Formulations, Journal of Computer Physics 79(1) (1988) 12- 49.
[21] Sethian J.A., Level Set Methods and Fast Marching Methods:
Evolving Interfaces in Geometry, Fluid Mechanics, Computer
Vision, and Materials Science, Second edition, Monograph on
Applied and computationa Mathematics, Cambridge University
Press (1999).
3495
International Journal of Pure and Applied Mathematics Special Issue
[22] Jayadevappa D., Srinivas Kumar S., Murthy D.S., A New
Deformable model based on Level sets for medical Image
segmentation, International Journal of Computer Science 36(3)
(2009), 199-207.
[23] Hai Min, Xiao-Feng Wang, De-Shuang Huang, Wei Jia, A novel
dual minimization based level set method for image
segmentation, Neurocomputing 214 (2016), 910-926.
[24] Sanping Zhou, Jinjun Wang, Mengmng Zhang, Quing, Cue,
Yihong Gong, Correntropy-based level set method for medical
image segmentation and bias correction, Neuro computing 234
(2017), 216-229.
[25] Sijie Niu, Qiang Chen, Luis De Sisternes, Zexuan Ji, Zeming
Zhou, Daniel L.R., Robust noise region-based active contour
model via local similarity factor for image segmentation,
Neurocomputing, Pattern Recognition 61 (2017), 104-119.
[26] Xilinx Inc., www.xilinx.com
3496
3497
3498
