333
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
AUTOMATIC KNEE SEGMENTATION USING EAGLE
ALGORITHM WITH MULTI STOCHASTIC OBJECTIVE
PROCESS
C. Rini
Research Scholar, Department of ECE, Kalasalingam Academy of Research and Education.
Krishnankoil, Virudhunagar (Dt), (India).
E-mail: rinidass2002@yahoo.co.in
ORCID: https://orcid.org/0000-0002-4413-5850
B. Perumal
Associate Professor, Department of ECE, Kalasalingam Academy of Research and Education.
Krishnankoil, Virudhunagar (Dt), (India).
E-mail: palanimet@gmail.com
ORCID: https://orcid.org/0000-0003-4408-9396
M. Pallikonda Rajasekaran
Professor, Department of ECE, Kalasalingam Academy of Research and Education.
Krishnankoil, Virudhunagar (Dt), (India).
E-mail: mpraja80@gmail.com
ORCID: https://orcid.org/0000-0001-6942-4512
V. Muneeswaran
Assistant Professor, Department of ECE, Kalasalingam Academy of Research and Education.
Krishnankoil, Virudhunagar (Dt), (India).
E-mail: munees.klu@gmail.com
ORCID: https://orcid.org/0000-0001-8061-8529
Recepción:
11/11/2019
Aceptación:
21/12/2020
Publicación:
30/11/2021
Citación sugerida:
Rini, C., Perumal, B., Rajasekaran, M. P., y Muneeswaran, V. (2021). Automatic knee segmentation
using eagle algorithm with multi stochastic objective process. 3C Tecnología. Glosas de innovación aplicadas
a la pyme, Edición Especial, (noviembre, 2021), 333-353. https://doi.org/10.17993/3ctecno.2021.
specialissue8.333-353
334
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
ABSTRACT
In our living world, Osteoarthritis (OA) is known to be foremost sicknesses which aect
the knee region especially aect hoarier people. The reason for OA in a person may be
due to ageing, malformed joints, rough cartilage, genetics eects or continuous repetitive
stress towards the joint. Magnetic Resonance Imaging (MRI) enacts a vigorous role in the
medical eld for detecting issues regarding bone structure, cartilage, and meniscus region
and tibia bone. Though it provides the details about the bone, it is not useful to detect clearly
about the faults in the bone due to many unfavorable conditions like poor segmentation,
broken pixels or some other natural issues including the shakes while clicking the image,
blurring, etc. Besides, manual calculations have some unexceptional error with partial
accuracy. Hence automated segmentation should be implemented for achieving perfection
in accuracy and the bone segmentation. In our work, we proposed eagle algorithm as
the segmentation method which provides an improved accuracy in contrast with other
traditional methods. The performance is measured by the metrics such as thickness, mean
and Standard Deviation (SD).
KEYWORDS
Pre-Processing, Contrast Enhancement, Stochastic Multi-Objective Process, Levy Walk
Random Process, Eagle Algorithm.
335
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
1. INTRODUCTION
Generally, 3-D Magnetic Resonance (MR) imaging delegates the therapies regarding
noninvasive, high-resolution as well as isotropic voxels. On behalf of quantitative studies
including dimensional (3-D) reconstruction and dimension of the cartilage, truthful
segmentation of the images is obligatory. The main objective of our work is to improve and
authenticate software for automated segmentation and thickness representing of articular
cartilage from three-dimensional (3-D) gradient-echo MR images of the knee. Inelegantly,
random noise occurs throughout the MR image processing and hence the quality of the
image is vitiated. Besides, the medical diagnostic tasks and noise damage many image
processing and investigation charges like registration, segmentation, super resolution and
visualization, etc. Hence the suppression of noise is essential to recover the image quality.
Many methods and algorithms are being introduced for attaining clear segmentations are
discussed in the next section.
Classically, joints are the regions where two bones meet in our body. Habitually, 360 joints
are there in our human body which include 86 skull joints, six throat joints, 66 thorax
joints and 76 in the spine and pelvis region. In case of limbs, 32 joints found in each upper
limb and 31 joints in each lower limb. Among that, knee joint is solitary and it connect
the leg and thigh of our human body. The knee joint is a hinge type synovial joint provide
the exion and extension of the leg. Knee joint is formed by articulations among patella
bone, femur bone (thigh bone) and tibia bone (leg bone). There are two articulations in the
knee joint. One of them is located between the femur and tibia bone and the latter is built
between the patella and femur (Peterfy et al., 2008); Eckstein, Burstein, & Link, 2006).
On the other hand, Condyle is dened as the round prominence at the end of a bone,
a part of the joint which articulate with another bone. It is also said to be one of the
markings or features of bones. In the femur bone of the knee joint, condyles are two types
of condyles namely medial condyle and lateral condyle. The joints are shaped by means of
the condyles of femur bone and tibia bone. The interior structure of knee joint comprise
of synovial cavity covering synovial uid, articular cartilage, meniscus region (semi-
lunar cartilages), cruciate ligaments and burse. Articular cartilage is a silky, white tissue
that wrap the ends of bones together with bones to form joints. In the mechanism of joints,
336
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
if the cartilage is ne, it is easier to move. It certies the bones to glide over each other with
a very slight friction.
Articular cartilage which covers patella bone, tibia bone and femur bone are frequently
injured by wounds or normal wear and tear. When the thickness of the cartilage is diminished,
the growth of Osteoarthritis (OA) begins. Osteoarthritis (OA) is the most common form of
arthritis, aect millions of people in worldwide. It occurs when the protective cartilage on
the ends of the bones wears down over time. Osteoarthritis symptoms can be eectively
managed, although the underlying process cannot be reversed. Maintenance of vigorous
weight and other treatments may leads for slow progression of the disease and help recover
pain and joint function. The cartilage joint is completely vanished whoever suering from
severe OA.
Figure 1. Knee joint for healthy and affected people.
Source: own elaboration.
During the cartilage joint reduction, the synovial uid drops its lubrication ability which
results in pain, inammation and eased movement of the joint. The bones get scratch
while gliding without uid and new tiny bones called spurs start to develop. Therefore, the
cartilage function (shock absorber) is constrained due to the stress on joint result in severe
inuence. The growth of Osteoarthritis (OA) may be stopped or diminished by introducing
a vigorous biomarkers to resolve the issue. An organization named Osteoarthritis Initiative
(OAI) has the main objective to detect and analyze the risk factors of Osteoarthritis which
337
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
are performing by several private health organizations. According to their survey, 50.2%
of the people among the age 65 and 74 years and 97.7% of people of age eighty-four and
above, are being aected by this disease.
Body mass Index (BMI) should be maintained for reduction of Osteoarthritis (OA). People
with high BMI rage falls under the respected disease and leads to give heavy burden at the
joint. The normal value of BMI range for both male and female is 18.5 - 25. People having
BMI beyond this range is considered to be overweight and they have the chances to suer
from Osteoarthritis (OA). It was 51.36% for the people with a BMI below 25 and 100%
for the people with a BMI above 40. The cartilage damage in the joints of knee can be
observed by means of Magnetic Resonance Images (MRI). MRI scan is used to scrutinize
the in-vivo and in-vitro structures of the human body meanwhile it is non-invasive and has
great resolution soft tissue contrast. Sagittal, coronal and axial planes are three dierent
planes where MRI scan the joints. The knee joint MRI images without Osteoarthritis
(OA) as publicized in Figure 2(a). At the same time, Figure 2(b) shows the MRI images
regarding knee joint of an OA aected person. With numerous time series of MR images,
the physicians detect femur, tibia and cartilage of knee bones and observe the inuences
of OA.
Magnetic Resonance Imaging (MRI) create the magnetic properties of assured atomic
nuclei. The human body is mostly covered by water. The hydrogen nuclei present in the
water act as compass needles which are incompletely aligned by a strong magnetic eld
in the scanner. Radio waves rotates the nuclei and oscillate in the magnetic eld while
returning to equilibrium successively. For examining the tissues, the waves concurrently emit
radio signals which are detected by antennas. MRI does not have process like radioactivity
or ionization comparing to the traditional methods. The range of Radio Frequency (RF) for
the operation is normal and won’t aect the body. The signal of MRI is sensitive to a wide
range of eects like nuclear mobility, molecular structure, ow and diusion.
338
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 2. a) MRI of normal knee, b) MRI of OA knee.
Source: own elaboration.
As a consequence, MRI is a very exible technique that provides the interior results of both
structure and function. Initially, the output image of MRI is processed with quantization.
In case of knee joints, MRI images are captured for detecting obvious output regarding
meniscal lesions, cartilage thickness, reshaping the bones, osteophytes, bone marrow lesions
and synovitis (Pelletier et al., 2008). The image segmentation is required to observe the
exact issue happened on the knee whereas the manual calculations have partial accuracy.
Hence, some auto segmentation methods are introduced for detecting and computing the
structure of the bone clearly. In our work, we proposed eagle algorithm (Yang & Deb,
2010) for segmentation process which executes accurate image quality and reduced pixel
breakdown. When compare with the existing method, the accuracy is greatly enhanced and
the thickness can be measured obviously.
2. RELATED WORKS
Early quantitative MRI studies assured promising morphological cartilage metrics for
describing the status of disease and perhaps to screen its progression (Norman, Pedoia, &
Majumdar, 2018). Nonetheless, the recent studies utilize larger OA cohorts by means of
demonstrating MRI articular cartilage biomarkers may have limited approachability to
disease progression. On other hand, the study has sustained to grow novel and better image
segmentation tools (Kubassova, Boyle, & Pyatnizkiy, 2005; Yin et al., 2010). In the present
339
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
decades, most quantitative analysis methods require uninterrupted human involvement
and also should be procient for accurate cartilage segmentation (Kaumann et al., 2003).
Image acquisition and denoising are the two techniques used to reduce the noise from the
images. Average of number of the similar samples that obtained is the result of image
acquisition technique. Conventionally, the image is denoised by denoising techniques which
acquire less time for computation (Liu et al., 2010).
Novel segmentation algorithm (Peterfy, Schneider, & Nevitt, 2008) presented knee cartilages
with level-based segmentation method and new template data. It gives consequence of
three cartilage tissues such as tibia, femur and bula bone. The segmentation of cartilage
is grim in atlas-based segmentation (Pelletier et al., 2008), since cartilage intensity varies
by thin or fat and the boundary of the cartilages are girdled by the muscle tissues. A main
shortcoming of atlas-based approach is the large inconsistency of the articular cartilage.
The potential for preference is equivalent to the target delay in accuracy (Raynauld et al.,
2004).
The multi-atlas based algorithm (Pavlyukevich, 2007) overcome atlas based algorithm
with promising results where it combine simple label fusion approach but not utilize any
other correction methods to report abnormalities like osteophytes produced besides the
joint margins. Alternatively, Local Weighted Vote (LWV) algorithm is built where the multi
atlas data merges and provide the result (Rini et al., 2020). The segmentation structure
grips three procedures including multiple-atlas building, applying a locally weighted vote
(LWV) and adjusting the region. In case of atlas building process, every training cases are
recorded to a target image - a non-rigid registration method and the nest coordinated
atlases are designated. However, LWV algorithm was usually applied for integrating the
data from these atlases and produce the initial segmentation outcome. At last, in the region
adjustment procedure, the statistical data of bone, cartilage, and surrounding regions is
calculated from the preliminary segmentation result. The statistical data absorbed the
automated determination of the seed points in all regions of bone intended for the graph-
cut based method.
The locus-correlated background by voxel subsampling technique is preferred than
uniform or Gaussian subsampling to distinguish the objects of interest from supplementary
340
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
objects. These supplementary objects have similar and close structures that frequently
knock into MR images, particularly for objects focusing on highly curved and complex
shapes (Stammberger et al., 1999). In this approach, cartilage compartments are segmented
underprivileged of prior segmentation of bones or determination of Bone Cartilage
Interface (BCI) demanding multiple MR images of a participant. Recently, a novel method
is introduced called U-net model whose results show longitudinal precision proves the
heftiness of algorithm, free of the ground truth denition (Norman et al., 2018).
The recent convolutional neural networks produce fast, accurate, and precise automatic
segmentations in cartilage and meniscus compartments which are invariant across patients
with OA. This approach also have continuous compensation of computational speed and
established ecacy in extracting relaxation times and morphologic features which used in
the prediction and intensive care of joint degeneration in OA. Local Coordinate System
(LCS) is a method used to segment the femoral and tibia cartilage regions based on 2-D
active contour algorithm (Kaumann et al., 2003). This method obtains a uniform sketch
of cartilage dimensions. Cartilage regions were usually segmented in the 3D-MRI scan
and developed into oset maps with certain gaps acquired from an inter-slice distance. The
oset maps were expressed by Local Coordinate System (LCS).
The gaps are lled with bi-cubic interpolation process. Several segmentation methods
of knee joints are performed on semi-automatic segmentation such as B-spline snakes
(Stammberger et al., 1999), active contours (Lynch et al., 2000; Raynauld et al., 2004),
directional gradient vector ow snakes (Tang et al., 2006), region-growing scheme
(Stammberger et al., 1999) and extended watershed algorithm, where others are achieved
habitually akin to voxel classication algorithm (Stammberger et al., 1999) and statistical
shape model (Fripp et al., 2007). These approaches explain cartilage boundaries in two-
or three-dimensional image spaces. Owing to the curved and thin structure of cartilage
in core, the cartilage segmentation job are quite complex in stimulation. In the work of
Grau et al. (2004), it does not permit informal control of segmentation outow or objective
tessellation but suer with lack of stability and excessive sensitivity to noise.
The reported techniques used basic signal analysis such as directional edge lters (Wolf,
Weierich, & Niemann, 1997), mathematical morphology (Dogdas, Shattuck, & Leahy,
341
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2002), gray-level classication (Folkesson et al., 2005), histogram analysis (Kubassova,
Boyle, & Pyatnizkiy, 2005), and other techniques including the hybrid signal and model,
e.g., texture level-set (Lorigo et al., 1998) and model tting. Nonetheless, these solutions
needed a specic tuning which is responsible for the mutable image quality or the signal
corrosion owing to disease. This tuning trusts on initial manual contribution, additional
signal data or cooperating tuning by making the independent use of these methods dicult
in the context of a large scale process. Alternatively, an algebraic strategy called Douglas-
Rachford Splitting method is presented in the work of Rini, Perumal, and Rajasekaran
(2020) which is segmented by means of proximal splitting method. Various applications of
image processing have been demonstrated in the literature (Muneeswaran & Rajasekaran,
2016, 2017, 2019c).
3. MATERIALS AND METHODS
The block diagram of the proposed Eagle Algorithm (EA) method is shown in Figure
3. The input image is the MRI images of the Osteoarthritis (OA) aected knee image.
In case of image processing, image acquisition is mandatory to transform the analogous
image into a digitalized form. It is done by the process of sampling and quantization. The
spatial resolution of the digitalized image is determined by the sampling rate, whereas the
quantization level controls the number of gray levels in the digitized image. An acquisitioned
image is send as input to the preprocessing where the contrast and the illumination of the
image is completely adjusted. Eagle algorithm extract the thickness of the cartilage region
from the femur, tibia or patella bone and segment the portion. The image is then quantized
for desired number of outputs. The output image is the noiseless and hassle free MRI image
which can detect the thickness of the cartilage precisely.
Figure 3. Block diagram of the proposed method.
Source: own elaboration.
342
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
3.1. PRE-PROCESSING
The purpose of preprocessing is to convert the RGB image into gray image. The image
cannot be processed under the RGB image. Hence it is converted into the binary image
(grey scale image). The gray scale image suer with the issues like contrast, illumination and
blur. In our proposed method, these problems are resolved by using Contrast Enhancement
Method (CHE). CHE provides perfect lightning eect regarding the background and the
Anisotropic Filter (AF) eradicate the noise in image. The Anisotropic Filter (AF) alters the
image to concentrate on the extracted portion where the edges are turned to be blurred.
This lter is non-linear type that permit dispersion in the consistent area and obstructs
at the boundaries. The partial dierential equation (PDE) of anisotropic lter is given by
Equation (1).
(1)
Where
- gradient operator, div - divergence operator, A
0
- Initial image and
diusion coecient, which is represented as
(2)
The input image (A) is a gray level 3-D image from MRI scanning, where the gray value of
a pixel at a point (a′, b′, c′) is specied by x = (a′, b′, c′). Thus, the slices of the 3-D image are
signied as 2-D images, Aa′.
The 2D images are outlined as 3D image limits having gray levels. The image space S is
the collection of all points of (a′, b′, c′) and the slices for the groups are Ωa′=ℎ= {(b′, c′)/
Aa′=ℎ> 0}, idemfor Ωb′=j, Ωc′=t and these spaces are called as axial, coronal and sagittal
slices respectively. The 3-D image A can be dened as a discrete function of the Euclidian
3-D space and their measurements are represented as,
343
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
For the given bone region (φ), the femur region or tibia region are in the same MR image
A, then the region is dened by two factors (u,v) as given in equation (4),
The bone region (φ) is a reference region which have a volume of interest in the whole
cartilage. The values of a′, b′ and c′ indices are integer values, since the values of the
3-D image are digital. Similarly a set D is computed by implicating the equation through
intensity level l which is discrete.
The voxels is a set of unit cubes in R
3
which is similar to the coordinate axes and midpoints
are located in Z
3
.
3.2. STOCHASTIC MULTI OBJECTIVE OPTIMIZATION
Generally, a regular optimization problem without including any noise or uncertainty, can
be given by
Where s = (s1, s2... sd)
T
is the vector of design variables.
Monte Carlo method is a sampling technique to determine µf. After the samples drawn
randomly, we have
(6)
Where M
i
denotes the number of samples.
3.3. EAGLE ALGORITHM
The scavenging behavior of eagles especially the behavior of golden eagle is motivating.
Generally, an eagle feeds in its own region by ying freely in a random manner similar
to the Levy ights. When the prey is found by eagle, automatically it will alter its search
strategy to a concentrated chasing strategies to catch the prey as eciently as possible. Thus,
344
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
two important mechanisms are involved in an eagle’s hunting strategy: random search by
Levy ight (or walk) and intensive chase by fastening its aim on the objective. The work
of Reynolds and Frye works with fruit ies, see the sights at their landscape by means of a
series of straight ight paths interrupted by an unexpected 900 turn, prominent to a Levy-
ight-style sporadic scale-free search pattern. However, light also can be connected to Levy
ights (Barthelemy, Bertolotti, & Wiersma, 2008). Consequently, this kind of behavior has
been utilized in the process of optimization and optimal search, and preliminary results
depicted with its enhanced capability (Pavlyukevich, 2007).
In our proposed work, let us fetishize the two-stage approach of an eagle’s foraging
behavior. At rst, assumption is made that an eagle has to do the Levy walk in the entire
domain. After the detection of prey, it alters to a chase strategy. Next, the chase strategy
can be measured as an intensive local search using any optimization techniques. Likewise,
the output image of the Anisotropic Filter (AF) is scanned randomly for the segmentation
process. When it detects the area of region where have to be segment, then the area is
intensely chased and segmented. Since the area of the region is chasing intensely without
considering any other region, the pixel is broken in reduced number. In the initial step, EA
is a two-stage strategy rather than a simple iterative method which going to conglomerate
a better randomization technique of global search with an intensive and well-organized
local search method. Additionally, EA applies Levy walk rather than simple randomization,
which resource that the global search space can be reconnoitered prociently. In fact, the
studies show that Levy walk is more ecient than simple random-walk exploration. The
Levy walk has a random step length being drawn from a Levy distribution which is given by
(7)
Where v represents the innite variance with innite mean. Here the steps of the eagle
motion is fundamentally a random walk process by means of a power-law step-length
distribution with a substantial tail. The special case (λ = 3) corresponds to Brownian motion,
whereas the case (λ = 1) has a physical appearance of stochastic excavating, which may be
more resourceful to escape being surrounded in local optima. In our proposed work, we use
the eagle algorithm to do the local search, since the eagle algorithm was designed to crack
multimodal global optimization problems.
345
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
4. RESULTS
For our proposed work, we have taken an input set with three regions of knee bone namely
femur, tibia and patella region. These three samples are tested with every process of Eagle
Algorithm (EA) as follows. Initially, the input images of the knee bone region are the gray
scale or RGB image which covers the cartilage region as revealed in Figure 4. The input
image is converted to gray scale image under the process of preprocessing. Anisotropic
Filter (AF) lters the brightness whereas the Contrast Enhancement Method (CHE) adjust
the brightness of the image.
Figure 4. The input images of femur, tibia and patella of knee bone region.
Source: own elaboration.
In case of the binary images, the thresholding process is mandatory. Our technique works
with the Otsu method for binarization which further provide contrast enhancement in the
binary images. Otsu method produce best results in contrast with the other threshold lters
and also ensures the clear view about the thickness of the images as shown in Figure 6.
Figure 5. Otsu method output images.
Source: own elaboration.
The contrast enhancement for altering the eect of the color images at the cartilage region
as shown in Figure 6. It shown clearly about the cartilage region at where we can obviously
view the cartilage region. Comparing the Figure 5 and Figure 6, it is obvious about the
optimal dierences in the region dierentiation.
346
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 6. Contrast enhancement.
Source: own elaboration.
The Anisotropic Filter (AF) is included with the process to remove the noise and region of
interest (ROI) of the image as shown in Figure 7. From this depicted gure, we can observe
the removal of noise at the segmenting region.
Figure 7. Anisotropic Filter (AF) image.
Source: own elaboration.
As we discussed above, the Eagle Algorithm (EA) is categorized under levy walk process
for extracting the area pattern and intensive chasing for identifying the thickness of the
pattern. The stochastic multi objective process deliberates the random search to extract that
where the thickness has to be determined. After it founds the region, it concentrates on the
desired pattern whereas another surround with blur as shown in Figure 8.
347
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 8. Stochastic multi objective process.
Source: own elaboration.
The above gure illustrates the probability of achieving maximum thickness for several
input set images. The objective function of the scale measures the value of the accuracy
in the images. It is found that the images are steadily removed by the noise and after the
completion of the search. In contrast with the traditional method, less computation time
is required in this process. The power law regression of the set of data sets of MRI images
are depicted in Figure 9.
Figure 9. Power law regression.
Source: own elaboration.
348
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
4.1. PERFORMANCE METRICS
The quality measurement of the output image is dignied by means of the thickness, mean
and the standard deviation as publicized in Table 1. The proposed work greatly enhanced
accuracy of the image when compare with the existing method.
Table 1. Performance metrics of input images using eagle algorithm.
Inputs Thickness Mean Standard Deviation
1 0.35503 0.04237 0.016560
2 0.45060 0.04183 0.015588
3 0.24184 0.04351 0.015193
4 0.19432 0.04329 0.013595
5 0.30043 0.04190 0.015678
6 0.53720 0.04266 0.014890
7 0.39266 0.04251 0.015023
8 0.40385 0.04414 0.013999
9 0.39356 0.04380 0.013750
10 0.27988 0.04380 0.013653
Source: own elaboration.
Thus, the eagle algorithm extracts the specic pattern to segment and concentrate on
that pattern by clear segmentation with multi stochastic process and the better accuracy is
achieved.
5. CONCLUSIONS
Our work, we proposed Eagle Algorithm (EA) for the enhancement of the segmentation
process of knee images. The pre-processed image is a binary image with heavy noise. The
noise is reduced by using Anisotropic Filter (AF). The brightness of the image is controlled
and stabilized by Contrast Enhancement Method (CHE). The segmentation is done with
two steps namely levy walk and intensive chasing. After the random search, the area is
extracted accurately with intensive chasing. The algorithm xes the cartilage part of the
bone and segment with better clarity. The performance of the algorithm is analyzed using
mat lab simulations. The thickness, mean and standard deviation result shows that the
proposed method has developed result in contrast with the traditional methods.
349
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
REFERENCES
Barthelemy, P., Bertolotti, J., & Wiersma, D. S. (2008). A Lévy ight for light. Nature,
453(7194), 495-498. https://www.nature.com/articles/nature06948
Dogdas, B., Shattuck, D. W., & Leahy, R. M. (2002). Segmentation of the skull in 3D
human MR images using mathematical morphology. In Medical Imaging 2002: Image
Processing (Vol. 4684, pp. 1553-1562). International Society for Optics and Photonics.
https://ui.adsabs.harvard.edu/abs/2002SPIE.4684.1553D/abstract
Eckstein, F., Burstein, D., & Link, T. M. (2006). Quantitative MRI of cartilage and
bone: degenerative changes in osteoarthritis. NMR in Biomedicine: An International
Journal Devoted to the Development and Application of Magnetic Resonance In vivo, 19(7), 822-
854. https://www.researchgate.net/publication/6720453_Quantitative_MRI_of_
cartilage_and_bone_Degenerative_changes_in_osteoarthritis
Folkesson, J., Dam, E., Olsen, O. F., Pettersen, P., & Christiansen, C. (2005).
Automatic Segmentation of the Articular Cartilage in Knee MRI Using a
Hierarchical Multi-class Classication Scheme. In Duncan, J.S., & Gerig, G. (eds.)
Medical Image Computing and Computer-Assisted Intervention – MICCAI 2005. Lecture
Notes in Computer Science, vol 3749. Springer, Berlin, Heidelberg. https://doi.
org/10.1007/11566465_41
Fripp, J., Crozier, S., Wareld, S. K., & Ourselin, S. (2007). Automatic segmentation
of the bone and extraction of the bone–cartilage interface from magnetic resonance
images of the knee. Physics in Medicine & Biology, 52(6), 1617. https://pubmed.ncbi.
nlm.nih.gov/17327652/
Grau, V., Mewes, A. U. J., Alcaniz, M., Kikinis, R., & Wareld, S. K. (2004).
Improved watershed transform for medical image segmentation using prior
information. IEEE transactions on medical imaging, 23(4), 447-458. https://ieeexplore.
ieee.org/document/1281998
Kaumann, C., Gravel, P., Godbout, B., Gravel, A., Beaudoin, G., Raynauld,
J. P., ... & de Guise, J. A. (2003). Computer-aided method for quantication of
350
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
cartilage thickness and volume changes using MRI: validation study using a synthetic
model. IEEE transactions on Biomedical Engineering, 50(8), 978-988. https://ieeexplore.
ieee.org/document/1213850
Kubassova, O., Boyle, R. D., & Pyatnizkiy, M. (2005). Bone segmentation in
metacarpophalangeal MR data. In International Conference on Pattern Recognition and
Image Analysis (pp. 726-735). Springer, Berlin, Heidelberg.
Li, L., Muneeswaran, V., Ramkumar, S., Emayavaramban, G., & Gonzalez, G. R.
(2019). Metaheuristic FIR lter with game theory based compression technique-A
reliable medical image compression technique for online applications. Pattern
Recognition Letters, 125, 7-12. https://doi.org/10.1016/j.patrec.2019.03.023
Liu, H., Yang, C., Pan, N., Song, E., & Green, R. (2010). Denoising 3D MR images
by the enhanced non-local means lter for Rician noise. Magnetic resonance imaging,
28(10), 1485-1496. https://doi.org/10.1016/j.mri.2010.06.023
Lorigo, L. M., Faugeras, O., Grimson, W. E. L., Keriven, R., & Kikinis, R. (1998).
Segmentation of bone in clinical knee MRI using texture-based geodesic active
contours. In Wells W.M., Colchester, A., Delp, S. (eds.) Medical Image Computing and
Computer-Assisted Intervention — MICCAI’98. Lecture Notes in Computer Science, vol
1496. Springer, Berlin, Heidelberg. https://doi.org/10.1007/BFb0056309
Lynch, J. A., Zaim, S., Zhao, J., Stork, A., Peterfy, C. G., & Genant, H. K. (2000).
Cartilage segmentation of 3D MRI scans of the osteoarthritic knee combining user
knowledge and active contours. In Medical Imaging 2000: Image Processing (Vol. 3979,
pp. 925-935). International Society for Optics and Photonics.
Muneeswaran, V., & Rajasekaran, M. P. (2016). Analysis of particle swarm optimization
based 2D FIR lter for reduction of additive and multiplicative noise in images. In
International Conference on Theoretical Computer Science and Discrete Mathematics (pp. 165-
174). Springer, Cham.
Muneeswaran, V., & Rajasekaran, M. P. (2016). Performance evaluation of radial basis
function networks based on tree seed algorithm. In 2016 International Conference on
351
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Circuit, Power and Computing Technologies (ICCPCT) (pp. 1-4). IEEE. https://ieeexplore.
ieee.org/document/7530267
Muneeswaran, V., & Rajasekaran, M. P. (2017). Beltrami-Regularized Denoising Filter
Based on Tree Seed Optimization Algorithm: An Ultrasound Image Application.
In Satapathy, S., Joshi, A. (eds.) Information and Communication Technology for Intelligent
Systems (ICTIS 2017) - Volume 1. ICTIS 2017. Smart Innovation, Systems and
Technologies, vol 83. Springer, Cham. https://doi.org/10.1007/978-3-319-63673-
3_54
Muneeswaran, V., & Rajasekaran, M. P. (2018). Gallbladder shape estimation using
tree-seed optimization tuned radial basis function network for assessment of acute
cholecystitis. In Intelligent Engineering Informatics (pp. 229-239). Springer, Singapore.
Muneeswaran, V., & Rajasekaran, M. P. (2019a). Automatic segmentation of
gallbladder using bio-inspired algorithm based on a spider web construction model.
The Journal of Supercomputing, 75(6), 3158-3183. https://doi.org/10.1007/s11227-
017-2230-4
Muneeswaran, V., & Rajasekaran, M. P. (2019b). Automatic segmentation of
gallbladder using intuitionistic fuzzy based active contour model. In Microelectronics,
Electromagnetics and Telecommunications (pp. 651-658). Springer, Singapore.
Muneeswaran, V., & Rajasekaran, M. P. (2019c). Local contrast regularized contrast
limited adaptive histogram equalization using tree seed algorithm—an aid for
mammogram images enhancement. In Smart intelligent computing and applications (pp.
693-701). Springer, Singapore.
Norman, B., Pedoia, V., & Majumdar, S. (2018). Use of 2D U-Net convolutional
neural networks for automated cartilage and meniscus segmentation of knee MR
imaging data to determine relaxometry and morphometry. Radiology, 288(1), 177-
185. https://pubmed.ncbi.nlm.nih.gov/29584598/
Pavlyukevich, I. (2007). Lévy ights, non-local search and simulated annealing. Journal of
Computational Physics, 226(2), 1830-1844. https://doi.org/10.1016/j.jcp.2007.06.008
352
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Pelletier, J. P., Raynauld, J. P., Abram, F., Haraoui, B., Choquette, D., & Martel-
Pelletier, J. (2008). A new non-invasive method to assess synovitis severity in relation
to symptoms and cartilage volume loss in knee osteoarthritis patients using MRI.
Osteoarthritis and cartilage, 16, S8-S13. https://pubmed.ncbi.nlm.nih.gov/18672386/
Peterfy, C. G., Schneider, E., & Nevitt, M. (2008). The osteoarthritis initiative: report
on the design rationale for the magnetic resonance imaging protocol for the knee.
Osteoarthritis and cartilage, 16(12), 1433-1441. https://www.ncbi.nlm.nih.gov/pmc/
articles/PMC3048821/
Raynauld, J. P., Martel‐Pelletier, J., Berthiaume, M. J., Labonté, F., Beaudoin, G.,
De Guise, J. A., ... & Hochberg, M. C. (2004). Quantitative magnetic resonance
imaging evaluation of knee osteoarthritis progression over two years and correlation
with clinical symptoms and radiologic changes. Arthritis & Rheumatism: Ocial Journal
of the American College of Rheumatology, 50(2), 476-487. https://pubmed.ncbi.nlm.nih.
gov/14872490/
Rini, C., Perumal, B., & Rajasekaran, M. P. (2020). Automatic knee joint segmentation
using Douglas-Rachford splitting method. Multimedia Tools and Applications, 79(9),
6599-6621. https://doi.org/10.1007/s11042-019-08303-8
Schmid, J., & Magnenat-Thalmann, N. (2008). MRI bone segmentation using
deformable models and shape priors. In International conference on medical image computing
and computer-assisted intervention (pp. 119-126). Springer, Berlin, Heidelberg.
Stammberger, T., Eckstein, F., Michaelis, M., Englmeier, K. H., & Reiser, M.
(1999). Interobserver reproducibility of quantitative cartilage measurements:
comparison of B-spline snakes and manual segmentation. Magnetic resonance imaging,
17(7), 1033-1042. https://pubmed.ncbi.nlm.nih.gov/10463654/
Tang, J., Millington, S., Acton, S. T., Crandall, J., & Hurwitz, S. (2006). Surface
extraction and thickness measurement of the articular cartilage from MR images
using directional gradient vector ow snakes. IEEE Transactions on Biomedical Engineering,
53(5), 896-907. https://ieeexplore.ieee.org/document/1621141
353
https://doi.org/10.17993/3ctecno.2021.specialissue8.333-353
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Wolf, M., Weierich, P., & Niemann, H. (1997). Automatic Segmentation and
3D-Registration of a femoral bone in MR images of the knee. Pattern Recognition and
Image Analysis (Advances in Mathematical Theory and Applications), 7(1), 152-165.
Yang, X. S., & Deb, S. (2010). Eagle strategy using Lévy walk and rey algorithms for
stochastic optimization. In Nature Inspired Cooperative Strategies for Optimization (NICSO
2010) (pp. 101-111). Springer, Berlin, Heidelberg.
Yin, Y., Zhang, X., Williams, R., Wu, X., Anderson, D. D., & Sonka, M. (2010).
LOGISMOS—layered optimal graph image segmentation of multiple objects and
surfaces: cartilage segmentation in the knee joint. IEEE transactions on medical imaging,
29(12), 2023-2037. https://pubmed.ncbi.nlm.nih.gov/20643602/
