International Journal on Future Revolution in Computer Science & Communication Engineering (IJFRCSCE)

ISSN: 2454-4248 Volume: 8 Issue: 1

DOI: https://doi.org/10.17762/ijfrcsce.v8i1.2084
Article Received: 24 December 2021 Revised: 16 February 2022 Accepted: 22 February 2022 Publication: 31 March 2022
____________________________________________________________________________________________________________________

An Efficient Medical Image Processing Approach

Based on a Cognitive Marine Predators Algorithm
Dr. Sunita Chaudhary
Professor, Computer Science and Engineering,
Marudhar Engineering College,
Bikaner, Rajasthan,
ORCID ID-0000-0001-8913-4897
choudhary.sunita@marudhar.ac.in

Abstract
Image processing aims to enhance the image's quality such that it is simple for both people and robots to understand. Medical image
processing and Biomedical signal processing have many conceptual similarities. Medical image processing involves evaluation,
enhancement, and presentation. The focus of medical imaging is on obtaining photographs for both therapeutic and diagnostic reasons. In the
existing Marine Predator Algorithm, different disadvantages are experienced when various automated optimization algorithms are used to the
problem of ECG categorization. The proposed method follows the flow outlined here: data collection, image preprocessing using histogram
equalization, segmentation using the Otsu threshold algorithm, feature extraction using the contour method, feature selection using the
Neighborhood Component Analysis (NCA) algorithm, and Cognitive Marine Predator Algorithm (CMPA) as the proposed method. By using
the Cognitive Marine Predators Algorithm (CMPA), base layers are fused to use the greatest feasible parameters, producing enhanced high-
quality output images. Finally, the image processing performance is analyzed. The proposed approaches overcome the drawbacks of existing
algorithms and increase the quality of medical images efficiently.

Keywords: Medical image processing, Cognitive Marine Predator Algorithm (CMPA), Neighborhood Component Analysis (NCA), Imaging
technologies

I. INTRODUCTION development of medical image analysis methods. As a

When it comes to protecting human health and monitoring result, open software platforms may facilitate the translation
illness, medical imaging and image interpretation are of innovative technological advances to the MRI hardware
critical. Medical imaging equipment has seen a surge in use makers, doctors, and the biomedical sector [2]. Medical
in recent years, thanks to advancements in technology. imaging may benefit from the fact that various datasets of
Medical imaging technology development is still the same patient, such as different MR sequences, PETCT
undergoing intense scrutiny. In 2018, an estimated 18.1 scans, or innovative hybrid CT-MR imaging modalities, are
million people will be diagnosed with cancer, and 9.6 people often accessible. As a result, even if a scan was obtained
will die from the disease. To eradicate many deadly using a different modality, leveraging redundant information
illnesses, early identification is essential. When compared to from across the datasets may assist recover lower-quality or
the global population, it is understandable that it is difficult lower-resolution images. With this, the inherent difficulty of
to get an early diagnosis. Expertise is scarce in low-income single-image restoration is alleviated: the illusion of data
nations. Medical photographs, in particular, need a lengthy that is not there in the original picture. To make medical
examination because of their size and complexity [1]. Image picture improvement easier to understand, we've integrated
processing methods like MRI have grown fast in recent the technique with adaptive image filters. This separates the
decades, prompting the development of digital image service's outputs from the image undergoing processed [3].
processing algorithms specifically designed for these It is very challenging to analyze medical pictures for illness
frequently enormous imaging datasets (3-dimensional and detection and diagnosis because of the growing complexity
higher-dimensional). Quantitative information may be of the images. Furthermore, as can be seen from the
extracted from images using techniques including aberration provided image, the area is heavily diseased. Image
reduction, identification, and separation by anatomical processing approaches utilized for segmenting and
segmentation, and spatial image registration. The academic categorizing tumor-infected areas are to be evaluated in this
research community is primarily responsible for the data analysis. Although medical images must be analyzed in
8
IJFRCSCE | March 2022, Available @ http://www.ijfrcsce.org
 International Journal on Future Revolution in Computer Science & Communication Engineering (IJFRCSCE)
ISSN: 2454-4248 Volume: 8 Issue: 1
DOI: https://doi.org/10.17762/ijfrcsce.v8i1.2084
Article Received: 24 December 2021 Revised: 16 February 2022 Accepted: 22 February 2022 Publication: 31 March 2022
____________________________________________________________________________________________________________________
detail to obtain useful information. Hence, we proposed the processing-based approaches for detecting cancers of the
CMPA for efficient image visualization in medical image lung, brain, and liver. In an attempt to handle enormous
processing. datasets and offer reliable and efficient findings in the
diagnosis of cancer, mechanized and computer-aided
Contributions to the paper detection systems (CAD) with artificial intelligence are
➢ The histogram equalization method is used to utilized as detection techniques.
improve the contrast in the raw dataset.
➢ The Otsu threshold technique is used to convert an III. PROPOSED APPROACHES
image into a collection of regions of pixels in the In this paper, we proposed the CMPA for medical image
dataset. processing. Medical image processing includes the use and
➢ The contour method for feature extraction is used to investigation of 3D datasets of the human internal organs
detect and extract the boundaries in the dataset. that are often acquired from a CT or MRI scanner to
➢ The Neighborhood Component Analysis (NCA) diagnose disorders or direct medical treatments like surgical
algorithm is used to identify the relevant feature planning or experimentation. Radiologists, engineers, and
subsets. doctors use medical image processing to learn more about
the anatomy of specific individuals or populations. Figure 1
The remainder of the paper is structured as follows: In part depicts the schematic representation of the proposed
II, a literature survey is provided. Described in Part III is a methodology.
proposed approach. Part IV includes a result and discussion.
Part V is the conclusion section.

II. LITERATURE SURVEY

Study [4] provided a brief history of deep learning and its
use in medicine in the recent decade. Furthermore, three
examples of future research directions for COVID-19
medical image processing are provided from China, Korea,
and Canada. Study [5] described the most advanced deep
learning architecture and its optimization. Deep learning
approaches in medical imaging and accessible data are
covered as the session makes a decision. Various CNN Figure 1: Schematic representation of the proposed
structure and their medical imaging applications have been methodology
thoroughly discussed by the study [6]. There has also been a
A. Data collection
comparison of the high tech in medical imaging
An MRI scan dataset comprised 319 images from distinct
classification with different available techniques. The latest
individuals. BRAMSIT is made up of 319 images depicting
work including potential future directions is discussed in
a wide range of topics. Each subject is identified by a unique
this study [7] for deep learning algorithms applied in clinical
reference number, age, and an axial identifying number.
image analysis. Deep learning in the subject of medical
There are many examples of normal scan pictures of
image processing, is a valuable source of information. After
BRAMSIT patients shown in Figure 2. Figure 3 depicts the
that the difficulties in applying it to medical imaging and
aberrant and ground truth MRI scans of five individuals
some unresolved research issues. Deep learning is used to
[11].
develop an effective medical image processing system.
Web-based image processing systems are subjected to a
series of security checks by the creators. When it comes to
evaluating security in medical image processing, study [8]
recommends using FAHP and TOPSIS, two techniques that
use fuzzy analysis and similarity to an ideal solution. Using
Morphing, the study [9] offered a new method for recreating
3D point clouds from 2D images. Mimics are used to create
Figure 2: Sample of MRI normal scan images
a 3D surface model of soft tissue (live) from a sequence of
2D CT scans. Study [10] provided an overview of image
9
IJFRCSCE | March 2022, Available @ http://www.ijfrcsce.org
 International Journal on Future Revolution in Computer Science & Communication Engineering (IJFRCSCE)
ISSN: 2454-4248 Volume: 8 Issue: 1
DOI: https://doi.org/10.17762/ijfrcsce.v8i1.2084
Article Received: 24 December 2021 Revised: 16 February 2022 Accepted: 22 February 2022 Publication: 31 March 2022
____________________________________________________________________________________________________________________
grey level. To increase the contrast between the foreground
and background areas, both components might be equalized
independently.

Automated thresholding of probability shape-based images

is done using Otsu's approach. As a normalized sum of two
classes' variations, the Otsu approach seeks the threshold
level that minimizes intraclass variance.
Figure 3: sample of MRI abnormal scan images
ArgMax
B. Image processing using histogram equalization V0 = {K H (G(AH ) − G(A))2 + K Z (G(AZ ) −
Vc
Raw images are subjected to image preprocessing, which G(A))2 } (4)
involves reducing abnormalities and enhancing significant
aspects of the images so that they may be used in subsequent The optimal threshold Vc is used in this equation.
post-processing steps. The image preprocessing technique Underlying pictures K H , as well as K Z , is their PDF
produced a set of n sub-histograms. [0 255] is the range of counterparts. For example, the mean image quality of AH
the histogram of the images being used. We may begin by and AZ is defined as G(AH ) and G(AZ ) whereas the average
setting ia , and ra are also the weakest and greatest intensity quality of the image is defined as G(A).
values of the sub histograms of image DK (W), and the range
of W -1 is [ia , ra ]. D. Feature extraction using the contour method
A histogram equalization step is required before contour
Generalized probabilities DK (W) are applied to every extraction can begin. In general, historical traditions have
individual histogram by the equation (1). low quality since the materials decay with time owing to
factors such as storage conditions. As an independent
D(w)−Dmin pa
DK (W) = Dmax ( ) ia ≤ w ≤ ra (1) variable, the active contour approach specifies a generalized
Dmax −Dmin
function of "power" with a continuous curve that is made up
Where pa the accumulated probability density of the ith sub- of two terms: one for internal and one for external sources of
histogram is, Dmax is the highest and Dmin is the lowest. energy. An image feature extraction that transforms into an
intuitive function-solving method with exceptionally
Following is an equation that normalizes the resulting rigorous mathematical reasoning may be used to establish
weighted probabilities Dk (m), the goal boundary. Active contour models are also known as
functional active contour models. Furthermore, significant
D (m)
Dkq (m) = ∑h−1k (2) structural changes, such as the splitting or merging of
m=0 Dk (m)
curves, cannot be implemented by these models. In this
A normalized weighted probability of input image case, just one target profile may be derived. It is also
histogram Dkq (m) is often used. difficult to extract the full return on investment (ROI) from
brain images at one time because of the varying image
Using the probability mass function of the input picture, contrast, thus it is required to complete the segmentation of
histogram equalization generates a uniformly distributed numerous areas and combine certain post-processing
histogram. The following equation is used to equalize each techniques to complete ROI extraction.
sub-histogram a with a range of [ra − ia ].
E. Feature selection using neighborhood components
Oa (w) = ia + (ra − ia ) ∗ Tkq (m) 1 ≤ a ≤ q (3) analysis
The process of selecting the best k features from among the
C. Segmentation using Otsu threshold dataset's features based on the algorithm's assessment results
The best method for separating an item from the backdrop is known as feature selection. There has been a wide range
of a picture is threshold segmentation. The histogram is of assessment criteria for feature selection approaches to
divided into two portions, the target area and the date. Based on the learning technique utilized, these criteria
background region, using an appropriate threshold between are generally separated into two groups: dependent and
the object's lower grey level and the background's higher independent. The NCA method, which relies on dependent
10
IJFRCSCE | March 2022, Available @ http://www.ijfrcsce.org
 International Journal on Future Revolution in Computer Science & Communication Engineering (IJFRCSCE)
ISSN: 2454-4248 Volume: 8 Issue: 1
DOI: https://doi.org/10.17762/ijfrcsce.v8i1.2084
Article Received: 24 December 2021 Revised: 16 February 2022 Accepted: 22 February 2022 Publication: 31 March 2022
____________________________________________________________________________________________________________________
criteria, was used in our research to examine the impact of Where, a population's, m-dimensional coordinates are
RBC characteristics on categorization. Algorithm selection represented by Vam , and n is the population size. d is the
is based on prediction and response values. In the method, solution dimension. The search space's upper and lower
the brain parameters were utilized as input parameters, and bounds are zr and hr, respectively, and a random integer is
the best-discriminating parameters were grouped into a a rand.
smaller set. The following are the specifics of how we Phase (ii): When the predator is going faster than the prey.
arrived at the final list of features: MPA has a worldwide search engine and a rapid
convergence time.
𝐻 = {(𝑦1 , 𝑦1 ), … , (𝑦𝑗 , 𝑦𝑗 ), … , (𝑦𝐾 , 𝑦𝐾 )} (5)
step ⃗ R ⊗ (Gh
⃗ sizea = B ⃗ itea − B
⃗ R ⊗ Pre⃗ ya )a = 1, … q
There are d-dimensional vectors, and there are K- {
⃗ ⊗ step
Pre⃗ ya = Pre⃗ ya + D. B ⃗ sizea
dimensional vectors, and there is a K-class label for each of
(10)
them. Weighting vector w, which may be used to determine
the best feature set for closest neighbor classification, is a In this case, step⃗ sizea denotes the step size. The normal
major purpose. The equation shows the weighted distance distribution of Brownian motion generates a random number
between two samples, va , and vm , about the weighting ⃗ R , and ⊗ the multiplicative operator is represented
vector, B
vector K.
by Pre⃗ ya . D is the step control factor is set at 0.5, and the
Pk (va , vm ) = ∑bj=1 k 2j |vaj − vmj | (6) random number generator generates⃗⃗⃗B from a distribution of
values between 0 and 1. In this case, the current iteration
th
Wm is the weight of the j feature, as shown in the figure. A number is Iter. The maximum number of iterations is Max
kernel function k that yields big values for tiny kj may be Iter. There are three key drawbacks of MPA: the inability to
used to establish a link between the probability Dam and the build a varied starting population with high productivity,
weighted distance Pk . It is possible to define Dam as shown and the incapacity to explore the search area more generally
in the following equation (7). and widely.
IV. RESULT AND DISCUSSION
w(Pk (Va ,Vm ))
Dam = ∑q (7) In this paper, we proposed the CMPA for medical image
m=1m≠a w((Pk (Va ,Vm ))
processing. Accuracy, F1-score, recall, and precision are
𝑧
If a = m, then pam = 0. This is how we get kz = (exp - ), analyzed with proposed and existing methods. Existing
𝜎
the kernel function looking for each point's likelihood of approaches such as hybrid deep convolution neural network
being picked as the reference point is affected by the kernel [HDCNN], multimodal deep guided filtering [MMDGF],
width. According to the equation, Va likelihood of being deep neural network [DNN], and neutrosophic set [NS] are
correctly classified is equation (8). compared with the proposed method.

Da = ∑m oam pam (8)

F. Cognitive marine predators algorithm

As a way of gathering information on how marine predators
hunt, CMPA has been developed. There are three periods in
which marine predators and their prey move at various rates,
which affects how well an MPA performs. In this phase, the
prey is moving quicker than the predator can keep up with
it. Two distinct phases:

Phase (i): When both the prey and the predator are traveling
at about the same pace.

Vam = hr + rand(zr − hr) a = 0 … q, m = 0. . d (9)

Figure 4: Comparison of the accuracy

11
IJFRCSCE | March 2022, Available @ http://www.ijfrcsce.org
 International Journal on Future Revolution in Computer Science & Communication Engineering (IJFRCSCE)
ISSN: 2454-4248 Volume: 8 Issue: 1
DOI: https://doi.org/10.17762/ijfrcsce.v8i1.2084
Article Received: 24 December 2021 Revised: 16 February 2022 Accepted: 22 February 2022 Publication: 31 March 2022
____________________________________________________________________________________________________________________
Figure 4 depicts the comparison of the accuracy. The
accuracy of a method is measured by utilizing a metric that
considers a technique works effectively throughout all of its
constituent parts. If every problem is equally important, this
is a good thing. The value is determined by dividing the
total number of ideas by the total number of true statements.
𝑇𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 +𝑇𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (11)
𝑇𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 +𝑇𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒+𝐹𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 +𝐹𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒

The suggested work was found to be more accurate (98%)

than the existing methods like HDCNN (45%), MMDGF
(69%), DNN (76%), and NS (82%).

Figure 6: Comparison of the precision

Figure 6 depicts the comparison of the precision. Positive

predictive value (PPV) is another term for a measure of
accuracy. Precision is a metric for determining the number
of correct class predictions out of a given sample. In other
words, it's a measure of how well things went as opposed to
how well they were projected to go. To find out how precise
a measurement is, apply the formula below:

True positive
Precision = (13)
Total predicted positive

Figure 5: Comparison of the recall The proposed method CMPA shows more significance
(98%) precision than the other existing methods like
Figure 5 comparison of the recall. In the medical image HDCNN (77%), MMDGF (85%), DNN (57%), and NS
processing of information systems, recall assesses how (66%).
successfully a suggested technique locates the supporting
information that a person has desired to be returned to it. It
has been determined that a set of measures in the following
way:

𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒
Recall= (12)
𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑐𝑡𝑢𝑎𝑙 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠

The suggested work was found to be more recall (98%) than

the existing methods like HDCNN (57%), MMDGF (66%),
DNN (75%), and NS (85%).

Figure 7: Comparison of the F1-score

12
IJFRCSCE | March 2022, Available @ http://www.ijfrcsce.org
 International Journal on Future Revolution in Computer Science & Communication Engineering (IJFRCSCE)
ISSN: 2454-4248 Volume: 8 Issue: 1
DOI: https://doi.org/10.17762/ijfrcsce.v8i1.2084
Article Received: 24 December 2021 Revised: 16 February 2022 Accepted: 22 February 2022 Publication: 31 March 2022
____________________________________________________________________________________________________________________
Figure 7 depicts the comparison of the F1 score. Precision existing methods. If we analyze more results in the future,
and recall are combined in the F1-Score. It's sometimes then the performance of this research will be increased
referred to as the "frequency mean" since it's the average of effectively.
the two. Another approach to determining an "average" of
numbers, the harmonic mean is often touted as being more References
suited for ratios (such as precision and recall) than the [1]. Öztürk, Ş., Ahmad, R. and Akhtar, N., 2020. Variants of
Artificial Bee Colony algorithm and its applications in
standard statistical distribution. Throughout this
medical image processing. Applied soft computing, 97,
investigation, the proposed work CMPA has more f1-score
p.106799.
(98%) than the existing methods like HDCNN (56%), [2]. Tournier, J.D., Smith, R., Raffelt, D., Tabbara, R.,
MMDGF (66%), DNN (77%), and NS (85%). Dhollander, T., Pietsch, M., Christiaens, D., Jeurissen, B.,
Yeh, C.H. and Connelly, A., 2019. MRtrix3: A fast,
Discussion flexible and open software framework for medical image
Due to the general limitations of the existing method, processing and visualization. Neuroimage, 202, p.116137.
figures clearly illustrate that the proposed methodology [3]. Stimpel, B., Syben, C., Schirrmacher, F., Hoelter, P.,
improves the existing one. The disadvantages of the current Dörfler, A. and Maier, A., 2019. Multi-modal deep guided
techniques are as follows. The proposed method is filtering for comprehensible medical image processing.
compared to the existing methods like hybrid deep IEEE transactions on medical imaging, 39(5), pp.1703-
1711.
convolution neural network [HDCNN], multimodal deep
[4]. Bhattacharya, S., Maddikunta, P.K.R., Pham, Q.V.,
guided filtering [MMDGF], deep neural network [DNN] and
Gadekallu, T.R., Chowdhary, C.L., Alazab, M. and Piran,
neutrosophic set [NS] are more efficient for the CMPA for M.J., 2021. Deep learning and medical image processing
medical image process. The HDCNN quality of the medical for coronavirus (COVID-19) pandemic: A survey.
image process was impacted by the usage of more images Sustainable cities and society, 65, p.102589.
than was necessary. Generally, the MMDGF image should [5]. Razzak, M.I., Naz, S. and Zaib, A., 2018. Deep learning
be decoupled from the processed output to minimize the for medical image processing: Overview, challenges and
impact of the adversary approach on the guidance image. the future. Classification in BioApps, pp.323-350.
DNN-based medical image analysis is vulnerable to a wide [6]. Mohapatra, S., Swarnkar, T. and Das, J., 2021. Deep
convolutional neural network in medical image processing.
range of assaults; however, the changes to the classier are so
In Handbook of deep learning in biomedical engineering
minor that they have limited effect on the diagnosis
(pp. 25-60). Academic Press.
outcomes when medical specialists are engaged. The great [7]. Mishra, S., Tripathy, H.K. and Acharya, B., 2021. A
throughput and imprecise handling capabilities of precise analysis of deep learning for medical image
neutrosophic set-based algorithms have garnered interest in processing. In Bio-inspired neurocomputing (pp. 25-41).
tackling image processing tasks. As a result, the method for Springer, Singapore.
medical image processing developed by the CMPA is more [8]. Algarni, A., Ahmad, M., Attaallah, A., Agrawal, A.,
effective. Kumar, R. and Khan, R.A., 2020. A fuzzy multi-objective
covering-based security quantification model for mitigating
V. CONCLUSION the risk of the web-based medical image processing
In this paper, the field of medical image processing, areas system. International Journal of Advanced Computer
such as illness identification, segmentation, detection, and Science and Applications, 11(1).
[9]. Cheng, Q., Sun, P., Yang, C., Yang, Y. and Liu, P.X.,
image normalization have benefited greatly from the fast-
2020. A morphing-Based 3D point cloud reconstruction
evolving technology. The paper proposed CMPA for framework for medical image processing. Computer
medical image processing regarding brain tumor methods and programs in biomedicine, 193, p.105495.
classification. This research analyzes 319 images from the [10]. Kaur, C. and Garg, U., 2021. Artificial intelligence
BRAMSIT database. Histogram equalization is used to techniques for cancer detection in medical image
image preprocessing, Ostu threshold is used for processing: A review. Materials Today: Proceedings.
segmentation, contour method is used for feature extraction, [11]. Tamilselvi, R., Nagaraj, A., Beham, M.P. and Sandhiya,
and neighborhood components analysis is used for feature M.B., 2020, February. BRAMSIT: A Database for Brain
selection. The experimental results are provided as accuracy Tumor Diagnosis and Detection. In 2020 Sixth
International Conference on Bio Signals, Images, and
(98%), recall (99%), precision (98%), and F1 score (98%).
Instrumentation (ICBSII) (pp. 1-5). IEEE.
The suggested approach has higher efficiency than the

13
IJFRCSCE | March 2022, Available @ http://www.ijfrcsce.org
 International Journal on Future Revolution in Computer Science & Communication Engineering (IJFRCSCE)
ISSN: 2454-4248 Volume: 8 Issue: 1
DOI: https://doi.org/10.17762/ijfrcsce.v8i1.2084
Article Received: 24 December 2021 Revised: 16 February 2022 Accepted: 22 February 2022 Publication: 31 March 2022
____________________________________________________________________________________________________________________
[12]. Kumar, K.A., Prasad, A.Y. and Metan, J., 2022. A hybrid
deep CNN-Cov-19-Res-Net Transfer learning archetype
for an enhanced Brain tumor Detection and Classification
scheme in medical image processing. Biomedical Signal
Processing and Control, 76, p.103631.
[13]. Stimpel, B., Syben, C., Schirrmacher, F., Hoelter, P.,
Dörfler, A. and Maier, A., 2019. Multi-modal deep guided
filtering for comprehensible medical image processing.
IEEE transactions on medical imaging, 39(5), pp.1703-
1711.
[14]. Verma, S. and Agrawal, R., 2021. Deep neural network in
medical image processing. In Handbook of Deep Learning
in Biomedical Engineering (pp. 271-292). Academic Press.
[15]. Palanisamy, T.S.C.A., Jayaraman, M., Vellingiri, K. and
Guo, Y., 2019. Optimization-based neutrosophic set for
medical image processing. In Neutrosophic set in medical
image analysis (pp. 189-206). Academic Press.

14
IJFRCSCE | March 2022, Available @ http://www.ijfrcsce.org