Research Article Ant Colony Optimization-Enabled CNN Deep Learning Technique For Accurate Detection of Cervical Cancer
Research Article Ant Colony Optimization-Enabled CNN Deep Learning Technique For Accurate Detection of Cervical Cancer
Research Article Ant Colony Optimization-Enabled CNN Deep Learning Technique For Accurate Detection of Cervical Cancer
Research Article
Ant Colony Optimization-Enabled CNN Deep Learning
Technique for Accurate Detection of Cervical Cancer
Received 7 August 2022; Revised 3 October 2022; Accepted 7 February 2023; Published 21 February 2023
Copyright © 2023 R. Kavitha et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Cancer is characterized by abnormal cell growth and proliferation, which are both diagnostic indicators of the disease. When
cancerous cells enter one organ, there is a risk that they may spread to adjacent tissues and eventually to other organs. Cancer
of the cervix of the uterus often initially manifests itself in the uterine cervix, which is located at the very bottom of the uterus.
Both the growth and death of cervical cells are characteristic features of this condition. False-negative results provide a
significant moral dilemma since they may cause women to get an incorrect diagnosis of cancer, which in turn can result in the
woman’s premature death from the disease. False-positive results do not raise any significant ethical concerns; but they do
require a patient to go through an expensive and time-consuming treatment process, and they also cause the patient to
experience tension and anxiety that is not warranted. In order to detect cervical cancer in its earliest stages in women, a
screening procedure known as a Pap test is often performed. This article describes a technique for improving images using
Brightness Preserving Dynamic Fuzzy Histogram Equalization. To individual components and find the right area of interest,
the fuzzy c-means approach is applied. The images are segmented using the fuzzy c-means method to find the right area of
interest. The feature selection algorithm is the ACO algorithm. Following that, categorization is carried out utilizing the CNN,
MLP, and ANN algorithms.
disease that has not yet spread to other regions of the body,
cervical cancer in its early stages does not produce any
symptoms. Early identification and treatment are both
completely curative and preventive when it comes to this Uterus
condition since it has a prolonged premalignant phase.
According to the National Cancer Institute in the United
States, cervical cancer is the fourth most frequent illness
found in women and is one of the most common causes of Cervix
cancer death in women worldwide. Cervical cancer is also
one of the most common causes of cancer mortality in males Vagina
[1]. The development of abnormal cells in the cervix’s lining
is known as cervical cancer. Squamous cell carcinoma is the Normal Cervical cancer
most typical type of cervical cancer. Cervical cancer symp-
toms include bleeding after intercourse, in between cycles,
or during menopause; watery, red, possibly thick, and foul-
smelling vaginal discharge; and pain in the pelvis or during
sexual activity. Numerous illnesses were linked to cervical
instances. The most common cervical illness among symp- Figure 1: Cervical cancer symptoms.
tomatic southern Ethiopian women was cervical carcinoma.
When compared to other cervical cases, the high prevalence During mass screening operations, a large number of sam-
of cervical cancer was seen in postmenopausal women. The ples need to be analyzed, which requires more cytotechni-
incidence of cervical cancer [2] and the mortality rate have cians with higher expertise as well as additional time. The
both decreased in industrialized countries because to the quality of medical photographs has significantly increased
increased use of frequent and mandated screening. On the because of the developments in medical imaging technology,
other hand, cancer is more prevalent in countries with low which has made it possible to diagnose sickness at an earlier
and moderate incomes, where resources are few and people stage [3, 4].
are not properly educated about the hazards posed by the ill- Screening may be performed by an automated diagnostic
ness. More women die from cancer in these countries. There tool without the introduction of observer bias if healthy
are a number of problems associated with automatic screen- samples are correctly classified as normal and those with
ing methods. Particles from the surfaces and surrounding sickness are correctly classified as abnormal. In the event
areas of the cervix are carefully extracted during a Pap smear that a patient is given an incorrect diagnosis during screen-
procedure so that they may be inspected under a microscope ing, they may be subjected to unnecessary follow-up and
for cervical cancer or cell changes that could lead to the dis- treatment even when they do not really have the condition.
ease. With a Pap smear, other issues like infections or If a test has a false-negative or false-positive result, the one
inflammatory disorders may be found. It is frequently done who has the ailment will interpret the result as normal,
simultaneously with a pelvic examination and may also be whereas the individual who does not have the illness would
done simultaneously with a test for particular types of interpret the result as abnormal. Negative smear tests were
human papillomavirus (HPV). These downsides include a examined to assess the diagnostic accuracy of recurrence.
decreased sensitivity, uncertainty surrounding cost effi- A total of 122 negative smear results were obtained from
ciency, and the inability to detect cases of early abnormali- women who did not have cervix disease, and 61 of them
ties. It is shown in Figure 1. came form 41 women whose negative smear results were
The question of whether or not to utilize an automated reported within 5.5 years of their CC diagnosis (used as con-
screening system as opposed to a manual screening method trol group). At review, it was discovered that every test from
has been hotly contested for decades, and the issue has not a woman without cervical cancer came back negative. On the
been addressed to anyone’s satisfaction. They look for early other hand, 27.1% of cancer cases from cytologists were
sickness detection, first before indications emerge. This has recoded as positive. As a direct consequence of this, patients
the advantage of allowing for much early medical interven- often report feeling of anxiety and depression [5, 6]. Figure 2
tion for the illness. A problem should only be treated early shows the different stages of cervical cancer that can be seen
if doing so leads to a better health outcome than delaying in the patients.
treatment. It has been passionately debated for decades Patients who have precancerous lesions will get treat-
whether to use an automated screening system instead of a ment as quickly as feasible utilizing computer-assisted
manual screening technique, and the problem has not been screening technologies. These approaches have the potential
resolved to everyone’s satisfaction. Pathologists are required to discover abnormal Pap smear results in patients. The suc-
to examine each subimage on a separate slide under a micro- cess of an automated screening product is limited by the
scope in the event of manually screening of Pap smear pic- high cost of the screening equipment, the price of screening
tures in order to make a medical diagnosis. In the case of activities, and the challenge of achieving high accuracy while
manual screening of Pap smear photos, pathologists are simultaneously minimizing the number of false-negative
required to inspect each subimage on a separate slide under samples produced. It is necessary to develop classification
a microscope in order to arrive at a diagnosis of sickness. strategies that are both effective and trustworthy if
BioMed Research International 3
CERVICAL CANCER
Uterus
Cancer
Cervix tissue
Cancer
tissue Bleeding
Vagina
Healty
cervix Cancer
tissue Bleeding
(viewed
from below)
automated systems are to be capable of correctly distinguish- intelligence.” Software applications are capable of predicting
ing between diseased and healthy samples. Screening sam- events more correctly without specific guidelines thanks to
ples may now be done more quickly, consistently, and machine learning (ML), a type of artificial intelligence (AI).
accurately than ever before with the use of machine learning Machine learning algorithms take previous information as
algorithms and medical image processing techniques, all at a input and estimate future accurate output. Medical experts
cheaper cost than traditional visual screening methods [7]. may quickly gain a new viewpoint thanks to the ML catego-
The precancer stages of cervical cancer are shown in rization of cancer. In reality, the use of machine learning
Figure 3. methods may broaden the tumor prognosis window. The
Image collection, noise reduction, segmentation of ML classification of cancer might provide medical profes-
regions of interest, feature extraction, and classification are sionals a different perspective in a very short amount of
among the major procedures that comprise a Pap smear time. The use of machine learning algorithms may, in fact,
screening system. The key steps is the Pap smear screening broaden the scope of cancer prediction [8].
technique. There may be further processes as well. There The classification power of ML is most beneficial in bio-
are several stages to the design process, and each one must logical applications when it is employed in combination with
be completed successfully for the classification and diagnosis genomic and proteomic data. Because of the growing
of illness to be done appropriately. False-negative test results amount of data available on cancer, machine learning is
are a significant ethical consideration since they can result in often used in the process of identifying and diagnosing can-
a wrong diagnosis of female cancer, which could cause the cer. The ML approach is reflective of the rising trend toward
patient to pass away from the condition. Untrue tests do personalized and predictive medicine in medicine generally.
not provide a severe ethical dilemma, but they do subject This article contains a methodology that uses Brightness
the patient to an expensive treatment method and undue Preserving Dynamic Fuzzy Histogram. For improving image
anxiety. Other steps may also be included. The process of contrast, fuzzy logic-based histogram equalization (FHE) is
design consists of multiple steps, and the successful comple- suggested. As opposed to conventional crisp histograms,
tion of each stage is essential for correctly classifying and fuzzy histograms are built using fuzzy set theory in order
diagnosing sickness. False-negative test results are a major to more effectively handle the ambiguity of grey level data.
ethical concern since they may lead to the incorrect diagno- The fuzzy distribution is divided into two subhistograms
sis of cancer in women, which can then result in the patient’s based on the median value of the original image in the sec-
death from the disease. False-positive results do not raise a ond stage, and each subhistogram is then separately bal-
serious ethical issue; but they do require a patient to undergo anced to preserve contrast enhancement. Two well-known
an expensive therapeutic process which causes the patient to metrics utilized to assess the both qualitative and quantita-
experience unnecessary worry. It is of the utmost impor- tive assessments of the proposed FHE approach are average
tance to develop a computer-aided diagnosis approach for information contents (AIC) and natural image quality eval-
cervical cancer treatment, in particular in developing coun- uator (NIQE) index for different pictures. Equalization for
tries and poorer nations, where the incidence and fatality image enhancement. Images are segmented to detect correct
rates of the disease are much greater than in wealthy nations. region of interest using the fuzzy c-means technique. In this
Researchers are increasingly turning to machine learning method, the author transformed the colour image to gray-
(ML), a research approach that is in the process of undergo- scale and applied a median filter to lessen the level of noise
ing fast development. Their goal is to improve cancer detec- and improve the representation of the image before using
tion and treatment. It falls under the umbrella of “artificial the fuzzy c-means clustering algorithm. Clusters and
4 BioMed Research International
classification are the third and fourth steps of this process, final image. Filtering is one component of image enhance-
respectfully. The centre value may be calculated thanks to ment, and its purposes include reducing distracting noise
the patch-based sorting. A data collection is divided into and bringing attention to certain aspects of an image.
N clusters using the fuzzy c-means (FCM) data clustering The authors Lu et al. [8] offered a comparative analysis
method, with each data point in the dataset to some extent of a variety of techniques for improving the quality of the
belonging to each cluster. A data point will have a high pictures. Approaches from the spatial domain, the frequency
degree of membership in a cluster, for instance, if it is domain, and the fuzzy domain were all used throughout the
near to the cluster’s center. ACO algorithm is used for fea- investigation. Approaches based on histograms and fuzzy
ture selection. Then, classification is performed using logic have both been shown to be effective. The author of
CNN, MLP, and ANN algorithms. Accuracy of ACO- the research believes that the evaluation of the fuzzy-based
CNN is highest among all classifiers used in this experi- enhancement’s K factor might perhaps be automated by
mental study. employing ant colony optimization to provide a more accu-
Both aberrant cell growth and reproduction, which are rate representation of the image.
both diagnostic indications of the illness, are characteristics According to the author, the Poisson noise is introduced
of cancer. The membrane of the uterine cervix, which is into cell images as a result of the noneven distribution of
found at the base of the uterus, is where cervical cancer first photons. According to the findings of the study, an adaptive
shows symptoms. Cell death and uncontrollable cervical cell Wiener filter has the potential to effectively minimize the
proliferation are symptoms of the illness. The fourth most Poisson noise. Even if the picture qualities change from
prevalent disease in women and one of the leading causes one location to another, it is still feasible to denoise photo-
of cancer-related deaths in women globally is cervical graphs by using an adaptive Wiener filter [9].
cancer. Bihistogram equalization is a method that was created by
The remainder of this paper is organized as follows: Sec- Tang and Isa [10] with the purpose of increasing the quality
tion 2 covers the literature survey. Section 3 covers the meth- of grayscale pictures. The input histogram is used to gener-
odologies, and the results are analyzed in Section 4. Section 5 ate not one but two subhistograms. The picture’s oversatura-
contains concluding remarks as well as future scope of the tion was brought under control by clipping the histogram,
research. and then, the output that was generated was equalized and
mixed to make the finished product. In general, the perfor-
2. Literature Survey mance of the authors’ technique is superior to that of exist-
ing histogram-based improvement algorithms, as shown by
The augmentation of a photograph is done with the inten- a comparison with those algorithms.
tion of calling attention to certain aspects in order to facili- It was found by Kumbhar et al. [11] that the main attri-
tate future research. A machine learning-based programme butes of an image may be retrieved for enhanced diagnosis if
called automated visual assessment (AVE) scans digital the features of medial photos are increased. The images were
images of the cervix for indications of cancer or precancer. run through filters before any sharpening or noise reduction
An AVE-positive result indicates the existence of troubling methods were applied to them. In addition to that, an effect
lesions that are either suggestive of cancer or enhance the called “Adaptive Histogram Equalization and Average Fil-
possibility that cancer will develop in the near future. An ter” was added to the photo in order to make it seem better.
AVE-negative outcome, on the other hand, indicates a cervix In the Herlev dataset, Chankong et al. [12] segregated
that is not at higher risk for cancer. The methodologies may the nucleus from the cytoplasm by using a patch-based ver-
be broken down into two primary categories: the spatial sion of the fuzzy c-means clustering algorithm. In this
domain and the frequency domain. The values of a picture’s method, the author converted the colour image to grayscale
pixels are inputted directly into mathematical calculations and used a median filter to reduce the amount of noise and
that are performed in the spatial domain. The picture is smooth out the appearance of the picture. The third and
transformed using a Fourier transformation in the frequency fourth processes in this process are called clustering and seg-
domain first, and then, the inverse of the image that was mentation, respectively. The patch-based sorting allowed for
converted using the Fourier transform is used to create the the calculation of the central value.
BioMed Research International 5
According to Sharma and Mangat [13], an additional work is able to capture the nonlinear link that exists between
improvement was made to the work, and the author intricate patterns and make accurate predictions. For the
improved the “fuzzy c-means (FCM)” clustering technique purposes of picture recognition and analysis, deep neural
by working with various numbers of clusters rather of just network topologies such as the convolution neural network
one. This was done in order to reinforce the clustering of (CNN) are often used [18]. A system will automatically learn
the data. The author recommended combining techniques features at many different levels of abstraction in order for it
to segmentation with methods for locating areas of interest to be able to learn every conceivable characteristic. This is
in order to improve the accuracy of the segmentation pro- necessary for the system to be able to learn everything.
cess. There is a possibility that the present accuracy of the The precision of the cell segmentation is a crucial com-
Herlev dataset may be improved using a variety of feature ponent in the overall success of the traditional machine
variations, improved noise reduction techniques, and seg- learning approaches. Taha et al. [19] proposed a technique
mentation methods (93.7 percent). for categorizing cells without the need for segmentation.
Saha et al. [14] introduced a circularly shaped function Their approach included the use of a deep feature learning
that put a restriction on the form of the cluster in order to convolution neural network. They were successful in assign-
enhance the border of the nucleus. This was done in order ing categories to the Herlev dataset with a rate of accuracy of
to make the nucleus seem more complete. 98 percent, which is an admirable achievement.
A categorization system that is based on texture and was Hyeon et al. [20] trained a model that can differentiate
created by Mariarputham and Stephen [15] is used to cate- between healthy and sick cervical cells by using a convolu-
gorize smear photographs into a total of seven separate tional neural network (CNN) and retrieving feature vectors
groups. In terms of the accuracy of classification, it was from images of cervical cells. These collected characteristics
shown that SVM had a precision of 97.38 percent for normal were trained with the use of an SVM classifier, which
squamous, 93.89 percent for intermediate squamous, 87.33 resulted in a success rate of 78 percent overall. The study
percent for mild dysplasia, and 58.52 percent for severe dys- presented here proposes a classification model for prostate
plasia. The RBF, linear, and quadratic SVM kernels were cancer that makes use of deep learning methods and
used in the classification process. achieves an accuracy of 80.1% on training sets and 78.1%
A wide range of neural network architectures were put to on testing sets, respectively.
the test by Devi et al. [16] for the purpose of disease diagno- Devi et al. [16] investigated a variety of neural network
sis. ANN designs such as the multilayered perceptron may architectures, convolutional neural network architecture,
be helpful when it comes to the acceleration of the detection and feedforward network architecture. According to the
process. Both a feedforward network and a knowledge-based results, the performance of ANN might be enhanced by
neural network were utilized in order to map the input pic- including a learning capacity that assists in the improvement
tures with the rules and extract the features that are neces- of its efficiency through the evaluation of exceptional
sary for classification. As a result, the classification results performance.
exhibited by the network were superior, and it had a decent
accuracy rate. The results of the classification carried out 3. Methodology
using the artificial neural network (ANN) method are better
and have a high rate of accuracy. This article contains a methodology that uses Brightness
Athinarayanan et al. [17] developed a categorization sys- Preserving Dynamic Fuzzy Histogram Equalization for
tem for cervical disease by analyzing images obtained from image enhancement. Images are segmented to detect correct
Pap smears. In order to improve the rough set text on cooc- region of interest using the fuzzy c-means technique. ACO
currence matrix, we made use of the ERSTCM and CABS algorithm is used for feature selection. Then, classification
descriptors as well as the concatenated feature extraction is performed using CNN, MLP, and ANN algorithms. Accu-
approach (CFE). Using a classifier, it is possible to measure racy of ACO-CNN is highest among all classifiers used in
the performance of the extracted features by comparing this experimental study. Interstrategies modelled after ants
them to statistical criteria such as sensitivity, specificity, are referred to as “ants.” Biological ants’ pheromone-based
and accuracy. This can be done with the help of a classifica- communication is frequently the dominating paradigm.
tion algorithm (FL-HKSVM). The performance of the For many optimization projects requiring some kind of
concatenated feature extraction method was superior than graph, local search algorithms combined with artificial ants
that of the other two classifiers. Feature extraction is the pro- have emerged as the preferred approach. This framework is
cess of converting raw data into manageable quantitative shown in Figure 4.
properties while conserving the original dataset’s character- By using the technique for fuzzy histogram equalization,
istics. It yields superior results when compared to utilizing the method known as “Brightness Preserving Dynamic
machine learning on the raw data directly. Since CNNs can Fuzzy Histogram Equalization” has been updated in order
automatically generate features from time series information to increase both its brightness and its contrast. It was chosen
and frequency represented images, they are most frequently to first use a Gaussian kernel to smooth out the image histo-
employed in healthcare applications. After that, a classifier gram and then to segment the troughs for dynamic equaliza-
network uses these features to do classification and regres- tion. This was done after the decision was made. Histograms
sion. A neural network that has hundreds of hidden layers are processed using crisp histograms so that the contrast
is referred to as a deep learning network. This kind of net- may be made better. Brightness Preserving Dynamic Fuzzy
6 BioMed Research International
97
96
95
94
93
92
91
90
89
88
87
ACO CNN ACO MLP ACO ANN
Accuracy %
Sensitivity %
Specificity %
Figure 5: Accuracy, sensitivity, and specificity of machine learning techniques for diagnosis of cervical cancer on the Herlev dataset.
It was Professors Bernard Widrow and Ted Hoff, back in c-means technique. ACO algorithm is used for feature selec-
1960, who created a mathematical model. Information may tion. Then, classification is performed using CNN, MLP, and
be processed in an analogous manner to the way the human ANN algorithms. Accuracy of ACO-CNN is highest among
brain works. Using this strategy, researchers were able to all classifiers used in this experimental study. Various
classify and forecast chronic illnesses like breast cancer. authors have investigated similar studies in their results.
ANN [25] is a powerful tool for revealing BC’s simple con- The result trend of the current study is similar to the find-
nections. An ML process was used to create and construct ings of various researchers. A deep learning method for cer-
this algorithm for a specific purpose, such as pattern vical cancer detection using Pap smear images is proposed
cataloguing. Edges of the ANN model refer to the connec- by Taha et al. [19]. They use the Herlev public database for
tions made between layers. Input, intermediate, and output single cell Pap smears to demonstrate the effectiveness of
layers are formed by the arrangement of neurons. Edges link this novel approach, and the results of the experiments show
the neurons, and each side has a vertex, which is termed that our suggested system works noticeably better than other
“wait” in this context. The underlying concept of ANN is cutting edge techniques.
that of a network of neurons.
An MLP algorithm is a kind of machine learning algo- 5. Conclusion
rithm that recreates the decision-making and learning pro-
cesses of a natural artificial neural network (ANN) by Since abnormal cell growth and multiplication are diagnostic
using several fully connected layers. MLP used three differ- signs of cancer, this condition is regarded as its distinguish-
ent layers for its nodes: (A) the initial layer, also known as ing feature. Once malignant cells have invaded an organ, it is
the input layer; (B) the intermediate layer, also known as possible that they will spread to the tissues nearby before
the hidden layer; and (C) the output layer. The back propa- eventually reaching other organs. The uterus cervix, which
gation method is used in order to affect an update to the is located at the very bottom of the uterus, is frequently
weight of the MLP. MLP allows for the recognition of non- where uterine cervical cancer first manifests itself. Both the
linear datasets that are separable [26]. growth and death of cervical cells are hallmarks of this con-
dition. Both of these procedures are signs. False-negative
4. Result Analysis results pose a huge ethical problem since they may result
in women being given the wrong cancer diagnosis, which
The Herlev dataset consists of 242 normal cells, 650 images may cause the woman to pass away from the disease too
are used to train the model, and the remaining 267 images soon. In this sense, erroneous negative outcomes might be
were used to test the model [27]. A model’s performance to blame for the early demise of females. False-positive
can be gauged by its sensitivity and specificity. In contrast results do not lead to any significant ethical issues, but they
to specificity, which measures how well the model predicts do make a patient undergo an expensive and time-
real negatives, sensitivity measures how well the model pre- consuming treatment process, as well as put them through
dicts true positives. The results are shown in Figure 5. unneeded stress and concern. Women are frequently sub-
Brightness Preserving Dynamic Fuzzy Histogram Equal- jected to a monitoring procedure called a Pap test in an
ization is used for image enhancement. Images are seg- effort to find cervical cancer in its earliest stages, when it is
mented to detect correct region of interest using the fuzzy most curable. By using the procedures described in this
8 BioMed Research International
article, which make use of a method called Brightness Pre- tional Journal of Engineering Research and Technology, vol. 2,
serving Dynamic Fuzzy Histogram Equalization, image no. 4, pp. 2359–2364, 2013.
quality can be enhanced. Images are segmented using the [12] T. Chankong, N. Theera-Umpon, and S. Auephanwiriyakul,
fuzzy c-means technique to determine the proper area of “Automatic cervical cell segmentation and classification in
interest. The ACO algorithm is used to select the character- Pap smears,” Computer Methods and Programs in Biomedi-
istics. Then, classification is performed with the aid of the cine, vol. 113, no. 2, pp. 539–556, 2014.
CNN, MLP, and ANN algorithms. Of all the classifiers used [13] B. Sharma and K. K. Mangat, “Various techniques for classifi-
in this experimental inquiry, the ACO-CNN classifier had cation and segmentation of cervical cell images-a review,”
the highest accuracy. International Journal of Computer Applications, vol. 147,
no. 9, pp. 16–20, 2016.
[14] R. Saha, M. Bajger, and G. Lee, “Spatial shape constrained
Data Availability fuzzy c-means (fcm) clustering for nucleus segmentation in
Pap smear images,” in 2016 International conference on digital
The corresponding author will provide data on request. image computing: techniques and applications (DICTA), pp. 1–
8, Gold Coast, QLD, Australia, 2016.
Conflicts of Interest [15] E. J. Mariarputham and A. Stephen, “Nominated texture based
cervical cancer classification,” Computational and Mathemat-
There is no conflict of interest. ical Methods in Medicine, vol. 2015, Article ID 586928, 10
pages, 2015.
[16] M. A. Devi, S. Ravi, J. Vaishnavi, and S. Punitha, “Classifica-
References tion of cervical cancer using artificial neural networks,” Proce-
dia Computer Science, vol. 89, pp. 465–472, 2016.
[1] R. Shafabakhsh, R. J. Reiter, H. Mirzaei, S. N. Teymoordash,
and Z. Asemi, “Melatonin: a new inhibitor agent for cervical [17] S. Athinarayanan, M. Srinath, and R. Kavitha, “Computer
cancer treatment,” Journal of Cellular Physiology, vol. 234, aided diagnosis for detection and stage identification of cer-
no. 12, pp. 21670–21682, 2019. vical cancer by using Pap smear screening test images,”
[2] US Preventive Services Task Force, S. J. Curry, A. H. Krist ICTACT Journal on Image & Video Processing, vol. 6,
et al., “Screening for cervical cancer,” Journal of the American no. 4, 2016.
Medical Association, vol. 320, no. 7, pp. 674–686, 2018. [18] L. Zhang, L. Lu, I. Nogues, R. M. Summers, S. Liu, and J. Yao,
[3] S. Pankaj, S. Nazneen, S. Kumari et al., “Comparison of con- “DeepPap: deep convolutional networks for cervical cell classi-
ventional Pap smear and liquid-based cytology: a study of cer- fication,” IEEE Journal of Biomedical and Health Informatics,
vical cancer screening at a tertiary care center in Bihar,” Indian vol. 21, no. 6, pp. 1633–1643, 2017.
Journal of Cancer, vol. 55, no. 1, pp. 80–83, 2018. [19] B. Taha, J. Dias, and N. Werghi, “Classification of cervical-
[4] A. S. Zamani, L. Anand, K. P. Rane et al., “Performance of cancer using Pap-smear images: a convolutional neural
machine learning and image processing in plant leaf disease network approach,” in Annual Conference on Medical
detection,” Journal of Food Quality, vol. 2022, Article ID Image Understanding and Analysis, pp. 261–272, Cham,
1598796, 7 pages, 2022. 2017.
[5] N. Dong, L. Zhao, C. H. Wu, and J. F. Chang, “Inception v3 [20] J. Hyeon, H.-J. Choi, K. N. Lee, and B. D. Lee, “Automating
based cervical cell classification combined with artificially papanicolaou test using deep convolutional activation feature,”
extracted features,” Applied Soft Computing, vol. 93, article in 2017 18th IEEE International Conference on Mobile Data
106311, 2020. Management (MDM), pp. 382–385, Daejeon, Korea (South),,
[6] R. Veluri, I. Patra, M. Naved et al., “Learning analytics using 2017.
deep learning techniques for efficiently managing educational [21] D. Sheet, H. Garud, A. Suveer, M. Mahadevappa, and
institutes,” Materials Today: Proceedings, vol. 51, pp. 2317– J. Chatterjee, “Brightness preserving dynamic fuzzy histogram
2320, 2022. equalization,” IEEE Transactions on Consumer Electronics,
[7] T. Zhang, Y. M. Luo, P. Li et al., “Cervical precancerous lesions vol. 56, no. 4, pp. 2475–2480, 2010.
classification using pre-trained densely connected convolu- [22] J. R. Tang, N. A. Mat Isa, and E. S. Ch’ng, “A fuzzy-c-means-
tional networks with colposcopy images,” Biomedical Signal clustering approach: quantifying chromatin pattern of non-
Processing and Control, vol. 55, article 101566, 2020. neoplastic cervical squamous cells,” PLoS One, vol. 10,
[8] J. Lu, E. Song, A. Ghoneim, and M. Alrashoud, “Machine no. 11, article e0142830, 2015.
learning for assisting cervical cancer diagnosis: an ensemble [23] S. Singh and R. R. Janghel, “Early diagnosis of Alzheimer’s dis-
approach,” Future Generation Computer Systems, vol. 106, ease using ACO optimized deep CNN classifier,” in Ubiquitous
pp. 199–205, 2020. Intelligent Systems. Smart Innovation, Systems and Technolo-
[9] T. Deepa and A. N. Rao, “A study on denoising of poisson gies, P. Karuppusamy, I. Perikos, and F. P. García Márquez,
noise in Pap smear microscopic image,” Indian Journal of Sci- Eds., vol. 243, Springer, Singapore, 2022.
ence and Technology, vol. 9, no. 45, 2016. [24] A. Ghoneim, G. Muhammad, and M. S. Hossain, “Cervical
[10] J. R. Tang and N. A. M. Isa, “Adaptive image enhancement cancer classification using convolutional neural networks and
based on bi-histogram equalization with a clipping limit,” extreme learning machines,” Future Generation Computer Sys-
Computers & Electrical Engineering, vol. 40, no. 8, pp. 86– tems, vol. 102, pp. 643–649, 2020.
103, 2014. [25] J. Zhang, C. Li, Y. Yin, J. Zhang, and M. Grzegorzek, “Applica-
[11] U. Kumbhar, V. Patil, and S. Rudrakshi, “Enhancement of tions of artificial neural networks in microorganism image
medical images using image processing in matlab,” Interna- analysis: a comprehensive review from conventional
BioMed Research International 9