diagnostics

Review
Pap Smear Images Classification Using Machine Learning:
A Literature Matrix
Nur Ain Alias 1 , Wan Azani Mustafa 1,2, * , Mohd Aminudin Jamlos 3 , Hiam Alquran 4 , Hafizul Fahri Hanafi 5 ,
Shahrina Ismail 6 and Khairul Shakir Ab Rahman 7

1 Faculty of Electrical Engineering & Technology, Universiti Malaysia Perlis, UniCITI Alam Campus,
Sungai Chuchuh, Padang Besar 02100, Perlis, Malaysia
2 Advanced Computing (AdvCOMP), Centre of Excellence, Universiti Malaysia Perlis (UniMAP),
Pauh Putra Campus, Arau 02600, Perlis, Malaysia
3 Faculty of Electronic Engineering & Technology, Universiti Malaysia Perlis, UniCITI Alam Campus,
Sungai Chuchuh, Padang Besar 02100, Perlis, Malaysia
4 Department of Biomedical Systems and Informatics Engineering, Yarmouk University, Irbid 21163, Jordan
5 Department of Computing, Faculty of Art, Computing and Creative Industry, Universiti Pendidikan Sultan
Idris, Tanjong Malim 35900, Perak, Malaysia
6 Faculty of Science and Technology, Universiti Sains Islam Malaysia (USIM),
Bandar Baru Nilai 71800, Negeri Sembilan, Malaysia
7 Department of Pathology, Hospital Tuanku Fauziah, Kangar 02000, Perlis, Malaysia
* Correspondence: [email protected]

Abstract: Cervical cancer is regularly diagnosed in women all over the world. This cancer is the sev-
enth most frequent cancer globally and the fourth most prevalent cancer among women. Automated
and higher accuracy of cervical cancer classification methods are needed for the early diagnosis of
cancer. In addition, this study has proved that routine Pap smears could enhance clinical outcomes
by facilitating the early diagnosis of cervical cancer. Liquid-based cytology (LBC)/Pap smears for
Citation: Alias, N.A.; Mustafa, W.A.;
Jamlos, M.A.; Alquran, H.; Hanafi,
advanced cervical screening is a highly effective precancerous cell detection technology based on
H.F.; Ismail, S.; Rahman, K.S.A. Pap cell image analysis, where cells are classed as normal or abnormal. Computer-aided systems in
Smear Images Classification Using medical imaging have benefited greatly from extraordinary developments in artificial intelligence
Machine Learning: A Literature (AI) technology. However, resource and computational cost constraints prevent the widespread use
Matrix. Diagnostics 2022, 12, 2900. of AI-based automation-assisted cervical cancer screening systems. Hence, this paper reviewed the
https://doi.org/10.3390/ related studies that have been done by previous researchers related to the automation of cervical
diagnostics12122900 cancer classification based on machine learning. The objective of this study is to systematically
Academic Editors: Christoph Palm review and analyses the current research on the classification of the cervical using machine learning.
and Costin Teodor Streba The literature that has been reviewed is indexed by Scopus and Web of Science. As a result, for
the published paper access until October 2022, this study assessed past approaches for cervical cell
Received: 13 October 2022
classification based on machine learning applications.
Accepted: 19 November 2022
Published: 22 November 2022
Keywords: cervical cancer; cell classification; review; SLR
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations. 1. Introduction
Cervical cancer is a major disease that seriously threatens women’s health [1,2]. This
cancer is also known as the second most commonly affected and killer type of cancer among
Copyright: © 2022 by the authors.
women around the world [3]. It results from a chronic infection of the skin and mucosal
Licensee MDPI, Basel, Switzerland. cells in women’s vaginal regions. The fact that this cancer does not manifest any signs
This article is an open access article when it first appears is the most alarming feature of it [4]. In a report by Elakkiya et al. [5],
distributed under the terms and it has been mentioned that this type of cancer is curable with early detection in the early
conditions of the Creative Commons stage. Unfortunately, the mortality rate is getting higher among women around the world
Attribution (CC BY) license (https:// that have been affected by this cancer [6–8]. The traditional method of manual inspection,
creativecommons.org/licenses/by/ also known as a Pap smear examination, is inaccurate due to human error that may lead to
4.0/). a false patient diagnosis [9,10]. The technology for automated cervical cancer screening

Diagnostics 2022, 12, 2900. https://doi.org/10.3390/diagnostics12122900 https://www.mdpi.com/journal/diagnostics

Diagnostics 2022, 12, 2900 2 of 16

is indeed very significant for lessening the risk of cervical cancer. However, the existing
approach using machine learning has drawbacks, including poor generalisation capacity in
complicated situations, as well as low efficiency, accuracy, and generalisation ability [11].
Several studies have attempted to investigate the ability of machine learning to classify
cervical cancer cells for the purpose of enhancing manual screening [10,12,13]. The most
often used approach for predicting characteristics from a high-dimensional collection of
cancer imaging data is the random forest approach [14–16]. However, if a large number
of decision trees are utilised, the random forest approach might become too sluggish and
ineffective for real-time predictions [10]. In addition, current classification approaches,
such as deep learning (DL) or hand-crafted techniques, mostly rely on single detection
structures and have high processing complexity and low accuracy [17].
Cervical cytopathology image classification is an important method for diagnosing cer-
vical cancer [18]. Moreover, previous studies have mentioned that cervical cell classification
has important clinical consequences in cervical cancer screening at an early stage [19,20].
The effective classification of Pap smear cell images may be used to create automated and
precise cervical cancer classification systems for early diagnosis [9]. To aid in the earlier
detection and diagnosis procedure of cervical cancer, the proper screening of Pap smear
images is vital. In a study by Janiesch et al. [21], DL has had a huge impact on several
sectors of science in recent years. It has resulted in considerable advances in speech recogni-
tion [22–24] and image recognition [25,26]. Prior to the development of DL, many of these
tasks were regarded as being beyond the capabilities of computers, even in science fiction
literature. However, DL methods are proposed to be able to compensate for the problem
through computer-aided systems for cancer cell classification [8,12,13,17,24,26–30]. Hence,
this study’s objective is to review the current development in technologies for cervical cell
classification using machine learning.

2. Review Method
The use of modern technology in medical research has increased as a result of its
advancement [31], particularly in the fields of image processing, drug discovery, computer-
aided diagnosis, and cancer research [32]. A growing number of academics have adopted
DL, which is known as the most popular machine learning technique for processing medical
images. The medical community sees a bright future for disease prediction through
machine learning [33]. There are numerous methods for cervical cell classification, in
which scientists have classified cervical cancer using these methods. In recent studies, the
convolutional neural network is one of the methods used to classify cell cancer through a DL
process [11,18,26–28,34]. Apart from that, many researchers have attempted to automate the
detection of cervical cancer cells using other approaches of machine learning by applying
a classifier, clustering algorithm, Random Forest, Ada Boost, MLP Algorithm, feature
extraction network, DeepCyto, Support Vector Machine (SVM), and others. The findings of
all the proposed methods above are able to detect cancer cells with their methodologies
in classifying the characteristics of cervical cancer cells. Several methods are able to
achieve high-performance evaluation values in terms of accuracy, specificity, sensitivity,
precision, and f-measure. Thus, this has shown that machine learning can compensate for
the traditional ways of diagnosing cancer and likely detect cancer in the early stage through
a screening process. A comprehensive review based on advanced searching related to the
classification of cervical cancer using machine learning is the main objective of this paper.
Comprehensive review approaches are about to become the “new normal” in organiz-
ing research reviews [35]. Advanced evaluation is one of the most significant discussions
currently taking place globally. The review technique entails three major steps in choosing
numerous relevant papers for this study. The identification step is the first step in writing
a comprehensive literature review, which includes the search for research items that may
be relevant to the predetermined research question. Next, screening is done to select the
inclusion and exclusion from the total searched papers. Finally, the third stage is to deter-
mine the eligibility of the paper by reviewing the abstract to identify the relevant topic and
Diagnostics 2022, 12, 2900 3 of 16

subtopic of the screened papers. The scientific literature is then reviewed and summarised
in an effort to discover, choose, and evaluate the key research that has contributed to
the classification of cervical cancer cells. Last, but not least, the aim of this paper is to
provide suggestions for more research in response to the issues raised in this work. In this
study, a comprehensive literature review is conducted using the specific method, which is
a recognised best practise. Essentially, the purpose of publication rules is to aid authors in
assessing the accuracy of a review by supplying pertinent and necessary information. A
comprehensive review also draws attention to the randomised investigations assessments
survey, which could be a key component of systematic analysis reports for various sorts
of studies. Due to their reliability, the Web of Science and Scopus databases were used to
analyse the research’s methodology. This section also covers the identification, screening,
eligibility, and data abstraction, as discussed in the four main subsections.

2.1. Preliminary Identification

The identification phase involves searching for study materials relevant to the prede-
termined research issue of classification of a cervical cancer cell. The keywords used are
‘classification of cervical cell’. Therefore, the first step was to detect keywords and search for
similar, equivalent phrases in dictionaries, thesauri, encyclopaedias, and previous research.
As a result, after determining all relevant phrases, search strings for the Web of Science and
Scopus databases were created (see Table 1). Thus, during the first part of the advanced
searching procedure, this study effectively obtained 3048 publications from the databases.

Table 1. The search strings.

Scopus Web of Science

Classification of cervical cancer cell (Abstract)
TITLE-ABS-KEY (classification AND cervical
or classification of cervical cancer cell (Title) or
AND cancer AND cell) AND (LIMIT-TO
classification of cervical cancer cell (Author
(SRCTYPE, “j”)) AND (LIMIT-TO (PUBSTAGE,
Key-words) and 2022 (Publication Years) and
“final”)) AND (LIMIT-TO (DOCTYPE, “ar”))
Article (Document Types) and English
AND (EXCLUDE (SUBJAREA, “MEDI”)) AND
(Languages) and Oncology or Engineering
(LIMIT-TO (PUBYEAR, 2022)) AND (LIMIT-TO
Biomedical or Engineering Electrical Electronic
(LANGUAGE, “English”))
(Web of Science Categories).
Date of access: October 2022
Date of access: October 2022

2.2. Screening
The collection of possibly relevant research items is examined for content that matches
the predefined research question(s) during the screening step. Content-related criteria that
are frequently used in the screening phase include the selection of research items based on
the classification of cervical cancer cells using machine learning. In this step, all duplicate
papers will be removed from the list of searched papers. The first stage of the screening
excluded 2999 publications, while the second stage examined 49 papers based on different
exclusion and inclusion criteria of this study (see Table 2). The literature (research papers)
was the first criterion utilised because it is the primary source of practical recommendations.
It also includes reviews, meta-synthesis, meta-analyses, books, book series, chapters, and
conference proceedings that were not included in the most recent study. Furthermore, the
review was confined to publications in English. It is vital to remember that the strategy
only focused on the year 2022. In all, four publications were rejected based on duplication
criteria.
Diagnostics 2022, 12, x FOR PEER REVIEW 4 of 18

remember that the strategy only focused on the year 2022. In all, four publications were
Diagnostics 2022, 12, 2900 4 of 16
rejected based on duplication criteria.

Table 2. The selection criterion for searching.

Table 2. The selection criterion for searching.
Criterion Inclusion Exclusion
Criterion
Language Inclusion
English Exclusion
Non-English
Language
Published Year English
2022 Non-English
<2022
Published Year Journal
2022 (only research <2022
Sources type Conference proceeding
Sources type articles)
Journal (only research articles) Conference proceeding
Document Type
Document Type
Article
Article
Letter, Review, Conference, Note
Letter, Review, Conference, Note
Computer Science and Besides Computer Science and Engi-
Research Area Computer Science and Besides Computer Science and
Research Area Engineering neering
Engineering Engineering

2.3. Eligibility
2.3. Eligibility
The final review sample is generated after all inclusion and exclusion criteria have
The final review sample is generated after all inclusion and exclusion criteria have
been met. A thorough disclosure of the full list of research items included in this sample
been met. A thorough disclosure of the full list of research items included in this sample is
is required, since readers will not know which research items exactly form the foundation
required, since readers will not know which research items exactly form the foundation for
for the review’s study results otherwise. The third level, termed eligibility, includes 45
the review’s study results otherwise. The third level, termed eligibility, includes 45 articles
articles in total. At this point, all article titles and significant content were carefully exam-
in total. At this point, all article titles and significant content were carefully examined to
ined to ensure
ensure that thethat the inclusion
inclusion criteriacriteria
were metwere
andmet
thatand
the that thewere
articles articles were to
relevant relevant to
the present
thestudy’s
presentresearch
study’s aims.
research aims. As a consequence, 18 publications were excluded,
As a consequence, 18 publications were excluded, since their titlesince
their
andtitle and abstract
abstract were notwere not significantly
significantly related
related to to thepurpose
the study’s study’s purpose
based onbased on em-
empirical data.
pirical data. Finally, 27 papers were made available for evaluation
Finally, 27 papers were made available for evaluation (see Figure 1). (see Figure 1).

(a) (b)
Figure
Figure1. (a) Autonomy
1. (a) of women’s
Autonomy cervix.
of women’s (b) Cervical
cervix. cell cytology
(b) Cervical imagesimages
cell cytology of Herlev datasetdataset
of Herlev Cer-
vical [36].
Cervical [36].

2.4.
2.4. Data
Data Abstraction
Abstraction and
and Analysis
Analysis
AnAn integrative
integrative analysis
analysis was wasusedused as one
as one of the
of the assessment
assessment strategies
strategies in study
in this this study
to
to examine
examine and synthesise
and synthesise a variety
a variety of research
of research designsdesigns (quantitative,
(quantitative, qualitative,
qualitative, and mixed and
mixed methods).
methods). The goal of The
thegoal of the competent
competent study was to study was relevant
identify to identify relevant
topics topics and
and subtopics.
subtopics.
The stage of Thedatastage of data
collection wascollection
the firstwasstepthe
in first step in the development
the development of the theme. of Figure
the theme.
2
Figure 2 shows how the authors meticulously analysed a compilation
shows how the authors meticulously analysed a compilation of 27 publications for asser- of 27 publications
for or
tions assertions
materialor material
relevant to relevant
the topics toofthe
thetopics
currentof the current
study. study. The
The authors thenauthors
evaluatedthen
evaluated
the the currentstudies
current significant significant
relatedstudies related
to cervical to cervical
cancer cancer cell classification.
cell classification. The methodol- The
methodology
ogy used in all used in all
studies, asstudies,
well asasthe well as the research
research results, results,
are beingareinvestigated.
being investigated. Next,
Next, the
the author
author collaborated
collaborated withco-authors
with other other co-authors to develop
to develop themes themes
based onbased on the evidence
the evidence in this
in this study’s context. A log was kept throughout the data
study’s context. A log was kept throughout the data analysis process to record analysis process toany
record
anal-any
analyses, viewpoints, riddles, or other thoughts relevant to the data interpretation.
yses, viewpoints, riddles, or other thoughts relevant to the data interpretation. Finally, the Finally,
the authors
authors compared
compared the results
the results to seetoifsee if there
there werewereanyany inconsistencies
inconsistencies in the
in the theme
theme design
design
process.
process. It It
is is worth
worth noting
noting that,
that, if if there
there are
are any
any disagreements
disagreements between
between the
the concepts,
concepts, thethe
authors discuss them amongst themselves. The produced themes were eventually tweaked
to ensure consistency. The analysis selection was carried out by two experts, one in public
health (Khairul Shakir Ab Rahman—expert medical doctor in pathology) and the other in
biomedical science (Wan Azani Mustafa—expert in biomedical computing), to determine
 Diagnostics 2022, 12, x FOR PEER REVIEW 5 of 18

Diagnostics 2022, 12, 2900 authors discuss them amongst themselves. The produced themes were eventually 5 of 16
tweaked to ensure consistency. The analysis selection was carried out by two experts, one
in public health (Khairul Shakir Ab Rahman—expert medical doctor in pathology) and
the other in biomedical science (Wan Azani Mustafa—expert in biomedical computing),
the validity ofthe
to determine thevalidity
problems. Theproblems.
of the expert review phase ensures
The expert the clarity,
review phase ensuresimportance,
the clarity,and
suitability of each subtheme by establishing the domain validity.
importance, and suitability of each subtheme by establishing the domain validity.

Search Keywords
‘Classification of Cervical Cancer
Cell’
Preliminary identification

Preliminary search
Identify the related paper
based on the keyword search
(n=3048)
Scopus Web of Sciences
Record of preliminary Record of preliminary
search (n=2827) search (n=221)

Inclusion and exclusion Inclusion and exclusion

criteria screening criteria :
Scopus (n=30) & WoS (n=19) • Language
Total = 49 • Published Year
• Sources Type
• Document Type
Screening

• Research Area
Records excluded
(n=2999)

Duplicate screening record

removal (n=4)

Exclusion of full text

Title and abstract screening criteria:
for eligibility (n=45) • Out of field
Eligibility

• Title not related

• Abstract not
significant
• Objective defer
Records excluded
(n=18)
Full text

Record included in the

qualitative analysis (n=27)

Figure 2. Flow diagram of the proposed advanced search study.

3. Results and Findings

Figure 2. Flow diagram of the proposed advanced search study.
The cancer disease’s significance has increased, as public health worries about the
region’s development and success. Microscopic image-based analysis has been extensively
used in pathological research and disease diagnosis. However, the misauthentication
of cell lines due to pathologists’ errors has been identified as a severe issue. Therefore,
a comparative evaluation of the proposed model was conducted to illustrate the utility
of feature selection and class imbalance based on the classifier’s accuracy, sensitivity,
precision, F-measure, and specificity. The goals of this study were to improve the efficiency
and accuracy of an early cervical cancer clinical diagnosis and evaluate the application of a
cell classification algorithm in conjunction with multispectral imaging in cervical cancer
Diagnostics 2022, 12, 2900 6 of 16

screening. Therefore, researchers have come out with various approaches to overcome the
shortcomings of the previously proposed approach for classification. The convolutional
neural network is one of the famous methods for the classification of infected cells. In this
study, a total of 27 articles were extracted and analysed based on the advanced searching.

Classification of Cells Based on Machine Learning Approach

There are numerous approaches that have been done by previous researchers in the
area of cell classification. The methodology and results of the approaches are summarised to
make it easier to compare the findings of the studies. Table 3 illustrates the summary of the
prevailing works related to the classification of cervical cancer cells using machine learning.

Table 3. Summary of the prevailing works.

Title Class and Database Methodology Results

CVM-Cervix: A Database: CRIC
hybrid cervical dataset, SIPaKMeD
Pap smear image dataset and Framework CVM-Cervix Effective and
classification combination of based on deep learning. potential of the
framework using CRIC and proposed
CNN, visual SIPaKMeD Type: Machine CVM-Cervix
transformer, and learning—Neural network proved.
multilayer Class: CRIC-6 class,
perceptron [18] SIPaKMeD 5 Class
All models’
A Comparative accuracy >905.
Analysis of Deep The individual
An ensemble novel transfer model
Learning Models Database: Thin Prep convolutional neural had high
for Automated Pap dataset network (CNN) and a CNN
Cross-Preparation with autoencoder (AE). variability in
Diagnosis of Class: 4 Bethesda performance,
Multi-Cell Liquid class Type: Machine while CNN and
Pap Smear learning—Neural network AE CNN did not.
ResNet101
Images [37] accuracy is
92.65%.
Segmentation—hybrid
system that incorporates
two binary image patches
obtained by a 19-layered
convolutional neural
network (ConvNet) model
Multi-class with an enhanced deep high Accuracy 99.12%
nucleus detection dimensional dissimilarity Specificity 99.45%
and classification Database: Herlev translation (HDDT). Sensitivity
using a deep dataset, SIPaKMeD A Pre trained Resnet-50 99.25%
convolutional dataset, CRIC model. Execution time
neural network dataset T-distribution stochastic 99.6248 s
with enhanced neighbour embedding The proposed
high dimensional Class: Herlev-7
class, SIPaKMeD-5 (t-SNE) for down-sampled. model is more
dissimilarity Class, CRIC-6 class Classification using a effective
translation model multi-class weighted kernel compared to
on cervical extreme learning machine existing
cells [38] (WKELM) classifier via a approaches.
sparse multicanonical
correlation (SMCCA)
method.
Type: Machine
learning—Neural network
Diagnostics 2022, 12, 2900 7 of 16

Table 3. Cont.

Title Class and Database Methodology Results

ShufflenetV2
results:-
Classification using hybrid Accuracy 96.18%
Hybrid loss function with label Precision 96.30%
Loss-Constrained Database: Herlev smoothing to improve the Recall 96.23%
Lightweight dataset distinguishing power of Specificity 99.08%
Convolutional lightweight convolutional GhostnetV2
Neural Networks Class: Herlev-7 neural networks (CNN) results:-
for Cervical Cell class Accuracy 96.39%
Classification [39] Type: Machine Precision 96.42%
learning—Neural network Recall 96.39%
Specificity
99.09%
Integration of Single Shot
Accuracy 90.80%
MultiBox Detector with the Mean average
Detection of Database: Herlev positive and negative
precision (mAP)
cervical cells dataset features to address the
based on problem of insufficient is 81.53%, which
improved SSD Class: Herlev-7 is 7.54% and
sensitivity for small objects. 4.92% higher
network [20] class
Type: Machine than YOLO and
classical SSD.
learning—Neural network
Detection using the YOLO
algorithm.
Feature extraction
generalization by adding Mean average
Detection of the dense block and S3Pool precision (mAP)
cervical cancer Database: Herlev algorithm on the basis of the
cells in a complex is 78.87%, which
dataset is 7.54% and
feature extraction network
situation based on DarkNet-53. 4.92% higher
improved Class: Herlev-7
Clustering algorithm of than YOLO (You
YOLOv3 class
network [1] improved algorithm Only Look Once)
k-means++. and classical SSD

Type: Machine
learning—Neural network
Data augmentation module
of DTWCT module and
Pap smear-based Database: Mendeley convolutional neural
cervical cancer LBC SIPaKMeD networks (CNN).
detection using Classification using ResNet Average Pap
dataset
residual neural 18, which defines four smear detection
networks deep Class: Mendeley classes of sources for Pap index (PDI) is
LBC SIPaKMeD—4 99%.
learning smear cell images.
architecture [27] class
Type: Machine
learning—Neural network
Convolutional neural
network (L-PCNN) that
Cervical cell multi- integrates global context
classification Database: information and attention Accuracy 98.89%.
algorithm using SIPaKMeD dataset mechanism. Sensitivity 99.9%.
global context Improved ResNet-50 Specificity 99.8%.
information and Class: backbone network for F-measure
attention SIPaKMeD—5 class feature extraction. 99.89%.
mechanism [3]
Type: Machine
learning—Neural network
Diagnostics 2022, 12, 2900 8 of 16

Table 3. Cont.

Title Class and Database Methodology Results

Novel classification using
DeepCyto.
Principal component
DeepCyto: a Database: Herlev analysis and machine DeepCyto is a
hybrid framework dataset, SIPaKMeD learning ensemble for powerful tool for
for cervical cancer dataset, LBC dataset classification of Pap smear precise feature
classification images.
using deep feature Class: Herlev-7 extraction and
Artificial neural network Pap smear image
fusion of cytology class, SIPaKMeD-5 with feature fusion vectors
class, LBC-4 class classification.
images [40] as an input for classification.

Type: Machine
learning—Neural network
Transfer learning using Accuracy 99.03%.
Classification of deep learning convolutional Kappa coefficient
Cervical Cytology Database: Cervix93 neural network. showing perfect
Overlapping Cell cervical cytology Cutting edge pretrained agreement.
image networks: AlexNet, AlexNet proved a
Images with
ImageNet, and Places 365. successful
Transfer Learning Class: 3 class assistive tool for
Architectures [29] Type: Machine cervical cancer
learning—Neural network detection.
Detection using intelligent
deep convolutional neural
network.
Classification
(IDCNN-CDC) model using
biomedical Pap smear
images. Higher
Noise removal using performance of
Optimal deep
Database: Herlev Gaussian Filter. the proposed
convolution Segmentation using the
dataset technique in
neural network
Tsallis entropy technique terms of
for cervical cancer Class: Herlev-7
diagnosis with the dragonfly sensitivity,
class optimization. specificity,
model [8]
Deep learned feature using accuracy, and
SqueezeNet. F-Score.
Classification using
weighted extreme learning
machine (ELM).
Type: Machine
learning—Neural network
Novel Modified Firefly
Optimization Algorithm
with Deep
Learning-enabled cervical
cancer classification
(MFFOA-DL3) model.
The findings of a
Noise removal using
comprehensive
Modified Bilateral Filtering
comparison
metaheuristics (BF)-based. investigation
with stacked Database: Herlev
dataset Segmentation technique of
sparse denoising revealed that the
Kapur’s entropy-based MFFOA-DL3
autoencoder Class: Herlev-7 image to define affected model
model for cervical class area. outperformed
cancer Generate feature vectors
classification [9]. other recent
using EfficientNet. approaches.
Classification of the cell
using MFFOA with Stacked
Sparse Denoising
Autoencoder (SSDA) model.
Type: Machine
learning—Neural network
Diagnostics 2022, 12, 2900 9 of 16

Table 3. Cont.

Title Class and Database Methodology Results

An effective hybrid deep
learning technique using The proposed
Small-Object method identifies
Detection-Generative and classifies
Imaging based Adversarial Networks cervical
Database: Herlev (SOD-GAN) with premalignant
cervical cancer
diagnostics using dataset, Colposcopy Fine-tuned Stacked
and malignant
images, Clinical Autoencoder (F-SAE).
small object Generation and diseases based on
references deep
detection— discrimination of the
generative cervical cell using characteristics
Class: not
adversarial applicable Region-based without the
networks [5] necessity for
Convolutional Neural
Network (RCNN). initial
classification and
Type: Machine segmentation.
learning—Neural network
MULTP, the
proposed texture
Segmentation of the image descriptor, is a
using a thresholding generic operator
Cervical cancer approach. that may be used
diagnosis based Feature extraction by to characterise
on modified applying a texture texture features
uniform local Database: Herlev descriptor titled modified of images in
ternary patterns dataset uniform local ternary numerous
and feed forward patterns (MULTP). computer vision
multilayer Class: Herlev-7 Classification of the cell issues. In
network class using an optimized addition, the
optimized by multilayer feed-forward suggested
genetic optimization
neural network.
algorithm [41] approach may be
Type: Machine utilised to
learning—Neural network increase
performance in
deep networks.
Augmentation process of
image enhancement, image
flipping, and image rotating
to reduce the number of
parameters necessary.
Segmentation of the Accuracy 99.80%.
cancer-affected regions with Specificity 99%.
Early cervical the help of kernel weighted Sensitivity
Database: Herlev 98.83%.
cancer diagnosis fuzzy local information F-Score 97.8.
dataset
using Sooty c-means clustering Improvement of
tern-optimized Class: Herlev-7 (KWFLICM) model. 28.5% better than
CNN-LSTM class Classification using the Random Forest
classifier [11] Sooty Tern Optimization and 19.46% better
(STO) algorithm with than ensemble
CNN-based long short-term classifier.
memory classifier
(CNN-LSTM).
Type: Machine
learning—Neural network
Boruta analysis and SVM Boruta analysis
Hybrid Model for
method for an efficient shows a better
Detection of performance
Cervical Cancer Database: Private feature selection and
Using Causal prediction of the model for approach
Class: not the cervical cell dataset. compared to the
Analysis and
applicable existing
Machine Learning Type: Machine techniques
Techniques [10] learning—Linear model available.
Diagnostics 2022, 12, 2900 10 of 16

Table 3. Cont.

Title Class and Database Methodology Results

Cervical Cancer Feature extraction using
Classification Database: Herlev ResNet-101.
Using Combined dataset Classification using Support
Machine Learning vector Machine (SVM). Accuracy 97.30%.
and Deep Class: Herlev-2
Learning class Type: Machine
Approach [42] learning—Linear model
Accuracy 98.70%.
Deep features extraction Sensitivity
using deep Convolutional 98.20%.
Neural Network of Specificity
pretrained Visual Geometry 98.90%.
Auxiliary Database: Herlev Group-19 (VGG-19). The suggested
classification of dataset, SIPaKMeD Hand-crafted images novel screening
cervical cells dataset, BJTU undergo the process of methodology is
based on the dataset feature selection, clustering, promising for
multi-domain and dimensionality early cervical
hybrid deep Class: Herlev-2&7 reduction. cancer detection,
learning class, SIPaKMeD-5 Classification using a with
framework [17] Class, BJTU-7 class Support Vector Machine multi-domain
(SVM) classifier. and hybrid
characteristics
Type: Machine proving realistic
learning-Linear model in clinical
practise.
All machine
learning
An Evaluation of architectures
Computational Database: Z-Stack gave outstanding
cellular microscopy Machine learning
Learning-based nuclei
proliferation images architecture of Random segmentation in
Methods for the Forest, Ada Boost, and MLP
Segmentation of provided by the algorithm. cervical cancer
HCS Pharma cells but did not
Nuclei in Cervical solve the
Cancer Cells from Type: Machine
Class: not overlapping
Microscopic learning—Nonlinear model
applicable nuclei and
Images [43] Z-stack
segmentation
problems.
Analysis of the different
supervised machine
learning techniques. Maximum
The classification algorithm accuracy
Database: Dataset used Artificial Neural achieved using
Prognosis of of 858 cervical Network, Bayesian XG-Boost with
Cervical Cancer cancer patients with Network, SVM, Random complete features
Disease by 36 risk factors and Tree, Logistic Tree and 94.94%.
Applying one outcome XG-Boost Tree. This approach
Machine Learning variable Selection algorithm for offers much
Techniques [4] Class: not feature selection: relief rank, potential for
applicable wrapper method, and clinical use and
LASSO regression. cervical cancer
cell detection.
Type: Machine
learning—Nonlinear model
Diagnostics 2022, 12, 2900 11 of 16

Table 3. Cont.

Title Class and Database Methodology Results

Is the aspect ratio
of cells important
in deep learning?
A robust
comparison of Twenty-two deep learning Deep learning
deep learning models were used to classify models are
Database- the cervical cancer cells into robust to changes
methods for SIPaKMeD dataset
multi-scale two categories of standard in the aspect ratio
cytopathology cell Class: not and scaled datasets. of cervical cells in
image applicable cervical
classification: Type: Machine cytopathological
From learning—Nonlinear model images.
convolutional
neural networks
to visual
transformers [44]
A novel fast hybrid fuzzy
classification algorithm with
feature reduction for
medical images.
Integration of
quantum-based
A Fast Hybrid grasshopper computing Established the
Database: Herlev algorithm (QGH) with a importance of the
Classification
dataset feature selection
Algorithm with fuzzy clustering technique
Feature Reduction for feature extraction. on the accuracy
Class: Herlev-7
for Medical class The second integration of of the proposed
Images [45] the fusion technique utilises classifier
QGH with the fuzzy
c-means algorithm to
determine the best features.
Type: Machine
learning—Nonlinear model
Geometrical and feature
extraction using a novel
approach of modified fuzzy
c-means. Minimum
Augmentation of the images accuracy 94.15%.
Cervical Cancer using Principal Component Maximum
Classification Analysis (PCA) to maintain accuracy 96.28%.
Database: Herlev Average accuracy
from Pap Smear dataset the uncorrelated features
Images Using and thus reduce the 94.86%.
algorithm processing time. Sensitivity
Modified Fuzzy C Class: Herlev-7 97.96%.
Means, PCA, and class Classification of the Pap Specificity
KNN [7] smear image into normal 83.65%.
and abnormal cells using K F1-Score 96.87%.
Nearest Neighbour (KNN). Precision 96.31%.
Type: Machine
learning—Nonlinear model
A novel manual features
and voting mechanism to
achieve data expansion in
Database: Herlev semi-supervised learning.
A Semi-supervised dataset, SIPaKMeD Clarity function to filter out
Deep Learning dataset higher-quality images,
Method for annotating a small amount Accuracy 91.94%.
Cervical Cell Class: Herlev-7 of the high-quality images,
Classification [28] class, SIPaKMeD-5 and voting mechanism for
Class balancing and training data.
Type: Machine
learning—Classifier
Diagnostics 2022, 12, 2900 12 of 16

Table 3. Cont.

Title Class and Database Methodology Results

Provide improvement in the
Ensemble feature Higher value of
efficiency of the clustering
extraction model performance
with optimal network algorithm of
Database: 962 achieved.
kernelized histopathological Optimal Kernelized Fuzzy
Precision
clustering C-Means (OKFCM).
cervical images. Accurate histopathological Specificity
algorithm for Recall
identifying cancer Class: Not image ensemble-based AUC
from cervical applicable feature extraction model. Accuracy
histopathology FPR
Type: Machine FNR
images [46]
learning—Clustering
Feature extraction using the
discrete coefficient
Cervical cancer transform (DCT) and Haar
diagnosis based transform.
on cytology Pap Database: Classification using seven
smear image SIPaKMeD dataset different machine learning
classification algorithms for normal and Accuracy 81.11%
Class:
using fractional abnormal Pap smear images.
SIPaKMeD—5 class
coefficient and Optimization of feature
machine learning extraction using fractional
classifiers [36] coefficient to form the five
different sizes of feature
vectors.
Accuracy:-
4-class
liquid-based
cytology Pap
Taming transformers smear dataset
(CCG-taming transformers). 98.79%.
Improve the encoder 5-Class
Database: SIPAKMeD
Liquid-based structure by introducing 99.58%.
Improving SE-block and 7-Class Herlev
cytology, Herlev
cervical cancer MultiRes-block. 99.88%.
classification with dataset, SIPaKMeD Layer Normalization to
dataset, Inception score
imbalanced standardize the data. (IS) 3.75.
datasets SMOTE-Tomek Links to Frechet inception
combining taming Class: Liquid-based
balance the source data set distance (FID)
transformers with cytology—4 class,
and the number of samples 0.71.
T2T-ViT [19] Herlev dataset—7
and weights. Recall 0.32.
class, SIPaKMeD
dataset—5 class Classification using Precision 0.65.
Tokens-to-Token Vision Novel approach
Transformers (T2T-ViT) that applies the
combing transfer learning. transformer to
the generation
and recognition
of cervical cancer
cell images.

Among the plausible explanations for these findings is that several previous attempts
have been made in the area of cervical cell classification. Convolutional neural network
(CNN) is currently one of the best approaches for the classification process. Several tech-
niques have been studied by previous researchers related to CNN, such as CNN-based long
short-term memory classifier, region-based classifier, lightweight, ResNet-50, and others.
For example, Chitra et al. [11] introduced a technique of classification using the Sooty
Tern Optimization (STO) algorithm with a CNN-based long short-term memory classifier
(CNN-LSTM) and achieved better performance results compared to other literature reviews.
The results achieved in the study have shown that the accuracy is 99.80%, specificity is 99%,
sensitivity is 98.83%, and F-score is 97.8. Their findings show an improvement of 28.5%
better than Random Forest and 19.46% better than the ensemble classifier. The findings
are consistent with the findings of the past study by Li et al. [3], which also achieved
Diagnostics 2022, 12, 2900 13 of 16

almost similar values of accuracy, specificity, sensitivity, and F-measure as the previous
study by Chitra et al. [11]. The method applied was a pulse convolutional neural network
(PCNN) that integrates a global context information and attention mechanism with an
improved ResNet-50 backbone network for feature extraction. Other than that, in a study
by Liu et al. [44], a conclusion was made that DL models are robust to changes in the aspect
ratio of cervical cells in cervical cytopathological images. The above findings contradict the
study by Elakkiya et al. [5], which proposed a method of identification and classification
of cervical premalignant and malignant diseases based on deep characteristics without
the necessity for initial classification and segmentation. The findings of the literature have
come out with four different methods of classification used by previous studies, which
are neural network-based classification, linear model classification, nonlinear classifica-
tion model, and others. It is apparent from Figure 3 that the most popular approach in
classification of cervical cells using machine learning is the neural network. The findings
Diagnostics 2022, 12, x FOR PEER REVIEW
also provide evidence that the classification method based on the neural network15have of 18

resulted in a higher accuracy level when compared to the other approaches. As illustrated
in Figure 3, the neural network approach was able to achieve higher accuracy, which was
more
higherthan 99%, and
accuracy, proved
which was to be able
more than to detect
99%, andthe class to
proved of be
theable
cellto
being tested
detect in the
the class of
different datasets.
the cell being tested in the different datasets.

Classification performance accuracy, %

120
99.12 99.89 99.8
98.7 99.88
100 94.86
Performance accuracy, %

80

60

40

20

0
Neural network Linear model Non-linear model Others

Figure 3. Graph of classification performance accuracy.

Figure 3. Graph of classification performance accuracy.

4.4.Discussion
Discussionand
andConclusions
Conclusions
Cervicalcancer
Cervical cancerisisthe
thesecond
secondmostmostcommon
commonfemalefemalecancer
cancerworldwide.
worldwide. ItIt isisvital
vitaltoto
detect
detectititearlier
earlierwith
withlow-cost,
low-cost,high-accuracy
high-accuracyautomated
automatedscreening
screeningtechnologies,
technologies,especially
especially
in
incountries
countrieswithwithlimited
limitedmedical
medical resources.
resources. Following
Following breast cancer,
breast this this
cancer, malignancy is theis
malignancy
second
the secondbiggest causecause
biggest of death among
of death women
among in developing
women countries.
in developing Utilising
countries. automatic
Utilising auto-
identification, cervical intraepithelial neoplasia (CIN) can be effectively
matic identification, cervical intraepithelial neoplasia (CIN) can be effectively avoided. avoided. The only
way to avoid morbidity is to detect the problem as soon as possible.
The only way to avoid morbidity is to detect the problem as soon as possible. Since the Since the traditional
Pap smear test
traditional Papassesses the abnormalities
smear test of the cell by hand,
assesses the abnormalities of thethecellclinical
by hand, testthe
used to detect
clinical test
cervical
used to cancer
detect is more prone
cervical cancertoisfalse-negative
more prone toand false-positive
false-negative andresults. Detecting
false-positive and
results.
classifying
Detecting and Papclassifying
smear cellPap images
smear is cell
significantly complex in cervical
images is significantly complexcancer screening.
in cervical cancer
Patients gain from earlier medical therapy when cancer is detected, diagnosed,
screening. Patients gain from earlier medical therapy when cancer is detected, diagnosed, and classi-
fied
andearly. This early.
classified study’s goal
This is to review
study’s goal isthe DL techniques
to review to automatetocancer
the DL techniques diagnosis
automate cancer
and classification in order to ensure that patient’s health conditions
diagnosis and classification in order to ensure that patient’s health conditions improve improve over time.
The findings of a comprehensive comparison investigation revealed
over time. The findings of a comprehensive comparison investigation revealed that the that the MFFOA-DL3
model
MFFOA-DL3outperformed
model other recent approaches.
outperformed other recent The suggested method
approaches. identifies
The suggested and classi-
method iden-
fies cervical premalignant and malignant illnesses based on deep characteristics
tifies and classifies cervical premalignant and malignant illnesses based on deep charac- without
teristics without the need for initial classification and segmentation. The Boruta analysis
shows a better performance approach compared to the existing techniques available.
DeepCyto is a powerful tool for precise feature extraction and Pap smear image classifi-
cation. The suggested novel screening methodology of auxiliary classification for cervical
Diagnostics 2022, 12, 2900 14 of 16

the need for initial classification and segmentation. The Boruta analysis shows a better per-
formance approach compared to the existing techniques available. DeepCyto is a powerful
tool for precise feature extraction and Pap smear image classification. The suggested novel
screening methodology of auxiliary classification for cervical cells based on a multi-domain
hybrid DL framework is promising for early cervical cancer detection, with multi-domain
and hybrid characteristics proving realistic in clinical practise. All machine learning archi-
tectures gave outstanding nuclei segmentation in cervical cancer cells but did not solve
the overlapping nuclei and Z-stack segmentation problems. Besides that, when it comes
to coping with the complexities of large-scale data and identifying prognostic patterns,
machine learning has been demonstrated to outperform traditional statistical models. It
has a lot of clinical potential for enhancing cervical cancer treatment. However, the limi-
tations of prediction studies and models, such as simplification, insufficient information,
overfitting, and lack of interpretability, indicate that additional efforts are required to im-
prove the accuracy, reliability, and practicality of clinical outcome predictions. This review
paper hopes to gain significance for the better design and methodology of cervical cancer
classification, with the objective of aiding the diagnosis process of this cancer detection.
The proposed system architecture simplifies the detection and classification of cervical cells
in Pap smear images, resulting in the early detection of cervical cancer and possibly an
increase in women’s survival rates.

Author Contributions: Conceptualisation, W.A.M.; methodology, N.A.A. and W.A.M.; formal analy-
sis, N.A.A. and S.I.; investigation, N.A.A. and K.S.A.R.; resources, H.A. and M.A.J.; writing—original
draft preparation, N.A.A., H.A. and W.A.M.; writing—review and editing, N.A.A., H.F.H. and S.I.;
visualisation, H.F.H. and K.S.A.R.; supervision, W.A.M.; project administration, N.A.A. and W.A.M.
and funding acquisition, W.A.M. and M.A.J. All authors have read and agreed to the published
version of the manuscript.
Funding: This work was supported by the Ministry of Higher Education Malaysia under the Funda-
mental Research Grant Scheme FRGS/1/2021/SKK0/UNIMAP/02/1.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Acknowledgments: The authors thank the authors of the dataset for making it available online,
and they would also like to thank the anonymous reviewers for their contributions to enhancing
this paper.
Conflicts of Interest: The authors declare that they have no conflict of interest to report regarding
the present study.

References
1. Jia, D.; He, Z.; Zhang, C.; Yin, W.; Wu, N.; Li, Z. Detection of cervical cancer cells in complex situation based on improved
YOLOv3 network. Multimed. Tools Appl. 2022, 81, 8939–8961. [CrossRef]
2. Balaji, G.N.; Suryanarayana, S.V.; Sengathir, J. Enhanced boykov’s graph cuts based segmentation for cervical cancer detection.
EAI Endorsed Trans. Pervasive Health Technol. 2021, 7, e3. [CrossRef]
3. Li, J.; Dou, Q.; Yang, H.; Liu, J.; Fu, L.; Zhang, Y.; Zheng, L.; Zhang, D. Cervical cell multi-classification algorithm using global
context information and attention mechanism. Tissue Cell 2022, 74, 101677. [CrossRef] [PubMed]
4. Kumawat, G.; Vishwakarma, S.K.; Chakrabarti, P.; Chittora, P.; Chakrabarti, T.; Lin, J. Prognosis of Cervical Cancer Disease by
Applying Machine Learning Techniques. J. Circuits Syst. Comput. 2022, 2350019. [CrossRef]
5. Elakkiya, R.; Teja, K.S.S.; Jegatha Deborah, L.; Bisogni, C.; Medaglia, C. Imaging based cervical cancer diagnostics using small
object detection—Generative adversarial networks. Multimed. Tools Appl. 2022, 81, 191–207. [CrossRef]
6. Lia, M.; Horn, L.C.; Sodeikat, P.; Höckel, M.; Aktas, B.; Wolf, B. The diagnostic value of core needle biopsy in cervical cancer: A
retrospective analysis. PLoS ONE 2022, 17, e0262257. [CrossRef]
7. Lavanya Devi, N.; Thirumurugan, P. Cervical Cancer Classification from Pap Smear Images Using Modified Fuzzy C Means,
PCA, and KNN. IETE J. Res. 2021, 68, 1591–1598. [CrossRef]
8. Waly, M.I.; Sikkandar, M.Y.; Aboamer, M.A.; Kadry, S.; Thinnukool, O. Optimal deep convolution neural network for cervical
cancer diagnosis model. Comput. Mater. Contin. 2022, 70, 3297–3309. [CrossRef]
Diagnostics 2022, 12, 2900 15 of 16

9. Vaiyapuri, T.; Alaskar, H.; Syed, L.; Aljohani, E.; Alkhayyat, A.; Shankar, K.; Kumar, S. Modified metaheuristics with stacked
sparse denoising autoencoder model for cervical cancer classification. Comput. Electr. Eng. 2022, 103, 108292. [CrossRef]
10. Lilhore, U.K.; Poongodi, M.; Kaur, A.; Simaiya, S.; Algarni, A.D.; Elmannai, H.; Vijayakumar, V.; Tunze, G.B.; Hamdi, M. Hybrid
Model for Detection of Cervical Cancer Using Causal Analysis and Machine Learning Techniques. Comput. Math. Methods Med.
2022, 2022, 4688327. [CrossRef]
11. Chitra, B.; Kumar, S. Early cervical cancer diagnosis using Sooty tern-optimized CNN-LSTM classifier. Int. J. Imaging Syst. Technol.
2022, 32, 1846–1860. [CrossRef]
12. Kaur, H.; Sharma, R.; Kaur, L. Automated Cervical Cancer Image Analysis using Deep Learning Techniques from Pap-Smear
Images: A Literature Review. In Proceedings of the 2021 9th International Conference on Reliability, Infocom Technologies and
Optimization (Trends and Future Directions) (ICRITO), Noida, India, 3–4 September 2021; pp. 1–7. [CrossRef]
13. Diniz, D.N.; Rezende, M.T.; Bianchi, A.G.C.; Carneiro, C.M.; Luz, E.J.S.; Moreira, G.J.P.; Ushizima, D.M.; de Medeiros, F.N.S.;
Souza, M.J.F. A deep learning ensemble method to assist cytopathologists in pap test image classification. J. Imaging 2021, 7, 111.
[CrossRef]
14. Jahan, S.; Islam, M.D.S.; Islam, L.; Rashme, T.Y.; Prova, A.A.; Paul, B.K.; Islam, M.D.M.; Mosharof, M.K. Automated invasive
cervical cancer disease detection at early stage through suitable machine learning model. SN Appl. Sci. 2021, 3, 806. [CrossRef]
15. Diniz, D.N.; Rezende, M.T.; Bianchi, A.G.C.; Carneiro, C.M.; Ushizima, D.M.; de Medeiros, F.N.S.; Souza, M.J.F. A hierarchical
feature-based methodology to perform cervical cancer classification. Appl. Sci. 2021, 11, 4091. [CrossRef]
16. Ali, M.M.; Ahmed, K.; Bui, F.M.; Paul, B.K.; Ibrahim, S.M.; Quinn, J.M.W.; Moni, M.A. Machine learning-based statistical analysis
for early stage detection of cervical cancer. Comput. Biol. Med. 2021, 139, 104985. [CrossRef] [PubMed]
17. Zhang, C.; Jia, D.; Li, Z.; Wu, N. Auxiliary classification of cervical cells based on multi-domain hybrid deep learning framework.
Biomed. Signal Process. Control 2022, 77, 103739. [CrossRef]
18. Liu, W.; Li, C.; Xu, N.; Jiang, T.; Rahaman, M.M.; Sun, H.; Wu, X.; Hu, W.; Chen, H.; Sun, C.; et al. CVM-Cervix: A hybrid cervical
Pap-smear image classification framework using CNN, visual transformer and multilayer perceptron. Pattern Recognit. 2022, 130,
108829. [CrossRef]
19. Zhao, C.; Shuai, R.; Ma, L.; Liu, W.; Wu, M. Improving cervical cancer classification with imbalanced datasets combining taming
transformers with T2T-ViT. Multimed. Tools Appl. 2022, 81, 24265–24300. [CrossRef]
20. Jia, D.; Zhou, J.; Zhang, C. Detection of cervical cells based on improved SSD network. Multimed. Tools Appl. 2022, 81, 13371–13387.
[CrossRef]
21. Janiesch, C.; Zschech, P.; Heinrich, K. Machine learning and deep learning. Electron. Mark. 2021, 31, 685–695. [CrossRef]
22. Huang, J.; Wang, T.; Zheng, D.; He, Y. Nucleus segmentation of cervical cytology images based on multi-scale fuzzy clustering
algorithm. Bioengineered 2020, 11, 484–501. [CrossRef] [PubMed]
23. Jin, X.; Che, J.; Chen, Y. Weed identification using deep learning and image processing in vegetable plantation. IEEE Access 2021,
9, 10940–10950. [CrossRef]
24. Maier, A.; Syben, C.; Lasser, T.; Riess, C. A gentle introduction to deep learning in medical image processing. Z. Med. Phys. 2019,
29, 86–101. [CrossRef]
25. Tekchandani, H.; Verma, S.; Londhe, N.D.; Jain, R.R.; Tiwari, A. Computer aided diagnosis system for cervical lymph nodes in CT
images using deep learning. Biomed. Signal Process. Control 2022, 71, 103158. [CrossRef]
26. Yaman, O.; Tuncer, T. Exemplar pyramid deep feature extraction based cervical cancer image classification model using pap-smear
images. Biomed. Signal Process. Control 2022, 73, 103428. [CrossRef]
27. Sellamuthu Palanisamy, V.; Athiappan, R.K.; Nagalingam, T. Pap smear based cervical cancer detection using residual neural
networks deep learning architecture. Concurr. Comput. Pract. Exp. 2022, 34, e6608. [CrossRef]
28. Zhao, S.; He, Y.; Qin, J.; Wang, Z.; Urzua, U. A Semi-supervised Deep Learning Method for Cervical Cell Classification. Anal. Cell.
Pathol. 2022, 2022, 4376178. [CrossRef]
29. Mulmule, P.V.; Kanphade, R.D. Classification of Cervical Cytology Overlapping Cell Images with Transfer Learning Architectures.
Biomed. Pharmacol. J. 2022, 15, 277–284. [CrossRef]
30. Chen, W.; Gao, L.; Li, X.; Shen, W. Lightweight convolutional neural network with knowledge distillation for cervical cells
classification. Biomed. Signal Process. Control 2022, 71, 103177. [CrossRef]
31. Chen, M.; Li, W.; Hao, Y.; Qian, Y.; Humar, I. Edge cognitive computing based smart healthcare system. Futur. Gener. Comput.
Syst. 2018, 86, 403–411. [CrossRef]
32. Shamim Hossain, M.; Amin, S.U.; Alsulaiman, M.; Muhammad, G. Applying deep learning for epilepsy seizure detection and
brain mapping visualization. ACM Trans. Multimed. Comput. Commun. Appl. 2019, 15, 1–17. [CrossRef]
33. Lu, J.; Song, E.; Ghoneim, A.; Alrashoud, M. Machine learning for assisting cervical cancer diagnosis: An ensemble approach.
Futur. Gener. Comput. Syst. 2020, 106, 199–205. [CrossRef]
34. Elakkiya, R.; Subramaniyaswamy, V.; Vijayakumar, V.; Mahanti, A. Cervical Cancer Diagnostics Healthcare System Using Hybrid
Object Detection Adversarial Networks. IEEE J. Biomed. Health Inform. 2022, 26, 1464–1471. [CrossRef] [PubMed]
35. Hiebl, M.R.W. Sample Selection in Systematic Literature Reviews of Management Research. Organ. Res. Methods 2021. [CrossRef]
36. Kalbhor, M.; Shinde, S.V.; Jude, H. Cervical cancer diagnosis based on cytology pap smear image classification using fractional
coefficient and machine learning classifiers. TELKOMNIKA Telecommun. Comput. Electron. Control 2022, 20, 1091–1102. [CrossRef]
Diagnostics 2022, 12, 2900 16 of 16

37. Karasu Benyes, Y.; Welch, E.C.; Singhal, A.; Ou, J.; Tripathi, A. A Comparative Analysis of Deep Learning Models for Automated
Cross-Preparation Diagnosis of Multi-Cell Liquid Pap Smear Images. Diagnostics 2022, 12, 1838. [CrossRef]
38. Karri, M.; Annavarapu, C.S.R.; Mallik, S.; Zhao, Z.; Acharya, U.R. Multi-class nucleus detection and classification using deep
convolutional neural network with enhanced high dimensional dissimilarity translation model on cervical cells. Biocybern. Biomed.
Eng. 2022, 42, 797–814. [CrossRef]
39. Chen, W.; Shen, W.; Gao, L.; Li, X. Hybrid Loss-Constrained Lightweight Convolutional Neural Networks for Cervical Cell
Classification. Sensors 2022, 22, 3272. [CrossRef]
40. Shinde, S.; Kalbhor, M.; Wajire, P. DeepCyto: A hybrid framework for cervical cancer classification by using deep feature fusion
of cytology images. Math. Biosci. Eng. 2022, 19, 6415–6434. [CrossRef]
41. Fekri-Ershad, S.; Ramakrishnan, S. Cervical cancer diagnosis based on modified uniform local ternary patterns and feed forward
multilayer network optimized by genetic algorithm. Comput. Biol. Med. 2022, 144, 105392. [CrossRef]
42. Alquran, H.; Mustafa, W.A.; Qasmieh, I.A.; Yacob, Y.M.; Alsalatie, M.; Al-Issa, Y.; Alqudah, A.M. Cervical Cancer Classification
Using Combined Machine Learning and Deep Learning Approach. Comput. Mater. Contin. 2022, 72, 5117–5134. [CrossRef]
43. Maylaa, T.; Windal, F.; Benhabiles, H.; Maubon, G.; Maubon, N.; Vandenhaute, E.; Collard, D. An Evaluation of Computational
Learning-based Methods for the Segmentation of Nuclei in Cervical Cancer Cells from Microscopic Images. Curr. Comput. Aided
Drug Des. 2022, 18, 81–94. [CrossRef]
44. Liu, W.; Li, C.; Rahaman, M.M.; Jiang, T.; Sun, H.; Wu, X.; Hu, W.; Chen, H.; Sun, C.; Yao, Y.; et al. Is the aspect ratio of cells
important in deep learning? A robust comparison of deep learning methods for multi-scale cytopathology cell image classification:
From convolutional neural networks to visual transformers. Comput. Biol. Med. 2022, 141. [CrossRef]
45. Mahmoud, H.A.H.; Alarfaj, A.A.; Hafez, A.M. A Fast Hybrid Classification Algorithm with Feature Reduction for Medical
Images. Appl. Bionics Biomech. 2022, 2022, 1367366. [CrossRef]
46. Baghialaxmi, R.L.P.S.; Kirubagari, B. Ensemble feature extraction model with optimal kernelized clustering algorithm for
identifying the cancer from cervical histopathology images. J. Theor. Appl. Inf. Technol. 2022, 100, 4866–4877.