A Project Report on

Deep Multi-Model Feature Extraction and Classification for

Cervical Cell Image Classification

Submitted in the partial fulfillment of the requirements for the award of degree

BACHELOR OF ENGINEERING

in

ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING

by

PRANAV RAJ B (160121729050)

V BHANU SAI (160121729061)

Under the guidance of

Ms. Prachi Pramod Dusa

Assistant Professor, AI&ML

DEPARTMENT OF ARTIFICIAL INTELLIGENCE AND MACHINE

LEARNING

CHAITANYA BHARATHI INSTITUTE OF

TECHNOLOGY

(An Autonomous Institution, Affiliated to Osmania University, Approved by AICTE,

Accredited by NAAC with A++ Grade and Programs Accredited by NBA)
Chaitanya Bharathi Post, Gandipet, Kokapet (Vill.), Hyderabad, Ranga Reddy - 500 075,
Telangana
 www.cbit.ac.in

CERTIFICATE

This is to certify that the project titled “Deep Multi-Model Feature Extraction and

Classification for Cervical Cell Image Classification” is the bonafide work carried out by

Pranav Raj B (160121729050) and V Bhanu Sai (160121729061), students of B.E. (AIML) of

Chaitanya Bharathi Institute of Technology(A), Hyderabad, affiliated to Osmania University,

Hyderabad, Telangana(India) during the academic year 2024-2025, submitted in partial

fulfilment of the requirements for the award of the degree in Bachelor of Engineering Artificial

Intelligence & Machine Learning and that the project has not formed the basis for the award

previously of any other degree, diploma, fellowship or any other similar title.

Supervisor​ ​ ​ ​ ​ ​ Head of the Department

Ms. Prachi Pramod Dusa​ ​ ​ Dr. Y. Rama Devi

Asst. Professor, AI&ML​ ​ ​ ​ Professor, Head of Dept. of AI&ML

Project Coordinator External Examiner

Smt.K.Mary Sudha Rani

Asst. Professor, AI&ML

1
 CHAITANYA BHARATHI INSTITUTE OF
TECHNOLOGY
(An Autonomous Institution, Affiliated to Osmania University, Approved by AICTE,
Accredited by NAAC with A++ Grade and Programs Accredited by NBA)
Chaitanya Bharathi Post, Gandipet, Kokapet (Vill.), Hyderabad, Ranga Reddy - 500 075,
Telangana
www.cbit.ac.in

Vision of Institute

To be the Centre of Excellence in Technical Education and Research.

Mission of Institute

To address the Emerging needs through Quality Technical Education and

Advanced Research.

Quality Policy

CBIT imparts value based Technical Education and Training to meet the
requirements of students, Industry, Trade/ Profession, Research and
Development Organizations for Self-sustained growth of Society.

2
 DEPARTMENT OF ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING

Vision of the Department

To produce professionals in artificial intelligence and machine learning through the best possible
education, acquire international recognition as a destination, and advance society in exciting and
creative ways.

Mission of the Department

• Impart rigorous training to generate knowledge through the state-of-the-art concepts and
technologies in Artificial Intelligence and Machine Learning.

• Develop technical proficiency in students through creativity and leadership.

• Encourage lifelong learning, social responsibility, environmental conservation, and

professional ethics.

• Establish centres of excellence in leading areas of computer and artificial intelligence

disciplines.

Program Educational Objectives (PEOs)

PEO1: Using a solid foundation in mathematical, scientific, engineering, and current computing
principles, formulate, analyse, and resolve engineering issues.
PEO2: Apply artificial intelligence theory and concepts to analyse the requirements, realise
technical specifications, and design engineering solutions.
PEO3: Through cross-disciplinary projects and a variety of professional activities, demonstrate
technical proficiency, AI competency, and foster collaborative learning and a sense of
teamwork.
PEO4: Provide graduates with solid knowledge, competence, and soft skills that will enable
them to ethically contribute to societal demands and achieve sustainable advancement in
emerging computer technologies through lifelong learning.

3
 CHAITANYA BHARATHI INSTITUTE OF
TECHNOLOGY
(An Autonomous Institution, Affiliated to Osmania University, Approved by AICTE,
Accredited by NAAC with A++ Grade and Programs Accredited by NBA)
Chaitanya Bharathi Post, Gandipet, Kokapet (Vill.), Hyderabad, Ranga Reddy - 500 075,
Telangana
www.cbit.ac.in

DEPARTMENT OF ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING

Program Outcomes (POs)

PO1.Engineering Knowledge: Apply the knowledge of mathematics, science, engineering
fundamentals and an engineering specialization for the solution of complex engineering
problems
PO2.Problem analysis: Identify, formulate, research literature, and analyse complex
engineering problems reaching substantiated conclusions using first principles of mathematics,
natural sciences, and engineering sciences.
PO3.Design/development of solutions: Design solutions for complex engineering problems
and design system components or processes that meet the specified needs with appropriate
consideration for public health and safety, and cultural, societal, and environmental
considerations.
PO4.Conduct investigations of complex problems: Use research-based knowledge and
research methods including design of experiments, analysis and interpretation of data, and
synthesis of the information to provide valid conclusions.
PO5.Modern tool usage: Create, select, and apply appropriate techniques, resources, and
modern engineering and IT tools, including prediction and modelling to complex engineering
activities, with an understanding of the limitations.
PO6.The engineer and society: Apply reasoning informed by the contextual knowledge to
assess societal, health, safety, legal, and cultural issues and the consequent responsibilities
relevant to the professional engineering practice.

4
PO7.Environment and sustainability: Understand the impact of the professional engineering
solutions in societal and environmental contexts, and demonstrate the knowledge of, and need
for sustainable development.
PO8.Ethics: Apply ethical principles and commit to professional ethics and responsibilities and
norms of the engineering practice. Individual and team work: Function effectively as an
individual, and as a member or leader in diverse teams, and in multidisciplinary settings.
PO9.Communication: Communicate effectively on complex engineering activities with the
engineering community and with the society at large, such as, being able to comprehend and
write effective reports and design documentation, make effective presentations, and give and
receive clear instructions.
PO10.Project management and finance: Demonstrate knowledge and understanding of the
engineering and management principles and apply these to one’s own work, as a member and
leader in a team, to manage projects and in multidisciplinary environments.
PO11.Life-long learning: Recognise the need for, and have the preparation and ability to
engage in independent and life-long learning in the broadest context of technological change.

Program Specific Outcomes (PSOs)

PSO1. Apply the principal concepts of AI Engineering to design, develop, deploy and prototype
AI Subsystems.

PSO2. Apply the knowledge gained pertaining to data storage, data analytics and AI concepts to
solve real world business problems.

PSO3. Apply, analyse, design, develop, and test principles of AI concepts on Intelligent
Systems.

5
 DEPARTMENT OF ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING

PROJECT PART-2 (20AMC18)

Course Outcomes (COs):

CO1.Demonstrate a sound technical knowledge of their selected topic.​
CO2.Design engineering solutions to complex problems utilizing a systematic
approach.​
CO3.Conduct investigations by using research-based knowledge and methods to
provide valid conclusions.​
CO4.Create/select/use modern tools for the modelling, prediction and
understanding the limitation of complex engineering solutions.​
CO5.Communicate with engineers and the community at large in written and oral
forms.​
CO6.Demonstrate the knowledge, skills and attitudes of a professional engineer.

CO-PO/PSO Articulation Matrix:​

PO1 PO2 PO3 PO4 PO5 PO6 PO7 PO8 PO9 PO10 PO11 PO12 PSO1 PSO2 PSO3

CO1

CO2

CO3

CO4

CO5

6
 DECLARATION

We hereby declare that the project entitled “Deep Multi-Model Feature Extraction and
Classification for Cervical Cell Image Classification” submitted for the B.E (AIML) degree is
our original work and the project has not formed the basis for the award of any other degree,
diploma, fellowship or any other similar titles.

Pranav Raj B

(160121729050)

V Bhanu Sai

(160121729061)

Place: CBIT-Hyderabad

Date:

7
 ACKNOWLEDGEMENT

The satisfaction that accompanies the successful completion of the project would be incomplete
without the mention of the people who made it possible, whose constant guidance and
encouragement crown all the efforts with success.

We show gratitude to our honourable Principal Dr. C. V. Narasimhulu, for providing all
facilities and support.

We are particularly thankful for Dr. Y.Rama Devi, the Head of the Department, Department of
Artificial Intelligence and Machine Learning, her guidance, intense support, and encouragement,
which helped us to mould our project into a successful one.

We wish to express our deep sense of gratitude to our supervisor Ms. Prachi Pramod Dusa,
Assistant Professor, Department of Artificial Intelligence and Machine Learning, for her able
guidance and useful suggestions, which helped us in completing the project on time.

We are particularly thankful to our Project Coordinator Smt. K. Mary Sudha Rani, Assistant
Professor, Department of Artificial Intelligence and Machine Learning, her guidance, intense
support, and encouragement, which helped us to mould our project into a successful one.

We also thank all the staff members of the AIML department for their valuable support and
generous advice.

Pranav Raj B (160121729050)

V Bhanu Sai (160121729061)

Place: CBIT-Hyderabad

Date:

8
 TABLE OF CONTENTS

S.No Title Page No.

Abstract 10
List of Tables 11
List of Figures 11
1. INTRODUCTION 12
1.1. Problem Definition, Significance, and Objectives 12
1.2. Beneficiaries 13
1.3. Methodology 14
1.4. Evaluation Metric: Accuracy 14
1.5. Datasets 14
2. LITERATURE SURVEY 16
2.1. Existing Works 16

3. PROPOSED SYSTEM 19
3.1. Dataset description 19
3.2. Architecture Diagram 20
3.3. Modules & Frameworks Description 22
4. TESTING AND RESULTS 24
4.1. Testing on a Dataset Image 24

5. DISCUSSIONS 25
5.1. Accuracies Achieved 25
5.2. Future Scope 26
6. CONCLUSIONS 27
7. RESEARCH PAPER STATUS 28
7.1. Literature Paper 28

REFERENCES 29

9
 ABSTRACT

Cervical cancer is among the major causes of cancer mortality in women, especially in
low-resource environments where early and precise diagnosis is imperative. In this study, we
introduce a hybrid deep learning system for automated cervical cell classification from the
SIPaKMeD dataset. Our method takes advantage of the feature extraction strength of Vision
Transformer (ViT), EfficientNet, and MobileNet to extract varied and complementary visual
representations. The combined features are classified using a Support Vector Machine (SVM),
selected due to its robustness and generalization performance on small datasets. The trained
system is implemented in a web-based application with a Flask backend and a React frontend,
providing accessible real-time predictions in operational healthcare settings. Experimental results
show high classification performance, confirming the merit of unifying transformer-based and
convolutional architectures for cervical cancer screening. The system provides a scalable,
lightweight solution focused on promoting early diagnosis and enhanced outcomes, particularly
in under-resourced settings.

Keywords: Cervical Cancer Detection, Deep Learning, Vision Transformer (ViT), EfficientNet,
MobileNet, Feature Extraction, Support Vector Machine (SVM), SIPaKMeD Dataset, Flask,
React

10
 LIST OF TABLES
Table No. Table Caption Page No.

1.​ Descriptions of the most common datasets used in cervical cancer detection 14
architectures.

LIST OF FIGURES
Figure No. Figure Caption Page No.

1.​ Example image of SIPaKMeD dataset image 20

2.​ Block diagram of a proposed Cervical cancer detection architecture. 20

3.​ Output Screen of Website 24

11
1. INTRODUCTION

1.1 Problem Definition, Significance, and Objectives

Cervical cancer remains a leading cause of cancer-related deaths among women globally,
particularly in low-resource settings where access to timely and accurate screening methods is
limited. Traditional diagnostic procedures like Pap smear interpretation are highly dependent on
skilled pathologists and are prone to inter-observer variability, making early detection a
challenge. With the growing availability of annotated medical image datasets and the rapid
advancements in artificial intelligence, there is an urgent need to develop automated, scalable,
and accurate diagnostic systems to support clinical decision-making and reduce diagnostic
delays.

This project addresses that need by leveraging deep learning-based feature extraction using
modern architectures—Vision Transformer (ViT), EfficientNet, and MobileNet—to learn rich
and meaningful representations from cervical cell images in the SIPaKMeD dataset. To further
enhance performance and eliminate redundant information, a Genetic Algorithm (GA) is applied
for optimal feature selection from the concatenated features of all three models. The selected
features are then classified using a Support Vector Machine (SVM), ensuring high-precision
categorization across five cervical cell types.

Unlike traditional diagnostic approaches or single-model deep learning solutions, our method
combines the strengths of multiple architectures with evolutionary optimization to maximize
classification accuracy and efficiency. The final trained model is embedded into a web-based
diagnostic tool using Flask for the backend and React for the frontend, creating a user-friendly
interface for real-time prediction. This project aims to deliver a robust and accessible tool for
early-stage cervical cancer detection, with the potential for broader application across other
cytology-based diagnostics.

1.2 Methodologies

This project presents a hybrid deep learning and evolutionary approach for the automated
classification of cervical cell images to support early detection of cervical cancer. The core

12
methodology involves feature extraction using three advanced pretrained models—Vision
Transformer (ViT), EfficientNet, and MobileNet—each known for its unique architectural
strengths. ViT excels in capturing long-range dependencies in images, EfficientNet offers a
highly parameter-efficient framework, and MobileNet is optimized for fast inference with limited
computational resources.

Each model is used to extract high-dimensional features from the input cytology images in the
SIPaKMeD dataset. These features are then flattened and concatenated to create a
comprehensive feature vector that encapsulates the strengths of all three models. However, since
this high-dimensional feature space may contain redundant or irrelevant information, a Genetic
Algorithm (GA) is employed for feature selection. GA mimics the process of natural selection
to iteratively select the most informative subset of features, thereby enhancing classification
performance and reducing computational overhead.

The refined features obtained from the GA are fed into a Support Vector Machine (SVM)
classifier, chosen for its ability to create robust decision boundaries in high-dimensional spaces.
The model is trained to accurately predict the cell category, achieving balanced performance
across all classes.

For real-world deployment, the final model is integrated into a full-stack web application using a
Flask backend and React frontend. This interface allows users—such as clinicians or lab
technicians—to upload cervical cell images and receive classification results in real time,
enabling faster and more accessible diagnostics in diverse healthcare settings, especially those
with limited expert availability.

1.3. Outline of Results

●​ The expected results of this project are focused on achieving high accuracy and
robustness in classifying cervical cell images into their respective categories (such as
superficial-intermediate, parabasal, koilocytotic, etc.). Through the use of multiple
pre-trained deep learning architectures, the model is designed to capture a wide range of
spatial and semantic features, leading to better generalization compared to using a single
network.

13
 ●​ Initial experiments show that combining feature representations from ViT, EfficientNet,
and MobileNet, followed by classification with an SVM, yields improved performance
over standalone models. Accuracy, precision, recall, and F1-score are used as evaluation
metrics to assess the effectiveness of the system. The model is tested using a hold-out
validation set, ensuring unbiased performance reporting.

●​ The deployment of the system as a web application further demonstrates its applicability
in real-world scenarios. Users can interact with the system through a browser interface,
upload microscopic images, and receive predictions within seconds, showcasing the
project's focus on both accuracy and usability.

1.4. Evaluation Metric: Accuracy

The primary evaluation metric used to assess the performance of the cervical cell classification
model is accuracy. Accuracy measures the proportion of correctly classified images relative to
the total number of images evaluated. It provides a straightforward indication of the model’s
effectiveness in assigning the correct class label to each input image from the SIPaKMeD
dataset:

Accuracy = (C/T) x 100

1.5. Datasets

The various datasets used for implementation of Cervical cancer detection using deep learning
are described in Table 1.

Dataset Name Description

SIPaKMeD 4049 segmented cervical cell images (5 classes: dyskeratotic, koilocytotic,

metaplastic, parabasal, superficial)

Herlev Dataset 917 images from Pap smear slides classified into 7 classes (normal to
carcinoma).

14
Cervix93 93 high-resolution cervical images labeled by domain experts, including
abnormal regions.

Cervical Cancer (Risk 858 samples with 32 attributes including demographic info, habits, and
Factors) – UCI medical history. Useful for risk prediction.

CRIC Cervix Dataset 400+ cervical cell images captured and labeled for
segmentation/classification tasks.

Table 1: Descriptions of the most common datasets used in cervical cancer detection
architectures.

15
2. LITERATURE SURVEY

2.1. Existing Works

In 2020, Xue et al. [1] employed transfer learning based on advanced neural networks
like Inception-V3, Xception, VGG-16, and ResNet-50 for extracting features of cervical
histopathological images. A weighted voting-based Ensemble Learning strategy was introduced
to enhance classification performance by combining strengths from multiple models. The
framework successfully classified three stages of cervical cancer differentiation—well, moderate,
and poorly differentiated. It reported high accuracy rates: 97.03% for AQP staining, 98.61% for
poorly differentiated VEGF staining images, and an overall accuracy of 98.37% in classifying
benign vs. malignant cells on the Herlev dataset. This study filled a gap in literature by
addressing differentiation stage classification in cervical histopathological images—an essential
step in diagnosis and treatment planning.

Sampaio et al. [2] proposed a mobile-oriented IoT framework for automated cervical
lesion detection using deep learning object detection models, focusing on Faster R-CNN. The
methodology transferred traditional cytology knowledge into mobile frameworks, leveraging a
new expert-annotated dataset of liquid-based cytology images. The system, designed for
smartphones, reported a cross-validation mAP of 0.20315, AR of 0.46572, and a 4-minute
average analysis time per sample. This mobile solution could be pivotal for low-resource
settings, enabling accessible screening.

Pramanik et al. [3] presented the first fuzzy distance-based ensemble approach for
cervical cancer detection using deep models. Three transfer learning models—Inception V3,
MobileNet V2, and Inception ResNet V2—were augmented with additional layers for
data-specific feature extraction. Distance metrics (Euclidean, Manhattan, Cosine) were
calculated and predictions combined via a product rule. The ensemble achieved 96.96%
accuracy, outperforming individual models (Inception V3: 95.30%, MobileNet V2: 93.92%,
Inception ResNet V2: 96.44%). This deep learning-fuzzy ensemble improved classification
efficiency and accuracy in medical imaging.

16
 Qin et al. [4] proposed a multi-task feature fusion model combining manual features with
deep learning methods for cervical cell classification. One model branch fit manual features; the
other classified cell categories. Utilizing multi-task learning and label smoothing, the model
achieved 81.88% (five-class) and 99.52% (two-class) accuracy. This dual-feature approach
reduced dependency on large labeled datasets and addressed mislabeling—critical in clinical
scenarios.

Fang et al. [5] introduced DeepCell, a deep CNN model using multi-size kernels and
dilated convolutions to improve feature pattern learning for cervical cell images. Without manual
segmentation, it showed superior accuracy, precision, recall, and F-score over other models on
the Herlev and SIPaKMeD datasets. This segmentation-free, high-accuracy classifier supports
improved diagnostics and treatment planning.

Maurya et al. [6] proposed ViT-CNN and CNN-LSTM models for cervical smear image
classification. ViT-CNN combined Vision Transformers and CNNs; CNN-LSTM utilized
CNN-extracted features fed into LSTM networks. The methodology incorporated attention maps
and data augmentation, achieving 97.65% accuracy with ViT-CNN and 95.80% with
CNN-LSTM. Attention maps enhanced interpretability, revealing critical image features. This
was among the first ViT-based ensemble strategies for cervical cell classification.

Pacal et al. [7] used CNN and Vision Transformer models to detect cervical cancer from
Pap smear images. Data augmentation increased dataset diversity, and ensemble learning boosted
prediction accuracy. Using over 60 models, their work—focused on the SIPaKMeD
dataset—demonstrated ViT models outperforming CNNs. This large-scale study emphasized
hospital-grade classification accuracy through innovative model combinations.

Kaur et al. [8] developed MLNet, a lightweight deep learning model improving cervical
cell classification. It tackled overfitting and parameter tuning via CNNs, data augmentation, and
Generative Adversarial Networks (GANs). Using 65% of data for training, 15% for validation,
and 20% for testing, MLNet achieved 99.36% accuracy on the Herlev dataset, outperforming
other models. It also showed high sensitivity and specificity and was evaluated on SIPaKMeD
and Mendeley LBC datasets.

17
 Sahoo et al. [9] proposed a fuzzy rank-based ensemble model using three pre-trained
classifiers. The end-to-end architecture incorporated CutOut, MixUp, and CutMix for
augmentation. Achieving 97.18% accuracy and 97.16% F1 on SIPaKMeD, and 99.22% accuracy
and 99.19% F1 on Mendeley LBC, the method improved decision-making by ranking classifiers
based on prediction confidence, tackling complexity in cervical cancer classification.

Kang et al. [10] introduced CerviSegNet-DistillPlus, an efficient framework using

DeepLabV3+ enhanced with knowledge distillation and model pruning. It incorporated Double
Swin-Transformer and Compressor Blocks for robust feature extraction while maintaining
lightweight design for clinical deployment. The model simulated diverse imaging conditions,
improving robustness and generalizability. The combination of pruning and distillation in a
compact architecture distinguished this model from prior approaches.

18
3. PROPOSED SYSTEM

3.1. Dataset Description

The SIPaKMeD dataset, short for Single-cell Images for Pap Smear Medical Dataset, was
developed by the University of Ioannina in Greece. It's specifically designed for research on
automatic cervical cell classification using Pap smear cytology. The dataset contains a total of
4049 high-quality RGB images, each representing a segmented single cell extracted from Pap
smear slides.

Each image is manually labeled by experts and falls into one of five categories, representing
different types of cervical cells:

●​ Superficial-Intermediate (SI) – normal mature squamous epithelial cells​

●​ Parabasal (PB) – immature squamous cells, often seen in atrophic smears​

●​ Koilocytotic (K) – indicative of HPV infection, showing perinuclear halos​

●​ Dyskeratotic (D) – abnormal keratinized cells, often precancerous​

●​ Metaplastic (M) – transformation zone cells, part of normal repair or precancerous

activity​

The original whole-slide images were segmented using a semi-automated method to isolate
single cells. The final images are roughly 200x200 pixels in size (dimensions can vary slightly),
stored in standard image formats BMP, and organized into folders by class label.

SIPaKMeD is ideal for training and evaluating image classification models, feature extraction
algorithms, or cell-level diagnostics in cervical cancer screening pipelines

19
 Figure 1: An example image of the SIPaKMeD dataset files.

3.2. Architecture Diagram

Figure 2: Architecture diagram of the proposed system.

This architecture diagram illustrates a hybrid deep learning and machine learning pipeline for
cervical cancer detection using cytology images. Here's a detailed description of each component
in the workflow:

1. Input: Cytology Image

●​ The process begins with a Pap smear cytology image of a cervical cell, likely from the
SIPaKMeD dataset.​

20
 ●​ These images contain morphological information critical for detecting abnormal cell
types.

2. Deep Feature Extraction

●​ The image is fed into three pre-trained deep learning models:​

○​ ViT (Vision Transformer) – captures global context using attention mechanisms.​

○​ EfficientNet – known for its parameter-efficient architecture and strong

performance.​

○​ MobileNet – a lightweight CNN ideal for fast, efficient feature extraction.​

●​ Each model extracts high-dimensional deep features from the input image.

3. Feature Concatenation

●​ The features extracted from ViT, EfficientNet, and MobileNet are concatenated into a
single unified feature vector.​

●​ This fusion combines diverse representations of the image, increasing the discriminative
power.

4. Optimal Feature Selection – Genetic Algorithm

●​ The concatenated feature vector is often high-dimensional and may contain redundant or
irrelevant features.​

●​ A Genetic Algorithm (GA) is used for optimal feature selection, simulating natural
selection to identify the most relevant subset of features.​

●​ This step improves efficiency and model generalization.

5. Final Classification – SVM

●​ The selected optimal features are fed into a Support Vector Machine (SVM) classifier.​

●​ SVM performs the final classification, predicting the cell type (e.g., normal vs. abnormal)
based on the selected features.​

21
3.3. Modules & Frameworks Description

3.3.1 Deep Feature Extractors

The deep feature extractors are composed of three distinct pre-trained convolutional and
transformer-based architectures: Vision Transformer (ViT), EfficientNet, and MobileNet. Each
model is responsible for extracting high-level discriminative features from the input cytology
image. ViT operates by dividing the image into fixed-size patches and processing them through
transformer blocks, allowing it to model long-range spatial dependencies. EfficientNet employs a
compound scaling method to balance depth, width, and resolution, producing compact yet
powerful feature representations. MobileNet, on the other hand, uses depthwise separable
convolutions to reduce computational cost while maintaining accuracy, making it suitable for
efficient feature extraction. These three models together capture diverse visual characteristics,
from fine-grained texture details to abstract structural patterns.

3.3.2 Feature Concatenation Layer

Following the individual extraction process, the output feature vectors from ViT, EfficientNet,
and MobileNet are concatenated into a single unified feature representation. This feature fusion
layer ensures that the combined vector encapsulates complementary information from multiple
receptive fields and architectures. The concatenated vector benefits from the unique strengths of
each backbone: global contextual awareness from ViT, balanced representation from
EfficientNet, and computational efficiency from MobileNet. This unified feature set serves as a
robust input for subsequent selection and classification tasks.

3.3.3 Genetic Algorithm for Feature Selection

Due to the high dimensionality of the fused feature vector, a Genetic Algorithm (GA) is utilized
to perform optimal feature selection. GA is a biologically inspired metaheuristic that evolves a
population of candidate solutions through selection, crossover, and mutation processes. In this
context, each individual in the population represents a binary mask determining which features to
retain. The fitness function is typically based on the classification performance (e.g., SVM

22
accuracy). By iteratively selecting the most informative and non-redundant features, GA
improves generalization, reduces overfitting, and enhances the classifier’s efficiency.

3.3.4 Support Vector Machine Classifier

The selected optimal features are finally passed to a Support Vector Machine (SVM), a
powerful supervised classifier that constructs a hyperplane in a high-dimensional space to
separate the classes. SVM is particularly effective in small-to-medium datasets with
high-dimensional features, which is typical in medical image analysis. The use of SVM ensures
robust decision boundaries and high classification accuracy, making it suitable for detecting
subtle morphological differences in cervical cells. The final output corresponds to the predicted
class label of the cytology image (e.g., normal or abnormal cell types).

23
4. TESTING AND RESULTS

4.1 Testing on an dataset image

Figure 3: Output Screen of the Website.

24
5. DISCUSSIONS

5.1. Accuracies Achieved

The classification performance of the proposed hybrid model was evaluated using standard
metrics such as precision, recall, F1-score, and accuracy across all five cervical cell classes in the
SIPaKMeD dataset. The results are summarized in the classification report, which reflects the
robustness and reliability of the proposed approach.

●​ im_Dyskeratotic: Achieved a precision of 0.95 and a recall of 0.99, resulting in a strong

F1-score of 0.97, indicating that the model is highly effective at identifying dyskeratotic
cells with minimal false negatives.​

●​ im_Koilocytotic: Obtained a precision of 0.97 and a recall of 0.92, with an F1-score of

0.94. While the precision is high, the slightly lower recall suggests a minor proportion of
missed positive instances for this class.​

●​ im_Metaplastic: Recorded a precision and recall of 0.95 and 0.96, respectively, yielding
an F1-score of 0.96, confirming the model’s balanced performance in classifying

25
 metaplastic cells.​

●​ im_Parabasal: Scored 0.97 in both precision and recall, with a consistent F1-score of
0.97, demonstrating stable performance in recognizing this class.​

●​ im_Superficial-Intermediate: Demonstrated exceptional performance with precision and

recall both at 0.99, and an F1-score of 0.99, indicating near-perfect classification of these
cell types.​

Overall, the model achieved a macro average F1-score of 0.97 and a weighted average F1-score
of 0.96, alongside a high overall accuracy of 97%. These results highlight the proposed model’s
ability to generalize across multiple cervical cell types while maintaining consistency and
precision.

5.2 Future Scope

The proposed hybrid framework demonstrates high accuracy in cervical cell classification;
however, there are several opportunities for further enhancement. Future work can focus on
integrating explainable AI (XAI) techniques to improve the interpretability of the model, which
is crucial in medical decision-making. Additionally, expanding the dataset with more diverse and
real-world clinical samples can help in improving the model’s generalization ability. Deployment
of the model in a web or mobile-based diagnostic tool can also be explored to facilitate real-time
screening in remote and under-resourced areas. Moreover, extending the framework to
multi-modal data (e.g., histopathological images or patient metadata) could lead to more
comprehensive cervical cancer diagnostic systems.

26
6.CONCLUSION

This research successfully proposed a hybrid deep learning framework for automated cervical
cancer cell classification using the SIPaKMeD dataset. By integrating multiple deep feature
extractors—Vision Transformer (ViT), EfficientNet, and MobileNet—the system effectively
captured rich and diverse features from cytology images. These features were concatenated and
refined through a Genetic Algorithm-based feature selection technique to ensure optimal
relevance and dimensionality reduction. The selected features were then classified using a
Support Vector Machine (SVM), yielding a high overall accuracy of 97%, with balanced
precision, recall, and F1-scores across all five cell categories.

The proposed architecture demonstrates the power of hybridization in feature learning and
selection for medical image classification. The results indicate strong generalization and
robustness across different cell types, making the model a promising tool for aiding pathologists
in early and reliable cervical cancer diagnosis.

Moreover, the success of this pipeline highlights the potential of combining modern vision
architectures with evolutionary optimization and classical machine learning classifiers for
improved diagnostic performance. The framework’s modularity also allows for future integration
of additional models and data modalities, paving the way for a more holistic cervical cancer
screening system.

27
7. RESEARCH PAPER STATUS

7.1. Literature Paper

28
 REFERENCES

[1] D. Xue et al., "An Application of Transfer Learning and Ensemble Learning Techniques for
Cervical Histopathology Image Classification," in IEEE Access, vol. 8, pp. 104603–104618,
2020, doi: 10.1109/ACCESS.2020.2999816.

[2] A. F. Sampaio, L. Rosado and M. J. M. Vasconcelos, "Towards the Mobile Detection of

Cervical Lesions: A Region-Based Approach for the Analysis of Microscopic Images," in IEEE
Access, vol. 9, pp. 152188–152205, 2021, doi: 10.1109/ACCESS.2021.3126486.

[3] R. Pramanik, M. Biswas, S. Sen, L. A. de Souza Júnior, J. P. Papa, and R. Sarkar, "A Fuzzy
Distance-Based Ensemble of Deep Models for Cervical Cancer Detection," Computational
Methods and Programs in Biomedicine, vol. 219, p. 106776, 2022, doi:
10.1016/j.cmpb.2022.106776.

[4] J. Qin, Y. He, J. Ge, and Y. Liang, "A Multi-Task Feature Fusion Model for Cervical Cell
Classification," in IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 9, pp.
4668–4678, Sept. 2022, doi: 10.1109/JBHI.2022.3180989.

[5] M. Fang, X. Lei, B. Liao and F. -X. Wu, "A Deep Neural Network for Cervical Cell
Classification Based on Cytology Images," in IEEE Access, vol. 10, pp. 130968–130980, 2022,
doi: 10.1109/ACCESS.2022.3230280.

[6] R. Maurya, N. N. Pandey, and M. K. Dutta, "VisionCervix: Papanicolaou Cervical Smears

Classification Using Novel CNN-Vision Ensemble Approach," in Biomedical Signal Processing
and Control, vol. 79, part 2, p. 104156, 2023, doi: 10.1016/j.bspc.2022.104156.

[7] I. Pacal and S. Kılıcarslan, "Deep Learning-Based Approaches for Robust Classification of
Cervical Cancer," Neural Computing and Applications, vol. 35, pp. 18813–18828, 2023, doi:
10.1007/s00521-023-08757-w.

[8] M. Kaur, D. Singh, V. Kumar, and H. -N. Lee, "MLNet: Metaheuristics-Based Lightweight
Deep Learning Network for Cervical Cancer Diagnosis," in IEEE Journal of Biomedical and
Health Informatics, vol. 27, no. 10, pp. 5004–5014, Oct. 2023, doi: 10.1109/JBHI.2022.3223127.

29
[9] P. Sahoo, S. Saha, S. Mondal, M. Seera, S. K. Sharma, and M. Kumar, "Enhancing
Computer-Aided Cervical Cancer Detection Using a Novel Fuzzy Rank-Based Fusion," in IEEE
Access, vol. 11, pp. 145281–145294, 2023, doi: 10.1109/ACCESS.2023.3346764.

[10] J. Kang and N. Li, "CerviSegNet-DistillPlus: An Efficient Knowledge Distillation Model

for Enhancing Early Detection of Cervical Cancer Pathology," in IEEE Access, vol. 12, pp.
85134–85149, 2024, doi: 10.1109/ACCESS.2024.3415395.

[11] T. Chen, W. Zheng, H. Hu, C. Luo, J. Chen, C. Yuan, W. Lu, D. Z. Chen, H. Gao, and J. Wu,
"A Corresponding Region Fusion Framework for Multi-Modal Cervical Lesion Detection," in
IEEE Transactions on Computational Biology and Bioinformatics, doi:
10.1109/TCBB.2022.3178725.

[12] M. Al Qathrady, A. Sh, T. Ali, A. Q. Rehman, S. Alqhtani, M. S. Alshehri, M. Irfan, S.

Rahman, L. A. B. Eljak, and L. Ahmed, "A Novel Web Framework for Cervical Cancer
Detection System: A Machine Learning Breakthrough," in IEEE Access, doi:
10.1109/ACCESS.2024.3377124.

[13] N. Youneszade, M. Marjani, S. Kumar, and K. Ray, "A Predictive Model to Detect Cervical
Diseases Using Convolutional Neural Network Algorithms and Digital Colposcopy Images," in
IEEE Access, doi: 10.1109/ACCESS.2023.3285409.

[14] Elakkiya, R., Subramaniyaswamy, V., Vijayakumar, V. and Mahanti, A., "Cervical Cancer
Diagnostics Healthcare System Using Hybrid Object Detection Adversarial Networks," in IEEE
Journal of Biomedical and Health Informatics, 2021, doi: 10.1109/JBHI.2021.3094311.

[15] M. K. Nour, I. Issaoui, A. Edris, A. Mahmud, M. Assiri and S. S. Ibrahim, "Computer Aided
Cervical Cancer Diagnosis Using Gazelle Optimization Algorithm With Deep Learning Model,"
in IEEE Access, vol. 12, pp. 13046–13054, 2024, doi: 10.1109/ACCESS.2024.3351883.

[16] M. E. Plissiti, P. Dimitrakopoulos, G. Sfikas, C. Nikou, O. Krikoni, and A. Charchanti,

"Sipakmed: A New Dataset for Feature and Image Based Classification of Normal and
Pathological Cervical Cells in Pap Smear Images," in Proc. 25th IEEE Int. Conf. Image
Processing (ICIP), Athens, Greece, 2018, pp. 3144–3148, doi: 10.1109/ICIP.2018.8451588.

30
[17] N. M. Fahad, S. Azam, S. Montaha, et al., "Enhancing Cervical Cancer Diagnosis with
Graph Convolution Network," Multimedia Tools and Applications, vol. 83, pp. 75343–75367,
2024, doi: 10.1007/s11042-024-18608-y.

[18] S. K. Mathivanan, D. Francis, S. Srinivasan, V. Khatavkar, P. K, and M. A. Shah,

"Enhancing Cervical Cancer Detection and Robust Classification Through a Fusion of Deep
Learning Models," Scientific Reports, vol. 14, no. 1, p. 10812, May 2024, doi:
10.1038/s41598-024-61063-w.

[19] B. A. Mohammed, E. M. Senan, Z. G. Al-Mekhlafi, M. Alazmi, A. M. Alayba, A. A.

Alanazi, A. Alreshidi, and M. Alshahrani, "Hybrid Techniques for Diagnosis with WSIs for
Early Detection of Cervical Cancer Based on Fusion Features," Applied Sciences, vol. 12, no. 17,
p. 8836, 2022, doi: 10.3390/app12178836.

31