S J C INSTITUTE OF TECHNOLOGY

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

The Project work Phase-II Review-II
On
“Advanced Meta-Heuristic Algorithm Based on Particle Swarm and Al-
Biruni Earth Radius Optimization Methods for Oral Cancer Detection”
PRESENTED BY
Ananya G R 1SJ20CS011
Anusha S V 1SJ20CS016
Poornima Paramanatti 1SJ20CS101
Prakruthi K 1SJ20CS103
UNDER THE GUIDANCE OF
Dr. Manjunatha Kumar B H Dr. Shrihari M R Dr. Seshaiah M Dr. Manjunatha Kumar B H
PROFESSOR AND HOD Associate Professor Associate Professor PROFESSOR AND HOD
Dept. Of CSE,SJCIT Dept. Of CSE,SJCIT Dept. Of CSE,SJCIT Dept. Of CSE,SJCIT
 CONTENTS
I. ABSTRACT
II. INTRODUCTION
III. LITERATURE SURVEY
IV. PROBLEM IDENTIFICATION & DEFINITION
V. SIGNIFICANCE & RELEVANCE OF WORK
VI. OBJECTIVES AND METHODOLOGY
VII. PLAN OF EXECUTION(Gantt Chart)
VIII.SOFTWARE REQUIREMENTS SPECIFICATION
IX.ANALYSIS
X. DESIGN
XI.IMPLEMENTATION\
XII. TESTING
XIII. RESULTS AND DISCUSION
XIV. CONCLUSION AND FUTURE ENHANCEMENTS
Bibliography
Dept. of CSE, SJCIT 2
 ABSTRACT
An approach for oral cancer detection by combining two optimization techniques: Particle Swarm Optimization
(PSO) and the AlBiruni Earth Radius Optimization method. The problem of oral cancer detection as an
optimization challenge, leveraging PSO's efficient exploration of solution spaces to identify relevant features
and optimize parameters. The Al-Biruni Earth Radius Optimization method is integrated to enhance the
algorithm's performance further. Through comprehensive experiments on real-world oral cancer datasets,
hybrid metaheuristic algorithm demonstrates significant improvements in accuracy, sensitivity, and specificity
compared to conventional methods, offering promising prospects for early diagnosis and treatment planning.
This showcases the potential of hybridizing PSO and the Al-Biruni Earth Radius Optimization method in
healthcare and optimization domains, setting the stage for future applications in medical diagnostics and
beyond.

Dept. of CSE, SJCIT 3

 INTRODUCTION
• Oral cancer is a pressing public health concern, often diagnosed at late stages with severe
consequences.
• Timely and accurate detection is pivotal for better outcomes and lower mortality rates.
• This project work pioneers an innovative approach by combining two optimization methods, Particle
Swarm Optimization (PSO) and the Al-Biruni Earth Radius Optimization method.
• This fusion of computational intelligence and optimization aims to revolutionize oral cancer detection.
• By synergizing PSO's efficient exploration of complex solution spaces and the potential of the Al-
Biruni Earth Radius Optimization method, we target improved feature selection and parameter
optimization, enhancing accuracy and sensitivity.
• It holds the potential not only to advance oral cancer diagnostics but also to contribute insights to
healthcare optimization and computational intelligence domains

Dept. of CSE, SJCIT 4

 LITERATURE SURVEY
Sl.No Paper Name Summary Algorithm used Accuracy Advantages Limitations/Future
work
01
“Abdelaziz along This research paper focuses on the crucial 1.Convolutional Neural The PSOBER algorithm It may require significant
with others issue of oral cancer detection, particularly in Network (CNN) offers superior precision in computational resources
proposed a paper low and middle-income countries where it is a 2. Deep Belief oral cancer detection for optimization,
"Advanced Meta- significant health concern. The validity and Network (DBN) through adaptive potentially limiting its
97.35%
Heuristic reliability of the proposed approach are further 3.Particle Swarm optimization of CNN and applicability in resource-
Algorithm Based confirmed through rigorous statistical tests, Optimization (PSO) DBN parameters. Its constrained settings.
on Particle Swarm including One-way analysis-of-variance 4.Al-Biruni Earth robustness, scalability, and Additionally, further
and Al-Biruni (ANOVA) and Wilcoxon signed-rank tests. Radius (BER) potential to reduce validation on larger and
Earth Radius While the proposed methodology shows great Optimization: BER specialist dependency hold diverse datasets is
Optimization promise and efficiency for specialists, it is Optimization is also promise for improving necessary to confirm its
Methods for Oral recommended that future research extends to used in the hybrid early cancer diagnosis, generalizability and
Cancer Detection” larger-scale datasets to validate the findings optimization algorithm particularly in resource- robustness across various
and potentially uncover additional oral cancer (PSOBER) constrained settings. clinical scenarios.
detection features.

Dept. of CSE, SJCIT 5

02
Roshan Alex This paper addresses the global health issue of 1.ResNet-101 The utilization of deep learning Roshan Alex
Welikala oral cancer by leveraging automation for early 2. Faster R-CNN algorithms, such as ResNet-101 and Welikala
proposed a paper diagnosis. It discusses the creation of a Faster R-CNN, in the MeMoSA proposed a
"Automated comprehensive library of annotated oral project offers a transformative paper
Detection and lesion images through the MeMoSA project. advantage by enabling automated "Automated
87.07%
Classification of The study utilizes deep neural networks, identification and classification of Detection and
Oral Lesions including ResNet-101 and Faster R-CNN, to oral lesions. These algorithms exhibit Classification of
using Deep develop automated systems for detecting and promising potential, with high F1 Oral Lesions
Learning for classifying oral lesions. Initial results indicate scores in image classification and using Deep
Early Detection promising potential for deep learning in lesion detection, signifying a Learning for
of Oral Cancer" tackling this challenging task, with image substantial leap toward low-cost, Early Detection
classification achieving high F1 scores for early diagnosis of oral cancer, of Oral Cancer"
lesion identification and referral, while object especially crucial in low-and middle
detection shows promise for identifying income countries.
lesions requiring referral.

Dept. of CSE, SJCIT 6

03
Amel Ali This paper addresses the significance of early 1.Advanced Al- This automation reduces the reliance Evaluating the
Alhussan along breast cancer detection, given its high Biruni Earth on manual diagnosis, overcomes proposed
with others mortality rate. While AI-based methods have Radius (ABER) image analysis challenges, and is approach using
proposed a paper been proposed, they face challenges like Optimization rigorously validated, promising additional larger
"Classification of overlapping regions and inadequate training. Algorithm earlier and more reliable detection of datasets and
Breast Cancer Statistical tests like ANOVA and Wilcoxon 2.Transfer breast cancer, potentially improving comparing the
Using Transfer reinforce the effectiveness of this approach Learning with patient outcomes. proposed
92.95%
Learning and over current methods, emphasizing its AlexNet methodology
Advanced Al- potential for improving breast 3.Convolutional with more
Biruni Earth cancer diagnosis. Neural Network recent
Radius (CNN) approaches.
Optimization”

Dept. of CSE, SJCIT 7

04
Abdelaziz This paper discusses the potential Transfer Learning using This approach ensures a more precise The future aims
proposed a paper threat of the monkeypox virus GoogleNet and effective diagnosis of of this work
"Classification of becoming a global pandemic, 2. Meta-Heuristic monkeypox skin lesions, potentially include
Monkeypox although it is less lethal and Optimization Algorithms: aiding in early detection and assessing the
Images Based on infectious than COVID-19. The i. Al-Biruni Earth Radius containment of the virus's spread. proposed
Transfer author proposes using machine Algorithm method with
91.8%.
Learning and the learning, particularly image-based ii. Sine Cosine Algorithm large-scale
Al-Biruni Earth diagnosis techniques, to improve the iii. Particle Swarm datasets and
Radius diagnosis of monkeypox. Optimization (PSO) engineering
Optimization 3. Binary Hybrid Algorithm optimization
Algorithm” for Feature Selection problems to
4. Hybrid Algorithm for clearly identify
Neural Network Parameter its advantages
Optimization and
disadvantages

Dept. of CSE, SJCIT 8

05
Harikumar oral cancer using a hybrid ABC-PSO 1.Hybrid ABC-PSO 1.BLDA classifier effectively The study offers
Rajaguru classifier and a Bayesian Linear Classifier manages large and noisy datasets valuable insights
and Sunil Discriminant Analysis (BLDA) classifier. 2.Bayesian Linear 2.ABC-PSO classifier- 100% into oral cancer
Kumar The BLDA classifier is employed to handle Discriminant Analysis accuracy classification,
Prabhakar large and noisy datasets, focusing on the (BLDA) Classifier 3.Early Detection suggesting the
proposed a linear relationship between targets and Techiniques potential of
91.66 %
paper “Oral feature vectors with Gaussian noise. The 3.Gaussian Mixture these methods in
Cancer results demonstrate that the Hybrid ABC- Models (GMM) mitigating
Classificati PSO classifier achieves 100% accuracy Multi-Layer Perceptrons global oral
on from across all stages of oral cancer, while the (MLP) cancer risks.
Hybrid BLDA classifier attains accuracy rates 4.Artificial Neural
ABC-PSO ranging from 74.49% to 91.66% across Networks
and different stages. The study offers valuable 5.Principal Component
Bayesian insights into oral cancer classification, Analysis (PCA)
LDA” suggesting the potential of these methods in
mitigating global oral cancer risks.

Dept. of CSE, SJCIT 9

06 Hani Attar along AIoMT (Artificial Intelligence of Medical Convolutional Neural 1.Improved Diagnosis Data
with others Things) approach to improve the Networks (CNNs) 2.Stain Normalization Dependency:
proposed a paper diagnosis of Oral Squamous Cell Random Forest (RF) 3. BPSO is employed for feature The quality and
“Diagnosis of Carcinoma (OSCC). OSCC is a Artificial Neural selection quantity of the
Oral Squamous significant subtype of head and neck Networks (ANN) dataset used for
Cell Carcinoma squamous cell carcinoma (HNSCC), and Techniques: training and
Using Deep early detection is crucial for successful 1.Deep testing can
96.3%
Neural Networks therapy. The paper proposes using deep learning(VGG16, influence the
and Binary learning models (VGG16, AlexNet, AlexNet, ResNet50, model's
Particle Swarm ResNet50, and Inception V3) for feature Inception V3) performance
Optimization on extraction from histopathological images 2. Binary Particle
Histopathological and Binary Particle Swarm Optimization Swarm Optimization
Images: An (BPSO) for feature selection. (BPSO)
AIoMT 3.Reinhard Stain
Approach” Normalization
4. XGBoost Classifier

Dept. of CSE, SJCIT 10

07 Rayed Al Ghamdi BERTL-HIALCCD, for better detection 1.ShuffleNet for feature 1. BERTL-HIALCCD- The algorithmic
along with others of lung and colorectal cancer using extraction. technique: reduces reliance on details of AER
proposed a paper histopathology images. It uses transfer 2. Deep Convolutional manual diagnosis. and COA are not
“Al-Biruni Earth learning with advanced shuffle nets and Recurrent Neural 2. Transfer learning enhances described in
Radius deep convolutional recurrent neural Network (DCRNN) for feature extraction depth, making it
Optimization networks (DCRNN). Al-Biruni Earth classification difficult to
with Transfer radius optimization and Koati 3. AER- hyperparameter assess their
80.20%
Learning Based optimization algorithms are used for optimization applicability in
Histopathological hyperparameter tuning. 4. Coati Optimization other contexts
Image Analysis Algorithm (COA) for
for Lung and further hyperparameter
Colon Cancer tuning.
Detection”

Dept. of CSE, SJCIT 11

08 R. Dharani1 It addresses the limitations of traditional 1.Bag histogram of 1.Early Detection The description
and S. diagnostic methods and employs digital oriented gradients 2.Feature Selection of the FPSO
Revathy image processing. The key features include 2.Wavelet transform 3.deep learning algorithm is
proposed a the use of the bag histogram of oriented 3.Zernike Moments limited
paper “ gradients, wavelet features, and Zernike 4.Fuzzy Particle Swarm
93.21%
DEEPORCD: Moments for feature extraction. The fuzzy Optimization (FPSO)
Detection of particle swarm optimization algorithm 5.Convolutional Neural
Oral Cancer (FPSO) is used for feature selection, and Networks (CNNs) for
using Deep classification is performed using image classification
Learning” Convolutional Neural Networks (CNNs).
The proposed method achieves improved
performance in terms of Recall Rate,
Classification Accuracy, Precision Rate, and
Error Rate compared to other approaches

Dept. of CSE, SJCIT 12

09 Xiao-Bo Yanet This research paper proposes a new method 1.Radial Basis Improved Accuracy Interpretability
along with of cancer prediction using a Radial Basis Function Neural Enhanced Reliability Computational
others, proposed Function Neural Network (RBFNN) Network (RBFNN) Increased Stability Complexity
“Cancer optimized by Particle Swarm Optimization 2.Particle Swarm Reduced Complexity
Prediction Based (PSO). The method aims to improve the Optimization (PSO) Standardization
84%
on Radical Basis accuracy, reliability, and stability of cancer
Function Neural prediction. The experimental results show
Network with that the proposed method can greatly and
Particle Swarm effectively enhance the accuracy of cancer
Optimization” prediction.

Dept. of CSE, SJCIT 13

10 Dr. D.Ravikumar The study employed CNNs and transfer 1.Convolutional High Accuracy: Improved Small and
et al, proposed learning to classify oral squamous cell Neural Networks Performance:. Imbalanced
“Classification carcinoma, utilizing a dataset of 1200 (CNN) Effective Feature Data Dataset.
Of Oral images. Various CNN architectures were 2.Binary Particle Augmentation Limited
Squamous Cell fine-tuned, with the concatenated model Swarm 90.67 Comprehensive evaluation
Carcinoma Using achieving the best performance, yielding an Optimization.
Convolutional accuracy of 96.67%, precision of 0.99, recall 3.Performance
Neural Networks of 0.94, and an F1-score of 0.97. Transfer Evaluation Metrics
And Binary learning and data augmentation techniques
Particle Swarm were crucial contributors to the enhanced
Optimization On classification accuracy, highlighting the
Oral effectiveness of these methods in improving
Histopathologica oral squamous cell carcinoma diagnosis
l Images”

Dept. of CSE, SJCIT 14

 PROBLEM IDENTIFICATION & DEFINITION

Problem Domain:
• Development of an accurate and efficient oral cancer detection system using a combination of deep learning and
meta-heuristic optimization algorithms. The goal is to achieve a higher accuracy than previous approaches and
contribute to improving the early detection of oral cancer.

Detailed Analysis:
• Project methodology combines the power of Particle Swarm Optimization (PSO) and Al-Biruni Earth Radius
Optimization (ABERO) to enhance the accuracy of oral cancer detection.
• PSO is employed to optimize feature selection and model parameters, improving the overall performance of the
detection system.

Dept. of CSE, SJCIT 15

 SIGNIFICANCE & RELEVANCE OF WORK
The proposed work addresses critical challenges in oral cancer detection by leveraging advanced meta-heuristic
algorithms, namely Particle Swarm Optimization (PSO) and the Al Biruni Earth Radius Algorithm (ABERA). Oral
cancer poses a significant public health concern due to its high morbidity and mortality rates, especially when
diagnosed late. Conventional diagnostic methods often lack the sensitivity and specificity needed for early
detection, resulting in delayed treatment and poorer outcomes. The complexity of oral cancer further complicates
accurate diagnosis, making it crucial to develop more effective detection strategies.

Our approach integrates PSO and ABERA to optimize feature selection and classification processes, enhancing the
accuracy and reliability of oral cancer detection models. PSO efficiently explores complex datasets to identify
discriminative biomarkers and patterns indicative of oral cancer, enabling early diagnosis and intervention. This
innovative methodology represents a significant advancement in oral cancer detection, empowering clinicians with
improved diagnostic tools to combat this pervasive health challenge effectively. By harnessing the synergistic
capabilities of PSO and ABER, our work promises to unveil new insights into oral cancer biomarkers, paving the
way for more precise and personalized diagnostic strategies.
Dept. of CSE, SJCIT 16
 OBJECTIVES & METHODOLOGY

OBJECTIVES

This Project aims to implement a hybrid meta-heuristic algorithm that combines Particle Swarm Optimization (PSO)
and the AI-Biruni Earth Radius Optimization method for effective oral cancer detection.
• Evaluate the proposed meta-heuristic algorithm's performance under varying conditions, including changes in
image quality, noise levels, and dataset characteristics, to ensure reliability in real-world scenarios
• Implement and benchmark the proposed algorithm against existing state-of-the-art approaches for oral cancer
detection, encompassing traditional image processing methods, machine learning techniques, and other meta-
heuristic algorithms.
• Fine-tune algorithmic parameters and mechanisms to optimize convergence speed and exploration-exploitation
balance, based on a thorough analysis of sensitivity, effectiveness, and predefined performance metrics

Dept. of CSE, SJCIT 17

METHODOLOGY

The methodology employed focuses on image analysis and optimization techniques. The primary goal is to enhance
the accuracy and efficiency of image-based classification tasks.

Methodology is divided into five key stages, each serving a crucial role in achieving the objectives.
• Dataset Preprocessing
• Feature Extraction with CNN

• PSOBER Optimization Algorithm

• Parameter Optimization for CNN and DBN

• Evaluation Metrics

Dept. of CSE, SJCIT 18

PLAN OF EXECUTION

Dept. of CSE, SJCIT 19

 SOFTWARE REQUIREMENTS SPECIFICATION

Software Requirements Specification for Anaconda Distribution:

1. Objective: The Anaconda distribution aims to simplify package management and deployment for data science
practitioners by providing an extensive collection of Python and R packages, tools, and resources.

2. Supported Languages: Anaconda supports Python and R programming languages, catering to the needs of data
scientists working in diverse environments.

3. Package Management System: Anaconda utilizes the conda package management system, which analyzes the
current environment to ensure compatibility and manage dependencies effectively during installations.

4. Included Packages: The distribution comes pre-installed with over 250 essential packages for data science,
including popular ones like numpy, pandas, scipy, scikit-learn, TensorFlow, and PyTorch.

5. Additional Packages: Users can access over 7500 additional open-source packages from PyPI and Anaconda
Cloud, enhancing the functionality of the distribution to suit specific project requirements.
Dept. of CSE, SJCIT 20
6. Graphical User Interface (GUI): Anaconda Navigator provides a user-friendly GUI alternative to the command-
line interface, enabling users to manage packages, environments, and channels seamlessly.

7. Compatibility and Portability: Anaconda supports Windows, Linux, and MacOS platforms, ensuring compatibility
across different operating systems for seamless deployment.

8. Community and Support: With over 20 million users worldwide and a large community, Anaconda offers support
for individual, team, and enterprise editions, ensuring users can access help and resources as needed.
9. Integration with Leading Technologies: Anaconda is optimized for machine learning tasks, including neural
networks, predictive analytics, data visualization, and bias mitigation, and integrates seamlessly with leading
technologies in the field.

10. Ease of Use and Accessibility: Anaconda prioritizes accessibility and ease of use, making it suitable for both
novice and experienced data scientists, with features like Jupyter Notebooks, PyCharm IDE, Spyder IDE, Glueviz,
and Orange 3 for various data science tasks and workflows.

Dept. of CSE, SJCIT 21

 ANALYSIS

MODULES

1. Data Preprocessing Module

• Prepare the dataset for analysis and modeling.
• Feature Normalization: Scale the features to a similar range to prevent any single feature from dominating due
to its scale.
• Data Splitting: Split the dataset into training, validation, and testing sets to ensure the model can be trained,
tuned, and evaluated properly

2. Feature Selection Module (PSO-Based)

• Identify the most relevant features for predicting oral cancer outcomes using Particle Swarm Optimization.

Dept. of CSE, SJCIT 22

• Swarm Initialization: Generate an initial population of particles, with each particle representing a

potential solution (i.e., a subset of features).

• Fitness Function: Define a metric (e.g., accuracy, sensitivity, specificity) to evaluate the effectiveness of a
• given subset of features.
• Optimization Loop: Iteratively update particles' positions and velocities based on their personal best and
• the global best positions, aiming to find the optimal feature subset.
3. Parameter Optimization Module (AERO-Based)
• Optimize the parameters of the prediction model using the AlBiruni Earth Radius Optimization method.
• Agent Initialization: Similar to swarm initialization in PSO, but here agents represent different parameter sets
for the model.
• Optimization Process: Use the AERO method to adjust the agents' positions (parameters) based on their
performance and the performance of their neighbors, striving for the best set of model parameters.

Dept. of CSE, SJCIT 23

4. Modeling and Training Module
• Train a prediction model using the selected features and optimized parameters.
• Model Selection: Choose an appropriate machine learning model for the task (e.g., SVM, neural network).
• Training Process: Train the model on the training dataset, utilizing the features and parameters identified by the
previous modules.
• Validation: Use the validation set to fine-tune the model and prevent overfitting.

5. Evaluation and Testing Module

• Assess the model’s performance on unseen data and compare it to baseline models.
• Performance Metrics: Calculate key metrics such as accuracy, sensitivity, and specificity on the testing set.
• Comparison: Compare these metrics against those of models developed with conventional approaches to
demonstrate improvements.

Dept. of CSE, SJCIT 24

6. Integration and Deployment Module
• Make the model accessible to end-users, typically healthcare professionals, for practical use in diagnosing oral
cancer.
• User Interface (UI): Develop a user-friendly interface for users to interact with the model, inputting patient data
and receiving predictions.
• Integration: Integrate the system with existing healthcare IT systems, ensuring it can receive input and provide
outputs seamlessly.
• Feedback Loop: Implement a mechanism for collecting user feedback and real-world performance data to
continuously refine and improve the model.

Dept. of CSE, SJCIT 25

 DESIGN

Fig 1.The architecture of the proposed system

Dept. of CSE, SJCIT 26
Fig 2. Illustrates the data flow within a Deep Belief Fig 3. Outlines the flow of data within a Convolutional
Network (DBN) model. Neural Network (CNN) model.
Dept. of CSE, SJCIT 27
Fig 4. Depicts the data flow within a Streamlit web application designed for
oral cancer detection.
Dept. of CSE, SJCIT 28
IMPLEMENTATION

Dept. of CSE, SJCIT 29

Parameter optimization using DBN
• Prepare Data: Organize images into folders for cancerous and non-cancerous cases.
• Augment Data: Increase the diversity of images through techniques like rotation, zooming, and flipping.

• Preprocess Data: Resize images and adjust pixel values for uniformity.

• Build Model: Utilize a pre-trained DenseNet169 model as a base, adding layers for classification.

• Train Model: Use augmented data to train the model, initially with fixed layers and then fine-tune by adjusting
some layers. Evaluate Model: Assess the models accuracy and effectively in classifying cancerous and non-
cancerous images using test data.
• Visualize Results: Plot training and validation metrics to understand the model's performance and identify any
issues.

Dept. of CSE, SJCIT 30

 TESTING

Testing is a critical aspect of ensuring the effectiveness and reliability of an oral cancer detection system.
Through comprehensive testing, we validate the accuracy and performance of image processing algorithms and
diagnostic modules that form the backbone of the system. Our testing strategy encompasses unit testing to
validate internal program logic, integration testing to assess component interactions, functional testing to ensure
adherence to technical requirements, and system testing to evaluate overall system performance. White box and
black box testing techniques are employed to verify the system's functionality from both internal and user
perspectives. This rigorous testing approach aims to identify and rectify errors, ensuring that the oral cancer
detection system meets its objectives of accuracy, reliability, and usability in real-world clinical settings.

Dept. of CSE, SJCIT 31

 Test Cases

Test Case Name Objective Input Expected Output Pass/Fail

Validate valid
Valid oral lesion Process inputs
Input Validation image inputs
images (e.g., PNG, without errors and Pass
Test produce accurate
JPG) correct predictions
outputs

Correctly identify
Validate internal
Test cases with and classify oral
Logic Validation program logic of
known lesions based on Pass
Test the detection
characteristics predetermined
algorithm
criteria

Dept. of CSE, SJCIT 32

Test Case Name Objective Input Expected Output Pass/Fail

Test integration of Preprocessed

Image
image Raw oral lesion images compatible
Preprocessing Pass
preprocessing with images with diagnostic
Integration Test
the main algorithm algorithm

Validate
Accurate
integration of
Algorithm Preprocessed predictions and
diagnostic Pass
Integration Test images classifications for
algorithms within
oral lesions
the system

Verify correct Acceptance and

Input Formatting formatting of Various types of processing of input
Pass
Test entries for oral input data data according to
cancer detection formatting rules

Dept. of CSE, SJCIT 33

Test Case Name Objective Input Expected Output Pass/Fail

Evaluate system Maintained

Large datasets,
Performance and performance and responsiveness and
concurrent user Fail
Scalability Test scalability under accuracy under
access
various conditions heavy usage

High level of code

Ensure thorough
Code Coverage Test cases covering coverage to
testing of internal Pass
Test decision branches identify potential
code paths
issues

Dept. of CSE, SJCIT 34

RESULTS AND DISCUSSION

Fig 5. Interface for oral cancer detection

Dept. of CSE, SJCIT 35

Fig 6. System input image

Dept. of CSE, SJCIT 36

Fig 7. System output image
Dept. of CSE, SJCIT 37
Figure 8. CNN Training and Validation Accuracy Plot Figure 9.CNN Training and Validation Loss Plot

Dept. of CSE, SJCIT 38

Figure 10. DBN Training and Validation Accuracy Plot Figure 11. DBN Training and Validation Loss Plot

Dept. of CSE, SJCIT 39

 Table 1. Performance comparison between Algorithms

Algorithm Accuracy F1-Score Sensitivity Precision

CNN 0.85 0.80 0.82 0.78
DBN 0.82 0.78 0.79 0.76

Random Forest 0.75 0.70 0.72 0.68

Support Vector
0.78 0.75 0.76 0.73
Machine (SVM)
K Nearest
0.72 0.68 0.70 0.65
Neighbor (KNN)

Dept. of CSE, SJCIT 40

 Table 2. Performance Comparison Of Optimization Methods

Optimization Method Accuracy F1-Score Sensitivity Precision

Hybrid PSO with
0.85 0.81 0.82 0.79
ABER (PSOBER)
Particle Swarm
0.83 0.79 0.80 0.76
Optimization (PSO)
Al-Biruni Earth Radius
0.80 0.76 0.78 0.74
Optimization (ABER)
Ant Colony
0.81 0.77 0.78 0.74
Optimization (ACO)
Stimulated Annealing 0.77 0.73 0.74 0.70

Differential Evaluation 0.82 0.78 0.79 0.75

Dept. of CSE, SJCIT 41

 CONCLUSION AND FUTURE ENHANCEMENTS
The "Advanced Meta-Heuristic Algorithm for Detecting Oral Cancer" represents a significant advancement in the
integration of meta-heuristic optimization techniques with deep-learning models for medical image analysis.
Through collaborative efforts, this research has established a robust foundation for further exploration and
development in the field of oral cancer detection. By leveraging the synergies between meta-heuristic optimization
and deep learning, this algorithm demonstrates promising potential for enhancing diagnostic accuracy in oral cancer
detection, ultimately contributing to improved patient outcomes. Moving forward continued research and
development in this area hold the key to further refining and optimizing the algorithm. This advancement has the
potential to push the boundaries of oral cancer detection, offering the possibility of earlier and more precise
diagnoses, thus significantly benefiting patients.

Dept. of CSE, SJCIT 42

 BIBLIOGRAPHY
[1] Harikumar, Rajaguru, and Sunil Kumar Prabhakar. "Oral cancer classification from hybrid ABC-PSO and
Bayesian LDA." In Communication and Electronics Systems (ICCES), 2017 2nd International Conference on. IEEE, 2017.
[2] Deif, Mohanad A., Hani Attar, Ayman Amer, Ismail A. Elhaty, Mohammad R. Khosravi, and Ahmed AA Solyman.
"Diagnosis of oral squamous cell carcinoma using deep neural networks and binary Particle Swarm optimization on
histopathological images: an AIoMT approach." Computational Intelligence and Neuroscience 2022 (2022).
[3] Dharani, R., and S. Revathy. "DEEPORCD: detection of oral cancer using deep learning." In Journal of physics:
conference series, vol. 1911, no. 1, p. 012006. IOP Publishing, 2021
[4] AlGhamdi, Rayed, Turky Omar Asar, Fatmah Y. Assiri, Rasha A. Mansouri, and Mahmoud Ragab. "Al-Biruni earth
radius optimization with transfer learning based histopathological image analysis for lung and colon cancer
detection." Cancers 15, no. 13 (2023): 3300.
[5] Yan, Xiao-Bo, Wei-Qing Xiong, Liang Hu, and Kuo Zhao. "Cancer prediction based on radical basis function neural
network with particle swarm optimization." Asian Pacific Journal of Cancer Prevention 15, no. 18 (2014): 7775-7780

Dept. of CSE, SJCIT 43

[6] Ravikumar, D., B. Kavitha, R. Mekala, D. Sasikala, and P. Dineshkumar. "Classification Of Oral Squamous Cell
Carcinoma Using Convolutional Neural Networks And Binary Particle Swarm Optimization On Oral
Histopathological Images." Journal of Pharmaceutical Negative Results (2023): 405-414.

[7] Alhussan, Amel Ali, Abdelaziz A. Abdelhamid, S. K. Towfek, Abdelhameed Ibrahim, Laith Abualigah, Nima
Khodadadi, Doaa Sami Khafaga, Shaha Al-Otaibi, and Ayman Em Ahmed. "Classification of breast cancer using
transfer learning and advanced al-biruni earth radius optimization." Biomimetics 8, no. 3 (2023): 270.
[8] Welikala, Roshan Alex, Paolo Remagnino, Jian Han Lim, Chee Seng Chan, Senthilmani Rajendran, Thomas George
Kallarakkal, Rosnah Binti Zain et al. "Automated detection and classification of oral lesions using deep learning for
early detection of oral cancer." IEEE Access 8 (2020): 132677-132693.
[9] Myriam, Hadjouni, Abdelaziz A. Abdelhamid, El-Sayed M. El-Kenawy, Abdelhameed Ibrahim, Marwa Metwally Eid,
Mona M. Jamjoom, and Doaa Sami Khafaga. "Advanced meta-heuristic algorithm based on Particle Swarm and Al-
biruni Earth Radius optimization methods for oral cancer detection." IEEE Access 11 (2023): 23681-23700.

Dept. of CSE, SJCIT 44

THANK YOU

Dept. of CSE, SJCIT 45