ENHANCING SKIN CANCER DIAGNOSIS

USING PCA FOR FEATURE REDUCTION

AND SVM FOR CLASSIFICATION

GROUP NO: 03
SUPERVISOR NAME: DR. RAVI AJMEERA

19/12/2024
GROUP NO: 03 MEMBERS’
DETAILS
S. Member Name Institute of Affiliation
No.
1. Dr.VVSSS Chakravarthy Raghu Engineering College

2. Dr.D.Sriramulu Aditya Institute of Technology and Management

3. Dr.BSSV Ramesh Babu Raghu Engineering College

4. Dr.Srinivasa Rao Sura GITAM Deemed to be University

5. Dr.P.Tejomurthi Seshadri Rao Guldvalleru Engineering

College,Gudlavalleru

AICTE QIP 2024 - IIIT KOTTAYAM 2

APPROVAL FROM
SUPERVISOR
Attach a screenshot of email approval of the presentation slides

AICTE QIP 2024 - IIIT KOTTAYAM 3

CONTENTS
 Introduction
 Literature Review
 Problem Statement
 Feasibility Study
 Objectives
 Proposed Methodology
 Results and Discussion
 Summary
 Publications (If Any)
 References

AICTE QIP 2024 - IIIT KOTTAYAM 4

INTRODUCTION
 Skin cancers are caused by the formation of abnormal cells that, depending on
their nature and intensity, may infiltrate or spread to different areas of the body.
Since our skin is the most exposed organ to environmental toxins, it is the most
vulnerable.
Skin cancer can be divided into three types: basal cell skin cancer (BCC),
squamous cell skin cancer (SCC), and melanoma.
The first two types of skin cancer are classified as non-melanoma and rarely lead
to death.
The most lethal type of skin cancer is melanoma. UV radiation from the sun is the
main cause of major skin cancer.
Melanoma, a highly aggressive form of skin cancer, represents a critical global
health challenge due to its rapid progression and high mortality rates.
Traditional diagnostic methods such as biopsies and visual examinations by
dermatologists are invasive, time-intensive, costly, and prone to subjective errors.
These limitations highlight the urgent need for innovative approaches that offer
accurate, non-invasive, and cost-effective melanoma detection. AICTE QIP 2024 - IIIT KOTTAYAM 5
 INTRODUCTION: TYPES OF
CANCERS

Actinic Keratosis Basal Cell Carcinom

 It is the most common form of skin cancers.

 A rough, scaly patch on the skin caused by years
 A type of skin cancer that begins in the
of sun exposure.
 Actinic Keratoses usually affects older adults. basal cells.
 Basal cells produce new skin cells as old
 Reducing sun exposure can help reduce the risk
ones die.
 Limiting sun exposure can help prevent
these cells from becoming cancerous.
AICTE QIP 2024 - IIIT KOTTAYAM 6
 INTRODUCTION: TYPE OF
CANCERS

Cherry Nevus Dermatofibro

 A small red dome-shaped bump on ma
 It is referred to as benign, fibrous
the skin ranging between 0.5 – 6 histiocytomas of the skin.
mm in diameter.  It is a commonly occurring cutaneous entity
 Usually several are present, typically usually centred within the skin’s dermis.
on the chest and arms and  The growths often develop after some type
increasing in number with age. of small trauma to the skin, including a
 This type of skin growth is not a puncture from a splinter or bug bite.
cause for concern unless it bleeds
often or changes in size, shape or
color. AICTE QIP 2024 - IIIT KOTTAYAM 7
INTRODUCTION: TYPES OF
CANCERS
 It is the most serious type of skin
cancers.
 It occurs when the pigment-producing
cells that give colour to the skin
become cancerous.
 Symptoms might include a new, unusual
growth or a change in an existing mole.
 Melanomas can occur anywhere on the
body
Melanoma

AICTE QIP 2024 - IIIT KOTTAYAM 8

INTRODUCTION
 The advent of machine learning (ML) and artificial intelligence (AI) has
revolutionized melanoma detection. Utilizing advanced image analysis, object
recognition, and classification techniques, these technologies promise
diagnostic precision and speed unattainable with conventional methods.

 Algorithms such as adaptive snake, support vector machines (SVMs), and

artificial neural networks (ANNs) have achieved accuracy rates exceeding
94%, largely due to the use of comprehensive datasets like the International
Skin Imaging Collaboration (ISIC) repository and the PH2 dataset.

 This study explores the design and implementation of a fully automated

melanoma detection system leveraging state-of-the-art segmentation, feature
extraction, and classification methods.

 By addressing challenges such as dataset bias, computational cost, and model

interpretability, this system aspires to enhance diagnostic accessibility,
efficiency, and reliability, ultimately advancing melanoma detection globally.
AICTE QIP 2024 - IIIT KOTTAYAM 9
 LITERATURE REVIEW
Reference Proposed Methodology/Technique
[1] A Pushpalatha, P Dharani, R Dharini, and J
Deep convolutional neural networks, such as Inception-
Gowsalya. Skin cancer classification detection using
v4, ResNet-152, and DenseNet-161, were trained for
cnn and svm. Journal of Physics: Confer- ence Series,
melanoma and squamous cell carcinoma detection and
page 012148, 2021.
seborrheickeratosis classification.
[2] Ebrahim Nasr Esfahani, Shadrokh Samavi, Nader
Karimi, S.M.Reza Soroushmehr, Mohammad Jafari, In this paper an implementation of a deep learning system
Kevin Ward, and Kayvan Najarian. Melanoma on a computer server, equipped with graphic processing unit
detection by analysis of clinical images using (GPU), is proposed for detection of melanoma lesions.
convolutional neural network. 2016.
[3] A Murugan, S Anu H Nair, and KP Sanal Kumar.
In this paper the watershed segmentation method is
Detection of skin cancer using svm, random forest
implemented for segmentation. The extracted segments are
and knn classifiers. Journal of medical systems, page
subjected to feature extraction. The features extracted are
269, 2019.
shape, ABCD rule and GLCM.

AICTE QIP 2024 - IIIT KOTTAYAM 10

 LITERATURE REVIEW
In order to categorize dermoscopy pictures into
[4] K Chatrapathy, M Lakshmi Prasad, Ajmeera Kiran,
benign or malignancy lesions, this research
Pundru Chan- dra Shaker Reddy, G Charles Babu, and N
introduces two innovative hybrid CNN strategies
Partheeban. Skin cancer classification using a hybrid
that use a support-vector-machine (SVM) at the
convolutional neural network with svm clas- sifier. In
output layer. In order to classify data, the
2023 Global Conference on Information Technologies and
characteristics mined by the first &second CNN
Com- munications (GCITC), pages 1–6. IEEE, 2023.
strategies are combined and supplied to the
SVM.
[5] Duggani Keerthana, Vipin Venugopal, Malaya Kumar Deep convolutional neural network (CNN) have shown
Nath, and Mad- husudhan Mishra. Hybrid convolutional great separability across many fine-grained object
neural networks with svm clas- sifier for classification of classes. Methods: This article presents two novel
skin cancer. Biomedical Engineering Advances, page hybrid CNN models with an SVM classifier at the
100069, 2023. output layer for classifying dermoscopy images into
either benign or melanoma lesions.
In this research, a novel skin cancer detection is
[6] Ashwani Kumar, Mohit Kumar, Ved Prakash Bhardwaj,
performed using a modified falcon finch deep
Sunil Kumar, and Shitharth Selvarajan. A novel skin
Convolutional neural network classifier (Modified
cancer detection model using modified finch deep cnn
Falcon finch deep CNN) that efficiently detects the
classifier model. Scientific Reports, page 11235, 2024.
disease with higher efficiency.
In this research, a novel skin cancer detection is
[7]Saket S Chaturvedi, Jitendra V Tembhurne, and Tausif performed using a modified falcon finch deep
Diwan. A multi- class skin cancer classification using Convolutional neural network classifier (Modified
deep convolutional neural networks. Multimedia Tools Falcon finch deep CNN) that efficiently detects the
and Applications, pages 28477–28498, disease with higher efficiency.
AICTE QIP 2024 - IIIT KOTTAYAM 11
PROBLEM STATEMENT
• Melanoma is a dangerous type of skin cancer that requires early detection to
ensure effective treatment and save lives. However, identifying whether a skin
lesion is malignant (cancerous) or benign (non-cancerous) can be challenging.
• This project aims to use artificial intelligence (AI) and machine learning (ML)
techniques to help doctors and researchers detect melanoma early.
• Specifically, Support Vector Machines (SVM) will be used to classify skin lesions as
malignant or benign. To improve the model’s performance and reduce the time it
takes for the system to work, Principal Component Analysis (PCA) will be applied
to simplify the data.
• A dataset of skin images will be used to train the SVM model, and the model’s
performance will be measured using metrics such as accuracy and a confusion
matrix. This study will also examine how reducing data features with PCA impacts
the accuracy of the classification.
• The goal is to develop a reliable and efficient tool that supports early diagnosis of
skin cancer, helping doctors make timely decisions and saving lives.
AICTE QIP 2024 - IIIT KOTTAYAM 12
FEASIBILITY STUDY
• This study evaluates the viability of using Support Vector Machines (SVM) with Principal
Component Analysis (PCA) for early melanoma detection using dermoscopic images.
• Technically, the approach is feasible as it relies on well-established machine learning
methods supported by open-source tools like Python's Scikit-learn and modern
computational resources.
• PCA reduces dimensionality, improving computational efficiency and ensuring faster
model training and prediction. Operationally, the system can be integrated into
diagnostic tools to assist dermatologists in accurately distinguishing malignant from
benign lesions, enhancing early detection.
• Economically, the use of publicly available datasets and open-source libraries
minimizes costs while offering potential healthcare savings through early intervention.
• With a structured timeline, including preprocessing, model development, and
evaluation, the study is achievable within a reasonable time frame, making it both
practical and impactful.

AICTE QIP 2024 - IIIT KOTTAYAM 13

OBJECTIVES
1.To develop and evaluate a Support Vector Machine (SVM)-based
classification model for distinguishing between malignant and benign skin
lesions using dermoscopic image data.

2.To investigate the impact of Principal Component Analysis (PCA) in reducing

feature dimensionality and improving the performance, computational
efficiency, and accuracy of the SVM model.

3.To assess the model's classification performance through metrics such as

accuracy, precision, recall, F1-score, and confusion matrix, highlighting
improvements achieved by integrating PCA.

AICTE QIP 2024 - IIIT KOTTAYAM 14

PROPOSED METHODOLOGY
• The research begins with the collection and preprocessing of a comprehensive
dermoscopic image dataset, where images are standardized, enhanced, and
features such as texture, color, and shape are extracted.
• Principal Component Analysis (PCA) is then applied to reduce the dimensionality of
the feature space, improving computational efficiency and focusing on the most
significant features.
• A Support Vector Machine (SVM) classifier is developed and trained using the PCA-
transformed features, with model evaluation conducted using metrics such as
accuracy, precision, recall, F1-score, and confusion matrix.
• The performance of the SVM model is compared with and without PCA to assess
the impact of dimensionality reduction on classification accuracy. Finally, the
results are analyzed to highlight improvements, demonstrating the effectiveness
of combining PCA with SVM for early melanoma detection.

AICTE QIP 2024 - IIIT KOTTAYAM 15

 TOOLS & TECHNIQUES FOR
IMAGE PROCESSING
•We focus on filtering, segmentation, feature extraction, PCA, and image classification.

•These techniques help improve, analyze, and manipulate images for real-world tasks like
medical imaging.

Filtering:
•Filtering removes unwanted noise from images to improve quality.

•It is used as a preprocessing step for tasks like feature extraction and object detection.

•Techniques like median filtering and Gaussian filtering are common for noise removal and smoothing.

AICTE QIP 2024 - IIIT KOTTAYAM 16

TOOLS & TECHNIQUES FOR
IMAGE PROCESSING
Median Filter: Segmentation:
•The median filter removes “salt-and- •Segmentation divides an image into
pepper” noise while keeping the edges of meaningful parts or regions.
the image sharp.

•It works by replacing a pixel with the •This makes it easier to identify areas of
median value of its neighboring pixels. interest, such as tumors in medical images.

•This filter is widely used in medical •Common methods include thresholding, edge
imaging and tasks where edge detection, and the watershed algorithm.
preservation is important.

AICTE QIP 2024 - IIIT KOTTAYAM 17

TOOLS & TECHNIQUES FOR
IMAGE PROCESSING
Thresholding:
• Thresholding converts grayscale images into black-and-white (binary) images by
setting a threshold.
• Otsu’s Thresholding automatically finds the best threshold to separate pixels into
two classes.
• Morphological operations like erosion help refine binary
Hereimages
:B(x,y):by removing
Output small
binary image.
unwanted details.
T: Global threshold.

AICTE QIP 2024 - IIIT KOTTAYAM 18

TOOLS & TECHNIQUES FOR
IMAGE PROCESSING
Morphological operations: Feature Extraction:

Morphological operations analyze and process •Feature extraction simplifies

complex data by identifying
binary images using structuring elements (SE). important patterns or
characteristics.
Erosion:
•It reduces the amount of data
while keeping the meaningful
Reduces the size of objects in the image
information.
· A: Input image.
•In images, features like edges,
· B: Structuring element. textures, and shapes are
extracted using tools like edge
· (B)z​: Translated structuring element. detection or deep learning.

AICTE QIP 2024 - IIIT KOTTAYAM 19

TOOLS & TECHNIQUES FOR
IMAGE PROCESSING
Principal Component Analysis (PCA): Image Classification:

•PCA reduces the size of high-dimensional •Image classification categorizes images

data while preserving its most important into predefined classes like benign or
features. malignant skin cancer.

•It identifies principal components that •The process involves improving image
capture the maximum variation in data. quality, detecting objects, extracting
features, and training a model.
•PCA is widely used in image compression,
data visualization, and machine learning to •It is widely used in medical diagnostics,
simplify datasets. facial recognition, and self-driving
vehicles.

AICTE QIP 2024 - IIIT KOTTAYAM 20

TOOLS & TECHNIQUES FOR
IMAGE PROCESSING
Supervised Classification: Unsupervised Classification:

•In supervised classification, the user •In unsupervised classification, the

provides training data to help the system system groups pixels or data based on
learn patterns. similarities without user input.

•The algorithm uses these patterns to classify •The software uses clustering techniques
the rest of the data into specific classes. like k-means or neural networks to find
patterns.
•Popular techniques include decision trees,
•The user interprets the groupings to
support vector machines (SVM), and neural
networks. understand the real-world features
represented by the classes.

AICTE QIP 2024 - IIIT KOTTAYAM 21

SUPPORT VECTOR MACHINE
(SVM)
•SVM is a supervised learning technique
Steps in SVM:
used to classify data, such as images.
•It finds the best boundary (hyperplane) 1.Input the image data.
2.Extract features like textures, edges, or
that separates data into two classes, like
color patterns.
benign or malignant.
•SVM uses kernel functions to handle 3.Apply the kernel function to separate the
data into classes.
complex, non-linear data and is very
4.Use the hyperplane to classify the image
accurate for small datasets.
(e.g., benign or malignant).

Fig.1.Block diagram for skin cancer

classification using SVM

AICTE QIP 2024 - IIIT KOTTAYAM 22

RESULTS AND DISCUSSION
Methodology:
The project follows a systematic approach involving several stages:

1.Image Preprocessing: This process converts images to grayscale, modifies the histogram,
and uses noise filtering to enhance contrast.
2.Segmentation: segment the Region of Interest (ROI) using Otsu thresholding, followed by
filling and opening operations to isolate the lesion area, making the melanoma features
clear for further analysis .
3. Feature Extraction: Extracts relevant features (texture, color, shape) from the segmented
lesion using methods like the Gray Level Co-occurrence Matrix (GLCM) and ABCD
parameters (Asymmetry, Border, Color, and Diameter).
4.Dimensionality Reduction: This method uses principal component analysis (PCA) to select
the most significant features for efficient classification.
5.Classification with SVM: Trains the SVM classifier to categorize lesions as benign or
malignant based on the extracted features.

AICTE QIP 2024 - IIIT KOTTAYAM 23

RESULTS AND
DISCUSSION
1.Preprocessing:
The dataset of skin cancer images has been
created with two labels: benign and malignant.
The label count is as shown in Fig.2.
Initially, images are converted to grayscale to
simplify the data and reduce computational
complexity. Subsequently, histogram
equalization is applied to modify the image's
intensity distribution, improving contrast and
visual clarity. Finally, noise filtering techniques,
using median filtering, is employed to reduce
noise artifacts and further enhance image
quality.

Fig.2. Labels

AICTE QIP 2024 - IIIT KOTTAYAM 24

 RESULTS AND DISCUSSION

(a) original Image (b) Gray Scale Image (a) Equalizer Image (b) Filtered Image

Fig.3.Preprocessing Images

AICTE QIP 2024 - IIIT KOTTAYAM 25

 RESULTS AND DISCUSSION
2.Segmentation:
• The segmentation process begins with Otsu thresholding, a technique that automatically
determines the optimal threshold value to differentiate between the lesion and the
surrounding skin.
• Once the initial segmentation is obtained, morphological operations such as filling and
opening are applied to refine the segmented region.
• Filling operation fills the holes within the segmented lesion, while the opening operation
removes small objects and noise, ensuring a clear and well-defined lesion area. The defined
segmentation process effectively isolates the region of interest, making the melanoma
features more prominent and facilitating accurate feature extraction for subsequent
classification tasks.

(d) Smoothing
(a) Original Image (b) Threshold
( c) Filled AICTE QIP 2024 - IIIT KOTTAYAM 26
Fig.4. Segmentation Process Images
RESULTS AND DISCUSSION
3.Features Extraction
Feature extraction in image processing refers to
 mean_red  circularit  correlatio
detecting and extracting significant patterns or
 mean_gree y n
attributes from images to simplify their representation
n  diameter  hist_entro
for machine learning models.
 mean_blue  contrast py
Techniques such as edge detection (like Sobel or Canny  asymmetry  homogen  hist_skew
filters) are used to identify boundaries. At the same  perimeter eity ness
time, a Histogram of Oriented Gradients (HOG) captures  energy  hist_kurto
Table 1: List of features of interest
texture and shape details, and Local Binary Patterns sis
(LBP) focus on analyzing textures.  hist_mean

These extracted features convert image data into

numerical forms, making it easier for Support Vector
AICTE QIP 2024 - IIIT KOTTAYAM 27
Machines (SVM) to perform classification or analysis
RESULTS AND DISCUSSION
4.Classification
4.1 Classification using only the SVM method
Table 2. Classification report

Type precision reca f1- suppo

ll score rt
Benin 0.80 0.74 0.77 3469

Malin 0.75 0.80 0.78 3307

Fig.5. Confusion m Matrix of SVM

AICTE QIP 2024 - IIIT KOTTAYAM 28

RESULTS AND DISCUSSION
4.2 PCA+SVM
The PCA has been considered a preliminary step to reduce the number of features
to 5.
Table 3.Classification report
Type precision recal f1- suppor
l score t
Benin 0.81 0.78 0.79 3469

Malin 0.78 0.81 0.79 3307

Fig.6. Confusion m Matrix PCA+SVM

AICTE QIP 2024 - IIIT KOTTAYAM 29
SUMMARY
Parameter SVM SVM+PCA

Average 0.769 0.789

cross-
validation
Accuracy 77.35% 79.34%

Classification metrics
SVM SVM+PCA
Type precision reca f1- suppo Type precision reca f1- suppo
ll score rt ll score rt
Benin 0.80 0.74 0.77 3469 Benin 0.81 0.78 0.79 3469

Malin 0.75 0.80 0.78 3307 Malin 0.78 0.81 0.79 3307

AICTE QIP 2024 - IIIT KOTTAYAM 30

SUMMARY
• In conclusion, the evaluation demonstrates that incorporating Principal Component
Analysis (PCA) with the Support Vector Machine (SVM) model significantly
improves its performance.
• The cross-validation average score increased from 0.769 to 0.789, and test
accuracy rose from 77.35% to 79.34%.
• The "Benign" and "Malignant" classes showed higher precision, recall, and F1-
scores with PCA, indicating better classification capabilities.
• Additionally, the macro and weighted averages for precision, recall, and F1-scores
improved from 0.77 to 0.79, reflecting a more balanced and effective model.
• These findings highlight the effectiveness of combining PCA with SVM to enhance
overall accuracy, generalization, and consistency in classification tasks.
AICTE QIP 2024 - IIIT KOTTAYAM 31
FUTURE SCOPE
• Hybridizing SVM with CNN and other neural network (NN) models will combine the
advantages of both techniques in classification, considering the apparent
performance excellence of the NN-based techniques.
• Furthermore, the machine-learning classification algorithms can also be hybridized
with evolutionary computing tools like genetic algorithms, particle swarm
optimization, etc., which can also be a good scope for future scope.
• Similarly, the other suggested improvement is the real-time application, which
involves handling data modeling in the real-time environment.

AICTE QIP 2024 - IIIT KOTTAYAM 32

PUBLICATIONS (IF ANY)

AICTE QIP 2024 - IIIT KOTTAYAM 33

REFERENCES
1. M. I. Faizi, S. M. Adnan, “Improved Segmentation Model for Melanoma Lesion
Detection Using Normalized Cross-Correlation-Based k-Means Clustering”, IEEE
Access, Vol 12, 2024.
2. Viomesh Singh, Kavita A. Sultanpure, Harshwardhan Patil, “Frontier Machine
Learning Techniques for Melanoma Skin Cancer Identification and
Categorization: An In-Depth Review”, Oral Oncology Reports, Vol 9, 100217,
2024.
3. Seeja R D, Suresh A, “Deep Learning Based Skin Lesion Segmentation and
Classification of Melanoma Using Support Vector Machine (SVM)”, Asian Pacific
Journal of Cancer Prevention, Vol 20, 2019.
4. Hiam Alquran, Isam Abu Qasmich, et al., “The Melanoma Skin Cancer Detection
and Classification Using Support Vector Machine”, IEEE Jordan Conference on
Applied Electrical Engineering and Computing Technologies, 2017.
5. Nourelhoda M. Mahmoud, Ahmed M. Soliman, “Early Automated Detection
System for Skin Cancer Diagnosis Using Artificial Intelligent Techniques”,
Scientific Reports, 2024.
6. Atheer Bassel, et al., “Automatic Malignant and Benign Skin Cancer
Classification Using a Hybrid Deep Learning Approach”, Diagnostics, 12,
AICTE QIP 2024 - IIIT 2472,
KOTTAYAM 34