SlideShare a Scribd company logo
1 of 8
Abstract-Image Classification in medical imaging
applications is extremely important. In our work,
we have used deep learning with CNN to build a
model to classify skin cancer to detect if it’s benign
or malignant. Datasets of skin cancer is challenging
because of format and shape. It’s required effort to
detect different features.Wetrainedandtestedour
model on a dataset from the ISIC (International Skin
Image Collaboration) Archive. The first step is to
prepare data of images with less noise to make it
easy for detection. Then, building CNN model with
four convolutional layers, two max pooling layers,
two dropout layers and 1 flatten layer, two dense
layers .The results of the test produced accuracy of
.8030 .withmoreefforts we canbuildmoreefficient
model to get higher accuracy.
Index Terms-CNN, Skin Cancer, CNN Layers,
Convolution layer, Image Classification, Deep
Learning, Malignant Skin Cancer, Benign Skin
Cancer, Neural Network, Neurons, Kernel Filters,
Cross-Validation.
I.INTRODUCTION
mage classification is a computer vision
problem[1]. It is assigning pixels in the image into
several classes or categories of LULC (land use  land
cover)based on the applicationof statistical decision
rules or logical decision rules [2]. In order to classify
a dataset into differentclasses or categoriesthe data
and the classes’ relationship must be well-
understood. so image classifier depends on the role
of discriminant whether discriminant value highest
for one class and lower for other and this is called
multiclass (like the case of recognition of characters
or digits from bitmaps of scanned text) or
discriminantvaluepositivefor one classand negative
for another which is called two class (like the case of
distinguishing between text and graphics in scanned
document).
To achieve this, computer must be trained as
training is the main key of the classification success.
Important aspects of accurate classification is
feature sets which are data elements attributes
based on assigning elements in to various classes, it
could be quantitative or qualitative and learning
techniques that could be supervised learning or
unsupervised learning.
We have chosen one of the algorithms used in
image classification which is CNN (Convolutional
neural network) [3] [4] and we choose to handle the
I
Image Classification
Skin cancer detection with CNN
Eman Othman Fathy , Dina Alaa Ahmed , Dai Alaa Hassan, Mohammed IbrahimGab Allah,
Mohammed Cherif Mohammed, Nihal Mansour Abd-Elbari,Hager IsmailAbd-Elaty.
CairoUniversityFacultyof Engineering –December,2019
problems this algorithms faces in the cancer an
especially skin cancer.
In the coming sections you will see why we choose
CNN specifically and why skin cancer.
Why CNN?
There are many types of neural networks out there
like ANN (Artificial neural network), RNN (Recursive
neural network) [5], DNN (Deep neural network) [6]
and lastly but not last CNN (Convolutional Neural
Network) which we choose specifically because it is
the best for image classification tasks as ANN is best
for pattern recognition [7] and other neural
networks are best for other specific application.
CNN is a supervised learning algorithm and it is
different from other neural networks that is has at
least one convolutional layer. The convolution layer
focuses on detecting lower level features such as
edge detection and gradient orientation and others.
CNN has manyadvantagessuch that it trainsfaster,
needs fewer samples and reduces the chances of
overfitting. CNN requires less preprocessing
compared to other classification algorithms. It can
also learn filters itself while other primitive methods
need them to be hand-engineered.
Brief introduction about cancer:
Cancer occurs when normal cells start dividing
uncontrollably and spread into surrounding tissues.
Some of cancer basics are:
 A mass called a tumor form due to the
multiplication of the cells.
 If a tumor is malignant then it’s cancerous.
This means that they invade neighboring
tissues because of their uncontrolled growth.
 Tumors can also travel to remote organs
throw the bloodstream.
 Tumors also invade the surrounding tissue's
space and take the oxygen and nutrients they
need to survive and function.
Skin cancer:
Skin cancer is the most common of all human
cancers. It is abnormal growth of skin cells. Most
often occurs on skin exposed to the sun. But it can
also occur on areas of your skin not ordinarily
exposed to sunlight (or specifically ultraviolet (UV)
radiation) [8].
Much of the damage to DNA in skin cells results
fromultraviolet(UV) radiation that found in sunlight.
But sun exposure doesn't explain skin cancers that
occur on areas that rarely see the light of day. This
indicates that other factors may cause skin cancer,
such as being exposed to toxic substances or having
a condition that weakens your immune system.
Skin cancer begins in your skin's top layer (the
epidermis). The epidermis is a thin layer that
provides a protective cover of skin cells that your
body continually sheds.
Treatment is dependent on the specific type of
cancer,locationof the cancer,ageofthe patient, and
whether the cancer is primary or a recurrence.
Currently, surgical excision is the most common
way of treatment for skin cancers. The aim of
reconstructive surgery is restoration of normal
appearance and function. The choice of technique in
reconstruction is dictated by the size and location of
the defect.
Types of skin cancer:
There are two main types of skin cancer are
melanoma and non-melanoma. Melanoma is the
19th most commonly occurring cancer in humans.
Non- melanoma is the 5th most commonly occurring
cancer in humans. It had over 1 million diagnoses
worldwide in 2018.
Non-melanoma skin cancer has two major types:
basal cell carcinoma (BCC) and squamous cell
carcinoma (SCC).
Basal cell carcinoma (BCC) is most often a benign
form of skin cancer. It's usually caused by exposure
to ultraviolet (UV) rays, and it’s the first most
occurring form of all skin cancers.
Squamous cell carcinoma (SCC) is the second most
occurring form of all skin cancers. It is also usually
caused by exposure to ultraviolet (UV) rays from the
sun or tanning beds. SCC is a slow-growing skin
cancer and it could become hard to treat ifit spreads
to the tissues, bones, and nearby lymph nodes. But if
SCC was caught early it’s easy to be treated.
Melanoma skin cancer is not common but it is the
most serious type of skin cancer.it can be developed
anywhere on the skin but there are some areas that
has a higher risk to be affected. The most common
affected areas in men are: chest and the back. In
women, the legsarethe most common affected area
[9].
II.MATHEMATICALTHEORY
CNN are a type of Neural Network Architecture.
They are composed of the following layers: conv,
pooling, ReLU, fully Connected With at least one
convolutional layer.Theinput and output of all layers,
except fully connected layers, are 3d volumes.
Figure 1.output of the neuron
A convolution layer consists of a set of filters. Each
neuron is connected to a small volume in the input
volume of the layer (called its receptive field) that
coversa small regionverticallyandhorizontallyofthe
input volume and covers it’s full depth. As shown at
(Figure1)the output of the neuron isthe dot product
of its weights ( 𝑤𝑖) and vectorization of its receptive
field (𝑥 𝑖) plus the neuron bias𝑏.
Like other neural network architectures, gradient
descent is used to iteratively modify the weights and
biases of the CNN to minimize the loss function.
𝐿((𝑥𝑖, 𝑦𝑖), 𝑊)𝑤ℎ𝑒𝑟𝑒(𝑥 𝑖, 𝑦𝑖) are the set of training
images and labels and W being the weight matrix of
the CNN. The gradient of the loss function with
respect to the parameters of the CNN
𝛻 𝑊 𝐿((𝑥 𝑖, 𝑦𝑖), 𝑊)is first calculated using the
backpropagation algorithm. Its negative is the
direction in which nudging the parameters of the
CNN would result inthe greatestdecreaseinthe loss.
We reiterate this procedure till we find a satisfying
model. In practice for large data sets of images
(millions) a loss function that takes as input a small
batch of images randomly sampled from the data set
is used instead of the entire data set hence greatly
reducing the computational cost of learning at the
expense of some error in the gradient. This
technique is sometimes called Batch Gradient
Descent (BGD) or Stochastic Gradient Descent (SGD).
III.METHODS
It has been proved that CNN is very successful in
solving image classification problems. Research
works based on CNNimproved the best performance
for many image databases, including the NORB
database, the MNIST database and the CIFAR10
dataset [10].
It is very good at learning the global and local
structures from image data. General image objects
like human faces or hand written numbers have
obvious local and global structures, hence simple
local features can be combined to become more
complex features such as shapes, corners and
eventually the objects [11].
Recently, CNN has joined the medical imaging
analysis such as the knee cartilage segmentation.
Different from these published studies, training a
large neural network with many parameters results
a potential problem which is over-fitting training a
large neural network with many parameters.
The convolutional neural network is typically
consisting of three layers, convolution layer, pooling
layer, and fully connected layer [12]. CNN may
contain optional layers such as dropout layer to help
increasing the efficiency and reducing the over-
fitting, however, the three basic layers (convolution,
pooling, and fully connected layers) are the most
popular architecture for CNN.
Figure 2- CNN Architecture
A. Convolution Layer
The convolution layer performs the most
computational load so basically, it’s the central
structure for the CNN. The layers are like a set of
learnable filters. This layer performs a dot product
between two matrices, the first matrix is the kernels
which is a set of filters that are smaller in size than
the image itself. The kernel filter is used to add any
wanted effect to the image, for example sharpen,
unsharpen, blurs or edge detection. These filters can
be seen as a transformation of the image as
illustrated in (Figure 3).
The second matrix is window matrix of the
receptive field which is typically a fraction of image
with predefined sliding and size. Mostly the kernel
and receptive filed are extended to three channels
because most of the images are composed of three
channels (RGB). The kernel filter or matrix travel
across the input image to produce smaller image
that maps the input image.
Figure 3. Kernel matrix
B. Pooling layer
The main importance of the pooling layer in CNN is
reducing the size of the output layer which produced
from the previous layer by reducing the size of
features the computation time is reduced and the
overfitting .so simply this layer acts as a down
sampling filter, in order to do that the pooling layer
algorithm combines all the neighboring elements in
the convolution layer output matrices.
Pooling could be done by several ways like max
pooling and average pooling; however, the most
popular wayis the maxpooling which we used .in this
work, we selected a (2D) maxpooling layer with
pooling size = 2 × 2[4]. The pooling size must be
chosen as increasing the pooling dimension makes
the down sampling more important. This layer
algorithm generates one element in the output
matrix by selecting the maximum value of the 4
neighboring element in the input matrix as shown in
(Figure 4).
Figure 4. Max pooling with a 2x2 pooling size
C. Fully connected layer
The last layer is the fully connected layer also
known as the dense layer. The results of the
convolutional layers (flattened matrix) are fed
through neural layers to generate a prediction as
illustrated in (Figure 5). For example, if the image is
of a cat some features should have high probability
such as a fur feature.
Figure 5.fully connected layer
Other layers:
A. Dropout layer:
This layer lies after the pooling layer. The dropout
is a regularizationtechnique, where some number of
layer outputs (neurons) are randomly ignored or
“dropped out” during training (Figure 6) which
improves the performance.
At the start of each iteration a drop-out map is
initialized to contain the on or off state of each
corresponding neuron. The dropped neurons during
the training iterations is the ones with the off state
on the drop-out map. The neurons are turned back
on during the testing. Applying this technique forces
the network to learn features in a distributed way. It
also reduces the overfitting and improves
generalization.
Figure 6.Drop-out illustration
B. Flatten layer:
The Flatten layer transforms a two-dimensional
matrix of features into a vector that can be fed into
a fully connected neural network classifier as
illustrated in (Figure 7). The flatten layer lies
between the convolutional/maxpool layers and the
fully connected layer. It combines all the found local
features of the previous convolutional layers.
Figure 7.flattening layer
IV. EXPERIMENTAL WORK
Dataset
The dataset is taken from the ISIC (International
Skin Image Collaboration) Archive. This dataset
contains a balanced dataset of images of benign skin
moles and malignant skin moles. It consists of 1800
pictures of benign moles and 1497 pictures of
malignant classified moles. The pictures have all
been resized to low resolution (224x224x3) RGB.The
task of this kernel is to create a model, which can
classify a mole visually into benign and malignant.
Eight Steps for building the model:
Figure 8.steps for building the model
Convolutional layers:
In model 1 there is two convolutional layer and in
model 2 we used four exact convolutional layer with
the same Kernel size of model 1 which enables the
neural network to model more complex
relationships and make better analogy for curve
fitting.
There are many other ways to increase the
accuracy of a network of existing depth like shuffling
the data, RMSprop (or other associated techniques),
Dropout, Supervised layer wise training (shameless
plug), changing model hyper parameters, or
even reducing the model complexity (has worked
plenty of times for me).
In this code we has tried to detect these classes of
moles using CNN with keras tensorflow in backend
and analyzed the result to see how the model can be
beneficial.
Cross-validation
Cross-validation is a procedure used to evaluate
CNN models on a limited data sample .It is a popular
simple train/test split method. We used k (the
number of groups that a given data sample is to be
split into) equals 3. This method is dependents on
splitting dataset into k groups then train each group
at k-1 groups eachgroup is trained on k-1 groups and
tested on 1 group.
V. RESULTSANDDISCUSSION
‘Accuracy of Learning-rate’ controls how quickly or
slowly a neural network model learns a problem.
‘Accuracy of cross validation’ determines the
performance of the model after training with cross
validation method.
‘Final accuracy’ determines the performance after
testing the model .It is the final accuracy, which
shows the ability of the CNN model to recognize
benign and malignant skin cancer.
Model 1 Model 2
Number of
Convolutional
layers
2 4
Accuracy of
learning rate
Accuracy of
cross
validation
train method
with k=3
69.06% (+/- 3.90%) 75.50% (+/- 3.16%)
Final accuracy .68181 .80303
Adding more convolutional layers to CNN model:
Adding more layers will help to extract more
features. Increases the number of weights in the
network, and thus the model complexity and can
increase the accuracy. but In case of small training
set, over fitting will occur and then accuracy will be
reduced on the test data .so adding more layers is
not always the best option. There is a limit that we
should not overshoot it otherwise, we tend to ‘over fit’
the data instead of extracting features. Over fitting can
make us face some errors like wrong positives.
For more understanding of over fitting problem,
we will explain an example. Consider that we train a
model for detecting cats. If all cats’ features are
detected and we add more layers, if some of these
cats wear bells, it may start detecting this bell as a
part of the cat and this will cause a problem. If there
is a cat without bell, it may classify it as not a cat and
may sometimes classify another animal with bell as
cat. Thishappen because adding morelayers beyond
a certain threshold leads to finding irregularities in
the data.
VI. FUTURE WORK
We will try to improve the CNN model with the use of
hyper-parameters optimization to get a model with
higher feature detection and higher accuracy.
VII. CONCLUSLON
Imageclassificationreferstoa process in computer
vision that can classify an image according to its
visual content and extractinformationclassesfroma
multiband raster image. One of the most important
applications of image classification is the medical
field where the medical image data is produced by
various imaging modalities for treatment planning
and diagnosis. In this paper, we specifically
presented the skin cancer and how to detect if it is
benign or malignant. Skin cancer consists of two
main types: melanoma and non-melanoma but
melanoma is one of the most dangerous types of
cancer, the early detection and treatment for it is
critical in the therapy. With the help of the image
classification we can supply a code with huge
number of data and pictures of differenttypes of skin
cancer which helps doctors to determine if it is
dangerous or not and to maintain its way of
treatment. We also introduced the CNN which
represents a huge breakthrough in image
recognition as we used Keras sequential API to add
two convolutional layers at a time starting from the
input which enable the neural network to model
more complex relationships and make better analog
for curve fitting.
REFERENCES
[1]K. He, X. Zhang, S. Ren, and J. Sun, “Deepresidual
learning for image recognition,” in Proceedings of
the IEEE conference on computer vision and pattern
recognition, pp. 770-778, 2016.
[2] Best Accuracy Land Use/Land Cover (LULC)
Classification to Derive Crop Types Using
Multitemporal, Multisensor, and Multi-Polarization
SAR Satellite Images.
[3] Contrasting Convolutional Neural Network (CNN)
with Multi-Layer Perceptron (MLP) for Big Data
Analysis.
[4] Hough-CNN: Deep learning for segmentation of
deep brain regions in MRI and ultrasound.
[5]Deep Recurrent Neural Networks for
Hyperspectral Image Classification.
[6] Polarimetric SAR Image Classification Using Deep
Convolutional Neural Networks.
[7] User Activity Recognition in Smart Homes Using
Pattern Clustering Applied to Temporal ANN
Algorithm.
[8] Dermatologist-level classification of skin cancer
with deep neural networks.
[9] Computer Aided Melanoma Skin Cancer
Detection Using Image Processing.
[10] Combining CNN with Hand-Crafted Features for
Image Classification.
[11] Patch-based Convolutional Neural Network for
Whole Slide Tissue Image Classification.
[12] Convolutional Neural Network for Diagnosing
Skin Cancer.
[13] Medical Image Classification with Convolutional
Neural Network.
APPENDICES
Abbreviations:
 CNN stands for Convolutional Neural
Networks
 LULC stands for (land use  land cover)
 ANN stands for Artificial neural network
 RNN stands for Recursive neural network
 DNN stands for Deep neural network
 BCC stands for basal cell carcinoma
 SCC stands for squamous cell carcinoma
 UV stands for ultraviolet

More Related Content

What's hot

Convolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep LearningConvolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep LearningMohamed Loey
 
Image classification with Deep Neural Networks
Image classification with Deep Neural NetworksImage classification with Deep Neural Networks
Image classification with Deep Neural NetworksYogendra Tamang
 
Object detection with deep learning
Object detection with deep learningObject detection with deep learning
Object detection with deep learningSushant Shrivastava
 
Brain Tumour Detection.pptx
Brain Tumour Detection.pptxBrain Tumour Detection.pptx
Brain Tumour Detection.pptxRevolverRaja2
 
Multiple object detection
Multiple object detectionMultiple object detection
Multiple object detectionSAURABH KUMAR
 
Object Detection & Tracking
Object Detection & TrackingObject Detection & Tracking
Object Detection & TrackingAkshay Gujarathi
 
Object detection
Object detectionObject detection
Object detectionSomesh Vyas
 
Convolution Neural Network (CNN)
Convolution Neural Network (CNN)Convolution Neural Network (CNN)
Convolution Neural Network (CNN)Basit Rafiq
 
Object Detection using Deep Neural Networks
Object Detection using Deep Neural NetworksObject Detection using Deep Neural Networks
Object Detection using Deep Neural NetworksUsman Qayyum
 
Skin melanoma stage detection - CNN.pptx
Skin melanoma stage detection - CNN.pptxSkin melanoma stage detection - CNN.pptx
Skin melanoma stage detection - CNN.pptxVishalLabde
 
Machine Learning - Breast Cancer Diagnosis
Machine Learning - Breast Cancer DiagnosisMachine Learning - Breast Cancer Diagnosis
Machine Learning - Breast Cancer DiagnosisPramod Sharma
 
Deep learning based object detection basics
Deep learning based object detection basicsDeep learning based object detection basics
Deep learning based object detection basicsBrodmann17
 
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical Analysis
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical AnalysisBrain Tumor Segmentation using Enhanced U-Net Model with Empirical Analysis
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical AnalysisMD Abdullah Al Nasim
 
Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...
Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...
Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...Universitat Politècnica de Catalunya
 
Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...
Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...
Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...Zabir Al Nazi Nabil
 
Introduction to object detection
Introduction to object detectionIntroduction to object detection
Introduction to object detectionBrodmann17
 
Brain Tumor Detection using CNN
Brain Tumor Detection using CNNBrain Tumor Detection using CNN
Brain Tumor Detection using CNNMohammadRakib8
 

What's hot (20)

Convolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep LearningConvolutional Neural Network Models - Deep Learning
Convolutional Neural Network Models - Deep Learning
 
Image classification with Deep Neural Networks
Image classification with Deep Neural NetworksImage classification with Deep Neural Networks
Image classification with Deep Neural Networks
 
ANIMATION SEQUENCE
ANIMATION SEQUENCEANIMATION SEQUENCE
ANIMATION SEQUENCE
 
Object detection with deep learning
Object detection with deep learningObject detection with deep learning
Object detection with deep learning
 
Brain Tumour Detection.pptx
Brain Tumour Detection.pptxBrain Tumour Detection.pptx
Brain Tumour Detection.pptx
 
Multiple object detection
Multiple object detectionMultiple object detection
Multiple object detection
 
Object Detection & Tracking
Object Detection & TrackingObject Detection & Tracking
Object Detection & Tracking
 
Object detection
Object detectionObject detection
Object detection
 
Convolution Neural Network (CNN)
Convolution Neural Network (CNN)Convolution Neural Network (CNN)
Convolution Neural Network (CNN)
 
Introduction to OpenCV
Introduction to OpenCVIntroduction to OpenCV
Introduction to OpenCV
 
Object Detection using Deep Neural Networks
Object Detection using Deep Neural NetworksObject Detection using Deep Neural Networks
Object Detection using Deep Neural Networks
 
Skin melanoma stage detection - CNN.pptx
Skin melanoma stage detection - CNN.pptxSkin melanoma stage detection - CNN.pptx
Skin melanoma stage detection - CNN.pptx
 
Machine Learning - Breast Cancer Diagnosis
Machine Learning - Breast Cancer DiagnosisMachine Learning - Breast Cancer Diagnosis
Machine Learning - Breast Cancer Diagnosis
 
Deep learning based object detection basics
Deep learning based object detection basicsDeep learning based object detection basics
Deep learning based object detection basics
 
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical Analysis
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical AnalysisBrain Tumor Segmentation using Enhanced U-Net Model with Empirical Analysis
Brain Tumor Segmentation using Enhanced U-Net Model with Empirical Analysis
 
Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...
Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...
Skin Lesion Detection from Dermoscopic Images using Convolutional Neural Netw...
 
You only look once
You only look onceYou only look once
You only look once
 
Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...
Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...
Automatic Skin Lesion Segmentation and Melanoma Detection: Transfer Learning ...
 
Introduction to object detection
Introduction to object detectionIntroduction to object detection
Introduction to object detection
 
Brain Tumor Detection using CNN
Brain Tumor Detection using CNNBrain Tumor Detection using CNN
Brain Tumor Detection using CNN
 

Similar to Image Classification And Skin cancer detection

A convolutional neural network for skin cancer classification
A convolutional neural network for skin cancer classificationA convolutional neural network for skin cancer classification
A convolutional neural network for skin cancer classificationIJICTJOURNAL
 
IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...
IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...
IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...IRJET Journal
 
RECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIER
RECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIERRECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIER
RECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIERADEIJ Journal
 
Hybrid channel and spatial attention-UNet for skin lesion segmentation
Hybrid channel and spatial attention-UNet for skin lesion segmentationHybrid channel and spatial attention-UNet for skin lesion segmentation
Hybrid channel and spatial attention-UNet for skin lesion segmentationIAESIJAI
 
IRJET - Classification of Cancer Images using Deep Learning
IRJET -  	  Classification of Cancer Images using Deep LearningIRJET -  	  Classification of Cancer Images using Deep Learning
IRJET - Classification of Cancer Images using Deep LearningIRJET Journal
 
Segmentation and Classification of Skin Lesions Based on Texture Features
Segmentation and Classification of Skin Lesions Based on Texture FeaturesSegmentation and Classification of Skin Lesions Based on Texture Features
Segmentation and Classification of Skin Lesions Based on Texture FeaturesIJERA Editor
 
Skin Cancer Detection and Classification
Skin Cancer Detection and ClassificationSkin Cancer Detection and Classification
Skin Cancer Detection and ClassificationDr. Amarjeet Singh
 
Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM
Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOMComputer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM
Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOMCSCJournals
 
Comparing the performance of linear regression versus deep learning on detect...
Comparing the performance of linear regression versus deep learning on detect...Comparing the performance of linear regression versus deep learning on detect...
Comparing the performance of linear regression versus deep learning on detect...journalBEEI
 
20469-38932-1-PB.pdf
20469-38932-1-PB.pdf20469-38932-1-PB.pdf
20469-38932-1-PB.pdfIjictTeam
 
SKIN CANCER ANALYSIS USING CNN
SKIN CANCER ANALYSIS USING CNNSKIN CANCER ANALYSIS USING CNN
SKIN CANCER ANALYSIS USING CNNIRJET Journal
 
Melanoma Skin Cancer Detection using Deep Learning
Melanoma Skin Cancer Detection using Deep LearningMelanoma Skin Cancer Detection using Deep Learning
Melanoma Skin Cancer Detection using Deep LearningIRJET Journal
 
A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...
A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...
A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...Editor IJCATR
 
97202107
9720210797202107
97202107IJRAT
 
IRJET- Skin Cancer Prediction using Image Processing and Deep Learning
IRJET- Skin Cancer Prediction using Image Processing and Deep LearningIRJET- Skin Cancer Prediction using Image Processing and Deep Learning
IRJET- Skin Cancer Prediction using Image Processing and Deep LearningIRJET Journal
 
IRJET -Malignancy Detection using Pattern Recognition and ANNS
IRJET -Malignancy Detection using Pattern Recognition and ANNSIRJET -Malignancy Detection using Pattern Recognition and ANNS
IRJET -Malignancy Detection using Pattern Recognition and ANNSIRJET Journal
 
Brain tumor detection using cnn
Brain tumor detection using cnnBrain tumor detection using cnn
Brain tumor detection using cnndrubosaha
 
IRJET-Bone Tumor Detection from MRI Images using Machine Learning: A Review
IRJET-Bone Tumor Detection from MRI  Images using Machine Learning: A ReviewIRJET-Bone Tumor Detection from MRI  Images using Machine Learning: A Review
IRJET-Bone Tumor Detection from MRI Images using Machine Learning: A ReviewIRJET Journal
 
Convolutional Neural Networks
Convolutional Neural NetworksConvolutional Neural Networks
Convolutional Neural NetworksTayleeGray
 
Computer Aided System for Detection and Classification of Breast Cancer
Computer Aided System for Detection and Classification of Breast CancerComputer Aided System for Detection and Classification of Breast Cancer
Computer Aided System for Detection and Classification of Breast CancerIJITCA Journal
 

Similar to Image Classification And Skin cancer detection (20)

A convolutional neural network for skin cancer classification
A convolutional neural network for skin cancer classificationA convolutional neural network for skin cancer classification
A convolutional neural network for skin cancer classification
 
IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...
IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...
IRJET - Automated Segmentation and Detection of Melanoma Skin Cancer using De...
 
RECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIER
RECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIERRECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIER
RECOGNITION OF SKIN CANCER IN DERMOSCOPIC IMAGES USING KNN CLASSIFIER
 
Hybrid channel and spatial attention-UNet for skin lesion segmentation
Hybrid channel and spatial attention-UNet for skin lesion segmentationHybrid channel and spatial attention-UNet for skin lesion segmentation
Hybrid channel and spatial attention-UNet for skin lesion segmentation
 
IRJET - Classification of Cancer Images using Deep Learning
IRJET -  	  Classification of Cancer Images using Deep LearningIRJET -  	  Classification of Cancer Images using Deep Learning
IRJET - Classification of Cancer Images using Deep Learning
 
Segmentation and Classification of Skin Lesions Based on Texture Features
Segmentation and Classification of Skin Lesions Based on Texture FeaturesSegmentation and Classification of Skin Lesions Based on Texture Features
Segmentation and Classification of Skin Lesions Based on Texture Features
 
Skin Cancer Detection and Classification
Skin Cancer Detection and ClassificationSkin Cancer Detection and Classification
Skin Cancer Detection and Classification
 
Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM
Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOMComputer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM
Computer Vision for Skin Cancer Diagnosis and Recognition using RBF and SOM
 
Comparing the performance of linear regression versus deep learning on detect...
Comparing the performance of linear regression versus deep learning on detect...Comparing the performance of linear regression versus deep learning on detect...
Comparing the performance of linear regression versus deep learning on detect...
 
20469-38932-1-PB.pdf
20469-38932-1-PB.pdf20469-38932-1-PB.pdf
20469-38932-1-PB.pdf
 
SKIN CANCER ANALYSIS USING CNN
SKIN CANCER ANALYSIS USING CNNSKIN CANCER ANALYSIS USING CNN
SKIN CANCER ANALYSIS USING CNN
 
Melanoma Skin Cancer Detection using Deep Learning
Melanoma Skin Cancer Detection using Deep LearningMelanoma Skin Cancer Detection using Deep Learning
Melanoma Skin Cancer Detection using Deep Learning
 
A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...
A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...
A Novel DBSCAN Approach to Identify Microcalcifications in Cancer Images with...
 
97202107
9720210797202107
97202107
 
IRJET- Skin Cancer Prediction using Image Processing and Deep Learning
IRJET- Skin Cancer Prediction using Image Processing and Deep LearningIRJET- Skin Cancer Prediction using Image Processing and Deep Learning
IRJET- Skin Cancer Prediction using Image Processing and Deep Learning
 
IRJET -Malignancy Detection using Pattern Recognition and ANNS
IRJET -Malignancy Detection using Pattern Recognition and ANNSIRJET -Malignancy Detection using Pattern Recognition and ANNS
IRJET -Malignancy Detection using Pattern Recognition and ANNS
 
Brain tumor detection using cnn
Brain tumor detection using cnnBrain tumor detection using cnn
Brain tumor detection using cnn
 
IRJET-Bone Tumor Detection from MRI Images using Machine Learning: A Review
IRJET-Bone Tumor Detection from MRI  Images using Machine Learning: A ReviewIRJET-Bone Tumor Detection from MRI  Images using Machine Learning: A Review
IRJET-Bone Tumor Detection from MRI Images using Machine Learning: A Review
 
Convolutional Neural Networks
Convolutional Neural NetworksConvolutional Neural Networks
Convolutional Neural Networks
 
Computer Aided System for Detection and Classification of Breast Cancer
Computer Aided System for Detection and Classification of Breast CancerComputer Aided System for Detection and Classification of Breast Cancer
Computer Aided System for Detection and Classification of Breast Cancer
 

Recently uploaded

Microscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptxMicroscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptxpurnimasatapathy1234
 
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...srsj9000
 
Churning of Butter, Factors affecting .
Churning of Butter, Factors affecting  .Churning of Butter, Factors affecting  .
Churning of Butter, Factors affecting .Satyam Kumar
 
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICSAPPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICSKurinjimalarL3
 
complete construction, environmental and economics information of biomass com...
complete construction, environmental and economics information of biomass com...complete construction, environmental and economics information of biomass com...
complete construction, environmental and economics information of biomass com...asadnawaz62
 
Biology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptxBiology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptxDeepakSakkari2
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.eptoze12
 
Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024hassan khalil
 
Application of Residue Theorem to evaluate real integrations.pptx
Application of Residue Theorem to evaluate real integrations.pptxApplication of Residue Theorem to evaluate real integrations.pptx
Application of Residue Theorem to evaluate real integrations.pptx959SahilShah
 
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...VICTOR MAESTRE RAMIREZ
 
Internship report on mechanical engineering
Internship report on mechanical engineeringInternship report on mechanical engineering
Internship report on mechanical engineeringmalavadedarshan25
 
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...Soham Mondal
 
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)dollysharma2066
 
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort serviceGurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort servicejennyeacort
 
SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )Tsuyoshi Horigome
 
Electronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdfElectronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdfme23b1001
 
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerStudy on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerAnamika Sarkar
 
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfCCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfAsst.prof M.Gokilavani
 
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130Suhani Kapoor
 

Recently uploaded (20)

Microscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptxMicroscopic Analysis of Ceramic Materials.pptx
Microscopic Analysis of Ceramic Materials.pptx
 
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
Gfe Mayur Vihar Call Girls Service WhatsApp -> 9999965857 Available 24x7 ^ De...
 
Churning of Butter, Factors affecting .
Churning of Butter, Factors affecting  .Churning of Butter, Factors affecting  .
Churning of Butter, Factors affecting .
 
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICSAPPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
 
complete construction, environmental and economics information of biomass com...
complete construction, environmental and economics information of biomass com...complete construction, environmental and economics information of biomass com...
complete construction, environmental and economics information of biomass com...
 
Biology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptxBiology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptx
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.
 
Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024
 
Application of Residue Theorem to evaluate real integrations.pptx
Application of Residue Theorem to evaluate real integrations.pptxApplication of Residue Theorem to evaluate real integrations.pptx
Application of Residue Theorem to evaluate real integrations.pptx
 
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Serviceyoung call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
 
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
 
Internship report on mechanical engineering
Internship report on mechanical engineeringInternship report on mechanical engineering
Internship report on mechanical engineering
 
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
OSVC_Meta-Data based Simulation Automation to overcome Verification Challenge...
 
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
Call Us ≽ 8377877756 ≼ Call Girls In Shastri Nagar (Delhi)
 
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort serviceGurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
 
SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )
 
Electronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdfElectronically Controlled suspensions system .pdf
Electronically Controlled suspensions system .pdf
 
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerStudy on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
 
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfCCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
 
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
VIP Call Girls Service Hitech City Hyderabad Call +91-8250192130
 

Image Classification And Skin cancer detection

  • 1. Abstract-Image Classification in medical imaging applications is extremely important. In our work, we have used deep learning with CNN to build a model to classify skin cancer to detect if it’s benign or malignant. Datasets of skin cancer is challenging because of format and shape. It’s required effort to detect different features.Wetrainedandtestedour model on a dataset from the ISIC (International Skin Image Collaboration) Archive. The first step is to prepare data of images with less noise to make it easy for detection. Then, building CNN model with four convolutional layers, two max pooling layers, two dropout layers and 1 flatten layer, two dense layers .The results of the test produced accuracy of .8030 .withmoreefforts we canbuildmoreefficient model to get higher accuracy. Index Terms-CNN, Skin Cancer, CNN Layers, Convolution layer, Image Classification, Deep Learning, Malignant Skin Cancer, Benign Skin Cancer, Neural Network, Neurons, Kernel Filters, Cross-Validation. I.INTRODUCTION mage classification is a computer vision problem[1]. It is assigning pixels in the image into several classes or categories of LULC (land use land cover)based on the applicationof statistical decision rules or logical decision rules [2]. In order to classify a dataset into differentclasses or categoriesthe data and the classes’ relationship must be well- understood. so image classifier depends on the role of discriminant whether discriminant value highest for one class and lower for other and this is called multiclass (like the case of recognition of characters or digits from bitmaps of scanned text) or discriminantvaluepositivefor one classand negative for another which is called two class (like the case of distinguishing between text and graphics in scanned document). To achieve this, computer must be trained as training is the main key of the classification success. Important aspects of accurate classification is feature sets which are data elements attributes based on assigning elements in to various classes, it could be quantitative or qualitative and learning techniques that could be supervised learning or unsupervised learning. We have chosen one of the algorithms used in image classification which is CNN (Convolutional neural network) [3] [4] and we choose to handle the I Image Classification Skin cancer detection with CNN Eman Othman Fathy , Dina Alaa Ahmed , Dai Alaa Hassan, Mohammed IbrahimGab Allah, Mohammed Cherif Mohammed, Nihal Mansour Abd-Elbari,Hager IsmailAbd-Elaty. CairoUniversityFacultyof Engineering –December,2019
  • 2. problems this algorithms faces in the cancer an especially skin cancer. In the coming sections you will see why we choose CNN specifically and why skin cancer. Why CNN? There are many types of neural networks out there like ANN (Artificial neural network), RNN (Recursive neural network) [5], DNN (Deep neural network) [6] and lastly but not last CNN (Convolutional Neural Network) which we choose specifically because it is the best for image classification tasks as ANN is best for pattern recognition [7] and other neural networks are best for other specific application. CNN is a supervised learning algorithm and it is different from other neural networks that is has at least one convolutional layer. The convolution layer focuses on detecting lower level features such as edge detection and gradient orientation and others. CNN has manyadvantagessuch that it trainsfaster, needs fewer samples and reduces the chances of overfitting. CNN requires less preprocessing compared to other classification algorithms. It can also learn filters itself while other primitive methods need them to be hand-engineered. Brief introduction about cancer: Cancer occurs when normal cells start dividing uncontrollably and spread into surrounding tissues. Some of cancer basics are:  A mass called a tumor form due to the multiplication of the cells.  If a tumor is malignant then it’s cancerous. This means that they invade neighboring tissues because of their uncontrolled growth.  Tumors can also travel to remote organs throw the bloodstream.  Tumors also invade the surrounding tissue's space and take the oxygen and nutrients they need to survive and function. Skin cancer: Skin cancer is the most common of all human cancers. It is abnormal growth of skin cells. Most often occurs on skin exposed to the sun. But it can also occur on areas of your skin not ordinarily exposed to sunlight (or specifically ultraviolet (UV) radiation) [8]. Much of the damage to DNA in skin cells results fromultraviolet(UV) radiation that found in sunlight. But sun exposure doesn't explain skin cancers that occur on areas that rarely see the light of day. This indicates that other factors may cause skin cancer, such as being exposed to toxic substances or having a condition that weakens your immune system. Skin cancer begins in your skin's top layer (the epidermis). The epidermis is a thin layer that provides a protective cover of skin cells that your body continually sheds. Treatment is dependent on the specific type of cancer,locationof the cancer,ageofthe patient, and whether the cancer is primary or a recurrence. Currently, surgical excision is the most common way of treatment for skin cancers. The aim of reconstructive surgery is restoration of normal appearance and function. The choice of technique in reconstruction is dictated by the size and location of the defect. Types of skin cancer: There are two main types of skin cancer are melanoma and non-melanoma. Melanoma is the 19th most commonly occurring cancer in humans. Non- melanoma is the 5th most commonly occurring cancer in humans. It had over 1 million diagnoses worldwide in 2018. Non-melanoma skin cancer has two major types: basal cell carcinoma (BCC) and squamous cell carcinoma (SCC). Basal cell carcinoma (BCC) is most often a benign form of skin cancer. It's usually caused by exposure to ultraviolet (UV) rays, and it’s the first most occurring form of all skin cancers.
  • 3. Squamous cell carcinoma (SCC) is the second most occurring form of all skin cancers. It is also usually caused by exposure to ultraviolet (UV) rays from the sun or tanning beds. SCC is a slow-growing skin cancer and it could become hard to treat ifit spreads to the tissues, bones, and nearby lymph nodes. But if SCC was caught early it’s easy to be treated. Melanoma skin cancer is not common but it is the most serious type of skin cancer.it can be developed anywhere on the skin but there are some areas that has a higher risk to be affected. The most common affected areas in men are: chest and the back. In women, the legsarethe most common affected area [9]. II.MATHEMATICALTHEORY CNN are a type of Neural Network Architecture. They are composed of the following layers: conv, pooling, ReLU, fully Connected With at least one convolutional layer.Theinput and output of all layers, except fully connected layers, are 3d volumes. Figure 1.output of the neuron A convolution layer consists of a set of filters. Each neuron is connected to a small volume in the input volume of the layer (called its receptive field) that coversa small regionverticallyandhorizontallyofthe input volume and covers it’s full depth. As shown at (Figure1)the output of the neuron isthe dot product of its weights ( 𝑤𝑖) and vectorization of its receptive field (𝑥 𝑖) plus the neuron bias𝑏. Like other neural network architectures, gradient descent is used to iteratively modify the weights and biases of the CNN to minimize the loss function. 𝐿((𝑥𝑖, 𝑦𝑖), 𝑊)𝑤ℎ𝑒𝑟𝑒(𝑥 𝑖, 𝑦𝑖) are the set of training images and labels and W being the weight matrix of the CNN. The gradient of the loss function with respect to the parameters of the CNN 𝛻 𝑊 𝐿((𝑥 𝑖, 𝑦𝑖), 𝑊)is first calculated using the backpropagation algorithm. Its negative is the direction in which nudging the parameters of the CNN would result inthe greatestdecreaseinthe loss. We reiterate this procedure till we find a satisfying model. In practice for large data sets of images (millions) a loss function that takes as input a small batch of images randomly sampled from the data set is used instead of the entire data set hence greatly reducing the computational cost of learning at the expense of some error in the gradient. This technique is sometimes called Batch Gradient Descent (BGD) or Stochastic Gradient Descent (SGD). III.METHODS It has been proved that CNN is very successful in solving image classification problems. Research works based on CNNimproved the best performance for many image databases, including the NORB database, the MNIST database and the CIFAR10 dataset [10]. It is very good at learning the global and local structures from image data. General image objects like human faces or hand written numbers have obvious local and global structures, hence simple local features can be combined to become more complex features such as shapes, corners and eventually the objects [11].
  • 4. Recently, CNN has joined the medical imaging analysis such as the knee cartilage segmentation. Different from these published studies, training a large neural network with many parameters results a potential problem which is over-fitting training a large neural network with many parameters. The convolutional neural network is typically consisting of three layers, convolution layer, pooling layer, and fully connected layer [12]. CNN may contain optional layers such as dropout layer to help increasing the efficiency and reducing the over- fitting, however, the three basic layers (convolution, pooling, and fully connected layers) are the most popular architecture for CNN. Figure 2- CNN Architecture A. Convolution Layer The convolution layer performs the most computational load so basically, it’s the central structure for the CNN. The layers are like a set of learnable filters. This layer performs a dot product between two matrices, the first matrix is the kernels which is a set of filters that are smaller in size than the image itself. The kernel filter is used to add any wanted effect to the image, for example sharpen, unsharpen, blurs or edge detection. These filters can be seen as a transformation of the image as illustrated in (Figure 3). The second matrix is window matrix of the receptive field which is typically a fraction of image with predefined sliding and size. Mostly the kernel and receptive filed are extended to three channels because most of the images are composed of three channels (RGB). The kernel filter or matrix travel across the input image to produce smaller image that maps the input image. Figure 3. Kernel matrix B. Pooling layer The main importance of the pooling layer in CNN is reducing the size of the output layer which produced from the previous layer by reducing the size of features the computation time is reduced and the overfitting .so simply this layer acts as a down sampling filter, in order to do that the pooling layer algorithm combines all the neighboring elements in the convolution layer output matrices. Pooling could be done by several ways like max pooling and average pooling; however, the most popular wayis the maxpooling which we used .in this work, we selected a (2D) maxpooling layer with pooling size = 2 × 2[4]. The pooling size must be chosen as increasing the pooling dimension makes the down sampling more important. This layer algorithm generates one element in the output matrix by selecting the maximum value of the 4 neighboring element in the input matrix as shown in (Figure 4). Figure 4. Max pooling with a 2x2 pooling size
  • 5. C. Fully connected layer The last layer is the fully connected layer also known as the dense layer. The results of the convolutional layers (flattened matrix) are fed through neural layers to generate a prediction as illustrated in (Figure 5). For example, if the image is of a cat some features should have high probability such as a fur feature. Figure 5.fully connected layer Other layers: A. Dropout layer: This layer lies after the pooling layer. The dropout is a regularizationtechnique, where some number of layer outputs (neurons) are randomly ignored or “dropped out” during training (Figure 6) which improves the performance. At the start of each iteration a drop-out map is initialized to contain the on or off state of each corresponding neuron. The dropped neurons during the training iterations is the ones with the off state on the drop-out map. The neurons are turned back on during the testing. Applying this technique forces the network to learn features in a distributed way. It also reduces the overfitting and improves generalization. Figure 6.Drop-out illustration B. Flatten layer: The Flatten layer transforms a two-dimensional matrix of features into a vector that can be fed into a fully connected neural network classifier as illustrated in (Figure 7). The flatten layer lies between the convolutional/maxpool layers and the fully connected layer. It combines all the found local features of the previous convolutional layers. Figure 7.flattening layer IV. EXPERIMENTAL WORK Dataset The dataset is taken from the ISIC (International Skin Image Collaboration) Archive. This dataset contains a balanced dataset of images of benign skin moles and malignant skin moles. It consists of 1800 pictures of benign moles and 1497 pictures of malignant classified moles. The pictures have all been resized to low resolution (224x224x3) RGB.The task of this kernel is to create a model, which can classify a mole visually into benign and malignant.
  • 6. Eight Steps for building the model: Figure 8.steps for building the model Convolutional layers: In model 1 there is two convolutional layer and in model 2 we used four exact convolutional layer with the same Kernel size of model 1 which enables the neural network to model more complex relationships and make better analogy for curve fitting. There are many other ways to increase the accuracy of a network of existing depth like shuffling the data, RMSprop (or other associated techniques), Dropout, Supervised layer wise training (shameless plug), changing model hyper parameters, or even reducing the model complexity (has worked plenty of times for me). In this code we has tried to detect these classes of moles using CNN with keras tensorflow in backend and analyzed the result to see how the model can be beneficial. Cross-validation Cross-validation is a procedure used to evaluate CNN models on a limited data sample .It is a popular simple train/test split method. We used k (the number of groups that a given data sample is to be split into) equals 3. This method is dependents on splitting dataset into k groups then train each group at k-1 groups eachgroup is trained on k-1 groups and tested on 1 group. V. RESULTSANDDISCUSSION ‘Accuracy of Learning-rate’ controls how quickly or slowly a neural network model learns a problem. ‘Accuracy of cross validation’ determines the performance of the model after training with cross validation method. ‘Final accuracy’ determines the performance after testing the model .It is the final accuracy, which shows the ability of the CNN model to recognize benign and malignant skin cancer.
  • 7. Model 1 Model 2 Number of Convolutional layers 2 4 Accuracy of learning rate Accuracy of cross validation train method with k=3 69.06% (+/- 3.90%) 75.50% (+/- 3.16%) Final accuracy .68181 .80303 Adding more convolutional layers to CNN model: Adding more layers will help to extract more features. Increases the number of weights in the network, and thus the model complexity and can increase the accuracy. but In case of small training set, over fitting will occur and then accuracy will be reduced on the test data .so adding more layers is not always the best option. There is a limit that we should not overshoot it otherwise, we tend to ‘over fit’ the data instead of extracting features. Over fitting can make us face some errors like wrong positives. For more understanding of over fitting problem, we will explain an example. Consider that we train a model for detecting cats. If all cats’ features are detected and we add more layers, if some of these cats wear bells, it may start detecting this bell as a part of the cat and this will cause a problem. If there is a cat without bell, it may classify it as not a cat and may sometimes classify another animal with bell as cat. Thishappen because adding morelayers beyond a certain threshold leads to finding irregularities in the data. VI. FUTURE WORK We will try to improve the CNN model with the use of hyper-parameters optimization to get a model with higher feature detection and higher accuracy. VII. CONCLUSLON Imageclassificationreferstoa process in computer vision that can classify an image according to its visual content and extractinformationclassesfroma multiband raster image. One of the most important applications of image classification is the medical field where the medical image data is produced by various imaging modalities for treatment planning
  • 8. and diagnosis. In this paper, we specifically presented the skin cancer and how to detect if it is benign or malignant. Skin cancer consists of two main types: melanoma and non-melanoma but melanoma is one of the most dangerous types of cancer, the early detection and treatment for it is critical in the therapy. With the help of the image classification we can supply a code with huge number of data and pictures of differenttypes of skin cancer which helps doctors to determine if it is dangerous or not and to maintain its way of treatment. We also introduced the CNN which represents a huge breakthrough in image recognition as we used Keras sequential API to add two convolutional layers at a time starting from the input which enable the neural network to model more complex relationships and make better analog for curve fitting. REFERENCES [1]K. He, X. Zhang, S. Ren, and J. Sun, “Deepresidual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770-778, 2016. [2] Best Accuracy Land Use/Land Cover (LULC) Classification to Derive Crop Types Using Multitemporal, Multisensor, and Multi-Polarization SAR Satellite Images. [3] Contrasting Convolutional Neural Network (CNN) with Multi-Layer Perceptron (MLP) for Big Data Analysis. [4] Hough-CNN: Deep learning for segmentation of deep brain regions in MRI and ultrasound. [5]Deep Recurrent Neural Networks for Hyperspectral Image Classification. [6] Polarimetric SAR Image Classification Using Deep Convolutional Neural Networks. [7] User Activity Recognition in Smart Homes Using Pattern Clustering Applied to Temporal ANN Algorithm. [8] Dermatologist-level classification of skin cancer with deep neural networks. [9] Computer Aided Melanoma Skin Cancer Detection Using Image Processing. [10] Combining CNN with Hand-Crafted Features for Image Classification. [11] Patch-based Convolutional Neural Network for Whole Slide Tissue Image Classification. [12] Convolutional Neural Network for Diagnosing Skin Cancer. [13] Medical Image Classification with Convolutional Neural Network. APPENDICES Abbreviations:  CNN stands for Convolutional Neural Networks  LULC stands for (land use land cover)  ANN stands for Artificial neural network  RNN stands for Recursive neural network  DNN stands for Deep neural network  BCC stands for basal cell carcinoma  SCC stands for squamous cell carcinoma  UV stands for ultraviolet