8th sem.pptx

BANGALORE TECHNOLOGICAL INSTITUTE
Department of Computer Science & Engineering
SKIN CANCER DETECTION USING
DEEPLEARNING
BY
_________________
Prasant Poudel 1BH19CS069
Sanjay Malla 1BH19CS094
Snehal Karki 1BH19CS103
Jitendra Kohar 1BH19CS034
VTU FINAL YEAR PROJECT
Under the Guidance of
Mrs.Manjula Devi P
(Assistant Prof.)

ABSTRACT
Skin cancer is a major public health concern, and early detection is key to improving patient outcomes. In this
project, we propose to develop a deep learning-based system for the detection of skin cancer. The system will
be trained on a large dataset of images of skin lesions, and will use convolutional neural networks (CNNs) to
classify images as benign or malignant. The system will be evaluated on a test dataset and compared to the
performance of human experts. We expect that the deep learning system will be able to achieve higher levels
of accuracy, speed, and efficiency than current methods, and will have the potential to improve access to skin
cancer diagnosis in resource-limited settings. Overall, the goal of this project is to contribute to the early
detection and treatment of skin cancer, and to improve patient outcomes.

INTRODUCTION
Skin cancer is the most common type of cancer worldwide, and early detection is crucial for effective treatment.
Deep learning, a subset of machine learning, has shown great promise in the field of medical image analysis,
including skin cancer detection.
Deep learning models use large amounts of data to learn patterns and make predictions, and have been trained on
extensive datasets of skin images to identify potential skin lesions that could be cancerous. These models have been
shown to achieve high accuracy in identifying skin cancer, often outperforming human experts.
There are different types of skin cancer, including melanoma, basal cell carcinoma, and squamous cell carcinoma.
Each type of cancer presents differently on the skin, and deep learning models can be trained to identify specific
characteristics of each type. For example, melanomas often have irregular borders, varied coloration, and an
asymmetrical shape, while basal cell carcinomas often have a pearly or translucent appearance and a rolled edge.

We are focusing on the category only..
bkl=Benign keratosis (solar lentigo / seborrheic keratosis / lichen planus-like keratosis)
nv=Melanocytic nevus
df=Dermatofibroma
mel=Melanoma
vasc=Vascular lesion
bcc=Basal cell carcinoma
akiec=Actinic keratosis
7 classes of skin cancer

Title Author Year Summary Drawback
Skin Cancer
Detection Using
Combine
Decision of
Deep Learner
Azhar Imran, Arslan
Nasir, Muhammad
Bilal,Guangmin
Sun,Abdulkareem
Alzahrani And
Abdullah Almuhaimeed
2022 The proposed deep learning-based
ensemble approach is developed in
two stages. In the first stage, three
deep learning models of VGG,
CapsNet, and ResNet have been
developed using malignant and
benign images obtained from the
International Skin Imaging
Collaboration (ISIC) skin cancer
images repository. In the second
stage, the findings of deep learners
have been combined using majority
weighting. The accuracy of the model
is 95%.
The model was train
for only binary
classification which
is either cancer or
not a cancer output.
And the number of
image which is used
is less.
LITERATURE SURVEY

Automatic Skin
Cancer Detection in
Dermoscopy Images
Based on Ensemble
Lightweight Deep
Learning Network
Lisheng WeiI, Kun Ding
And Huosheng
Hu
2020 In this paper, they proposed an
efficient and lightweight melanoma
classification network based on
MobileNet, DenseNet-121. The
proposed dermoscopy image lesion
recognition method includes three
steps: image preprocessing, model
construction and model training, and
model fusion.
They train the model
on ISIC 2016 which is
older dataset and that
contain 900 image
only. And as a input
pair of positive and
negative data is
should be given.

A Deep Learning
Approach Based on
Explainable Artificial
Intelligence for Skin
Lesion Classification
Muhammad
Umar,Natasha
Nigar,Muhammad
Kashif
Shahzad,Shahid
Islam And Douhadji
Abalo
2022 Resnet-18 transfer learning
algorithm is used for this model.
The dataset ISIC 1019 was used for
training the model. Total 8 classes
was classified.
The hyperparameter that is set for
the training the model was:
They doesn’t use large
and different type of
dataset. They have
tried the only pretrain
model.The resizing of
image was done in
small patch which
could affect the
classifier performance.

TRADITIONAL TECHNOLOGY

EXISTING SYSTEM
1. DeepSkin: DeepSkin is a deep learning model that has been trained on a large dataset of skin images to classify
skin lesions as either benign or malignant. The model uses a convolutional neural network (CNN) to extract
features from the images and has achieved high accuracy in identifying skin cancer, with an area under the curve
(AUC) of 0.94.
2. ISIC (International Skin Imaging Collaboration) Archive: The ISIC Archive is a public database of skin images
that has been annotated by dermatologists. Researchers can use this database to train deep learning models for
skin cancer detection. The database includes images of different types of skin lesions, such as melanoma, basal
cell carcinoma, and squamous cell carcinoma.
3. SkinLesionClassifier: SkinLesionClassifier is a deep learning model that has been trained on a dataset of skin
images to classify skin lesions as benign, malignant, or unclassified. The model uses a CNN to extract features
from the images and has achieved high accuracy in identifying skin cancer, with an AUC of 0.91.
4. DeepSquint: DeepSquint is a deep learning model that has been developed to detect skin cancer using

PROPOSED SYSTEM
● In the purpose system dataset are the combination from HAM1000,,ISIC dataset present in the kaggle
website.
● The system is divided into three phase where first two is training and testing and third one is for
implementing the GUI part.
● First step is to gather the dataset from from different website.
● The dataset will be divided into 80-10-10% of training ,testing and validation purpose.
● All the dataset are of image data which are of different and different variant of cancer classes.
● Preprocessing of image is necessary for improvement of model, preprocessing include:
■ Collection of the dataset
■ Resizing the image to 224*224*3 shape
■ Formating the image to the respective classes folder
● For the training of model we are using tensorflow keras library for Deep Learning and Sklearn for
Machine learning.

● Also another option is to use Transfer Learning models which are Alex Net, VGG-16, VGG-19, Inception.
● The architecture will be as below:
Basic Architecture

PURPOSE MODEL

WORK FLOW DIAGRAM

● For the implementation part which will contain two approach which are website and mobile
application.
● For the website, the website develop react will be use having the interactive UI and animated front
end.
● For the collection of new dataset from user and user info store purpose MongoDB database will be
used.
● The website will contain basic user info form and they can login through their Email address.
● The user need to upload the image on image Section which will give the result of either cancer or
non cancer.
● If cancer is detected the what type of cancer it is. The classes of cancer data is 7.
REAL TIME IMPLEMENTATION

MODULES
1. Data Collection
2. Image Preprocessing
3. Dataset Split for training and testing
4. Training the model
5. Model Evaluation
6. Front End development

Data has been collected from publicly available dataset from Kaggle website which contain multiple type of
data.
From the website we have downloaded ISIC2019 and Ham10000 dataset. Which contains the image of 30k
of 7 different classes of cancer with the metadata containing image_id ,label,age,gender,lesion location are
available.
By combing the both dataset we got imbalance
Data where some of classes contain 10k and some of classes contain only 250 so we have to balance the
image in every classes for better accuracy.
ISIC2019 https://www.kaggle.com/datasets/salviohexia/isic-2019-skin-lesion-images-for-classification
HAM10000 https://www.kaggle.com/datasets/kmader/skin-cancer-mnist-ham10000
1.Data Collection

2.Image Preprocessing
Image preprocessing refers to the steps taken to prepare an image for further analysis or processing. Here are some
general steps for image preprocessing
1. Image Resizing: The image present in dataset in the resolution of 1024*1024*3 so we have converted it to
246*246*3.
2. Image Classification: We classify the image using the metadata csv file to the respective label directory.
3. Image Augmentation: Some of the classes doesn’t have sufficient image so using the data augmentation
additional image was generated.
4. Image Enhancement: Enhancing an image involves improving the visual quality of an image by adjusting
various parameters like contrast, brightness, and sharpness. This can be done using various enhancement
techniques like histogram equalization or contrast stretching.

Dataset split is an important step in machine learning, where a dataset is divided into subsets for training, validatio
and testing. The purpose of this is to evaluate the performance of a machine learning model and prevent overfitting.
The typical split ratio is 80:10:10, where 80% of the data is used for training, 10% for validation, and 10% for testing.
1. Training Set: The training set is used to train the model, i.e., the model is fit to this data by updating the mod
parameters using an optimization algorithm. The model learns the underlying patterns and relationships in t
training data.
2. Validation Set: The validation set is used to evaluate the performance of the model during the training phase. It
used to tune the hyperparameters of the model, such as the learning rate, regularization parameter, or the numb
of hidden layers. The model is not trained on this data; instead, it is used to estimate the generalization error
the model.
3. Testing Set: The testing set is used to evaluate the final performance of the model after the training and validati
phases. It is used to estimate the accuracy of the model in making predictions on unseen data.
3. Dataset Spliting

4. Model Training
There are several deep learning architectures that can be used for image-based skin cancer
detection, such as VGG, ResNet, and Inception. The choice of architecture depends on the size of the
dataset, the computational resources available, and the performance of the model on the validation
set.
The deep learning model is trained on the training set using an optimizer such as stochastic gradient
descent (SGD) or Adam. The loss function used for training depends on the classification problem,
but typically involves minimizing the cross-entropy loss between the predicted and true labels.
Hyperparameters such as the learning rate, batch size, and number of epochs need to be tuned to
optimize the performance of the model on the validation set.

5.Model Evaluation
Performance Metrics: The next step is to choose appropriate performance metrics. The metrics
selected should be relevant to the problem being solved and should provide meaningful insights into
the model's performance. In the case of a skin cancer detection system, the following metrics can be
used:
● Accuracy: This metric measures the percentage of correctly predicted skin cancer cases out of
all the cases.
● Precision: Precision is the fraction of true positive cases among all the positive predictions.
● Recall: Recall is the fraction of true positive cases among all the actual positive cases.
● F1 Score: The F1 score is the harmonic mean of precision and recall.

SNAPSHOT
Image Preprocessing:

Data Augmentation:

Imbalance Dataset
Balance Dataset

Model Evaluation

OBJECTIVE
The objective of a skin cancer detection project using deep learning would be to develop a system that can
accurately and efficiently identify skin cancers, such as melanoma, basal cell carcinoma, and squamous cell
carcinoma. This could be achieved by training a deep learning model on a large dataset of images of skin
lesions, and then using the model to classify new images.
There are several potential benefits to such a system. First, it could help to improve the accuracy of skin
cancer diagnosis, as deep learning models can often achieve higher levels of accuracy than human experts.
Second, it could reduce the time and cost of diagnosis, as the system can analyze images quickly and
without the need for a human expert to review them. Finally, the system could help to increase access to
skin cancer diagnosis, especially in areas where there is a shortage of trained dermatologists.
Overall, the main goal of a skin cancer detection project using deep learning would be to improve the early
detection and treatment of skin cancer, which can help to improve patient outcomes and save lives.

THANK YOU

8th sem.pptx

More Related Content

What's hot (20)

Similar to 8th sem.pptx (20)

Recently uploaded (20)

8th sem.pptx