Major project

PHASE I
CONTENTS

 Introduction

 Problem Statement

 Literature Survey

 Proposed system

 Tools Required

 Conclusion
LITERATURE
SURVEY
Automatic Lesion Detection System (ALDS) for Skin Cancer Classification Using SVM and Neural
Classifiers

 In the proposed research work of cancerous mole, evaluation and classification system has been

implemented for the detection of malignant melanoma.

 The algorithm consists of five steps starting from the input phase of preprocessing ranging to the analysis

in the form of likelihood of Lesion Malignancy:

 Preprocessing, segmentation of cancer mole, feature extraction, feature selection and classification.

 Model used: Supervised learning techniques. i.e., Support Vector Machines (SVM) and the Artificial

Neural Networks (ANN). SVM assigns weights to all the selected features excluding diameter as it will

vary for each test image.

 Accuracy of this is approximated to be 80%

Conclusion:

 Active contours and watershed approaches are used to segment out the cancerous area automatically
from the dataset image with increased efficiency, whereas the classification of cancer mole using SVM
was practiced using research findings.

 On the other hand, ANN classifier is implemented as second level classifier to witness the results
obtained from SVM and to check the cases where SVM fails to classify.
Meta-learning for skin cancer detection using Deep Learning techniques

 For this study, a ResNet50 model was used for all the experiments. This is a variation of the ResNet
architecture that consists of 50 convolutional layers.

 A set of three different experiments were conducted for this study. These experiments are listed
below. 

• Bias detection

• Bench-marking experiments on ResNet model

• Pre-trained ResNet model

 The experiment consists of measuring the performance of a classification model on recognizing which
dataset an image comes from.
 For this experiment, a ResNet model was used with random weights initialization.
 
 Results:

• The results obtained with these experiments demonstrate the benefit of using knowledge of
an unrelated domain to predict the target class of the medical data as compared with the
performance obtained with random initialization. (needs trimming)

 Conclusion:

• The results from this study suggest that popular deep learning models, such as ResNet, can
extract knowledge of data from everyday objects and generalize for the classification of
medical data, specifically the skin cancer moles, tackling one of the main challenges in the
application of deep learning in medical domains, the scarcity of the data.
Tailored Deep Neural Network Designs for Detection of Skin Cancer from Dermoscopy Images

 They introduce CancerNet-SCa, a suite of deep neural network designs tailored for the detection

of skin cancer from dermoscopy images

 International Skin Imaging Collaboration (ISIC) dataset is used, which is an open-source public

access archive of skin images consisting of 23,906 dermoscopy images.

 Methods:

• Data preparation

• Machine design

• Network architecture
 Result and conclusion:
• The benefits of computer-aided strategies such as CancerNet-SCa are two-fold.
• Not only do they provide dermatologists with valuable second opinions during diagnosis, they
also save time by acting as pre-screening tools in the diagnostic process.
Transfer Learning with Ensembles of Deep Neural Networks for Skin Cancer Detection in Imbalanced Data
Sets

 In this paper the author suggest two methods called Transfer learning and Ensembles learning of neural
networks
 Transfer learning: A way to exploit available training data collected for a different classification task than the
one at hand.
 Ensembles learning: The idea is to combine the output of multiple classifiers, called base-learners, by using
their outputs as the input to another model that is called a meta-learner, to obtain a consensus classification.
 Number of CNN base learners are trained on input images (scaled to size between 32x32 pixels to 256x256
pixels)
 Two out of 6 base learners are pre-trained with different datasets.
 All prediction from base learner is provided as an input to SVM meta learner to obtain final
classification

Conclusion:
The proposed method compared favourably against other machine learning based techniques including
three deep learning based techniques, making it a promising approach for skin cancer detection
especially on imbalanced datasets.
RECENT ADVANCES IN DEEP LEARNING APPLIED TO SKIN CANCER DETECTION

 An overview of the recent advances reported in this field as well as a discussion about the challenges and

opportunities for improvement in the current models.

 Deep learning models convolutional Neural Networks (CNN) have been achieving remarkable results in

this field.

 Another trend in this field is to adopt an ensemble of deep models instead of a single method.

 Limitations:

• One of the main concerns of applying deep learning for this task is the lack of training data.

• Lack of open clinical data is a limiting factor for this task.

Fig.3.1 Summary of recent deep learning models proposed for the skin cancer detection
 Conclusion:
• Despite the remarkable results reported, we indicated that there are rooms for improvement,
especially for the way the results should be presented.
• In this context, we believe that in the future this task needs to be addressed as a variant of the
visual and question answering (VQA) problem.
• Addressing skin cancer detection as a VQA problem increases the difficulty of the problem.
• Another aspect they believe will become a trend soon is the use of three types of skin cancer
images: clinical, dermoscopic and histopathological.
• The use of lighter models is necessary to make the apps available in remote places.
A Smartphone based Application for Skin Cancer Classification Using Deep Learning with Clinical
Images and Lesion Information

 This application is based on a Convolutional Neural Network (CNN) trained on clinical images and
patients demographics, both collected from smartphones.

 Methods:
1.Deep model to classify skin cancer
2.Data Balancing
 
 Result:
• The results obtained with clinical information presents an average balanced accuracy of 85% and a
recall of 96%
• Regarding the data balancing approach, the weighted loss function presented the best results, but
the approach based on the mutation operator of differential evolution is competitive