Volume 8, Issue 8, August – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Skin Cancer Classification

Nikita Mali (Student)1; Prajwal Thawari (Student)2; Uday Jawheri (Student)3; Sulaxan Jadhav4
DY. Patil International University

Abstract:- Skin cancer is a common and dangerous form To further enhance the performance and
of cancer. This is a dangerous kind of cancer, and early generalizability of the Deep Learning model, we employ
detection is crucial for successful treatment. Malignant transfer learning techniques. By leveraging pre-trained CNN
tumors develop when healthy, normal skin cells models, such as VGGNet or ResNet, as initialization for the
experience genetic alterations and begin to grow network, this can turn out as an advantage since the
uncontrolled. An important risk factor for skin cancer is knowledge learned from vast datasets can be used for the
ultraviolet (UV) radiation from the sun and artificial specific purpose of skin cancer classification.
sources like tanning beds. Skin cancer can develop in
other places of the body, although it often occurs on the  Motivation:
face, neck, arms, and hands because of their exposure to Skin cancer is a common as well as fatal disease that
sunlight. Data imbalance issues are brought on by the affects millions of individuals worldwide. As the most
significant discrepancy between data from several prevalent kind of cancer, its prevalence has gradually
healthcare industry classifications. Deep learning models increased over time. Despite advances in medical
frequently train on one class more than others due to technology, accurate and early skin cancer diagnosis
problems with data imbalance. The dataset utilized is remains a serious difficulty. Due to a lack of access to trained
skin cancer MNIST: HAM10000, which contains seven dermatologists, time-consuming physical examinations, and
kinds of skin lesions. The seven forms of skin lesions are the subjectivity of visual judgments, delayed diagnosis and
as follows: melanocytic nevi (nv), melanoma (mel), inferior treatment outcomes are common. As a result, there is
benign keratosis (bkl), basal cell carcinoma (bcc), actinic a critical need for fast and dependable techniques of skin
keratosis (akiec), vascular lesions (vasc), and cancer categorization to aid medical professionals in correct
dermatofibroma (df). These are used to categorize skin detection and appropriate therapy.
cancer based on mutations and variations. Deep learning
models (inception v3, resnet, vgg16, and mobile net) and The motivation behind this research stems from the
deep learning techniques such as data augmentation, potential impact it can have on the healthcare sector. By using
image normalization, and image standardization were the potential of artificial intelligence and machine learning,
used to classify skin cancer. the aim is to develop a robust and automated skin cancer
classification system. A system like this might not only help
I. INTRODUCTION dermatologists in their diagnosis process, but it could also
be available in areas with low medical resources,
Skin cancer is a significant public health concern democratizing healthcare and increasing patient outcomes.
worldwide, with its incidence steadily rising over the past few
decades. It is estimated that one in every three cancers Furthermore, previous studies on skin cancer
diagnosed globally is a skin cancer. Early detection and classification has shown excellent results, demonstrating the
accurate diagnosis are pivotal in effectively managing skin effectiveness of deep learning models in image analysis
cancer cases, as they directly impact treatment outcomes tasks. However, many of these researches are constrained by
and patient survival rates. Traditional methods of diagnosing small datasets or lack full comparisons with other cutting-
skin cancer primarily rely on visual examination by edge approaches. To solve these constraints through this
dermatologists, which can be subjective and prone to human research, one solution is by leveraging a big and varied
error. In recent years, there has been a growing interest in dataset and conducting a thorough examination of several
leveraging advanced technologies, such as deep learning, to classification techniques. By doing so, it will provide
aid in the automated and accurate classification of skin significant insights to the scientific community, improving
cancer. understanding of skin cancer categorization and maybe
encouraging further advances in the field.
The paper discusses a deep learning approach
specifically designed for skin cancer classification. Deep Ultimately, the successful development of an accurate
Learning harnesses the power of convolutional neural and accessible skin cancer classification system holds the
networks (CNNs) to extract high-level features from potential to save lives, reduce healthcare costs, and alleviate
dermatoscopic images. By training on a substantial dataset of the burden on healthcare systems worldwide. It is our hope
annotated skin lesion images, Deep Learning models can that this research will pave the way for practical and
learn to differentiate between different types of skin cancer scalable solutions in the early detection and management of
based on these extracted features. skin cancer, ultimately leading to improved patient care and
outcomes.

IJISRT23AUG2018 www.ijisrt.com 2979

Volume 8, Issue 8, August – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. LITERATURE REVIEW  Skin Cancer Detection from Smartphone Imagery using
Convolutional Neural Network (7):
Skin cancer is a major worldwide health problem, with This study addressed the practical difficulty of
growing incidence rates and serious consequences for detecting skin cancer using smartphone photos. The
patient outcomes. Machine learning techniques, particularly researchers created a convolutional neural network (CNN)
deep learning, have opened up new avenues for the architecture that was refined using a smartphone picture
automated identification and categorization of skin cancer. dataset, giving promising resultsfor real-world applications.
These are the research publications that have been
investigated for various machine learning techniques in skin  Deep Learning-based Skin Cancer Classification using
cancer detection andclassification in this literature review. Transfer Learning (8):
This study adopts transfer learning techniques to
 A Deep Learning Model for Skin Cancer Detection (1): leverage pre-trained CNN models for skin cancer
This significant study showed a deep learning model classification. The research demonstrates how transfer
built on a huge dataset of skin pictures. The model learning can significantly reduce training time and improve
distinguished between benign and malignant skin lesions classification accuracy.
with outstanding accuracy. The study also developed the
idea of explainability in deep learning models, which Machine learning techniques, particularly deep
provides insights into the algorithm's decision- making learning models, were demonstrated to be successful in skin
process. cancer detection and classification in these research studies.
The use of huge and diverse datasets, along with improved
 An Ensemble Learning Approach for Skin Cancer architectures and transfer learning, has resulted in major
Classification (2): advances in this field. Many machine learning algorithms,
In this work, ensemble learning, a technique that such as as support vector machines, random forests, and
integrates many models for greater performance, was used ensemble learning approaches, have been investigated.
to increase skin cancer classification accuracy. The research Deep learning models, particularly CNNs, have emerged as
produced greater sensitivity and specificity rates than a dominant force due to their capacity to learn detailed
individual models by harnessing the capabilities of several patterns and characteristics from raw picture data
classifiers. automatically. CNNs have demonstrated outstanding
performance in image-based applications, making them an
 A Machine Learning Approach for the Detection of Skin excellent candidate for skin cancer classification, where
Cancer (3): precise feature extraction is critical.
The study investigates the use of machine learning
methods, such as support vector machines and random Several papers make use of transfer learning, a
forests, for skin cancer diagnosis. The authors obtained prominent approach that enables researchers to utilize pre-
competitive findings and emphasized the significance using trained CNN models on large-scale picture datasets and
feature engineering in boosting classification performance. fine-tune them for specific objectives such as skin cancer
classification. Transfer learning reduces the requirement for a
 Machine Learning and Its Application in Skin Cancer largeamount of data and processing resources while keeping
(4): the learnt characteristics from the original dataset. Another
This complete review paper investigated several popular strategy in skin cancer classification is ensemble
machine learning algorithms for skin cancer diagnosis, such learning, which tries to aggregate predictions from
as support vector machines, random forests, and neural numerous models to improve overall accuracy and
networks. The report examined the advantages and generalization. Ensemble learning improves robustness and
disadvantages of each methodology and gave a road map for minimizes the danger of overfitting by aggregating decisions
scholars to investigate other approaches. from several models, making it a useful method for dealing
with the intricacies of skin cancer classification problems.
 Skin Sight: AI-Powered Skin Cancer Detector (5):
Skin Sight revealed a unique mobile app that used One notable trends in the reviewed papers is Several
machine learning algorithms to detect skin cancer in studies have been conducted to look into the use of
smartphone photographs. The research emphasized the smartphone imaging for skin cancer screening, taking
potential of such apps to empower individuals for early self- advantage of the widespread availability of mobile devices.
detection and sparked debate about ethical concerns. These mobile-based technologies have the potential to
revolutionize skin cancer screening, allowing for early
 Deep Skin: A Deep Learning Approach for Skin Cancer identification and intervention, particularly in poor areas
Classification (6): with limited access to professional healthcare. In addition,
Deep Skin suggested a revolutionary deep learning several articles highlight the significance of interpretability
architecture designed exclusively for skin cancer and explainability in deep learning models for skin cancer
categorization. The model used attention methods to focus categorization. Due to the complexity of deep learning
on important aspects, resulting in enhanced interpretability models' structures, attempts have been made to add attention
and performance. mechanisms and visualization tools. These techniques
enable researchers to comprehend the model's decision-

IJISRT23AUG2018 www.ijisrt.com 2980

Volume 8, Issue 8, August – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
making process and give significant insights to  Uncertainty Estimation:
dermatologists, hence increasing confidence and adoption of Along with their forecasts, skin cancer classification
automated diagnostic systems. However, the reviewed models should be able to provide probability estimates.
papers show that machine learning has the potential to Understanding the model's degree of confidence in its
revolutionize skin cancer detection and management. diagnosis is critical for lowering false positives and false
Interdisciplinary collaboration involving medical specialists, negatives, particularly in critical scenarios.
machine learning practitioners, and ethicists will play a
critical role in molding the future of skin cancer  Integrating Multi-Modal Data:
categorization technologies as this subject evolves. The Integrating several data sources, such as dermoscopic
development of strong, accessible, and ethical AI- powered pictures, clinical information, patient history, and genetic
skin cancer classification systems has the potential to data, has the potential to improve the accuracy and
improve patient outcomes and save millions of lives by reliability of skin cancer classification models. Investigating
aiding early diagnosis. multimodal techniques may result in more complete and
robust classifiers.
III. GAPS FOUND
 Interpretability and Explainability:
 Data Imbalance and Generalization: While deep learning models have demonstrated
Many skin cancer datasets suffer from class imbalance, outstanding performance in skin cancer classification, their
with a significantly larger number of benign samples black-box nature raises questions regarding their
compared to malignant ones. Imbalanced datasets can bias interpretability. Medical practitioners and patients must
models towards the majority class, reducing their capacity to understand how the models make their judgments. As a
reliably detect uncommon cases. To make skin cancer result, research concentrating on approaches to improve the
classifications trustworthy and applicable globally, interpretability and explainability of deep learning models in
procedures for dealing with data imbalance and ensuring skin cancer classification is critical for obtaining clinician
strong generalization across varied skin types, races, and trust and acceptance.
circumstances must be developed.
Filling these research gaps will help to design more
 Mobile Application Ethical Considerations: reliable, clear, and practical machine learning models for
The rise of AI-powered mobile skin cancer screening skin cancer classification. It may make considerable
applications raises ethical questions about data privacy, progress toward early diagnosis, individualized treatment
security, and potential misdiagnosis. More study is needed to strategies, and improved patient outcomes in the battle
address the ethical issues of making such programs available against skin cancer byconstantly refining these models.
to the general public, such as informed permission, data
protection, and responsible usage standards.  Objectives:

 Validation on Diverse Populations:  Improve Interpretability and Explainability:

Some research rely solely on datasets from specific Seek to improve the deep learning model's
locations, which may restrict the generalizability of their interpretability and explainability for skin cancer
findings to a larger population. To verify the efficiency of classification. Using approaches like attention mechanisms
skin cancer classifiers across different populations, or saliency maps to gain insight into the features that
validation on varied datasets covering distinct geographic influence the model's conclusions. The goal is to create a
locations and skin types is required. more transparent and trustworthy system that can be easily
implemented in clinical settings, while also encouraging
 Integration into Clinical Workflows: collaboration between medical practitioners and AI
To be effective in real-world clinical practice, skin algorithms.
cancer classification models must seamlessly integrate into
existing clinical workflows. Understanding how  Address Data Imbalance:
dermatologists and other medical professionals can Provide ways to address data imbalance within the skin
effectively use AI-based tools to augment their decision- lesion dataset, as benign instances frequently outnumber
making process is a crucial area for future research. malignant cases. Study oversampling, undersampling, and
data augmentation strategies to guarantee that the model is
 Longitudinal Data and Disease Progression: not skewed towards the majority class. The goal is to obtain
The majority of present studies is focused on the balanced classification performance across benign and
binary categorization of benign and malignant lesions. The malignant cases while lowering classification errors.
incorporation of longitudinal data and the study of disease
development over time can give useful insights into the  Assess Model Uncertainty:
dynamic nature of skin cancer and enhance early Analyze approaches for evaluating model uncertainty to
identification and treatment planning. quantify the amount of trust in the deep learning model's
predictions. The model may generate probabilistic outputs by
measuring uncertainty, which is especially useful in crucial
or confusing instances. The goal is to boost diagnostic

IJISRT23AUG2018 www.ijisrt.com 2981

Volume 8, Issue 8, August – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
confidence and alert medical personnel to instances that  Dataset Screening:
require additional investigation or expert advice. The dataset used in this work is obtained from MNIST:
HAM10000, which includes 10015 pictures of seven distinct
 Transfer Learning and Model Optimization: types of skin lesions. This dataset serves as the foundation
Investigate transfer learning strategies that harness for the skin cancer classification task. The HAM10000
information from large-scale picture datasets using pre- dataset (Human Against Machine with 10,000 training
trained neural network designs such as VGG or ResNet. To photos) is a popular and freely available dataset for skin
speed training and perhaps increase model performance, cancer research. It contains 10,015 dermatoscopic photos of
fine-tune these pre-trained models using the skin cancer skin diseases from seven diagnostic categories, including
dataset. Furthermore, use hyperparameter optimization melanoma, nevus, seborrheic keratosis, basal cell carcinoma,
approaches to fine- tune the model architecture for best skin actinic keratosis, benign keratosis, and dermatofibroma. The
cancerclassification performance. HAM10000 dataset has been extensively utilized in the
development and testing of machine learning algorithms and
 Deployment and Usability: deep learning models for skin cancer classification. It
Consider the practical implementation of the skin includes a wide variety of lesion types, allowing researchers
cancer classification model, with an emphasis on usability, to train and test their algorithms on a large set of skin scans.
speed, and resource efficiency. To guarantee the model's The dataset has been used in a variety of scientific studies,
practicality in clinical applications, evaluate its performance allowing researchers to investigate various elements of skin
in real-time or near-real-time settings. The goal is to develop cancer detection, classification, and image processing
a user-friendly tool that fits easily into dermatologists' approaches.
workflows and assists in the efficient and accurate
identification of skin cancer.

 Evaluate Clinical Impact:

Assess the clinical effect and possible advantages of
incorporating the AI- based skin cancer categorization
system into everyday dermatological practices. Conduct
user studies in collaboration with medical specialists to
assess the influence on diagnostic accuracy, patient
outcomes, and effort reduction. The goal is to demonstrate
the created model's practical applicability and value in
improving skin cancer diagnosis and patient treatment.

 Experimentation: Fig 1 Seven Different Diagnostic Categories of Skin Lesions

 Segmentation:  Convolutional Neural Networks (CNNs):

Image segmentation techniques are used to identify and CNNs are widely used machine learning and computer
delineate regions of interest within skin lesion images, vision techniques for image classification tasks, including
enabling more precise analysis and feature extraction, Deep skin cancer detection. They are well-suited for learning
Learning Architectures: Besides CNNs, other deep learning hierarchical features from images, making them effective.
architectures like Recurrent Neural Networks (RNNs) and
Transformer-based models have been explored for Image Segmentation: Image segmentation techniques
sequential data in dermatological image analysis. are applied to identify and delineate regions of interest
within skin lesion images. This enables more precise analysis
 Dataset Balancing: and feature extraction, which can be beneficial for accurate
The skin lesion photos are resampled to balance the classification.
dataset, ensuring that each class (skin condition) hasa similar
number of samples. This step is important to prevent biases  Deep Learning Architectures:
during model training and improve overall performance. In addition to CNNs, other deep learning architectures
such as Recurrent Neural Networks (RNNs) and
 Architecture and Transfer Learning: Transformer-based models are explored for sequential data in
The methodology involves using the encoding element dermatological image analysis. These architectures may be
of a pre-trained segmentation model to extract features from useful for tasks involving sequential patterns or temporal
the skin lesion images. This pre-trained model is then fine- information in the data.
tuned to classify the seven distinct skin conditions. Transfer
learning leverages the knowledge learned from a large dataset  Model Evaluation:
to improve the model's performance on the specific skin The classification performance is evaluated on both the
lesion dataset. test set and the validation set. The accuracy achieved on the
test set is 0.82534, and on the validation set, it is 0.808204.
This indicates that the model is capable of accurately
classifying skin lesions.

IJISRT23AUG2018 www.ijisrt.com 2982

Volume 8, Issue 8, August – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IV. RESULTS AND DISCUSSION  Melanoma:
There were 41 instances of Melanoma in the dataset.
All of them were correctly classified as Mel (true positives),
as indicated by the value of 41 in the sixth row, sixth
column cell.

 Vascular Lesions (VL):

There were 17 instances of Vascular Lesions in the
dataset. All of them were correctly classified as VL (true
positives), as indicated by the value of 17 in the seventh
row, seventh column cell.

 In Conclusion, the Model Appears to Perform well on this

Under Sampled Dataset, with all Classes Accurately
Classified.

Fig 2 Confusion Matrix

 Confusion Matrix of Under Sampling in the above

Confusion Matrix of Under Sampling:
The rows represent the actual classes, while the
columns represent the predicted classes. The diagonal
elements (from the top-left to bottom-right) indicate the
correct predictions for each class. Off- diagonal elements
represent misclassifications.

 Based on the values provided in the confusion matrix,

here's a brief analysis for each class:
Fig 3 Confusion Matrix for Oversampling
 Actinic Keratoses (AK):
There were 45 instances of Actinic Keratoses in the
dataset. All of them were correctly classified as AK (true
positives), as indicated by the value of 45 in the top-left cell.

 Basel Cell Carcinoma (BCC):

There were 31 instances of Basel Cell Carcinoma in the
dataset. All of them were correctly classified as BCC (true
positives), as indicated by the value of 31 in the second row,
second column cell.

 Benign Keratoses (BK):

There were 36 instances of Benign Keratoses in the
dataset. All of them were correctly classified as BK (true
positives), as indicated by the value of 36 in the third row, Fig 4 Training and Validation Accuracy
third column cell.

 DermatoFibroma (DF):
There were 21 instances of DermatoFibroma in the
dataset. All of them were correctly classified as DF (true
positives), as indicated by the value of 21 in the fourth row,
fourthcolumn cell.

 Menatocytic Nevi (MN):

There were 38 instances of Menatocytic Nevi in the
dataset. All of them were correctly classified as MN (true
positives), as indicated by the value of 38 in the fifth row,
fifth column cell.
Fig 5 Training and Validation Loss

IJISRT23AUG2018 www.ijisrt.com 2983

Volume 8, Issue 8, August – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 By Analysing these Metrics, the Model’s Performance is FUTURE SCOPE
Evaluated:
 Improved Accuracy: As the field of deep learning
 Training Loss (0.52 or 52%): evolves, models will become more accurate and robust.
The training loss represents the error between the Advances in network designs, data augmentation
predicted values and the actual values during the training methods, and transfer learning will all help to improve
phase. A lower training loss generally indicates that the detection and classification of skin cancer.
model has learned to fit the training data well. A value of  Larger and more diverse datasets of skin lesion pictures:
0.52 suggests that the model has made considerable progress The availability of larger and more diverse datasets of
in learning from the training data. skin lesion images will allow for deeper and more
extensive training of deep learning models. Access to
 Validation Loss (0.63 or 63%): high-quality information can result in more generic
The validation loss measures the error on a separate models that perform well across a variety of populations
dataset that was not used during training. It helps to assess and skin types.
how well the model generalizes to unseen data. A validation  Real-Time Diagnosis: With improved deep learning
loss slightly higher than the training loss (0.52) is normal and models and developments in hardware capabilities, real-
expected. It indicates that the model is not overfitting the time skin cancer categorization systems may be
training data and is capable of generalizing to some extent. developed in the future. These technologies may be
easily incorporated into cellphones or mobile devices,
 Accuracy (0.79 or 79%): allowing users to undertake early skin lesion
Accuracy is the proportion of correct predictions that examinations.
the model made on the validation set. In this case, the model  Personalized Medicine: Deep learning models have the
achieved an accuracy of 79%. This percentage indicates the potential to help in customized treatment by recognizing
proportion of correctly classified instances out of all unique aspects of skin lesions that are unique to people.
instances in the validation set. This might lead to more successful outcomes if treatment
approaches are customized to a patient's individual skin
Therefore, the model's performance seems to be cancer features.
decent. The relatively close values of training loss and  Multi-Class Classification: While current deep learning
validation loss suggest that the model is not overfitting, models primarily distinguish between benign and
which is a good sign. An accuracy of 79% indicates that the malignant lesions or a few specific types of skin cancer,
model is reasonably capableof making correct predictions. future research may expand to handle more complex
multi-class classification tasks, such as distinguishing
V. CONCLUSION between various subtypes skin cancer

Skin cancer is a rapidly spreading disease that is REFERENCES

mostly caused by a person's exposure to UV radiation from
the sun. Given the limited resources available, early [1]. Sreevidya R. C.,1Jalaja G,1Sajitha N,2D. Lakshmi
diagnosis is critical. Dermatologists frequently fail to Padmaja,3S. Nagaprasad,4Kumud Pant,5and Yekula
diagnose skin cancer in its early stages, thus accurate Prasanna Kumar” Role of Artificial Intelligence
diagnosis and identification are critical for effective skin and Deep Learning in Easier Skin Cancer
cancer preventive methods. We wanted to increase the Detection through Antioxidants Present in Food”.
accuracy and efficiency of skin cancer diagnosis by using [link]
the capability of convolutional neural networks (CNNs) with [2]. O T Jones, R N Matin, M van der Schaar, K
a huge collection of skin lesion photos. We established the Prathivadi Bhayankaram, C K I Ranmuthu, M S
usefulness of Deep Skin in properly diagnosing skin lesions Islam, D Behiyat, R Boscott, N Calanzani, J Emery,
through the experimentation and review. On the test set, the H C Williams, F M Walter “Artificial intelligence
model surpassed prior state-of-the- art approaches, and machine learning algorithms for early
achieving a classification accuracy of 0.82534. detection of skin cancer in community and
primary care settings: a systematic review.” [link]
Deep Skin showed great sensitivity and specificity, [3]. M. Krishna Monika, N. Arun Vignesh, Ch. Usha
showing that it can identify both malignant and benign skin Kumari, M.N.V.S.S. Kumar, E. Laxmi Lydia “Skin
lesions with exceptional precision. A transfer learning cancer detection and classification using machine
approach, such as ResNet, was used to train the model. learning.” [link]
Different training and evaluation ratios were used, including [4]. Kinnor Das, Clay J. Cockerell, Anant Patil, Paweł
80:20, 70:30, and 40:60. Undersampling and oversampling Pietkiewicz, Mario Giulini, Stephan
approaches were compared. The findings give important Grabbe, and Mohamad Goldust “Machine learning
insights into the use of CNNs for accurate detection and open and its application in skin cancer”.[link]
the path for future advances in using deep learning [5]. Praveen Banasode*, Minal Patil, Nikhil Ammanag”
techniques to treat skin cancer. In conclusion, this study A Melanoma Skin Cancer Detection Using
demonstrates the efficacy of deep learning algorithms in Machine Learning Technique: Support Vector
skin cancer categorization. Machine.” [link]

IJISRT23AUG2018 www.ijisrt.com 2984

Volume 8, Issue 8, August – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[6]. Mehwish Zafar, Muhammad Imran Sharif,
Muhammad Imran Sharif, seifedine Kadry, Syed
Ahmed Chan Bukhari, and Hafiz Tayyab Rauf “Skin
lesion analysis and cancer detection based on
deep/machine learning Techniques: A
comprehensive survey.” [link]
[7]. Snehal Vijay Kamble, Dr. P.R. Gumble "A survey
on diagnosis on skin cancer based on image
processing using Machine Learning."[link]
[8]. Mario Fernando Jojoa Acosta, Liesle Yail Caballero
Tovar, Maria Begonya Garcia- Zapirain & Winston
Spencer Percybrooks "Melanoma diagnosis using
deep learning techniques on dermatoscopic
images."[link]
[9]. K. Dittakan, P. Nulek, and K. Prutsachainimmit
"Skin Cancer Detection from Smartphone
Imagery using Convolutional Neural
Network."[link]
[10]. Kirill Solovev, Nicolas Pröllochs” Deep learning-
based skin cancer classification using transfer
learning.” [link]

IJISRT23AUG2018 www.ijisrt.com 2985