International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878 (Online), Volume-8 Issue-6, March 2020
Skin Diseases Prediction using Deep Learning
Framework
Padmavathi S, Mathu Mithaa E, Kiruthika T, Ruba M
Abstract—Dermatological diseases are found to induce a serious
impact on the health of millions of people as everyone is affected Artificial Intelligence (AI) stimulates human intelligence in
by almost all types of skin disorders every year. Since the human machines to make them think like humans to inherit the
analysis of such diseases takes some time and effort, and current learning and problem-solving traits that the human brains
methods are only used to analyse singular types of skin diseases,
exhibit and thereby taking the best actions that allow them
there is a need for a more high-level computer-aided expertise in
the analysis and diagnosis of multi-type skin diseases. This paper to achieve its goal in minimal time. It makes the machines
proposes an approach to use computer-aided techniques in deep mimic it and execute tasks, from the simplest to those that
learning neural networks such as Convolutional neural networks are even more complex using its goals such as learning,
(CNN) and Residual Neural Networks (ResNet) to predict skin reasoning, and perception. Some of the previous
diseases real-time and thus provides more accuracy than other benchmarks that used artificial intelligence are not
neural networks.
considered to embody them since they are now made as an
Index Terms: Deep Learning, Skin disease, Convolutional inherited function in computers. Machines are wired using a
Neural Network, Residual Neural Network cross-disciplinary approach based on mathematics,
computer science, linguistics, psychology, and more. The
I. INTRODUCTION computers are thus provided with the potential to grasp even
the complex theories without being specifically programmed
Dermatological diseases are one of the complicated using deep learning which humans find difficult to perceive
branches of science due to its complexities in the diagnosis [5]. They are used to process huge complex datasets in the
of diseases and their variation in the changing environment. healthcare industry to analyse them into medical insights
Skin diseases are the most common among them especially [6]. Unlike machine learning, deep learning uses huge
prone to spread and can prove to be fatal leading to skin datasets and fewer classifiers resulting in more training time.
cancer if not treated in its earlier stages [1]. The occurrence To identify a skin disease, a variety of visual clues
of skin cancer is now increasing in numbers than the may be used such as the individual lesion morphology, the
incidence of other new types of cancer of the lung, breast body size distribution, colour, scaling and arrangement of
combined [2]. Research induces that one-fifth of the people lesions. By analysing the individual components separately,
are likely to be affected by skin cancer in their lifetime and the complexity of the recognition process is quite increased
hence making it diagnosis more challenging [3]. Thus the [7, 8] and the human-engineered feature extraction method
computer-based diagnosis of such diseases comes into play is not applicable for its classification. On the contrary, hand-
due to its ability to produce the result within a short interval crafted features are just devoted to a limited variety of skin
of time with more accuracy than human analysis using diseases due to its diverse nature and are not suitable to be
laboratory methods. The most prevalent technology used for applied to classes and datasets. One way to solve this
the prediction of diseases is Artificial Intelligence using problem is to use feature learning [9] which eliminates the
Deep Learning where AI can be used to develop algorithms need for feature engineering and allows the machine to
to learn the behaviour and patterns of the disease and in decide which feature to use by itself. Though many
turn, the machines make use of these learning methods to classification systems that use feature-learning have been
process the images and predict [19]. developed [1, 5], most of them are restricted to dermo copy
or histopathology images [23, 11]. They are mainly used for
detecting mitosis, which is a cancer indicator [12]. For most
of the cases, transfer learning can be utilized to train a deep
Convolutional Neural Network (CNN) [13]. Also in transfer
Manuscript received on February 10, 2020. learning, instead of training the network from randomly
Revised Manuscript received on February 20, 2020. initialized parameters, a pre-trained network can be used by
Manuscript published on March 30, 2020. fine-tuning its weights by continuing the back propagation.
* Correspondence Author The reason is that the results of some of the initial layers of
Padmavathi.S,Department of Computer Science and Engineering,Sri
Krishna college of Technology,Coimbatore,India.
a well-trained network contain certain generic features like
MathuMithaa.E, Department of Computer Science and blobs, edges that are used in many tasks and such features
Engineering,Sri Krishna college of Technology,Coimbatore,India. can be applied directly to a new dataset. For the proposed
Kiruthika.T, Department of Computer Science and Engineering,Sri
Krishna college of Technology,Coimbatore,India.
skin diagnosis system, transfer learning is done by fine-
Ruba.M, Department of Computer Science and Engineering,Sri tuning ImageNet [14], which is a pre-trained model along
Krishna college of Technology,Coimbatore,India. with Caffe [15], which is a framework in deep learning that
is used for efficient and
© The Authors. Published by Blue Eyes Intelligence Engineering and
Sciences Publication (BEIESP). This is an open access article under the CC expressive CNN training.
BY-NC-ND license ([Link]
Retrieval Number: F9038038620/2020©BEIESP
DOI: 10.35940/ijrte.F9038.038620 Published By:
Journal Website: [Link] Blue Eyes Intelligence Engineering
4781 & Sciences Publication
Skin Diseases Prediction using Deep Learning Framework
II. METHODOLOGY marked information, they mean the gatherings of tests that
have been labelled to at least one name.
Since the human analysis of skin diseases takes
some time and effort, and current methods are only used to Overfitting: A significant thought in Deep Learning is that
analyse singular types of skin diseases, there is a need for a the estimation of the objective capacity that has been
more high-level computer-aided expertise in the analysis prepared utilizing preparing information, sums up to new
and diagnosis of multi-type skin diseases. information. Speculation works best if the sign or the
By using the appropriate methods the dataset is example that is utilized as the preparation information has a
studied and then by applying various techniques and high sign to commotion proportion.
algorithms the skin disease can be predicted. Comparison
among algorithms helps to achieve the best one which Regularization: Regularization is the strategy to assess a
provides high accuracy. favoured multifaceted nature of the Deep Learning model
with the goal that the model sums up and the over-fit/under-
fit issue is maintained a strategic distance from. This is
Dataset: This dataset consists of 10015 dermatoscopic
finished by including a punishment the various parameters
images categorized into 7 different classes. A complete
of the model along these lines diminishing the opportunity
dataset is employed to train the system model. A dataset is
of the model.
split into training set and validating/testing set.
Validation/testing set will tune the parameters and is used
Parameter and Hyper-Parameter: Parameters are design
only to assess the effectiveness and efficiency of the system.
factors that can be believed to be inside to the model as they
can be assessed from the preparation information. Hyper
Data pre-processing: Data Pre-handling is a procedure that
parameters of a model are set and tuned relying upon a mix
is utilized to change over the crude information into a
of certain heuristics and the experience and area information
spotless informational collection. As it were, at whatever
on the information researcher.
point the information is assembled from various sources it is
gathered in crude arrangement which isn't achievable for the
III. RESIDUAL NEURAL
examination. For accomplishing better outcomes from the
NETWORK METHOD
applied model in deep Learning ventures the organization of
the information must be in a legitimate way. Some
predefined Deep Learning model needs data in a Residual Neural Network (ResNet) is a design of VGG’s
predetermined configuration, for instance, Random Forest 3x33x3. It is a design methodology that follows convolution
calculation doesn't bolster invalid qualities, in this manner layers. There are two layers in this block with an equal
to execute arbitrary timberland calculation invalid qualities number of output channels [10]. It follows the function of
must be overseen from the first crude informational index. activation and a layer for the process of normalization.
There can be a case where the variables from a user can be
Another viewpoint is that informational collection ought to added before the unit activation function omitting the above-
be arranged so that more than one Machine Learning and mentioned layer operations. To add the output of those
Deep Learning calculations are executed in one layers the output must be in the same phase as that of input.
informational index, and best out of them is picked. For additional operation, we can add a 1x11x1 layer for
transforming the input to get the shape that we need. Images
Model development: Deep Learning algorithm is the are 64 x 64 x 3 (3-channel RGB images). With a total of 2
hypothesis set that is taken at the beginning before the classes, it also has a stack (15, 4, 6) with convolution layers
training starts with real-world data. Linear Regression (64,128,256,512) and these will be performed by ResNet. It
algorithm is used which uses some set of functions and has 64 filters in total before the spatial dimensions are
chooses one that will fit the most. reduced, convolution is the first layer. Finally, we have 3
sets of modules.
Training: While training for deep learning, pass an
algorithm with training data. CNN and ResNet finds
patterns in the training data such that the input parameters
correspond to the target. The output of the training process
is a machine learning model which can be used to make
predictions.
Focus on: The objective is whatever the yield of the
information factors. It could be the individual classes that
the information factors are mapped to if there should be an Fig.1 Single Residual Block
occurrence of an order issue or the yield esteem results in a
relapse issue. In the event that the preparation set is
Each Residual module present in the convolution layer will
considered, at that point the objective is the preparation
learn the CONV filters i.e., 32, 32 and 128. After the spatial
yield esteems that will be considered.
dimensions are reduced [16].
Labels: Labels are the last yield. The yield classes is likely
to be the 1names. When information researchers discuss
Retrieval Number: F9038038620/2020©BEIESP
DOI: 10.35940/ijrte.F9038.038620 Published By:
Journal Website: [Link] 4782 Blue Eyes Intelligence Engineering
& Sciences Publication
International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878 (Online), Volume-8 Issue-6, March 2020
Then, stacking the residual modules of 4 sets, each of three
layers with the filters 64, 64, and 256. The dimensions are
reduced once again so that finally we can stack 6 sets of the
residual module, the layer can learn 128, 128, and 512
filters. For the last time, we reduce the spatial dimensions,
before the performance of average pooling and classifier
which is softmax that applied.
IV. CONVOLUTIONAL NEURAL NETWORK
Convolutional Neural Network (CNN or ConvNet) is a deep Fig.3 Basic Structure of CNN
learning algorithm used for analysing visual imagery that
can accept an input image and attaches some significance to
each point in the image which makes is easy to differentiate
the points [17]. CNN model uses a feed forward network.
The flow of information occurs in the forward direction in a
feed forward neural network [18]. Convolution neural
network uses MLP, which is Multilayer Perceptron, to do
convolutional processes in which each neuron in a layer is
associated with every other neuron in the other layers [19].
CNN techniques are well known for its better performance
in image recognition. A small amount of pre-processing is
needed for CNN. Several filters are used in the convolution
layer; each filter makes a 2-dimensional activation map by
moving across the input data [10]. One of the metrics for
evaluating classification models is accuracy. Accuracy can Fig.4 Proposed Framework
be calculated by a fraction of predictions our system got
right [20]. CNN is one of the types of neural network which
is highly used in the computer science field.
Fig.5 CNN Model Accuracy
Fig.2 CNN model diagram
CNN model consists of an input layer, it is composed of
artificial input neurons and output layer and also has
multiple hidden layers. The hidden layers of CNN are
convolutional layer, pooling layer, fully connected layer,
receptive field and weights [21]. Feature extraction and
classification parts are the two components of the CNN
Fig.6 CNN Model Loss
model. The feature extraction part is performed by the
convolution and pooling layer [22]. The fully connected V. RESULT
layers then act as a classifier on top of these features and The better outcome achieved to predict and prevent the
assign a probability to provide final output [10]. dermatological diseases using the techniques in Deep
Learning Neural Networks (CNN) and Residual Neural
Networks (ResNet) to predict skin diseases and thus
provides more accuracy than other neural networks. The
accuracy rate is tabulated
below:
Retrieval Number: F9038038620/2020©BEIESP
DOI: 10.35940/ijrte.F9038.038620 Published By:
Journal Website: [Link] Blue Eyes Intelligence Engineering
4783 & Sciences Publication
Skin Diseases Prediction using Deep Learning Framework
Table 1 Experimental results Services. IEEE Transactions on Services Computing, 2015, 8(1):65-
78.
16. Guo K, Liang Z, Tang Y, et al. SOR: An optimized semantic ontology
retrieval algorithm for heterogeneous multimedia big data. Journal of
CLASSIFIERS TEST ACCURACY Computational Science, 2018, 28: 455-465.
17. IoffeS ,Szegedy C . Batch normalization: accelerating deep network
training by reducing internal covariate shift[C]// International
CNN 77% Conference on International Conference on Machine Learning.
[Link], 2015.
18. [Link]
ResNet 68% convolutional-neural-networks-the-eli5-way-3bd2b1164a53
19. Lindsay R K, Feigenbaum E A, Buchanan B G, et al. Applications of
artificial intelligence for chemical inference: the DENDRAL project.
20. [Link]
VI. CONCLUSION 21. Li S, Da Xu L, Zhao S. The internet of things: a survey. Information
Systems Frontiers, 2015, 17(2): 243-259.
The feasibility of building a universal skin disease 22. Guo K, Xu T, Kui X, et al. iFusion: Towards efficient intelligence
fusion for deep learning from real-time and heterogeneous data.
classification system has been investigated using deep CNN. Information Fusion, 2019, 51: 215-223.
Better accuracy can be obtained by providing a training set 23. History of artificial intelligence [EB/OL].
with more variance and also by increasing its size. Also, [Link] of artificial intelligence. April
note that the images retrieved by the networks are closely 14, 2019.
24. Long E , Lin H , Liu Z , et al. An artificial intelligence platform for
related to the ground truth. We may need to design a the multihospital collaborative management of congenital cataracts.
hierarchical classification algorithm using the retrieved Nature Biomedical Engineering, 2017, 1(2):0024.
images to improve the accuracy. Thus by using ensemble
features as well as deep learning, predictions can be AUTHORS PROFILE:
achieved with a higher rate than previous models. It is also
found that Convolution Neural Networks performs well
compared to Residual Neural Network in the diagnosis of Padmavathi.S, completed her M.E. degree in 2013
skin diseases. at Angel college of engineering and Technology
,Tirupur . She has about Five years of experience in
teaching . Her areas of interest Network
REFERENCES: Security , Grid computing and Wireless Technology .
She published her papers in various International
1. DuanY , Fu G , Zhou N , et al. Everything as a Service (XaaS) on the /National Journals . She is a member in IAENG.
Cloud: Origins, Current and Future Trends[C]// 2015 IEEE 8th
International Conference on Cloud Computing (CLOUD). IEEE, MathuMithaa.E, UG student in Sri Krishna College
2015. Of Technology in the Department of Computer
2. Sheng Ren, Deepak Kumar Jain, KehuaGuo, Tao Xu,Tao Chi. Science and Engineering.
Towards Efficient Medical Lesion Image Super-Resolution based on
Deep Residual Networks. Signal Processing: Image
Communication75(2019):1-10.
3. Zhou M, Rong Z, Zeng D, et al. Services in the Cloud Computing era:
A survey[C]// Universal Communication Symposium. 2010. Kiruthika.T, UG student in Sri Krishna College Of
4. Bakkar N, et al. Artificial intelligence in neurodegenerative disease Technology in the Department of Computer Science
research: use of IBM Watson to identify additional RNA-binding and Engineering.
proteins altered in amyotrophic lateral sclerosis.
Actaneuropathologica, 2018, 135(2): 227-247.
5. Automated Skin Disease Identification using Deep Learning
Algorithm-
6. Sourav Kumar Patnaik, Mansher Singh Sidhu, YaagyanikaGehlot, Ruba.M, UG student in Sri Krishna College Of
Bhairvi Sharma and P. Muthu Technology in the Department of Computer Science
7. [Link] and Engineering.
8. P.-Q. Huang, Y. Wang, K. Wang, and Z.-[Link] optimization
approach for joint offloading decision and resource allocation in
cooperative mobile edge computing. IEEE Transactions on
Cybernetics, in press. DOI: 10.1109/TCYB.2019.2916728
9. Duan Y, Fu G, Zhou N, et al. Everything as a Service (XaaS) on the
Cloud: Origins, Current and Future Trends[C]// IEEE International
Conference on Cloud Computing. 2015.
10. Weiss S M, Kulikowski C A, Amarel S, et al. A model-based method
for computer-aided medical decision-making. Artificial intelligence,
1978, 11(1):145-172 .
11. Sreeja N.K., Sankar A. Pattern matching based classification using
Ant Colony Optimization based feature selection. 2015
12. A. Masood and A. A. Al-Jumaily Computer Aided Diagnostic
Support System for Skin Cancer: A Review of Techniques and
Algorithms. International Journal of Biomedical Imaging. 2013, 2013:
323268.
13. Elhoseny M, Gustavo Ram´ırez-Gonzalez, Abu-Elnasr O M, et al.
Secure Medical Data Transmission Model for IoT-based Healthcare
Systems. IEEE Access, 2018, PP(99):1-1.
14. Esteva A , Kuprel B , Novoa R A , et al. Dermatologist-level
classification of skin cancer with deep neural networks. Nature, 2017,
542(7639):115-118.
15. AlcarazCalero J M ,Aguado J G . MonPaaS: An Adaptive Monitoring
Platformas a Service for Cloud Computing Infrastructures and
Retrieval Number: F9038038620/2020©BEIESP
DOI: 10.35940/ijrte.F9038.038620 Published By:
Journal Website: [Link] 4784 Blue Eyes Intelligence Engineering
& Sciences Publication
