People in today‘s world are busy and occupied all of the time.

We often
overlook minor illnesses in our bodies as a result of our fast-paced lives. Skin disease
is one of them. It is the most widespread disease on the planet. People generally
consider dermatology problems as ephemeral, though that‘s not always the case. If
the skin disease is not appropriately recognized, it might cause serious complications.
The project‘s current models use segmentation techniques such as edge detection;
however, our study seeks to diagnose skin disorders with high accuracy rates utilizing
Deep learning. Our goal is to use CNN (convolutional neural networks) and transfer
learning from the Inception model to identify and categorize skin disorders. This
research seeks to inform victims of the potential repercussions and assist doctors in
making an initial diagnosis. This is being implemented in both a web application and
an Android application.

TABLE OF CONTENTS

5
Chapter No. TITLE Page No.
ABSTRACT v
LIST OF FIGURES viii
LIST OF ABBREVIATIONS ix
1 INTRODUCTION 1
1.1 OVERVIEW 1
1.2 DEEP LEARNING 2
1.3
2 LITERATURE SURVEY 4
2.1. RELATED WORK 4

3 AIM AND SCOPE OF THE PROJECT 6

3.1. EXISTING SYSTEM 6
3.2. PROPOSED SYSTEM 6
3.3.
8 OBJECTIVE 7

4 METHODOLOGY 8
4.1. HARDWARE REQUIREMENTS 8
4.2. SOFTWARE REQUIRMENTS 8
4.3. SYSTEM ARCHITECTURE 8
4.4. PYTHON AND JUPYTER 9
4.5. DEEP LEARNING 16
4.6. TRANSFER LEARNING 26
4.7. ALGORITHMS AND ARCHITECURES 30
4.7.1. CONVOLUTIONAL NEURAL NETWORK 30
4.7.2. INCEPTION NEURAL NETWORK 36
4.7.3. LIBRARIES AND FRAMEWORKS
MODEL 43
4.8. MODULES 51
4.8.1. TRAINING AND TESTING THE MODEL .

51
4.8.2. DEPLOYING THE MODEL 51

6
5 RESULTS AND DISCUSSION 52
5.1. PERFORMANCE ANALYSIS 52

6 CONCLUSION 53
6.1. CONCLUSION 53

REFERENCES 54
APPENDICES 60
A. SOURCE CODE 63
B. SCREENSHOTS 117
C. PLAGIARISM REPORT 120
D. JOURNAL PAPER 121

LIST OF FIGURES

7
Fig No. Figure Name Page No.
4.3.1 SYSTEM ARCHITECTURE 9
4.5.1 ADVANTAGE OF DEEP LEARNING 17
4.5.3 SINGLE LAYER PERCEPTRON 18
4.5.3.2 WORKING OF A PERCEPTRON 20
4.5.4 MULTILAYER PERCEPTRON 21
4.5.4.2 FORWARD PROPAGATION 22
4.5.5.4.1 BACKWARD PROPAGATION 23
COMPUTING DERIVATIVES IN BACK
4.5.5.4.2 24
PROPAGATION
PERFORMANCE OF TRANSFER LEARNING
4.6.1 27
MODEL
4.6.2 ILLUSTRATION OF TRANSFER LEARNING 28
4.7.2.1.1 WORKING OF CONVOLUTIONAL LAYERS 32
4.7.2.1.2 PADDING IN CNN 33
4.7.2.1.3 ReLU GRAPH 34
4.7.2.1.4 MAX POOLING 35
4.7.3.1 INCEPTION ARCHITECTURE 36
4.7.3.2 A UNIT IN INCEPTION 40
4.7.3.3 1x1 CONVOLUTION 41
5.1 TRAINING ACCURACY OF THE MODEL 53

LIST OF ABBREVIATIONS
8
ABBREVIATIONS EXPANSION

ANN Artificial Neural network

BGR Blue Green Red

RGB Decision Tree

CNN Convolutional neural network

Multi Layer Perceptron
MLP
Very Deep Convolutional
VGG Networks for Large-Scale Image
Recognition

9
 CHAPTER 1

INTRODUCTION

1.1 OVERVIEW

AI is a broad term for intelligence exhibited by machines; we are witnessing a wide-

reaching integration of AI into human life in the next few decades. Digital assistants,
for example, help us with day-to-day tasks and give us information when we need it.
Web searching is becoming increasingly accurate and efficient thanks to AI‘s
predictive capabilities; an example is AI-powered web search. It has increased and
can now give we accurate answers to our questions, which would have taken hours of
research before. The disease detection or natural catastrophe forecasts are far more
accurate than they were twenty years ago.

With the massive amount of data generated every day, it is natural that AI plays an
essential role in the future of daily life. Artificial intelligence has been a topic of
discussion for a long time, and it seems that it is finally reaching a point in which it
can impact our day-to-day lives. Deep learning is a subdivision of AI. It aims at
developing algorithms that analyze vast datasets to learn how to perform specific
jobs.

Deep learning involves neural networks made up of multiple layers. These networks
mimic the behaviour of the brain. Neural networks are at the core, and they provide
the best results with unstructured data. This form of AI allows for more accessible
training and better returns on investment. There are many advantages; one
advantage is the high accuracy it provides with unstructured data. Deep learning
outperforms machine learning when it comes to addressing complicated problems.
The deep learning models‘ accuracy is high, and their predictions are close to
humans. Most people do not know how to take care of it when it comes to skin. It is a
sensitive and delicate organ that needs care and affection. An estimated 1 in 3 people
have a skin disease at any time.

1
The human body has many potential entry points for parasites and bacteria to get
inside and infect us with various ailments. Without a proper diagnosis, the disease will
likely worsen. Some common skin diseases are eczema, psoriasis, acne, rosacea,
and ichthyosis. Classifying integumentary system disorders using deep learning is
one of the most challenging tasks for medical professionals. The accuracy of a
computer-based diagnosis for a patient can help save time and money by providing a
more accurate diagnosis. An important step in diagnosing a medical condition is to
classify it as a certain type of disorder. This classification may be done using, for
example, the ICD codes or the DSM-5 codes. To classify the group of International
Classification of Diseases (ICD) codes into a suitable category, we use deep learning
approach. Our results show that our method outperforms current methods in terms of
both accuracy and efficiency.

Our goal is to four classify integumentary system disorders namely Acne, Psoriasis,
Ringworm and Eczema using Deep learning. The model is trained using images of
each disease as input. The images are fed to the deep learning layers, which adjust
the weights of the neurons for every epoch. This model uses an Inception v3 neural
network with 48 layers and an additional five dense layers for training photos of
distinct classes. After completion of training the testing phase carried out to determine
the accuracy of the model.

1.2 DEEP LEARNING

Deep learning allows computational models of multiple processing layers to learn and
represent data with multiple levels of abstraction mimicking how the brain perceives
and understands multimodal information, thus implicitly capturing intricate structures
of large‐scale data. Deep learning is a rich family of methods, encompassing neural
networks, hierarchical probabilistic models, and a variety of unsupervised and
supervised feature learning algorithms. The recent surge of interest in deep learning
methods is due to the fact that they have been shown to outperform previous state-of-

2
the-art techniques in several tasks, as well as the abundance of complex data from
different sources.

Convolutional neural networks (CNNs) or convnets, for short – have in recent years
achieved results which were previously considered to be purely within the human
realm. In this chapter we introduce CNNs, and for this we first consider regular neural
networks, and how these methods are trained. After introducing the convolution, we
introduce CNNs. They are very similar to the regular neural networks as they are also
made up of neurons with learnable weights. But, in contrast to MLPs, CNNs make the
explicit assumption that inputs have specific structure like images. This allows
encoding this property into the architecture by sharing the weights for each location in
the image and having neurons respond only locally.

3
 CHAPTER 2

LITERATURE SURVEY

2.1 RELATED WORK

[2] Digital dermatology: Skin disease detection model using image processing, this
paper uses algorithms like DCT (Direct Cosine Transform), SVD (Single Value
Decomposition) and DWT (Direct Wavelet Transform) According to this paper, the
attained accuracy for detecting skin disorders is 80%. The model's reliability would
have increased greatly if neural networks had been utilized to train the images.

[7] Diagnosis of skin diseases using Convolutional Neural Networks, this paper solely
relies on convolutional neural network (CNN) and achieved accuracy of seventy
percent. The paper claims that accuracy can be improved if larger dataset is used.
The model's accuracy would have improved significantly if the transfer learning
approach had been used. Our model was also trained with around one thousand two
hundred pictures of skin diseases and tested with two hundred images but achieved
around ninety percent accuracy.

[3] Face Skin Disease Detection with Textural Feature Extraction, this paper aims at
classifying skin face skin disease detection using K-nearest neighbour algorithm
which produced an accuracy of eighty percent. If Deep learning is utilized and transfer
learning is implemented then the efficiency of the model would have enhanced
greatly.

Integumentary system disorders affect 71 percent of the world‘s population, almost

three-fourths of humankind. Ailments of the integumentary system are the third most
prevalent illness, and many who suffer from them do not consider them a significant
problem that requires medical attention [2].

Depending on the skin state, the appearance of skin conditions in dermoscopy

images might differ significantly. Hair, texture, and bubbles may further obfuscate the

4
distinction between skin infections and neighbouring clear skin. As a result,
dermoscopy image classification is a complicated process [4]. We are deploying
cutting-edge deep learning technologies to take on this complex task. When
compared towards other ml algorithms, deep learning has been used to perform
exceptionally challenging classification and segmentation jobs [32], [33] with great
accuracy. The framework of the system is based primarily on convolutional layers
[31]. Characteristics in the initial layers of deep learning network topologies are
typically broad, while high-level features are often particular. Yosinski et al. [34]
demonstrated that using model parameters transmitted from different jobs to initialize
networks enhances results compared to arbitrary parameters [24].

5
 CHAPTER 3

AIM AND SCOPE OF THE PROJECT

3.1 EXISTING SYSTEM

From the above literature survey, it can be inferred that some previous models used
segmentation techniques with low accuracy, deep learning with CNN with less
reliability and some models tried to achieve high accuracy with small dataset and fell
flat.

The majority of the previous models concentrated on identifying whether the

melanoma is benignant or malignant (due to abundance of datasets available on
melanoma) and did not cover on other diseases like psoriasis, ringworm, etc.

3.2 PROPOSED SYSTEM

Our proposal is to make an interactive web application for the patient to get an
instant diagnosis of the disease and a pdf with do‘s and dont‘s and a general
medication to reduce the severity of the disease. Our model uses a small dataset
and implements deep learning technology along with CNN and transfer learning.

The use of transfer learning using Inception v3 in the project sets apart from other
models. Our model achieved accuracy on par with models which the dataset of ten
thousand images with around thousand images.

3.3 OBJECTIVE

The project aims to perform diagnosis of integumentary system disorders like

Ringworm, Melanoma, Psoriasis and Measles using deep learning employing