DermaLens: INTELLIGENT SKIN DISEASE IDENTIFICATION AND

CARE GUIDANCE SYSTEM

Project Team

Sl. No. Reg. No. Student Name

1 21ETCS002115 SABA NIDA

2 21ETCS002075 ANKITHA R

3 21ETCS002411 VAISHNAVI PAI N

4 21ETCS002079 CHANDANA S

B. Tech. in Computer Science and Engineering

FACULTY OF ENGINEERING AND TECHNOLOGY

M. S. RAMAIAH UNIVERSITY OF APPLIED SCIENCES
Bengaluru - 560 054
 FACULTY OF ENGINEERING AND TECHNOLOGY

Certificate
This is to certify that the Project titled “DermaLens: INTELLIGENT SKIN DISEASE
IDENTIFICATION AND CARE GUIDANCE SYSTEM” is a bonafide work carried out in
the Department of Computer Science and Engineering by Saba Nida bearing Reg.
No. 21ETCS002115 respectively in partial fulfilment of requirements for the award
of B. Tech. Degree in Computer Science
and Engineering of M. S. Ramaiah University of Applied Sciences.
JANUARY– 2025

Mr. Prabhakar A
Associate Professor – Dept. of CSE

Dr. Rinki Sharma Dr Sarat Kumar Maharana

Professor and Head – Dept. of CSE Professor and Dean-FET
 FACULTY OF ENGINEERING AND TECHNOLOGY

Certificate

This is to certify that the Project titled “DermaLens: INTELLIGENT SKIN DISEASE
IDENTIFICATION AND CARE GUIDANCE SYSTEM” is a bonafide work carried out in
the Department of Computer Science and Engineering by Ankitha R bearing Reg.
No. 21ETCS002075 respectively in partial fulfilment of requirements for the award
of B. Tech. Degree in Computer Science
and Engineering of M. S. Ramaiah University of Applied Sciences.
JANUARY– 2025

Mr. Prabhakar A
Associate Professor – Dept. of CSE

Dr. Rinki Sharma Dr Sarat Kumar Maharana

Professor and Head – Dept. of CSE Professor and Dean-FET

1
 FACULTY OF ENGINEERING AND TECHNOLOGY

Certificate

This is to certify that the Project titled “DermaLens: INTELLIGENT SKIN DISEASE
IDENTIFICATION AND CARE GUIDANCE SYSTEM” is a bonafide work carried out in
the Department of Computer Science and Engineering by Vaishnavi Pai N bearing
Reg. No. 21ETCS002411 respectively in partial fulfilment of requirements for the
award of B. Tech. Degree in Computer Science
and Engineering of M. S. Ramaiah University of Applied Sciences.
JANUARY– 2025

Mr. Prabhakar A
Associate Professor – Dept. of CSE

Dr. Rinki Sharma Dr Sarat Kumar Maharana

Professor and Head – Dept. of CSE Professor and Dean-FET

1
 FACULTY OF ENGINEERING AND TECHNOLOGY

Certificate
This is to certify that the Project titled “DermaLens: INTELLIGENT SKIN DISEASE
IDENTIFICATION AND CARE GUIDANCE SYSTEM” is a bonafide work carried out in
the Department of Computer Science and Engineering by Chandana S bearing Reg.
No. 21ETCS002079 respectively in partial fulfilment of requirements for the award
of B. Tech. Degree in Computer Science
and Engineering of M. S. Ramaiah University of Applied Sciences.
JANUARY– 2025

Mr. Prabhakar A
Associate Professor – Dept. of CSE

Dr. Rinki Sharma Dr Sarat Kumar Maharana

Professor and Head – Dept. of CSE Professor and Dean-FET

1
 Declaration

DermaLens: INTELLIGENT SKIN DISEASE IDENTIFICATION AND CARE GUIDANCE SYSTEM

The project work is submitted in partial fulfilment of academic requirements for the award of
B. Tech. Degree in the Department of Computer Science and Engineering of the Faculty of
Engineering and Technology of M. S. Ramaiah University of Applied Sciences. The project report
submitted herewith is a result of our own work and in conformance to the guidelines on plagiarism
as laid out in the University Student Handbook. All sections of the text and results which have been
obtained from other sources are fully referenced. We understand that cheating and plagiarism
constitute a breach of university regulations, hence this project report has been passed through
plagiarism check and the report has been submitted to the supervisor.

Sl. No. Reg. No. Student Name

1 21ETCS002115 SABA NIDA

2 21ETCS002075 ANKITHA R

3 21ETCS002411 VAISHNAVI PAI N

4 21ETCS002079 CHANDANA S

Date : 10th Feb 2025

1
 Acknowledgements

Our sincere gratitude goes to all those who supported and guided us throughout this project.

Mr. Prabhakar A, our esteemed guide and mentor, deserves our deepest gratitude for his
valuable insight, constant encouragement, and unrelenting support throughout the research
and development of this project. His expertise and constructive criticism helped us
tremendously in improving our work.

We express our gratitude to Dr. Rinki Sharma, Head of Department of Computer Science and
Engineering, and Dr Sarat Kumar Maharana, Dean of Faculty of Engineering and Technology,
for providing all essential resources and a congenial atmosphere for learning during this
project.

We thank M. S. Ramaiah University of Applied Sciences for giving us the opportunity to look
into and work on this project in an environment conducive to research and innovation.

Last but not the least, our heartfelt thanks go to all the family members for their unconditional
support, patience, and encouragement during the entire journey. Their undiluted faith in us
has always been our strength and inspiration.
 Abstract

As a huge medical entity, skin diseases affect millions of lives worldwide. Early diagnosis
becomes an important means to prevent serious complications and to handle them effectively.
The traditional method of diagnosing the presence of a skin condition takes time and requires
a considerable investment, thus limiting its reach to a number of people.

This project presents an AI-empowered Intelligent Skin Disease Identification and Care
Guidance System using Convolutional Neural Networks (CNNs), particularly trained on the
DermNet dataset using the DenseNet architecture. The system allows users to upload images
of their skin conditions, whereupon they will receive a preliminary diagnosis as well as
information on the severity of the disease, possible treatment options, and local
dermatologists.''

The accuracy achieved on training stands at 90.5% over 48 epochs, which demonstrates that
skin disease classification is done effectively by the system. In order to improve usability and
accessibility, this platform combines Google Maps API for information about dermatologists.
It also provides the users with information pertaining to the skin disease, severity assessments,
and feedback mechanisms to enhance the accuracy and relevance of the platform.

Future improvements will seek to diversify the data set so as to improve model accuracy and
pursue the development of a mobile app to provide a seamless user experience. The goal is to
bridge AI-powered dermatology solutions and accessible healthcare for early and effective skin
disease management.
 Table of Contents

Acknowledgements 2
Abstract 3
List of Tables 5
List of Figures 6
1. Introduction 7
1.1 Introduction 7
1.2 Literature Survey 8
1.3 Conclusion 9
2. Background Theory 10
2.1 Background Theory: 10
2.2 Background of Existing Application 12
2.3 Conclusion 12
3. Aim and Objectives 13
3.1 Title 13
3.2 Aim 13
3.3 Objectives 13
3.4 Functional Requirements 14
3.5 Method and Methodology 15
3.6 Conclusion 16
4. Problem Solving 17
4.1 Design 17
4.1.1 System Design 17
4.1.2 Sequence Diagram 20
4.2 Implementation 22
4.3 Testing 29
5. Results 30
5.1 Model Prediction 30
5.2 Frontend UI 31
6. Project Costing 34
6.1 Project Cost Estimation 34
7. Conclusions and Suggestions for Future Work 35
7.1 Conclusion 35
7.2 Future Implementations 35
References 37
 List of Tables

Table 1………………………………………………………………………………………………………..…….………….........…8
Table 2 ………………………………………………………..………………………………………………………………………..15
Table 3 ……………………………………………………………………………........................................................29
Table 4 ……………………………………………………………………………………................................................29
Table 5.……….……………………………………………………………………………………………………………………..….34
 List of Figures

Figure 1………………………………………………………………………………………………………..…….……………..…17
Figure 2 ………………………………………………………..…………………………………………………………………..…20
Figure 3 …………………………………………………………………………….....................................................22
Figure 4 …………………………………………………………………………………….............................................27
Figure 5.……….………………………………………………………………………………………………………………………28
Figure 6 ……………………………………………………………………………………………………………………………….28
Figure 7 ……………………………………………………………………………………………………………………………….30
Figure 8 ……………………………………………………………………………………………………………………………….30
Figure 9 ……………………………………………………………………………………………………………………………….31
Figure10..……………………………………………………………………………………………………………………………..31
Figure 11 ………………………………………………………………………………………………………………………….....32
Figure 12 ………………………………………………………………………………………………………………………….....32
Figure 13 ………………………………………………………………………………………………………………………….....33
 1. Introduction

In this chapter, the project is introduced with its theme followed by its purpose. Then a
literature survey is documented stating most popular existing applications under domain of
current project and their key features. In the end, current project and its output will be
compared with the existing application and in the end, key features of current project will be
evaluated.

1.1 Introduction
Skin diseases are among the most common medical conditions, affecting people of all ages.
These conditions vary from mild irritation to chronic and severe diseases that, if left
undiagnosed and untreated, can turn life-threatening. The need for an efficient and cost-
effective diagnostic system has never been greater, especially in underdeveloped regions with
limited access to dermatologists.
Traditional skin disease diagnosis involves face-to-face consultations, comprehensive medical
tests, and high-resolution imaging techniques that are not accessible at all times. Many people
are also ignorant about the onset of early symptoms for skin diseases; hence, they develop
much later due to late detection. Based on this challenge, we came up with DermaLens, a web
application based on CNN for accurate identification of skin diseases through image study. The
proposed method applies the DenseNet architecture, which is known to focus effectively on
handling medical images with accuracy.
Using a dataset of 23 skin diseases from DERMNET, the system gives diagnoses with an
accuracy of 90.5%. Besides classification, the platform offers educational content in articles
and videos to improve users' knowledge about their skin condition. Additionally, it provides
nearby dermatologists and, therefore, allows users to seek professional medical service
whenever needed.
Our research on the topic has also been published in the International Journal of Progressive
Research in Engineering Management and Science IJPREMS, which thereby gives scientific
weight and importance to our approach. The study elaborates on medical image classification
capability using deep learning models and is an effort at understanding the prospective AI-
based tools in dermatological diagnosis that increase healthcare accessibility. Another great
feature of DermaLens is the real-time feedback system. It allows the user to express
experiences with the system, and hence, their experiences can become invaluable input that
could improve model accuracy and usability. This gives the system an evolution over time due
to its continuous adaptation toward both technological advances and user needs. DermaLens
aims to bridge the gap between medical expertise and accessibility by combining AI-driven
diagnostics with practical healthcare resources. Our final goal is to empower individuals with
the knowledge and tools necessary for early skin disease detection and management,
ultimately improving overall skin health and quality of life.

1.2 Literature Review

Below is the tabulated study of Papers available on this domain. The results are as follows.

Table 1 Literature Review

Conclusion
S. Journal Research
Author(s) Findings Conclusions Limitations of Published
NO and Year Focus
Work
Journal of Review of Deep
Deep Highlights
Medical various deep More learning is
learning- the success
Imaging learning diverse promising
1 Jain, S., et al. based skin of deep
and Health models for datasets are but requires
disease learning in
Informatic skin disease required more
classification dermatology
s, 2021 classification validation
AI can
Classification Showed Deep neural Dataset enhance
Esteva, A.,
of skin dermatologi networks diversity was dermatology
Kuprel, B., Nature,
2 cancer with st-level can help low, but needs to
Novoa, R. A., et 2017
deep neural classification dermatologi potential for be combined
al.
networks accuracy sts diagnose bias with clinical
experience
Presented VGG
Shown
VGG networks
Deep CNNs strong
networks High offer a good
Simonyan, K., for large- performance
3 ICLR, 2015 with very computation baseline for
Zisserman, A. scale image in image
deep al cost deep
recognition recognition
architecture learning
tasks
s models
State-of-the-
Introduced DenseNets
Huang, G., Liu, art
Densely DenseNet Requires improve
Z., Van Der performance
connected architecture more learning
4 Maaten, L., CVPR, 2017 on image
convolution to improve memory and efficiency
Weinberger, K. classification
al networks gradient computation and feature
Q. tasks
flow reuse
obtained
MobileNet- MobileNet-
Utilized
based skin Mobile- Accuracy is based
Velasco, J., MobileNet
disease based dependent models
Pascion, C., CNN for
5 arXiv, 2019 classification approach on camera provide real-
Alberio, J. W., et mobile skin
using provides quality and time skin
al. disease
smartphone accessibility lighting disease
detection
s classification
Detection of
Proposed a Skin diseases Needs a Machine
skin disease
Procedia detection can be bigger learning can
based on
Computer system detected dataset for assist
6 ALenezi, N. S. A. image
Science, based on effectively more dermatologi
processing
2019 machine using generalizatio sts in
and machine
learning machine n diagnosis
learning
 learning
techniques
Machine Explored Machine
learning various ML ML models Limited learning is a
Bhadula, S., et IJITEE, algorithms algorithms showed datasets and promising
7
al. 2019 for skin for skin promising overfitting solution for
disease disease accuracy problems skin disease
detection classification detection

1.3 Conclusion

DermaLens takes a big step forward in the sphere of AI dermato-diagnosis, providing a very
easy, fast, and accurate means of identifying skin diseases. A DenseNet-based CNN model was
trained on the DERMNET dataset, thereby achieving high classification accuracy, with an
average of 90.5%. The user receives immediate and accurate results. Besides detection of skin
diseases, the system renders educational information and recommendations for
dermatologists, while also collecting user feedback in real-time to further improve the user
experience.
DermaLens holds promise of transforming dermatological care by furthering research on a
larger dataset and better model training algorithms to create an interface between medical
expertise and accessibility. This technology especially in the under-represented regions with
fewer dermatologists will enable patients to adopt an informed approach that helps in early
diagnosis, subsequent treatment, and health outcomes.
 2. Background Theory

This is a deep learning-based project on the classification of skin disease using the DenseNet
architecture. The architecture applies the design of a convolutional neural network that helps
in feature propagation, reduces redundant computations, and develops a model, which then
gets trained on the DERMNET dataset. This dataset includes images representing 23 different
types of skin diseases. Data pre-processing, such as oversampling and under sampling,
helped to balance the dataset. To train this model, Kaggle is used for 48 epochs, with an
accuracy of 90.5%.

The solution was made more user-friendly by developing a web-based interface, thereby
integrating it with Flask for the back end and using HTML and CSS for the front end. The skin
image of the user could be uploaded into the system, and they could find the results
immediately based on the classification. There are also Google Integrated Location Services
to suggest nearby dermatologists, and Google Search API and YouTube API fetch relevant
articles and videos for further guidance. Libraries such as NumPy, TensorFlow, and Pillow
were used for image processing and deep learning model implementation. The system is
deployed using Visual Studio Code (VS Code) and continuously improved through customer
feedback.

2.1 Background Theory:

This section gives complete knowledge and understanding of different technologies and
framework that were required in this project for its completion. The background Theory is
listed below:

2.1.1 DenseNet
Densely Connected Convolutional Networks or DenseNet for short is one of the architectures
in deep learning that enhances the reuse of features and reduces the vanishing gradient
problem. While traditional CNNs learn layers individually, DenseNet ensures that all
preceding layers give additional information to each layer for better learning efficiency and
reduced computations. In this project, the DenseNet will be used to classify skin disease by
optimizing the feature extraction based on its densely connected layers with high accuracy.

2.1.2 Flask
Flask is one of the simplest lightweight web frameworks for Python to support rapid web
application development and deployment. In fact, with this framework, one can readily have
the major features of a web application including routing, request handling, and template
rendering and also include all the other needed libraries. Here, Flask was used to seamlessly
create a back-end that integrated the trained deep learning model into the front end,
allowing a user to upload an image to get diagnostic results.

2.1.3 HTML & CSS

HTML and CSS are the backbone of web development. HTML is used to structure the content
of web pages, while CSS is used to enhance their visual presentation. In this project, HTML
and CSS are used to design an intuitive and user-friendly interface that enables users to
interact with the skin disease detection system effortlessly.

2.1.4 Google Integrated Location Services

Google Integrated Location Services provides geolocation-based functionalities allowing
applications to obtain the location of the user, and accordingly deliver location-based
services. In this project, it is integrated so that the user can get dermatologists nearby in his
location; hence, easy access to a professional medical consultation.

2.1.5 Google Search API & YouTube API

The Google Search API permits web content to be dynamically retrieved within applications,
and the YouTube API provides access to video resources. In this project, these APIs are used
to fetch videos and articles associated with detected skin conditions. The result of this feature
creates user awareness about the detected disease through educational content.

2.1.6 NumPy
NumPy (Numerical Python) is a fundamental library for numerical computing in Python. It
supports multi-dimensional arrays, mathematical operations, and linear algebra functions,
making it indispensable for scientific computing. In this project, NumPy is used to handle
image data efficiently and perform matrix computations essential for deep learning model
operations.

2.1.7 TensorFlow
TensorFlow is an open-source deep learning framework developed by Google. It provides a
flexible platform for building, training, and deploying machine learning models. In this
project, TensorFlow is used to implement and train the DenseNet-based skin disease
classification model, leveraging GPU acceleration for optimized performance.
2.1.8 Pillow
Pillow is the Python Imaging Library that has broad capabilities in the processing of images,
such as resizing, filtering, and changing format. In this project, it was used to perform pre-
 processing of the images like resizing and normalization before feeding the images into the
deep learning model.

2.1.9 Visual Studio Code

Visual Studio Code is a popular and versatile code editor that is used for its light weight and
availability of extensive plugin support. Here, it will be used in this project in writing,
debugging, and even deploying Flask-based web applications while ensuring efficient
development and seamless integration of the deep learning model into the application.

2.2 Background of Existing Application

While there are various projects and applications related to skin disease detection, each
project tends to have unique features and implementations based on specific requirements.
Few notable project in a similar domain are as follow:

1. DermAssist by Google – A web-based AI tool that analyzes images of the skin to provide
possible conditions and references to medical data. It doesn't give any direct medical
diagnosis but is a supporting tool.
2. SkinVision – Mobile application using AI to analyze the skin lesion, giving risk scores for
melanoma and other related conditions. The app provides suggestions but requires
subscriptions from users to access more comprehensive features.
3. Miiskin – It tracks changes in the skin over time. It does not give a direct diagnosis but helps
the users monitor moles and skin conditions for potential concerns.
4. Aysa – A mobile application that allows users to upload images of skin conditions for AI-
based preliminary analysis. It provides educational resources and suggests whether a
doctor's visit is needed.
5. DermaCompare- The use of AI in the analysis of moles from images taken in different times
will help in identifying possible skin cancer risks. It allows remote dermatological monitoring.
6. AI-Based Dermatology Research- Other studies like CNN-SVM-based mobile applications
have also been successful in attaining high accuracies in the classification of skin diseases.
These systems often use pre-trained models and various machine learning techniques to
improve outcomes.
2.3 Conclusion
The proposed system is a deep learning-powered web application that classifies skin diseases
and enhances user accessibility to dermatological care. Employing the DenseNet architecture
and leveraging Flask for backend integration, the model achieves a high classification
accuracy of 90.5%. The platform also integrates Google APIs for dermatologist
recommendations and educational resources. Compared to existing applications, this system
provides a well-rounded solution for users seeking early skin disease detection and
information.
Future improvements could be extended for a larger dataset to enhance accuracy, mobile
application integration for wider accessibilities, and improvement of the recommendation
system by using more advanced AI techniques. Since there are continuous improvements and
integration of real-time feedback in this project, it is sure to impact dermatological care as
well as awareness considerably.
 3. Aim and Objectives

In this section, the actual dissection of project is done and each module is built piece by piece
in order to complete the project. In design section, the application has been built in
accordance with functional requirements based on which diagrams like Use Case and Low-
Level Sequence Diagram is drawn. In implementation section, snips of important code is
displayed with their explanation given below. In testing section, all functional requirements
are tested and the result is analysed resulting in status of test condition. This chapter serves
as the compass for our project, delineating the overarching goal and specific
objectives that chart our course. It begins with a clear articulation of the project's title and
aim, providing a concise yet comprehensive overview. Delving deeper, we outline the
objectives essential for the project's fulfilment, each playing a crucial role in achieving our
desired outcome. The section meticulously documents functional requirements, offering a
detailed perspective
through the lens of various diagrams. Methodologies are presented to articulate the
procedural steps guiding the fulfilment of these objectives. As the chapter concludes, a
succinct summary ties together the aim, objectives, and methodologies, setting the stage for
a structured and purposeful project execution.

3.1 Title
DermaLens: Intelligent Skin Disease Identification and Care Guidance System
3.2 Aim
To empower users with accessible, reliable skin health insights and resources, fostering
informed and effective care.
3.3 Objectives

The objectives of the proposed Project are listed below:

1. To review the literature on skin disease detection.
2. To develop and train a deep learning model to identify various skin diseases from skin
images
3. To design a website that allows user to upload the photos of their skin to get the possible
diagnosis.
4. To provide severity, relevant articles and videos recommendations based on the detected
disease.
5. To find nearby dermatologists using integrated location services based on user opinion.
6. To provide a feedback option for the betterment of the website.
7. To validate the developed website by considering real time input from the user and
providing services to the user accordingly.
3.4 Functional Requirements
The functional requirements for this project are mentioned below.
FR1: The system should enable users to upload images of their conditions and let it classify
and determine what kind of skin condition they are suffering from.
FR2: The system should process the uploaded images against the CNN model, DenseNet.
FR3: The system should provide a diagnosis by identifying one among the 23 diseases.
FR4: The system should show relevant articles and videos relating to the diagnosed skin
condition.
FR5: The system should suggest treatment guidelines of the detected disease.
FR6: The system shall suggest dermatologists according to the location of the user using
Google API.
FR7: The system shall permit users to rate the correctness of the diagnosis and
recommendations.
FR8: The system shall be able to accept user registration and authentication.
FR9: The system shall implement secure login with account management functionality and
password reset.
FR10: The system must provide real-time validation of its predictions by acquiring feedback
from users.
FR11: The system should enable a user to obtain a second opinion by consulting with a
dermatologist through the website.

The functional Requirements for User Interface module are mentioned below:
FR1: Interface should be user-friendly as well as intuitive.
FR2: The system should support responsive design for mobile devices, tablets, and desktops.
FR3: It will provide easy navigation for uploading images and viewing results.
FR4: It will allow users to filter and search through articles and videos.
FR5: It should present interactive elements like buttons, drop-down menus, and forms.
Functional Requirements for Image Processing & Model Integration:
FR1: Image preprocessing should occur to ensure standard input to the model.
FR2: It should classify a skin disease correctly with a relatively high accuracy value.
FR3: The system should provide each prediction with appropriate confidence scores.
FR4: The system improves predictions over time based on the feedback of its users.
FR5: Allow integration with later model updates if any for an improved accuracy score.
Non-Functional Requirements:

NFR1: The system or software should load pages and provide responses to user inputs within
2-3 seconds.
NFR2: The system or software should be available 24/7, ensuring users can access it
whenever needed.
NFR3: The system or software should be available on operating systems like windows/
android.
NFR4: The system or software should be well structured and modular, making it easy to
maintain, update and add new features without major disruptions.
NFR5: User medical data and images should be anonymous to protect personal identities
when stored or processed.

3.5 Methods/Methodology
Table 2 Methods and Methodology
3.6 Conclusion
This particular chapter focused on defined title and Aim of the project correctly. Later this
chapter included required objectives that were required to be fulfilled to complete this
project. Functional Requirements is documented successfully in this section followed by
method and methodologies which is successfully tabulated in order to know the steps used in
completing the objectives including resources used.
 4. Problem Solving

In this section, the actual dissection of project is done and each module is built piece by piece
in order to complete the project. In design section, the application has been built in
accordance with functional requirements based on which diagrams like Use Case and Low
Level Sequence Diagram is drawn. In implementation section, snips of important code is
displayed with their explanation given below. In testing section, all functional requirements
are tested and the result is analysed resulting in status of test condition.

4.1 Design
Design is necessary when it comes to development since it acts as a blueprint for entire process
from requirement making to finished (final product). Hence, in this section designs specific to
this project like use case and sequence diagram are attached.

Figure 1. System Architecture

Process:

1. Preprocessing:

• Cleaning and preprocessing raw data to remove inconsistencies and anomalies.

• Carry out data transformation, normalization, and feature engineering to prepare

datasets for analysis.

2. Feature Extraction:

• The key features extracted from the datasets will be those relevant to crop analysis,
such as temperature, humidity, soil type, crop growth duration, and market prices.

• Use statistical methods and domain knowledge to select informative features.

3. Machine Learning Model Selection:

• Choose a suitable machine learning algorithm for Skin disease detection—

CNN,Densenet etc.

• Consider factors such as model interpretability, accuracy, and scalability.

4. Model Training:

• Train the chosen machine learning model using historical data on weather conditions,
soil properties, and crop performance.

• In the Hyperparameters Tuning section, tune the hyperparameters of the model to get
better predictive accuracy.

5. Real-time Data Updates:

• Implement mechanisms for fetching and incorporating real-time data other relevant
information.

6. User Interface Integration:

• Integrate the results , articles recommendation and related videos recommendation

confidence rate and allows users to connect with dermatologists near them through
Google maps.

• Offer an intuitive interface where users can provide the details of land and receive
information on suggested crops.

7. Feedback Mechanism:

• Allow users to provide feedback on recommended results based on their experiences

and outcomes.

• Incorporate user feedback to continuously refine and improve the recommendation

engine.

8. Performance Monitoring and Evaluation:

• Monitor the performance of the recommendation engine by tracking metrics such as

prediction accuracy, user satisfaction, and crop yield.

• Conduct periodic evaluations and assessments to validate the effectiveness of the crop
analysis and recommendation module.
1. SEQUENCE DIAGRAM OF THE SYSTEM

Figure 2 Sequence Diagram

The Intelligent Skin Disease Identification and Care Guidance System is designed to assist
users in identifying skin diseases through AI-driven image analysis. The system workflow, as
depicted in the image, involves multiple components interacting sequentially to provide a
comprehensive diagnosis and care guidance experience.
System Workflow Overview
1. User Interaction
o The user uploads an image of the affected skin area to the system.
o The system processes the image using a DenseNet-based Convolutional Neural
Network (CNN) model.
2. Image Processing and Analysis
o The CNN model classifies the skin disease by analyzing the uploaded image.
o If classification is successful, the system provides a diagnosis result.
3. Additional Resources and Recommendations
o The system retrieves relevant articles from Google Search Engine to help users
understand their condition.
o It fetches relevant YouTube videos using the YouTube API for further educational
support.
o Severity analysis is conducted, and dermatologist recommendations are provided using
the Google API.
4. User Feedback and Data Storage
o Users can rate the diagnosis and recommendations, and their feedback is stored in the
database.
o If the system is unable to diagnose the condition, it notifies the user accordingly.
5. Error Handling
o If an invalid image is uploaded, the system returns an error message.
This system enhances accessibility to dermatological care, reducing dependency on
immediate clinical visits.
4.2 Implementation

Figure 3 Dermnet Dataset Code Snippet

The code begins with importing necessary libraries: os for directory and file manipulation,
matplotlib.pyplot for graphing the distribution of the dataset, random for randomly selecting
images, and shutil for copying, moving, and deleting files.

Then, the dataset directory is defined as dataset_dir = '/kaggle/input/dermnet', and two

subdirectories, train_dir and test_dir, are defined by appending this path with 'train' and
'test', respectively. The list of class names is then retrieved by reading the subdirectory
names under train_dir using os.listdir(train_dir), and sorted for good measure.

Two empty lists, train_data_counts and

test_data_counts,are created to keep track of how many images each class has available in
the training data and the testing data. A for loop loops through each
class_name within class_names and builds the appropriate paths for the training class
directories and testing class directories. The amount of image files present within each class
directory is retrieved via len(os.listdir(class_folder_path))
and is appended to train_data_counts and test_data_counts.

A new list, combined_data_counts, is generated to hold the total of training and testing
images per class by adding matching elements in train_data_counts and
test_data_counts. A bar chart is used to plot the distribution of the dataset with
plt.bar(class_names, combined_data_counts), with class names on the x-
axis and the count of images in each class on the y-axis. The title and labels of the
chart are determined by plt.xlabel(), plt.ylabel(), and plt.title(), and class names are
rotated by plt.xticks(rotation=60, ha='right') for readability.

A loop goes through each class index and count and employs plt.text(i, count, str(count),
ha='center', va='bottom') to show the count values above each bar in the plot. At last,
plt.tight_layout() optimizes the layout for the best spacing, and plt.show() renders the chart.

Following visualization, dataset balancing functions are established.

count_images(folder) is used to get the count of files within a directory.
The function oversample(folder, target_count) randomly copies images in a class folder until
the number of images is attained equal to target_count. It randomly picks images from the
folder with random.choice(image_files), renames them, and stores a copy with shutil.copy().

Likewise, the undersample(folder, target_count) function decreases the number of images in

a folder to the target count. It picks additional images randomly with random.sample()
and removes them with os.remove(). The move_images(source_folder, destination_folder,
count) function moves a specified number of images from the source folder to the
destination folder randomly with shutil.move().

The preprocess_dataset(training_dir, test_dir, output_dir, oversample_target=1405,

undersample_target=1300) function preprocesses the dataset by first copying
all the images to a new combined directory. It then balances the dataset by oversampling
classes with too few images or undersampling classes with too many images.

After balancing, the data is divided into training, validation, and test sets in the ratio 8:1:1.
The function makes new train, validation, and test folders and moves images
accordingly with move_images(). The line shutil.rmtree(combined_dir) deletes the
temporary combined folder after processing.

Next, the preprocessed dataset

is plotted in the same manner as the original dataset. New directories (train_dir, test_dir,
validation_dir) are created under the dataset directory. Three lists (train_data_counts,
test_data_counts, validation_data_counts) are initialized to store image
counts by class. There is an iteration over each class, counting images in each dataset
split and appending them to the respective lists.

The sum of train, test, and validation images per class is calculated and visualized in a
bar plot. The count values above each bar are inserted using the plt.text()
function, and the plot is rendered using the plt.show() function.

This entire process balances the dataset and

properly splits it for training and testing machine learning model.
Frontend Code Snippet

This HTML file acts as a web page. It serves as an end-user page to view the results of a skin
disease prediction model from input images. The page gives basic information about an image
uploaded by a user, the expected disease, confidence level, severity, and other
recommendations for further reading, as well as related videos. Further, it provides easy
navigation buttons and elements on the page for better user experience.
This document starts with the <!DOCTYPE html> standard, indicating to the arrangement of
an HTML5 document. Its <html> element defines this document as lang="en", which states
the language of the page, while the <head> section metadata includes character encoding with
UTF-8 so that the page is compatible with many text formats and includes viewport meta tags
for a responsive design. It identifies the title of the page as "Prediction Result," and an external
CSS file (/static/style.css) to style the page.
The next <body> section consists of two buttons at the top where the user can navigate either
back to the homepage (/) or to find nearby dermatologists (/dermatologists). These buttons
on the click event are redirected through the onclick attribute and are made look better using
the class navigate-button.
Main content is wrapped inside a <div> of class container. The introductory part of this section
is titled as <h1>Prediction Result</h1>. Following that, it is enclosed within another <div> with
the image-container class, where the outputted image will be shown with the <img> tag. Here,
the {{ image_url }} is essentially a placeholder through which the image URL gets dynamically
loaded from backend. The image has restrictions of maximum 300-pixel widths so as to ensure
a consistent display. After the image, there are some key prediction details. The name of the
disease is shown as <p><strong>Disease:</strong> {{ class_name }}</p>, where {{ class_name
}} will dynamically insert the predicted class name. Similarly, the model's confidence score is
shown rounded to two decimal places using {{ confidence | round(2) }}%. The disease severity
is also shown dynamically.
The webpage also has sections for recommended articles and videos, enhancing user
engagement and educational content. These are displayed as unordered lists (<ul>), with each
article or video being listed as a hyperlink. The {% for article in articles %} and {% for video in
videos %} loops go through lists of recommended articles and videos and dynamically generate
links, such as <a href="{{ article.url }}" target="_blank">{{ article.title }}</a>, that open in a new
tab.
At the bottom of the page is an extra button named "Submit Another". The user can return to
the home page for another prediction by clicking this button, which is styled with class back-
button for aesthetic purposes.
It represents an easy, user-friendly way to show what a machine learning model predicts in
terms of input images, their predictions, related educational materials, and other operations
that can be performed on such a prediction page. Dynamic usage of placeholders with {{ }} and
template loops using {% %} indicates integration into a backend framework like Flask or Django
for providing real-time information based on outputs from the models.
 Figure 4. Frontend Code Snippet

The style.css is applied to give a beautiful and interactive look to the skin disease prediction
webpage. It starts with the general reset for common HTML elements such as body, h1, h2, p,
ul, li, and a that remove default margin and padding so that it has a consistent layout. In a
clean and modern style, it is maintained, with a white background (#f4f4f9), dark gray to read
the texts, and major containers centered at white backgrounds along with subtle drop
shadows for rounded corners and producing a card like effect. It has a light blue teal #006d77
color for all headings that evoke a professional calm look while all subheadings are styled soft
gray. The image shown is with a border-radius and a drop shadow for better presentation.
Buttons and interactivity elements are well structured with a seamless user experience. The
top-level buttons come on a teal blue background with round edges and have a hover effect
in which the color darkens, thus appealing and easy to interact with. Underlined blue links for
lists of recommended articles and videos, which turn green during hover. Special positioning
is done on primary navigation buttons: "Home" (#home), "Find Dermatologists Nearby"
(#dermatologist), and "Feedback" (#feedback), ensuring accessibility. Error and success
messages are styled with fixed positions at the top-right corner of the screen with red
background for errors and green for success messages. This is a well-structured and aesthetic
CSS file that ensures a smooth and interactive user experience on the prediction results page.
USER INTERFACE SCREENSHOTS:

Figure 5. Home Screen UI

Figure 6. Exception handling Screen

4.3 Testing
All modules’ functional requirements are tested and are tabulate below:
User Interface Module:

Table 3 User Interface Module

FR No. Expected Obtained Result
FR1 Interface should be user- Users found the interface easy Pass
friendly as well as intuitive. to navigate and interact with.

FR2 The system should support The interface was responsive Pass
responsive design for mobile and adapted well to various
devices, tablets and desktops. devices and screen sizes.
FR3 It will provide easy navigation Uploaded image successfully Pass
for uploading images and and getting the results.
viewing results.
FR4 It will allow user to filter and For detected disease getting Pass
search through articles and the articles and videos.
videos.

Image Processing and Model Integration :

Table 4 Image Processing and Model Integration

FR No. Expected Obtained Result
FR1 Image processing should occur It is preprocessing and giving Pass
to ensure standard input to the the result.
model.
FR2 It should classify a skin disease Detect the skin disease Pass
correctly with a relatively high correctly with high accuracy.
accuracy value.
FR3 The system should provide each The confidence is predicting Pass
prediction with appropriate correctly.
confidence scores.

FR4 The system improves prediction By users feedback our website Pass
over time based on the is improving.
feedback of its users.
 5. Results

5.1 Model Prediction

Using the DenseNet architecture, the validation dataset is evaluated at the end of each epoch
to compute validation accuracy and assess the model's performance. DenseNet updates its
weights through training data, leveraging its densely connected layers for efficient feature
propagation and reuse, repeating the process over multiple epochs. Figure 10. shows the
model training with 48 epochs. Validation data at each epoch ensures the reliability of the
learned weights. Approximately 90.5% accuracy is obtained.

Figure 7. Model Training with 48 epochs

Accuracy and loss graphs are generated, comparing training and validation metrics. These
graphs are crucial for evaluating DenseNet's performance and determining the most accurate
model configuration. Figure 11. represents the Training and Validation accuracy curve. Figure
12. represents the Training and Validation loss curve.

Figure 8. Training and Validation Accuracy Curve

Training and validation accuracy curve represents how well the model is performing in terms
of classification accuracy. Here Training and validation curve is increasing over time that shows
model is performing well. The loss curve represents the model’s error on the training set. Loss
curve is decreasing over time.

Figure 9. Training and Validation loss curve

5.2 Frontend UI

The website is designed to assist users in diagnosing skin conditions and providing
comprehensive support. Users can upload photos of their skin, which are analyzed by a CNN
deep learning model using DenseNet to predict potential diseases and assess their severity.
Based on the detected condition, the platform offers relevant articles, videos, and
dermatologist recommendations to help users better understand and manage their condition.
Additionally, integrated location services enable users to find nearby dermatologists for
further consultation. The website also collects feedback from users to continuously improve
its accuracy, functionality, and user experience. It will be validated using real-time user inputs
to ensure reliable predictions and personalized services. Figure 13. shows the workflow and
architecture of the DermaLens website. Figure 14. represents Overview and user interface of
the website. Figure 15. represents the recommendation of dermatologists nearby.
Figure 10. Workflow and Architecture of the DermaLens Website

Figure 11. Overview of the website

Figure 12. Recommendation of Dermatologists nearby

Figure 13. Feedback page
 6. Project Costing

This chapter deals with costing of this project which gives an overall estimation of expenses
that was required to complete this project. This Covers expenses of testing devices, Platform
and Hardware cost, Human Resource Cost and a grand total of Entire cost.

6.1 Project Cost Estimation

The cost of project is summarised in a tabular form displayed below:

Table 5 Cost Estimation
Serial Resources and Work Cost(Rs)
Number
1 Laptop (for development) 60,000/-

TOTAL 60,000/-

As this initiative did not involve the creation of a tangible model, there were no expenditures
associated with the development of a physical prototype. Nevertheless, the investment in
crafting the software through concurrent acquisition of new technological skills led to an
increase in human resource expenditures, as detailed in Table 8.
 7. Conclusions and Suggestions for Future Work

7.1 Conclusion

Skin diseases are one of the health problems common to millions of people around the world.
Early diagnosis is essential to prevent complications and give a proper treatment. But the
common diagnosis methods proved to be expensive and take longer time plus they often
cannot be accessed by many people.

Here, we presented Intelligent Skin Disease Identification and Care Guidance System using
Deep Learning based on a DenseNet-based Convolutional Neural Network. By this AI-based
Skin Disease Identification and Care Guidance System, one can identify 23 different types of
skin disease concerning images uploaded by the user with a high level of accuracy. Apart from
diagnosis, the system grades the disease severity level and also provides educational articles,
videos, and treatment recommendations.

Another notable feature of our system is a location-based service that provides information
about nearby dermatologists when a user requires immediate medical attention. A feedback
mechanism also has been integrated that allows users to rate the diagnosis and
recommendation ensuring continuous improvement of the system.

Our platform was real-time tested to prove its effectiveness in preliminary assessment and
guidance. Ultimately, it will benefit the middle class because it will bring AI-powered
dermatological assistance closer to them which may serve as an early diagnosis and awareness
tool for them to make informed decisions about their skin health before consulting a doctor.

7.2 Suggestion for Future Work

Improvements planned for the existing Intelligent Skin Disease Identification and Care
Guidance System, which is one of the richest and most powerful systems for detecting skin
diseases in an early phase, include the following:

1. Improved Model Accuracy and Dataset Expansion

Development of large, more diverse datasets for improved classification accuracy to cover
an even greater variety of skin diseases.

Train the model on additional skin diseases to cover a broader range.

Fine-tune the DenseNet CNN model with even more sophisticated architecture to improve
precision and recall.
 2. Mobile Application Development

Developing an Android and iOS mobile application for comprehensive accessibility,

For offline diagnosis in areas without connectivity.

3. Real-Time Dermatologist Consultation

Teleconsultation with dermatologists.An AI-assisted chatbot for real-time tips on skincare

and basic treatments. Improved Severity Assessment and Customized Recommendations

4. Improved severity gradings by adding more medical parameters.

Customized skin care recommendations based on past history of conditions experienced

by the user.

5. Integration with Wearable Devices

Integration with the smartwatches and skin sensors to keep a track of skin health over time
is also being pursued Multi-Language Support & Accessibility. Multi-language support will
be implemented to make the system more inclusive. Providing vision-impaired individuals
with voice-based assistance at the user interface enhancement features.

These future advancements will further increase the usability, accuracy, and accessibility of
AI-enhanced dermatological assistance, making it more effective and widely available.
 8. References

1. Rimi, T. A., Sultana, N., & Foysal, M. F. A. (2020, May). Derm-NN: skin diseases detection
using convolutional neural network. In 2020 4th International Conference on Intelligent
Computing and Control Systems (ICICCS) (pp. 1205-1209). IEEE.

2. Elngar, A. A., Kumar, R., Hayat, A., & Churi, P. (2021, August). Intelligent system for skin
disease prediction using machine learning. In Journal of Physics: Conference Series (Vol.
1998, No. 1, p. 012037). IOP Publishing.

3. Maduranga, M. W. P., & Nandasena, D. (2022). Mobile-based skin disease diagnosis

system using convolutional neural networks (CNN). International Journal of Image,
Graphics and Signal Processing, 12(3), 47.

4. Velasco, J., Pascion, C., Alberio, J. W., Apuang, J., Cruz, J. S., Gomez, M. A., ... & Jorda
Jr, R. (2019). A smartphone-based skin disease classification using mobilenet cnn. arXiv
preprint arXiv:1911.07929.

5. Inthiyaz, S., Altahan, B. R., Ahammad, S. H., Rajesh, V., Kalangi, R. R., Smirani, L. K., ...
& Rashed, A. N. Z. (2023). Skin disease detection using deep learning. Advances in
Engineering Software, 175, 103361.

6. Sharma, M., Jain, B., Kargeti, C., Gupta, V., & Gupta, D. (2021). Detection and diagnosis
of skin diseases using residual neural networks (RESNET). International Journal of Image
and Graphics, 21(05), 2140002.

7. ALEnezi, N. S. A. (2019). A method of skin disease detection using image processing and
machine learning. Procedia Computer Science, 163, 85-92.

8. Bhadula, Shuchi, et al. "Machine learning algorithms based skin disease detection."
International Journal of Innovative Technology and Exploring Engineering (IJITEE) 9.2
(2019): 4044-4049.

9. Likhitha, S., & Baskar, R. (2022, December). Skin cancer classification using CNN in
comparison with support vector machine for better accuracy. In 2022 11th International
Conference on System Modeling & Advancement in Research Trends (SMART) (pp. 1298-
1302). IEEE.
10. Aziz, A., Hartono, R., & Abdilah, R. (2020, June). Decision Support System for Detection
of Skin Diseases in Smart Health development planning. In IOP Conference Series:
Materials Science and Engineering (Vol. 858, No. 1, p. 012051). IOP Publishing.
11. Kumar, N & Kumar, P & Pramodh, K & Karuna, Yepuganti. (2019). Classification of Skin
diseases using Image processing and SVM. 1-5. 10.1109/ViTECoN.2019.8899449.

12. Li, H., Pan, Y., Zhao, J., & Zhang, L. (2021). Skin disease diagnosis with deep learning: A
review. Neurocomputing, 464, 364-393.

13. Patnaik, S. K., Sidhu, M. S., Gehlot, Y., Sharma, B., & Muthu, P. (2018). Automated skin
disease identification using deep learning algorithm. Biomedical & Pharmacology
Journal, 11(3), 1429.

14. Niranjana, R., Hemadarshana, T., Ilakkya, S., Krishnan, R. S., Epziba, J. J., & Preetha, T.
(2024, June). Enhanced Skin Diseases Prediction using DenseNet-121: Leveraging Dataset
Diversity for High Accuracy Classification. In 2024 3rd International Conference on
Applied Artificial Intelligence and Computing (ICAAIC) (pp. 1270-1278). IEEE.

15. Pramesti, E., Mutiara, A. B., & Refianti, R. (2023, December). Implementation of Deep
Learning Using Convolutional Neural Network for Skin Disease Classification with
DenseNet-201 Architecture. In 2023 Eighth International Conference on Informatics and
Computing (ICIC) (pp. 1-6). IEEE.

16. Bello, A., Ng, S. C., & Leung, M. F. (2024). Skin Cancer Classification Using Fine-Tuned
Transfer Learning of DENSENET-121. Applied Sciences, 14(17), 7707.

17. Zhang, B., Zhou, X., Luo, Y. et al. Opportunities and Challenges: Classification of Skin
Disease Based on Deep Learning. Chin. J. Mech. Eng. 34, 112 (2021).
 Page 1 of 31 - Cover Page Submission ID trn:oid:::1:3149401001

Supriya M S
Chandana_CSE_Skin disease detection system.docx
CSE

csemsruas

M. S. Ramaiah University Of Applied Sciences

Document Details

Submission ID

trn:oid:::1:3149401001 27 Pages

Submission Date 5,667 Words

Feb 8, 2025, 4:59 PM GMT+5:30

32,884 Characters

Download Date

Feb 8, 2025, 5:17 PM GMT+5:30

File Name

Chandana_CSE_Skin_disease_detection_system.docx

File Size

3.0 MB

Page 1 of 31 - Cover Page Submission ID trn:oid:::1:3149401001

 Page 2 of 31 - Integrity Overview Submission ID trn:oid:::1:3149401001

16% Overall Similarity

The combined total of all matches, including overlapping sources, for each database.

Filtered from the Report

Bibliography

Quoted Text

Match Groups Top Sources

31 Not Cited or Quoted 16% 7% Internet sources

Matches with neither in-text citation nor quotation marks
6% Publications
0 Missing Quotations 0% 15% Submitted works (Student Papers)
Matches that are still very similar to source material

0 Missing Citation 0%
Matches that have quotation marks, but no in-text citation

0 Cited and Quoted 0%

Matches with in-text citation present, but no quotation marks

Integrity Flags
1 Integrity Flag for Review
Our system's algorithms look deeply at a document for any inconsistencies that
Hidden Text would set it apart from a normal submission. If we notice something strange, we flag
it for you to review.
19 suspect characters on 1 page
Text is altered to blend into the white background of the document. A Flag is not necessarily an indicator of a problem. However, we'd recommend you
focus your attention there for further review.

Page 2 of 31 - Integrity Overview Submission ID trn:oid:::1:3149401001

 Page 3 of 31 - Integrity Overview Submission ID trn:oid:::1:3149401001

Match Groups Top Sources

31 Not Cited or Quoted 16% 7% Internet sources

Matches with neither in-text citation nor quotation marks
6% Publications
0 Missing Quotations 0% 15% Submitted works (Student Papers)
Matches that are still very similar to source material

0 Missing Citation 0%
Matches that have quotation marks, but no in-text citation

0 Cited and Quoted 0%

Matches with in-text citation present, but no quotation marks

Top Sources
The sources with the highest number of matches within the submission. Overlapping sources will not be displayed.

1 Student papers

M S Ramaiah University of Applied Sciences 13%

2 Internet

eitca.org <1%

3 Student papers

Global Banking Training <1%

4 Student papers

Notre Dame of Marbel University <1%

5 Student papers

Anglia Ruskin University <1%

6 Student papers

Lovely Professional University <1%

7 Internet

e-hir.org <1%

8 Student papers

Queen's University of Belfast <1%

9 Student papers

University of East London <1%

10 Internet

www.irejournals.com <1%

Page 3 of 31 - Integrity Overview Submission ID trn:oid:::1:3149401001

 Page 4 of 31 - Integrity Overview Submission ID trn:oid:::1:3149401001

11 Publication

Ankur Gupta. "Next Generation Computing and Information Systems", CRC Press,… <1%

12 Publication

Poonam Tanwar, Tapas Kumar, K. Kalaiselvi, Haider Raza, Seema Rawat. "Predicti… <1%

13 Internet

fastercapital.com <1%

14 Publication

"Data Science and Applications", Springer Science and Business Media LLC, 2024 <1%

15 Publication

"Innovative Data Communication Technologies and Application", Springer Scienc… <1%

16 Publication

Md. Ferdous Wahid, Md. Jahid Hasan, Md. Shahin Alom, Shamim Mahbub. "Perfor… <1%

17 Publication

Saleh Alyahyan. "A novel NEMONET framework for enhanced RCC detection and s… <1%

18 Publication

Surya S Durbha, Jibonananda Sanyal, Lexie Yang, Sangita S Chaudhari, Ujwala Bh… <1%

19 Publication

Takehiro Yamamoto. "Reranking and Classifying Search Results Exhaustively Bas… <1%

20 Internet

iarjset.com <1%

Page 4 of 31 - Integrity Overview Submission ID trn:oid:::1:3149401001