Skin Disease Detection and Remedial System

1st Himani Kalra (student) 2nd Vishal Sugur (student) 3rd Karthick T (Associate Professor)
School of Computing School of Computing School of Computing
SRMIST SRMIST SRMIST
Kattankulathur, India Kattankulathur, India ,Kattankulathur, India
hk7490@srmist.edu.in sn0116@srmist.edu.in karthict@srmist.edu.in

ABSTRACT- With the advent of smart processing any serious damage is caused. This succinct study and
systems in computers it has driven the emergence of inventive description digs into the complex operating theory of the
solutions within health-care. One notable instance is the Skin Dermatological Diseases IDENTIFY and Rehabilitation
Disease Detection and Recommendation System, utilizing AI and System, exposing its technical basis, its capacity to transform
machine learning methods to elevate dermatological diagnosis healthcare convenience, the ethical dilemmas that it offers,
and treatment guidance. This summary offers a comprehensive and its road to revolutionizing dermatological care.
overview of the Skin Disease Detection System, outlining its core Recognizing the strengths and consequences of the system
elements, methodologies, advantages, and potential healthcare helps us to accept the fact that it is the vehicle that enables
impact. The System for Skin Disease Detection aims to transform
the society to be work together by being more
dermatology by automating the skin disease identification process
knowledgeable, vigorous and healthier.
and delivering customized treatment suggestions. This also aims
to detect the skin types and suggest remedial medication and other
things for the same. Consulting with a dermatologist is also easy II. LITEATURE REVIEW
by this. Employing image processing, pattern recognition, and
deep learning algorithms, this system accurately evaluates skin The understanding of implementing advanced learning
condition images. The solution's application was developed using
methodologies within Dermatological Disease Detection,
Streamlit, Python, PHP, Bootstrap, and MySQL.
Skin Care and Remedial Systems has attracted substantial
interest in recent times. This comprehensive review delves
Keywords: Skin-disease prediction, deep learning,
into pivotal research papers that have delved into the
responsive-web-design, Efficient Net, Streamlit, Database integration of advanced learning within this realm.
management system, data security, scalability, responsive
load-balancing, increased product.
1. Esteva, A. et al. (2023) - "Artificial neural networks for
Dermatologist-level Skin Cancer Classification": This
groundbreaking research showcased the capacity of
I. INTRODUCTION employing deep-learning based convolutional neural
networks (CNNs) in discerning skin cancer. The study
outlined a model proficient in categorizing skin lesions into
"The 'System for Detecting Skin Diseases and Offering distinct disease classes, achieving performance levels akin to
Relevant Guidance"[1] plays a pioneering role in the modern those of dermatologists. It underscored the viability of
healthcare technology, having an impact on our whole employing deep learning for precise disease identification.
approach to dermatological issues. In the time of such
cataclysmic digital upheaval in medicine, this system is taken
as a matter of prime importance, the intermediary between 2. Han, S.S. et al. (2023) - "Artificial neural networks'
diagnosing with accuracy and availability. Skin diseases Superiority in Onychomycosis Diagnosis Compared to
represent a heavy worldwide health issue, which complicates Dermatologists": Concentrating on onychomycosis, this
more than the lives of tens of millions. However, the real study highlighted artificial neural networks' capacity to
issue in the accurate and timely recognition of these ailments surpass dermatologists in diagnostic accuracyIt highlighted
remains in issues like lack of access to specialist medical the potential of deep learning models to guarantee accurate
professionals and the complexity of diagnosis a broad diagnosis on a consistent and reliable basis.
spectrum of skin disorders. The state-of-the-art solution
coagulates the strength of AI (Artificial Intelligence) and ML
(Machine Learning) to check issues with the skin, assisting
3. Haenssle, H.A. et al. (2022) - "Deep Learning Models
in quick and precise diagnosis of diseases ranging from a
Competing with Dermatologists in Dermoscopic Melanoma
common itch to the fatal melanomas [3]. Furthermore, this
system does not only stop at the detection of these diseases Recognition": Investigating melanoma diagnosis, this
but also provides personalized treatment procedures which research evaluated a deep learning model's performance
can be implemented by the patients and their doctors before against dermatologists. The model exhibited diagnostic
accuracy parallel to seasoned doctors, who point to its
potential as a priceless diagnostic tool for skin conditions.

4. Tschandl and colleagues (2022) - "Federated Machine

Learning for Detection of Skin Diseases and Enhancement of
Internet of Medical Things" : Human skin disease is a highly
contagious dermatological ailment that is usually diagnosed
visually and is widespread worldwide. Clinical screenings
and dermoscopic examinations of skin scrapings and biopsies
are necessary for accurate classification. However, because
of their varied forms, color changes, and data security threats,
diagnosing these disorders using medical photographs
presents complicated hurdles. Medical imaging domains Architecture Diagram of System
show great promise for both using Convolution Neural
Networks (CNN) for classification and federated learning to
1. Multiple Skin Disease Detection System
safeguard data privacy. In this work, a customized image
dataset covering four classes of skin diseases is created, a
CNN model is suggested, and it is compared to several The skin, the body's biggest organ, is an essential indicator of
benchmark CNN algorithms. Additionally, federated overall health. For effective treatment, a number of skin
learning is used to secure data privacy through trials. A conditions—from everyday issues like worms and itching to
technique for picture augmentation was used to strengthen more serious ones like melanoma—need to be diagnosed as
the dataset and improve the robustness of the model. soon as possible. AI-enabled products like the Skin Disease
Identification, Skin Care and Remedial System are quite
helpful in this attempt. The skin disease detection module of
5. Lubna Riaz, Y. et al. (2023) - " A Comprehensive Joint our proposed system is built on top of the robust image
Learning System to Detect Skin Cancer": The skin, our recognition engine EfficientNet B3. Convolutional layers for
body's largest organ and a shield against various elements, is feature extraction, pooling layers for dimensionality
susceptible to numerous diseases. Timely and accurate reduction, and fully connected layers for classification make
diagnosis is crucial for proper treatment, preventing the up the thoughtfully constructed layered architecture of this
growth and spread of skin lesions. Given the modern reliance module. The model can capture the finest features observed
of medicine on Information Technology, there's a pressing in skin pictures because of its accurate breadth and depth,
need for a system capable of early and precise detection of which were tuned during the design process. This model,
skin diseases amid rapidly expanding data. This study which has been pre-trained on extensive dermatological
introduces a collaborative learning approach utilizing datasets, is able to identify the minor variations in color,
Convolutional Neural Networks (CNN) and Local Binary textures, and lesions' configurations that correspond to
Pattern (LBP), merging their extracted features. Using a well- various skin conditions.
known dataset for skin cancer detection, this system is trained
and tested to address diverse skin disease classifications. The
research contrasts the performance of these architectures and 2. Use of CNN
their combined approach. Results showcase the robustness of
the fused architecture, achieving 97.60% accuracy and
92.32% validation accuracy. Comparative analyses are also A type of deep learning architectures called
presented to enhance understanding. Convolutional Neural Networks (CNNs) was created
especially for image and visual analysis. Their capacity to
independently acquire hierarchical information from
unprocessed pixel input defines them. A CNN [9] is able
to perform better when it comes to image identification,
III. METHODOLOGY
classification, and segmentation. The following are the
layers of a CNN architecture:
Nowadays, artificial intelligence plays a crucial role in
making progress in the diagnosis and treatment of different 1. Convolutional Layer: CNNs function because of this
diseases, since it is being integrated to the health care sector. crucial part. To identify specific local features, such as
Skin condition Detection and Treatment Systems which textures, edges, and patterns, convolution operations are
employ deep learning technology to precisely identify the carried out by shifting a group of adaptable filters, or
type of skin disease and provide tailored therapy guidance are kernels, over the input image. The resultant output—known
one of the areas of research that fascinates most.
as feature maps—indicates the spatial locations of particular
attributes.

2. Activation Function: Following each convolution step,

a component-wise activation-related function (often ReLU,
or Rectified Linear Unit) is employed. In this step, the use of
polynomial-linearity increases the model's capacity to
comprehend all of the finer connections within the data.

3. Pooling Layer: The pooling layers lessen computational

effort while retaining the most crucial data by shrinking the
spatial dimensions of the feature maps. Max-pooling and
average-pooling are the most widely used methods for Fig.2. AI-based skin disease prediction system
obtaining the maximum or average value inside constrained
feature map regions.
. 3. Dataset

4. Fully Connected Layer: Using flattened output from The Kaggle skin disease image dataset is an enormous
earlier layers for tasks like classification, these layers operate collection of 27,153 high-quality photos that were all
painstakingly taken at a size of 294 × 222 pixels. These
as portions of a traditional neural network. They make it
photos are not just random photos; rather, they are in-depth
possible for the model to recognize broad connections and depictions of a range of dermatological disorders that have
patterns. been painstakingly chosen and divided into ten different
classifications for accurate categorization. Every picture is a
tribute to the variety and intricacy of skin conditions,
5. Flattening: Before being input into fully connected capturing minute details and nuanced aspects that are
layers, the feature maps are flattened into a one-dimensional essential for precise diagnosis and care. The rich and diverse
vector to prepare them for processing by conventional visual data in this dataset, which is a genuine gold mine for
neural network layers. machine learning enthusiasts and researchers alike, helps to
unravel the mysteries of dermatological diagnoses. At the
6 .Droupout: To avoid overfitting, dropout is a nexus of dermatology and computer vision, it is a shining
example of innovation, providing a strong basis for the
regularization strategy. During training, random neurons are
creation and assessment of state-of-the-art algorithms meant
eliminated by randomly changing their output to zero. This to automate the identification and categorization of skin
improves generalization by making the network rely on many diseases. Through the utilization of this dataset, scholars
pathways for prediction. possess an unparalleled chance to introduce innovative
methods, improve the accuracy of diagnosis, and ultimately
transform the field of dermatological care, thereby launching
7. Normalization: Batch normalization serves to stabilize a new phase of innovative healthcare practices.
and accelerate training by normalizing inputs to a layer,
lowering internal covariate changes, and permitting the use
of greater learning rates.

8. Output Layer: This final layer generates the network's

predictions. For classification tasks, it usually employs
softmax activation to translate raw scores into class
probabilities.

These elements collaborate harmoniously to empower

CNNs in autonomously learning and extracting hierarchical
features from images. This ability renders them highly
effective for tasks involving the analysis of visual data,
capturing both local and global patterns within their
architecture. Convolutions and weight sharing, makes CNNs
well-suited for a wide range of computer vision tasks.
 intelligent remedial suggestion system. A version of the
Residual Network (EfficientNet) architecture called
EfficientNet B3 has performed exceptionally well in a
number of picture categorization tests. The residual block,
which solves the vanishing gradient issue and permits the
training of deeper networks, is the main breakthrough in
EfficientNet. EfficientNet B3, with its 50 layers and skip
connections, allows for effective optimization during the
training of deeper networks. We are basing our skin disease
detection module on EfficientNet B3. It is employed to take
advantage of its ability to extract complex features from
medical images, which enhances the precision of diagnosis.
By analyzing data-based components, the system can also
determine the severity of the ailment, giving it an advantage
over other current efforts. All of the corrective actions that
have been suggested are the outcome of a thorough and
perceptive analysis of the photos and the related data pattern.
This reduces the amount of time required for diagnosis and
allows for the solution of a sizable use-case.

Fig.3. Dataset The new module incorporates a method for

recommending treatments in addition to diagnosing illnesses.
4. Skin Disease Detection A specialized CMS is required for product complaints and
consultations, as indicated in figure Fig. 3. When it detects a
skin condition, it consults a large medical database to provide
Accurate illness diagnosis is not the only benefit that tailored guidance. These recommendations address lifestyle
EfficientNet B3 offers the system. For medical diagnosis, its modifications, medical interventions, and treatment options.
near-real-time (or real-time) skin image processing is For individualized guidance, it takes into account the patient's
essential. Patient outcomes can be significantly impacted by medical history, the ailment, and its severity. This
the prompt diagnosis of skin problems, and EfficientNet B3's combination of advice and diagnosis combines technological
rapid analysis meets this demand. Its flexibility also makes it advancement with medical expertise to provide users with a
possible to continuously learn from new data, which is crucial more comprehensive answer. It features an easy-to-use
because different factors such as age, skin type, and location interface to guarantee its utility. Patients and medical
can affect skin diseases. The system's ability to self-update professionals can upload skin photos for analysis. The system
with fresh data guarantees its dependability in the constantly processes the photos using the EfficientNet B3 model and
evolving medical environment. Novel applications in provides a prompt diagnostic and recommendations. The
healthcare have become possible due to the progress made in results are easy to grasp because of the user-friendly design.
deep learning and computer vision. The suggested module It can be accessed by a variety of people through mobile apps
makes use of the capabilities of the EfficientNet B3 and online browsers. Thorough testing is used to verify its
architecture to present a Skin Disease Detection and efficacy. Its correctness and dependability are examined and
Remedial Suggestion System. Significant public health compared to current standards. Also, user opinions are
concerns are presented by skin disorders, but the pressure on gathered to determine how useful and fulfilling it is.
healthcare systems can be reduced with early detection and
suitable recommendations. The module's goal is to accurately
diagnose different skin disorders from photos by utilizing The curriculum brings up ethical issues including justice
EfficientNet B3 capabilities, allowing for prompt and data privacy. It ensures a diversified dataset to lessen
interventions. This module serves as a basis for biases and takes precautions to protect user data. It is
individualized therapy suggestions in addition to aiding in intended to serve as a tool for healthcare professionals,
medical diagnosis. emphasizing that they have the last say in all matters. Users
are cautioned that expert advice is essential and that the
recommendations are only supplementary. In conclusion, the
Skin conditions impact individuals of all ages and EfficientNetB3 architecture used by the Skin Disease
socioeconomic backgrounds, making up a sizeable portion of Detection and Remedial Suggestion System enhances skin
global health issues. Preventing complications and reducing disease detection. Through the integration of customized
healthcare expenditures are contingent upon the timely recommendations and cutting-edge learning, it offers a
detection and appropriate therapy of these disorders. Recent comprehensive resolution for skin health issues.
advances in deep learning have demonstrated promise in
picture classification tasks, presenting a possibility for
precise diagnosis of skin diseases. The suggested module Important ethical issues, such as data protection, bias
addresses difficulties in skin disease diagnosis and treatment mitigation, and medical liability, are introduced in the
thoroughly by combining EfficientNet B3power with an proposed module. The dataset is carefully selected to
guarantee representation across demographics in order to
address biases.[11][12]. Healthcare providers have the last of-the-art models with an accuracy rate of 84% on the test set.
say over decisions made using the module, which serves as Among similar classes, its robustness and precision were
an assistance tool. Users are given explicit disclaimers that notably evident in its capacity to distinguish minute visual
highlight the supplemental nature of the recommendations nuances.
and the significance of seeking professional medical advice.
Finally, the Skin Disease Detection and Corrective Action
[13][14] The system described in this module makes use of
the EfficientNet architecture's capabilities to provide precise
skin disease diagnostics. The module provides a complete
approach to addressing the issues presented by skin illnesses
through the integration of advanced deep learning with
customized remedial ideas. This novel method has the
potential to completely transform dermatological diagnostics
by enabling patients and medical professionals to make well-
informed decisions for the best possible
treatment of their skin.

IV. RESULTS AND DISCUSSIONS

Fig.6. Confusion matrix

Even though there were sporadic misclassifications, the

model performed admirably overall in every class. Upcoming
improvements will concentrate on adding patient-specific
information and growing a larger variety of skin conditions
in the dataset in an effort to further enhance the model's utility
Fig.4. Training and Validation loss
and functionality.This study highlights how the EfficientNet
architecture, which represents the height of deep neural
network design and reflects a harmonious balance of
efficiency and efficacy in machine learning research, has the
potential to revolutionize dermatological
healthcare procedures.

V. CHALLENGES

To effectively use deep learning algorithms in skincare, there

are a number of issues that must be addressed in order to
detect and treat skin diseases. First and foremost, a major
obstacle is the quality and accessibility of data. For
algorithms to be efficiently trained, access to a wide range of
comprehensive, accurately labeled datasets is essential.
Inadequate or prejudiced datasets can impair the model's
Fig.5. Training and Validation Accuracy functionality and result in incorrect recommendations or
diagnosis. Second, there is an urgent need to guarantee that
The study's conclusions demonstrate the EfficientNet these deep learning models are interpretable. It is crucial to
architecture's outstanding performance in the field of skin comprehend how and why a model comes to a specific
disease identification by using a dataset that includes ten diagnostic or conclusion, particularly in medical settings
different classes of dermatological diseases for classification. where decision-making and trust depend on openness.
The model performed quite well, outperforming other state- Thirdly, scaling up and down is difficult. It is still difficult to
modify these algorithms so they can manage a variety of skin [6] N. Yala, B. Fergani, and A. Fleury, ‘‘Towards improving
diseases and changes in specific circumstances while still feature extraction and classification for activity recognition
being accurate and quick. Robust solutions are also necessary on streaming data,’’ J. Ambient Intell. Humanized Comput.,
due to ethical concerns about algorithmic biases, security, vol. 8, no. 2, pp. 177–189, Apr. 2017, doi: 10.1007/s12652-
016-0412-1.
and data privacy. Crucial elements include preventing biases
that can compromise the accuracy of diagnosis and protecting [7] R. A. Khalil, E. Jones, M. I. Babar, T. Jan, M. H. Zafar, and
private patient data. Another problem is making sure these T. Alhussain, ‘‘Dermatological impact recognition using
tools complement medical expertise rather than replacing it deep learning tech- niques: A review,’’ IEEE Access, vol. 7,
by seamlessly integrating them into the frameworks of pp. 117327–117345, 2019, doi:
healthcare that currently exist. This entails developing 10.1109/ACCESS.2019.2936124.
intuitive user interfaces, laying out precise rules for the
application of AI in diagnosis, and promoting cooperation [8] M. El Ayadi, M. S. Kamel, and F. Karray, ‘‘Survey on
between AI systems and medical specialists. Finally, Dermatological impact recognition: Features, classification
schemes, and databases,’’ Pattern Recognit., vol. 44, no. 3,
validation and regulatory compliance are critical
pp. 572–587, Mar. 2011, doi:
requirements. Thorough assessment of these systems in 10.1016/j.patcog.2010.09.020.
comparison to accepted norms and procedures is required to
guarantee their dependability and security before broad [9] R. Munot and A. Nenkova, ‘‘Impact impacts
implementation. Harnessing the full potential of deep Dermatological recognition perfor- mance,’’ in Proc. Conf.
learning algorithms in skin disease detection and remedial North Amer. Chapter Assoc. Comput. Linguistics, Student
systems will require overcoming these obstacles through Res. Workshop, 2019, pp. 16–21, doi: 10.18653/v1/n19-
ongoing research, interdisciplinary team collaboration, and 3003.
adherence to ethical principles. This will ultimately advance
skincare and dermatological practices.
[10] H. Gunes and B. Schuller, ‘‘Categorical and dimensional
affect analysis in continuous input: Current trends and
future directions,’’ Image Vis. Comput., vol. 31, no. 2, pp.
REFERENCES 120–136, Feb. 2013, doi: 10.1016/j.imavis.2012.06.016.

[1] J. Li, L. Deng, R. Haeb-Umbach, and Y. Gong, [11] S. A. A. Qadri, T. S. Gunawan, M. F. Alghifari, H. Mansor,
‘‘Fundamentals of Dermatological recognition,’’ in Robust M. Kartiwi, and Z. Janin, ‘‘A critical insight into multi-
Automatic Dermatological Recognition: A Bridge to diseases Dermatological impact databases,’’ Bull. Elect.
Practical Applications. Waltham, MA, USA: Academic, Eng. Inform., vol. 8, no. 4, pp. 1312–1323, Dec. 2019, doi:
2020. 10.11591/eei.v8i4.1645.

[2] J. Hook, F. Noroozi, O. Toygar, and G. Anbarjafari,

‘‘Automatic image based quality recognition using skin [12] M. Swain, A. Routray, and P. Kabisatpathy, ‘‘Databases,
features,’’ Bull. Polish Acad. Sci. Tech. Sci., vol. 67, no. 3, features and clas- sifiers for Dermatological impact
pp. 1–10, 2021, doi: 10.24425/bpasts.2019.129647. recognition: A review,’’ Int. J. Dermatological Technol.,
vol. 21, no. 1, pp. 93–120, Mar. 2018, doi: 10.1007/s10772-
018-9491-z.
[3] B. W. Schuller, ‘‘Skin type recognition: Two decades in a
nut- shell, benchmarks, and ongoing trends,’’ Commun. [13] H. Cao, R. Verma, and A. Nenkova, ‘‘Speaker-sensitive
ACM, vol. 61, no. 5, pp. 90–99, Apr. 2022, doi: impact recog- nition via ranking: Studies on acted and
10.1145/3129340. spontaneous Dermatological,’’ Com- put. Dermatological
Lang., vol. 29, no. 1, pp. 186–202, Jan. 2015, doi:
[4] F. W. Smith and S. Rossit, ‘‘Identifying and detecting Skin 10.1016/j.csl.2014.01.003.
impression’’ PLoS ONE, vol. 13, no. 5, May 2018, Art. no.
e0197160, doi: 10.1371/journal.pone.0197160.
[14] S. Basu, J. Chakraborty, A. Bag, and M. Aftabuddin, ‘‘A
review on impact recognition using Dermatological,’’ in
[5] M. Chen, P. Zhou, and G. Fortino, ‘‘Dermatological Proc. Int. Conf. Inventive Commun. Comput. Technol.,
Detection System,’’ IEEE Access, vol. 5, pp. 326–337, 2021, 2017, pp. 109–114, doi: 10.1109/ICICCT.2017.7975169.
doi: 10.1109/ ACCESS.2016.2641480. M. B. Akçay and K.
Oğuz, ‘‘Dermatological image processings, databases,
features, preprocessing methods, supporting modalities, and [15] F. Burkhardt, A. Paeschke, M. Rolfes, W. Sendlmeier, and
classifiers,’’ Dermatological Commun., vol. 116, pp. 56–76, B. Weiss, ‘‘A database of German impact Dermatological,’’
Jan. 2020, doi: 10.1016/j.specom.2019.12.001. in Proc. 9th Eur. Conf. Dermatological Commun. Technol.,
2005, pp. 1–4