A review of A Method of Skin Disease Detection Using Image Processing And

Machine Learning
Endale Aragaw
ID: BDU1500524

Faculty of Computing, Bahir Dar Institute of Technology, Bahir Dar University, Bahir Dar,
Ethiopia, Email:[email protected]

ABSTRACT
The abstract provides a clear and concise overview of a study on skin disease detection using image
processing techniques. It highlights the prevalence of skin diseases, the limitations of current
diagnostic methods, and the role of computer vision in detection. The study proposes a simple and
cost-effective approach for detecting skin diseases using digital images of affected skin areas,
which involves resizing the input color image to extract features using a pretrained convolutional
neural network and classifying the features using a Multiclass Support Vector Machine (SVM).
The results of the study are impressive, with the system successfully detecting three different types
of skin diseases with 100% accuracy. The abstract also provides relevant background information
on skin diseases and their causes, as well as the specific context of the study in Saudi Arabia,
where skin diseases are common due to the hot and dry climate. Overall, the abstract is well-written
and effectively communicates the key findings and contributions of the study.

Keywords: Skin diseases; Image Processing; Computer Vision; Machine Learning

1. INTRODUCTION
Skin diseases are a major health concern worldwide, with skin cancer being the most common
form of cancer in many countries. Early detection and diagnosis of skin diseases can significantly
improve patient outcomes and survival rates. However, the visual diagnosis of skin diseases can
be challenging, even for experienced dermatologists. In recent years, there has been a growing
interest in the use of image processing and machine learning techniques to aid in the detection and
diagnosis of skin diseases. The introduction of the paper provides a comprehensive overview of
the importance of early detection and treatment of skin diseases, and the potential of image
processing and machine learning techniques in improving the accuracy and efficiency of skin
disease diagnosis.
The introduction begins by highlighting the significant impact of skin diseases on public health,
with skin diseases affecting people of all ages and demographics worldwide. It also emphasizes
the importance of early detection and treatment of skin diseases in preventing further
complications and improving patient outcomes.
The authors then proceed to discuss the traditional method of skin disease diagnosis, which
involves visual inspection of the affected area by dermatologists. The authors note that this method
can be time-consuming and prone to errors, which underscores the need for automated methods of
skin disease detection.
The authors then highlight recent advances in image processing and machine learning, which have
opened up new possibilities for the automated detection and diagnosis of skin diseases. The
authors note that these techniques can be used to analyze images of skin lesions and accurately
classify them into different categories of skin diseases, which can significantly improve the
accuracy and efficiency of skin disease diagnosis.
The introduction also provides a brief overview of related work in the field of skin disease
detection using image processing and machine learning. The authors note that numerous studies
have been conducted in recent years to develop automated methods for skin disease detection,
employing various techniques such as texture analysis, color-based segmentation, and deep
learning algorithms.
Finally, the introduction presents the main contribution of the paper, which is a novel method for
skin disease detection using image processing and machine learning. The authors note that the
proposed method is based on a combination of feature extraction, image segmentation,
and classification algorithms, and is designed to accurately classify skin lesions into different
categories of skin diseases.
Overall, the introduction provides a comprehensive and well-organized overview of the
importance of skin disease detection, the potential of image processing and machine learning
techniques in improving skin disease diagnosis, and the main contribution of the paper. The
authors present a clear and concise argument for the significance of the proposed method and
provides a solid foundation for the rest of the paper.

2. OBJECTIVE
The objective of the paper is to propose a method for the automated detection of skin diseases
using image processing and machine learning techniques. The authors aim to develop a system
that can accurately classify skin into various disease categories, such as melanoma, basal cell
carcinoma, and squamous cell carcinoma, using digital images of the affected area. The proposed
method involves pre-processing the images, extracting relevant features, and training a machine
learning model to classify the images. The authors hope that this research will contribute to the
early detection and treatment of skin diseases, ultimately improving patient outcomes.

3. METHODOLOGY
A. Tools and techniques
Based on the methodology section of the paper, the authors propose a system for the detection,
extraction, and classification of skin disease images using a combination of preprocessing, feature
extraction, and classification techniques.
The preprocessing step involves resizing the input images to a standardized size, which is
necessary for ensuring that the system can extract the same number of features from all images.
Resizing the images also helps to reduce processing time and improve system performance.
However, it's worth noting that there may be other image preprocessing techniques that could be
used to improve the accuracy of the system, such as contrast enhancement or noise reduction.
For feature extraction, the authors propose using a pre-trained convolutional neural
network (CNN), specifically the AlexNet model, which has been shown to be effective for visual
tasks and natural language processing. Using a pre-trained CNN is a common approach for feature
extraction in image classification tasks, as it allows for the use of the powerful feature extraction
capabilities of the CNN without the need for extensive training on the specific dataset. However,
it's worth noting that there are many other pre-trained CNN models available that could be used
for feature extraction, and the choice of model may affect the accuracy of the system.
Finally, the classification step involves using a support vector machine (SVM) to train a classifier
using the extracted features from the training set. SVMs are a commonly used machine learning
algorithm for image classification tasks, as they can effectively separate different classes of data
in high-dimensional feature spaces.
Overall, the methodology section of the paper provides a clear and concise description of the
proposed system for skin disease image classification. The use of a pre-trained CNN for feature
extraction and an SVM for classification is a well-established approach for image classification
tasks, and the authors provide a clear justification for their choices. However, the paper does not
provide a detailed evaluation of the performance of the proposed system, which would be
necessary to determine its effectiveness in practice.
One potential limitation of the proposed system is that it relies solely on visual features extracted
from the skin disease images. While visual features can be informative for skin disease diagnosis,
there may be other types of data, such as patient history or genetic information, that could also be
informative for classification. Incorporating other types of data into the classification system could
potentially improve its accuracy.
B. Dataset Preparation
In this study, 100 skin images were used by several dermatological disease patients, also were
taken from the Internet. The proposed system can successfully detect 3 different skin diseases with
an accuracy of 100%. They authors used 20 of images for validation purpose and 80 images for
training purpose. The system works well. The detection rate of the system is 100 The detection
rate of diseases is very high 100%.
Based on the paper, it seems that the authors of the paper have compiled a dataset of skin disease
images, which could be useful for developing machine learning models or other computer vision
applications related to dermatology.
The dataset contains 80 images for each of the four skin diseases, which gives a total of 320 images.
It's worth mentioning that the dataset has an equal number of images for each disease category,
which could be helpful to ensure that the model is not biased towards any specific disease.
However, the description doesn't provide much information about the quality of the images or how
they were selected. It would be helpful to know whether the images were reviewed by
dermatologists or other medical professionals to ensure that they accurately represent each disease.
In addition, 80 images per disease category is a relatively small sample size, which could limit the
effectiveness of any machine learning model trained on this dataset. It's possible that a larger
dataset with more diverse images could lead to better results.
While the dataset described in the paper could be a useful resource for researchers working on skin
disease classification or other related tasks, its small size and lack of information about image
quality and selection could limit its usefulness in some contexts.
Another potential issue with the dataset is that it was compiled by collecting images from different
websites. This means that there may be variability in terms of the image quality, resolution,
lighting conditions, and other factors that could affect the accuracy of any machine learning
model trained on the dataset. The authors don't provide any details about how they addressed these
potential issues, so it's difficult to know how reliable the dataset is.
The dataset is its relatively small size. While 80 images per disease category is a decent starting
point, it may not be enough to train a robust machine learning model that can accurately classify
skin diseases. Ideally, a larger dataset with more diverse images would be better for this purpose.
It's also worth noting that the dataset only includes four categories of skin diseases: normal skin,
melanoma, eczema, and psoriasis. While these are certainly important skin conditions to be able
to diagnose, there are many other types of skin diseases that could be relevant for dermatological
diagnosis. Therefore, a larger and more diverse dataset that includes a wider range of skin diseases
would be more useful for training a comprehensive machine learning model. But, while the dataset
described in the paper could be a useful starting point for researchers working on skin disease
classification, it has some limitations that should be taken into account. Researchers should be
cautious when using this dataset, and should consider supplementing it with additional data
sources to ensure the accuracy and reliability of any machine learning models that are developed.
The dataset described in the paper is that there is no information provided about the demographics
of the patients whose skin disease images were included. It's possible that certain skin conditions
may be more prevalent or have different manifestations in different populations, so having a
diverse and representative sample of patients would be important for developing accurate
diagnostic tools.
The description of the dataset doesn't mention whether the images were taken at different stages
of disease progression or severity. It's possible that images taken at different stages could have
different visual characteristics that could affect the accuracy of any machine learning
model trained on the dataset.
Finally, while the equal distribution of images between disease categories is intended to prevent
bias, it's possible that there are other sources of bias in the dataset that aren't immediately apparent.
For example, the images may have been selected based on certain visual characteristics that are
more common in certain skin diseases, which could affect the accuracy of any machine learning
model trained on the dataset.
Overall, while the dataset described in the paper could be a useful starting point for researchers
working on skin disease classification, it has some limitations that should be considered.
Researchers should be cautious when using this dataset and should consider supplementing it with
additional data sources to ensure the accuracy and reliability of any machine learning models that
are developed.

4. CRITIQUES
A. Strength
• The paper addresses an important problem of automated skin disease detection, which can
potentially contribute to the early diagnosis and treatment of skin diseases.
• The proposed incorporates both image processing and machine learning techniques, which
are widely used in computer vision applications and have shown promising results in
medical image analysis.
• The paper provides a detailed description of the methodology, including the steps involved
in pre-processing, feature extraction, and classification, which can be helpful for
researchers and practitioners in the field.
• The paper presents experimental results on a dataset of skin lesion images and compares
the performance of different machine learning algorithms, demonstrating the effectiveness
of the proposed method.
B. Weaknesses:
• The paper lacks a detailed discussion of the limitations and challenges of the proposed
method. For example, it is unclear how the method would perform on images with poor
quality or different imaging modalities, or how it would generalize to different populations
or ethnicities.
• The paper does not provide a thorough analysis of the features used in the proposed method,
or how they were selected. This makes it difficult to assess the relevance and importance
of the features in the classification task.
• The paper does not compare the proposed method with other state-of-the-art methods for
skin disease detection, which would provide a better understanding of its strengths and
weaknesses.
• The paper does not provide access to the dataset used in the experiments or the code for
the proposed method, which limits the reproducibility and evaluation of the results by other
researchers.