You are on page 1of 6

Cognitive Inspired Machine Learning for Skin Cancer Disease

𝐷𝑟. 𝐹𝑎𝑤𝑎𝑑 𝑁𝑎𝑠𝑒𝑒𝑚1𝑠𝑡 , 𝐴𝑑𝑒𝑒𝑙 𝐴ℎ𝑚𝑒𝑑 2𝑛𝑑 , 𝑇𝑎𝑙ℎ𝑎 𝑍𝑎𝑓𝑎𝑟 3𝑟𝑑 , 𝑆𝑎𝑖𝑟𝑎 𝑇𝑎𝑟𝑖𝑞4𝑡ℎ , 𝑀𝑎ℎ𝑛𝑜𝑜𝑟 𝑆𝑎𝑟𝑑𝑎𝑟 5𝑡ℎ

1. Dr. Fawad Naseem – Ph.D Scholar, Lecturer Department of Computer Science, Superior University
Lahore, Pakistan.
2. Adeel Ahmed – Mphil Scholar, Department of Computer Science, Superior University Lahore,
Pakistan.
3. Talha Zafar – Mphil Scholar, Department of Computer Science, Superior University Lahore,
Pakistan.
4. Saira Tariq – Mphil Scholar, Department of Computer Science, Superior University Lahore,
Pakistan.
5. Mahnoor Sardar – Mphil Scholar, Department of Computer Science, Superior University Lahore,
Pakistan.

Abstract:
One of the most common types of cancer, skin cancer, has been on the rise in recent decades. Skin cancer
is caused by a number of dermatological conditions. Based on the texture, color, morphological
characteristics, and structure, there are several different forms of skin cancer. The traditional method of
identifying skin cancer requires time and money in order to produce the expected results. Nowadays, skin
cancer is categorized by medical science using a variety of digital technology-based methods. The reliable
and learning classification system technique for an automatic way of classifying skin cancer is machine
learning-based. Malignant and benign skin cancers are distinguished using the different existing and
proposed methods of deep neural networks, support vector machines, neural networks, random forests, and
K-nearest neighbors. In this study, a strategy for classifying melanoma and benign skin tumors was
presented based on the stacking of classifiers with three folds.
………………….to be continue……………………

Keywords: skin cancer, machine learning, Artificial Intelligence

Introduction:
It is challenging for medical research to find a way to diagnose and treat cancer in people. The most common
type of cancer is skin cancer. One of the tumors’ with the fastest rate of growth and greatest risk is
melanoma. It is exceedingly challenging to treat this cancer when it is advanced. Reducing the number of
cancer patients is the aim of early detection and treatment of this type of cancer. One of the worst forms of
skin cancer is malignant melanoma. Because the patient's health issues worsen over time, early detection
of this type of cancer is especially important. Melanoma starts when abnormalities form in the melanin, the
pigment that provides skin its colour. HPV has the capacity to penetrate our skin's lower layers, get into the
bloodstream, and then spread to other parts of our bodies.
Early skin cancer diagnosis and identification necessitate the use of computer assisted technology and
methodologies. By using computer assisted methods and tools, clinical diagnosis for cancer detection can
be made more accurately. Dermoscopy is the most important non-invasive technique for identifying
malignant, benign, and other pigmented skin malignancies. The classic techniques for melanoma detection
and major feature identification use eye-based examination and monitoring of colour changes in the skin.
This time-tested method for finding skin cancer depends on the appearance and colour of the skin's surface.
Dermoscopy enables more accurate classification of different types of skin cancer based on how they seem
and how they develop. Dermatologists use their expertise to review dermoscopy images. Due to the
complexity and subjectivity of human interpretation, computerized analysis of dermoscopy images has
become a significant research area. Dermoscopy images can be used to detect cancer and increase the
accuracy of skin cancer diagnosis.
………………..to be continue………………..

Literature Review:
Melanoma is a skin condition that frequently results in death. The uncontrolled expansion of genetically
altered melanocytic stem cells is thought to be the first stage in the development of melanoma. On the other
hand, melanoma detection is a very difficult task. As a result, skin cancer detection and identification greatly
improve pathologists' ability to diagnose the disease. In this study, a deep learning method is demonstrated
for accurately identifying the type of melanoma that is present at an early stage. The suggested method
distinguishes between nodular melanoma, superficial spreading, and lesion maligna. This enables the early
detection of the virus and the prompt isolation and treatment required to halt future infection transmission.
The deep layer structure of the convolutional neural network, a neural network technique, is an example of
both deep learning and the conventional non-parametric machine learning approach. As seen by its total
accuracy of 88.83 percent, our suggested technique, which is a two-stage learning platform, has excellent
projected accuracy at each level. This applies to all classification algorithms, including DT, RF, and GBT.
Skin illnesses can now be identified much more rapidly and precisely because to improvements in laser and
photonics-based medical technologies. Nonetheless, the expense of such a diagnosis is still prohibitive and
high. Hence, image processing methods aid in the beginning development of an automated dermatology
screening system. The classification of skin diseases relies heavily on the extraction of features. In a number
of methods, computer vision plays a part in the detection of skin conditions. Skin infections are prevalent
in Saudi Arabia as a result of the deserts and the hot climate. This project advances the study of skin disease
detection. We suggested a method for identifying skin conditions based on image processing. This
technique uses a digital photograph of the affected skin area to identify the kind of disease through image
analysis. A camera and a computer are the only pricey pieces of equipment needed for our straightforward,
quick method. The method uses a colour image's inputs as its basis. Then use a pretrained convolutional
neural network to resize the image and extract features. After that, a multiclass SVM was used to classify
the feature. The user is then presented the results, which include the type of disease, its distribution, and its
severity. With a 100% accuracy rate, the system can accurately identify three different kinds of skin
illnesses.
In this study, machine learning (ML) algorithms such as logistic regression, linear discriminant analysis, k-
nearest neighbours classifier, decision tree classifier, and Gaussian naive Bayes were tested, while deep
learning (DL) models were either based on a custom Convolutional Neural Network model or used transfer
learning by using previously trained models (VGG16, Xception and ResNet50). We discover that all DL
models outperform ML models, with accuracies up to 0.88. Accuracy for ML models is below 0.72, but
with ensemble learning, it can go up to 0.75. We test DL models on a larger and more unbalanced dataset
in order to further evaluate their performance. Pre-trained models perform remarkably well for skin tumour
classification after fine-tuning, according to metrics like the F-score and accuracy. This is especially true
for VGG16, which shows metrics of 0.70 and 0.88 on the larger database and an F-score and accuracy of
0.88 on the smaller database.
In order to solve these problems and assist dermatologists, machine-learning (ML) and deep-learning (DL)
techniques have been created. Early and correct diagnosis of SC is essential to improve patient survival
rates. With the intention of giving researchers just starting out in this field a reliable starting point, we give
a thorough literature analysis of current research on the use of machine learning to categorize skin lesions.
Many electronic databases were searched using inclusion/exclusion filters, and for this review, only those
documents that accurately and fully detailed the steps taken and reported the outcomes were chosen. Sixty-
eight publications were chosen, the majority of which use DL methods, particularly convolutional neural
networks (CNN), while the remainder use machine learning (ML) or hybrid ML/DL methods for the
identification and classification of skin cancer. Many ML and DL techniques perform well as classifiers of
skin lesions. The encouraging outcomes attained thus far portend the potential integration of these methods
in clinical practice in the not-too-distant future.
By combining the results of numerous repetitions, the CNN approach might be used to categorize images
of skin lesions. Then, for fine-tuning, several transfers learning models, including Resnet50, InceptionV3,
and Inception Resnet, were applied. This study's originality and contribution is the use of ESRGAN as a
preprocessing step, in addition to testing out a variety of models (including the intended CNN, Resnet50,
InceptionV3, and Inception Resnet). Results from our created model were equivalent to those from the
pretrained model. Simulations utilizing the ISIC 2018 skin lesion dataset demonstrated the effectiveness of
the suggested approach. The CNN model outperformed the Resnet50 (83.7%), InceptionV3 (85.8%), and
Inception Resnet (84%) models in terms of accuracy.
Deep neural network-based categorization techniques for computer-assisted skin cancer typically produce
predictions based solely on photos of skin lesions. While showing encouraging results, it is still feasible to
perform better by taking into account patient demographics, which are significant indicators that human
professionals take into account when screening for skin lesions. In this paper, we discuss the challenge of
applying deep learning models for skin cancer classification to combine image and metadata elements. In
our proposed metadata processing block, a novel approach, we enhanced the most pertinent feature
collected from the images throughout the classification pipeline by using metadata to support data
classification. We contrasted the suggested strategy with two additional combination approaches—one
based on features concatenation and the other on the meta net. Findings from two distinct skin lesion
datasets demonstrate that our method outperforms existing combination approaches in 6 to 10 scenarios
and improves classification for all tested models.
One of the most serious types of cancer is skin cancer. Unrepaired DNA breaks in skin cells, which result
in genetic flaws or mutations on the skin, are the primary cause of skin cancer. Because skin cancer is more
treatable in its early stages and has a tendency to gradually spread to other body areas, it is best identified
at an early stage. Early detection of skin cancer symptoms is necessary due to the rising incidence of cases,
high death rate, and expensive medical treatments. Given the gravity of these problems, researchers have
created a number of early-detection methods for skin cancer. Skin cancer is detected and benign skin cancer
from melanoma is distinguished using lesion features including symmetry, colour, size, form, etc. This
research gives a thorough analysis of deep learning methods for skin cancer early detection. The topic of
skin cancer diagnosis-related research articles from reputable publications were examined. For easier
understanding, research findings are presented via tools, graphs, tables, methodologies, and frameworks.
Skin cancer can be detected early thanks to AI. Deep convolutional neural networks, for instance, can be
used to construct a system that assesses skin image data to detect skin cancer. For skin cancer to be
effectively treated and to have better outcomes, early detection is crucial. Specialists are capable of
precisely diagnosing cancer, but due to their scarcity, it is necessary to create automated systems that can
do it quickly and effectively. This will help save lives and lessen the financial and medical burdens placed
on patients. ML can be really helpful in this situation. In this post, we go through the foundations of machine
learning and how it might help with skin cancer diagnosis. All clinical and photographic data are required
for AI research. It is necessary to record the sensitivities, specificities, and performance in future research
and in actual environments. AI is not a danger to dermatologists' knowledge; rather, in the next years, it
may be employed to enhance clinical practice. If practicing dermatologists have a greater understanding of
AI concepts, they will be able to deliver better skin care. Protecting patient information, having access to
enormous datasets, and training the AI algorithms to improve diagnosis accuracy are some of the hurdles
in implementing AI for the diagnosis of skin cancer.
In order to differentiate between malignant and benign skin lesions in several picture modalities, including
dermoscopic, clinical, and histopathological images, researchers have created AI solutions, specifically
deep learning algorithms. In this review, we discuss advancement in the digital image-based AI solutions
for the diagnosis of skin cancer, along with some challenges and future opportunities to improve these AI
systems. Despite the various claims that AI systems can classify different skin lesions with greater accuracy
than dermatologists, these AI systems are still in the very early stages of clinical application.
We use convolutional neural networks to tackle the classification issue for skin cancer. Many cancer cases
are incorrectly identified in the early stages, which can have serious repercussions, including patient death.
However, there are instances where individuals present with other issues and doctors diagnose skin cancer.
This results in wasteful time and financial outlays for additional diagnosis. Using transfer learning
architecture and deep neural networks, we overcome both of the aforementioned issues in this work. We
trained and tested our network using ISIC databases that are available to the public. Our model outperforms
prior state-of-the-art methods with accuracy of 0.935, precision of 0.94, recall of 0.77, F1 score of 0.85,
and ROC-AUC of 0.861. By utilizing very deep convolutional neural networks with transfer learning and
fine-tuning them on dermoscopy pictures, we demonstrated that improved diagnostic accuracy could be
attained in comparison to experienced doctors and clinicians. Despite the lack of a preprocessing step in
this article, the experimental findings are quite encouraging. In order to help dermatologists, these models
may be simply integrated into dermoscopy equipment or even on smartphones.
Sr. Year Title Citations Author
1 2021 “A machine learning model for skin 74 Viswanatha Reddy Allugunti
disease classification using convolution
neural network ”
2 2019 “A method of skin cancer disease 67 Nawal Soliman Alkolifi
detection using image processing and Alenezi
machine learning ”
3 2022 “Machine learning and deep learning 9 Solene Bechelli, Jerome
algorithms for skin cancer classification Delhommelle
from dermoscopic image ”
4 2022 “ Machine learning approach for skin cancer 58 Flavia Grignaffini, Francesco
classification from dermoscopic images: A Barbuto et all.
systematic review ”
5 2022 “ Detection of skin cancer based on skin 5 Walaa Gouda, Najm Us Sama
lesion image using deep learning ”
6 2021 “ An attention based mechanism to combine 37 Andre G. C. Pacheco, Renato
image and metadata in deep learning models A. Krohling
applied to kin cancer classification ”
7 2021 “ Skin cancer detection a review using deep 71 Mehwish Dildar, Shumaila
learning techniques ” Akram et all.
8 2021 “ Machine learning and its application in 15 Kinnor Das, Anant Patil et all.
skin cancer ”
9 2020 “Artificial intelligence based image 135 Manu Goyal, Thomas
classification method for diagnosis of skin Knackstedt et all.
cancer ”
10 2021 “ Convolutional neural network for 26 Abhinav Sagar, Dheeba Jacob
classifying melanoma images ”

You might also like