A Review of Skin Cancer Classifications and

Management Strategies
Muthamizhan.M B. Sathyasri
Department of Electronics and communication, Department of Electronics and communication,
Vel Tech Rangarajan Dr Sagunthala R & D Institute of Science Vel Tech Rangarajan Dr Sagunthala R & D Institute of Science
and Technology and Technology
Avadi, Chennai, India Avadi, Chennai, India
[email protected] [email protected]
possible through various methods. This research systematic
ally reviews deep learning algorithms for early detection, e
Abstract—Three different kinds of skin cancer can be xamining high-quality studies and utilizing tools, graphs, ta
discovered on Earth: Cutaneous malignancies include bles, methodologies, and frameworks. Schierbeck et al. (20
melanoma, basal cell carcinoma and squamous cell 19) report 15,000 annual skin cancer cases in Denmark (po
carcinoma of skin. The various types of skin cancer have pulation 5.7 million), caused mainly by UVR, immunosupp
their causes, effects, or outcomes and the approaches of
ressive treatment, and irradiation.
handling the treatments separately. Out of all types of
cancer, skin cancer is the most widespread type of tumor. Kadampur and Al Riyaee (2020) identify melanoma, ba
Consequently, the aim of this review is to present a detailed sal cell carcinoma, and squamous cell carcinoma as prevale
study on skin cancer subtypes by reviewing the various types nt skin cancers, each with unique challenges. This study re
of skin cancer, their molecular characteristics, risk factors views these cancers' molecular mechanisms, risk factors, a
and clinical manifestations. Moreover, it explores the nd symptoms. It highlights diagnostic advancements like m
available modern diagnostic procedures like molecular olecular testing and dermoscopy, and explores melanoma
testing, dermoscopy, and histopathology that facilitated
management strategies including genetic profiling and tailo
better staging of cancers and better early diagnosis. The
study also explores the change of management strategies
red treatments. Public health programs and sun protection a
currently in use such as radiation therapy, surgical excisions, re emphasized for prevention, advocating a multidisciplinar
immunotherapies, photodynamic therapy, targeted therapies y approach to enhance therapy and patient outcomes. The
among others. It is therefore genetic screening and more so work of Berhil, Benlahmar, & Labani, (2020), reveals how
the personalized medicine which are right at the fore front of human capital analysis with the help of HR analytics positi
the treatment options for the advanced melanoma cases. vely affects profitability of companies. This review provide
Health promotion practices, including community awareness s a comprehensive analysis of HR risks and challenges and
and sun protection measures are other topics of debate. The the latest AI advancements addressing these issues. Summa
presentation of a concept focusing on the involvement of a rizing IT solutions from 2008 to 2018, it serves as a referen
plurality of disciplines in skin cancer patients care is one of ce for computer scientists, outlining how AI can transform
the goals of this research, and thus contributing to better HR processes and decision-making..
results in the fate of patients with this illness. To accomplish
this goal, both the old and the new approach will be looked at Kliegr, Bahník, & Fürnkranz (2021) emphasize that un
both ways. derstanding machine learning interpretability of models req
uire delving into cognitive science beyond syntactic compr
Keywords— Skin cancer, Melanoma, Basal cell carcinoma ehension. They discuss twenty cognitive biases affecting in
(BCC), Squamous cell carcinoma (SCC), Clinical symptoms, terpretability and provide debiasing solutions, calling for e
Diagnostic methods, Dermoscopy, Molecular testing, Skin mpirical studies to bridge the gap between cognitive psych
cancer staging, Skin cancer prevention ology and machine learning. Hajiarbabi (2023) introduces a
technique for differentiating melanoma from benign skin a
I. INTRODUCTION bnormalities using preprocessing steps, image augmentatio
The three commonly occurring skin cancer types in the n, and a CNN. A three-stage analysis focuses on lesion cent
world are melanocytic carcinoma, basalioma, and squamou res, with results combined via a fully connected neural net
s cell carcinoma. A primary type of skin cancer is melanom work. The method achieves 88.5% recall, 91.75% precision,
a, and most skin cancer incidents are melanoma. Each of th 94.42% accuracy, and an AUC of 0.94, demonstrating effe
ese subtypes presents certain difficulties in tracing the deve ctiveness.
lopment of the disease, its prognosis, and therapy. Squamo
us cell carcinoma and basal cell carcinoma are less dangero Lu & Firoozeh (2022) highlight the significant issue of
us and most frequently occurring compared to the melano skin cancer, 40% of all cancers, with over 300,000 new me
ma that is known to be highly invasive and fatal. The fact i lanoma cases in 2018. Traditional treatments like surgery, r
s that to enhance the diagnosis, prediction, and therapy of n adiation, and chemotherapy have been the mainstay of canc
umerous forms of skin cancer it is critical to have extensive er care for decades, but new image processing technologies
knowledge about the molecular profile and essential risk in show promise for early detection and treatment. An enhanc
dicators associated with these pathological conditions. ed XceptionNet model, with improved performance metric
s, demonstrates that modern machine learning methods can
II. LITERATURE SURVEY significantly enhance diagnosis accuracy. In the work of K
hanzode & Sarode (2020), the history of ET is described fr
Dildar et al. (2021) identify skin diseases, particularly s
om the concept of traditional form of AI to the integration
kin cancer from genetic mutations as highly perilous. Early
of AI in the creation of ICAI systems. The paper synthesize
detection, crucial due to high fatality and treatment costs is
s literature to explore applications of AI in education, emph ying challenges and suggesting improvements. AI and ML
asizing its potential for academic professionals and researc advancements could significantly enhance safe nanomateri
hers. al production.
According to Ullah, et al. (2020), smart cities aim Early melanoma detection can save lives and is ac
to tackle urbanization, energy management, environmental hievable through computer-aided approaches. This study pr
sustainability, and quality of life using ICT. AI, ML, and D oposes an ensemble technique combining CNNs and visual
RL are emerging as crucial in addressing these complex iss texture feature extraction for automatic melanoma detectio
ues. Key applications include energy-saving smart grids, in n. It uses VGG-19 and a proposed network model for imag
telligent transportation systems, cybersecurity, and improv e categorization, with kernel Principal Component Analysi
ed 5G communication services using UAVs. The role of th s (kPCA) for feature dimension reduction. The technique w
ese technologies in smart healthcare systems is also examin as tested on ISIC 2016, ISIC 2019, and PH2 datasets.
ed. The report identifies research roadblocks and suggests a
Rojaramani et al.(2020) Melanoma, a lethal skin conditi
pproaches to enhance smart city initiatives with AI, ML, an
on, can be malignant or non-malignant. Benign melanoma i
d DRL. Hermosilla, et al. (2024), reviews computer-assiste
nvolves non-cancerous moles, while malignant melanoma,
d technology for early skin cancer detection via dermatosco
the worst type, causes bleeding sores and spreads slowly fr
pic image processing. It analyses 45 studies, focusing on al
om melanocytes. Early detection is crucial for effective tre
gorithms, accuracy, and validation. Despite advancements,
atment. Dermatology imaging technology aims to detect m
challenges like image quality and reliance on human interp
elanoma early, often using stable monitoring and diagnosti
retation persist. Further research is needed to improve pract
c success. Biopsies, though painful and time-consuming, ar
icality and reliability of these technologies, paving the way
e commonly used for diagnosis. Computerized approaches
for future advancements in skin cancer diagnostics.
offer more accurate and time-saving alternatives.
Lemaignan, et al. (2017), this study identifies cog
nitive and decisional challenges for effective human-robot III. SKIN CANCER IMAGE CAPTURING METHODS
collaboration. Essential cognitive abilities for robots includ A. Image acquisition
e geometric reasoning, multi-perspective evaluation, and af
fordance analysis. Knowledge models accounting for huma Mishra et al.(2017) Image acquisition is the first step in
n-robot differences, multi-modal communication, and awar video or image processing, allowing the camera to capture
eness-based task planning are crucial. The paper emphasize and transform images into editable data. It involves three st
s explicit knowledge management for natural human-robot ages: an energy-focusing optical system, energy reflected fr
interactions and calls for incorporating human-level meani om the object, and a device measuring the energy used. Var
ngs into AI decisions. It highlights the importance of robust ious cameras serve different purposes: x-ray sensitive came
knowledge management for successful human-robot collab ras for x-ray images, infrared-sensitive cameras for infrared
oration. images, and visual spectrum cameras for everyday photos.
Image processing begins with capturing images using meas
Thrall, et al. (2018), the ongoing development in c urable energy, such as light or electromagnetic waves.
omputing power, large datasets, and deep learning algorith
ms have sparked interest in AI in imaging. AI presents both B. Image Acquisition Model
opportunities and risks, requiring considerations of nomenc Creating an image involves an illumination source and
lature, data sharing, and validation across imaging platform objects that reflect or absorb energy. Light can be produced
s and patients. AI can help radiologists prioritize cases and by methods such as radar, infrared energy, computer-gener
enhance diagnostic results. Though some fear AI may repla ated patterns, ultrasonic energy, and X-rays. The type of ca
ce radiologists, this paper argues that AI will complement h mera sensor used depends on the lighting conditions. ‘Imag
uman efforts for better and faster diagnoses. Current limitat e acquisition’ refers to capturing an image, where light ener
ions include lack of knowledge and computer resources, ad gy hitting the sensor is transformed into digital images. Sen
dressable by technical support. AI's impact on imaging will sors convert incoming illumination into measurable voltage
be measured by patient health, time saved, improved diagn s. Charge-Coupled Devices (CCDs) and Complementary
ostic precision, and more free time for radiologists. Radiolo Metal-Oxide Semiconductors (CMOS) are common types
gists are expected to lead AI integration in medical imagin of image sensors. CCD sensors offer high-resolution image
g. s with minimal noise, while CMOS sensors are more susce
ptible to noise but consume less power. Image acquisition i
From Collenette, et al. (2023), the current study d nvolves a two-dimensional array of cells that convert light i
emonstrates the importance of explainability in AI tools for nto electrons, which are then processed to form a digital im
legal thinking, making solutions easily consumable for end age.
users. Experimental evaluation showed 97% accuracy in re
al-world scenarios, with user polls indicating high understa C. Skin Image Preprocessing
ndability. AI supports fast and effective legal decisions, de Studied in 2016 by Fioravante Oliveira and other resear
monstrating its potential for transparency and reliability in chers. Effective detection and analysis of pigmented skin le
complex legal domains. sions require pre-processing. Challenges in image segment
D. A. Winkler, explores advancements in robotics, ation include hairs, reflected light, shadows, skin lines, and
automation, and AI for nanomaterial synthesis and analysi air bubbles. The blue channel accentuates lesion regions, a
s. While automation has progressed, nanomaterials still req nd the original RGB image can be used in vector processin
uire further development for ML methods in data analysis. g without converting to other colour spaces like CIE or HS
The study examines traditional and deep learning approach V. Scalar techniques can be applied to each colour channel
es to nano safety and nanomaterial characterization, identif
and combined or processed using vector data processing al Sobel operator — Using the Gradient gives a resulting i
gorithms. mage is smoothed while the edges are enhanced by this sha
rpening filter, which makes use of a coefficient. In the mid
Artifacts from lighting changes can affect skin lesion se
dle, that it employs a weight value of 2. The total of all the
gmentation. A quadratic function models shading effects, r
masks is zero, as predicted for a derivative operator, as can
educing these effects when converting images back to RG
be seen in the examples.
B from HSV. Otsu's thresholding technique (Glaister et al.
2013) segments colour-corrected images. Novel approache Using multiple preprocessing filters in image processin
s, such as Monte Carlo sampling, correct lighting variance i g is common to enhance training datasets for computer visi
n macroscopic images. The approach uses Otsu's histogram on and machine learning. Noise reduction filters are often a
bimodality metric to find the best weights for converting an pplied before masks to focus on specific areas of an image.
RGB image to a grey-level image. Shade-of-grey methods Banerjee et al. (2020) highlight that dermoscopy, a costly d
(Barata, et al. 2014) correct dermoscopy images based on li iagnostic method for skin cancer, can be replaced with Blu
ght source hue. Motif-based set theory may be used to impr m's 'tape dermatoscopy' technique, which is cost-effective
ove the appearance of skin blemishes. without compromising image quality. This involves using a
transparent adhesive on the lesion after immersing the regi
Colour Morphological filtering enhances lesion areas b
on in immersion fluid. Yacin Sikkandar et al. (2021) descri
y reducing noise and low-contrast boundaries. Hair remova
be a two-step pre-processing method for skin lesion images.
l in macroscopic and dermoscopy images is crucial as it inf
The initial step includes format conversion and area of inte
luences lesion borders. Early methods identify hair removal
rest (ROI) recognition, followed by hair removal, as hair si
stages, particularly for dense black hair, using thresholds to
gnificantly impacts detection and classification. The image
create binary masks. Anisotropic diffusion and median filte
is first converted to grayscale, and then morphological ima
rs reduce hair visibility and enhance segmentation accuracy
ge processing, specifically a top hat filter, is used to identif
while retaining important lesion data. These filters effectiv
y and remove thick, black hair.
ely handle noisy images, and iterative application of anisotr
opic diffusion reduces high-frequency noise and enhances IV. SKIN CANCER – FEATURE EXTRACTION METHODS
edges. With the help of the supplied improvements, the ne
w method is desired not only to enhance edges but also to d TABLE I. QUALITATIVE COMPARISON MODELS WITH WELL KNOWN
enoise images without disregarding important edges. Aniso MODELS USED FOR FEATURE EXTRACTION
tropic diffusion filter iteration 150 was the final check that Title Classes Features
halted the smoothening process. A methodological app Benign and Mela 1. The form is characteriz
roach to the classificati noma: ed by the area, the level of its a
D. Image Processing: Filters for Noise Reduction and on of dermoscopy ima symmetry, compactness, and d
Edge Detection ges 2 Classes iameter.
Smoothing Filters are used for blurring and noise reduc 2. Colour features include
tion. Two types are Smoothing Linear Filters and Smoothin colour space, colour moment,
g Non-Linear Filters. An average filter determines the pixel centroids, standard deviation,
values' average in a region using a mask. A weighted avera mean and centroid diameter.
ge filter elevates specific pixels by multiplying their values. Computer Aided Diag Melanomas and 1.Color
Larger masks increase blurring. Convolution uses more pix nosis of Mealanocytic Mealanocytic:
Lesions 2 Classes 2.Texure
els to determine the average result. Median filters, a type of
non-linear filter, remove salt and pepper noise by sorting pi Pattern classification Benign Color and texture properties
xel values in a region, calculating the median, and assignin of dermoscopy Melanocytic can be used to derive
g it to the pixel. Sharpening Filters highlight transition in in images: a perceptually lesion and multiscale texture features and
tensity. Types include The Laplacian (Second Derivative), uniform model Melanoma. colour symmetry.
2 Classes
The Gradient (First Order Derivative), and Sobel Operators.
These filters show intensity changes around edges. The fir Automating melanoma Benign, The size comprises of the
st derivative shows pixel intensity fluctuations, while the se prediction and Dysplastic lesion bounding rectangle, area,
cond derivative indicates changes in direction. detection and Melanoma: traverse, and ploar
3classes measurements. How rounded
The Laplacian — Second Derivative used in this proce it is and how small it can be
dure is defined as follows, made. Colour defines the
present colour space
respectively.
Different facets of the slope
Segmentation and Melanoma, Color features: color space
Classification of skin Bullae,Seborrheic with color moment and RGB
lesions for disease keratosis,Shingles histogram. Texture features:
(1) diagnosis and GLCM with Haralick features
Squamous cell:
The Gradient — First Order Derivative is good for spot 5 Classes
ting preprocessing flaws. In image processing, the first deri
vatives are computed based on the gradient’s magnitude. T
he pace at which the gradient changes direction is expresse
d by this magnitude. This filter removes all of the image’s i
sotropic features.
 Banerjee et al. (2020) emphasize the importance rity criteria. Many methods begin with scalar photos, conve
of early lesion diagnosis in fighting skin cancer, rting the original color image to scalar data, like greyscale,
highlighting dermoscopy as a vital non-intrusive tool for for more efficient processing. Segmentation methods for ve
melanoma detection. Melanocytic skin lesions can be ctor images rely on data from individual color channels in t
benign, questionable, or malignant. The widely accepted he original image, which is computationally intensive and r
ABCD rule, introduced by Stolz in 1985, includes equires specific colour spaces. Table II discusses the variou
asymmetry, border irregularity, color variation, diameter s techniques used in the segmentation strategy.
beyond 6 millimeters, and evolution. Diagnosing early-
stage melanoma is challenging due to minor symmetry in Banerjee et al. (2020) state that after pre-processing,
shape and color. The Menzies technique classifies moles segmentation identifies the boundaries of the affected
region. Traditional methods like thresholding, area
based on one color and symmetry distribution. Yacin
augmenting, and clustering are time-consuming and
Sikkandar et al. (2021) utilize the Inception v4 model for
complex, making them less effective for melanoma
feature extraction from segmented skin lesion images. diagnosis. New approaches are replacing these methods.
Older Inception techniques divide repeating blocks into Yacin Sikkandar et al. (2021) describe semi-automated
sub-networks stored in the storage region. A simple filter image segmentation using a graph where each pixel is a
count tweak in Inception models maintains network node, with additional nodes for background and foreground
quality and improves computation and training rates. links. The graph is segmented using Min-Cut/Max-Flow
Tensor Flow-based Inception techniques minimize tensor and Grabcut models, with the Gaussian Mixture Model
matrices for backpropagation. Inception-v4 creates typical (GMM) providing area details. Tong et al. (2021) mention
possibilities for blocks of various grid sizes, as shown in traditional segmentation methods like threshold gradient
Table I, comparing models used for feature extraction. vector flow, region growth, and morphology-based
techniques, while CNN-based algorithms, like Full
A. Skin Lesion Segmentation convolutional residual networks (FcRN) and U-Net
networks, enhance segmentation by integrating multi-scale
TABLE II. SEGMENTATION METHODS AND ITS DIFFERENT
TECHNIQUES contextual information and addressing challenges such as
fixed receptive fields and gradient disappearance.
Segmentation Technique
Method
Region-based Statistical region merging Generative adversarial networks (GANs) improve
cutaneous lesion segmentation. Bi et al. (2017) proposed a
Iterative stochastic region merging multi-stage Fully Convolutional Network (FCN) with both
Region growing low-level and high-level semantic pyramids. Hasan et al. (2
020) introduced depth-wise separable convolution in the ds
Active contour-based Gradient vector flow
Net for better pixel space properties. Skin lesion networks,
Region-based active contour algorithm such as those using encoder-decoder FCNs, dense blocks, a
Active contour without edges nd CRFs (Adegun et al. 2020), enhance performance with l
ess model complexity, though border segmentation remains
Expectation-maximization level set challenging. Ashour et al. (2021) discuss segmentation tech
Adaptive snake niques that divide images into areas of interest using regio
n-based and edge-based approaches. Techniques like Partic
Level set
le Swarm Optimization (PSO), Ant Colony Optimization
AI-based Evolutionary computation (ACO), and Krill Herd methods enhance segmentation. Opt
k-means clustering imized Laplacian of Gaussian (LoG) energy and parallel co
mputing improve the Active Contour (AC) method. Shaha
Neural networks matnia and Ebadzadeh utilized revised PSO and the Bat alg
Fuzzy logic orithm to avoid local minima.
Thresholding-based Ensemble
Yuan et al. (2017) introduced a 19-layer DCNN-based
Renyi’s entropy approach, using the Jaccard Distance for loss function, test
Otsu’s thresholding ed with ISBI 2016 and PH2 datasets. Goyal et al. (2017) de
monstrated multi-class segmentation algorithms using the 2
Statistics 017 ISBI challenge dataset.
Adaptive thresholding
V. PARAMETER TUNING METHODS FOR SKIN IMAGE
Fuzzy logic
CLASSIFICATION
Iterative thresholding Banerjee et al. (2020) conducted a two-phase evaluatio
Edge-based Edge detectors n of methods to locate skin lesions using YOLOv3. The mo
del's ability to identify lesion locations in training images
Other methods Dynamic programming
was assessed using the IOU metric, claiming locations with
Hill-climbing algorithm an IOU score above 80%. The second phase evaluated sens
Factors such as equivalent grey levels or hues can itivity (Sen), specificity (Spe), the dice coefficient (Dic), an
impact the segmentation process, alongside similarity criter d the JACARD index (Acc). Sensitivity measured how wel
ia. Skin lesion identification in images can be achieved usi l lesions were segmented, specificity quantified segmented
ng thresholding and region-based segmentation with simila lesions, the dice coefficient examined relationships with gr
ound truth masks, and accuracy showed total pixel-wise se plified feature extraction, inspired by Gabor and HSV filter
gmentation performance. For the assessment metrics listed s, with significant GPU and dataset advancements. Barata e
above, the formula is as follows: t al. (2013) proposed an approach for melanoma identificati
on using global and local characteristics, extracting texture,
IOU = Area of Overlap/Area of Union shape, and color features through segmentation and histogr
Sen = TP/TP+FN ams. Esteva et al. (2017) used GoogleNet to classify over 1
Spe = TN/TN+FP 00,000 skin lesion datasets. Zhang et al. (2018) employed d
Dic = 2xTP/(2xTP)+FP+FN eep learning to classify four skin disorders using dermatosc
Jac = TP/TP+FN+FP opic images, achieving 86% accuracy on 1,067 images, hig
hlighting the importance of domain expertise and hierarchi
Acc = TP+TN/TP+TN+FN+FP cal structure. Table 5 represents the qualitative comparison
True positive (TP), true negative (TN), false positive (F models with well-known models used for classification.
P), and false negative (FN) represent the outcomes in imag
e segmentation. TP pixels are accurately recognized and se TABLE IV. QUALITATIVE COMPARISON MODELS WITH WELL KNOWN
gmented, whereas FN pixels are not. Non-lesion pixels are MODELS USED FOR CLASSIFICATION
TN if correctly identified as non-lesion, while FP pixels are Classifiers Classes Title
inaccurately recognized as lesion pixels. Deep learning and 5 Classes:
Segmentation and
machine learning methods are employed for image classific combined KNN and Squamous cell,
Classification of skin
ation. Table 4 presents the image classification techniques SVM classifiers, Shingles, Seborrheic
lesions for disease
used by these methods. SVM, and KNN keratosis, Bullae,
diagnosis.
Melanoma
Pattern classification
TABLE III. IMAGE CLASSIFICATION USED BY DEEP LEARNING AND AdaBoostMC 2 Classes:
of dermoscopy
MACHINE LEARNING METHODS Adaptive Boosting Melanoma and
images: a
Multi channel Benign Melanocytic
Author & perceptually uniform
Objective Method classification lesion
Year model
Lee et al. Using pre-trained data and a Fine-tuned Neural 3 Classes: Automated
2018 Fine-Tune Neural Network, KNN classifer Melanoma, Dysplatic Melanoma
it is easy to create and update lesion and Benign Recognisation.
a new challenge. A methodological
2 classes:
By storing the data across the NN, SVM, 2-KNN, approach to the
Benign and
network nodes, Artificial and Decision Trees classification of
Melanoma
Neural Networks are dermoscopy images
Noord et al. Artificial Neural Computer Aided
effective at recognizing non- 2 classes:
2017 Networks Statistical learning Diagnosis of
linear relationships between Melanoma and
the dependent and SVM Mealanocytic
Mealanocytic
independent parameters. Lesions
Models of Convolutional
Neural Networks are
effective in selecting crucial VII. CONCLUSION
Naranjo- characteristics automatically.
Convolutional Neural The skin cancer, including melanoma, BCC, and SCC,
Torres et al. Instead of keeping the
Networks
2020 network node’s training data poses a significant global health challenge. Differentiating t
in auxiliary memory, the ypes of skin cancer through clinical signs, risk factors, and
CNN model uses multi-layer molecular markers is crucial. New diagnostic tests such as
perceptrons to store it.
Structured and unstructured
molecular tests, dermoscopy, and histopathology have impr
data may be processed using oved early diagnosis and patient outcomes. The dynamic fi
Deep Neural Networks- eld of skin cancer treatment includes targeted drugs, radiati
Yadav et al. Deep Neural
2019
based algorithms.
Networks on therapy, immunotherapy, and excision procedures, offer
Unlabeled data may still be ing hope for complex cases like metastasized melanoma. G
used by the models and
produce a better result.
enetic and molecular screening enable tailored treatment ap
proaches. Prevention through community campaigns and s
VI. SKIN LESION CLASSIFICATION un protection is vital. Eradicating this pervasive disease req
uires enhancing patient outcomes through interdisciplinary
Researchers (Lima et al. 2021; Hameed et al. 2018) eva approaches, individualized treatments, and preventive meas
luated 1,011 images using a 7-point system, creating multi ures. Cross-disciplinary research and collaboration are esse
modal CNNs that process clinical and dermoscopic images, ntial for effective skin cancer management.
along with metadata. They used two Inception v3 CNNs, e
ach representing one of seven criteria, trained to detect abn REFERENCES
ormalities. The ISIC2018 (HAM10000) collection included
10,015 dermoscopic images for identifying benign, necroti
[1] Kadampur, M. A., & Al Riyaee, S. (2020). Skin cancer detection:
c, actinic keratosis, or melanoma lesions, using two pre-trai Applying a deep learning based model driven architecture in the
ned Inception-v3 CNNs for combined output. ISIC archive cloud for classifying dermal cell images. Informatics in Medicine
(2018) datasets, testing Multi-Crop, Recurrent output modu Unlocked, 18, 100282.I. S. Jacobs and C. P. Bean, “Fine particles,
les, and Patch-based Attention against baseline deep-featur thin films and exchange anisotropy,” in Magnetism, vol. III, G. T.
Rado and H. Suhl, Eds. New York: Academic, 1963, pp. 271–350.
e extractors like Inception-v3, DenseNet-121, and SEResne
[2] Berhil, S., Benlahmar, H., & Labani, N. (2020). A review paper on
xt50. They explored four methods for balancing multi-class artificial intelligence at the service of human resources
classification datasets. Bhardwaj et al. (2021) noted that ad management. Indonesian Journal of Electrical Engineering and
vances in image processing and machine learning have sim Computer Science, 18(1), 32-40.
[3] Kliegr, T., Bahník, Š., & Fürnkranz, J. (2021). A review of possible [21] Tong, X, Wei, J, Sun, B, Su, S, Zuo, Z & Wu, P 2021, ‘ASCU-Net:
effects of cognitive biases on interpretation of rule-based machine attention gate, spatial and channel attention u-net for skin lesion
learning models. Artificial Intelligence, 295, 103458. segmentation’, Diagnostics, vol. 11(3), pp.501.
[4] Hajiarbabi, M. (2023). Skin cancer detection using multi-scale deep [22] Yu, Z, Jiang, X, Zhou, F, Qin, J, Ni, D, Chen, S, Lei, B & Wang, T
learning and transfer learning. Journal of Medical Artificial 2018,‘Melanoma recognition in dermoscopy images via aggregated
Intelligence, 6. deep convolutional features’, IEEE Transactions on Biomedical
[5] Lu, X., & Firoozeh Abolhasani Zadeh, Y. A. (2022). Deep Engineering, vol. 66(4), pp.1006-1016.
Learning‐Based Classification for Melanoma Detection Using [23] Song, L, Lin, J, Wang, ZJ & Wang, H 2019, ‘Dense-residual
XceptionNet. Journal of Healthcare Engineering, 2022(1), attention network for skin lesion segmentation’, Proceedings of the
2196096. International Workshop on Machine Learning in Medical
[6] Khanzode, K. C. A., & Sarode, R. D. (2020). Advantages and Imaging,Springer, Cham, pp. 319-327.
disadvantages of artificial intelligence and machine learning: A [24] Hasan, MK, Dahal, L, Samarakoon, PN, Tushar, FI& Martí, R
literature review. International Journal of Library & Information 2020,‘DSNet: Automatic dermoscopic skin lesion segmentation’,
Science (IJLIS), 9(1), 3. Computers in Biology and Medicine, vol. 120, p.103738.
[7] Ullah, Z., Al-Turjman, F., Mostarda, L., & Gagliardi, R. (2020). [25] Adegun, AA & Viriri, S 2020, ‘FCN-based DenseNet framework
Applications of artificial intelligence and machine learning in smart for automated detection and classification of skin lesions in
cities. Computer Communications, 154, 313-323. dermoscopy images’, IEEE Access, vol. 8, pp.150377-150396.
[8] Hermosilla, P., Soto, R., Vega, E., Suazo, C., & Ponce, J. (2024). [26] Ashour, AS, Nagieb, RM, El-Khobby, HA, Abd Elnaby, MM &
Skin Cancer Detection and Classification Using Neural Network Dey, N 2021, ‘Genetic algorithm-based initial contour optimization
Algorithms: A Systematic Review. Diagnostics, 14(4), 454. for skin lesion border detection’, Multimedia Tools and
[9] Lemaignan, S., Warnier, M., Sisbot, E. A., Clodic, A., & Alami, R. Applications, vol. 80(2), pp.2583-2597.
(2017). Artificial cognition for social human–robot interaction: An [27] K. Ding Azad, M, Bozorg-Haddad, O & Chu, X 2018, ‘Flower
implementation. Artificial Intelligence, 247, 45-69. pollination algorithm (FPA)’, In Advanced optimization by nature-
[10] Thrall, J. H., Li, X., Li, Q., Cruz, C., Do, S., Dreyer, K., & Brink, J. inspired algorithms pp. 59-67. Springer, Singapore.
(2018). Artificial intelligence and machine learning in radiology: [28] Isah, OR, Usman, AD & Tekanyi, AM 2017, ‘Medical image
opportunities, challenges, pitfalls, and criteria for success. Journal segmentation through bat-active contour algorithm’.
of the American College of Radiology, 15(3), 504-508. [29] Goyal, M, Yap, MH & Hassanpour, S 2017, ‘Multi-class semantic
[11] Collenette, J., Atkinson, K., & Bench-Capon, T. (2023). segmentation of skin lesions via fully convolutional networks’,
Explainable AI tools for legal reasoning about cases: A study on the arXiv preprint arXiv:1711.10449.
European Court of Human Rights. Artificial Intelligence, 317, [30] Y, Chao, M & Lo, YC 2017, ‘Automatic skin lesion segmentation
103861. using deep fully convolutional networks with jaccard distance’,
[12] Winkler, D. A. (2020). Role of artificial intelligence and machine IEEE transactions on medical imaging, vol.36(9), pp.1876-1886.
learning in nanosafety. Small, 16(36), 2001883. [31] Lee, YC, Jung, SH & Won, HH 2018, ‘WonDerM: Skin lesion
[13] Dildar, M, Akram, S, Irfan, M, Khan, H.U, Ramzan, M, Mahmood, classification with fine-tuned neural networks’, arXiv preprint
AR, Alsaiari, SA, Saeed, AHM, Alraddadi, MO & Mahnashi, MH arXiv:1808.03426.
2021, ‘Skin cancer detection: a review using deep learning [32] Van Noord, N & Postma, E 2017, ‘Learning scale-variant and
techniques’, International journal of environmental research and scale-invariant features for deep image classification’, Pattern
public health, vol. 18(10), pp.5479. Recognition, vol. 61, pp.583-592.
[14] Schierbeck, J, Vestergaard, T & Bygum, A 2019,‘Skin cancer [33] Yadav, SS & Jadhav, SM 2019, ‘Deep convolutional neural
associated genodermatoses: a literature review’, Acta Derm network based medical image classification for disease diagnosis’,
Venereol, vol. 99(4), pp.360-369. Journal of Big Data, vol. 6(1), pp.1-18.
[15] Rojaramani, D 2020, ‘An analysis for melanoma diagnosis in [34] Naranjo-Torres, J, Mora, M, Hernández-García, R, Barrientos, RJ,
dermascopic images using integrated segmentation and multiple Fredes, C & Valenzuela, A 2020,‘A review of convolutional neural
feature extraction’. network applied to fruit image processing’, Applied Sciences, vol.
[16] Mishra, VK, Kumar, S & Shukla, N 2017, ‘Image acquisition and 10(10), pp.3443.
techniques to perform image acquisition’, SAMRIDDHI: A Journal [35] Lima, DM, Rodrigues-Jr, JF, Brandoli, B, Goeuriot, L & Amer-
of Physical Sciences, Engineering and Technology, vol. 9(01), Yahia, S 2021, ‘Dermadl: advanced convolutional neural networks
pp.21-24. for computer-aided skin-lesion classification’, SN Computer
[17] Oliveira, RB, Mercedes Filho, E, Ma, Z, Papa, JP, Pereira, AS& Science, vol. 2(4), pp.1-13.
Tavares, JMR 2016, ‘Computational methods for the image [36] Harangi, B, Baran, A&Hajdu, A 2020,‘Assisted deep learning
segmentation of pigmented skin lesions: a review’, Computer framework for multi-class skin lesion classification considering a
methods and programs in biomedicine, vol. 131, pp.127-141. binary classification support’, Biomedical Signal Processing and
[18] Banerjee, S, Singh, SK, Chakraborty, A, Das, A & Bag, R 2020, Control, vol. 62, pp.102041.
‘Melanoma diagnosis using deep learning and fuzzy logic’, [37] Bhardwaj, A & Rege, PP 2021,‘Skin lesion classification using
Diagnostics, vol. 10(8), pp.577. deep learning’, Proceedings of the Advances in Signal and Data
[19] Yacin Sikkandar, M, Alrasheadi, BA, Prakash, NB, Hemalakshmi, Processing, pp. 575-589. Springer, Singapore.
GR, Mohanarathinam, A & Shankar, K 2021, ‘Deep learning based [38] Esteva, A, Kuprel, B, Novoa, RA, Ko, J, Swetter, SM, Blau, HM &
an automated skin lesion segmentation and intelligent classification Thrun, S 2017, ‘Dermatologist-level classification of skin cancer
model’, Journal of ambient intelligence and humanized computing, with deep neural networks’, nature, vol. 542(7639), pp.115-118.
vol. 12(3), pp.3245-3255.
[39] Zhang, X, Wang, S, Liu, J & Tao, C 2018, ‘Towards improving
[20] Szegedy, C, Ioffe, S, Vanhoucke, V & Alemi, AA 2017, diagnosis of skin diseases by combining deep neural network and
‘Inception-v4, inception-resnet and the impact of residual human knowledge’, BMC medical informatics and decision
connections on learning’, In Thirty-first AAAI conference on making, vol. 18(2), pp.69-76.
artificial intelligence.